You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@tvm.apache.org by tq...@apache.org on 2020/05/19 17:23:42 UTC

[incubator-tvm-site] branch asf-site updated: Docs build at Tue May 19 10:23:27 PDT 2020

This is an automated email from the ASF dual-hosted git repository.

tqchen pushed a commit to branch asf-site
in repository https://gitbox.apache.org/repos/asf/incubator-tvm-site.git


The following commit(s) were added to refs/heads/asf-site by this push:
     new 7fdf1c1  Docs build at Tue May 19 10:23:27 PDT 2020
7fdf1c1 is described below

commit 7fdf1c1c32ccc5cf5683c0196833d17aa8e4f1c3
Author: tqchen <tq...@octoml.ai>
AuthorDate: Tue May 19 10:23:28 2020 -0700

    Docs build at Tue May 19 10:23:27 PDT 2020
---
 .../from_tflite.py                                 |   19 +-
 .../from_tflite.ipynb                              |    2 +-
 docs/_sources/api/links.rst.txt                    |   25 +
 docs/_sources/api/python/contrib.rst.txt           |    6 +-
 docs/_sources/api_links.rst.txt                    |   25 -
 docs/_sources/contribute/code_guide.rst.txt        |   41 +
 docs/_sources/contribute/pull_request.rst.txt      |   11 +-
 docs/_sources/dev/convert_layout.rst.txt           |   52 +-
 docs/_sources/dev/relay_add_pass.rst.txt           |    2 +-
 docs/_sources/index.rst.txt                        |    4 +-
 docs/_sources/langref/index.rst.txt                |    1 +
 docs/_sources/langref/relay_pattern.rst.txt        |  141 +
 .../tutorials/autotvm/sg_execution_times.rst.txt   |   16 +-
 .../tutorials/autotvm/tune_conv2d_cuda.rst.txt     |   42 +-
 .../tutorials/autotvm/tune_simple_template.rst.txt |   20 +-
 .../tutorials/cross_compilation_and_rpc.rst.txt    |    2 +-
 .../tutorials/dev/low_level_custom_pass.rst.txt    |   26 +-
 .../tutorials/dev/sg_execution_times.rst.txt       |    6 +-
 .../frontend/deploy_model_on_android.rst.txt       |    2 +-
 .../tutorials/frontend/deploy_prequantized.rst.txt |    2 +-
 .../tutorials/frontend/deploy_ssd_gluoncv.rst.txt  |   78 +-
 docs/_sources/tutorials/frontend/from_onnx.rst.txt |    4 +-
 .../tutorials/frontend/from_tensorflow.rst.txt     |    2 +-
 .../tutorials/frontend/from_tflite.rst.txt         |   19 +-
 .../tutorials/frontend/sg_execution_times.rst.txt  |   34 +-
 docs/_sources/tutorials/language/reduction.rst.txt |  135 +-
 docs/_sources/tutorials/language/scan.rst.txt      |   90 +-
 .../tutorials/language/schedule_primitives.rst.txt |  189 +-
 .../tutorials/language/sg_execution_times.rst.txt  |   18 +-
 docs/_sources/tutorials/language/tensorize.rst.txt |  161 +-
 .../tutorials/language/tuple_inputs.rst.txt        |   69 +-
 .../tutorials/optimize/opt_conv_cuda.rst.txt       |    2 +-
 .../tutorials/optimize/opt_conv_tensorcore.rst.txt |  232 +-
 docs/_sources/tutorials/optimize/opt_gemm.rst.txt  |  266 +-
 .../tutorials/optimize/sg_execution_times.rst.txt  |   10 +-
 docs/_sources/tutorials/relay_quick_start.rst.txt  |    4 +-
 docs/_sources/tutorials/sg_execution_times.rst.txt |    8 +-
 docs/_sources/tutorials/topi/intro_topi.rst.txt    |  615 ++--
 .../tutorials/topi/sg_execution_times.rst.txt      |    4 +-
 docs/_sources/vta/install.rst.txt                  |  247 +-
 .../tutorials/autotvm/sg_execution_times.rst.txt   |    4 +-
 .../vta/tutorials/autotvm/tune_relay_vta.rst.txt   |   20 +-
 .../frontend/deploy_classification.rst.txt         |    4 +-
 .../tutorials/frontend/deploy_detection.rst.txt    |    2 +-
 .../tutorials/frontend/sg_execution_times.rst.txt  |    6 +-
 .../_sources/vta/tutorials/matrix_multiply.rst.txt |  194 +-
 .../vta/tutorials/optimize/convolution_opt.rst.txt |  425 +--
 .../tutorials/optimize/matrix_multiply_opt.rst.txt |  302 +-
 .../tutorials/optimize/sg_execution_times.rst.txt  |    6 +-
 .../vta/tutorials/sg_execution_times.rst.txt       |    6 +-
 .../_sources/vta/tutorials/vta_get_started.rst.txt |   87 +-
 docs/api/links.html                                |  251 ++
 docs/api/python/autotvm.html                       |    2 +-
 docs/api/python/contrib.html                       |   30 +-
 docs/api/python/driver.html                        |    2 +-
 docs/api/python/error.html                         |   13 +-
 docs/api/python/graph_runtime.html                 |    2 +-
 docs/api/python/index.html                         |    4 +-
 docs/api/python/ir.html                            |    2 +-
 docs/api/python/ndarray.html                       |    2 +-
 docs/api/python/relay/analysis.html                |    2 +-
 docs/api/python/relay/backend.html                 |    2 +-
 docs/api/python/relay/frontend.html                |    4 +-
 docs/api/python/relay/image.html                   |    2 +-
 docs/api/python/relay/index.html                   |  234 +-
 docs/api/python/relay/nn.html                      |  136 +-
 docs/api/python/relay/testing.html                 |    2 +-
 docs/api/python/relay/transform.html               |   11 +-
 docs/api/python/relay/vision.html                  |    2 +-
 docs/api/python/rpc.html                           |  217 +-
 docs/api/python/runtime.html                       |    6 +-
 docs/api/python/target.html                        |    2 +-
 docs/api/python/te.html                            |  247 +-
 docs/api/python/tir.html                           |    2 +-
 docs/api/python/topi.html                          |    6 +-
 docs/api/python/vta/index.html                     |    2 +-
 docs/api_links.html                                |  251 --
 docs/contribute/code_guide.html                    |   33 +-
 docs/contribute/code_review.html                   |    2 +-
 docs/contribute/committer_guide.html               |    2 +-
 docs/contribute/community.html                     |    2 +-
 docs/contribute/document.html                      |    2 +-
 docs/contribute/error_handling.html                |    2 +-
 docs/contribute/git_howto.html                     |    2 +-
 docs/contribute/index.html                         |    2 +-
 docs/contribute/pull_request.html                  |   13 +-
 docs/contribute/release_process.html               |    2 +-
 docs/deploy/android.html                           |    2 +-
 docs/deploy/cpp_deploy.html                        |    2 +-
 docs/deploy/hls.html                               |    2 +-
 docs/deploy/index.html                             |    2 +-
 docs/deploy/integrate.html                         |    2 +-
 docs/dev/benchmark.html                            |    2 +-
 docs/dev/codebase_walkthrough.html                 |    2 +-
 docs/dev/convert_layout.html                       |   49 +-
 docs/dev/debugger.html                             |    2 +-
 docs/dev/hybrid_script.html                        |    2 +-
 docs/dev/index.html                                |    6 +-
 docs/dev/inferbound.html                           |    2 +-
 docs/dev/introduction_to_module_serialization.html |    2 +-
 docs/dev/relay_add_op.html                         |    2 +-
 docs/dev/relay_add_pass.html                       |    4 +-
 docs/dev/relay_bring_your_own_codegen.html         |    2 +-
 docs/dev/relay_intro.html                          |    2 +-
 docs/dev/relay_op_strategy.html                    |    2 +-
 docs/dev/relay_pass_infra.html                     |    2 +-
 docs/dev/runtime.html                              |    2 +-
 docs/dev/security.html                             |    2 +-
 docs/dev/virtual_machine.html                      |    2 +-
 docs/doxygen/algorithm_8h.html                     |    2 +-
 docs/doxygen/algorithm_8h__incl.svg                | 1865 +++++-----
 docs/doxygen/algorithm_8h_source.html              |   32 +-
 docs/doxygen/analyzer_8h.html                      |   14 +-
 docs/doxygen/analyzer_8h__dep__incl.svg            |  795 ++---
 docs/doxygen/analyzer_8h__incl.svg                 | 1867 +++++-----
 docs/doxygen/analyzer_8h_source.html               |   55 +-
 docs/doxygen/annotated.html                        |  542 +--
 docs/doxygen/annotation_8h.html                    |    2 +-
 docs/doxygen/annotation_8h__incl.svg               | 1615 ++++-----
 docs/doxygen/annotation_8h_source.html             |   18 +-
 docs/doxygen/array__utils_8h.html                  |    2 +-
 docs/doxygen/array__utils_8h__incl.svg             | 2211 ++++++------
 docs/doxygen/array__utils_8h_source.html           |    4 +-
 docs/doxygen/attr__registry__map_8h.html           |  139 +
 docs/doxygen/attr__registry__map_8h__dep__incl.svg |  365 ++
 docs/doxygen/attr__registry__map_8h__incl.svg      |   39 +
 docs/doxygen/attr__registry__map_8h_source.html    |  112 +
 docs/doxygen/autodiff_8h.html                      |    2 +-
 docs/doxygen/autodiff_8h__incl.svg                 | 1831 +++++-----
 docs/doxygen/autodiff_8h_source.html               |    2 +-
 docs/doxygen/base_8h.html                          |   10 +-
 docs/doxygen/base_8h__dep__incl.svg                |  480 +--
 docs/doxygen/base_8h__incl.svg                     | 1649 ++++-----
 docs/doxygen/base_8h_source.html                   |   30 +-
 docs/doxygen/batch__matmul_8h.html                 |    6 +-
 docs/doxygen/batch__matmul_8h__incl.svg            | 2220 ++++++------
 docs/doxygen/batch__matmul_8h_source.html          |    8 +-
 docs/doxygen/bias__add_8h.html                     |    6 +-
 docs/doxygen/bias__add_8h__incl.svg                | 2204 ++++++------
 docs/doxygen/bias__add_8h_source.html              |    6 +-
 docs/doxygen/bitserial_8h.html                     |    2 +-
 docs/doxygen/bitserial_8h__incl.svg                | 1865 +++++-----
 docs/doxygen/bitserial_8h_source.html              |   68 +-
 docs/doxygen/bound_8h.html                         |    8 +-
 docs/doxygen/bound_8h__dep__incl.svg               |  876 +++--
 docs/doxygen/bound_8h__incl.svg                    | 1903 +++++------
 docs/doxygen/bound_8h_source.html                  |   12 +-
 docs/doxygen/broadcast_8h.html                     |   16 +-
 docs/doxygen/broadcast_8h__incl.svg                | 2330 ++++++-------
 docs/doxygen/broadcast_8h_source.html              |   56 +-
 docs/doxygen/buffer_8h.html                        |    8 +-
 docs/doxygen/buffer_8h__dep__incl.svg              |  892 ++---
 docs/doxygen/buffer_8h__incl.svg                   | 1625 ++++-----
 docs/doxygen/buffer_8h_source.html                 |   66 +-
 docs/doxygen/c__backend__api_8h__incl.svg          |   36 +-
 docs/doxygen/c__backend__api_8h_source.html        |   10 +-
 docs/doxygen/c__runtime__api_8h.html               |  196 +-
 docs/doxygen/c__runtime__api_8h__dep__incl.svg     | 1143 +++----
 docs/doxygen/c__runtime__api_8h__incl.svg          |   26 +-
 docs/doxygen/c__runtime__api_8h_source.html        |   30 +-
 docs/doxygen/classes.html                          |  252 +-
 docs/doxygen/classtvm_1_1Array-members.html        |    2 +-
 docs/doxygen/classtvm_1_1Array.html                |   16 +-
 .../classtvm_1_1AttrRegistryMap-members.html       |  110 +
 docs/doxygen/classtvm_1_1AttrRegistryMap.html      |  331 ++
 ...tvm_1_1AttrRegistryMapContainerMap-members.html |  110 +
 .../classtvm_1_1AttrRegistryMapContainerMap.html   |  325 ++
 ...1_1AttrRegistryMapContainerMap__coll__graph.svg |   25 +
 .../classtvm_1_1AttrRegistryMap__coll__graph.svg   |   26 +
 docs/doxygen/classtvm_1_1BaseExpr.html             |    2 +-
 docs/doxygen/classtvm_1_1BaseExprNode.html         |    2 +-
 .../classtvm_1_1BaseExprNode__inherit__graph.svg   | 1374 ++++----
 .../classtvm_1_1BaseExpr__inherit__graph.svg       |  328 +-
 docs/doxygen/classtvm_1_1Constructor-members.html  |    2 +-
 docs/doxygen/classtvm_1_1Constructor.html          |   10 +-
 docs/doxygen/classtvm_1_1EnvFunc-members.html      |    2 +-
 docs/doxygen/classtvm_1_1EnvFunc.html              |   10 +-
 docs/doxygen/classtvm_1_1FloatImmNode.html         |    2 +-
 .../classtvm_1_1FloatImmNode__coll__graph.svg      |   93 +-
 docs/doxygen/classtvm_1_1GenericOpMap-members.html |  110 -
 docs/doxygen/classtvm_1_1GenericOpMap.html         |  337 --
 .../classtvm_1_1GenericOpMap__coll__graph.svg      |   25 -
 .../doxygen/classtvm_1_1GlobalTypeVar-members.html |    2 +-
 docs/doxygen/classtvm_1_1GlobalTypeVar.html        |   10 +-
 .../classtvm_1_1GlobalTypeVarNode-members.html     |    2 +-
 docs/doxygen/classtvm_1_1GlobalTypeVarNode.html    |   12 +-
 .../classtvm_1_1GlobalTypeVarNode__coll__graph.svg |  122 +-
 docs/doxygen/classtvm_1_1GlobalVar-members.html    |    2 +-
 docs/doxygen/classtvm_1_1GlobalVar.html            |    8 +-
 .../doxygen/classtvm_1_1GlobalVarNode-members.html |    2 +-
 docs/doxygen/classtvm_1_1GlobalVarNode.html        |   12 +-
 .../classtvm_1_1GlobalVarNode__coll__graph.svg     |  166 +-
 docs/doxygen/classtvm_1_1IRModule-members.html     |    4 +-
 docs/doxygen/classtvm_1_1IRModule.html             |   22 +-
 docs/doxygen/classtvm_1_1IRModuleNode-members.html |   20 +-
 docs/doxygen/classtvm_1_1IRModuleNode.html         |  102 +-
 docs/doxygen/classtvm_1_1IntImmNode.html           |    2 +-
 .../classtvm_1_1IntImmNode__coll__graph.svg        |   93 +-
 docs/doxygen/classtvm_1_1Map-members.html          |    2 +-
 docs/doxygen/classtvm_1_1Map.html                  |   66 +-
 docs/doxygen/classtvm_1_1MapNode-members.html      |    2 +-
 docs/doxygen/classtvm_1_1MapNode.html              |   14 +-
 docs/doxygen/classtvm_1_1Op-members.html           |    6 +-
 docs/doxygen/classtvm_1_1Op.html                   |   42 +-
 docs/doxygen/classtvm_1_1OpAttrMap-members.html    |  113 +
 docs/doxygen/classtvm_1_1OpAttrMap.html            |  260 ++
 .../doxygen/classtvm_1_1OpAttrMap__coll__graph.svg |   64 +
 .../classtvm_1_1OpAttrMap__inherit__graph.svg      |   43 +
 docs/doxygen/classtvm_1_1OpMap-members.html        |  110 -
 docs/doxygen/classtvm_1_1OpMap.html                |  338 --
 docs/doxygen/classtvm_1_1OpMap__coll__graph.svg    |   25 -
 docs/doxygen/classtvm_1_1OpNode-members.html       |   33 +-
 docs/doxygen/classtvm_1_1OpNode.html               |   44 +-
 docs/doxygen/classtvm_1_1OpRegEntry-members.html   |  117 +
 docs/doxygen/classtvm_1_1OpRegEntry.html           |  593 ++++
 .../classtvm_1_1OpRegEntry__coll__graph.svg        |   32 +
 docs/doxygen/classtvm_1_1OpRegistry-members.html   |  117 -
 docs/doxygen/classtvm_1_1OpRegistry.html           |  568 ----
 .../classtvm_1_1OpRegistry__coll__graph.svg        |   32 -
 docs/doxygen/classtvm_1_1Op__coll__graph.svg       |   18 +-
 docs/doxygen/classtvm_1_1Op__inherit__graph.svg    |   18 +-
 docs/doxygen/classtvm_1_1PrimExprNode.html         |    4 +-
 .../classtvm_1_1PrimExprNode__coll__graph.svg      |   93 +-
 .../classtvm_1_1PrimExprNode__inherit__graph.svg   | 1104 +++---
 docs/doxygen/classtvm_1_1PrimTypeNode.html         |    2 +-
 .../classtvm_1_1PrimTypeNode__coll__graph.svg      |  131 +-
 docs/doxygen/classtvm_1_1RelayExpr.html            |    2 +-
 .../classtvm_1_1RelayExpr__inherit__graph.svg      |  324 +-
 docs/doxygen/classtvm_1_1SHashReducer.html         |    4 +-
 docs/doxygen/classtvm_1_1TensorTypeNode.html       |    2 +-
 .../classtvm_1_1TensorTypeNode__coll__graph.svg    |  173 +-
 ..._1_1arith_1_1ConstIntBoundAnalyzer-members.html |    9 +-
 ...classtvm_1_1arith_1_1ConstIntBoundAnalyzer.html |   31 +-
 .../classtvm_1_1relay_1_1AltPattern-members.html   |  108 +
 docs/doxygen/classtvm_1_1relay_1_1AltPattern.html  |  198 ++
 ...lasstvm_1_1relay_1_1AltPatternNode-members.html |  111 +
 .../classtvm_1_1relay_1_1AltPatternNode.html       |  257 ++
 ...tvm_1_1relay_1_1AltPatternNode__coll__graph.svg |   88 +
 ..._1_1relay_1_1AltPatternNode__inherit__graph.svg |   56 +
 ...lasstvm_1_1relay_1_1AltPattern__coll__graph.svg |   56 +
 ...stvm_1_1relay_1_1AltPattern__inherit__graph.svg |   56 +
 .../classtvm_1_1relay_1_1AttrPattern-members.html  |  108 +
 docs/doxygen/classtvm_1_1relay_1_1AttrPattern.html |  198 ++
 ...asstvm_1_1relay_1_1AttrPatternNode-members.html |  111 +
 .../classtvm_1_1relay_1_1AttrPatternNode.html      |  257 ++
 ...vm_1_1relay_1_1AttrPatternNode__coll__graph.svg |  111 +
 ...1_1relay_1_1AttrPatternNode__inherit__graph.svg |   56 +
 ...asstvm_1_1relay_1_1AttrPattern__coll__graph.svg |   56 +
 ...tvm_1_1relay_1_1AttrPattern__inherit__graph.svg |   56 +
 .../classtvm_1_1relay_1_1CallPattern-members.html  |  108 +
 docs/doxygen/classtvm_1_1relay_1_1CallPattern.html |  205 ++
 ...asstvm_1_1relay_1_1CallPatternNode-members.html |  113 +
 .../classtvm_1_1relay_1_1CallPatternNode.html      |  296 ++
 ...vm_1_1relay_1_1CallPatternNode__coll__graph.svg |  178 +
 ...1_1relay_1_1CallPatternNode__inherit__graph.svg |   58 +
 ...asstvm_1_1relay_1_1CallPattern__coll__graph.svg |   56 +
 ...tvm_1_1relay_1_1CallPattern__inherit__graph.svg |   56 +
 ...sstvm_1_1relay_1_1ConstantNode__coll__graph.svg |    2 +-
 .../classtvm_1_1relay_1_1DFPattern-members.html    |  106 +
 docs/doxygen/classtvm_1_1relay_1_1DFPattern.html   |  167 +
 ...stvm_1_1relay_1_1DFPatternCallback-members.html |  107 +
 .../classtvm_1_1relay_1_1DFPatternCallback.html    |  196 ++
 ..._1_1relay_1_1DFPatternCallbackNode-members.html |  110 +
 ...classtvm_1_1relay_1_1DFPatternCallbackNode.html |  252 ++
 ...relay_1_1DFPatternCallbackNode__coll__graph.svg |   85 +
 ...ay_1_1DFPatternCallbackNode__inherit__graph.svg |   39 +
 ..._1_1relay_1_1DFPatternCallback__coll__graph.svg |   38 +
 ...1relay_1_1DFPatternCallback__inherit__graph.svg |   38 +
 .../classtvm_1_1relay_1_1DFPatternFunctor.html     |  127 +
 ...Pattern_01_6n_00_01Args_8_8_8_08_4-members.html |  119 +
 ...nst_01DFPattern_01_6n_00_01Args_8_8_8_08_4.html |  668 ++++
 ...ern_01_6n_00_01Args_8_8_8_08_4__coll__graph.svg |   36 +
 ...m_1_1relay_1_1DFPatternFunctor__coll__graph.svg |   23 +
 ...classtvm_1_1relay_1_1DFPatternNode-members.html |  107 +
 .../classtvm_1_1relay_1_1DFPatternNode.html        |  188 ++
 ...stvm_1_1relay_1_1DFPatternNode__coll__graph.svg |   36 +
 ...m_1_1relay_1_1DFPatternNode__inherit__graph.svg |  237 ++
 ...sstvm_1_1relay_1_1DFPatternVisitor-members.html |  117 +
 .../classtvm_1_1relay_1_1DFPatternVisitor.html     |  446 +++
 ...m_1_1relay_1_1DFPatternVisitor__coll__graph.svg |   50 +
 ..._1relay_1_1DFPatternVisitor__inherit__graph.svg |   50 +
 ...classtvm_1_1relay_1_1DFPattern__coll__graph.svg |   37 +
 ...sstvm_1_1relay_1_1DFPattern__inherit__graph.svg |  227 ++
 ...sstvm_1_1relay_1_1DominatorPattern-members.html |  108 +
 .../classtvm_1_1relay_1_1DominatorPattern.html     |  204 ++
 ...m_1_1relay_1_1DominatorPatternNode-members.html |  112 +
 .../classtvm_1_1relay_1_1DominatorPatternNode.html |  274 ++
 ...1relay_1_1DominatorPatternNode__coll__graph.svg |   89 +
 ...lay_1_1DominatorPatternNode__inherit__graph.svg |   57 +
 ...m_1_1relay_1_1DominatorPattern__coll__graph.svg |   56 +
 ..._1relay_1_1DominatorPattern__inherit__graph.svg |   56 +
 .../classtvm_1_1relay_1_1ExprPattern-members.html  |  108 +
 docs/doxygen/classtvm_1_1relay_1_1ExprPattern.html |  197 ++
 ...asstvm_1_1relay_1_1ExprPatternNode-members.html |  110 +
 .../classtvm_1_1relay_1_1ExprPatternNode.html      |  240 ++
 ...vm_1_1relay_1_1ExprPatternNode__coll__graph.svg |  105 +
 ...1_1relay_1_1ExprPatternNode__inherit__graph.svg |   55 +
 ...asstvm_1_1relay_1_1ExprPattern__coll__graph.svg |   56 +
 ...tvm_1_1relay_1_1ExprPattern__inherit__graph.svg |   56 +
 docs/doxygen/classtvm_1_1relay_1_1Id-members.html  |    2 +-
 docs/doxygen/classtvm_1_1relay_1_1Id.html          |   10 +-
 .../classtvm_1_1relay_1_1IdNode-members.html       |    2 +-
 docs/doxygen/classtvm_1_1relay_1_1IdNode.html      |   10 +-
 .../classtvm_1_1relay_1_1IdNode__coll__graph.svg   |   38 +-
 .../classtvm_1_1relay_1_1OpImplementationNode.html |    8 +-
 ...sstvm_1_1relay_1_1OpSpecialization-members.html |    2 +-
 .../classtvm_1_1relay_1_1OpSpecialization.html     |   14 +-
 .../classtvm_1_1relay_1_1OpStrategy-members.html   |    2 +-
 docs/doxygen/classtvm_1_1relay_1_1OpStrategy.html  |   14 +-
 ...vm_1_1relay_1_1TupleGetItemPattern-members.html |  108 +
 .../classtvm_1_1relay_1_1TupleGetItemPattern.html  |  193 ++
 ..._1relay_1_1TupleGetItemPatternNode-members.html |  111 +
 ...asstvm_1_1relay_1_1TupleGetItemPatternNode.html |  252 ++
 ...lay_1_1TupleGetItemPatternNode__coll__graph.svg |   89 +
 ..._1_1TupleGetItemPatternNode__inherit__graph.svg |   57 +
 ..._1relay_1_1TupleGetItemPattern__coll__graph.svg |   57 +
 ...elay_1_1TupleGetItemPattern__inherit__graph.svg |   57 +
 .../classtvm_1_1relay_1_1TuplePattern-members.html |  108 +
 .../doxygen/classtvm_1_1relay_1_1TuplePattern.html |  191 ++
 ...sstvm_1_1relay_1_1TuplePatternNode-members.html |  110 +
 .../classtvm_1_1relay_1_1TuplePatternNode.html     |  240 ++
 ...m_1_1relay_1_1TuplePatternNode__coll__graph.svg |   97 +
 ..._1relay_1_1TuplePatternNode__inherit__graph.svg |   55 +
 ...sstvm_1_1relay_1_1TuplePattern__coll__graph.svg |   56 +
 ...vm_1_1relay_1_1TuplePattern__inherit__graph.svg |   56 +
 .../classtvm_1_1relay_1_1TypePattern-members.html  |  108 +
 docs/doxygen/classtvm_1_1relay_1_1TypePattern.html |  198 ++
 ...asstvm_1_1relay_1_1TypePatternNode-members.html |  111 +
 .../classtvm_1_1relay_1_1TypePatternNode.html      |  257 ++
 ...vm_1_1relay_1_1TypePatternNode__coll__graph.svg |  111 +
 ...1_1relay_1_1TypePatternNode__inherit__graph.svg |   56 +
 ...asstvm_1_1relay_1_1TypePattern__coll__graph.svg |   56 +
 ...tvm_1_1relay_1_1TypePattern__inherit__graph.svg |   56 +
 docs/doxygen/classtvm_1_1relay_1_1Var-members.html |    2 +-
 docs/doxygen/classtvm_1_1relay_1_1Var.html         |   10 +-
 .../classtvm_1_1relay_1_1VarNode-members.html      |    4 +-
 docs/doxygen/classtvm_1_1relay_1_1VarNode.html     |   16 +-
 .../classtvm_1_1relay_1_1VarPattern-members.html   |  108 +
 docs/doxygen/classtvm_1_1relay_1_1VarPattern.html  |  193 ++
 ...lasstvm_1_1relay_1_1VarPatternNode-members.html |  112 +
 .../classtvm_1_1relay_1_1VarPatternNode.html       |  283 ++
 ...tvm_1_1relay_1_1VarPatternNode__coll__graph.svg |  103 +
 ..._1_1relay_1_1VarPatternNode__inherit__graph.svg |   57 +
 ...lasstvm_1_1relay_1_1VarPattern__coll__graph.svg |   56 +
 ...stvm_1_1relay_1_1VarPattern__inherit__graph.svg |   56 +
 ...asstvm_1_1relay_1_1WildcardPattern-members.html |  107 +
 .../classtvm_1_1relay_1_1WildcardPattern.html      |  169 +
 ...vm_1_1relay_1_1WildcardPatternNode-members.html |  109 +
 .../classtvm_1_1relay_1_1WildcardPatternNode.html  |  219 ++
 ..._1relay_1_1WildcardPatternNode__coll__graph.svg |   54 +
 ...elay_1_1WildcardPatternNode__inherit__graph.svg |   54 +
 ...vm_1_1relay_1_1WildcardPattern__coll__graph.svg |   55 +
 ...1_1relay_1_1WildcardPattern__inherit__graph.svg |   55 +
 .../classtvm_1_1runtime_1_1DataType-members.html   |   24 +-
 docs/doxygen/classtvm_1_1runtime_1_1DataType.html  |   57 +-
 ...lasstvm_1_1runtime_1_1DataType__coll__graph.svg |   47 +-
 docs/doxygen/classtvm_1_1runtime_1_1DeviceAPI.html |    4 +-
 .../classtvm_1_1runtime_1_1NDArray-members.html    |   63 +-
 docs/doxygen/classtvm_1_1runtime_1_1NDArray.html   |   17 +
 ...classtvm_1_1runtime_1_1NDArray__coll__graph.svg |    2 +-
 ...sstvm_1_1runtime_1_1NDArray__inherit__graph.svg |    2 +-
 ...tvm_1_1runtime_1_1ObjectRef__inherit__graph.svg |    2 +-
 .../classtvm_1_1runtime_1_1PackedFunc-members.html |    2 +-
 .../doxygen/classtvm_1_1runtime_1_1PackedFunc.html |   18 +-
 docs/doxygen/classtvm_1_1runtime_1_1Registry.html  |    4 +-
 .../classtvm_1_1runtime_1_1TVMArgValue.html        |    4 +-
 .../classtvm_1_1runtime_1_1TVMArgsSetter.html      |   12 +-
 ...lasstvm_1_1runtime_1_1TVMMovableArgValue__.html |    6 +-
 .../classtvm_1_1runtime_1_1TVMPODValue__.html      |    2 +-
 .../classtvm_1_1runtime_1_1TVMRetValue.html        |    6 +-
 ...1_1TypedPackedFunc_3_01R_07Args_8_8_8_08_4.html |    8 +-
 ...untime_1_1vm_1_1VirtualMachine__coll__graph.svg |   16 +-
 .../classtvm_1_1te_1_1PlaceholderOpNode.html       |    2 +-
 ...tvm_1_1te_1_1PlaceholderOpNode__coll__graph.svg |  201 +-
 docs/doxygen/classtvm_1_1te_1_1Stage-members.html  |   37 +-
 docs/doxygen/classtvm_1_1te_1_1Stage.html          |   21 -
 .../classtvm_1_1te_1_1StageNode-members.html       |   23 +-
 docs/doxygen/classtvm_1_1te_1_1StageNode.html      |   19 +-
 .../classtvm_1_1te_1_1StageNode__coll__graph.svg   |  297 +-
 ...classtvm_1_1te_1_1StageNode__inherit__graph.svg |    2 +-
 .../classtvm_1_1te_1_1Stage__coll__graph.svg       |    2 +-
 .../classtvm_1_1te_1_1Stage__inherit__graph.svg    |    2 +-
 docs/doxygen/classtvm_1_1te_1_1TensorNode.html     |    2 +-
 .../classtvm_1_1te_1_1TensorNode__coll__graph.svg  |  149 +-
 docs/doxygen/classtvm_1_1tir_1_1AddNode.html       |    2 +-
 .../classtvm_1_1tir_1_1AddNode__coll__graph.svg    |  127 +-
 docs/doxygen/classtvm_1_1tir_1_1AllocateNode.html  |    2 +-
 ...lasstvm_1_1tir_1_1AllocateNode__coll__graph.svg |  227 +-
 docs/doxygen/classtvm_1_1tir_1_1AndNode.html       |    2 +-
 .../classtvm_1_1tir_1_1AndNode__coll__graph.svg    |  127 +-
 docs/doxygen/classtvm_1_1tir_1_1AnyNode.html       |    2 +-
 .../classtvm_1_1tir_1_1AnyNode__coll__graph.svg    |   93 +-
 docs/doxygen/classtvm_1_1tir_1_1BinaryOpNode.html  |    2 +-
 ...lasstvm_1_1tir_1_1BinaryOpNode__coll__graph.svg |  127 +-
 docs/doxygen/classtvm_1_1tir_1_1BroadcastNode.html |    2 +-
 ...asstvm_1_1tir_1_1BroadcastNode__coll__graph.svg |  127 +-
 .../doxygen/classtvm_1_1tir_1_1BufferLoadNode.html |    2 +-
 ...sstvm_1_1tir_1_1BufferLoadNode__coll__graph.svg |  113 +-
 docs/doxygen/classtvm_1_1tir_1_1BufferNode.html    |    2 +-
 .../classtvm_1_1tir_1_1BufferNode__coll__graph.svg |  181 +-
 .../classtvm_1_1tir_1_1CallNode-members.html       |   53 +-
 docs/doxygen/classtvm_1_1tir_1_1CallNode.html      |   26 +-
 .../classtvm_1_1tir_1_1CallNode__coll__graph.svg   |  227 +-
 ...classtvm_1_1tir_1_1CallNode__inherit__graph.svg |  112 +-
 docs/doxygen/classtvm_1_1tir_1_1CastNode.html      |    2 +-
 .../classtvm_1_1tir_1_1CastNode__coll__graph.svg   |  127 +-
 docs/doxygen/classtvm_1_1tir_1_1CmpOpNode.html     |    2 +-
 .../classtvm_1_1tir_1_1CmpOpNode__coll__graph.svg  |  127 +-
 docs/doxygen/classtvm_1_1tir_1_1DivNode.html       |    2 +-
 .../classtvm_1_1tir_1_1DivNode__coll__graph.svg    |  127 +-
 docs/doxygen/classtvm_1_1tir_1_1EQNode.html        |    2 +-
 .../classtvm_1_1tir_1_1EQNode__coll__graph.svg     |  127 +-
 docs/doxygen/classtvm_1_1tir_1_1FloorDivNode.html  |    2 +-
 ...lasstvm_1_1tir_1_1FloorDivNode__coll__graph.svg |  127 +-
 docs/doxygen/classtvm_1_1tir_1_1FloorModNode.html  |    2 +-
 ...lasstvm_1_1tir_1_1FloorModNode__coll__graph.svg |  127 +-
 docs/doxygen/classtvm_1_1tir_1_1GENode.html        |    2 +-
 .../classtvm_1_1tir_1_1GENode__coll__graph.svg     |  127 +-
 docs/doxygen/classtvm_1_1tir_1_1GTNode.html        |    2 +-
 .../classtvm_1_1tir_1_1GTNode__coll__graph.svg     |  127 +-
 docs/doxygen/classtvm_1_1tir_1_1LTNode.html        |    2 +-
 .../classtvm_1_1tir_1_1LTNode__coll__graph.svg     |  127 +-
 docs/doxygen/classtvm_1_1tir_1_1LetNode.html       |    2 +-
 .../classtvm_1_1tir_1_1LetNode__coll__graph.svg    |  157 +-
 docs/doxygen/classtvm_1_1tir_1_1LoadNode.html      |    2 +-
 .../classtvm_1_1tir_1_1LoadNode__coll__graph.svg   |  155 +-
 docs/doxygen/classtvm_1_1tir_1_1MaxNode.html       |    2 +-
 .../classtvm_1_1tir_1_1MaxNode__coll__graph.svg    |  127 +-
 docs/doxygen/classtvm_1_1tir_1_1MinNode.html       |    2 +-
 .../classtvm_1_1tir_1_1MinNode__coll__graph.svg    |  127 +-
 docs/doxygen/classtvm_1_1tir_1_1ModNode.html       |    2 +-
 .../classtvm_1_1tir_1_1ModNode__coll__graph.svg    |  127 +-
 docs/doxygen/classtvm_1_1tir_1_1MulNode.html       |    2 +-
 .../classtvm_1_1tir_1_1MulNode__coll__graph.svg    |  127 +-
 docs/doxygen/classtvm_1_1tir_1_1NENode.html        |    2 +-
 .../classtvm_1_1tir_1_1NENode__coll__graph.svg     |  127 +-
 docs/doxygen/classtvm_1_1tir_1_1NotNode.html       |    2 +-
 .../classtvm_1_1tir_1_1NotNode__coll__graph.svg    |  127 +-
 docs/doxygen/classtvm_1_1tir_1_1OrNode.html        |    2 +-
 .../classtvm_1_1tir_1_1OrNode__coll__graph.svg     |  127 +-
 docs/doxygen/classtvm_1_1tir_1_1RampNode.html      |    2 +-
 .../classtvm_1_1tir_1_1RampNode__coll__graph.svg   |  127 +-
 docs/doxygen/classtvm_1_1tir_1_1RealizeNode.html   |    2 +-
 ...classtvm_1_1tir_1_1RealizeNode__coll__graph.svg |  223 +-
 docs/doxygen/classtvm_1_1tir_1_1ReduceNode.html    |    2 +-
 .../classtvm_1_1tir_1_1ReduceNode__coll__graph.svg |  137 +-
 docs/doxygen/classtvm_1_1tir_1_1SelectNode.html    |    2 +-
 .../classtvm_1_1tir_1_1SelectNode__coll__graph.svg |  127 +-
 docs/doxygen/classtvm_1_1tir_1_1ShuffleNode.html   |    2 +-
 ...classtvm_1_1tir_1_1ShuffleNode__coll__graph.svg |  109 +-
 docs/doxygen/classtvm_1_1tir_1_1SizeVarNode.html   |    2 +-
 ...classtvm_1_1tir_1_1SizeVarNode__coll__graph.svg |  109 +-
 docs/doxygen/classtvm_1_1tir_1_1StringImmNode.html |    2 +-
 ...asstvm_1_1tir_1_1StringImmNode__coll__graph.svg |   93 +-
 docs/doxygen/classtvm_1_1tir_1_1SubNode.html       |    2 +-
 .../classtvm_1_1tir_1_1SubNode__coll__graph.svg    |  127 +-
 docs/doxygen/classtvm_1_1tir_1_1VarNode.html       |    2 +-
 .../classtvm_1_1tir_1_1VarNode__coll__graph.svg    |  109 +-
 .../classtvm_1_1transform_1_1PassInfo-members.html |    2 +-
 .../doxygen/classtvm_1_1transform_1_1PassInfo.html |   10 +-
 ...lasstvm_1_1transform_1_1Sequential-members.html |    2 +-
 .../classtvm_1_1transform_1_1Sequential.html       |   10 +-
 docs/doxygen/codegen_8h.html                       |    8 +-
 docs/doxygen/codegen_8h__incl.svg                  | 2262 ++++++-------
 docs/doxygen/codegen_8h_source.html                |   10 +-
 docs/doxygen/constant__utils_8h.html               |    8 +-
 docs/doxygen/constant__utils_8h__incl.svg          | 2396 ++++++-------
 docs/doxygen/constant__utils_8h_source.html        |   12 +-
 docs/doxygen/cublas_8h.html                        |    6 +-
 docs/doxygen/cublas_8h__incl.svg                   | 2284 ++++++-------
 docs/doxygen/cublas_8h_source.html                 |   10 +-
 docs/doxygen/cuda_2dense_8h.html                   |   14 +-
 docs/doxygen/cuda_2dense_8h__incl.svg              | 2188 ++++++------
 docs/doxygen/cuda_2dense_8h_source.html            |   32 +-
 docs/doxygen/cuda_2injective_8h.html               |   10 +-
 docs/doxygen/cuda_2injective_8h__incl.svg          | 2171 ++++++------
 docs/doxygen/cuda_2injective_8h_source.html        |   18 +-
 docs/doxygen/cuda_2normalization_8h.html           |    8 +-
 docs/doxygen/cuda_2normalization_8h__incl.svg      | 2255 ++++++-------
 docs/doxygen/cuda_2normalization_8h_source.html    |   22 +-
 docs/doxygen/cuda_2pooling_8h.html                 |   12 +-
 docs/doxygen/cuda_2pooling_8h__incl.svg            | 2164 ++++++------
 docs/doxygen/cuda_2pooling_8h_source.html          |   24 +-
 docs/doxygen/cuda_2reduction_8h.html               |   10 +-
 docs/doxygen/cuda_2reduction_8h__incl.svg          | 2171 ++++++------
 docs/doxygen/cuda_2reduction_8h_source.html        |   36 +-
 docs/doxygen/cuda_2softmax_8h.html                 |   10 +-
 docs/doxygen/cuda_2softmax_8h__incl.svg            | 2171 ++++++------
 docs/doxygen/cuda_2softmax_8h_source.html          |   22 +-
 docs/doxygen/data__layout_8h.html                  |    8 +-
 docs/doxygen/data__layout_8h__incl.svg             | 1895 +++++------
 docs/doxygen/data__layout_8h_source.html           |   88 +-
 docs/doxygen/data__type_8h.html                    |   10 +-
 docs/doxygen/data__type_8h__dep__incl.svg          | 1545 +++++----
 docs/doxygen/data__type_8h__incl.svg               |  104 +-
 docs/doxygen/data__type_8h_source.html             |   82 +-
 docs/doxygen/dataflow__matcher_8h.html             |  151 +
 docs/doxygen/dataflow__matcher_8h__incl.svg        | 1433 ++++++++
 docs/doxygen/dataflow__matcher_8h_source.html      |  118 +
 docs/doxygen/dataflow__pattern_8h.html             |  197 ++
 docs/doxygen/dataflow__pattern_8h__dep__incl.svg   |   52 +
 docs/doxygen/dataflow__pattern_8h__incl.svg        | 1443 ++++++++
 docs/doxygen/dataflow__pattern_8h_source.html      |  164 +
 docs/doxygen/dataflow__pattern__functor_8h.html    |  182 +
 .../dataflow__pattern__functor_8h__dep__incl.svg   |   33 +
 .../dataflow__pattern__functor_8h__incl.svg        | 1452 ++++++++
 .../dataflow__pattern__functor_8h_source.html      |  126 +
 docs/doxygen/debug_8h.html                         |    3 +-
 docs/doxygen/debug_8h__incl.svg                    | 1643 ++++-----
 docs/doxygen/debug_8h_source.html                  |   15 +-
 docs/doxygen/detail_2broadcast_8h.html             |    6 +-
 docs/doxygen/detail_2broadcast_8h__incl.svg        | 2301 +++++++------
 docs/doxygen/detail_2broadcast_8h_source.html      |   20 +-
 docs/doxygen/detail_2extern_8h.html                |    4 +-
 docs/doxygen/detail_2extern_8h__incl.svg           | 2227 ++++++------
 docs/doxygen/detail_2extern_8h_source.html         |   26 +-
 docs/doxygen/device__api_8h.html                   |    2 +-
 docs/doxygen/device__api_8h__incl.svg              |  581 ++--
 docs/doxygen/device__api_8h_source.html            |   54 +-
 docs/doxygen/device__copy_8h.html                  |    2 +-
 docs/doxygen/device__copy_8h__incl.svg             | 1615 ++++-----
 docs/doxygen/device__copy_8h_source.html           |   12 +-
 docs/doxygen/dilate_8h.html                        |    6 +-
 docs/doxygen/dilate_8h__incl.svg                   | 2294 ++++++-------
 docs/doxygen/dilate_8h_source.html                 |   20 +-
 docs/doxygen/dir_000004_000000.html                |    2 +-
 docs/doxygen/dir_000005_000000.html                |    2 +-
 docs/doxygen/dir_000007_000013.html                |    2 +-
 docs/doxygen/dir_000008_000007.html                |    2 +-
 docs/doxygen/dir_000012_000007.html                |    2 +-
 docs/doxygen/dir_000020_000000.html                |    2 +-
 .../dir_14640941691eabc830844bef8114fc0c_dep.svg   |    4 +-
 .../dir_194ecda214f05a38134392ac6a69b970_dep.svg   |    4 +-
 .../dir_501163e255a572c2b8f8622055dcc830_dep.svg   |    4 +-
 .../dir_519be2d4a83a987dbf989f1de527b870_dep.svg   |    4 +-
 .../dir_59fae77b9344561cec492d4c84fe16f0_dep.svg   |    4 +-
 .../dir_5da96592f3a7c442b838b075c58254c2_dep.svg   |    4 +-
 .../dir_63946bee875c6d52bce55e72a67a86ad.html      |    9 +
 .../dir_63946bee875c6d52bce55e72a67a86ad_dep.svg   |    8 +-
 .../dir_6cd4295f6ad5aa17e5b568d0e5b190e5.html      |    3 +
 .../dir_72c2f11201cd7636dc7624de0754daa5_dep.svg   |    4 +-
 .../dir_82fbc0b750acd2d16bf657eba2735040_dep.svg   |    4 +-
 .../dir_85f3d6180da4edc5eb489febd20e807a_dep.svg   |    4 +-
 .../dir_9028d5d964ed0471973ff1d6dc49c294_dep.svg   |    4 +-
 .../dir_919d148e3bc0d2ada10274b170555243_dep.svg   |    4 +-
 .../dir_9a72de697ed8769045f2ae88dc0db9c6_dep.svg   |    4 +-
 .../dir_a99f4dc995763900c0e382b7dd5d0038_dep.svg   |    4 +-
 .../dir_b4c7d8e826c599ba55146c099a14beb5_dep.svg   |    8 +-
 .../dir_cf1b2a4a187708c24d9394e639e6417e_dep.svg   |    4 +-
 .../dir_dc867ff9a37cad1764f1670dc7eba6c1_dep.svg   |    4 +-
 .../dir_e79ffd3285e304ad4c501fa62028ed74_dep.svg   |    4 +-
 docs/doxygen/driver__api_8h.html                   |   12 +-
 docs/doxygen/driver__api_8h__incl.svg              | 2186 ++++++------
 docs/doxygen/driver__api_8h_source.html            |    2 +-
 docs/doxygen/elemwise_8h.html                      |   20 +-
 docs/doxygen/elemwise_8h__incl.svg                 | 2108 ++++++------
 docs/doxygen/elemwise_8h_source.html               |   71 +-
 docs/doxygen/env__func_8h.html                     |    4 +-
 docs/doxygen/env__func_8h__dep__incl.svg           |  428 ++-
 docs/doxygen/env__func_8h__incl.svg                | 1165 +++----
 docs/doxygen/env__func_8h_source.html              |   44 +-
 docs/doxygen/error_8h.html                         |   10 +-
 docs/doxygen/error_8h__incl.svg                    | 1924 +++++------
 docs/doxygen/error_8h_source.html                  |    6 +-
 docs/doxygen/feature_8h.html                       |    6 +-
 docs/doxygen/feature_8h__incl.svg                  | 2207 ++++++------
 docs/doxygen/feature_8h_source.html                |   26 +-
 docs/doxygen/files.html                            |   42 +-
 docs/doxygen/flatten_8h.html                       |    6 +-
 docs/doxygen/flatten_8h__incl.svg                  | 2321 ++++++-------
 docs/doxygen/flatten_8h_source.html                |   10 +-
 docs/doxygen/functions.html                        |  257 +-
 docs/doxygen/functions_0x7e.html                   |    6 +-
 docs/doxygen/functions__.html                      |  377 ---
 docs/doxygen/functions_a.html                      |   37 +-
 docs/doxygen/functions_b.html                      |    6 +-
 docs/doxygen/functions_c.html                      |   27 +-
 docs/doxygen/functions_d.html                      |   12 +-
 docs/doxygen/functions_e.html                      |   11 +-
 docs/doxygen/functions_f.html                      |   13 +-
 docs/doxygen/functions_func_0x7e.html              |    3 +
 docs/doxygen/functions_func_a.html                 |   23 +-
 docs/doxygen/functions_func_c.html                 |   17 +-
 docs/doxygen/functions_func_d.html                 |    9 +-
 docs/doxygen/functions_func_e.html                 |    5 +-
 docs/doxygen/functions_func_f.html                 |    2 +-
 docs/doxygen/functions_func_g.html                 |   27 +-
 docs/doxygen/functions_func_h.html                 |    4 +-
 docs/doxygen/functions_func_i.html                 |   17 +-
 docs/doxygen/functions_func_l.html                 |    2 +-
 docs/doxygen/functions_func_m.html                 |    2 +-
 docs/doxygen/functions_func_n.html                 |    3 +-
 docs/doxygen/functions_func_o.html                 |   28 +-
 docs/doxygen/functions_func_p.html                 |    2 +-
 docs/doxygen/functions_func_r.html                 |    8 +-
 docs/doxygen/functions_func_s.html                 |   12 +-
 docs/doxygen/functions_func_t.html                 |   35 +-
 docs/doxygen/functions_func_v.html                 |   54 +-
 docs/doxygen/functions_g.html                      |   40 +-
 docs/doxygen/functions_h.html                      |    7 +-
 docs/doxygen/functions_i.html                      |   34 +-
 docs/doxygen/functions_j.html                      |    3 +-
 docs/doxygen/functions_k.html                      |    3 +-
 docs/doxygen/functions_l.html                      |    6 +-
 docs/doxygen/functions_m.html                      |   10 +-
 docs/doxygen/functions_n.html                      |   15 +-
 docs/doxygen/functions_o.html                      |   39 +-
 docs/doxygen/functions_p.html                      |   24 +-
 docs/doxygen/functions_r.html                      |   20 +-
 docs/doxygen/functions_rela.html                   |   27 +-
 docs/doxygen/functions_s.html                      |   15 +-
 docs/doxygen/functions_t.html                      |   51 +-
 docs/doxygen/functions_type.html                   |   19 +-
 docs/doxygen/functions_u.html                      |    5 +-
 docs/doxygen/functions_v.html                      |   68 +-
 docs/doxygen/functions_vars.html                   |   12 +
 docs/doxygen/functions_vars_a.html                 |    3 +
 docs/doxygen/functions_vars_c.html                 |    3 +
 docs/doxygen/functions_vars_e.html                 |    3 +
 docs/doxygen/functions_vars_f.html                 |    4 +
 docs/doxygen/functions_vars_g.html                 |    3 -
 docs/doxygen/functions_vars_i.html                 |    4 +-
 docs/doxygen/functions_vars_l.html                 |    3 +
 docs/doxygen/functions_vars_m.html                 |    3 +
 docs/doxygen/functions_vars_n.html                 |    9 +-
 docs/doxygen/functions_vars_o.html                 |    4 +
 docs/doxygen/functions_vars_p.html                 |   13 +-
 docs/doxygen/functions_vars_r.html                 |    3 +
 docs/doxygen/functions_vars_t.html                 |    6 +
 docs/doxygen/functions_vars_v.html                 |    3 +
 docs/doxygen/functions_w.html                      |    3 +-
 docs/doxygen/functor_8h.html                       |   18 +-
 docs/doxygen/functor_8h__dep__incl.svg             | 1247 +++----
 docs/doxygen/functor_8h__incl.svg                  |  122 +-
 docs/doxygen/functor_8h_source.html                |    6 +-
 docs/doxygen/fuse_8h.html                          |    2 +-
 docs/doxygen/fuse_8h__incl.svg                     | 2211 ++++++------
 docs/doxygen/fuse_8h_source.html                   |    2 +-
 docs/doxygen/generic_2default_8h.html              |   10 +-
 docs/doxygen/generic_2default_8h__incl.svg         | 2171 ++++++------
 docs/doxygen/generic_2default_8h_source.html       |   14 +-
 docs/doxygen/generic_2extern_8h.html               |   12 +-
 docs/doxygen/generic_2extern_8h__incl.svg          | 2206 ++++++------
 docs/doxygen/generic_2extern_8h_source.html        |   10 +-
 docs/doxygen/generic_2injective_8h.html            |   10 +-
 docs/doxygen/generic_2injective_8h__incl.svg       | 2171 ++++++------
 docs/doxygen/generic_2injective_8h_source.html     |   12 +-
 docs/doxygen/generic__func_8h.html                 |   23 +-
 docs/doxygen/generic__func_8h__incl.svg            | 1894 +++++------
 docs/doxygen/generic__func_8h_source.html          |   26 +-
 docs/doxygen/globals.html                          |   21 +
 docs/doxygen/globals_defs.html                     |    9 +
 docs/doxygen/globals_eval.html                     |    3 +
 docs/doxygen/globals_func.html                     |    9 +
 docs/doxygen/hierarchy.html                        | 1355 ++++----
 docs/doxygen/image_8h.html                         |    2 +-
 docs/doxygen/image_8h__incl.svg                    | 1865 +++++-----
 docs/doxygen/image_8h_source.html                  |   56 +-
 docs/doxygen/include_2tvm_2ir_2transform_8h.html   |   20 +-
 .../include_2tvm_2ir_2transform_8h__incl.svg       | 2024 +++++------
 .../include_2tvm_2ir_2transform_8h_source.html     |   72 +-
 .../doxygen/include_2tvm_2relay_2attrs_2nn_8h.html |    2 +-
 .../include_2tvm_2relay_2attrs_2nn_8h__incl.svg    | 1863 +++++-----
 .../include_2tvm_2relay_2attrs_2nn_8h_source.html  |  530 +--
 .../include_2tvm_2relay_2attrs_2transform_8h.html  |    2 +-
 ...lude_2tvm_2relay_2attrs_2transform_8h__incl.svg | 2335 ++++++-------
 ...de_2tvm_2relay_2attrs_2transform_8h_source.html |  152 +-
 .../include_2tvm_2relay_2qnn_2transform_8h.html    |    6 +-
 ...nclude_2tvm_2relay_2qnn_2transform_8h__incl.svg | 2145 ++++++------
 ...lude_2tvm_2relay_2qnn_2transform_8h_source.html |    2 +-
 .../doxygen/include_2tvm_2relay_2transform_8h.html |   30 +-
 .../include_2tvm_2relay_2transform_8h__incl.svg    | 2300 ++++++-------
 .../include_2tvm_2relay_2transform_8h_source.html  |   42 +-
 docs/doxygen/include_2tvm_2tir_2transform_8h.html  |    2 +-
 .../include_2tvm_2tir_2transform_8h__incl.svg      | 2182 ++++++------
 .../include_2tvm_2tir_2transform_8h_source.html    |    4 +-
 docs/doxygen/inherit_graph_10.svg                  | 1989 ++++++-----
 docs/doxygen/inherit_graph_100.svg                 |   15 +-
 docs/doxygen/inherit_graph_101.svg                 |   18 +-
 docs/doxygen/inherit_graph_102.svg                 |   57 +-
 docs/doxygen/inherit_graph_103.svg                 |   18 +-
 docs/doxygen/inherit_graph_104.svg                 |   15 +-
 docs/doxygen/inherit_graph_105.svg                 |   15 +-
 docs/doxygen/inherit_graph_106.svg                 |   18 +-
 docs/doxygen/inherit_graph_107.svg                 |   18 +-
 docs/doxygen/inherit_graph_108.svg                 |   12 +-
 docs/doxygen/inherit_graph_109.svg                 |   15 +-
 docs/doxygen/inherit_graph_110.svg                 |   12 +-
 docs/doxygen/inherit_graph_111.svg                 |   12 +-
 docs/doxygen/inherit_graph_112.svg                 |   15 +-
 docs/doxygen/inherit_graph_113.svg                 |   15 +-
 docs/doxygen/inherit_graph_114.svg                 |   17 +-
 docs/doxygen/inherit_graph_115.svg                 |   12 +-
 docs/doxygen/inherit_graph_116.svg                 |   15 +-
 docs/doxygen/inherit_graph_117.svg                 |   15 +-
 docs/doxygen/inherit_graph_118.svg                 |   14 +-
 docs/doxygen/inherit_graph_119.svg                 |   70 +-
 docs/doxygen/inherit_graph_120.svg                 |   68 +-
 docs/doxygen/inherit_graph_121.svg                 |   15 +-
 docs/doxygen/inherit_graph_122.svg                 |   19 +-
 docs/doxygen/inherit_graph_123.svg                 |   14 +-
 docs/doxygen/inherit_graph_124.svg                 |   68 +-
 docs/doxygen/inherit_graph_125.svg                 |   69 +-
 docs/doxygen/inherit_graph_126.svg                 |   28 +-
 docs/doxygen/inherit_graph_127.svg                 |   32 +-
 docs/doxygen/inherit_graph_128.svg                 |   15 +-
 docs/doxygen/inherit_graph_129.svg                 |   12 +-
 docs/doxygen/inherit_graph_13.svg                  |   15 +-
 docs/doxygen/inherit_graph_130.svg                 |   15 +-
 docs/doxygen/inherit_graph_131.svg                 |   28 +-
 ...inherit_graph_127.svg => inherit_graph_132.svg} |    0
 ...inherit_graph_128.svg => inherit_graph_133.svg} |    0
 ...inherit_graph_129.svg => inherit_graph_134.svg} |    0
 ...inherit_graph_130.svg => inherit_graph_135.svg} |    0
 ...inherit_graph_131.svg => inherit_graph_136.svg} |    0
 docs/doxygen/inherit_graph_14.svg                  |   38 +-
 docs/doxygen/inherit_graph_15.svg                  |   28 +-
 docs/doxygen/inherit_graph_16.svg                  |   15 +-
 docs/doxygen/inherit_graph_17.svg                  |   12 +-
 docs/doxygen/inherit_graph_18.svg                  |   40 +-
 docs/doxygen/inherit_graph_19.svg                  |   25 +-
 docs/doxygen/inherit_graph_20.svg                  |   15 +-
 docs/doxygen/inherit_graph_21.svg                  |   15 +-
 docs/doxygen/inherit_graph_22.svg                  |   12 +-
 docs/doxygen/inherit_graph_23.svg                  |   12 +-
 docs/doxygen/inherit_graph_24.svg                  |   14 +-
 docs/doxygen/inherit_graph_25.svg                  |   15 +-
 docs/doxygen/inherit_graph_26.svg                  |   15 +-
 docs/doxygen/inherit_graph_27.svg                  |   15 +-
 docs/doxygen/inherit_graph_28.svg                  |   14 +-
 docs/doxygen/inherit_graph_29.svg                  |   15 +-
 docs/doxygen/inherit_graph_30.svg                  |   14 +-
 docs/doxygen/inherit_graph_31.svg                  |   54 +-
 docs/doxygen/inherit_graph_32.svg                  |   14 +-
 docs/doxygen/inherit_graph_33.svg                  |   14 +-
 docs/doxygen/inherit_graph_34.svg                  |    6 +-
 docs/doxygen/inherit_graph_35.svg                  |   54 +-
 docs/doxygen/inherit_graph_36.svg                  |   17 +-
 docs/doxygen/inherit_graph_37.svg                  |   17 +-
 docs/doxygen/inherit_graph_38.svg                  |   14 +-
 docs/doxygen/inherit_graph_39.svg                  |   14 +-
 docs/doxygen/inherit_graph_40.svg                  |   17 +-
 docs/doxygen/inherit_graph_41.svg                  |   17 +-
 docs/doxygen/inherit_graph_42.svg                  |   14 +-
 docs/doxygen/inherit_graph_43.svg                  |   14 +-
 docs/doxygen/inherit_graph_44.svg                  |    4 +-
 docs/doxygen/inherit_graph_45.svg                  |   15 +-
 docs/doxygen/inherit_graph_46.svg                  |    4 +-
 docs/doxygen/inherit_graph_47.svg                  |    4 +-
 docs/doxygen/inherit_graph_48.svg                  |   15 +-
 docs/doxygen/inherit_graph_49.svg                  |   12 +-
 docs/doxygen/inherit_graph_5.svg                   | 3557 ++++++++++----------
 docs/doxygen/inherit_graph_50.svg                  |   15 +-
 docs/doxygen/inherit_graph_51.svg                  |   14 +-
 docs/doxygen/inherit_graph_52.svg                  |   15 +-
 docs/doxygen/inherit_graph_53.svg                  |   12 +-
 docs/doxygen/inherit_graph_54.svg                  |   17 +-
 docs/doxygen/inherit_graph_55.svg                  |   15 +-
 docs/doxygen/inherit_graph_56.svg                  |   12 +-
 docs/doxygen/inherit_graph_57.svg                  |   16 +-
 docs/doxygen/inherit_graph_58.svg                  |   15 +-
 docs/doxygen/inherit_graph_59.svg                  |   41 +-
 docs/doxygen/inherit_graph_60.svg                  |   14 +-
 docs/doxygen/inherit_graph_61.svg                  |   14 +-
 docs/doxygen/inherit_graph_62.svg                  |   40 +-
 docs/doxygen/inherit_graph_63.svg                  |   28 +-
 docs/doxygen/inherit_graph_64.svg                  |   41 +-
 docs/doxygen/inherit_graph_65.svg                  |   14 +-
 docs/doxygen/inherit_graph_66.svg                  |   27 +-
 docs/doxygen/inherit_graph_67.svg                  |   40 +-
 docs/doxygen/inherit_graph_68.svg                  |   28 +-
 docs/doxygen/inherit_graph_69.svg                  |   12 +-
 docs/doxygen/inherit_graph_70.svg                  |   15 +-
 docs/doxygen/inherit_graph_71.svg                  |   28 +-
 docs/doxygen/inherit_graph_72.svg                  |   14 +-
 docs/doxygen/inherit_graph_73.svg                  |   28 +-
 docs/doxygen/inherit_graph_74.svg                  |   15 +-
 docs/doxygen/inherit_graph_75.svg                  |   28 +-
 docs/doxygen/inherit_graph_76.svg                  |  175 +-
 docs/doxygen/inherit_graph_77.svg                  |   15 +-
 docs/doxygen/inherit_graph_78.svg                  |   12 +-
 docs/doxygen/inherit_graph_79.svg                  |   15 +-
 docs/doxygen/inherit_graph_80.svg                  |   27 +-
 docs/doxygen/inherit_graph_81.svg                  |  185 +-
 docs/doxygen/inherit_graph_82.svg                  |   15 +-
 docs/doxygen/inherit_graph_83.svg                  |   15 +-
 docs/doxygen/inherit_graph_84.svg                  |   15 +-
 docs/doxygen/inherit_graph_85.svg                  |   15 +-
 docs/doxygen/inherit_graph_86.svg                  |  120 +-
 docs/doxygen/inherit_graph_87.svg                  |   14 +-
 docs/doxygen/inherit_graph_88.svg                  |   14 +-
 docs/doxygen/inherit_graph_89.svg                  |   14 +-
 docs/doxygen/inherit_graph_90.svg                  |   16 +-
 docs/doxygen/inherit_graph_91.svg                  |   16 +-
 docs/doxygen/inherit_graph_92.svg                  |   17 +-
 docs/doxygen/inherit_graph_93.svg                  |   14 +-
 docs/doxygen/inherit_graph_94.svg                  |   14 +-
 docs/doxygen/inherit_graph_95.svg                  |   16 +-
 docs/doxygen/inherit_graph_96.svg                  |   12 +-
 docs/doxygen/inherit_graph_97.svg                  |   56 +-
 docs/doxygen/inherit_graph_98.svg                  |   19 +-
 docs/doxygen/inherit_graph_99.svg                  |   14 +-
 docs/doxygen/inherits.html                         |  244 +-
 docs/doxygen/int__set_8h.html                      |    4 +-
 docs/doxygen/int__set_8h__dep__incl.svg            |  880 ++---
 docs/doxygen/int__set_8h__incl.svg                 | 1727 +++++-----
 docs/doxygen/int__set_8h_source.html               |   32 +-
 docs/doxygen/int__solver_8h.html                   |    2 +-
 docs/doxygen/int__solver_8h__incl.svg              | 1729 +++++-----
 docs/doxygen/int__solver_8h_source.html            |   48 +-
 docs/doxygen/interpreter_8h.html                   |    4 +-
 docs/doxygen/interpreter_8h__incl.svg              | 2130 ++++++------
 docs/doxygen/interpreter_8h_source.html            |   66 +-
 docs/doxygen/ir_2adt_8h.html                       |   12 +-
 docs/doxygen/ir_2adt_8h__dep__incl.svg             | 1061 +++---
 docs/doxygen/ir_2adt_8h__incl.svg                  | 1567 ++++-----
 docs/doxygen/ir_2adt_8h_source.html                |   70 +-
 docs/doxygen/ir_2attrs_8h.html                     |   14 +-
 docs/doxygen/ir_2attrs_8h__dep__incl.svg           |  937 +++---
 docs/doxygen/ir_2attrs_8h__incl.svg                | 1593 ++++-----
 docs/doxygen/ir_2attrs_8h_source.html              |  216 +-
 docs/doxygen/ir_2expr_8h.html                      |   12 +-
 docs/doxygen/ir_2expr_8h__incl.svg                 | 1505 ++++-----
 docs/doxygen/ir_2expr_8h_source.html               |  156 +-
 docs/doxygen/ir_2function_8h.html                  |   12 +-
 docs/doxygen/ir_2function_8h__dep__incl.svg        | 1028 +++---
 docs/doxygen/ir_2function_8h__incl.svg             | 1679 ++++-----
 docs/doxygen/ir_2function_8h_source.html           |   24 +-
 docs/doxygen/ir_2module_8h.html                    |   23 +-
 docs/doxygen/ir_2module_8h__dep__incl.svg          | 1060 +++---
 docs/doxygen/ir_2module_8h__incl.svg               | 1856 +++++-----
 docs/doxygen/ir_2module_8h_source.html             |   59 +-
 docs/doxygen/ir_2op_8h.html                        |   33 +-
 docs/doxygen/ir_2op_8h__dep__incl.svg              |  251 +-
 docs/doxygen/ir_2op_8h__incl.svg                   | 2042 +++++------
 docs/doxygen/ir_2op_8h_source.html                 |  114 +-
 docs/doxygen/ir_2type_8h.html                      |   12 +-
 docs/doxygen/ir_2type_8h__dep__incl.svg            | 1340 ++++----
 docs/doxygen/ir_2type_8h__incl.svg                 | 1431 ++++----
 docs/doxygen/ir_2type_8h_source.html               |  170 +-
 docs/doxygen/local__response__norm_8h.html         |    6 +-
 docs/doxygen/local__response__norm_8h__incl.svg    | 2224 ++++++------
 docs/doxygen/local__response__norm_8h_source.html  |   16 +-
 docs/doxygen/mapping_8h.html                       |    6 +-
 docs/doxygen/mapping_8h__incl.svg                  | 2220 ++++++------
 docs/doxygen/mapping_8h_source.html                |    8 +-
 docs/doxygen/namespacemembers.html                 |   19 +-
 docs/doxygen/namespacemembers_c.html               |    6 +-
 docs/doxygen/namespacemembers_f.html               |   24 +-
 docs/doxygen/namespacemembers_func.html            |   19 +-
 docs/doxygen/namespacemembers_func_c.html          |    6 +-
 docs/doxygen/namespacemembers_func_l.html          |    2 +-
 docs/doxygen/namespacemembers_func_m.html          |   11 +-
 docs/doxygen/namespacemembers_func_o.html          |   23 +-
 docs/doxygen/namespacemembers_func_p.html          |    9 +-
 docs/doxygen/namespacemembers_func_r.html          |    5 +-
 docs/doxygen/namespacemembers_l.html               |    2 +-
 docs/doxygen/namespacemembers_m.html               |   11 +-
 docs/doxygen/namespacemembers_o.html               |   42 +-
 docs/doxygen/namespacemembers_p.html               |    9 +-
 docs/doxygen/namespacemembers_r.html               |    5 +-
 docs/doxygen/namespacemembers_s.html               |    4 +-
 docs/doxygen/namespacemembers_t.html               |    9 +
 docs/doxygen/namespacemembers_type.html            |   22 +-
 docs/doxygen/namespacemembers_vars.html            |   17 +-
 docs/doxygen/namespacetopi.html                    |  684 ++--
 docs/doxygen/namespacetvm.html                     |   47 +-
 docs/doxygen/namespacetvm_1_1relay.html            |  286 +-
 .../namespacetvm_1_1relay_1_1transform.html        |   32 +-
 docs/doxygen/namespacetvm_1_1runtime.html          |   72 +
 docs/doxygen/namespacetvm_1_1te.html               |   28 +-
 docs/doxygen/namespacetvm_1_1tir_1_1attr.html      |   17 -
 docs/doxygen/namespacetvm_1_1tir_1_1intrinsic.html |   56 +-
 docs/doxygen/namespacetvm_1_1transform.html        |   18 +-
 docs/doxygen/ndarray_8h.html                       |    5 +-
 docs/doxygen/ndarray_8h__dep__incl.svg             | 1459 ++++----
 docs/doxygen/ndarray_8h__incl.svg                  |  225 +-
 docs/doxygen/ndarray_8h_source.html                |   85 +-
 docs/doxygen/nn_2bnn_8h.html                       |    8 +-
 docs/doxygen/nn_2bnn_8h__incl.svg                  | 2319 ++++++-------
 docs/doxygen/nn_2bnn_8h_source.html                |   28 +-
 docs/doxygen/nn_2dense_8h.html                     |    6 +-
 docs/doxygen/nn_2dense_8h__incl.svg                | 2220 ++++++------
 docs/doxygen/nn_2dense_8h_source.html              |   12 +-
 docs/doxygen/nn_2pooling_8h.html                   |    2 +-
 docs/doxygen/nn_2pooling_8h__incl.svg              | 2323 ++++++-------
 docs/doxygen/nn_2pooling_8h_source.html            |   72 +-
 docs/doxygen/nn_2softmax_8h.html                   |    6 +-
 docs/doxygen/nn_2softmax_8h__incl.svg              | 2326 +++++++------
 docs/doxygen/nn_2softmax_8h_source.html            |   24 +-
 docs/doxygen/node_2container_8h.html               |   14 +-
 docs/doxygen/node_2container_8h__dep__incl.svg     | 1511 +++++----
 docs/doxygen/node_2container_8h__incl.svg          |  853 ++---
 docs/doxygen/node_2container_8h_source.html        |  154 +-
 docs/doxygen/node_8h.html                          |   18 +-
 docs/doxygen/node_8h__dep__incl.svg                | 1323 ++++----
 docs/doxygen/node_8h__incl.svg                     | 1249 +++----
 docs/doxygen/node_8h_source.html                   |    6 +-
 docs/doxygen/object_8h.html                        |   32 +-
 docs/doxygen/object_8h__dep__incl.svg              | 1017 +++---
 docs/doxygen/object_8h__incl.svg                   |   54 +-
 docs/doxygen/object_8h_source.html                 |  152 +-
 docs/doxygen/op__attr__types_8h.html               |   56 +-
 docs/doxygen/op__attr__types_8h__incl.svg          | 2454 +++++++-------
 docs/doxygen/op__attr__types_8h_source.html        |   44 +-
 docs/doxygen/op__strategy_8h.html                  |    8 +-
 docs/doxygen/op__strategy_8h__incl.svg             | 2455 +++++++-------
 docs/doxygen/op__strategy_8h_source.html           |   50 +-
 docs/doxygen/operation_8h.html                     |   22 +-
 docs/doxygen/operation_8h__dep__incl.svg           |  743 ++--
 docs/doxygen/operation_8h__incl.svg                | 2215 ++++++------
 docs/doxygen/operation_8h_source.html              |  138 +-
 docs/doxygen/packed__func_8h.html                  |   43 +-
 docs/doxygen/packed__func_8h__dep__incl.svg        | 1428 ++++----
 docs/doxygen/packed__func_8h__incl.svg             |  557 +--
 docs/doxygen/packed__func_8h_source.html           |  231 +-
 docs/doxygen/pad__utils_8h.html                    |    9 +-
 docs/doxygen/pad__utils_8h__incl.svg               | 2062 +++++++-----
 docs/doxygen/pad__utils_8h_source.html             |    7 +-
 docs/doxygen/pattern_8h.html                       |    6 +-
 docs/doxygen/pattern_8h__incl.svg                  | 1765 +++++-----
 docs/doxygen/pattern_8h_source.html                |    2 +-
 docs/doxygen/pattern__functor_8h.html              |   12 +-
 docs/doxygen/pattern__functor_8h__incl.svg         | 2314 ++++++-------
 docs/doxygen/pattern__functor_8h_source.html       |   34 +-
 docs/doxygen/ravel__unravel_8h.html                |    2 +-
 docs/doxygen/ravel__unravel_8h__incl.svg           | 2219 ++++++------
 docs/doxygen/ravel__unravel_8h_source.html         |   10 +-
 docs/doxygen/reduce_8h.html                        |    2 +-
 docs/doxygen/reduce_8h__incl.svg                   | 1615 ++++-----
 docs/doxygen/reduce_8h_source.html                 |   18 +-
 docs/doxygen/reduction_8h.html                     |   18 +-
 docs/doxygen/reduction_8h__incl.svg                | 2131 ++++++------
 docs/doxygen/reduction_8h_source.html              |   70 +-
 docs/doxygen/reflection_8h.html                    |   38 +-
 docs/doxygen/reflection_8h__dep__incl.svg          | 1585 +++++----
 docs/doxygen/reflection_8h__incl.svg               | 1137 +++----
 docs/doxygen/reflection_8h_source.html             |   56 +-
 docs/doxygen/registry_8h.html                      |   19 +-
 docs/doxygen/registry_8h__dep__incl.svg            |  426 ++-
 docs/doxygen/registry_8h__incl.svg                 |  603 ++--
 docs/doxygen/registry_8h_source.html               |   26 +-
 docs/doxygen/relay_2adt_8h.html                    |    8 +-
 docs/doxygen/relay_2adt_8h__incl.svg               | 2328 ++++++-------
 docs/doxygen/relay_2adt_8h_source.html             |  148 +-
 docs/doxygen/relay_2analysis_8h.html               |    6 +-
 docs/doxygen/relay_2analysis_8h__incl.svg          | 2401 +++++++------
 docs/doxygen/relay_2analysis_8h_source.html        |    6 +-
 docs/doxygen/relay_2attrs_2memory_8h.html          |    2 +-
 docs/doxygen/relay_2attrs_2memory_8h__incl.svg     | 2218 ++++++------
 docs/doxygen/relay_2attrs_2memory_8h_source.html   |   42 +-
 docs/doxygen/relay_2expr_8h.html                   |    8 +-
 docs/doxygen/relay_2expr_8h__dep__incl.svg         |  337 +-
 docs/doxygen/relay_2expr_8h__incl.svg              | 2291 ++++++-------
 docs/doxygen/relay_2expr_8h_source.html            |  208 +-
 docs/doxygen/relay_2expr__functor_8h.html          |   14 +-
 docs/doxygen/relay_2expr__functor_8h__incl.svg     | 2328 ++++++-------
 docs/doxygen/relay_2expr__functor_8h_source.html   |  122 +-
 docs/doxygen/relay_2function_8h.html               |    2 +-
 docs/doxygen/relay_2function_8h__incl.svg          | 2286 ++++++-------
 docs/doxygen/relay_2function_8h_source.html        |   52 +-
 docs/doxygen/relay_2op_8h.html                     |    4 +-
 docs/doxygen/relay_2op_8h__incl.svg                | 2389 ++++++-------
 docs/doxygen/relay_2op_8h_source.html              |    6 +-
 docs/doxygen/relay_2qnn_2attrs_8h.html             |    2 +-
 docs/doxygen/relay_2qnn_2attrs_8h__incl.svg        | 1615 ++++-----
 docs/doxygen/relay_2qnn_2attrs_8h_source.html      |   26 +-
 docs/doxygen/relay_2type_8h.html                   |   10 +-
 docs/doxygen/relay_2type_8h__dep__incl.svg         |  436 +--
 docs/doxygen/relay_2type_8h__incl.svg              | 2254 +++++++------
 docs/doxygen/relay_2type_8h_source.html            |   50 +-
 docs/doxygen/reorg_8h.html                         |    6 +-
 docs/doxygen/reorg_8h__incl.svg                    | 2073 ++++++------
 docs/doxygen/reorg_8h_source.html                  |   10 +-
 docs/doxygen/repr__printer_8h.html                 |    4 +-
 docs/doxygen/repr__printer_8h__dep__incl.svg       | 1293 +++----
 docs/doxygen/repr__printer_8h__incl.svg            |  140 +-
 docs/doxygen/repr__printer_8h_source.html          |   14 +-
 docs/doxygen/rocblas_8h.html                       |    2 +-
 docs/doxygen/rocblas_8h__incl.svg                  | 2216 ++++++------
 docs/doxygen/rocblas_8h_source.html                |   10 +-
 docs/doxygen/rocm_2dense_8h.html                   |   12 +-
 docs/doxygen/rocm_2dense_8h__incl.svg              | 2210 ++++++------
 docs/doxygen/rocm_2dense_8h_source.html            |   22 +-
 docs/doxygen/rocm_2injective_8h.html               |    8 +-
 docs/doxygen/rocm_2injective_8h__incl.svg          | 2114 ++++++------
 docs/doxygen/rocm_2injective_8h_source.html        |    6 +-
 docs/doxygen/rocm_2normalization_8h.html           |    6 +-
 docs/doxygen/rocm_2normalization_8h__incl.svg      | 2308 ++++++-------
 docs/doxygen/rocm_2normalization_8h_source.html    |    6 +-
 docs/doxygen/rocm_2pooling_8h.html                 |   12 +-
 docs/doxygen/rocm_2pooling_8h__incl.svg            | 2187 ++++++------
 docs/doxygen/rocm_2pooling_8h_source.html          |    4 +-
 docs/doxygen/rocm_2reduction_8h.html               |    8 +-
 docs/doxygen/rocm_2reduction_8h__incl.svg          | 2114 ++++++------
 docs/doxygen/rocm_2reduction_8h_source.html        |    6 +-
 docs/doxygen/rocm_2softmax_8h.html                 |    8 +-
 docs/doxygen/rocm_2softmax_8h__incl.svg            | 2114 ++++++------
 docs/doxygen/rocm_2softmax_8h_source.html          |    4 +-
 docs/doxygen/runtime_2container_8h.html            |   12 +-
 docs/doxygen/runtime_2container_8h__dep__incl.svg  |  883 ++---
 docs/doxygen/runtime_2container_8h__incl.svg       |  555 +--
 docs/doxygen/runtime_2container_8h_source.html     |  158 +-
 docs/doxygen/runtime_2crt_2memory_8h_source.html   |    2 +-
 docs/doxygen/runtime_2memory_8h.html               |    6 +-
 docs/doxygen/runtime_2memory_8h__dep__incl.svg     | 1265 +++----
 docs/doxygen/runtime_2memory_8h__incl.svg          |  122 +-
 docs/doxygen/runtime_2memory_8h_source.html        |   34 +-
 docs/doxygen/runtime_2module_8h.html               |    8 +-
 docs/doxygen/runtime_2module_8h__dep__incl.svg     | 1426 ++++----
 docs/doxygen/runtime_2module_8h__incl.svg          |  499 +--
 docs/doxygen/runtime_2module_8h_source.html        |   50 +-
 docs/doxygen/schedule_8h.html                      |    8 +-
 docs/doxygen/schedule_8h__dep__incl.svg            | 1033 +++---
 docs/doxygen/schedule_8h__incl.svg                 | 2143 ++++++------
 docs/doxygen/schedule_8h_source.html               |  163 +-
 docs/doxygen/schedule__pass_8h.html                |    2 +-
 docs/doxygen/schedule__pass_8h__incl.svg           | 2127 ++++++------
 docs/doxygen/schedule__pass_8h_source.html         |    2 +-
 docs/doxygen/search/all_0.js                       |   11 +-
 docs/doxygen/search/all_1.js                       |  286 +-
 docs/doxygen/search/all_10.js                      |  261 +-
 docs/doxygen/search/all_11.js                      |  144 +-
 docs/doxygen/search/all_12.js                      |  151 +-
 docs/doxygen/search/all_13.js                      |  379 ++-
 docs/doxygen/search/all_14.js                      |  615 ++--
 docs/doxygen/search/all_15.js                      |  395 +--
 docs/doxygen/search/all_16.js                      |  115 +-
 docs/doxygen/search/all_17.js                      |   95 +-
 docs/doxygen/search/all_18.js                      |   17 +-
 docs/doxygen/search/all_19.js                      |    3 +-
 docs/doxygen/search/all_1a.js                      |   25 +-
 docs/doxygen/search/all_1b.html                    |   26 -
 docs/doxygen/search/all_1b.js                      |   25 -
 docs/doxygen/search/all_2.js                       |  361 +-
 docs/doxygen/search/all_3.js                       |  290 +-
 docs/doxygen/search/all_4.js                       |  313 +-
 docs/doxygen/search/all_5.js                       |  197 +-
 docs/doxygen/search/all_6.js                       |  234 +-
 docs/doxygen/search/all_7.js                       |  219 +-
 docs/doxygen/search/all_8.js                       |   88 +-
 docs/doxygen/search/all_9.js                       |  195 +-
 docs/doxygen/search/all_a.js                       |  180 +-
 docs/doxygen/search/all_b.js                       |  140 +-
 docs/doxygen/search/all_c.js                       |  209 +-
 docs/doxygen/search/all_d.js                       |  172 +-
 docs/doxygen/search/all_e.js                       |  147 +-
 docs/doxygen/search/all_f.js                       |  162 +-
 docs/doxygen/search/classes_0.js                   |   10 +
 docs/doxygen/search/classes_11.js                  |    6 +
 docs/doxygen/search/classes_13.js                  |    6 +-
 docs/doxygen/search/classes_14.js                  |    2 +
 docs/doxygen/search/classes_2.js                   |    4 +-
 docs/doxygen/search/classes_3.js                   |   10 +
 docs/doxygen/search/classes_4.js                   |    4 +-
 docs/doxygen/search/classes_6.js                   |    1 -
 docs/doxygen/search/classes_7.js                   |    2 +-
 docs/doxygen/search/classes_9.js                   |    2 +-
 docs/doxygen/search/classes_c.js                   |    4 +-
 docs/doxygen/search/defines_0.js                   |    3 +-
 docs/doxygen/search/defines_4.js                   |    1 +
 docs/doxygen/search/defines_6.js                   |    1 +
 docs/doxygen/search/enumvalues_6.js                |    1 +
 docs/doxygen/search/files_0.js                     |    3 +-
 docs/doxygen/search/files_3.js                     |    7 +-
 docs/doxygen/search/functions_1.js                 |   25 +-
 docs/doxygen/search/functions_10.js                |    9 +-
 docs/doxygen/search/functions_11.js                |    6 +-
 docs/doxygen/search/functions_12.js                |   14 +-
 docs/doxygen/search/functions_13.js                |   12 +-
 docs/doxygen/search/functions_15.js                |    9 +-
 docs/doxygen/search/functions_19.js                |    1 +
 docs/doxygen/search/functions_2.js                 |    2 +-
 docs/doxygen/search/functions_3.js                 |   15 +-
 docs/doxygen/search/functions_4.js                 |    6 +-
 docs/doxygen/search/functions_5.js                 |    1 +
 docs/doxygen/search/functions_6.js                 |    4 +-
 docs/doxygen/search/functions_7.js                 |   15 +-
 docs/doxygen/search/functions_8.js                 |    2 +-
 docs/doxygen/search/functions_9.js                 |   11 +-
 docs/doxygen/search/functions_c.js                 |    6 +-
 docs/doxygen/search/functions_d.js                 |    3 +-
 docs/doxygen/search/functions_e.js                 |    2 +-
 docs/doxygen/search/functions_f.js                 |   11 +-
 docs/doxygen/search/related_0.js                   |    5 +-
 docs/doxygen/search/related_1.js                   |    4 +-
 docs/doxygen/search/related_2.js                   |    3 +-
 docs/doxygen/search/related_3.js                   |    2 +-
 docs/doxygen/search/related_4.js                   |    3 +-
 docs/doxygen/search/related_5.js                   |    6 +-
 docs/doxygen/search/related_6.js                   |    7 +-
 docs/doxygen/search/related_7.js                   |    5 +-
 docs/doxygen/search/related_8.js                   |   10 +-
 docs/doxygen/search/related_9.js                   |   11 +-
 docs/doxygen/search/related_a.js                   |    3 +-
 docs/doxygen/search/related_b.js                   |    6 +-
 docs/doxygen/search/related_c.js                   |   10 +-
 docs/doxygen/search/related_d.html                 |   26 -
 docs/doxygen/search/related_d.js                   |    8 -
 docs/doxygen/search/searchdata.js                  |    4 +-
 docs/doxygen/search/typedefs_1.js                  |    3 +-
 docs/doxygen/search/typedefs_2.js                  |    2 +-
 docs/doxygen/search/typedefs_5.js                  |   24 +-
 docs/doxygen/search/typedefs_7.js                  |    2 +-
 docs/doxygen/search/typedefs_b.js                  |    2 +-
 docs/doxygen/search/typedefs_d.js                  |    1 +
 docs/doxygen/search/variables_0.js                 |    2 +-
 docs/doxygen/search/variables_1.js                 |    4 +-
 docs/doxygen/search/variables_10.js                |    1 +
 docs/doxygen/search/variables_12.js                |   10 +-
 docs/doxygen/search/variables_14.js                |    1 +
 docs/doxygen/search/variables_3.js                 |    1 +
 docs/doxygen/search/variables_5.js                 |    1 +
 docs/doxygen/search/variables_6.js                 |    3 +-
 docs/doxygen/search/variables_7.js                 |    1 -
 docs/doxygen/search/variables_9.js                 |    3 +-
 docs/doxygen/search/variables_b.js                 |    1 +
 docs/doxygen/search/variables_c.js                 |    1 +
 docs/doxygen/search/variables_d.js                 |    6 +-
 docs/doxygen/search/variables_e.js                 |    4 +-
 docs/doxygen/search/variables_f.js                 |    5 +-
 docs/doxygen/serialization_8h.html                 |    2 +-
 docs/doxygen/serialization_8h__incl.svg            |  110 +-
 docs/doxygen/serializer_8h.html                    |    4 +-
 docs/doxygen/serializer_8h__dep__incl.svg          | 1457 ++++----
 docs/doxygen/serializer_8h__incl.svg               |  179 +-
 docs/doxygen/serializer_8h_source.html             |    2 +-
 docs/doxygen/span_8h.html                          |    6 +-
 docs/doxygen/span_8h__incl.svg                     | 1315 ++++----
 docs/doxygen/span_8h_source.html                   |   34 +-
 docs/doxygen/stmt_8h.html                          |   11 +-
 docs/doxygen/stmt_8h__dep__incl.svg                |  985 +++---
 docs/doxygen/stmt_8h__incl.svg                     | 1785 +++++-----
 docs/doxygen/stmt_8h_source.html                   |  361 +-
 docs/doxygen/stmt__functor_8h.html                 |   18 +-
 docs/doxygen/stmt__functor_8h__incl.svg            | 1807 +++++-----
 docs/doxygen/stmt__functor_8h_source.html          |   86 +-
 .../structtvm_1_1Map_1_1ValueConverter.html        |    4 +-
 .../structtvm_1_1relay_1_1AllocStorageAttrs.html   |    2 +-
 ..._1_1relay_1_1AllocStorageAttrs__coll__graph.svg |  133 +-
 .../structtvm_1_1relay_1_1AllocTensorAttrs.html    |    2 +-
 ...m_1_1relay_1_1AllocTensorAttrs__coll__graph.svg |  233 +-
 .../doxygen/structtvm_1_1relay_1_1ArangeAttrs.html |    2 +-
 ...ucttvm_1_1relay_1_1ArangeAttrs__coll__graph.svg |  185 +-
 .../structtvm_1_1relay_1_1ArgsortAttrs.html        |    2 +-
 ...cttvm_1_1relay_1_1ArgsortAttrs__coll__graph.svg |  131 +-
 .../structtvm_1_1relay_1_1BinaryConv2DAttrs.html   |    2 +-
 ..._1_1relay_1_1BinaryConv2DAttrs__coll__graph.svg |  225 +-
 .../structtvm_1_1relay_1_1BinaryDenseAttrs.html    |    2 +-
 ...m_1_1relay_1_1BinaryDenseAttrs__coll__graph.svg |  195 +-
 .../structtvm_1_1relay_1_1BitPackAttrs.html        |    2 +-
 ...cttvm_1_1relay_1_1BitPackAttrs__coll__graph.svg |  133 +-
 docs/doxygen/structtvm_1_1relay_1_1CastAttrs.html  |    2 +-
 ...tructtvm_1_1relay_1_1CastAttrs__coll__graph.svg |  133 +-
 .../structtvm_1_1relay_1_1CastHintAttrs.html       |    2 +-
 ...ttvm_1_1relay_1_1CastHintAttrs__coll__graph.svg |  135 +-
 .../doxygen/structtvm_1_1relay_1_1Conv1DAttrs.html |    2 +-
 ...ucttvm_1_1relay_1_1Conv1DAttrs__coll__graph.svg |  225 +-
 ...structtvm_1_1relay_1_1Conv1DTransposeAttrs.html |    2 +-
 ...1relay_1_1Conv1DTransposeAttrs__coll__graph.svg |  225 +-
 .../doxygen/structtvm_1_1relay_1_1Conv2DAttrs.html |    2 +-
 ...ucttvm_1_1relay_1_1Conv2DAttrs__coll__graph.svg |  225 +-
 ...structtvm_1_1relay_1_1Conv2DTransposeAttrs.html |    2 +-
 ...1relay_1_1Conv2DTransposeAttrs__coll__graph.svg |  225 +-
 .../structtvm_1_1relay_1_1Conv2DWinogradAttrs.html |    2 +-
 ..._1relay_1_1Conv2DWinogradAttrs__coll__graph.svg |  225 +-
 ..._1Conv2DWinogradNNPACKWeightTransformAttrs.html |    2 +-
 ...gradNNPACKWeightTransformAttrs__coll__graph.svg |  133 +-
 .../doxygen/structtvm_1_1relay_1_1Conv3DAttrs.html |    2 +-
 ...ucttvm_1_1relay_1_1Conv3DAttrs__coll__graph.svg |  225 +-
 .../structtvm_1_1relay_1_1Conv3DWinogradAttrs.html |    2 +-
 ..._1relay_1_1Conv3DWinogradAttrs__coll__graph.svg |  225 +-
 .../structtvm_1_1relay_1_1CropAndResizeAttrs.html  |    2 +-
 ...1_1relay_1_1CropAndResizeAttrs__coll__graph.svg |  183 +-
 ...tructtvm_1_1relay_1_1DeformableConv2DAttrs.html |    2 +-
 ...relay_1_1DeformableConv2DAttrs__coll__graph.svg |  225 +-
 docs/doxygen/structtvm_1_1relay_1_1DenseAttrs.html |    2 +-
 ...ructtvm_1_1relay_1_1DenseAttrs__coll__graph.svg |  189 +-
 .../structtvm_1_1relay_1_1Dilation2DAttrs.html     |    2 +-
 ...vm_1_1relay_1_1Dilation2DAttrs__coll__graph.svg |  185 +-
 .../doxygen/structtvm_1_1relay_1_1InitOpAttrs.html |    2 +-
 ...ucttvm_1_1relay_1_1InitOpAttrs__coll__graph.svg |  183 +-
 .../structtvm_1_1relay_1_1NdarraySizeAttrs.html    |    2 +-
 ...m_1_1relay_1_1NdarraySizeAttrs__coll__graph.svg |  135 +-
 .../doxygen/structtvm_1_1relay_1_1OneHotAttrs.html |    2 +-
 ...ucttvm_1_1relay_1_1OneHotAttrs__coll__graph.svg |  131 +-
 ...structtvm_1_1relay_1_1ReshapeAttrs-members.html |    2 +-
 .../structtvm_1_1relay_1_1ReshapeAttrs.html        |   10 +-
 ...cttvm_1_1relay_1_1ReshapeAttrs__coll__graph.svg |  146 +-
 .../doxygen/structtvm_1_1relay_1_1ResizeAttrs.html |    2 +-
 ...ucttvm_1_1relay_1_1ResizeAttrs__coll__graph.svg |  183 +-
 .../structtvm_1_1relay_1_1ShapeOfAttrs.html        |    2 +-
 ...cttvm_1_1relay_1_1ShapeOfAttrs__coll__graph.svg |  133 +-
 docs/doxygen/structtvm_1_1relay_1_1TopKAttrs.html  |    2 +-
 ...tructtvm_1_1relay_1_1TopKAttrs__coll__graph.svg |  133 +-
 ...structtvm_1_1relay_1_1qnn_1_1QuantizeAttrs.html |    2 +-
 ...1relay_1_1qnn_1_1QuantizeAttrs__coll__graph.svg |  133 +-
 ...ructtvm_1_1relay_1_1qnn_1_1RequantizeAttrs.html |    2 +-
 ...elay_1_1qnn_1_1RequantizeAttrs__coll__graph.svg |  133 +-
 ...vm_1_1runtime_1_1vm_1_1Instruction-members.html |   39 +-
 .../structtvm_1_1runtime_1_1vm_1_1Instruction.html |   50 +-
 ..._1runtime_1_1vm_1_1Instruction__coll__graph.svg |   16 +-
 ...vm_1_1runtime_1_1vm_1_1VMFrame__coll__graph.svg |   16 +-
 docs/doxygen/structtvm_1_1tir_1_1LENode.html       |    2 +-
 .../structtvm_1_1tir_1_1LENode__coll__graph.svg    |  127 +-
 docs/doxygen/structural__equal_8h.html             |    8 +-
 docs/doxygen/structural__equal_8h__dep__incl.svg   | 1285 ++++---
 docs/doxygen/structural__equal_8h__incl.svg        |  933 ++---
 docs/doxygen/structural__equal_8h_source.html      |   44 +-
 docs/doxygen/structural__hash_8h.html              |   10 +-
 docs/doxygen/structural__hash_8h__dep__incl.svg    | 1285 ++++---
 docs/doxygen/structural__hash_8h__incl.svg         |  941 +++---
 docs/doxygen/structural__hash_8h_source.html       |   46 +-
 docs/doxygen/tags_8h_source.html                   |    4 +-
 docs/doxygen/target_8h.html                        |   10 +-
 docs/doxygen/target_8h__incl.svg                   | 2078 ++++++------
 docs/doxygen/target_8h_source.html                 |   20 +-
 docs/doxygen/target__info_8h.html                  |    2 +-
 docs/doxygen/target__info_8h__incl.svg             | 1521 ++++-----
 docs/doxygen/target__info_8h_source.html           |   20 +-
 docs/doxygen/tensor_8h.html                        |   21 +-
 docs/doxygen/tensor_8h__dep__incl.svg              | 1108 +++---
 docs/doxygen/tensor_8h__incl.svg                   | 2019 +++++------
 docs/doxygen/tensor_8h_source.html                 |   54 +-
 docs/doxygen/tensor__intrin_8h.html                |    4 +-
 docs/doxygen/tensor__intrin_8h__dep__incl.svg      | 1052 +++---
 docs/doxygen/tensor__intrin_8h__incl.svg           | 2055 +++++------
 docs/doxygen/tensor__intrin_8h_source.html         |   30 +-
 docs/doxygen/tensor__type_8h.html                  |    8 +-
 docs/doxygen/tensor__type_8h__dep__incl.svg        |  446 +--
 docs/doxygen/tensor__type_8h__incl.svg             | 1529 ++++-----
 docs/doxygen/tensor__type_8h_source.html           |   30 +-
 docs/doxygen/tensor__utils_8h.html                 |    2 +-
 docs/doxygen/tensor__utils_8h__incl.svg            | 2211 ++++++------
 docs/doxygen/tensor__utils_8h_source.html          |   10 +-
 docs/doxygen/threading__backend_8h_source.html     |    8 +-
 docs/doxygen/tir_2analysis_8h.html                 |    2 +-
 docs/doxygen/tir_2analysis_8h__incl.svg            | 2125 ++++++------
 docs/doxygen/tir_2analysis_8h_source.html          |   18 +-
 docs/doxygen/tir_2expr_8h.html                     |   22 +-
 docs/doxygen/tir_2expr_8h__dep__incl.svg           |  934 +++--
 docs/doxygen/tir_2expr_8h__incl.svg                | 1749 +++++-----
 docs/doxygen/tir_2expr_8h_source.html              |  449 +--
 docs/doxygen/tir_2expr__functor_8h.html            |   10 +-
 docs/doxygen/tir_2expr__functor_8h__incl.svg       | 1585 ++++-----
 docs/doxygen/tir_2expr__functor_8h_source.html     |   86 +-
 docs/doxygen/tir_2function_8h.html                 |    4 +-
 docs/doxygen/tir_2function_8h__incl.svg            | 1993 +++++------
 docs/doxygen/tir_2function_8h_source.html          |   48 +-
 docs/doxygen/tir_2op_8h.html                       |   16 +-
 docs/doxygen/tir_2op_8h__dep__incl.svg             |  885 ++---
 docs/doxygen/tir_2op_8h__incl.svg                  | 1777 +++++-----
 docs/doxygen/tir_2op_8h_source.html                |  128 +-
 docs/doxygen/topi_2include_2topi_2nn_8h.html       |    8 +-
 docs/doxygen/topi_2include_2topi_2nn_8h__incl.svg  | 2395 ++++++-------
 .../doxygen/topi_2include_2topi_2nn_8h_source.html |   40 +-
 .../doxygen/topi_2include_2topi_2transform_8h.html |   16 +-
 .../topi_2include_2topi_2transform_8h__incl.svg    | 2272 ++++++-------
 .../topi_2include_2topi_2transform_8h_source.html  |   94 +-
 docs/doxygen/topi_2include_2topi_2util_8h.html     |    2 +-
 .../doxygen/topi_2include_2topi_2util_8h__incl.svg | 1519 ++++-----
 .../topi_2include_2topi_2util_8h_source.html       |   10 +-
 docs/doxygen/type__functor_8h.html                 |    8 +-
 docs/doxygen/type__functor_8h__incl.svg            | 2184 ++++++------
 docs/doxygen/type__functor_8h_source.html          |   50 +-
 docs/doxygen/type__relation_8h.html                |   10 +-
 docs/doxygen/type__relation_8h__dep__incl.svg      |  398 ++-
 docs/doxygen/type__relation_8h__incl.svg           | 1952 +++++------
 docs/doxygen/type__relation_8h_source.html         |   62 +-
 docs/doxygen/var_8h.html                           |    8 +-
 docs/doxygen/var_8h__dep__incl.svg                 | 1005 +++---
 docs/doxygen/var_8h__incl.svg                      | 1567 ++++-----
 docs/doxygen/var_8h_source.html                    |  108 +-
 docs/doxygen/vision_8h.html                        |    2 +-
 docs/doxygen/vision_8h__incl.svg                   | 1865 +++++-----
 docs/doxygen/vision_8h_source.html                 |  112 +-
 docs/doxygen/vm_8h.html                            |    6 +-
 docs/doxygen/vm_8h__incl.svg                       |  691 ++--
 docs/doxygen/vm_8h_source.html                     |  179 +-
 docs/doxygen/with_8h.html                          |    2 +-
 docs/doxygen/with_8h__dep__incl.svg                |  965 +++---
 docs/doxygen/with_8h_source.html                   |    8 +-
 docs/doxygen/x86_2bnn_8h.html                      |    8 +-
 docs/doxygen/x86_2bnn_8h__incl.svg                 | 2203 ++++++------
 docs/doxygen/x86_2bnn_8h_source.html               |   20 +-
 docs/doxygen/x86_2default_8h.html                  |    8 +-
 docs/doxygen/x86_2default_8h__incl.svg             | 2155 ++++++------
 docs/doxygen/x86_2default_8h_source.html           |   14 +-
 docs/doxygen/x86_2injective_8h.html                |    8 +-
 docs/doxygen/x86_2injective_8h__incl.svg           | 2203 ++++++------
 docs/doxygen/x86_2injective_8h_source.html         |   10 +-
 docs/faq.html                                      |    2 +-
 docs/frontend/tensorflow.html                      |    2 +-
 docs/genindex.html                                 |  102 +-
 docs/index.html                                    |    4 +-
 docs/install/docker.html                           |    2 +-
 docs/install/from_source.html                      |    2 +-
 docs/install/index.html                            |    2 +-
 docs/install/nnpack.html                           |    2 +-
 docs/jsdoc/fonts/OpenSans-Bold-webfont.eot         |  Bin 19544 -> 0 bytes
 docs/jsdoc/fonts/OpenSans-Bold-webfont.svg         | 1830 ----------
 docs/jsdoc/fonts/OpenSans-Bold-webfont.woff        |  Bin 22432 -> 0 bytes
 docs/jsdoc/fonts/OpenSans-BoldItalic-webfont.eot   |  Bin 20133 -> 0 bytes
 docs/jsdoc/fonts/OpenSans-BoldItalic-webfont.svg   | 1830 ----------
 docs/jsdoc/fonts/OpenSans-BoldItalic-webfont.woff  |  Bin 23048 -> 0 bytes
 docs/jsdoc/fonts/OpenSans-Italic-webfont.eot       |  Bin 20265 -> 0 bytes
 docs/jsdoc/fonts/OpenSans-Italic-webfont.svg       | 1830 ----------
 docs/jsdoc/fonts/OpenSans-Italic-webfont.woff      |  Bin 23188 -> 0 bytes
 docs/jsdoc/fonts/OpenSans-Light-webfont.eot        |  Bin 19514 -> 0 bytes
 docs/jsdoc/fonts/OpenSans-Light-webfont.svg        | 1831 ----------
 docs/jsdoc/fonts/OpenSans-Light-webfont.woff       |  Bin 22248 -> 0 bytes
 docs/jsdoc/fonts/OpenSans-LightItalic-webfont.eot  |  Bin 20535 -> 0 bytes
 docs/jsdoc/fonts/OpenSans-LightItalic-webfont.svg  | 1835 ----------
 docs/jsdoc/fonts/OpenSans-LightItalic-webfont.woff |  Bin 23400 -> 0 bytes
 docs/jsdoc/fonts/OpenSans-Regular-webfont.eot      |  Bin 19836 -> 0 bytes
 docs/jsdoc/fonts/OpenSans-Regular-webfont.svg      | 1831 ----------
 docs/jsdoc/fonts/OpenSans-Regular-webfont.woff     |  Bin 22660 -> 0 bytes
 docs/jsdoc/index.html                              |  209 --
 docs/jsdoc/scripts/linenumber.js                   |   25 -
 docs/jsdoc/scripts/prettify/Apache-License-2.0.txt |  202 --
 docs/jsdoc/scripts/prettify/lang-css.js            |    2 -
 docs/jsdoc/scripts/prettify/prettify.js            |   28 -
 docs/jsdoc/styles/jsdoc-default.css                |  358 --
 docs/jsdoc/styles/prettify-jsdoc.css               |  111 -
 docs/jsdoc/styles/prettify-tomorrow.css            |  132 -
 docs/jsdoc/tvm.GraphModule.html                    |  857 -----
 docs/jsdoc/tvm.NDArray.html                        |  760 -----
 docs/jsdoc/tvm.TVMConstant.html                    |  173 -
 docs/jsdoc/tvm.TVMContext.html                     |  171 -
 docs/jsdoc/tvm.TVMModule.html                      |  559 ---
 docs/jsdoc/tvm.TVMRuntime.html                     | 2288 -------------
 docs/jsdoc/tvm.html                                |  156 -
 docs/jsdoc/tvm_runtime.html                        |  343 --
 docs/jsdoc/tvm_runtime.js.html                     | 1325 --------
 docs/langref/hybrid_script.html                    |    6 +-
 docs/langref/index.html                            |    8 +-
 docs/langref/relay_adt.html                        |    3 +-
 docs/langref/relay_expr.html                       |    3 +-
 docs/langref/relay_op.html                         |    7 +-
 docs/langref/relay_pattern.html                    |  368 ++
 docs/langref/relay_type.html                       |    3 +-
 docs/objects.inv                                   |  Bin 14168 -> 14152 bytes
 docs/py-modindex.html                              |    4 +-
 docs/search.html                                   |    2 +-
 docs/searchindex.js                                |    2 +-
 docs/tutorials/autotvm/sg_execution_times.html     |   16 +-
 docs/tutorials/autotvm/tune_conv2d_cuda.html       |   44 +-
 docs/tutorials/autotvm/tune_relay_arm.html         |    2 +-
 docs/tutorials/autotvm/tune_relay_cuda.html        |    2 +-
 docs/tutorials/autotvm/tune_relay_mobile_gpu.html  |    2 +-
 docs/tutorials/autotvm/tune_relay_x86.html         |    2 +-
 docs/tutorials/autotvm/tune_simple_template.html   |   22 +-
 docs/tutorials/cross_compilation_and_rpc.html      |    4 +-
 docs/tutorials/dev/low_level_custom_pass.html      |   28 +-
 docs/tutorials/dev/relay_pass_infra.html           |    2 +-
 docs/tutorials/dev/sg_execution_times.html         |    8 +-
 docs/tutorials/frontend/build_gcn.html             |    2 +-
 .../frontend/deploy_model_on_android.html          |    4 +-
 docs/tutorials/frontend/deploy_model_on_rasp.html  |    2 +-
 docs/tutorials/frontend/deploy_prequantized.html   |    4 +-
 docs/tutorials/frontend/deploy_quantized.html      |    2 +-
 docs/tutorials/frontend/deploy_ssd_gluoncv.html    |   82 +-
 docs/tutorials/frontend/from_caffe2.html           |    2 +-
 docs/tutorials/frontend/from_coreml.html           |    2 +-
 docs/tutorials/frontend/from_darknet.html          |    2 +-
 docs/tutorials/frontend/from_keras.html            |    2 +-
 docs/tutorials/frontend/from_mxnet.html            |    2 +-
 docs/tutorials/frontend/from_onnx.html             |   10 +-
 docs/tutorials/frontend/from_pytorch.html          |    2 +-
 docs/tutorials/frontend/from_tensorflow.html       |    4 +-
 docs/tutorials/frontend/from_tflite.html           |   18 +-
 docs/tutorials/frontend/sg_execution_times.html    |   36 +-
 docs/tutorials/frontend/using_external_lib.html    |    2 +-
 docs/tutorials/index.html                          |    2 +-
 docs/tutorials/language/extern_op.html             |    2 +-
 docs/tutorials/language/intrin_math.html           |    2 +-
 docs/tutorials/language/reduction.html             |  137 +-
 docs/tutorials/language/scan.html                  |   92 +-
 docs/tutorials/language/schedule_primitives.html   |  191 +-
 docs/tutorials/language/sg_execution_times.html    |   20 +-
 docs/tutorials/language/tedd.html                  |    2 +-
 docs/tutorials/language/tensorize.html             |  163 +-
 docs/tutorials/language/tuple_inputs.html          |   71 +-
 docs/tutorials/optimize/opt_conv_cuda.html         |    4 +-
 docs/tutorials/optimize/opt_conv_tensorcore.html   |  234 +-
 docs/tutorials/optimize/opt_gemm.html              |  268 +-
 .../optimize/opt_matmul_auto_tensorcore.html       |    2 +-
 docs/tutorials/optimize/sg_execution_times.html    |   12 +-
 docs/tutorials/relay_quick_start.html              |  106 +-
 docs/tutorials/sg_execution_times.html             |   10 +-
 docs/tutorials/tensor_expr_get_started.html        |    2 +-
 docs/tutorials/topi/intro_topi.html                |  617 ++--
 docs/tutorials/topi/sg_execution_times.html        |    6 +-
 docs/typedoc/assets/css/main.css                   | 2679 +++++++++++++++
 docs/typedoc/assets/images/icons.png               |  Bin 0 -> 9615 bytes
 docs/typedoc/assets/images/icons@2x.png            |  Bin 0 -> 28144 bytes
 docs/typedoc/assets/images/widgets.png             |  Bin 0 -> 480 bytes
 docs/typedoc/assets/images/widgets@2x.png          |  Bin 0 -> 855 bytes
 docs/typedoc/assets/js/main.js                     |    1 +
 docs/typedoc/assets/js/search.json                 |    1 +
 docs/typedoc/classes/bytestreamreader.html         |  278 ++
 docs/typedoc/classes/cachedcallstack.html          |  718 ++++
 docs/typedoc/classes/dlcontext.html                |  278 ++
 docs/typedoc/classes/dldatatype.html               |  292 ++
 docs/typedoc/classes/environment.html              |  337 ++
 docs/typedoc/classes/ffilibrary.html               |  371 ++
 docs/typedoc/classes/graphruntime.html             |  434 +++
 docs/typedoc/classes/instance.html                 |  918 +++++
 docs/typedoc/classes/memory.html                   |  631 ++++
 docs/typedoc/classes/module.html                   |  323 ++
 docs/typedoc/classes/ndarray.html                  |  443 +++
 docs/typedoc/classes/packedfunccell.html           |  231 ++
 docs/typedoc/classes/rpcserver.html                |  338 ++
 docs/typedoc/classes/scalar.html                   |  221 ++
 docs/typedoc/classes/webgpucontext.html            |  319 ++
 docs/typedoc/enums/aynccallbackcode.html           |  164 +
 docs/typedoc/enums/rpcserverstate.html             |  213 ++
 docs/typedoc/enums/sizeof.html                     |  262 ++
 docs/typedoc/enums/typecode.html                   |  346 ++
 docs/typedoc/index.html                            | 1951 +++++++++++
 docs/typedoc/interfaces/disposable.html            |  193 ++
 docs/typedoc/interfaces/functioninfo.html          |  179 +
 docs/typedoc/interfaces/libraryprovider.html       |  219 ++
 docs/vta/dev/config.html                           |    2 +-
 docs/vta/dev/hardware.html                         |    2 +-
 docs/vta/dev/index.html                            |    2 +-
 docs/vta/index.html                                |    2 +-
 docs/vta/install.html                              |  271 +-
 docs/vta/tutorials/autotvm/sg_execution_times.html |    6 +-
 docs/vta/tutorials/autotvm/tune_relay_vta.html     |  204 +-
 .../tutorials/frontend/deploy_classification.html  |   16 +-
 docs/vta/tutorials/frontend/deploy_detection.html  |    4 +-
 .../vta/tutorials/frontend/sg_execution_times.html |    8 +-
 docs/vta/tutorials/index.html                      |    2 +-
 docs/vta/tutorials/matrix_multiply.html            |  196 +-
 docs/vta/tutorials/optimize/convolution_opt.html   |  427 +--
 .../tutorials/optimize/matrix_multiply_opt.html    |  304 +-
 .../vta/tutorials/optimize/sg_execution_times.html |    8 +-
 docs/vta/tutorials/sg_execution_times.html         |    8 +-
 docs/vta/tutorials/vta_get_started.html            |   89 +-
 1440 files changed, 203813 insertions(+), 186782 deletions(-)

diff --git a/docs/_downloads/02fc8627299fa0b05eb017773b471bfa/from_tflite.py b/docs/_downloads/02fc8627299fa0b05eb017773b471bfa/from_tflite.py
index 3273855..e01a4ec 100644
--- a/docs/_downloads/02fc8627299fa0b05eb017773b471bfa/from_tflite.py
+++ b/docs/_downloads/02fc8627299fa0b05eb017773b471bfa/from_tflite.py
@@ -21,25 +21,12 @@ Compile TFLite Models
 
 This article is an introductory tutorial to deploy TFLite models with Relay.
 
-To get started, Flatbuffers and TFLite package needs to be installed as prerequisites.
-A quick solution is to install Flatbuffers via pip
+To get started, TFLite package needs to be installed as prerequisite.
 
 .. code-block:: bash
 
-    pip install flatbuffers --user
-
-
-To install TFlite packages, you could use our prebuilt wheel:
-
-.. code-block:: bash
-
-    # For python3:
-    wget https://github.com/FrozenGene/tflite/releases/download/v1.13.1/tflite-1.13.1-py3-none-any.whl
-    pip3 install -U tflite-1.13.1-py3-none-any.whl --user
-
-    # For python2:
-    wget https://github.com/FrozenGene/tflite/releases/download/v1.13.1/tflite-1.13.1-py2-none-any.whl
-    pip install -U tflite-1.13.1-py2-none-any.whl --user
+    # install tflite
+    pip install tflite=2.1.0 --user
 
 
 or you could generate TFLite package yourself. The steps are the following:
diff --git a/docs/_downloads/0c30ce88b67b0e8d46494348ab36c9fb/from_tflite.ipynb b/docs/_downloads/0c30ce88b67b0e8d46494348ab36c9fb/from_tflite.ipynb
index 3364e3a..f9ff631 100644
--- a/docs/_downloads/0c30ce88b67b0e8d46494348ab36c9fb/from_tflite.ipynb
+++ b/docs/_downloads/0c30ce88b67b0e8d46494348ab36c9fb/from_tflite.ipynb
@@ -15,7 +15,7 @@
       "cell_type": "markdown",
       "metadata": {},
       "source": [
-        "\nCompile TFLite Models\n=====================\n**Author**: `Zhao Wu <https://github.com/FrozenGene>`_\n\nThis article is an introductory tutorial to deploy TFLite models with Relay.\n\nTo get started, Flatbuffers and TFLite package needs to be installed as prerequisites.\nA quick solution is to install Flatbuffers via pip\n\n.. code-block:: bash\n\n    pip install flatbuffers --user\n\n\nTo install TFlite packages, you could use our prebuilt wheel:\n\n.. code-block:: bash\n\n   [...]
+        "\nCompile TFLite Models\n=====================\n**Author**: `Zhao Wu <https://github.com/FrozenGene>`_\n\nThis article is an introductory tutorial to deploy TFLite models with Relay.\n\nTo get started, TFLite package needs to be installed as prerequisite.\n\n.. code-block:: bash\n\n    # install tflite\n    pip install tflite=2.1.0 --user\n\n\nor you could generate TFLite package yourself. The steps are the following:\n\n.. code-block:: bash\n\n    # Get the flatc compiler.\n    [...]
       ]
     },
     {
diff --git a/docs/_sources/api/links.rst.txt b/docs/_sources/api/links.rst.txt
new file mode 100644
index 0000000..8c22cf8
--- /dev/null
+++ b/docs/_sources/api/links.rst.txt
@@ -0,0 +1,25 @@
+..  Licensed to the Apache Software Foundation (ASF) under one
+    or more contributor license agreements.  See the NOTICE file
+    distributed with this work for additional information
+    regarding copyright ownership.  The ASF licenses this file
+    to you under the Apache License, Version 2.0 (the
+    "License"); you may not use this file except in compliance
+    with the License.  You may obtain a copy of the License at
+
+..    http://www.apache.org/licenses/LICENSE-2.0
+
+..  Unless required by applicable law or agreed to in writing,
+    software distributed under the License is distributed on an
+    "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+    KIND, either express or implied.  See the License for the
+    specific language governing permissions and limitations
+    under the License.
+
+Links to API References
+=======================
+
+This page contains links to API references that are build with different doc build system.
+
+* `C++ doyxgen API <doxygen/index.html>`_
+* `Typescript typedoc API <typedoc/index.html>`_
+* `Java Javadoc API <javadoc/index.html>`_
diff --git a/docs/_sources/api/python/contrib.rst.txt b/docs/_sources/api/python/contrib.rst.txt
index b482d30..8ac4e1f 100644
--- a/docs/_sources/api/python/contrib.rst.txt
+++ b/docs/_sources/api/python/contrib.rst.txt
@@ -48,9 +48,9 @@ tvm.contrib.dlpack
 .. automodule:: tvm.contrib.dlpack
     :members:
 
-tvm.contrib.emscripten
-~~~~~~~~~~~~~~~~~~~~~~
-.. automodule:: tvm.contrib.emscripten
+tvm.contrib.emcc
+~~~~~~~~~~~~~~~~
+.. automodule:: tvm.contrib.emcc
     :members:
 
 tvm.contrib.miopen
diff --git a/docs/_sources/api_links.rst.txt b/docs/_sources/api_links.rst.txt
deleted file mode 100644
index b2a66a5..0000000
--- a/docs/_sources/api_links.rst.txt
+++ /dev/null
@@ -1,25 +0,0 @@
-..  Licensed to the Apache Software Foundation (ASF) under one
-    or more contributor license agreements.  See the NOTICE file
-    distributed with this work for additional information
-    regarding copyright ownership.  The ASF licenses this file
-    to you under the Apache License, Version 2.0 (the
-    "License"); you may not use this file except in compliance
-    with the License.  You may obtain a copy of the License at
-
-..    http://www.apache.org/licenses/LICENSE-2.0
-
-..  Unless required by applicable law or agreed to in writing,
-    software distributed under the License is distributed on an
-    "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
-    KIND, either express or implied.  See the License for the
-    specific language governing permissions and limitations
-    under the License.
-
-Links to API References
-==================================
-
-This page contains links to API references that are build with different doc build system.
-
-* `C++ doyxgen API <doxygen/index.html>`_
-* `Javascript jsdoc API <jsdoc/index.html>`_
-* `Java Javadoc API <javadoc/index.html>`_
diff --git a/docs/_sources/contribute/code_guide.rst.txt b/docs/_sources/contribute/code_guide.rst.txt
index 3d135eb..c932e93 100644
--- a/docs/_sources/contribute/code_guide.rst.txt
+++ b/docs/_sources/contribute/code_guide.rst.txt
@@ -34,6 +34,47 @@ C++ Code Styles
   pass by value is better than pass by const reference in such cases.
 - Favor ``const`` member function when possible.
 
+We use `clang-format` to enforce the code style. Because different version
+of clang-format might change by its version, it is recommended to use the same
+version of the clang-format as the master.
+You can also use the following command via docker.
+
+.. code:: bash
+
+    docker/bash.sh tvmai/ci-lint clang-format-10 [path-to-file]
+
+
+clang-format is also not perfect, when necessary, you can use disble clang-format on certain code regions.
+
+.. code :: c
+
+   // clang-format off
+   void Test() {
+      // clang-format will be disabled in this region.
+   }
+   // clang-format on
+
+
+Because clang-format may not recognize macros, it is recommended to use macro like normal function styles.
+
+
+.. code :: c
+
+   #define MACRO_IMPL { custom impl; }
+   #define MACRO_FUNC(x)
+
+   // not preferred, because clang-format might recognize it as types.
+   virtual void Func1() MACRO_IMPL
+
+   // preferred
+   virtual void Func2() MACRO_IMPL;
+
+   void Func3() {
+     // preferred
+     MACRO_FUNC(xyz);
+   }
+
+
 Python Code Styles
 ------------------
 - The functions and classes are documented in `numpydoc <https://numpydoc.readthedocs.io/en/latest/>`_ format.
diff --git a/docs/_sources/contribute/pull_request.rst.txt b/docs/_sources/contribute/pull_request.rst.txt
index 25dc0da..7e0ba37 100644
--- a/docs/_sources/contribute/pull_request.rst.txt
+++ b/docs/_sources/contribute/pull_request.rst.txt
@@ -29,7 +29,14 @@ This is a quick guide to submit a pull request, please also refer to the detaile
     git rebase upstream/master
 
 - Make sure code style check pass by typing the following command, and all the existing test-cases pass.
-- ``docker/bash.sh tvmai/ci-lint ./tests/scripts/task_lint.sh``. (Note: You must install docker beforehand so you can run a docker image.)
+
+  .. code:: bash
+
+    # Reproduce the lint procedure in the CI.
+    docker/bash.sh tvmai/ci-lint ./tests/scripts/task_lint.sh
+    # Run clang-format check for all the files that changed since upstream/master
+    docker/bash.sh tvmai/ci-lint ./tests/lint/git-clang-format.sh upstream/master
+
 - Add test-cases to cover the new features or bugfix the patch introduces.
 - Document the code you wrote, see more at :ref:`doc_guide`
 - Send the pull request and fix the problems reported by automatic checks.
@@ -44,11 +51,11 @@ This is a quick guide to submit a pull request, please also refer to the detaile
 - The patch can be merged after the reviewers approve the pull request.
 
 
+
 CI Environment
 --------------
 We use docker container to create stable CI environments
 that can be deployed to multiple machines.
-You can find the prebuilt images in `<https://hub.docker.com/r/tvmai/>`_ .
 Because we want a relatively stable CI environment and make use of pre-cached image,
 all of the CI images are built and maintained by committers.
 
diff --git a/docs/_sources/dev/convert_layout.rst.txt b/docs/_sources/dev/convert_layout.rst.txt
index 7345c15..ee5350c 100644
--- a/docs/_sources/dev/convert_layout.rst.txt
+++ b/docs/_sources/dev/convert_layout.rst.txt
@@ -92,7 +92,7 @@ These steps happen for each operator in sequence, where ConvertLayout pass keeps
 .. code-block:: python
 
     @reg.register_convert_op_layout("nn.conv2d")
-    def convert_conv2d(attrs, inputs, tinfos, desired_layout):
+    def convert_conv2d(attrs, inputs, tinfos, desired_layouts):
         """Convert Layout pass registration for conv2d op.
 
         Parameters
@@ -103,8 +103,9 @@ These steps happen for each operator in sequence, where ConvertLayout pass keeps
             The args of the Relay expr to be legalized
         tinfos : list of types
             List of input and output types
-        desired_layout : str
-            The desired layout
+        desired_layouts : list of layout strings
+                List of layouts defining our desired
+                layout for the data and kernel inputs respectively.
 
         Returns
         -------
@@ -113,19 +114,30 @@ These steps happen for each operator in sequence, where ConvertLayout pass keeps
         """
 
         from tvm import relay
-        data_layout = attrs['data_layout']
-        kernel_layout = attrs['kernel_layout']
         data, weight = inputs
-        assert desired_layout == 'NCHW', \
-                "Currently only transformation to NCHW layout is supported."
-        if desired_layout == 'NCHW':
-            new_attrs = dict(attrs)
-            new_attrs['data_layout'] = desired_layout
-            new_attrs['kernel_layout'] = 'OIHW'
+        new_attrs = dict(attrs)
+
+        # We expect 2 desired layouts to be specified, one for the data and one for the kernel.
+        assert len(desired_layouts) == 2, "A desired layout is expected for both of nn.conv2d's inputs"
+
+        # Use the first entry in desired layouts which specifies the data layout.
+        # The expected ordering of layouts for this operator is defined by this function.
+        desired_data_layout, desired_kernel_layout = map(str, desired_layouts)
+
+        assert desired_data_layout != "default", "Data layout cannot be default"
+
+        new_attrs['data_layout'] = desired_data_layout
+
+        if desired_data_layout == 'NCHW':
+            if desired_kernel_layout != 'default':
+                new_attrs['kernel_layout'] = desired_kernel_layout
+            else:
+                new_attrs['kernel_layout'] = 'OIHW'
             # Actual insertion of layout transforms is taken care internally
             # by ConvertLayout pass.
             return relay.nn.conv2d(data, weight, **new_attrs)
-        return None
+
+        raise ValueError('Layout %s is not yet supported' % desired_data_layout)
 
 
 **FInferCorrectLayout - Layout inference** - Currently, this attribute is exposed only in C++. This function takes original input layouts and the new input layouts (passed from the previous operator or from the python callback for layout alteration), and infers the final data layouts. Layout inference is called for each operator. The usage might vary for different operator categories. For layout agnostic operators, we just want to return the new data layouts in this function. For lightly [...]
@@ -218,6 +230,8 @@ Second example is for a lightly-layout sensitive operator - batch normalization.
 
 ConvertLayout pass is extremely easy to use. The pass is not a part of default relay.build pipeline. The intended usage is to call it between the framework-to-relay parser and relay.build module call.
 
+In order to specify the layouts to convert to, we create a mapping of heavily-layout sensitive operators to a list of the desired layouts for that operator. The first example below specifies data layout, we allow the kernel layout to be automatically converted to one that is supported by TVM (for that particular data layout and operator). This is specified by the use of the "default" keyword. The second example shows how we could have also converted to a specific kernel layout of our cho [...]
+
 .. code-block:: python
 
     # TFlite framework to Relay parser - Default layout is NHWC
@@ -225,10 +239,13 @@ ConvertLayout pass is extremely easy to use. The pass is not a part of default r
                                              shape_dict=shape_dict,
                                              dtype_dict=dtype_dict)
 
+    # We assume our model's heavily-layout sensitive operators only consist of nn.conv2d
+    desired_layouts = {'nn.conv2d': ['NCHW', 'default']}
+
     # Convert the layout to NCHW
     # RemoveUnunsedFunctions is used to clean up the graph.
     seq = tvm.transform.Sequential([relay.transform.RemoveUnusedFunctions(),
-                                      relay.transform.ConvertLayout('NCHW')])
+                                    relay.transform.ConvertLayout(desired_layouts)])
     with relay.transform.PassContext(opt_level=3):
         mod = seq(mod)
 
@@ -236,6 +253,15 @@ ConvertLayout pass is extremely easy to use. The pass is not a part of default r
     with relay.build_config(opt_level=3):
          graph, lib, params = relay.build(mod, target, params=params)
 
+
+.. code-block:: python
+
+    desired_layouts = {'nn.conv2d': ['NCHW', 'OIHW']}
+    pass = relay.transform.ConvertLayout(desired_layouts)
+
+
+The ordering of the layouts is defined by the implementation of `register_convert_op_layout("OPNAME")`, you can refer to the docstring which should explicitly state the expected layout. In the examples above it's [data_layout, kernel_layout].
+
 Current implementation has support for almost all the operators commonly used in image classification models. However, if one encounters too many data layout transforms in the graph, it is highly likely that there is an operator whose layouts need special handling as described in Section 3. Some pull requests that can help in such a situation are
 
 - Layout inference for `Batch Norm <https://github.com/apache/incubator-tvm/pull/4600>`_ - Batch normalization falls into the category of lightly-sensitive operator. The PR shows how to handle the layout inference for batch norm.
diff --git a/docs/_sources/dev/relay_add_pass.rst.txt b/docs/_sources/dev/relay_add_pass.rst.txt
index 8a6f8be..3eb9586 100644
--- a/docs/_sources/dev/relay_add_pass.rst.txt
+++ b/docs/_sources/dev/relay_add_pass.rst.txt
@@ -292,7 +292,7 @@ pointed to by ``op->index``. The reason we need to check is because
 .. code:: c
 
     Expr VisitExpr_(const CallNode* call) final {
-      static auto op_stateful = Op::GetAttr<TOpIsStateful>("TOpIsStateful");
+      static auto op_stateful = Op::GetAttrMap<TOpIsStateful>("TOpIsStateful");
       Expr res = ExprMutator::VisitExpr_(call);
       call = res.as<CallNode>();
       // We don't constant fold function with zero arguments.
diff --git a/docs/_sources/index.rst.txt b/docs/_sources/index.rst.txt
index 258547a..5e3fa45 100644
--- a/docs/_sources/index.rst.txt
+++ b/docs/_sources/index.rst.txt
@@ -37,7 +37,7 @@ API Reference
 
    langref/index
    api/python/index
-   api_links
+   api/links
 
 Developer Guide
 ---------------
@@ -47,7 +47,7 @@ Developer Guide
    dev/index
 
 Frontends
-----------------
+---------
 .. toctree::
    :maxdepth: 1
 
diff --git a/docs/_sources/langref/index.rst.txt b/docs/_sources/langref/index.rst.txt
index 0d29611..dcea9fa 100644
--- a/docs/_sources/langref/index.rst.txt
+++ b/docs/_sources/langref/index.rst.txt
@@ -46,6 +46,7 @@ algebraic data types, and operators in Relay, respectively.
    relay_type
    relay_adt
    relay_op
+   relay_pattern
 
 Hybrid Script
 -------------
diff --git a/docs/_sources/langref/relay_pattern.rst.txt b/docs/_sources/langref/relay_pattern.rst.txt
new file mode 100644
index 0000000..7f81b9b
--- /dev/null
+++ b/docs/_sources/langref/relay_pattern.rst.txt
@@ -0,0 +1,141 @@
+..  Licensed to the Apache Software Foundation (ASF) under one
+    or more contributor license agreements.  See the NOTICE file
+    distributed with this work for additional information
+    regarding copyright ownership.  The ASF licenses this file
+    to you under the Apache License, Version 2.0 (the
+    "License"); you may not use this file except in compliance
+    with the License.  You may obtain a copy of the License at
+
+..    http://www.apache.org/licenses/LICENSE-2.0
+
+..  Unless required by applicable law or agreed to in writing,
+    software distributed under the License is distributed on an
+    "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+    KIND, either express or implied.  See the License for the
+    specific language governing permissions and limitations
+    under the License.
+
+
+=========================
+Pattern Matching in Relay
+=========================
+
+There are many places in TVM where we identify pure data-flow sub-graphs of the Relay program and attempt to transform them in some way example passes include fusion, quantization, external code generation, and device specific optimizations such as bitpacking, and layer slicing used by VTA. 
+
+Many of these passes today require a lots of boring boilerplate code in order to implement as well as requiring users to think in terms of visitors and AST matching. Many of these transformations can easily be described in terms of graph rewrites. In order to build a rewriter or other advanced machinery we first need a language of patterns to describe what we can match. 
+
+Such a language is not just useful for building a rewriter but also providing extension points for existing passes. For example the fusion pass could be parameterized by a set of fusion patterns which describes the capability of your hardware, and the quantization pass could take a set of patterns which describe which operators can be quantized on a given platform.
+
+In the backend world, we could use the same machinery to build a higher level API using bring your own code generation. This API takes set of patterns describing your hardware capabilities and an external compiler, providing a relatively smooth heterogeneous experience out of the box.
+
+Examples
+========
+
+There are quite a few properties that are worth matching of operators below we examine how to match tree properties, and expand on some use cases that are not fully explored in the prototype. The first example is a simple case where we want to match one operator with a single input OR another operator with a single input, see the below diagram for a graphical representation and corresponding code::
+
+    def test_match_op_or():
+        is_add_or_sub = is_op('add') | is_op('subtract')
+        assert is_add_or_sub.match(relay.op.op.get("add"))
+        assert is_add_or_sub.match(relay.op.op.get("subtract"))
+
+The next example is a dense operation with any operator that is marked element-wise::
+
+    def test_no_match_attr():
+        op = is_op('nn.dense').has_attr("TOpPattern", K_ELEMWISE)
+        op_pat = op(wildcard(), wildcard())
+        x = relay.var('x')
+        y = relay.var('y')
+        assert not op_pat.match(relay.op.nn.dense(x, y))
+
+The next example is matching a diamond with two inputs at the top of the diamond::
+
+    def test_match_diamond():
+        # Pattern
+        is_conv2d = is_op('nn.conv2d')(is_input(), is_input())
+        path1 = is_op('nn.relu')(is_conv2d)
+        path2 = is_op('nn.leaky_relu')(is_conv2d)
+        diamond = is_op('add')(path1, path2)
+
+        # Expr
+        inp = relay.var('input')
+        weight = relay.var('weight')
+        conv2d = relay.op.nn.conv2d(inp, weight)
+        relu = relay.op.nn.relu(conv2d)
+        leaky_relu = relay.op.nn.leaky_relu(conv2d, alpha=0)
+        out = relu + leaky_relu
+
+        # Check
+        assert diamond.match(out)
+
+The final example  is matching diamonds with a post-dominator relationship. We embed dominator analysis as type of matching in the pattern language in order to allow for pattern matching with unknown topology. This is important because we want to be able to use the language to describe fuse patterns, like elementwise operations followed by a conv2d::
+
+    def test_match_dom_diamond():
+        # Pattern
+        is_conv2d = is_op('nn.conv2d')(is_input(), is_input())
+        reduction = is_op('add')(wildcard(), wildcard())
+        diamond = dominates(is_conv2d, is_elemwise, reduction)
+
+        # Expr
+        inp = relay.var('input')
+        weight = relay.var('weight')
+        conv2d = relay.op.nn.conv2d(inp, weight)
+        relu = relay.op.nn.relu(conv2d)
+        leaky_relu = relay.op.nn.leaky_relu(conv2d, alpha=0)
+        out = relu + leaky_relu
+
+        # Check
+        assert diamond.match(out)
+
+Design
+======
+
+The pattern language proposed is designed to be a mirror of Relay's IR with additional support for common scenarios. The goal of the pattern language is to provide a regular-expression like capability for matching data-flow graphs and doing rewriting.
+
+The high level design is to introduce a language of patterns for now we propose the language as::
+
+    Pattern ::= expr
+            | *
+            | pattern(pattern1, ... patternN)
+            | has_type(pattern, type)
+            | has_attr(pattern, attr, attr_value)
+            | is_input(name)
+            | pattern1 `|` pattern2
+            | dominates(parent_pattern, path_pattern, child_pattern)
+
+The above language then provides a matching interface with both can select sub-graphs as well as verify that the graph does match the pattern.
+
+Expression Pattern
+******************
+
+Match a literal expression.
+
+Wildcard
+********
+
+Match any expression.
+
+Type Pattern
+************
+
+Check that the expression matched by the nested pattern has a particular type.
+
+Attribute Pattern
+*****************
+
+Check that the operator matched by the pattern has an attribute with a particular value.
+
+Input
+*****
+
+Check that the expression is an input, i.e has no parents and is a variable.
+
+
+Alternate
+*********
+
+Either match the first pattern or the second pattern.
+
+Domination
+**********
+
+Match child pattern, find a match for the parent pattern, insuring that the child ultimately dominates the parrent (i.e., no nodes outside the pattern use outputs of the parent), and that ever node betwen the child and the pattern matches the path pattern.
diff --git a/docs/_sources/tutorials/autotvm/sg_execution_times.rst.txt b/docs/_sources/tutorials/autotvm/sg_execution_times.rst.txt
index 27d7622..e97ae0b 100644
--- a/docs/_sources/tutorials/autotvm/sg_execution_times.rst.txt
+++ b/docs/_sources/tutorials/autotvm/sg_execution_times.rst.txt
@@ -5,11 +5,11 @@
 
 Computation times
 =================
-**00:52.017** total execution time for **tutorials_autotvm** files:
-
-- **00:27.259**: :ref:`sphx_glr_tutorials_autotvm_tune_simple_template.py` (``tune_simple_template.py``)
-- **00:24.212**: :ref:`sphx_glr_tutorials_autotvm_tune_conv2d_cuda.py` (``tune_conv2d_cuda.py``)
-- **00:00.147**: :ref:`sphx_glr_tutorials_autotvm_tune_relay_x86.py` (``tune_relay_x86.py``)
-- **00:00.133**: :ref:`sphx_glr_tutorials_autotvm_tune_relay_mobile_gpu.py` (``tune_relay_mobile_gpu.py``)
-- **00:00.133**: :ref:`sphx_glr_tutorials_autotvm_tune_relay_cuda.py` (``tune_relay_cuda.py``)
-- **00:00.133**: :ref:`sphx_glr_tutorials_autotvm_tune_relay_arm.py` (``tune_relay_arm.py``)
+**00:49.863** total execution time for **tutorials_autotvm** files:
+
+- **00:26.895**: :ref:`sphx_glr_tutorials_autotvm_tune_simple_template.py` (``tune_simple_template.py``)
+- **00:22.394**: :ref:`sphx_glr_tutorials_autotvm_tune_conv2d_cuda.py` (``tune_conv2d_cuda.py``)
+- **00:00.169**: :ref:`sphx_glr_tutorials_autotvm_tune_relay_x86.py` (``tune_relay_x86.py``)
+- **00:00.136**: :ref:`sphx_glr_tutorials_autotvm_tune_relay_cuda.py` (``tune_relay_cuda.py``)
+- **00:00.135**: :ref:`sphx_glr_tutorials_autotvm_tune_relay_arm.py` (``tune_relay_arm.py``)
+- **00:00.135**: :ref:`sphx_glr_tutorials_autotvm_tune_relay_mobile_gpu.py` (``tune_relay_mobile_gpu.py``)
diff --git a/docs/_sources/tutorials/autotvm/tune_conv2d_cuda.rst.txt b/docs/_sources/tutorials/autotvm/tune_conv2d_cuda.rst.txt
index 36c2227..8d6e82c 100644
--- a/docs/_sources/tutorials/autotvm/tune_conv2d_cuda.rst.txt
+++ b/docs/_sources/tutorials/autotvm/tune_conv2d_cuda.rst.txt
@@ -234,26 +234,26 @@ for this template
        7 unroll_explicit: OtherOption([0, 1]) len=2
     )
     Get devices for measurement successfully!
-    No: 1   GFLOPS: 0.00/0.00       result: MeasureResult(costs=(InstantiationError('Traceback (most recent call last):\n  [bt] (4) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7ff84dca8c61]\n  [bt] (3) /workspace/build/libtvm.so(+0x4b628c) [0x7ff84d3b728c]\n  [bt] (2) /workspace/build/libtvm.so(tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const+0x2fd) [0x7ff84d3b5d2d]\n  [bt] (1) /workspace/build/libtvm.so(tvm::tir::transform::PrimFunc [...]
-    No: 2   GFLOPS: 0.00/0.00       result: MeasureResult(costs=(InstantiationError('Traceback (most recent call last):\n  [bt] (4) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7ff84dca8c61]\n  [bt] (3) /workspace/build/libtvm.so(+0x4b628c) [0x7ff84d3b728c]\n  [bt] (2) /workspace/build/libtvm.so(tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const+0x2fd) [0x7ff84d3b5d2d]\n  [bt] (1) /workspace/build/libtvm.so(tvm::tir::transform::PrimFunc [...]
-    No: 3   GFLOPS: 0.00/0.00       result: MeasureResult(costs=(InstantiationError('Traceback (most recent call last):\n  [bt] (4) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7ff84dca8c61]\n  [bt] (3) /workspace/build/libtvm.so(+0x4b628c) [0x7ff84d3b728c]\n  [bt] (2) /workspace/build/libtvm.so(tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const+0x2fd) [0x7ff84d3b5d2d]\n  [bt] (1) /workspace/build/libtvm.so(tvm::tir::transform::PrimFunc [...]
-    No: 4   GFLOPS: 0.00/0.00       result: MeasureResult(costs=(InstantiationError('Traceback (most recent call last):\n  [bt] (4) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7ff84dca8c61]\n  [bt] (3) /workspace/build/libtvm.so(+0x4b628c) [0x7ff84d3b728c]\n  [bt] (2) /workspace/build/libtvm.so(tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const+0x2fd) [0x7ff84d3b5d2d]\n  [bt] (1) /workspace/build/libtvm.so(tvm::tir::transform::PrimFunc [...]
-    No: 5   GFLOPS: 0.00/0.00       result: MeasureResult(costs=(InstantiationError('Traceback (most recent call last):\n  [bt] (4) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7ff84dca8c61]\n  [bt] (3) /workspace/build/libtvm.so(+0x4b628c) [0x7ff84d3b728c]\n  [bt] (2) /workspace/build/libtvm.so(tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const+0x2fd) [0x7ff84d3b5d2d]\n  [bt] (1) /workspace/build/libtvm.so(tvm::tir::transform::PrimFunc [...]
-    No: 6   GFLOPS: 0.00/0.00       result: MeasureResult(costs=(InstantiationError('Traceback (most recent call last):\n  [bt] (4) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7ff84dca8c61]\n  [bt] (3) /workspace/build/libtvm.so(+0x4b628c) [0x7ff84d3b728c]\n  [bt] (2) /workspace/build/libtvm.so(tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const+0x2fd) [0x7ff84d3b5d2d]\n  [bt] (1) /workspace/build/libtvm.so(tvm::tir::transform::PrimFunc [...]
-    No: 7   GFLOPS: 0.00/0.00       result: MeasureResult(costs=(InstantiationError('Traceback (most recent call last):\n  [bt] (4) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7ff84dca8c61]\n  [bt] (3) /workspace/build/libtvm.so(+0x4b628c) [0x7ff84d3b728c]\n  [bt] (2) /workspace/build/libtvm.so(tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const+0x2fd) [0x7ff84d3b5d2d]\n  [bt] (1) /workspace/build/libtvm.so(tvm::tir::transform::PrimFunc [...]
-    No: 8   GFLOPS: 0.00/0.00       result: MeasureResult(costs=(InstantiationError('Traceback (most recent call last):\n  [bt] (4) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7ff84dca8c61]\n  [bt] (3) /workspace/build/libtvm.so(+0x4b628c) [0x7ff84d3b728c]\n  [bt] (2) /workspace/build/libtvm.so(tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const+0x2fd) [0x7ff84d3b5d2d]\n  [bt] (1) /workspace/build/libtvm.so(tvm::tir::transform::PrimFunc [...]
-    No: 9   GFLOPS: 0.00/0.00       result: MeasureResult(costs=(InstantiationError('Traceback (most recent call last):\n  [bt] (4) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7ff84dca8c61]\n  [bt] (3) /workspace/build/libtvm.so(+0x4b628c) [0x7ff84d3b728c]\n  [bt] (2) /workspace/build/libtvm.so(tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const+0x2fd) [0x7ff84d3b5d2d]\n  [bt] (1) /workspace/build/libtvm.so(tvm::tir::transform::PrimFunc [...]
-    No: 10  GFLOPS: 0.00/0.00       result: MeasureResult(costs=(InstantiationError('Traceback (most recent call last):\n  [bt] (4) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7ff84dca8c61]\n  [bt] (3) /workspace/build/libtvm.so(+0x4b628c) [0x7ff84d3b728c]\n  [bt] (2) /workspace/build/libtvm.so(tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const+0x2fd) [0x7ff84d3b5d2d]\n  [bt] (1) /workspace/build/libtvm.so(tvm::tir::transform::PrimFunc [...]
-    No: 11  GFLOPS: 0.00/0.00       result: MeasureResult(costs=(InstantiationError('Traceback (most recent call last):\n  [bt] (4) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7ff84dca8c61]\n  [bt] (3) /workspace/build/libtvm.so(+0x4b628c) [0x7ff84d3b728c]\n  [bt] (2) /workspace/build/libtvm.so(tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const+0x2fd) [0x7ff84d3b5d2d]\n  [bt] (1) /workspace/build/libtvm.so(tvm::tir::transform::PrimFunc [...]
-    No: 12  GFLOPS: 57.93/57.93     result: MeasureResult(costs=(0.003995959576923077,), error_no=0, all_cost=2.0548136234283447, timestamp=1588296329.510231)      [('tile_f', [-1, 1, 1, 1]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 1, 1, 1]), ('tile_rc', [-1, 16, 1]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 512), ('unroll_explicit', 1)],None,8533140
-    No: 13  GFLOPS: 0.00/57.93      result: MeasureResult(costs=(InstantiationError('Traceback (most recent call last):\n  [bt] (4) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7ff84dca8c61]\n  [bt] (3) /workspace/build/libtvm.so(+0x4b628c) [0x7ff84d3b728c]\n  [bt] (2) /workspace/build/libtvm.so(tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const+0x2fd) [0x7ff84d3b5d2d]\n  [bt] (1) /workspace/build/libtvm.so(tvm::tir::transform::PrimFunc [...]
-    No: 14  GFLOPS: 0.00/57.93      result: MeasureResult(costs=(InstantiationError('Traceback (most recent call last):\n  [bt] (4) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7ff84dca8c61]\n  [bt] (3) /workspace/build/libtvm.so(+0x4b628c) [0x7ff84d3b728c]\n  [bt] (2) /workspace/build/libtvm.so(tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const+0x2fd) [0x7ff84d3b5d2d]\n  [bt] (1) /workspace/build/libtvm.so(tvm::tir::transform::PrimFunc [...]
-    No: 15  GFLOPS: 0.00/57.93      result: MeasureResult(costs=(InstantiationError('Traceback (most recent call last):\n  [bt] (4) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7ff84dca8c61]\n  [bt] (3) /workspace/build/libtvm.so(+0x4b628c) [0x7ff84d3b728c]\n  [bt] (2) /workspace/build/libtvm.so(tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const+0x2fd) [0x7ff84d3b5d2d]\n  [bt] (1) /workspace/build/libtvm.so(tvm::tir::transform::PrimFunc [...]
-    No: 16  GFLOPS: 0.00/57.93      result: MeasureResult(costs=(InstantiationError('Traceback (most recent call last):\n  [bt] (4) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7ff84dca8c61]\n  [bt] (3) /workspace/build/libtvm.so(+0x4b628c) [0x7ff84d3b728c]\n  [bt] (2) /workspace/build/libtvm.so(tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const+0x2fd) [0x7ff84d3b5d2d]\n  [bt] (1) /workspace/build/libtvm.so(tvm::tir::transform::PrimFunc [...]
-    No: 17  GFLOPS: 0.00/57.93      result: MeasureResult(costs=(InstantiationError('Traceback (most recent call last):\n  [bt] (4) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7ff84dca8c61]\n  [bt] (3) /workspace/build/libtvm.so(+0x4b628c) [0x7ff84d3b728c]\n  [bt] (2) /workspace/build/libtvm.so(tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const+0x2fd) [0x7ff84d3b5d2d]\n  [bt] (1) /workspace/build/libtvm.so(tvm::tir::transform::PrimFunc [...]
-    No: 18  GFLOPS: 0.00/57.93      result: MeasureResult(costs=(InstantiationError('Traceback (most recent call last):\n  [bt] (4) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7ff84dca8c61]\n  [bt] (3) /workspace/build/libtvm.so(+0x4b628c) [0x7ff84d3b728c]\n  [bt] (2) /workspace/build/libtvm.so(tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const+0x2fd) [0x7ff84d3b5d2d]\n  [bt] (1) /workspace/build/libtvm.so(tvm::tir::transform::PrimFunc [...]
-    No: 19  GFLOPS: 789.30/789.30   result: MeasureResult(costs=(0.00029330012114537447,), error_no=0, all_cost=2.554238796234131, timestamp=1588296334.2122595)    [('tile_f', [-1, 2, 1, 1]), ('tile_y', [-1, 1, 7, 1]), ('tile_x', [-1, 1, 7, 1]), ('tile_rc', [-1, 32, 1]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 1)],None,9699801
-    No: 20  GFLOPS: 6.96/789.30     result: MeasureResult(costs=(0.03326839375,), error_no=0, all_cost=2.2470388412475586, timestamp=1588296335.6904595)    [('tile_f', [-1, 2, 1, 16]), ('tile_y', [-1, 1, 1, 1]), ('tile_x', [-1, 7, 1, 1]), ('tile_rc', [-1, 2, 32]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 512), ('unroll_explicit', 1)],None,7502165
+    No: 1   GFLOPS: 0.00/0.00       result: MeasureResult(costs=(InstantiationError('Traceback (most recent call last):\n  [bt] (4) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7fb02147c191]\n  [bt] (3) /workspace/build/libtvm.so(+0x4da061) [0x7fb020b49061]\n  [bt] (2) /workspace/build/libtvm.so(tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const+0x3ca) [0x7fb020b47afa]\n  [bt] (1) /workspace/build/libtvm.so(tvm::tir::transform::PrimFunc [...]
+    No: 2   GFLOPS: 0.00/0.00       result: MeasureResult(costs=(InstantiationError('Traceback (most recent call last):\n  [bt] (4) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7fb02147c191]\n  [bt] (3) /workspace/build/libtvm.so(+0x4da061) [0x7fb020b49061]\n  [bt] (2) /workspace/build/libtvm.so(tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const+0x3ca) [0x7fb020b47afa]\n  [bt] (1) /workspace/build/libtvm.so(tvm::tir::transform::PrimFunc [...]
+    No: 3   GFLOPS: 0.00/0.00       result: MeasureResult(costs=(InstantiationError('Traceback (most recent call last):\n  [bt] (4) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7fb02147c191]\n  [bt] (3) /workspace/build/libtvm.so(+0x4da061) [0x7fb020b49061]\n  [bt] (2) /workspace/build/libtvm.so(tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const+0x3ca) [0x7fb020b47afa]\n  [bt] (1) /workspace/build/libtvm.so(tvm::tir::transform::PrimFunc [...]
+    No: 4   GFLOPS: 0.00/0.00       result: MeasureResult(costs=(InstantiationError('Traceback (most recent call last):\n  [bt] (4) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7fb02147c191]\n  [bt] (3) /workspace/build/libtvm.so(+0x4da061) [0x7fb020b49061]\n  [bt] (2) /workspace/build/libtvm.so(tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const+0x3ca) [0x7fb020b47afa]\n  [bt] (1) /workspace/build/libtvm.so(tvm::tir::transform::PrimFunc [...]
+    No: 5   GFLOPS: 0.00/0.00       result: MeasureResult(costs=(InstantiationError('Traceback (most recent call last):\n  [bt] (4) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7fb02147c191]\n  [bt] (3) /workspace/build/libtvm.so(+0x4da061) [0x7fb020b49061]\n  [bt] (2) /workspace/build/libtvm.so(tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const+0x3ca) [0x7fb020b47afa]\n  [bt] (1) /workspace/build/libtvm.so(tvm::tir::transform::PrimFunc [...]
+    No: 6   GFLOPS: 0.00/0.00       result: MeasureResult(costs=(InstantiationError('Traceback (most recent call last):\n  [bt] (4) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7fb02147c191]\n  [bt] (3) /workspace/build/libtvm.so(+0x4da061) [0x7fb020b49061]\n  [bt] (2) /workspace/build/libtvm.so(tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const+0x3ca) [0x7fb020b47afa]\n  [bt] (1) /workspace/build/libtvm.so(tvm::tir::transform::PrimFunc [...]
+    No: 7   GFLOPS: 0.00/0.00       result: MeasureResult(costs=(InstantiationError('Traceback (most recent call last):\n  [bt] (4) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7fb02147c191]\n  [bt] (3) /workspace/build/libtvm.so(+0x4da061) [0x7fb020b49061]\n  [bt] (2) /workspace/build/libtvm.so(tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const+0x3ca) [0x7fb020b47afa]\n  [bt] (1) /workspace/build/libtvm.so(tvm::tir::transform::PrimFunc [...]
+    No: 8   GFLOPS: 0.00/0.00       result: MeasureResult(costs=(InstantiationError('Traceback (most recent call last):\n  [bt] (4) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7fb02147c191]\n  [bt] (3) /workspace/build/libtvm.so(+0x4da061) [0x7fb020b49061]\n  [bt] (2) /workspace/build/libtvm.so(tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const+0x3ca) [0x7fb020b47afa]\n  [bt] (1) /workspace/build/libtvm.so(tvm::tir::transform::PrimFunc [...]
+    No: 9   GFLOPS: 0.00/0.00       result: MeasureResult(costs=(InstantiationError('Traceback (most recent call last):\n  [bt] (4) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7fb02147c191]\n  [bt] (3) /workspace/build/libtvm.so(+0x4da061) [0x7fb020b49061]\n  [bt] (2) /workspace/build/libtvm.so(tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const+0x3ca) [0x7fb020b47afa]\n  [bt] (1) /workspace/build/libtvm.so(tvm::tir::transform::PrimFunc [...]
+    No: 10  GFLOPS: 0.00/0.00       result: MeasureResult(costs=(InstantiationError('Traceback (most recent call last):\n  [bt] (4) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7fb02147c191]\n  [bt] (3) /workspace/build/libtvm.so(+0x4da061) [0x7fb020b49061]\n  [bt] (2) /workspace/build/libtvm.so(tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const+0x3ca) [0x7fb020b47afa]\n  [bt] (1) /workspace/build/libtvm.so(tvm::tir::transform::PrimFunc [...]
+    No: 11  GFLOPS: 0.00/0.00       result: MeasureResult(costs=(InstantiationError('Traceback (most recent call last):\n  [bt] (4) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7fb02147c191]\n  [bt] (3) /workspace/build/libtvm.so(+0x4da061) [0x7fb020b49061]\n  [bt] (2) /workspace/build/libtvm.so(tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const+0x3ca) [0x7fb020b47afa]\n  [bt] (1) /workspace/build/libtvm.so(tvm::tir::transform::PrimFunc [...]
+    No: 12  GFLOPS: 70.79/70.79     result: MeasureResult(costs=(0.0032700486428571427,), error_no=0, all_cost=1.5509259700775146, timestamp=1589905599.9728117)    [('tile_f', [-1, 1, 1, 1]), ('tile_y', [-1, 1, 1, 7]), ('tile_x', [-1, 1, 1, 1]), ('tile_rc', [-1, 16, 1]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 1, 3]), ('auto_unroll_max_step', 512), ('unroll_explicit', 1)],None,8533140
+    No: 13  GFLOPS: 0.00/70.79      result: MeasureResult(costs=(InstantiationError('Traceback (most recent call last):\n  [bt] (4) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7fb02147c191]\n  [bt] (3) /workspace/build/libtvm.so(+0x4da061) [0x7fb020b49061]\n  [bt] (2) /workspace/build/libtvm.so(tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const+0x3ca) [0x7fb020b47afa]\n  [bt] (1) /workspace/build/libtvm.so(tvm::tir::transform::PrimFunc [...]
+    No: 14  GFLOPS: 0.00/70.79      result: MeasureResult(costs=(InstantiationError('Traceback (most recent call last):\n  [bt] (4) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7fb02147c191]\n  [bt] (3) /workspace/build/libtvm.so(+0x4da061) [0x7fb020b49061]\n  [bt] (2) /workspace/build/libtvm.so(tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const+0x3ca) [0x7fb020b47afa]\n  [bt] (1) /workspace/build/libtvm.so(tvm::tir::transform::PrimFunc [...]
+    No: 15  GFLOPS: 0.00/70.79      result: MeasureResult(costs=(InstantiationError('Traceback (most recent call last):\n  [bt] (4) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7fb02147c191]\n  [bt] (3) /workspace/build/libtvm.so(+0x4da061) [0x7fb020b49061]\n  [bt] (2) /workspace/build/libtvm.so(tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const+0x3ca) [0x7fb020b47afa]\n  [bt] (1) /workspace/build/libtvm.so(tvm::tir::transform::PrimFunc [...]
+    No: 16  GFLOPS: 0.00/70.79      result: MeasureResult(costs=(InstantiationError('Traceback (most recent call last):\n  [bt] (4) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7fb02147c191]\n  [bt] (3) /workspace/build/libtvm.so(+0x4da061) [0x7fb020b49061]\n  [bt] (2) /workspace/build/libtvm.so(tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const+0x3ca) [0x7fb020b47afa]\n  [bt] (1) /workspace/build/libtvm.so(tvm::tir::transform::PrimFunc [...]
+    No: 17  GFLOPS: 0.00/70.79      result: MeasureResult(costs=(InstantiationError('Traceback (most recent call last):\n  [bt] (4) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7fb02147c191]\n  [bt] (3) /workspace/build/libtvm.so(+0x4da061) [0x7fb020b49061]\n  [bt] (2) /workspace/build/libtvm.so(tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const+0x3ca) [0x7fb020b47afa]\n  [bt] (1) /workspace/build/libtvm.so(tvm::tir::transform::PrimFunc [...]
+    No: 18  GFLOPS: 0.00/70.79      result: MeasureResult(costs=(InstantiationError('Traceback (most recent call last):\n  [bt] (4) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7fb02147c191]\n  [bt] (3) /workspace/build/libtvm.so(+0x4da061) [0x7fb020b49061]\n  [bt] (2) /workspace/build/libtvm.so(tvm::transform::SequentialNode::operator()(tvm::IRModule, tvm::transform::PassContext const&) const+0x3ca) [0x7fb020b47afa]\n  [bt] (1) /workspace/build/libtvm.so(tvm::tir::transform::PrimFunc [...]
+    No: 19  GFLOPS: 794.56/794.56   result: MeasureResult(costs=(0.0002913596233480176,), error_no=0, all_cost=2.470529317855835, timestamp=1589905604.5863163)     [('tile_f', [-1, 2, 1, 1]), ('tile_y', [-1, 1, 7, 1]), ('tile_x', [-1, 1, 7, 1]), ('tile_rc', [-1, 32, 1]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 1)],None,9699801
+    No: 20  GFLOPS: 6.97/794.56     result: MeasureResult(costs=(0.033195469,), error_no=0, all_cost=1.6219539642333984, timestamp=1589905605.963692)       [('tile_f', [-1, 2, 1, 16]), ('tile_y', [-1, 1, 1, 1]), ('tile_x', [-1, 7, 1, 1]), ('tile_rc', [-1, 2, 32]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 1, 1]), ('auto_unroll_max_step', 512), ('unroll_explicit', 1)],None,7502165
 
 
 
@@ -307,7 +307,7 @@ and measure running time.
 
     Best config:
     [('tile_f', [-1, 2, 1, 1]), ('tile_y', [-1, 1, 7, 1]), ('tile_x', [-1, 1, 7, 1]), ('tile_rc', [-1, 32, 1]), ('tile_ry', [-1, 1, 3]), ('tile_rx', [-1, 3, 1]), ('auto_unroll_max_step', 1500), ('unroll_explicit', 1)],None,9699801
-    Time cost of this operator: 0.000355
+    Time cost of this operator: 0.000357
 
 
 
diff --git a/docs/_sources/tutorials/autotvm/tune_simple_template.rst.txt b/docs/_sources/tutorials/autotvm/tune_simple_template.rst.txt
index dc6fbd8..b5deab6 100644
--- a/docs/_sources/tutorials/autotvm/tune_simple_template.rst.txt
+++ b/docs/_sources/tutorials/autotvm/tune_simple_template.rst.txt
@@ -361,16 +361,16 @@ used to get the best config later.
  .. code-block:: none
 
     Get devices for measurement successfully!
-    No: 1   GFLOPS: 6.96/6.96       result: MeasureResult(costs=(0.0385797166,), error_no=0, all_cost=0.8728656768798828, timestamp=1588296295.960279)      [('tile_y', [-1, 8]), ('tile_x', [-1, 8])],None,33
-    No: 2   GFLOPS: 2.26/6.96       result: MeasureResult(costs=(0.11858004,), error_no=0, all_cost=2.1669492721557617, timestamp=1588296298.2292306)       [('tile_y', [-1, 8]), ('tile_x', [-1, 2])],None,13
-    No: 3   GFLOPS: 7.30/7.30       result: MeasureResult(costs=(0.0367570304,), error_no=0, all_cost=1.125162124633789, timestamp=1588296299.196056)       [('tile_y', [-1, 16]), ('tile_x', [-1, 128])],None,74
-    No: 4   GFLOPS: 4.41/7.30       result: MeasureResult(costs=(0.060877514,), error_no=0, all_cost=1.292130708694458, timestamp=1588296300.5514374)       [('tile_y', [-1, 16]), ('tile_x', [-1, 32])],None,54
-    No: 5   GFLOPS: 6.12/7.30       result: MeasureResult(costs=(0.0438933678,), error_no=0, all_cost=0.9459524154663086, timestamp=1588296301.6312118)     [('tile_y', [-1, 128]), ('tile_x', [-1, 8])],None,37
-    No: 6   GFLOPS: 6.52/7.30       result: MeasureResult(costs=(0.0411427822,), error_no=0, all_cost=1.0240263938903809, timestamp=1588296302.660769)      [('tile_y', [-1, 64]), ('tile_x', [-1, 8])],None,36
-    No: 7   GFLOPS: 26.27/26.27     result: MeasureResult(costs=(0.0102179166,), error_no=0, all_cost=0.4918959140777588, timestamp=1588296303.1997833)     [('tile_y', [-1, 1]), ('tile_x', [-1, 128])],None,70
-    No: 8   GFLOPS: 20.88/26.27     result: MeasureResult(costs=(0.0128584314,), error_no=0, all_cost=0.5001192092895508, timestamp=1588296303.777795)      [('tile_y', [-1, 4]), ('tile_x', [-1, 512])],None,92
-    No: 9   GFLOPS: 0.82/26.27      result: MeasureResult(costs=(0.3262719522,), error_no=0, all_cost=5.464502334594727, timestamp=1588296309.3735292)      [('tile_y', [-1, 512]), ('tile_x', [-1, 2])],None,19
-    No: 10  GFLOPS: 0.00/26.27      result: MeasureResult(costs=(RuntimeError('Traceback (most recent call last):\n  [bt] (3) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7ff84dca8c61]\n  [bt] (2) /workspace/build/libtvm.so(std::_Function_handler<void (tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*), tvm::runtime::RPCModuleNode::WrapRemote(void*)::{lambda(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*)#1}>::_M_invoke(std::_Any_data const&, tvm::runtime::TVMArgs&&, tvm::runtime:: [...]
+    No: 1   GFLOPS: 7.13/7.13       result: MeasureResult(costs=(0.0376472006,), error_no=0, all_cost=0.8645853996276855, timestamp=1589905568.5166452)     [('tile_y', [-1, 8]), ('tile_x', [-1, 8])],None,33
+    No: 2   GFLOPS: 2.31/7.13       result: MeasureResult(costs=(0.11604594640000002,), error_no=0, all_cost=2.1710281372070312, timestamp=1589905570.7549071)      [('tile_y', [-1, 8]), ('tile_x', [-1, 2])],None,13
+    No: 3   GFLOPS: 7.49/7.49       result: MeasureResult(costs=(0.0358356658,), error_no=0, all_cost=1.1851162910461426, timestamp=1589905571.7046235)     [('tile_y', [-1, 16]), ('tile_x', [-1, 128])],None,74
+    No: 4   GFLOPS: 4.48/7.49       result: MeasureResult(costs=(0.0598577358,), error_no=0, all_cost=1.3230500221252441, timestamp=1589905573.0249946)     [('tile_y', [-1, 16]), ('tile_x', [-1, 32])],None,54
+    No: 5   GFLOPS: 6.76/7.49       result: MeasureResult(costs=(0.0396876996,), error_no=0, all_cost=0.9506959915161133, timestamp=1589905574.0419118)     [('tile_y', [-1, 128]), ('tile_x', [-1, 8])],None,37
+    No: 6   GFLOPS: 6.59/7.49       result: MeasureResult(costs=(0.040720794399999996,), error_no=0, all_cost=0.9806015491485596, timestamp=1589905575.0622969)     [('tile_y', [-1, 64]), ('tile_x', [-1, 8])],None,36
+    No: 7   GFLOPS: 26.82/26.82     result: MeasureResult(costs=(0.010009581200000001,), error_no=0, all_cost=0.5352885723114014, timestamp=1589905575.5948904)     [('tile_y', [-1, 1]), ('tile_x', [-1, 128])],None,70
+    No: 8   GFLOPS: 21.86/26.82     result: MeasureResult(costs=(0.012282078,), error_no=0, all_cost=0.46282339096069336, timestamp=1589905576.173275)      [('tile_y', [-1, 4]), ('tile_x', [-1, 512])],None,92
+    No: 9   GFLOPS: 0.85/26.82      result: MeasureResult(costs=(0.3147072902,), error_no=0, all_cost=5.290054798126221, timestamp=1589905581.5827458)      [('tile_y', [-1, 512]), ('tile_x', [-1, 2])],None,19
+    No: 10  GFLOPS: 0.00/26.82      result: MeasureResult(costs=(RuntimeError('Traceback (most recent call last):\n  [bt] (5) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7fb02147c191]\n  [bt] (4) /workspace/build/libtvm.so(+0xe3fac2) [0x7fb0214aeac2]\n  [bt] (3) /workspace/build/libtvm.so(tvm::runtime::RPCWrappedFunc::operator()(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const+0x21f) [0x7fb0214b253f]\n  [bt] (2) /workspace/build/libtvm.so(tvm::runtime::RPCClientSession::CallF [...]
 
 
 
diff --git a/docs/_sources/tutorials/cross_compilation_and_rpc.rst.txt b/docs/_sources/tutorials/cross_compilation_and_rpc.rst.txt
index e3f17e7..380483b 100644
--- a/docs/_sources/tutorials/cross_compilation_and_rpc.rst.txt
+++ b/docs/_sources/tutorials/cross_compilation_and_rpc.rst.txt
@@ -235,7 +235,7 @@ device and returns the measured cost. Network overhead is excluded.
 
  .. code-block:: none
 
-    1.683e-07 secs/op
+    1.292e-07 secs/op
 
 
 
diff --git a/docs/_sources/tutorials/dev/low_level_custom_pass.rst.txt b/docs/_sources/tutorials/dev/low_level_custom_pass.rst.txt
index 85cee67..ba2519c 100644
--- a/docs/_sources/tutorials/dev/low_level_custom_pass.rst.txt
+++ b/docs/_sources/tutorials/dev/low_level_custom_pass.rst.txt
@@ -72,13 +72,18 @@ our customized lowering pass to manipulate the IR directly instead of using sche
 
  .. code-block:: none
 
-    PrimFunc([a, b, c]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      for (i, 0, 128) {
-        c[i] = (a[i] + b[i])
+    primfn(a_1: handle, b_1: handle, c_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {a: Buffer(a_2: handle, float32, [128], []),
+                 b: Buffer(b_2: handle, float32, [128], []),
+                 c: Buffer(c_2: handle, float32, [128], [])}
+      buffer_map = {c_1: c, b_1: b, a_1: a} {
+      for (i: int32, 0, 128) {
+        c_2[i] = ((float32*)a_2[i]) + (float32*)b_2[i]))
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -211,13 +216,18 @@ Thus, a good place to put this transformation pass is just after Phase 1.
 
  .. code-block:: none
 
-    PrimFunc([a, b, c]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      for (i.outer, 0, 16) {
-        c[ramp((i.outer*8), 1, 8)] = (a[ramp((i.outer*8), 1, 8)] + b[ramp((i.outer*8), 1, 8)])
+    primfn(a_1: handle, b_1: handle, c_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {a: Buffer(a_2: handle, float32, [128], []),
+                 b: Buffer(b_2: handle, float32, [128], []),
+                 c: Buffer(c_2: handle, float32, [128], [])}
+      buffer_map = {c_1: c, b_1: b, a_1: a} {
+      for (i.outer: int32, 0, 16) {
+        c_2[ramp((i.outer*8), 1, 8)] = ((float32x8*)a_2[ramp((i.outer*8), 1, 8)]) + (float32x8*)b_2[ramp((i.outer*8), 1, 8)]))
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
diff --git a/docs/_sources/tutorials/dev/sg_execution_times.rst.txt b/docs/_sources/tutorials/dev/sg_execution_times.rst.txt
index 0ac3c8a..b4b2855 100644
--- a/docs/_sources/tutorials/dev/sg_execution_times.rst.txt
+++ b/docs/_sources/tutorials/dev/sg_execution_times.rst.txt
@@ -5,7 +5,7 @@
 
 Computation times
 =================
-**00:00.457** total execution time for **tutorials_dev** files:
+**00:00.468** total execution time for **tutorials_dev** files:
 
-- **00:00.324**: :ref:`sphx_glr_tutorials_dev_relay_pass_infra.py` (``relay_pass_infra.py``)
-- **00:00.133**: :ref:`sphx_glr_tutorials_dev_low_level_custom_pass.py` (``low_level_custom_pass.py``)
+- **00:00.332**: :ref:`sphx_glr_tutorials_dev_relay_pass_infra.py` (``relay_pass_infra.py``)
+- **00:00.136**: :ref:`sphx_glr_tutorials_dev_low_level_custom_pass.py` (``low_level_custom_pass.py``)
diff --git a/docs/_sources/tutorials/frontend/deploy_model_on_android.rst.txt b/docs/_sources/tutorials/frontend/deploy_model_on_android.rst.txt
index 09f19aa..480f979 100644
--- a/docs/_sources/tutorials/frontend/deploy_model_on_android.rst.txt
+++ b/docs/_sources/tutorials/frontend/deploy_model_on_android.rst.txt
@@ -415,7 +415,7 @@ Execute on TVM
 
     TVM prediction top-1: tiger cat
     Evaluate inference time cost...
-    Mean inference time (std dev): 3.14 ms (0.27 ms)
+    Mean inference time (std dev): 2.91 ms (0.02 ms)
 
 
 
diff --git a/docs/_sources/tutorials/frontend/deploy_prequantized.rst.txt b/docs/_sources/tutorials/frontend/deploy_prequantized.rst.txt
index a72bfd3..0601927 100644
--- a/docs/_sources/tutorials/frontend/deploy_prequantized.rst.txt
+++ b/docs/_sources/tutorials/frontend/deploy_prequantized.rst.txt
@@ -352,7 +352,7 @@ Here we give an example of how to measure performance of TVM compiled models.
 
  .. code-block:: none
 
-    Elapsed average ms: 11.220273310000003
+    Elapsed average ms: 11.351677870000001
 
 
 
diff --git a/docs/_sources/tutorials/frontend/deploy_ssd_gluoncv.rst.txt b/docs/_sources/tutorials/frontend/deploy_ssd_gluoncv.rst.txt
index a724cb9..c8b8d36 100644
--- a/docs/_sources/tutorials/frontend/deploy_ssd_gluoncv.rst.txt
+++ b/docs/_sources/tutorials/frontend/deploy_ssd_gluoncv.rst.txt
@@ -169,45 +169,6 @@ Create TVM runtime and do inference
 
  .. code-block:: none
 
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 3, 512, 512), 'float32'), ('TENSOR', (64, 3, 7, 7), 'float32'), (2, 2), (3, 3, 3, 3), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 64, 128, 128), 'float32'), ('TENSOR', (64, 64, 1, 1), 'float32'), (1, 1), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 64, 128, 128), 'float32'), ('TENSOR', (64, 64, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 64, 128, 128), 'float32'), ('TENSOR', (256, 64, 1, 1), 'float32'), (1, 1), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 256, 128, 128), 'float32'), ('TENSOR', (64, 256, 1, 1), 'float32'), (1, 1), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 256, 128, 128), 'float32'), ('TENSOR', (128, 256, 1, 1), 'float32'), (2, 2), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 128, 64, 64), 'float32'), ('TENSOR', (128, 128, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 128, 64, 64), 'float32'), ('TENSOR', (512, 128, 1, 1), 'float32'), (1, 1), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 256, 128, 128), 'float32'), ('TENSOR', (512, 256, 1, 1), 'float32'), (2, 2), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 512, 64, 64), 'float32'), ('TENSOR', (128, 512, 1, 1), 'float32'), (1, 1), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 512, 64, 64), 'float32'), ('TENSOR', (256, 512, 1, 1), 'float32'), (2, 2), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 256, 32, 32), 'float32'), ('TENSOR', (256, 256, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 256, 32, 32), 'float32'), ('TENSOR', (1024, 256, 1, 1), 'float32'), (1, 1), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 512, 64, 64), 'float32'), ('TENSOR', (1024, 512, 1, 1), 'float32'), (2, 2), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 1024, 32, 32), 'float32'), ('TENSOR', (256, 1024, 1, 1), 'float32'), (1, 1), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 1024, 32, 32), 'float32'), ('TENSOR', (84, 1024, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 1024, 32, 32), 'float32'), ('TENSOR', (512, 1024, 1, 1), 'float32'), (2, 2), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 512, 16, 16), 'float32'), ('TENSOR', (512, 512, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 512, 16, 16), 'float32'), ('TENSOR', (2048, 512, 1, 1), 'float32'), (1, 1), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 1024, 32, 32), 'float32'), ('TENSOR', (2048, 1024, 1, 1), 'float32'), (2, 2), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 2048, 16, 16), 'float32'), ('TENSOR', (512, 2048, 1, 1), 'float32'), (1, 1), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 2048, 16, 16), 'float32'), ('TENSOR', (126, 2048, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 512, 16, 16), 'float32'), ('TENSOR', (512, 512, 3, 3), 'float32'), (2, 2), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 512, 8, 8), 'float32'), ('TENSOR', (126, 512, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 512, 8, 8), 'float32'), ('TENSOR', (512, 512, 1, 1), 'float32'), (1, 1), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 512, 8, 8), 'float32'), ('TENSOR', (512, 512, 3, 3), 'float32'), (2, 2), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 512, 4, 4), 'float32'), ('TENSOR', (126, 512, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 512, 4, 4), 'float32'), ('TENSOR', (256, 512, 1, 1), 'float32'), (1, 1), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 256, 4, 4), 'float32'), ('TENSOR', (256, 256, 3, 3), 'float32'), (2, 2), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 256, 2, 2), 'float32'), ('TENSOR', (84, 256, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 256, 2, 2), 'float32'), ('TENSOR', (256, 256, 1, 1), 'float32'), (1, 1), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 256, 2, 2), 'float32'), ('TENSOR', (256, 256, 3, 3), 'float32'), (2, 2), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 256, 1, 1), 'float32'), ('TENSOR', (84, 256, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 1024, 32, 32), 'float32'), ('TENSOR', (16, 1024, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 2048, 16, 16), 'float32'), ('TENSOR', (24, 2048, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 512, 8, 8), 'float32'), ('TENSOR', (24, 512, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 512, 4, 4), 'float32'), ('TENSOR', (24, 512, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 256, 2, 2), 'float32'), ('TENSOR', (16, 256, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
-    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 256, 1, 1), 'float32'), ('TENSOR', (16, 256, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
     Cannot find config for target=cuda, workload=('conv2d_nchw.cuda', ('TENSOR', (1, 3, 512, 512), 'float32'), ('TENSOR', (64, 3, 7, 7), 'float32'), (2, 2), (3, 3, 3, 3), (1, 1), 'float32'). A fallback configuration is used, which may bring great performance regression.
     Cannot find config for target=cuda, workload=('conv2d_nchw.cuda', ('TENSOR', (1, 64, 128, 128), 'float32'), ('TENSOR', (64, 64, 1, 1), 'float32'), (1, 1), (0, 0, 0, 0), (1, 1), 'float32'). A fallback configuration is used, which may bring great performance regression.
     Cannot find config for target=cuda, workload=('conv2d_nchw.cuda', ('TENSOR', (1, 64, 128, 128), 'float32'), ('TENSOR', (64, 64, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'float32'). A fallback configuration is used, which may bring great performance regression.
@@ -263,6 +224,45 @@ Create TVM runtime and do inference
     Cannot find config for target=cuda, workload=('conv2d_nchw_winograd.cuda', ('TENSOR', (1, 256, 2, 2), 'float32'), ('TENSOR', (16, 256, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'float32'). A fallback configuration is used, which may bring great performance regression.
     Cannot find config for target=cuda, workload=('conv2d_nchw.cuda', ('TENSOR', (1, 256, 1, 1), 'float32'), ('TENSOR', (16, 256, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'float32'). A fallback configuration is used, which may bring great performance regression.
     Cannot find config for target=cuda, workload=('conv2d_nchw_winograd.cuda', ('TENSOR', (1, 256, 1, 1), 'float32'), ('TENSOR', (16, 256, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 3, 512, 512), 'float32'), ('TENSOR', (64, 3, 7, 7), 'float32'), (2, 2), (3, 3, 3, 3), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 64, 128, 128), 'float32'), ('TENSOR', (64, 64, 1, 1), 'float32'), (1, 1), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 64, 128, 128), 'float32'), ('TENSOR', (64, 64, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 64, 128, 128), 'float32'), ('TENSOR', (256, 64, 1, 1), 'float32'), (1, 1), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 256, 128, 128), 'float32'), ('TENSOR', (64, 256, 1, 1), 'float32'), (1, 1), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 256, 128, 128), 'float32'), ('TENSOR', (128, 256, 1, 1), 'float32'), (2, 2), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 128, 64, 64), 'float32'), ('TENSOR', (128, 128, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 128, 64, 64), 'float32'), ('TENSOR', (512, 128, 1, 1), 'float32'), (1, 1), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 256, 128, 128), 'float32'), ('TENSOR', (512, 256, 1, 1), 'float32'), (2, 2), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 512, 64, 64), 'float32'), ('TENSOR', (128, 512, 1, 1), 'float32'), (1, 1), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 512, 64, 64), 'float32'), ('TENSOR', (256, 512, 1, 1), 'float32'), (2, 2), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 256, 32, 32), 'float32'), ('TENSOR', (256, 256, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 256, 32, 32), 'float32'), ('TENSOR', (1024, 256, 1, 1), 'float32'), (1, 1), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 512, 64, 64), 'float32'), ('TENSOR', (1024, 512, 1, 1), 'float32'), (2, 2), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 1024, 32, 32), 'float32'), ('TENSOR', (256, 1024, 1, 1), 'float32'), (1, 1), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 1024, 32, 32), 'float32'), ('TENSOR', (84, 1024, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 1024, 32, 32), 'float32'), ('TENSOR', (512, 1024, 1, 1), 'float32'), (2, 2), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 512, 16, 16), 'float32'), ('TENSOR', (512, 512, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 512, 16, 16), 'float32'), ('TENSOR', (2048, 512, 1, 1), 'float32'), (1, 1), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 1024, 32, 32), 'float32'), ('TENSOR', (2048, 1024, 1, 1), 'float32'), (2, 2), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 2048, 16, 16), 'float32'), ('TENSOR', (512, 2048, 1, 1), 'float32'), (1, 1), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 2048, 16, 16), 'float32'), ('TENSOR', (126, 2048, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 512, 16, 16), 'float32'), ('TENSOR', (512, 512, 3, 3), 'float32'), (2, 2), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 512, 8, 8), 'float32'), ('TENSOR', (126, 512, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 512, 8, 8), 'float32'), ('TENSOR', (512, 512, 1, 1), 'float32'), (1, 1), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 512, 8, 8), 'float32'), ('TENSOR', (512, 512, 3, 3), 'float32'), (2, 2), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 512, 4, 4), 'float32'), ('TENSOR', (126, 512, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 512, 4, 4), 'float32'), ('TENSOR', (256, 512, 1, 1), 'float32'), (1, 1), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 256, 4, 4), 'float32'), ('TENSOR', (256, 256, 3, 3), 'float32'), (2, 2), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 256, 2, 2), 'float32'), ('TENSOR', (84, 256, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 256, 2, 2), 'float32'), ('TENSOR', (256, 256, 1, 1), 'float32'), (1, 1), (0, 0, 0, 0), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 256, 2, 2), 'float32'), ('TENSOR', (256, 256, 3, 3), 'float32'), (2, 2), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 256, 1, 1), 'float32'), ('TENSOR', (84, 256, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 1024, 32, 32), 'float32'), ('TENSOR', (16, 1024, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 2048, 16, 16), 'float32'), ('TENSOR', (24, 2048, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 512, 8, 8), 'float32'), ('TENSOR', (24, 512, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 512, 4, 4), 'float32'), ('TENSOR', (24, 512, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 256, 2, 2), 'float32'), ('TENSOR', (16, 256, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
+    Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 256, 1, 1), 'float32'), ('TENSOR', (16, 256, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
 
 
 
diff --git a/docs/_sources/tutorials/frontend/from_onnx.rst.txt b/docs/_sources/tutorials/frontend/from_onnx.rst.txt
index ced58d3..fec093d 100644
--- a/docs/_sources/tutorials/frontend/from_onnx.rst.txt
+++ b/docs/_sources/tutorials/frontend/from_onnx.rst.txt
@@ -125,7 +125,7 @@ Compile the model with relay
 
  .. code-block:: none
 
-    /workspace/docs/../python/tvm/relay/frontend/onnx.py:1947: UserWarning: Mismatched attribute type in ' : kernel_shape'
+    /workspace/docs/../python/tvm/relay/frontend/onnx.py:1999: UserWarning: Mismatched attribute type in ' : kernel_shape'
 
     ==> Context: Bad node spec: input: "1" input: "2" output: "11" op_type: "Conv" attribute { name: "kernel_shape" ints: 5 ints: 5 } attribute { name: "strides" ints: 1 ints: 1 } attribute { name: "pads" ints: 2 ints: 2 ints: 2 ints: 2 } attribute { name: "dilations" ints: 1 ints: 1 } attribute { name: "group" i: 1 }
       warnings.warn(str(e))
@@ -151,7 +151,7 @@ Execute on TVM
 
  .. code-block:: none
 
-
    ...47%, 0.01 MB, 51 KB/s, 0 seconds passed
    ...94%, 0.02 MB, 102 KB/s, 0 seconds passed
    ...100%, 0.02 MB, 153 KB/s, 0 seconds passed
+
    ...47%, 0.01 MB, 41 KB/s, 0 seconds passed
    ...94%, 0.02 MB, 83 KB/s, 0 seconds passed
    ...100%, 0.02 MB, 124 KB/s, 0 seconds passed
     Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 32, 224, 224), 'float32'), ('TENSOR', (9, 32, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
     Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 64, 224, 224), 'float32'), ('TENSOR', (32, 64, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
     Cannot find config for target=llvm, workload=('conv2d_NCHWc.x86', ('TENSOR', (1, 1, 224, 224), 'float32'), ('TENSOR', (64, 1, 5, 5), 'float32'), (1, 1), (2, 2, 2, 2), (1, 1), 'NCHW', 'NCHW', 'float32'). A fallback configuration is used, which may bring great performance regression.
diff --git a/docs/_sources/tutorials/frontend/from_tensorflow.rst.txt b/docs/_sources/tutorials/frontend/from_tensorflow.rst.txt
index fd22f74..606fe21 100644
--- a/docs/_sources/tutorials/frontend/from_tensorflow.rst.txt
+++ b/docs/_sources/tutorials/frontend/from_tensorflow.rst.txt
@@ -197,7 +197,7 @@ Results:
 
     ANTLR runtime and generated code versions disagree: 4.8!=4.7.2
     ANTLR runtime and generated code versions disagree: 4.8!=4.7.2
-    /workspace/docs/../python/tvm/relay/frontend/tensorflow.py:2818: UserWarning: Ignore the passed shape. Shape in graphdef will be used for operator DecodeJpeg/contents.
+    /workspace/docs/../python/tvm/relay/frontend/tensorflow.py:2822: UserWarning: Ignore the passed shape. Shape in graphdef will be used for operator DecodeJpeg/contents.
       "will be used for operator %s." % node.name)
     /workspace/docs/../python/tvm/relay/frontend/tensorflow.py:620: UserWarning: DecodeJpeg: It's a pass through, please handle preprocessing before input
       warnings.warn("DecodeJpeg: It's a pass through, please handle preprocessing before input")
diff --git a/docs/_sources/tutorials/frontend/from_tflite.rst.txt b/docs/_sources/tutorials/frontend/from_tflite.rst.txt
index e7d8a86..7f8d67b 100644
--- a/docs/_sources/tutorials/frontend/from_tflite.rst.txt
+++ b/docs/_sources/tutorials/frontend/from_tflite.rst.txt
@@ -13,25 +13,12 @@ Compile TFLite Models
 
 This article is an introductory tutorial to deploy TFLite models with Relay.
 
-To get started, Flatbuffers and TFLite package needs to be installed as prerequisites.
-A quick solution is to install Flatbuffers via pip
+To get started, TFLite package needs to be installed as prerequisite.
 
 .. code-block:: bash
 
-    pip install flatbuffers --user
-
-
-To install TFlite packages, you could use our prebuilt wheel:
-
-.. code-block:: bash
-
-    # For python3:
-    wget https://github.com/FrozenGene/tflite/releases/download/v1.13.1/tflite-1.13.1-py3-none-any.whl
-    pip3 install -U tflite-1.13.1-py3-none-any.whl --user
-
-    # For python2:
-    wget https://github.com/FrozenGene/tflite/releases/download/v1.13.1/tflite-1.13.1-py2-none-any.whl
-    pip install -U tflite-1.13.1-py2-none-any.whl --user
+    # install tflite
+    pip install tflite=2.1.0 --user
 
 
 or you could generate TFLite package yourself. The steps are the following:
diff --git a/docs/_sources/tutorials/frontend/sg_execution_times.rst.txt b/docs/_sources/tutorials/frontend/sg_execution_times.rst.txt
index f9340b3..0297379 100644
--- a/docs/_sources/tutorials/frontend/sg_execution_times.rst.txt
+++ b/docs/_sources/tutorials/frontend/sg_execution_times.rst.txt
@@ -5,21 +5,21 @@
 
 Computation times
 =================
-**03:14.518** total execution time for **tutorials_frontend** files:
+**03:19.145** total execution time for **tutorials_frontend** files:
 
-- **00:32.295**: :ref:`sphx_glr_tutorials_frontend_deploy_prequantized.py` (``deploy_prequantized.py``)
-- **00:26.674**: :ref:`sphx_glr_tutorials_frontend_deploy_ssd_gluoncv.py` (``deploy_ssd_gluoncv.py``)
-- **00:19.755**: :ref:`sphx_glr_tutorials_frontend_from_tflite.py` (``from_tflite.py``)
-- **00:17.823**: :ref:`sphx_glr_tutorials_frontend_deploy_quantized.py` (``deploy_quantized.py``)
-- **00:17.595**: :ref:`sphx_glr_tutorials_frontend_from_tensorflow.py` (``from_tensorflow.py``)
-- **00:15.859**: :ref:`sphx_glr_tutorials_frontend_from_darknet.py` (``from_darknet.py``)
-- **00:11.209**: :ref:`sphx_glr_tutorials_frontend_from_pytorch.py` (``from_pytorch.py``)
-- **00:11.079**: :ref:`sphx_glr_tutorials_frontend_from_caffe2.py` (``from_caffe2.py``)
-- **00:08.838**: :ref:`sphx_glr_tutorials_frontend_deploy_model_on_android.py` (``deploy_model_on_android.py``)
-- **00:08.215**: :ref:`sphx_glr_tutorials_frontend_from_keras.py` (``from_keras.py``)
-- **00:06.489**: :ref:`sphx_glr_tutorials_frontend_deploy_model_on_rasp.py` (``deploy_model_on_rasp.py``)
-- **00:05.913**: :ref:`sphx_glr_tutorials_frontend_from_coreml.py` (``from_coreml.py``)
-- **00:04.606**: :ref:`sphx_glr_tutorials_frontend_from_mxnet.py` (``from_mxnet.py``)
-- **00:04.539**: :ref:`sphx_glr_tutorials_frontend_build_gcn.py` (``build_gcn.py``)
-- **00:02.177**: :ref:`sphx_glr_tutorials_frontend_using_external_lib.py` (``using_external_lib.py``)
-- **00:01.452**: :ref:`sphx_glr_tutorials_frontend_from_onnx.py` (``from_onnx.py``)
+- **00:32.751**: :ref:`sphx_glr_tutorials_frontend_deploy_prequantized.py` (``deploy_prequantized.py``)
+- **00:24.609**: :ref:`sphx_glr_tutorials_frontend_deploy_ssd_gluoncv.py` (``deploy_ssd_gluoncv.py``)
+- **00:20.814**: :ref:`sphx_glr_tutorials_frontend_deploy_quantized.py` (``deploy_quantized.py``)
+- **00:19.846**: :ref:`sphx_glr_tutorials_frontend_from_tflite.py` (``from_tflite.py``)
+- **00:17.846**: :ref:`sphx_glr_tutorials_frontend_from_tensorflow.py` (``from_tensorflow.py``)
+- **00:17.137**: :ref:`sphx_glr_tutorials_frontend_from_darknet.py` (``from_darknet.py``)
+- **00:12.561**: :ref:`sphx_glr_tutorials_frontend_from_caffe2.py` (``from_caffe2.py``)
+- **00:11.077**: :ref:`sphx_glr_tutorials_frontend_from_pytorch.py` (``from_pytorch.py``)
+- **00:09.115**: :ref:`sphx_glr_tutorials_frontend_deploy_model_on_android.py` (``deploy_model_on_android.py``)
+- **00:08.130**: :ref:`sphx_glr_tutorials_frontend_from_keras.py` (``from_keras.py``)
+- **00:06.590**: :ref:`sphx_glr_tutorials_frontend_deploy_model_on_rasp.py` (``deploy_model_on_rasp.py``)
+- **00:05.757**: :ref:`sphx_glr_tutorials_frontend_from_coreml.py` (``from_coreml.py``)
+- **00:04.672**: :ref:`sphx_glr_tutorials_frontend_build_gcn.py` (``build_gcn.py``)
+- **00:04.671**: :ref:`sphx_glr_tutorials_frontend_from_mxnet.py` (``from_mxnet.py``)
+- **00:02.045**: :ref:`sphx_glr_tutorials_frontend_using_external_lib.py` (``using_external_lib.py``)
+- **00:01.523**: :ref:`sphx_glr_tutorials_frontend_from_onnx.py` (``from_onnx.py``)
diff --git a/docs/_sources/tutorials/language/reduction.rst.txt b/docs/_sources/tutorials/language/reduction.rst.txt
index cc46cef..3ca10a3 100644
--- a/docs/_sources/tutorials/language/reduction.rst.txt
+++ b/docs/_sources/tutorials/language/reduction.rst.txt
@@ -92,16 +92,20 @@ Before doing anything, let us print out the IR code of default schedule.
 
  .. code-block:: none
 
-    PrimFunc([A, B]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      for (i, 0, n) {
-        B[(i*stride)] = 0f
-        for (k, 0, m) {
-          B[(i*stride)] = (B[(i*stride)] + A[((i*stride) + (k*stride))])
+    primfn(A_1: handle, B_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {A: Buffer(A_2: handle, float32, [n: int32, m: int32], [stride: int32, stride_1: int32], type="auto"),
+                 B: Buffer(B_2: handle, float32, [n], [stride_2: int32], type="auto")}
+      buffer_map = {B_1: B, A_1: A} {
+      for (i: int32, 0, n) {
+        B_2[(i*stride_2)] = 0f32
+        for (k: int32, 0, m) {
+          B_2[(i*stride_2)] = ((float32*)B_2[(i*stride_2)]) + (float32*)A_2[((i*stride) + (k*stride_1))]))
         }
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -129,17 +133,21 @@ axis by different factors. The result is a nested reduction.
 
  .. code-block:: none
 
-    PrimFunc([A, B]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      for (i.outer, 0, floordiv((n + 31), 32)) {
-        for (i.inner, 0, 32) {
-          if (likely((((i.outer*32) + i.inner) < n))) {
-            B[(((i.outer*32) + i.inner)*stride)] = 0f
+    primfn(A_1: handle, B_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {A: Buffer(A_2: handle, float32, [n: int32, m: int32], [stride: int32, stride_1: int32], type="auto"),
+                 B: Buffer(B_2: handle, float32, [n], [stride_2: int32], type="auto")}
+      buffer_map = {B_1: B, A_1: A} {
+      for (i.outer: int32, 0, floordiv((n + 31), 32)) {
+        for (i.inner: int32, 0, 32) {
+          if @likely((((i.outer*32) + i.inner) < n), dtype=bool, type="pure_intrin", index=0) {
+            B_2[(((i.outer*32) + i.inner)*stride_2)] = 0f32
           }
-          for (k.outer, 0, floordiv((m + 15), 16)) {
-            for (k.inner, 0, 16) {
-              if (likely((((i.outer*32) + i.inner) < n))) {
-                if (likely((((k.outer*16) + k.inner) < m))) {
-                  B[(((i.outer*32) + i.inner)*stride)] = (B[(((i.outer*32) + i.inner)*stride)] + A[((((i.outer*32) + i.inner)*stride) + (((k.outer*16) + k.inner)*stride))])
+          for (k.outer: int32, 0, floordiv((m + 15), 16)) {
+            for (k.inner: int32, 0, 16) {
+              if @likely((((i.outer*32) + i.inner) < n), dtype=bool, type="pure_intrin", index=0) {
+                if @likely((((k.outer*16) + k.inner) < m), dtype=bool, type="pure_intrin", index=0) {
+                  B_2[(((i.outer*32) + i.inner)*stride_2)] = ((float32*)B_2[(((i.outer*32) + i.inner)*stride_2)]) + (float32*)A_2[((((i.outer*32) + i.inner)*stride) + (((k.outer*16) + k.inner)*stride_1))]))
                 }
               }
             }
@@ -148,7 +156,7 @@ axis by different factors. The result is a nested reduction.
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -171,24 +179,29 @@ If we are building a GPU kernel, we can bind the rows of B to GPU threads.
 
  .. code-block:: none
 
-    PrimFunc([A, B]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      // attr [iter_var(blockIdx.x, , blockIdx.x)] thread_extent = floordiv((n + 31), 32)
-      // attr [iter_var(threadIdx.x, , threadIdx.x)] thread_extent = 32
-      if (likely((((blockIdx.x*32) + threadIdx.x) < n))) {
-        B[(((blockIdx.x*32) + threadIdx.x)*stride)] = 0f
-      }
-      for (k.outer, 0, floordiv((m + 15), 16)) {
-        for (k.inner, 0, 16) {
-          if (likely((((blockIdx.x*32) + threadIdx.x) < n))) {
-            if (likely((((k.outer*16) + k.inner) < m))) {
-              B[(((blockIdx.x*32) + threadIdx.x)*stride)] = (B[(((blockIdx.x*32) + threadIdx.x)*stride)] + A[((((blockIdx.x*32) + threadIdx.x)*stride) + (((k.outer*16) + k.inner)*stride))])
+    primfn(A_1: handle, B_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {A: Buffer(A_2: handle, float32, [n: int32, m: int32], [stride: int32, stride_1: int32], type="auto"),
+                 B: Buffer(B_2: handle, float32, [n], [stride_2: int32], type="auto")}
+      buffer_map = {B_1: B, A_1: A} {
+      attr [IterVar(blockIdx.x: int32, (nullptr), "ThreadIndex", "blockIdx.x")] "thread_extent" = floordiv((n + 31), 32);
+      attr [IterVar(threadIdx.x: int32, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 32 {
+        if @likely((((blockIdx.x*32) + threadIdx.x) < n), dtype=bool, type="pure_intrin", index=0) {
+          B_2[(((blockIdx.x*32) + threadIdx.x)*stride_2)] = 0f32
+        }
+        for (k.outer: int32, 0, floordiv((m + 15), 16)) {
+          for (k.inner: int32, 0, 16) {
+            if @likely((((blockIdx.x*32) + threadIdx.x) < n), dtype=bool, type="pure_intrin", index=0) {
+              if @likely((((k.outer*16) + k.inner) < m), dtype=bool, type="pure_intrin", index=0) {
+                B_2[(((blockIdx.x*32) + threadIdx.x)*stride_2)] = ((float32*)B_2[(((blockIdx.x*32) + threadIdx.x)*stride_2)]) + (float32*)A_2[((((blockIdx.x*32) + threadIdx.x)*stride) + (((k.outer*16) + k.inner)*stride_1))]))
+              }
             }
           }
         }
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -222,28 +235,33 @@ result B.rf. The factored dimension becomes the first dimension of B.rf.
 
  .. code-block:: none
 
-    PrimFunc([A, B]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      // attr [B.rf] storage_scope = "global"
-      allocate B.rf[float32 * (n*16)]
-      for (k.inner, 0, 16) {
-        for (i, 0, n) {
-          B.rf[((k.inner*n) + i)] = 0f
-          for (k.outer, 0, floordiv((m + 15), 16)) {
-            if (likely((((k.outer*16) + k.inner) < m))) {
-              B.rf[((k.inner*n) + i)] = (B.rf[((k.inner*n) + i)] + A[((i*stride) + (((k.outer*16) + k.inner)*stride))])
+    primfn(A_1: handle, B_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {A: Buffer(A_2: handle, float32, [n: int32, m: int32], [stride: int32, stride_1: int32], type="auto"),
+                 B: Buffer(B_2: handle, float32, [n], [stride_2: int32], type="auto")}
+      buffer_map = {B_1: B, A_1: A} {
+      attr [B.rf: handle] "storage_scope" = "global";
+      allocate(B.rf, float32, [(n*16)]) {
+        for (k.inner: int32, 0, 16) {
+          for (i: int32, 0, n) {
+            B.rf[((k.inner*n) + i)] = 0f32
+            for (k.outer: int32, 0, floordiv((m + 15), 16)) {
+              if @likely((((k.outer*16) + k.inner) < m), dtype=bool, type="pure_intrin", index=0) {
+                B.rf[((k.inner*n) + i)] = ((float32*)B.rf[((k.inner*n) + i)]) + (float32*)A_2[((i*stride) + (((k.outer*16) + k.inner)*stride_1))]))
+              }
             }
           }
         }
-      }
-      for (ax0, 0, n) {
-        B[(ax0*stride)] = 0f
-        for (k.inner.v, 0, 16) {
-          B[(ax0*stride)] = (B[(ax0*stride)] + B.rf[((k.inner.v*n) + ax0)])
+        for (ax0: int32, 0, n) {
+          B_2[(ax0*stride_2)] = 0f32
+          for (k.inner.v: int32, 0, 16) {
+            B_2[(ax0*stride_2)] = ((float32*)B_2[(ax0*stride_2)]) + (float32*)B.rf[((k.inner.v*n) + ax0)]))
+          }
         }
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -323,7 +341,7 @@ columns by threadIdx.x and finally do a cross thread reduction over threadIdx.x
         }
       }
       __syncthreads();
-      ((volatile float*)red_buf0)[(((((int)threadIdx.y) * 16) + ((int)threadIdx.x)))] = ((((((int)blockIdx.x) * 32) + ((int)threadIdx.y)) < n) ? B_rf[(0)] : 0.000000e+00f);
+      ((volatile float*)red_buf0)[(((((int)threadIdx.y) * 16) + ((int)threadIdx.x)))] = B_rf[(0)];
       __syncthreads();
       if (((int)threadIdx.x) < 8) {
         ((volatile float*)red_buf0)[(((((int)threadIdx.y) * 16) + ((int)threadIdx.x)))] = (((volatile float*)red_buf0)[(((((int)threadIdx.y) * 16) + ((int)threadIdx.x)))] + ((volatile float*)red_buf0)[((((((int)threadIdx.y) * 16) + ((int)threadIdx.x)) + 8))]);
@@ -332,10 +350,8 @@ columns by threadIdx.x and finally do a cross thread reduction over threadIdx.x
         ((volatile float*)red_buf0)[(((((int)threadIdx.y) * 16) + ((int)threadIdx.x)))] = (((volatile float*)red_buf0)[(((((int)threadIdx.y) * 16) + ((int)threadIdx.x)))] + ((volatile float*)red_buf0)[((((((int)threadIdx.y) * 16) + ((int)threadIdx.x)) + 1))]);
       }
       __syncthreads();
-      if (((((int)blockIdx.x) * 32) + ((int)threadIdx.y)) < n) {
-        if (((int)threadIdx.x) == 0) {
-          ((float*)B)[((((((int)blockIdx.x) * 32) + ((int)threadIdx.y)) * stride2))] = ((volatile float*)red_buf0)[((((int)threadIdx.y) * 16))];
-        }
+      if (((int)threadIdx.x) == 0) {
+        ((float*)B)[((((((int)blockIdx.x) * 32) + ((int)threadIdx.y)) * stride2))] = ((volatile float*)red_buf0)[((((int)threadIdx.y) * 16))];
       }
     }
 
@@ -394,20 +410,25 @@ Here is an example for 2D convolution with filter size = [3, 3] and strides = [1
 
  .. code-block:: none
 
-    PrimFunc([Input, Filter, Output]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      for (i, 0, (n - 2)) {
-        for (j, 0, (n - 2)) {
-          Output[((i*(n - 2)) + j)] = 0f
-          for (di, 0, 3) {
-            for (dj, 0, 3) {
-              Output[((i*(n - 2)) + j)] = (Output[((i*(n - 2)) + j)] + (Input[(((i + di)*stride) + ((j + dj)*stride))]*Filter[((di*3) + dj)]))
+    primfn(Input_1: handle, Filter_1: handle, Output_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {Input: Buffer(Input_2: handle, float32, [n: int32, n], [stride: int32, stride_1: int32], type="auto"),
+                 Filter: Buffer(Filter_2: handle, float32, [3, 3], []),
+                 Output: Buffer(Output_2: handle, float32, [(n - 2), (n - 2)], [])}
+      buffer_map = {Output_1: Output, Filter_1: Filter, Input_1: Input} {
+      for (i: int32, 0, (n - 2)) {
+        for (j: int32, 0, (n - 2)) {
+          Output_2[((i*(n - 2)) + j)] = 0f32
+          for (di: int32, 0, 3) {
+            for (dj: int32, 0, 3) {
+              Output_2[((i*(n - 2)) + j)] = ((float32*)Output_2[((i*(n - 2)) + j)]) + ((float32*)Input_2[(((i + di)*stride) + ((j + dj)*stride_1))])*(float32*)Filter_2[((di*3) + dj)])))
             }
           }
         }
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
diff --git a/docs/_sources/tutorials/language/scan.rst.txt b/docs/_sources/tutorials/language/scan.rst.txt
index d174f71..8e08ebc 100644
--- a/docs/_sources/tutorials/language/scan.rst.txt
+++ b/docs/_sources/tutorials/language/scan.rst.txt
@@ -98,22 +98,26 @@ To split on the time iteration, user can schedule on scan_op.scan_axis instead.
 
  .. code-block:: none
 
-    PrimFunc([X, scan]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      // attr [iter_var(blockIdx.x, , blockIdx.x)] thread_extent = floordiv((n + 255), 256)
-      // attr [iter_var(threadIdx.x, , threadIdx.x)] thread_extent = 256
-      if (likely((((blockIdx.x*256) + threadIdx.x) < n))) {
-        scan[(((blockIdx.x*256) + threadIdx.x)*stride)] = X[(((blockIdx.x*256) + threadIdx.x)*stride)]
+    primfn(X_1: handle, scan_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {X: Buffer(X_2: handle, float32, [m: int32, n: int32], [stride: int32, stride_1: int32], type="auto"),
+                 scan: Buffer(scan_2: handle, float32, [m, n], [stride_2: int32, stride_3: int32], type="auto")}
+      buffer_map = {scan_1: scan, X_1: X} {
+      attr [IterVar(blockIdx.x: int32, (nullptr), "ThreadIndex", "blockIdx.x")] "thread_extent" = floordiv((n + 255), 256);
+      attr [IterVar(threadIdx.x: int32, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 256;
+      if @likely((((blockIdx.x*256) + threadIdx.x) < n), dtype=bool, type="pure_intrin", index=0) {
+        scan_2[(((blockIdx.x*256) + threadIdx.x)*stride_3)] = (float32*)X_2[(((blockIdx.x*256) + threadIdx.x)*stride_1)])
       }
-      for (scan.idx, 0, (m - 1)) {
-        // attr [iter_var(blockIdx.x, , blockIdx.x)] thread_extent = floordiv((n + 255), 256)
-        // attr [iter_var(threadIdx.x, , threadIdx.x)] thread_extent = 256
-        if (likely((((blockIdx.x*256) + threadIdx.x) < n))) {
-          scan[(((scan.idx + 1)*stride) + (((blockIdx.x*256) + threadIdx.x)*stride))] = (scan[((scan.idx*stride) + (((blockIdx.x*256) + threadIdx.x)*stride))] + X[(((scan.idx + 1)*stride) + (((blockIdx.x*256) + threadIdx.x)*stride))])
+      for (scan.idx: int32, 0, (m - 1)) {
+        attr [IterVar(blockIdx.x, (nullptr), "ThreadIndex", "blockIdx.x")] "thread_extent" = floordiv((n + 255), 256);
+        attr [IterVar(threadIdx.x, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 256;
+        if @likely((((blockIdx.x*256) + threadIdx.x) < n), dtype=bool, type="pure_intrin", index=0) {
+          scan_2[(((scan.idx + 1)*stride_2) + (((blockIdx.x*256) + threadIdx.x)*stride_3))] = ((float32*)scan_2[((scan.idx*stride_2) + (((blockIdx.x*256) + threadIdx.x)*stride_3))]) + (float32*)X_2[(((scan.idx + 1)*stride) + (((blockIdx.x*256) + threadIdx.x)*stride_1))]))
         }
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -193,29 +197,34 @@ the body of scan to be compute_at locations outside the scan loop.
 
  .. code-block:: none
 
-    PrimFunc([X, scan]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      // attr [s1] storage_scope = "global"
-      allocate s1[float32 * 32]
-      for (i, 0, n) {
-        scan[(i*stride)] = X[(i*stride)]
-      }
-      for (scan.idx, 0, (m - 1)) {
-        for (i.outer, 0, floordiv((n + 31), 32)) {
-          for (i, 0, 32) {
-            if (likely((((i.outer*32) + i) < n))) {
-              s1[i] = (scan[((scan.idx*stride) + (((i.outer*32) + i)*stride))]*2f)
+    primfn(X_1: handle, scan_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {X: Buffer(X_2: handle, float32, [m: int32, n: int32], [stride: int32, stride_1: int32], type="auto"),
+                 scan: Buffer(scan_2: handle, float32, [m, n], [stride_2: int32, stride_3: int32], type="auto")}
+      buffer_map = {scan_1: scan, X_1: X} {
+      attr [s1: handle] "storage_scope" = "global";
+      allocate(s1, float32, [32]) {
+        for (i: int32, 0, n) {
+          scan_2[(i*stride_3)] = (float32*)X_2[(i*stride_1)])
+        }
+        for (scan.idx: int32, 0, (m - 1)) {
+          for (i.outer: int32, 0, floordiv((n + 31), 32)) {
+            for (i_1: int32, 0, 32) {
+              if @likely((((i.outer*32) + i_1) < n), dtype=bool, type="pure_intrin", index=0) {
+                s1[i_1] = ((float32*)scan_2[((scan.idx*stride_2) + (((i.outer*32) + i_1)*stride_3))])*2f32)
+              }
             }
-          }
-          for (i.inner, 0, 32) {
-            if (likely((((i.outer*32) + i.inner) < n))) {
-              scan[(((scan.idx + 1)*stride) + (((i.outer*32) + i.inner)*stride))] = (s1[i.inner] + X[(((scan.idx + 1)*stride) + (((i.outer*32) + i.inner)*stride))])
+            for (i.inner: int32, 0, 32) {
+              if @likely((((i.outer*32) + i.inner) < n), dtype=bool, type="pure_intrin", index=0) {
+                scan_2[(((scan.idx + 1)*stride_2) + (((i.outer*32) + i.inner)*stride_3))] = ((float32*)s1[i.inner]) + (float32*)X_2[(((scan.idx + 1)*stride) + (((i.outer*32) + i.inner)*stride_1))]))
+              }
             }
           }
         }
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -255,24 +264,29 @@ The following example demonstrates how we can build recurrence with two states.
 
  .. code-block:: none
 
-    PrimFunc([X, scan.v0, scan.v1]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      for (i, 0, n) {
-        scan.v0[(i*stride)] = X[(i*stride)]
+    primfn(X_1: handle, scan.v0_1: handle, scan.v1_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {scan.v0: Buffer(scan.v0_2: handle, float32, [m: int32, n: int32], [stride: int32, stride_1: int32], type="auto"),
+                 X: Buffer(X_2: handle, float32, [m, n], [stride_2: int32, stride_3: int32], type="auto"),
+                 scan.v1: Buffer(scan.v1_2: handle, float32, [m, l: int32], [stride_4: int32, stride_5: int32], type="auto")}
+      buffer_map = {scan.v1_1: scan.v1, scan.v0_1: scan.v0, X_1: X} {
+      for (i: int32, 0, n) {
+        scan.v0_2[(i*stride_1)] = (float32*)X_2[(i*stride_3)])
       }
-      for (i, 0, l) {
-        scan.v1[(i*stride)] = 0f
+      for (i_1: int32, 0, l) {
+        scan.v1_2[(i_1*stride_5)] = 0f32
       }
-      for (scan.idx, 0, (m - 1)) {
-        for (i, 0, n) {
-          scan.v0[(((scan.idx + 1)*stride) + (i*stride))] = (scan.v0[((scan.idx*stride) + (i*stride))] + X[(((scan.idx + 1)*stride) + (i*stride))])
+      for (scan.idx: int32, 0, (m - 1)) {
+        for (i_2: int32, 0, n) {
+          scan.v0_2[(((scan.idx + 1)*stride) + (i_2*stride_1))] = ((float32*)scan.v0_2[((scan.idx*stride) + (i_2*stride_1))]) + (float32*)X_2[(((scan.idx + 1)*stride_2) + (i_2*stride_3))]))
         }
-        for (i, 0, l) {
-          scan.v1[(((scan.idx + 1)*stride) + (i*stride))] = (scan.v1[((scan.idx*stride) + (i*stride))] + scan.v0[(scan.idx*stride)])
+        for (i_3: int32, 0, l) {
+          scan.v1_2[(((scan.idx + 1)*stride_4) + (i_3*stride_5))] = ((float32*)scan.v1_2[((scan.idx*stride_4) + (i_3*stride_5))]) + (float32*)scan.v0_2[(scan.idx*stride)]))
         }
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
diff --git a/docs/_sources/tutorials/language/schedule_primitives.rst.txt b/docs/_sources/tutorials/language/schedule_primitives.rst.txt
index 6a9e218..c75d8b1 100644
--- a/docs/_sources/tutorials/language/schedule_primitives.rst.txt
+++ b/docs/_sources/tutorials/language/schedule_primitives.rst.txt
@@ -83,15 +83,20 @@ schedule computes tensor in a serial manner in a row-major order.
 
  .. code-block:: none
 
-    PrimFunc([A, B, C]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      for (i, 0, m) {
-        for (j, 0, n) {
-          C[((i*stride) + (j*stride))] = (A[((i*stride) + (j*stride))]*B[((i*stride) + (j*stride))])
+    primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {A: Buffer(A_2: handle, float32, [m: int32, n: int32], [stride: int32, stride_1: int32], type="auto"),
+                 B: Buffer(B_2: handle, float32, [m, n], [stride_2: int32, stride_3: int32], type="auto"),
+                 C: Buffer(C_2: handle, float32, [m, n], [stride_4: int32, stride_5: int32], type="auto")}
+      buffer_map = {C_1: C, B_1: B, A_1: A} {
+      for (i: int32, 0, m) {
+        for (j: int32, 0, n) {
+          C_2[((i*stride_4) + (j*stride_5))] = ((float32*)A_2[((i*stride) + (j*stride_1))])*(float32*)B_2[((i*stride_2) + (j*stride_3))]))
         }
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -124,17 +129,21 @@ split
 
  .. code-block:: none
 
-    PrimFunc([A, B]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      for (i.outer, 0, floordiv((m + 31), 32)) {
-        for (i.inner, 0, 32) {
-          if (likely((((i.outer*32) + i.inner) < m))) {
-            B[(((i.outer*32) + i.inner)*stride)] = (A[(((i.outer*32) + i.inner)*stride)]*2f)
+    primfn(A_1: handle, B_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {A: Buffer(A_2: handle, float32, [m: int32], [stride: int32], type="auto"),
+                 B: Buffer(B_2: handle, float32, [m], [stride_1: int32], type="auto")}
+      buffer_map = {B_1: B, A_1: A} {
+      for (i.outer: int32, 0, floordiv((m + 31), 32)) {
+        for (i.inner: int32, 0, 32) {
+          if @likely((((i.outer*32) + i.inner) < m), dtype=bool, type="pure_intrin", index=0) {
+            B_2[(((i.outer*32) + i.inner)*stride_1)] = ((float32*)A_2[(((i.outer*32) + i.inner)*stride)])*2f32)
           }
         }
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -161,17 +170,21 @@ contrary with :code:`factor`.
 
  .. code-block:: none
 
-    PrimFunc([A, B]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      for (i.outer, 0, 32) {
-        for (i.inner, 0, floordiv((m + 31), 32)) {
-          if (likely(((i.inner + (i.outer*floordiv((m + 31), 32))) < m))) {
-            B[((i.inner + (i.outer*floordiv((m + 31), 32)))*stride)] = A[((i.inner + (i.outer*floordiv((m + 31), 32)))*stride)]
+    primfn(A_1: handle, B_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {A: Buffer(A_2: handle, float32, [m: int32], [stride: int32], type="auto"),
+                 B: Buffer(B_2: handle, float32, [m], [stride_1: int32], type="auto")}
+      buffer_map = {B_1: B, A_1: A} {
+      for (i.outer: int32, 0, 32) {
+        for (i.inner: int32, 0, floordiv((m + 31), 32)) {
+          if @likely(((i.inner + (i.outer*floordiv((m + 31), 32))) < m), dtype=bool, type="pure_intrin", index=0) {
+            B_2[((i.inner + (i.outer*floordiv((m + 31), 32)))*stride_1)] = (float32*)A_2[((i.inner + (i.outer*floordiv((m + 31), 32)))*stride)])
           }
         }
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -200,14 +213,18 @@ axises.
 
  .. code-block:: none
 
-    PrimFunc([A, B]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      for (i.outer, 0, floordiv((m + 9), 10)) {
-        for (j.outer, 0, floordiv((n + 4), 5)) {
-          for (i.inner, 0, 10) {
-            for (j.inner, 0, 5) {
-              if (likely((((i.outer*10) + i.inner) < m))) {
-                if (likely((((j.outer*5) + j.inner) < n))) {
-                  B[((((i.outer*10) + i.inner)*stride) + (((j.outer*5) + j.inner)*stride))] = A[((((i.outer*10) + i.inner)*stride) + (((j.outer*5) + j.inner)*stride))]
+    primfn(A_1: handle, B_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {A: Buffer(A_2: handle, float32, [m: int32, n: int32], [stride: int32, stride_1: int32], type="auto"),
+                 B: Buffer(B_2: handle, float32, [m, n], [stride_2: int32, stride_3: int32], type="auto")}
+      buffer_map = {B_1: B, A_1: A} {
+      for (i.outer: int32, 0, floordiv((m + 9), 10)) {
+        for (j.outer: int32, 0, floordiv((n + 4), 5)) {
+          for (i.inner: int32, 0, 10) {
+            for (j.inner: int32, 0, 5) {
+              if @likely((((i.outer*10) + i.inner) < m), dtype=bool, type="pure_intrin", index=0) {
+                if @likely((((j.outer*5) + j.inner) < n), dtype=bool, type="pure_intrin", index=0) {
+                  B_2[((((i.outer*10) + i.inner)*stride_2) + (((j.outer*5) + j.inner)*stride_3))] = (float32*)A_2[((((i.outer*10) + i.inner)*stride) + (((j.outer*5) + j.inner)*stride_1))])
                 }
               }
             }
@@ -216,7 +233,7 @@ axises.
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -247,13 +264,17 @@ fuse
 
  .. code-block:: none
 
-    PrimFunc([A, B]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      for (i.outer, 0, floordiv((m + 9), 10)) {
-        for (j.outer, 0, floordiv((n + 4), 5)) {
-          for (i.inner.j.inner.fused, 0, 50) {
-            if (likely((((i.outer*10) + floordiv(i.inner.j.inner.fused, 5)) < m))) {
-              if (likely((((j.outer*5) + floormod(i.inner.j.inner.fused, 5)) < n))) {
-                B[((((i.outer*10) + floordiv(i.inner.j.inner.fused, 5))*stride) + (((j.outer*5) + floormod(i.inner.j.inner.fused, 5))*stride))] = A[((((i.outer*10) + floordiv(i.inner.j.inner.fused, 5))*stride) + (((j.outer*5) + floormod(i.inner.j.inner.fused, 5))*stride))]
+    primfn(A_1: handle, B_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {A: Buffer(A_2: handle, float32, [m: int32, n: int32], [stride: int32, stride_1: int32], type="auto"),
+                 B: Buffer(B_2: handle, float32, [m, n], [stride_2: int32, stride_3: int32], type="auto")}
+      buffer_map = {B_1: B, A_1: A} {
+      for (i.outer: int32, 0, floordiv((m + 9), 10)) {
+        for (j.outer: int32, 0, floordiv((n + 4), 5)) {
+          for (i.inner.j.inner.fused: int32, 0, 50) {
+            if @likely((((i.outer*10) + floordiv(i.inner.j.inner.fused, 5)) < m), dtype=bool, type="pure_intrin", index=0) {
+              if @likely((((j.outer*5) + floormod(i.inner.j.inner.fused, 5)) < n), dtype=bool, type="pure_intrin", index=0) {
+                B_2[((((i.outer*10) + floordiv(i.inner.j.inner.fused, 5))*stride_2) + (((j.outer*5) + floormod(i.inner.j.inner.fused, 5))*stride_3))] = (float32*)A_2[((((i.outer*10) + floordiv(i.inner.j.inner.fused, 5))*stride) + (((j.outer*5) + floormod(i.inner.j.inner.fused, 5))*stride_1))])
               }
             }
           }
@@ -261,7 +282,7 @@ fuse
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -292,14 +313,18 @@ reorder
 
  .. code-block:: none
 
-    PrimFunc([A, B]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      for (i.inner, 0, 10) {
-        for (j.outer, 0, floordiv((n + 4), 5)) {
-          for (i.outer, 0, floordiv((m + 9), 10)) {
-            for (j.inner, 0, 5) {
-              if (likely((((i.outer*10) + i.inner) < m))) {
-                if (likely((((j.outer*5) + j.inner) < n))) {
-                  B[((((i.outer*10) + i.inner)*stride) + (((j.outer*5) + j.inner)*stride))] = A[((((i.outer*10) + i.inner)*stride) + (((j.outer*5) + j.inner)*stride))]
+    primfn(A_1: handle, B_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {A: Buffer(A_2: handle, float32, [m: int32, n: int32], [stride: int32, stride_1: int32], type="auto"),
+                 B: Buffer(B_2: handle, float32, [m, n], [stride_2: int32, stride_3: int32], type="auto")}
+      buffer_map = {B_1: B, A_1: A} {
+      for (i.inner: int32, 0, 10) {
+        for (j.outer: int32, 0, floordiv((n + 4), 5)) {
+          for (i.outer: int32, 0, floordiv((m + 9), 10)) {
+            for (j.inner: int32, 0, 5) {
+              if @likely((((i.outer*10) + i.inner) < m), dtype=bool, type="pure_intrin", index=0) {
+                if @likely((((j.outer*5) + j.inner) < n), dtype=bool, type="pure_intrin", index=0) {
+                  B_2[((((i.outer*10) + i.inner)*stride_2) + (((j.outer*5) + j.inner)*stride_3))] = (float32*)A_2[((((i.outer*10) + i.inner)*stride) + (((j.outer*5) + j.inner)*stride_1))])
                 }
               }
             }
@@ -308,7 +333,7 @@ reorder
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -339,15 +364,19 @@ in gpu programming.
 
  .. code-block:: none
 
-    PrimFunc([A, B]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      // attr [iter_var(blockIdx.x, , blockIdx.x)] thread_extent = floordiv((n + 63), 64)
-      // attr [iter_var(threadIdx.x, , threadIdx.x)] thread_extent = 64
-      if (likely((((blockIdx.x*64) + threadIdx.x) < n))) {
-        B[(((blockIdx.x*64) + threadIdx.x)*stride)] = (A[(((blockIdx.x*64) + threadIdx.x)*stride)]*2f)
+    primfn(A_1: handle, B_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {A: Buffer(A_2: handle, float32, [n: int32], [stride: int32], type="auto"),
+                 B: Buffer(B_2: handle, float32, [n], [stride_1: int32], type="auto")}
+      buffer_map = {B_1: B, A_1: A} {
+      attr [IterVar(blockIdx.x: int32, (nullptr), "ThreadIndex", "blockIdx.x")] "thread_extent" = floordiv((n + 63), 64);
+      attr [IterVar(threadIdx.x: int32, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 64;
+      if @likely((((blockIdx.x*64) + threadIdx.x) < n), dtype=bool, type="pure_intrin", index=0) {
+        B_2[(((blockIdx.x*64) + threadIdx.x)*stride_1)] = ((float32*)A_2[(((blockIdx.x*64) + threadIdx.x)*stride)])*2f32)
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -376,16 +405,21 @@ tensors at the root separately by default.
 
  .. code-block:: none
 
-    PrimFunc([A, B, C]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      for (i, 0, m) {
-        B[(i*stride)] = (A[(i*stride)] + 1f)
+    primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {A: Buffer(A_2: handle, float32, [m: int32], [stride: int32], type="auto"),
+                 B: Buffer(B_2: handle, float32, [m], [stride_1: int32], type="auto"),
+                 C: Buffer(C_2: handle, float32, [m], [stride_2: int32], type="auto")}
+      buffer_map = {C_1: C, B_1: B, A_1: A} {
+      for (i: int32, 0, m) {
+        B_2[(i*stride_1)] = ((float32*)A_2[(i*stride)]) + 1f32)
       }
-      for (i, 0, m) {
-        C[(i*stride)] = (B[(i*stride)]*2f)
+      for (i_1: int32, 0, m) {
+        C_2[(i_1*stride_2)] = ((float32*)B_2[(i_1*stride_1)])*2f32)
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -413,14 +447,19 @@ of computation of `C`.
 
  .. code-block:: none
 
-    PrimFunc([A, B, C]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      for (i, 0, m) {
-        B[(i*stride)] = (A[(i*stride)] + 1f)
-        C[(i*stride)] = (B[(i*stride)]*2f)
+    primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {A: Buffer(A_2: handle, float32, [m: int32], [stride: int32], type="auto"),
+                 B: Buffer(B_2: handle, float32, [m], [stride_1: int32], type="auto"),
+                 C: Buffer(C_2: handle, float32, [m], [stride_2: int32], type="auto")}
+      buffer_map = {C_1: C, B_1: B, A_1: A} {
+      for (i: int32, 0, m) {
+        B_2[(i*stride_1)] = ((float32*)A_2[(i*stride)]) + 1f32)
+        C_2[(i*stride_2)] = ((float32*)B_2[(i*stride_1)])*2f32)
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -451,13 +490,18 @@ tensor is required.
 
  .. code-block:: none
 
-    PrimFunc([A, B, C]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      for (i, 0, m) {
-        C[(i*stride)] = ((A[(i*stride)] + 1f)*2f)
+    primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {B: Buffer(B_2: handle, float32, [m: int32], [stride: int32], type="auto"),
+                 A: Buffer(A_2: handle, float32, [m], [stride_1: int32], type="auto"),
+                 C: Buffer(C_2: handle, float32, [m], [stride_2: int32], type="auto")}
+      buffer_map = {C_1: C, B_1: B, A_1: A} {
+      for (i: int32, 0, m) {
+        C_2[(i*stride_2)] = (((float32*)A_2[(i*stride_1)]) + 1f32)*2f32)
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -487,16 +531,21 @@ compute_root
 
  .. code-block:: none
 
-    PrimFunc([A, B, C]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      for (i, 0, m) {
-        B[(i*stride)] = (A[(i*stride)] + 1f)
+    primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {A: Buffer(A_2: handle, float32, [m: int32], [stride: int32], type="auto"),
+                 B: Buffer(B_2: handle, float32, [m], [stride_1: int32], type="auto"),
+                 C: Buffer(C_2: handle, float32, [m], [stride_2: int32], type="auto")}
+      buffer_map = {C_1: C, B_1: B, A_1: A} {
+      for (i: int32, 0, m) {
+        B_2[(i*stride_1)] = ((float32*)A_2[(i*stride)]) + 1f32)
       }
-      for (i, 0, m) {
-        C[(i*stride)] = (B[(i*stride)]*2f)
+      for (i_1: int32, 0, m) {
+        C_2[(i_1*stride_2)] = ((float32*)B_2[(i_1*stride_1)])*2f32)
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
diff --git a/docs/_sources/tutorials/language/sg_execution_times.rst.txt b/docs/_sources/tutorials/language/sg_execution_times.rst.txt
index 0e199e0..be4db3e 100644
--- a/docs/_sources/tutorials/language/sg_execution_times.rst.txt
+++ b/docs/_sources/tutorials/language/sg_execution_times.rst.txt
@@ -5,13 +5,13 @@
 
 Computation times
 =================
-**00:03.665** total execution time for **tutorials_language** files:
+**00:03.594** total execution time for **tutorials_language** files:
 
-- **00:01.305**: :ref:`sphx_glr_tutorials_language_intrin_math.py` (``intrin_math.py``)
-- **00:00.670**: :ref:`sphx_glr_tutorials_language_tensorize.py` (``tensorize.py``)
-- **00:00.498**: :ref:`sphx_glr_tutorials_language_reduction.py` (``reduction.py``)
-- **00:00.450**: :ref:`sphx_glr_tutorials_language_scan.py` (``scan.py``)
-- **00:00.287**: :ref:`sphx_glr_tutorials_language_extern_op.py` (``extern_op.py``)
-- **00:00.167**: :ref:`sphx_glr_tutorials_language_tedd.py` (``tedd.py``)
-- **00:00.153**: :ref:`sphx_glr_tutorials_language_schedule_primitives.py` (``schedule_primitives.py``)
-- **00:00.136**: :ref:`sphx_glr_tutorials_language_tuple_inputs.py` (``tuple_inputs.py``)
+- **00:01.251**: :ref:`sphx_glr_tutorials_language_intrin_math.py` (``intrin_math.py``)
+- **00:00.635**: :ref:`sphx_glr_tutorials_language_tensorize.py` (``tensorize.py``)
+- **00:00.489**: :ref:`sphx_glr_tutorials_language_reduction.py` (``reduction.py``)
+- **00:00.470**: :ref:`sphx_glr_tutorials_language_scan.py` (``scan.py``)
+- **00:00.281**: :ref:`sphx_glr_tutorials_language_extern_op.py` (``extern_op.py``)
+- **00:00.172**: :ref:`sphx_glr_tutorials_language_tedd.py` (``tedd.py``)
+- **00:00.156**: :ref:`sphx_glr_tutorials_language_schedule_primitives.py` (``schedule_primitives.py``)
+- **00:00.139**: :ref:`sphx_glr_tutorials_language_tuple_inputs.py` (``tuple_inputs.py``)
diff --git a/docs/_sources/tutorials/language/tensorize.rst.txt b/docs/_sources/tutorials/language/tensorize.rst.txt
index 56661d5..6e94968 100644
--- a/docs/_sources/tutorials/language/tensorize.rst.txt
+++ b/docs/_sources/tutorials/language/tensorize.rst.txt
@@ -69,18 +69,23 @@ The following lines describe the computation :code:`A * B^T` in TVM.
 
  .. code-block:: none
 
-    PrimFunc([A, B, C]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      for (i, 0, 1024) {
-        for (j, 0, 512) {
-          C[((i*512) + j)] = 0f
-          for (k, 0, 64) {
-            C[((i*512) + j)] = (C[((i*512) + j)] + (A[((i*64) + k)]*B[((j*64) + k)]))
+    primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {A: Buffer(A_2: handle, float32, [1024, 64], []),
+                 B: Buffer(B_2: handle, float32, [512, 64], []),
+                 C: Buffer(C_2: handle, float32, [1024, 512], [])}
+      buffer_map = {C_1: C, B_1: B, A_1: A} {
+      for (i: int32, 0, 1024) {
+        for (j: int32, 0, 512) {
+          C_2[((i*512) + j)] = 0f32
+          for (k: int32, 0, 64) {
+            C_2[((i*512) + j)] = ((float32*)C_2[((i*512) + j)]) + ((float32*)A_2[((i*64) + k)])*(float32*)B_2[((j*64) + k)])))
           }
         }
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -113,20 +118,25 @@ Thus we break down the matmul loops to make the innermost loops a (16x64) GEMV.
 
  .. code-block:: none
 
-    PrimFunc([A, B, C]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      for (i, 0, 1024) {
-        for (j.outer, 0, 32) {
-          for (j.inner, 0, 16) {
-            C[(((i*512) + (j.outer*16)) + j.inner)] = 0f
-            for (k, 0, 64) {
-              C[(((i*512) + (j.outer*16)) + j.inner)] = (C[(((i*512) + (j.outer*16)) + j.inner)] + (A[((i*64) + k)]*B[(((j.outer*1024) + (j.inner*64)) + k)]))
+    primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {A: Buffer(A_2: handle, float32, [1024, 64], []),
+                 B: Buffer(B_2: handle, float32, [512, 64], []),
+                 C: Buffer(C_2: handle, float32, [1024, 512], [])}
+      buffer_map = {C_1: C, B_1: B, A_1: A} {
+      for (i: int32, 0, 1024) {
+        for (j.outer: int32, 0, 32) {
+          for (j.inner: int32, 0, 16) {
+            C_2[(((i*512) + (j.outer*16)) + j.inner)] = 0f32
+            for (k: int32, 0, 64) {
+              C_2[(((i*512) + (j.outer*16)) + j.inner)] = ((float32*)C_2[(((i*512) + (j.outer*16)) + j.inner)]) + ((float32*)A_2[((i*64) + k)])*(float32*)B_2[(((j.outer*1024) + (j.inner*64)) + k)])))
             }
           }
         }
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -225,15 +235,20 @@ such placeholder can be put to let TVM automatically bind the inferred value for
 
  .. code-block:: none
 
-    PrimFunc([A, B, C]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      for (i, 0, 1024) {
-        for (j.outer, 0, 32) {
-          gemv_update(tvm_access_ptr(type_annotation(), C, ((i*512) + (j.outer*16)), 16, 2), tvm_access_ptr(type_annotation(), A, (i*64), 64, 1), tvm_access_ptr(type_annotation(), B, (j.outer*1024), 1024, 1), 16, 64, 64)
+    primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {A: Buffer(A_2: handle, float32, [1024, 64], []),
+                 B: Buffer(B_2: handle, float32, [512, 64], []),
+                 C: Buffer(C_2: handle, float32, [1024, 512], [])}
+      buffer_map = {C_1: C, B_1: B, A_1: A} {
+      for (i: int32, 0, 1024) {
+        for (j.outer: int32, 0, 32) {
+          @gemv_update(@tvm_access_ptr(@type_annotation(, dtype=float32, type="pure_intrin", index=0), C_2, ((i*512) + (j.outer*16)), 16, 2, dtype=handle, type="intrin", index=0), @tvm_access_ptr(@type_annotation(, dtype=float32, type="pure_intrin", index=0), A_2, (i*64), 64, 1, dtype=handle, type="intrin", index=0), @tvm_access_ptr(@type_annotation(, dtype=float32, type="pure_intrin", index=0), B_2, (j.outer*1024), 1024, 1, dtype=handle, type="intrin", index=0), 16, 64, 64, dtype=int32, [...]
         }
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -290,16 +305,112 @@ The importing needs to happen before the tensorized GEMV being executed.
 
  .. code-block:: none
 
-    PrimFunc([A, B, C]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      // attr [iter_var(i, )] pragma_import_llvm = "; ModuleID = '/tmp/tmpwfn404_d/input0.cc'\nsource_filename = \"/tmp/tmpwfn404_d/input0.cc\"\ntarget datalayout = \"e-m:e-i64:64-f80:128-n8:16:32:64-S128\"\ntarget triple = \"x86_64-pc-linux-gnu\"\n\n; Function Attrs: noinline nounwind optnone uwtable\ndefine dso_local i32 @gemv_update(float*, float*, float*, i32, i32, i32) #0 {\n  %7 = alloca float*, align 8\n  %8 = alloca float*, align 8\n  %9 = alloca float*, align 8\n  %10 = alloca i [...]
+    primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {A: Buffer(A_2: handle, float32, [1024, 64], []),
+                 B: Buffer(B_2: handle, float32, [512, 64], []),
+                 C: Buffer(C_2: handle, float32, [1024, 512], [])}
+      buffer_map = {C_1: C, B_1: B, A_1: A} {
+      attr [IterVar(i: int32, (nullptr), "DataPar", "")] "pragma_import_llvm" = "; ModuleID = '/tmp/tmpx2996fnn/input0.cc'
+    source_filename = "/tmp/tmpx2996fnn/input0.cc"
+    target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
+    target triple = "x86_64-pc-linux-gnu"
+
+    ; Function Attrs: noinline nounwind optnone uwtable
+    define dso_local i32 @gemv_update(float*, float*, float*, i32, i32, i32) #0 {
+      %7 = alloca float*, align 8
+      %8 = alloca float*, align 8
+      %9 = alloca float*, align 8
+      %10 = alloca i32, align 4
+      %11 = alloca i32, align 4
+      %12 = alloca i32, align 4
+      %13 = alloca i32, align 4
+      %14 = alloca i32, align 4
+      store float* %0, float** %7, align 8
+      store float* %1, float** %8, align 8
+      store float* %2, float** %9, align 8
+      store i32 %3, i32* %10, align 4
+      store i32 %4, i32* %11, align 4
+      store i32 %5, i32* %12, align 4
+      store i32 0, i32* %13, align 4
+      br label %15
+
+    15:                                               ; preds = %50, %6
+      %16 = load i32, i32* %13, align 4
+      %17 = load i32, i32* %10, align 4
+      %18 = icmp slt i32 %16, %17
+      br i1 %18, label %19, label %53
+
+    19:                                               ; preds = %15
+      store i32 0, i32* %14, align 4
+      br label %20
+
+    20:                                               ; preds = %46, %19
+      %21 = load i32, i32* %14, align 4
+      %22 = load i32, i32* %11, align 4
+      %23 = icmp slt i32 %21, %22
+      br i1 %23, label %24, label %49
+
+    24:                                               ; preds = %20
+      %25 = load float*, float** %8, align 8
+      %26 = load i32, i32* %14, align 4
+      %27 = sext i32 %26 to i64
+      %28 = getelementptr inbounds float, float* %25, i64 %27
+      %29 = load float, float* %28, align 4
+      %30 = load float*, float** %9, align 8
+      %31 = load i32, i32* %13, align 4
+      %32 = load i32, i32* %12, align 4
+      %33 = mul nsw i32 %31, %32
+      %34 = load i32, i32* %14, align 4
+      %35 = add nsw i32 %33, %34
+      %36 = sext i32 %35 to i64
+      %37 = getelementptr inbounds float, float* %30, i64 %36
+      %38 = load float, float* %37, align 4
+      %39 = fmul float %29, %38
+      %40 = load float*, float** %7, align 8
+      %41 = load i32, i32* %13, align 4
+      %42 = sext i32 %41 to i64
+      %43 = getelementptr inbounds float, float* %40, i64 %42
+      %44 = load float, float* %43, align 4
+      %45 = fadd float %44, %39
+      store float %45, float* %43, align 4
+      br label %46
+
+    46:                                               ; preds = %24
+      %47 = load i32, i32* %14, align 4
+      %48 = add nsw i32 %47, 1
+      store i32 %48, i32* %14, align 4
+      br label %20
+
+    49:                                               ; preds = %20
+      br label %50
+
+    50:                                               ; preds = %49
+      %51 = load i32, i32* %13, align 4
+      %52 = add nsw i32 %51, 1
+      store i32 %52, i32* %13, align 4
+      br label %15
+
+    53:                                               ; preds = %15
+      ret i32 0
+    }
+
+    attributes #0 = { noinline nounwind optnone uwtable "correctly-rounded-divide-sqrt-fp-math"="false" "disable-tail-calls"="false" "less-precise-fpmad"="false" "min-legal-vector-width"="0" "no-frame-pointer-elim"="true" "no-frame-pointer-elim-non-leaf" "no-infs-fp-math"="false" "no-jump-tables"="false" "no-nans-fp-math"="false" "no-signed-zeros-fp-math"="false" "no-trapping-math"="false" "stack-protector-buffer-size"="8" "target-cpu"="x86-64" "target-features"="+cx8,+fxsr,+mmx,+sse,+ss [...]
+
+    !llvm.module.flags = !{!0}
+    !llvm.ident = !{!1}
+
+    !0 = !{i32 1, !"wchar_size", i32 4}
+    !1 = !{!"clang version 9.0.1-+20191211110317+c1a0a213378-1~exp1~20191211221711.104 "}
+    ";
       for (i, 0, 1024) {
-        for (j.outer, 0, 32) {
-          gemv_update(tvm_access_ptr(type_annotation(), C, ((i*512) + (j.outer*16)), 16, 2), tvm_access_ptr(type_annotation(), A, (i*64), 64, 1), tvm_access_ptr(type_annotation(), B, (j.outer*1024), 1024, 1), 16, 64, 64)
+        for (j.outer: int32, 0, 32) {
+          @gemv_update(@tvm_access_ptr(@type_annotation(, dtype=float32, type="pure_intrin", index=0), C_2, ((i*512) + (j.outer*16)), 16, 2, dtype=handle, type="intrin", index=0), @tvm_access_ptr(@type_annotation(, dtype=float32, type="pure_intrin", index=0), A_2, (i*64), 64, 1, dtype=handle, type="intrin", index=0), @tvm_access_ptr(@type_annotation(, dtype=float32, type="pure_intrin", index=0), B_2, (j.outer*1024), 1024, 1, dtype=handle, type="intrin", index=0), 16, 64, 64, dtype=int32, [...]
         }
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
diff --git a/docs/_sources/tutorials/language/tuple_inputs.rst.txt b/docs/_sources/tutorials/language/tuple_inputs.rst.txt
index f26def8..2783f86 100644
--- a/docs/_sources/tutorials/language/tuple_inputs.rst.txt
+++ b/docs/_sources/tutorials/language/tuple_inputs.rst.txt
@@ -62,16 +62,22 @@ together in the next schedule procedure.
 
  .. code-block:: none
 
-    PrimFunc([A0, A1, B.v0, B.v1]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      for (i, 0, m) {
-        for (j, 0, n) {
-          B.v0[((i*stride) + (j*stride))] = (A0[((i*stride) + (j*stride))] + 2f)
-          B.v1[((i*stride) + (j*stride))] = (A1[((i*stride) + (j*stride))]*3f)
+    primfn(A0_1: handle, A1_1: handle, B.v0_1: handle, B.v1_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {A1: Buffer(A1_2: handle, float32, [m: int32, n: int32], [stride: int32, stride_1: int32], type="auto"),
+                 B.v0: Buffer(B.v0_2: handle, float32, [m, n], [stride_2: int32, stride_3: int32], type="auto"),
+                 A0: Buffer(A0_2: handle, float32, [m, n], [stride_4: int32, stride_5: int32], type="auto"),
+                 B.v1: Buffer(B.v1_2: handle, float32, [m, n], [stride_6: int32, stride_7: int32], type="auto")}
+      buffer_map = {B.v1_1: B.v1, B.v0_1: B.v0, A1_1: A1, A0_1: A0} {
+      for (i: int32, 0, m) {
+        for (j: int32, 0, n) {
+          B.v0_2[((i*stride_2) + (j*stride_3))] = ((float32*)A0_2[((i*stride_4) + (j*stride_5))]) + 2f32)
+          B.v1_2[((i*stride_6) + (j*stride_7))] = ((float32*)A1_2[((i*stride) + (j*stride_1))])*3f32)
         }
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -125,18 +131,24 @@ with :py:func:`te.comm_reducer` as below:
 
  .. code-block:: none
 
-    PrimFunc([idx, val, T.v0, T.v1]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      for (i, 0, m) {
-        T.v0[(i*stride)] = -1
-        T.v1[(i*stride)] = -2147483648
-        for (k, 0, n) {
-          T.v0[(i*stride)] = tvm_if_then_else((val[((i*stride) + (k*stride))] <= T.v1[(i*stride)]), T.v0[(i*stride)], idx[((i*stride) + (k*stride))])
-          T.v1[(i*stride)] = tvm_if_then_else((val[((i*stride) + (k*stride))] <= T.v1[(i*stride)]), T.v1[(i*stride)], val[((i*stride) + (k*stride))])
+    primfn(idx_1: handle, val_1: handle, T.v0_1: handle, T.v1_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {idx: Buffer(idx_2: handle, int32, [m: int32, n: int32], [stride: int32, stride_1: int32], type="auto"),
+                 val: Buffer(val_2: handle, int32, [m, n], [stride_2: int32, stride_3: int32], type="auto"),
+                 T.v1: Buffer(T.v1_2: handle, int32, [m], [stride_4: int32], type="auto"),
+                 T.v0: Buffer(T.v0_2: handle, int32, [m], [stride_5: int32], type="auto")}
+      buffer_map = {T.v0_1: T.v0, T.v1_1: T.v1, val_1: val, idx_1: idx} {
+      for (i: int32, 0, m) {
+        T.v0_2[(i*stride_5)] = -1
+        T.v1_2[(i*stride_4)] = -2147483648
+        for (k: int32, 0, n) {
+          T.v0_2[(i*stride_5)] = @tvm_if_then_else(((int32*)val_2[((i*stride_2) + (k*stride_3))]) <= (int32*)T.v1_2[(i*stride_4)])), (int32*)T.v0_2[(i*stride_5)]), (int32*)idx_2[((i*stride) + (k*stride_1))]), dtype=int32, type="intrin", index=0)
+          T.v1_2[(i*stride_4)] = @tvm_if_then_else(((int32*)val_2[((i*stride_2) + (k*stride_3))]) <= (int32*)T.v1_2[(i*stride_4)])), (int32*)T.v1_2[(i*stride_4)]), (int32*)val_2[((i*stride_2) + (k*stride_3))]), dtype=int32, type="intrin", index=0)
         }
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -177,23 +189,28 @@ in terms of operation.
 
  .. code-block:: none
 
-    PrimFunc([A0, A1, C]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      // attr [B.v0] storage_scope = "global"
-      allocate B.v0[float32 * n]
-      // attr [B.v1] storage_scope = "global"
-      allocate B.v1[float32 * n]
-      for (i, 0, m) {
-        for (j, 0, n) {
-          B.v0[j] = (A0[((i*stride) + (j*stride))] + 2f)
-          B.v1[j] = (A0[((i*stride) + (j*stride))]*3f)
+    primfn(A0_1: handle, A1_1: handle, C_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {A0: Buffer(A0_2: handle, float32, [m: int32, n: int32], [stride: int32, stride_1: int32], type="auto"),
+                 C: Buffer(C_2: handle, float32, [m, n], [stride_2: int32, stride_3: int32], type="auto"),
+                 A1: Buffer(A1_2: handle, float32, [m, n], [stride_4: int32, stride_5: int32], type="auto")}
+      buffer_map = {A1_1: A1, C_1: C, A0_1: A0} {
+      attr [B.v0: handle] "storage_scope" = "global";
+      allocate(B.v0, float32, [n]);
+      attr [B.v1: handle] "storage_scope" = "global";
+      allocate(B.v1, float32, [n]);
+      for (i: int32, 0, m) {
+        for (j: int32, 0, n) {
+          B.v0[j] = ((float32*)A0_2[((i*stride) + (j*stride_1))]) + 2f32)
+          B.v1[j] = ((float32*)A0_2[((i*stride) + (j*stride_1))])*3f32)
         }
-        for (j, 0, n) {
-          C[((i*stride) + (j*stride))] = (A1[((i*stride) + (j*stride))] + B.v0[j])
+        for (j_1: int32, 0, n) {
+          C_2[((i*stride_2) + (j_1*stride_3))] = ((float32*)A1_2[((i*stride_4) + (j_1*stride_5))]) + (float32*)B.v0[j_1]))
         }
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
diff --git a/docs/_sources/tutorials/optimize/opt_conv_cuda.rst.txt b/docs/_sources/tutorials/optimize/opt_conv_cuda.rst.txt
index 3203b0f..b651df7 100644
--- a/docs/_sources/tutorials/optimize/opt_conv_cuda.rst.txt
+++ b/docs/_sources/tutorials/optimize/opt_conv_cuda.rst.txt
@@ -293,7 +293,7 @@ latency of convolution.
 
  .. code-block:: none
 
-    Convolution: 19.551412 ms
+    Convolution: 19.545791 ms
 
 
 
diff --git a/docs/_sources/tutorials/optimize/opt_conv_tensorcore.rst.txt b/docs/_sources/tutorials/optimize/opt_conv_tensorcore.rst.txt
index c3a71a0..1b1232a 100644
--- a/docs/_sources/tutorials/optimize/opt_conv_tensorcore.rst.txt
+++ b/docs/_sources/tutorials/optimize/opt_conv_tensorcore.rst.txt
@@ -340,69 +340,75 @@ one time.
 
  .. code-block:: none
 
-    PrimFunc([A, W, Conv]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      // attr [iter_var(blockIdx.z, , blockIdx.z)] thread_extent = 196
-      // attr [Conv.wmma.accumulator] storage_scope = "wmma.accumulator"
-      allocate Conv.wmma.accumulator[float32 * 2048]
-      // attr [Apad.shared] storage_scope = "shared"
-      allocate Apad.shared[float16 * 12288]
-      // attr [W.shared] storage_scope = "shared"
-      allocate W.shared[float16 * 12288]
-      // attr [Apad.shared.wmma.matrix_a] storage_scope = "wmma.matrix_a"
-      allocate Apad.shared.wmma.matrix_a[float16 * 512]
-      // attr [W.shared.wmma.matrix_b] storage_scope = "wmma.matrix_b"
-      allocate W.shared.wmma.matrix_b[float16 * 1024]
-      // attr [iter_var(blockIdx.x, , blockIdx.x)] thread_extent = 2
-      // attr [iter_var(blockIdx.y, , blockIdx.y)] thread_extent = 4
-      // attr [iter_var(threadIdx.y, , threadIdx.y)] thread_extent = 4
-      // attr [iter_var(threadIdx.z, , threadIdx.z)] thread_extent = 2
-      for (n.c.init, 0, 2) {
-        for (o.c.init, 0, 4) {
-          for (nn.c.init, 0, 16) {
-            for (oo.c.init, 0, 16) {
-              Conv.wmma.accumulator[((((n.c.init*1024) + (o.c.init*256)) + (nn.c.init*16)) + oo.c.init)] = 0f
+    primfn(A_1: handle, W_1: handle, Conv_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {A: Buffer(A_2: handle, float16, [16, 14, 14, 16, 16, 16], []),
+                 W: Buffer(W_2: handle, float16, [3, 3, 16, 32, 16, 16], []),
+                 Conv: Buffer(Conv_2: handle, float32, [16, 14, 14, 32, 16, 16], [])}
+      buffer_map = {Conv_1: Conv, W_1: W, A_1: A} {
+      attr [IterVar(blockIdx.z: int32, (nullptr), "ThreadIndex", "blockIdx.z")] "thread_extent" = 196;
+      attr [Conv.wmma.accumulator: handle] "storage_scope" = "wmma.accumulator";
+      allocate(Conv.wmma.accumulator, float32, [2048]);
+      attr [Apad.shared: handle] "storage_scope" = "shared";
+      allocate(Apad.shared, float16, [12288]);
+      attr [W.shared: handle] "storage_scope" = "shared";
+      allocate(W.shared, float16, [12288]);
+      attr [Apad.shared.wmma.matrix_a: handle] "storage_scope" = "wmma.matrix_a";
+      allocate(Apad.shared.wmma.matrix_a, float16, [512]);
+      attr [W.shared.wmma.matrix_b: handle] "storage_scope" = "wmma.matrix_b";
+      allocate(W.shared.wmma.matrix_b, float16, [1024]);
+      attr [IterVar(blockIdx.x: int32, (nullptr), "ThreadIndex", "blockIdx.x")] "thread_extent" = 2;
+      attr [IterVar(blockIdx.y: int32, (nullptr), "ThreadIndex", "blockIdx.y")] "thread_extent" = 4;
+      attr [IterVar(threadIdx.y: int32, (nullptr), "ThreadIndex", "threadIdx.y")] "thread_extent" = 4;
+      attr [IterVar(threadIdx.z: int32, (nullptr), "ThreadIndex", "threadIdx.z")] "thread_extent" = 2 {
+        for (n.c.init: int32, 0, 2) {
+          for (o.c.init: int32, 0, 4) {
+            for (nn.c.init: int32, 0, 16) {
+              for (oo.c.init: int32, 0, 16) {
+                Conv.wmma.accumulator[((((n.c.init*1024) + (o.c.init*256)) + (nn.c.init*16)) + oo.c.init)] = 0f32
+              }
             }
           }
         }
-      }
-      for (ic.outer, 0, 8) {
-        for (kh, 0, 3) {
-          for (ax2, 0, 3) {
-            for (ax3, 0, 2) {
-              for (ax4.ax5.fused.outer, 0, 8) {
-                // attr [iter_var(threadIdx.x, , threadIdx.x)] thread_extent = 32
-                Apad.shared[((((((threadIdx.y*3072) + (threadIdx.z*1536)) + (ax2*512)) + (ax3*256)) + (ax4.ax5.fused.outer*32)) + threadIdx.x)] = tvm_if_then_else(((((1 <= (floordiv(blockIdx.z, 14) + kh)) && ((floordiv(blockIdx.z, 14) + kh) < 15)) && (1 <= (ax2 + floormod(blockIdx.z, 14)))) && ((ax2 + floormod(blockIdx.z, 14)) < 15)), A[(((((((((((blockIdx.x*6422528) + (threadIdx.y*1605632)) + (threadIdx.z*802816)) + (kh*57344)) + (blockIdx.z*4096)) + (ax2*4096)) + (ic.outer*512)) + (ax3 [...]
+        for (ic.outer: int32, 0, 8) {
+          for (kh: int32, 0, 3) {
+            for (ax2: int32, 0, 3) {
+              for (ax3: int32, 0, 2) {
+                for (ax4.ax5.fused.outer: int32, 0, 8) {
+                  attr [IterVar(threadIdx.x: int32, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 32;
+                  Apad.shared[((((((threadIdx.y*3072) + (threadIdx.z*1536)) + (ax2*512)) + (ax3*256)) + (ax4.ax5.fused.outer*32)) + threadIdx.x)] = @tvm_if_then_else(((((1 <= (floordiv(blockIdx.z, 14) + kh)) && ((floordiv(blockIdx.z, 14) + kh) < 15)) && (1 <= (ax2 + floormod(blockIdx.z, 14)))) && ((ax2 + floormod(blockIdx.z, 14)) < 15)), (float16*)A_2[(((((((((((blockIdx.x*6422528) + (threadIdx.y*1605632)) + (threadIdx.z*802816)) + (kh*57344)) + (blockIdx.z*4096)) + (ax2*4096)) + (ic.out [...]
+                }
               }
             }
-          }
-          for (ax1, 0, 3) {
-            for (ax2, 0, 2) {
-              // attr [iter_var(threadIdx.x, , threadIdx.x)] thread_extent = 32
-              W.shared[ramp((((((ax1*4096) + (ax2*2048)) + (threadIdx.y*512)) + (threadIdx.z*256)) + (threadIdx.x*8)), 1, 8)] = W[ramp(((((((((kh*393216) + (ax1*131072)) + (ic.outer*16384)) + (ax2*8192)) + (blockIdx.y*2048)) + (threadIdx.y*512)) + (threadIdx.z*256)) + (threadIdx.x*8)), 1, 8)]
+            for (ax1: int32, 0, 3) {
+              for (ax2_1: int32, 0, 2) {
+                attr [IterVar(threadIdx.x, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 32;
+                W.shared[ramp((((((ax1*4096) + (ax2_1*2048)) + (threadIdx.y*512)) + (threadIdx.z*256)) + (threadIdx.x*8)), 1, 8)] = (float16x8*)W_2[ramp(((((((((kh*393216) + (ax1*131072)) + (ic.outer*16384)) + (ax2_1*8192)) + (blockIdx.y*2048)) + (threadIdx.y*512)) + (threadIdx.z*256)) + (threadIdx.x*8)), 1, 8)])
+              }
             }
-          }
-          for (ic.inner, 0, 2) {
-            for (kw, 0, 3) {
-              for (ax0, 0, 2) {
-                for (ax4, 0, 16) {
-                  for (ax5, 0, 16) {
-                    Apad.shared.wmma.matrix_a[(((ax0*256) + (ax4*16)) + ax5)] = Apad.shared[((((((threadIdx.y*3072) + (ax0*1536)) + (kw*512)) + (ic.inner*256)) + (ax4*16)) + ax5)]
+            for (ic.inner: int32, 0, 2) {
+              for (kw: int32, 0, 3) {
+                for (ax0: int32, 0, 2) {
+                  for (ax4: int32, 0, 16) {
+                    for (ax5: int32, 0, 16) {
+                      Apad.shared.wmma.matrix_a[(((ax0*256) + (ax4*16)) + ax5)] = (float16*)Apad.shared[((((((threadIdx.y*3072) + (ax0*1536)) + (kw*512)) + (ic.inner*256)) + (ax4*16)) + ax5)])
+                    }
                   }
                 }
-              }
-              for (ax3, 0, 4) {
-                for (ax4, 0, 16) {
-                  for (ax5, 0, 16) {
-                    W.shared.wmma.matrix_b[(((ax3*256) + (ax4*16)) + ax5)] = W.shared[((((((kw*4096) + (ic.inner*2048)) + (threadIdx.z*1024)) + (ax3*256)) + (ax4*16)) + ax5)]
+                for (ax3_1: int32, 0, 4) {
+                  for (ax4_1: int32, 0, 16) {
+                    for (ax5_1: int32, 0, 16) {
+                      W.shared.wmma.matrix_b[(((ax3_1*256) + (ax4_1*16)) + ax5_1)] = (float16*)W.shared[((((((kw*4096) + (ic.inner*2048)) + (threadIdx.z*1024)) + (ax3_1*256)) + (ax4_1*16)) + ax5_1)])
+                    }
                   }
                 }
-              }
-              for (n.c, 0, 2) {
-                for (o.c, 0, 4) {
-                  for (nn.c, 0, 16) {
-                    for (oo.c, 0, 16) {
-                      for (ii, 0, 16) {
-                        Conv.wmma.accumulator[((((n.c*1024) + (o.c*256)) + (nn.c*16)) + oo.c)] = (Conv.wmma.accumulator[((((n.c*1024) + (o.c*256)) + (nn.c*16)) + oo.c)] + (float32(Apad.shared.wmma.matrix_a[(((n.c*256) + (nn.c*16)) + ii)])*float32(W.shared.wmma.matrix_b[(((o.c*256) + (ii*16)) + oo.c)])))
+                for (n.c: int32, 0, 2) {
+                  for (o.c: int32, 0, 4) {
+                    for (nn.c: int32, 0, 16) {
+                      for (oo.c: int32, 0, 16) {
+                        for (ii: int32, 0, 16) {
+                          Conv.wmma.accumulator[((((n.c*1024) + (o.c*256)) + (nn.c*16)) + oo.c)] = ((float32*)Conv.wmma.accumulator[((((n.c*1024) + (o.c*256)) + (nn.c*16)) + oo.c)]) + (cast(float32, (float16*)Apad.shared.wmma.matrix_a[(((n.c*256) + (nn.c*16)) + ii)]))*cast(float32, (float16*)W.shared.wmma.matrix_b[(((o.c*256) + (ii*16)) + oo.c)]))))
+                        }
                       }
                     }
                   }
@@ -411,19 +417,19 @@ one time.
             }
           }
         }
-      }
-      for (n.inner, 0, 2) {
-        for (o.inner, 0, 4) {
-          for (nn, 0, 16) {
-            for (oo, 0, 16) {
-              Conv[(((((((((blockIdx.x*12845056) + (threadIdx.y*3211264)) + (n.inner*1605632)) + (blockIdx.z*8192)) + (blockIdx.y*2048)) + (threadIdx.z*1024)) + (o.inner*256)) + (nn*16)) + oo)] = Conv.wmma.accumulator[((((n.inner*1024) + (o.inner*256)) + (nn*16)) + oo)]
+        for (n.inner: int32, 0, 2) {
+          for (o.inner: int32, 0, 4) {
+            for (nn: int32, 0, 16) {
+              for (oo: int32, 0, 16) {
+                Conv_2[(((((((((blockIdx.x*12845056) + (threadIdx.y*3211264)) + (n.inner*1605632)) + (blockIdx.z*8192)) + (blockIdx.y*2048)) + (threadIdx.z*1024)) + (o.inner*256)) + (nn*16)) + oo)] = (float32*)Conv.wmma.accumulator[((((n.inner*1024) + (o.inner*256)) + (nn*16)) + oo)])
+              }
             }
           }
         }
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -452,68 +458,74 @@ by mapping the 2D convolution to tensor intrinsics
 
  .. code-block:: none
 
-    PrimFunc([A, W, Conv]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      // attr [iter_var(blockIdx.z, , blockIdx.z)] thread_extent = 196
-      // attr [Conv.wmma.accumulator] storage_scope = "wmma.accumulator"
-      allocate Conv.wmma.accumulator[float32 * 2048]
-      // attr [Apad.shared] storage_scope = "shared"
-      allocate Apad.shared[float16 * 12288]
-      // attr [W.shared] storage_scope = "shared"
-      allocate W.shared[float16 * 12288]
-      // attr [Apad.shared.wmma.matrix_a] storage_scope = "wmma.matrix_a"
-      allocate Apad.shared.wmma.matrix_a[float16 * 512]
-      // attr [W.shared.wmma.matrix_b] storage_scope = "wmma.matrix_b"
-      allocate W.shared.wmma.matrix_b[float16 * 1024]
-      // attr [iter_var(blockIdx.x, , blockIdx.x)] thread_extent = 2
-      // attr [iter_var(blockIdx.y, , blockIdx.y)] thread_extent = 4
-      // attr [iter_var(threadIdx.y, , threadIdx.y)] thread_extent = 4
-      // attr [iter_var(threadIdx.z, , threadIdx.z)] thread_extent = 2
-      for (n.c.init, 0, 2) {
-        for (o.c.init, 0, 4) {
-          tvm_fill_fragment(Conv.wmma.accumulator, 16, 16, 16, ((n.c.init*4) + o.c.init), 0f)
+    primfn(A_1: handle, W_1: handle, Conv_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {A: Buffer(A_2: handle, float16, [16, 14, 14, 16, 16, 16], []),
+                 W: Buffer(W_2: handle, float16, [3, 3, 16, 32, 16, 16], []),
+                 Conv: Buffer(Conv_2: handle, float32, [16, 14, 14, 32, 16, 16], [])}
+      buffer_map = {Conv_1: Conv, W_1: W, A_1: A} {
+      attr [IterVar(blockIdx.z: int32, (nullptr), "ThreadIndex", "blockIdx.z")] "thread_extent" = 196;
+      attr [Conv.wmma.accumulator: handle] "storage_scope" = "wmma.accumulator";
+      allocate(Conv.wmma.accumulator, float32, [2048]);
+      attr [Apad.shared: handle] "storage_scope" = "shared";
+      allocate(Apad.shared, float16, [12288]);
+      attr [W.shared: handle] "storage_scope" = "shared";
+      allocate(W.shared, float16, [12288]);
+      attr [Apad.shared.wmma.matrix_a: handle] "storage_scope" = "wmma.matrix_a";
+      allocate(Apad.shared.wmma.matrix_a, float16, [512]);
+      attr [W.shared.wmma.matrix_b: handle] "storage_scope" = "wmma.matrix_b";
+      allocate(W.shared.wmma.matrix_b, float16, [1024]);
+      attr [IterVar(blockIdx.x: int32, (nullptr), "ThreadIndex", "blockIdx.x")] "thread_extent" = 2;
+      attr [IterVar(blockIdx.y: int32, (nullptr), "ThreadIndex", "blockIdx.y")] "thread_extent" = 4;
+      attr [IterVar(threadIdx.y: int32, (nullptr), "ThreadIndex", "threadIdx.y")] "thread_extent" = 4;
+      attr [IterVar(threadIdx.z: int32, (nullptr), "ThreadIndex", "threadIdx.z")] "thread_extent" = 2 {
+        for (n.c.init: int32, 0, 2) {
+          for (o.c.init: int32, 0, 4) {
+            @tvm_fill_fragment(Conv.wmma.accumulator, 16, 16, 16, ((n.c.init*4) + o.c.init), 0f32, dtype=handle, type="intrin", index=0)
+          }
         }
-      }
-      for (ic.outer, 0, 8) {
-        for (kh, 0, 3) {
-          for (ax2, 0, 3) {
-            for (ax3, 0, 2) {
-              for (ax4.ax5.fused.outer, 0, 8) {
-                // attr [iter_var(threadIdx.x, , threadIdx.x)] thread_extent = 32
-                Apad.shared[((((((threadIdx.y*3072) + (threadIdx.z*1536)) + (ax2*512)) + (ax3*256)) + (ax4.ax5.fused.outer*32)) + threadIdx.x)] = tvm_if_then_else(((((1 <= (floordiv(blockIdx.z, 14) + kh)) && ((floordiv(blockIdx.z, 14) + kh) < 15)) && (1 <= (ax2 + floormod(blockIdx.z, 14)))) && ((ax2 + floormod(blockIdx.z, 14)) < 15)), A[(((((((((((blockIdx.x*6422528) + (threadIdx.y*1605632)) + (threadIdx.z*802816)) + (kh*57344)) + (blockIdx.z*4096)) + (ax2*4096)) + (ic.outer*512)) + (ax3 [...]
+        for (ic.outer: int32, 0, 8) {
+          for (kh: int32, 0, 3) {
+            for (ax2: int32, 0, 3) {
+              for (ax3: int32, 0, 2) {
+                for (ax4.ax5.fused.outer: int32, 0, 8) {
+                  attr [IterVar(threadIdx.x: int32, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 32;
+                  Apad.shared[((((((threadIdx.y*3072) + (threadIdx.z*1536)) + (ax2*512)) + (ax3*256)) + (ax4.ax5.fused.outer*32)) + threadIdx.x)] = @tvm_if_then_else(((((1 <= (floordiv(blockIdx.z, 14) + kh)) && ((floordiv(blockIdx.z, 14) + kh) < 15)) && (1 <= (ax2 + floormod(blockIdx.z, 14)))) && ((ax2 + floormod(blockIdx.z, 14)) < 15)), (float16*)A_2[(((((((((((blockIdx.x*6422528) + (threadIdx.y*1605632)) + (threadIdx.z*802816)) + (kh*57344)) + (blockIdx.z*4096)) + (ax2*4096)) + (ic.out [...]
+                }
               }
             }
-          }
-          for (ax1, 0, 3) {
-            for (ax2, 0, 2) {
-              // attr [iter_var(threadIdx.x, , threadIdx.x)] thread_extent = 32
-              W.shared[ramp((((((ax1*4096) + (ax2*2048)) + (threadIdx.y*512)) + (threadIdx.z*256)) + (threadIdx.x*8)), 1, 8)] = W[ramp(((((((((kh*393216) + (ax1*131072)) + (ic.outer*16384)) + (ax2*8192)) + (blockIdx.y*2048)) + (threadIdx.y*512)) + (threadIdx.z*256)) + (threadIdx.x*8)), 1, 8)]
-            }
-          }
-          for (ic.inner, 0, 2) {
-            for (kw, 0, 3) {
-              for (ax0, 0, 2) {
-                tvm_load_matrix_sync(Apad.shared.wmma.matrix_a, 16, 16, 16, ax0, tvm_access_ptr(type_annotation(), Apad.shared, ((((threadIdx.y*3072) + (ax0*1536)) + (kw*512)) + (ic.inner*256)), 256, 1), 16, "row_major")
-              }
-              for (ax3, 0, 4) {
-                tvm_load_matrix_sync(W.shared.wmma.matrix_b, 16, 16, 16, ax3, tvm_access_ptr(type_annotation(), W.shared, ((((kw*4096) + (ic.inner*2048)) + (threadIdx.z*1024)) + (ax3*256)), 256, 1), 16, "row_major")
+            for (ax1: int32, 0, 3) {
+              for (ax2_1: int32, 0, 2) {
+                attr [IterVar(threadIdx.x, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 32;
+                W.shared[ramp((((((ax1*4096) + (ax2_1*2048)) + (threadIdx.y*512)) + (threadIdx.z*256)) + (threadIdx.x*8)), 1, 8)] = (float16x8*)W_2[ramp(((((((((kh*393216) + (ax1*131072)) + (ic.outer*16384)) + (ax2_1*8192)) + (blockIdx.y*2048)) + (threadIdx.y*512)) + (threadIdx.z*256)) + (threadIdx.x*8)), 1, 8)])
               }
-              for (n.c, 0, 2) {
-                for (o.c, 0, 4) {
-                  tvm_mma_sync(Conv.wmma.accumulator, ((n.c*4) + o.c), Apad.shared.wmma.matrix_a, n.c, W.shared.wmma.matrix_b, o.c, Conv.wmma.accumulator, ((n.c*4) + o.c))
+            }
+            for (ic.inner: int32, 0, 2) {
+              for (kw: int32, 0, 3) {
+                for (ax0: int32, 0, 2) {
+                  @tvm_load_matrix_sync(Apad.shared.wmma.matrix_a, 16, 16, 16, ax0, @tvm_access_ptr(@type_annotation(, dtype=float16, type="pure_intrin", index=0), Apad.shared, ((((threadIdx.y*3072) + (ax0*1536)) + (kw*512)) + (ic.inner*256)), 256, 1, dtype=handle, type="intrin", index=0), 16, "row_major", dtype=handle, type="intrin", index=0)
+                }
+                for (ax3_1: int32, 0, 4) {
+                  @tvm_load_matrix_sync(W.shared.wmma.matrix_b, 16, 16, 16, ax3_1, @tvm_access_ptr(@type_annotation(, dtype=float16, type="pure_intrin", index=0), W.shared, ((((kw*4096) + (ic.inner*2048)) + (threadIdx.z*1024)) + (ax3_1*256)), 256, 1, dtype=handle, type="intrin", index=0), 16, "row_major", dtype=handle, type="intrin", index=0)
+                }
+                for (n.c: int32, 0, 2) {
+                  for (o.c: int32, 0, 4) {
+                    @tvm_mma_sync(Conv.wmma.accumulator, ((n.c*4) + o.c), Apad.shared.wmma.matrix_a, n.c, W.shared.wmma.matrix_b, o.c, Conv.wmma.accumulator, ((n.c*4) + o.c), dtype=handle, type="intrin", index=0)
+                  }
                 }
               }
             }
           }
         }
-      }
-      for (n.inner, 0, 2) {
-        for (o.inner, 0, 4) {
-          tvm_store_matrix_sync(Conv.wmma.accumulator, 16, 16, 16, ((n.inner*4) + o.inner), tvm_access_ptr(type_annotation(), Conv, (((((((blockIdx.x*12845056) + (threadIdx.y*3211264)) + (n.inner*1605632)) + (blockIdx.z*8192)) + (blockIdx.y*2048)) + (threadIdx.z*1024)) + (o.inner*256)), 256, 2), 16, "row_major")
+        for (n.inner: int32, 0, 2) {
+          for (o.inner: int32, 0, 4) {
+            @tvm_store_matrix_sync(Conv.wmma.accumulator, 16, 16, 16, ((n.inner*4) + o.inner), @tvm_access_ptr(@type_annotation(, dtype=float32, type="pure_intrin", index=0), Conv_2, (((((((blockIdx.x*12845056) + (threadIdx.y*3211264)) + (n.inner*1605632)) + (blockIdx.z*8192)) + (blockIdx.y*2048)) + (threadIdx.z*1024)) + (o.inner*256)), 256, 2, dtype=handle, type="intrin", index=0), 16, "row_major", dtype=handle, type="intrin", index=0)
+          }
         }
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -549,7 +561,7 @@ be able to run on our build server
 
  .. code-block:: none
 
-    conv2d with tensor core: 6.307512 ms
+    conv2d with tensor core: 6.302243 ms
 
 
 
diff --git a/docs/_sources/tutorials/optimize/opt_gemm.rst.txt b/docs/_sources/tutorials/optimize/opt_gemm.rst.txt
index be88cfe..ab42635 100644
--- a/docs/_sources/tutorials/optimize/opt_gemm.rst.txt
+++ b/docs/_sources/tutorials/optimize/opt_gemm.rst.txt
@@ -118,8 +118,8 @@ Then we write a baseline implementation, the simplest way to write a matrix mult
 
  .. code-block:: none
 
-    Numpy running time: 0.007315
-    Baseline: 5.920522
+    Numpy running time: 0.007157
+    Baseline: 5.880575
 
 
 
@@ -142,18 +142,23 @@ Here is the generated IR using our baseline schedule.
 
  .. code-block:: none
 
-    PrimFunc([A, B, C]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      for (x, 0, 1024) {
-        for (y, 0, 1024) {
-          C[((x*1024) + y)] = 0f
-          for (k, 0, 1024) {
-            C[((x*1024) + y)] = (C[((x*1024) + y)] + (A[((x*1024) + k)]*B[((k*1024) + y)]))
+    primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {B: Buffer(B_2: handle, float32, [1024, 1024], []),
+                 A: Buffer(A_2: handle, float32, [1024, 1024], []),
+                 C: Buffer(C_2: handle, float32, [1024, 1024], [])}
+      buffer_map = {C_1: C, B_1: B, A_1: A} {
+      for (x: int32, 0, 1024) {
+        for (y: int32, 0, 1024) {
+          C_2[((x*1024) + y)] = 0f32
+          for (k: int32, 0, 1024) {
+            C_2[((x*1024) + y)] = ((float32*)C_2[((x*1024) + y)]) + ((float32*)A_2[((x*1024) + k)])*(float32*)B_2[((k*1024) + y)])))
           }
         }
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -201,7 +206,7 @@ fill 32 * 32 * sizeof(float) which is 4KB in the cache whose total size is 32KB
 
  .. code-block:: none
 
-    Opt1: 0.110557
+    Opt1: 0.108626
 
 
 
@@ -223,19 +228,24 @@ Here is the generated IR after blocking.
 
  .. code-block:: none
 
-    PrimFunc([A, B, C]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      for (x.outer, 0, 32) {
-        for (y.outer, 0, 32) {
-          for (x.inner.init, 0, 32) {
-            for (y.inner.init, 0, 32) {
-              C[((((x.outer*32768) + (x.inner.init*1024)) + (y.outer*32)) + y.inner.init)] = 0f
+    primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {A: Buffer(A_2: handle, float32, [1024, 1024], []),
+                 B: Buffer(B_2: handle, float32, [1024, 1024], []),
+                 C: Buffer(C_2: handle, float32, [1024, 1024], [])}
+      buffer_map = {C_1: C, B_1: B, A_1: A} {
+      for (x.outer: int32, 0, 32) {
+        for (y.outer: int32, 0, 32) {
+          for (x.inner.init: int32, 0, 32) {
+            for (y.inner.init: int32, 0, 32) {
+              C_2[((((x.outer*32768) + (x.inner.init*1024)) + (y.outer*32)) + y.inner.init)] = 0f32
             }
           }
-          for (k.outer, 0, 256) {
-            for (k.inner, 0, 4) {
-              for (x.inner, 0, 32) {
-                for (y.inner, 0, 32) {
-                  C[((((x.outer*32768) + (x.inner*1024)) + (y.outer*32)) + y.inner)] = (C[((((x.outer*32768) + (x.inner*1024)) + (y.outer*32)) + y.inner)] + (A[((((x.outer*32768) + (x.inner*1024)) + (k.outer*4)) + k.inner)]*B[((((k.outer*4096) + (k.inner*1024)) + (y.outer*32)) + y.inner)]))
+          for (k.outer: int32, 0, 256) {
+            for (k.inner: int32, 0, 4) {
+              for (x.inner: int32, 0, 32) {
+                for (y.inner: int32, 0, 32) {
+                  C_2[((((x.outer*32768) + (x.inner*1024)) + (y.outer*32)) + y.inner)] = ((float32*)C_2[((((x.outer*32768) + (x.inner*1024)) + (y.outer*32)) + y.inner)]) + ((float32*)A_2[((((x.outer*32768) + (x.inner*1024)) + (k.outer*4)) + k.inner)])*(float32*)B_2[((((k.outer*4096) + (k.inner*1024)) + (y.outer*32)) + y.inner)])))
                 }
               }
             }
@@ -244,7 +254,7 @@ Here is the generated IR after blocking.
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -290,7 +300,7 @@ In this tutorial, we chose to vectorize the inner loop row data since it is cach
 
  .. code-block:: none
 
-    Opt2: 0.126554
+    Opt2: 0.115723
 
 
 
@@ -312,16 +322,21 @@ Here is the generated IR after vectorization.
 
  .. code-block:: none
 
-    PrimFunc([A, B, C]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      for (x.outer, 0, 32) {
-        for (y.outer, 0, 32) {
-          for (x.inner.init, 0, 32) {
-            C[ramp((((x.outer*32768) + (x.inner.init*1024)) + (y.outer*32)), 1, 32)] = x32(0f)
+    primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {A: Buffer(A_2: handle, float32, [1024, 1024], []),
+                 B: Buffer(B_2: handle, float32, [1024, 1024], []),
+                 C: Buffer(C_2: handle, float32, [1024, 1024], [])}
+      buffer_map = {C_1: C, B_1: B, A_1: A} {
+      for (x.outer: int32, 0, 32) {
+        for (y.outer: int32, 0, 32) {
+          for (x.inner.init: int32, 0, 32) {
+            C_2[ramp((((x.outer*32768) + (x.inner.init*1024)) + (y.outer*32)), 1, 32)] = broadcast(0f32, 32)
           }
-          for (k.outer, 0, 256) {
-            for (k.inner, 0, 4) {
-              for (x.inner, 0, 32) {
-                C[ramp((((x.outer*32768) + (x.inner*1024)) + (y.outer*32)), 1, 32)] = (C[ramp((((x.outer*32768) + (x.inner*1024)) + (y.outer*32)), 1, 32)] + (x32(A[((((x.outer*32768) + (x.inner*1024)) + (k.outer*4)) + k.inner)])*B[ramp((((k.outer*4096) + (k.inner*1024)) + (y.outer*32)), 1, 32)]))
+          for (k.outer: int32, 0, 256) {
+            for (k.inner: int32, 0, 4) {
+              for (x.inner: int32, 0, 32) {
+                C_2[ramp((((x.outer*32768) + (x.inner*1024)) + (y.outer*32)), 1, 32)] = ((float32x32*)C_2[ramp((((x.outer*32768) + (x.inner*1024)) + (y.outer*32)), 1, 32)]) + (broadcast((float32*)A_2[((((x.outer*32768) + (x.inner*1024)) + (k.outer*4)) + k.inner)]), 32)*(float32x32*)B_2[ramp((((k.outer*4096) + (k.inner*1024)) + (y.outer*32)), 1, 32)])))
               }
             }
           }
@@ -329,7 +344,7 @@ Here is the generated IR after vectorization.
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -374,7 +389,7 @@ the access pattern for A matrix is more cache friendly.
 
  .. code-block:: none
 
-    Opt3: 0.061485
+    Opt3: 0.059947
 
 
 
@@ -396,16 +411,21 @@ Here is the generated IR after loop permutation.
 
  .. code-block:: none
 
-    PrimFunc([A, B, C]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      for (x.outer, 0, 32) {
-        for (y.outer, 0, 32) {
-          for (x.inner.init, 0, 32) {
-            C[ramp((((x.outer*32768) + (x.inner.init*1024)) + (y.outer*32)), 1, 32)] = x32(0f)
+    primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {A: Buffer(A_2: handle, float32, [1024, 1024], []),
+                 B: Buffer(B_2: handle, float32, [1024, 1024], []),
+                 C: Buffer(C_2: handle, float32, [1024, 1024], [])}
+      buffer_map = {C_1: C, B_1: B, A_1: A} {
+      for (x.outer: int32, 0, 32) {
+        for (y.outer: int32, 0, 32) {
+          for (x.inner.init: int32, 0, 32) {
+            C_2[ramp((((x.outer*32768) + (x.inner.init*1024)) + (y.outer*32)), 1, 32)] = broadcast(0f32, 32)
           }
-          for (k.outer, 0, 256) {
-            for (x.inner, 0, 32) {
-              for (k.inner, 0, 4) {
-                C[ramp((((x.outer*32768) + (x.inner*1024)) + (y.outer*32)), 1, 32)] = (C[ramp((((x.outer*32768) + (x.inner*1024)) + (y.outer*32)), 1, 32)] + (x32(A[((((x.outer*32768) + (x.inner*1024)) + (k.outer*4)) + k.inner)])*B[ramp((((k.outer*4096) + (k.inner*1024)) + (y.outer*32)), 1, 32)]))
+          for (k.outer: int32, 0, 256) {
+            for (x.inner: int32, 0, 32) {
+              for (k.inner: int32, 0, 4) {
+                C_2[ramp((((x.outer*32768) + (x.inner*1024)) + (y.outer*32)), 1, 32)] = ((float32x32*)C_2[ramp((((x.outer*32768) + (x.inner*1024)) + (y.outer*32)), 1, 32)]) + (broadcast((float32*)A_2[((((x.outer*32768) + (x.inner*1024)) + (k.outer*4)) + k.inner)]), 32)*(float32x32*)B_2[ramp((((k.outer*4096) + (k.inner*1024)) + (y.outer*32)), 1, 32)])))
               }
             }
           }
@@ -413,7 +433,7 @@ Here is the generated IR after loop permutation.
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -477,7 +497,7 @@ the corresponding value from the packed array.
 
  .. code-block:: none
 
-    Opt4: 0.072638
+    Opt4: 0.091496
 
 
 
@@ -499,23 +519,29 @@ Here is the generated IR after array packing.
 
  .. code-block:: none
 
-    PrimFunc([A, B, C]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      // attr [packedB] storage_scope = "global"
-      allocate packedB[float32x32 * 32768]
-      parallel (x, 0, 32) {
-        for (y, 0, 1024) {
-          packedB[ramp(((x*32768) + (y*32)), 1, 32)] = B[ramp(((y*1024) + (x*32)), 1, 32)]
-        }
-      }
-      for (x.outer, 0, 32) {
-        for (y.outer, 0, 32) {
-          for (x.inner.init, 0, 32) {
-            C[ramp((((x.outer*32768) + (x.inner.init*1024)) + (y.outer*32)), 1, 32)] = x32(0f)
+    primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {A: Buffer(A_2: handle, float32, [1024, 1024], []),
+                 B: Buffer(B_2: handle, float32, [1024, 1024], []),
+                 C: Buffer(C_2: handle, float32, [1024, 1024], [])}
+      buffer_map = {C_1: C, B_1: B, A_1: A} {
+      attr [packedB: handle] "storage_scope" = "global";
+      allocate(packedB, float32x32, [32768]) {
+        for (x: int32, 0, 32) "parallel" {
+          for (y: int32, 0, 1024) {
+            packedB[ramp(((x*32768) + (y*32)), 1, 32)] = (float32x32*)B_2[ramp(((y*1024) + (x*32)), 1, 32)])
           }
-          for (k.outer, 0, 256) {
-            for (x.inner, 0, 32) {
-              for (k.inner, 0, 4) {
-                C[ramp((((x.outer*32768) + (x.inner*1024)) + (y.outer*32)), 1, 32)] = (C[ramp((((x.outer*32768) + (x.inner*1024)) + (y.outer*32)), 1, 32)] + (x32(A[((((x.outer*32768) + (x.inner*1024)) + (k.outer*4)) + k.inner)])*packedB[ramp((((y.outer*32768) + (k.outer*128)) + (k.inner*32)), 1, 32)]))
+        }
+        for (x.outer: int32, 0, 32) {
+          for (y.outer: int32, 0, 32) {
+            for (x.inner.init: int32, 0, 32) {
+              C_2[ramp((((x.outer*32768) + (x.inner.init*1024)) + (y.outer*32)), 1, 32)] = broadcast(0f32, 32)
+            }
+            for (k.outer: int32, 0, 256) {
+              for (x.inner: int32, 0, 32) {
+                for (k.inner: int32, 0, 4) {
+                  C_2[ramp((((x.outer*32768) + (x.inner*1024)) + (y.outer*32)), 1, 32)] = ((float32x32*)C_2[ramp((((x.outer*32768) + (x.inner*1024)) + (y.outer*32)), 1, 32)]) + (broadcast((float32*)A_2[((((x.outer*32768) + (x.inner*1024)) + (k.outer*4)) + k.inner)]), 32)*(float32x32*)packedB[ramp((((y.outer*32768) + (k.outer*128)) + (k.inner*32)), 1, 32)])))
+                }
               }
             }
           }
@@ -523,7 +549,7 @@ Here is the generated IR after array packing.
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -581,7 +607,7 @@ write to C when all the block results are ready.
 
  .. code-block:: none
 
-    Opt5: 0.064131
+    Opt5: 0.087638
 
 
 
@@ -603,39 +629,45 @@ Here is the generated IR after blocking.
 
  .. code-block:: none
 
-    PrimFunc([A, B, C]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      // attr [packedB] storage_scope = "global"
-      allocate packedB[float32x32 * 32768]
-      // attr [C.global] storage_scope = "global"
-      allocate C.global[float32 * 1024]
-      parallel (x, 0, 32) {
-        for (y, 0, 1024) {
-          packedB[ramp(((x*32768) + (y*32)), 1, 32)] = B[ramp(((y*1024) + (x*32)), 1, 32)]
-        }
-      }
-      for (x.outer, 0, 32) {
-        for (y.outer, 0, 32) {
-          for (x.c.init, 0, 32) {
-            C.global[ramp((x.c.init*32), 1, 32)] = x32(0f)
+    primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {A: Buffer(A_2: handle, float32, [1024, 1024], []),
+                 B: Buffer(B_2: handle, float32, [1024, 1024], []),
+                 C: Buffer(C_2: handle, float32, [1024, 1024], [])}
+      buffer_map = {C_1: C, B_1: B, A_1: A} {
+      attr [packedB: handle] "storage_scope" = "global";
+      allocate(packedB, float32x32, [32768]);
+      attr [C.global: handle] "storage_scope" = "global";
+      allocate(C.global, float32, [1024]) {
+        for (x: int32, 0, 32) "parallel" {
+          for (y: int32, 0, 1024) {
+            packedB[ramp(((x*32768) + (y*32)), 1, 32)] = (float32x32*)B_2[ramp(((y*1024) + (x*32)), 1, 32)])
           }
-          for (k.outer, 0, 256) {
-            for (x.c, 0, 32) {
-              C.global[ramp((x.c*32), 1, 32)] = (C.global[ramp((x.c*32), 1, 32)] + (x32(A[(((x.outer*32768) + (x.c*1024)) + (k.outer*4))])*packedB[ramp(((y.outer*32768) + (k.outer*128)), 1, 32)]))
-              C.global[ramp((x.c*32), 1, 32)] = (C.global[ramp((x.c*32), 1, 32)] + (x32(A[((((x.outer*32768) + (x.c*1024)) + (k.outer*4)) + 1)])*packedB[ramp((((y.outer*32768) + (k.outer*128)) + 32), 1, 32)]))
-              C.global[ramp((x.c*32), 1, 32)] = (C.global[ramp((x.c*32), 1, 32)] + (x32(A[((((x.outer*32768) + (x.c*1024)) + (k.outer*4)) + 2)])*packedB[ramp((((y.outer*32768) + (k.outer*128)) + 64), 1, 32)]))
-              C.global[ramp((x.c*32), 1, 32)] = (C.global[ramp((x.c*32), 1, 32)] + (x32(A[((((x.outer*32768) + (x.c*1024)) + (k.outer*4)) + 3)])*packedB[ramp((((y.outer*32768) + (k.outer*128)) + 96), 1, 32)]))
+        }
+        for (x.outer: int32, 0, 32) {
+          for (y.outer: int32, 0, 32) {
+            for (x.c.init: int32, 0, 32) {
+              C.global[ramp((x.c.init*32), 1, 32)] = broadcast(0f32, 32)
             }
-          }
-          for (x.inner, 0, 32) {
-            for (y.inner, 0, 32) {
-              C[((((x.outer*32768) + (x.inner*1024)) + (y.outer*32)) + y.inner)] = C.global[((x.inner*32) + y.inner)]
+            for (k.outer: int32, 0, 256) {
+              for (x.c: int32, 0, 32) {
+                C.global[ramp((x.c*32), 1, 32)] = ((float32x32*)C.global[ramp((x.c*32), 1, 32)]) + (broadcast((float32*)A_2[(((x.outer*32768) + (x.c*1024)) + (k.outer*4))]), 32)*(float32x32*)packedB[ramp(((y.outer*32768) + (k.outer*128)), 1, 32)])))
+                C.global[ramp((x.c*32), 1, 32)] = ((float32x32*)C.global[ramp((x.c*32), 1, 32)]) + (broadcast((float32*)A_2[((((x.outer*32768) + (x.c*1024)) + (k.outer*4)) + 1)]), 32)*(float32x32*)packedB[ramp((((y.outer*32768) + (k.outer*128)) + 32), 1, 32)])))
+                C.global[ramp((x.c*32), 1, 32)] = ((float32x32*)C.global[ramp((x.c*32), 1, 32)]) + (broadcast((float32*)A_2[((((x.outer*32768) + (x.c*1024)) + (k.outer*4)) + 2)]), 32)*(float32x32*)packedB[ramp((((y.outer*32768) + (k.outer*128)) + 64), 1, 32)])))
+                C.global[ramp((x.c*32), 1, 32)] = ((float32x32*)C.global[ramp((x.c*32), 1, 32)]) + (broadcast((float32*)A_2[((((x.outer*32768) + (x.c*1024)) + (k.outer*4)) + 3)]), 32)*(float32x32*)packedB[ramp((((y.outer*32768) + (k.outer*128)) + 96), 1, 32)])))
+              }
+            }
+            for (x.inner: int32, 0, 32) {
+              for (y.inner: int32, 0, 32) {
+                C_2[((((x.outer*32768) + (x.inner*1024)) + (y.outer*32)) + y.inner)] = (float32*)C.global[((x.inner*32) + y.inner)])
+              }
             }
           }
         }
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -691,7 +723,7 @@ Futhermore, we can also utilize multi-core processors to do the thread-level par
 
  .. code-block:: none
 
-    Opt6: 0.013905
+    Opt6: 0.016215
 
 
 
@@ -713,39 +745,45 @@ Here is the generated IR after parallelization.
 
  .. code-block:: none
 
-    PrimFunc([A, B, C]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      // attr [packedB] storage_scope = "global"
-      allocate packedB[float32x32 * 32768]
-      parallel (x, 0, 32) {
-        for (y, 0, 1024) {
-          packedB[ramp(((x*32768) + (y*32)), 1, 32)] = B[ramp(((y*1024) + (x*32)), 1, 32)]
-        }
-      }
-      parallel (x.outer, 0, 32) {
-        // attr [C.global] storage_scope = "global"
-        allocate C.global[float32 * 1024]
-        for (y.outer, 0, 32) {
-          for (x.c.init, 0, 32) {
-            C.global[ramp((x.c.init*32), 1, 32)] = x32(0f)
+    primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {A: Buffer(A_2: handle, float32, [1024, 1024], []),
+                 C: Buffer(C_2: handle, float32, [1024, 1024], []),
+                 B: Buffer(B_2: handle, float32, [1024, 1024], [])}
+      buffer_map = {B_1: B, C_1: C, A_1: A} {
+      attr [packedB: handle] "storage_scope" = "global";
+      allocate(packedB, float32x32, [32768]) {
+        for (x: int32, 0, 32) "parallel" {
+          for (y: int32, 0, 1024) {
+            packedB[ramp(((x*32768) + (y*32)), 1, 32)] = (float32x32*)B_2[ramp(((y*1024) + (x*32)), 1, 32)])
           }
-          for (k.outer, 0, 256) {
-            for (x.c, 0, 32) {
-              C.global[ramp((x.c*32), 1, 32)] = (C.global[ramp((x.c*32), 1, 32)] + (x32(A[(((x.outer*32768) + (x.c*1024)) + (k.outer*4))])*packedB[ramp(((y.outer*32768) + (k.outer*128)), 1, 32)]))
-              C.global[ramp((x.c*32), 1, 32)] = (C.global[ramp((x.c*32), 1, 32)] + (x32(A[((((x.outer*32768) + (x.c*1024)) + (k.outer*4)) + 1)])*packedB[ramp((((y.outer*32768) + (k.outer*128)) + 32), 1, 32)]))
-              C.global[ramp((x.c*32), 1, 32)] = (C.global[ramp((x.c*32), 1, 32)] + (x32(A[((((x.outer*32768) + (x.c*1024)) + (k.outer*4)) + 2)])*packedB[ramp((((y.outer*32768) + (k.outer*128)) + 64), 1, 32)]))
-              C.global[ramp((x.c*32), 1, 32)] = (C.global[ramp((x.c*32), 1, 32)] + (x32(A[((((x.outer*32768) + (x.c*1024)) + (k.outer*4)) + 3)])*packedB[ramp((((y.outer*32768) + (k.outer*128)) + 96), 1, 32)]))
+        }
+        for (x.outer: int32, 0, 32) "parallel" {
+          attr [C.global: handle] "storage_scope" = "global";
+          allocate(C.global, float32, [1024]);
+          for (y.outer: int32, 0, 32) {
+            for (x.c.init: int32, 0, 32) {
+              C.global[ramp((x.c.init*32), 1, 32)] = broadcast(0f32, 32)
             }
-          }
-          for (x.inner, 0, 32) {
-            for (y.inner, 0, 32) {
-              C[((((x.outer*32768) + (x.inner*1024)) + (y.outer*32)) + y.inner)] = C.global[((x.inner*32) + y.inner)]
+            for (k.outer: int32, 0, 256) {
+              for (x.c: int32, 0, 32) {
+                C.global[ramp((x.c*32), 1, 32)] = ((float32x32*)C.global[ramp((x.c*32), 1, 32)]) + (broadcast((float32*)A_2[(((x.outer*32768) + (x.c*1024)) + (k.outer*4))]), 32)*(float32x32*)packedB[ramp(((y.outer*32768) + (k.outer*128)), 1, 32)])))
+                C.global[ramp((x.c*32), 1, 32)] = ((float32x32*)C.global[ramp((x.c*32), 1, 32)]) + (broadcast((float32*)A_2[((((x.outer*32768) + (x.c*1024)) + (k.outer*4)) + 1)]), 32)*(float32x32*)packedB[ramp((((y.outer*32768) + (k.outer*128)) + 32), 1, 32)])))
+                C.global[ramp((x.c*32), 1, 32)] = ((float32x32*)C.global[ramp((x.c*32), 1, 32)]) + (broadcast((float32*)A_2[((((x.outer*32768) + (x.c*1024)) + (k.outer*4)) + 2)]), 32)*(float32x32*)packedB[ramp((((y.outer*32768) + (k.outer*128)) + 64), 1, 32)])))
+                C.global[ramp((x.c*32), 1, 32)] = ((float32x32*)C.global[ramp((x.c*32), 1, 32)]) + (broadcast((float32*)A_2[((((x.outer*32768) + (x.c*1024)) + (k.outer*4)) + 3)]), 32)*(float32x32*)packedB[ramp((((y.outer*32768) + (k.outer*128)) + 96), 1, 32)])))
+              }
+            }
+            for (x.inner: int32, 0, 32) {
+              for (y.inner: int32, 0, 32) {
+                C_2[((((x.outer*32768) + (x.inner*1024)) + (y.outer*32)) + y.inner)] = (float32*)C.global[((x.inner*32) + y.inner)])
+              }
             }
           }
         }
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
diff --git a/docs/_sources/tutorials/optimize/sg_execution_times.rst.txt b/docs/_sources/tutorials/optimize/sg_execution_times.rst.txt
index 30c0408..6ecaba4 100644
--- a/docs/_sources/tutorials/optimize/sg_execution_times.rst.txt
+++ b/docs/_sources/tutorials/optimize/sg_execution_times.rst.txt
@@ -5,9 +5,9 @@
 
 Computation times
 =================
-**00:26.970** total execution time for **tutorials_optimize** files:
+**00:27.289** total execution time for **tutorials_optimize** files:
 
-- **00:25.245**: :ref:`sphx_glr_tutorials_optimize_opt_gemm.py` (``opt_gemm.py``)
-- **00:00.902**: :ref:`sphx_glr_tutorials_optimize_opt_conv_tensorcore.py` (``opt_conv_tensorcore.py``)
-- **00:00.675**: :ref:`sphx_glr_tutorials_optimize_opt_conv_cuda.py` (``opt_conv_cuda.py``)
-- **00:00.148**: :ref:`sphx_glr_tutorials_optimize_opt_matmul_auto_tensorcore.py` (``opt_matmul_auto_tensorcore.py``)
+- **00:25.575**: :ref:`sphx_glr_tutorials_optimize_opt_gemm.py` (``opt_gemm.py``)
+- **00:00.905**: :ref:`sphx_glr_tutorials_optimize_opt_conv_tensorcore.py` (``opt_conv_tensorcore.py``)
+- **00:00.664**: :ref:`sphx_glr_tutorials_optimize_opt_conv_cuda.py` (``opt_conv_cuda.py``)
+- **00:00.146**: :ref:`sphx_glr_tutorials_optimize_opt_matmul_auto_tensorcore.py` (``opt_matmul_auto_tensorcore.py``)
diff --git a/docs/_sources/tutorials/relay_quick_start.rst.txt b/docs/_sources/tutorials/relay_quick_start.rst.txt
index 939eaa7..3738575 100644
--- a/docs/_sources/tutorials/relay_quick_start.rst.txt
+++ b/docs/_sources/tutorials/relay_quick_start.rst.txt
@@ -92,7 +92,7 @@ to show the network structure.
       %3 = nn.batch_norm(%2, %bn0_gamma, %bn0_beta, %bn0_moving_mean, %bn0_moving_var, epsilon=2e-05f) /* ty=(Tensor[(1, 64, 112, 112), float32], Tensor[(64), float32], Tensor[(64), float32]) */;
       %4 = %3.0;
       %5 = nn.relu(%4) /* ty=Tensor[(1, 64, 112, 112), float32] */;
-      %6 = nn.max_pool2d(%5, pool_size=[3, 3], strides=[2, 2], padding=[1, 1]) /* ty=Tensor[(1, 64, 56, 56), float32] */;
+      %6 = nn.max_pool2d(%5, pool_size=[3, 3], strides=[2, 2], padding=[1, 1, 1, 1]) /* ty=Tensor[(1, 64, 56, 56), float32] */;
       %7 = nn.batch_norm(%6, %stage1_unit1_bn1_gamma, %stage1_unit1_bn1_beta, %stage1_unit1_bn1_moving_mean, %stage1_unit1_bn1_moving_var, epsilon=2e-05f) /* ty=(Tensor[(1, 64, 56, 56), float32], Tensor[(64), float32], Tensor[(64), float32]) */;
       %8 = %7.0;
       %9 = nn.relu(%8) /* ty=Tensor[(1, 64, 56, 56), float32] */;
@@ -223,7 +223,7 @@ in this example. Then the machine code will be generated as the module library.
 
  .. code-block:: none
 
-
    ...1%, 0.01 MB, 41 KB/s, 0 seconds passed
    ...3%, 0.02 MB, 81 KB/s, 0 seconds passed
    ...5%, 0.02 MB, 122 KB/s, 0 seconds passed
    ...7%, 0.03 MB, 162 KB/s, 0 seconds passed
    ...9%, 0.04 MB, 202 KB/s, 0 seconds passed
    ...11%, 0.05 MB, 243 KB/s, 0 seconds passed
    ...13%, 0.05 MB, 283 KB/s, 0 seconds passed
    ...15%, 0.06 MB, 322 KB/s, 0 seconds passed
    ...17%, 0.07 MB, 362 KB/s, 0 seconds passed
    ...19%, 0.08 MB, 401 KB/s, 0 seconds passed
    ...21%, 0.09 MB, 441 KB/s, 0 seconds passed
    ...23%, 0.09 MB, 480 KB/s, 0 seconds passed
    ...25%, 0.10 MB, 519 KB/s, 0 seconds passed
    ...27%, 0.11 MB, 558 KB/s, 0 seconds passed
    ...29%, 0.12 MB, 597 KB/s, 0 seconds passed
    ...31%, 0.12 MB, 636 KB/s, 0 seconds passed
    ...33%, 0.13 MB, 675 KB/s, 0 seconds passed
    ...35%, 0.14 MB, 713 KB/s, 0 seconds passed
    ...37%, 0.15 MB, 753 KB/s, 0 seconds passed
    ...39%, 0.16 MB, 789 KB/s, 0 seconds passed
    ...41%, 0.16 MB, 828 KB/s, 0 seconds pa
 ssed
    ...43%, 0.17 MB, 867 KB/s, 0 seconds passed
    ...45%, 0.18 MB, 906 KB/s, 0 seconds passed
    ...47%, 0.19 MB, 941 KB/s, 0 seconds passed
    ...49%, 0.20 MB, 979 KB/s, 0 seconds passed
    ...51%, 0.20 MB, 1018 KB/s, 0 seconds passed
    ...53%, 0.21 MB, 1056 KB/s, 0 seconds passed
    ...55%, 0.22 MB, 1094 KB/s, 0 seconds passed
    ...57%, 0.23 MB, 1133 KB/s, 0 seconds passed
    ...59%, 0.23 MB, 1169 KB/s, 0 seconds passed
    ...61%, 0.24 MB, 1207 KB/s, 0 seconds passed
    ...63%, 0.25 MB, 1241 KB/s, 0 seconds passed
    ...65%, 0.26 MB, 1279 KB/s, 0 seconds passed
    ...67%, 0.27 MB, 1317 KB/s, 0 seconds passed
    ...69%, 0.27 MB, 1356 KB/s, 0 seconds passed
    ...71%, 0.28 MB, 1394 KB/s, 0 seconds passed
    ...73%, 0.29 MB, 1432 KB/s, 0 seconds passed
    ...75%, 0.30 MB, 1466 KB/s, 0 seconds passed
    ...77%, 0.30 MB, 1504 KB/s, 0 seconds passed
    ...79%, 0.31 MB, 1541 KB/s, 0 seconds passed
    ...81%, 0.32 MB, 1579 KB/s, 0 seconds passed
    ...83%, 0.33
  MB, 1612 KB/s, 0 seconds passed
    ...85%, 0.34 MB, 1649 KB/s, 0 seconds passed
    ...87%, 0.34 MB, 1686 KB/s, 0 seconds passed
    ...89%, 0.35 MB, 1724 KB/s, 0 seconds passed
    ...91%, 0.36 MB, 1761 KB/s, 0 seconds passed
    ...93%, 0.37 MB, 1798 KB/s, 0 seconds passed
    ...95%, 0.38 MB, 1832 KB/s, 0 seconds passed
    ...97%, 0.38 MB, 1869 KB/s, 0 seconds passed
    ...99%, 0.39 MB, 1906 KB/s, 0 seconds passed
    ...100%, 0.40 MB, 1943 KB/s, 0 seconds passed
+
    ...1%, 0.01 MB, 35 KB/s, 0 seconds passed
    ...3%, 0.02 MB, 71 KB/s, 0 seconds passed
    ...5%, 0.02 MB, 107 KB/s, 0 seconds passed
    ...7%, 0.03 MB, 142 KB/s, 0 seconds passed
    ...9%, 0.04 MB, 178 KB/s, 0 seconds passed
    ...11%, 0.05 MB, 213 KB/s, 0 seconds passed
    ...13%, 0.05 MB, 248 KB/s, 0 seconds passed
    ...15%, 0.06 MB, 283 KB/s, 0 seconds passed
    ...17%, 0.07 MB, 318 KB/s, 0 seconds passed
    ...19%, 0.08 MB, 353 KB/s, 0 seconds passed
    ...21%, 0.09 MB, 387 KB/s, 0 seconds passed
    ...23%, 0.09 MB, 422 KB/s, 0 seconds passed
    ...25%, 0.10 MB, 457 KB/s, 0 seconds passed
    ...27%, 0.11 MB, 490 KB/s, 0 seconds passed
    ...29%, 0.12 MB, 525 KB/s, 0 seconds passed
    ...31%, 0.12 MB, 559 KB/s, 0 seconds passed
    ...33%, 0.13 MB, 593 KB/s, 0 seconds passed
    ...35%, 0.14 MB, 627 KB/s, 0 seconds passed
    ...37%, 0.15 MB, 661 KB/s, 0 seconds passed
    ...39%, 0.16 MB, 695 KB/s, 0 seconds passed
    ...41%, 0.16 MB, 729 KB/s, 0 seconds pa
 ssed
    ...43%, 0.17 MB, 763 KB/s, 0 seconds passed
    ...45%, 0.18 MB, 797 KB/s, 0 seconds passed
    ...47%, 0.19 MB, 830 KB/s, 0 seconds passed
    ...49%, 0.20 MB, 863 KB/s, 0 seconds passed
    ...51%, 0.20 MB, 897 KB/s, 0 seconds passed
    ...53%, 0.21 MB, 929 KB/s, 0 seconds passed
    ...55%, 0.22 MB, 963 KB/s, 0 seconds passed
    ...57%, 0.23 MB, 997 KB/s, 0 seconds passed
    ...59%, 0.23 MB, 1031 KB/s, 0 seconds passed
    ...61%, 0.24 MB, 1062 KB/s, 0 seconds passed
    ...63%, 0.25 MB, 1096 KB/s, 0 seconds passed
    ...65%, 0.26 MB, 1129 KB/s, 0 seconds passed
    ...67%, 0.27 MB, 1163 KB/s, 0 seconds passed
    ...69%, 0.27 MB, 1193 KB/s, 0 seconds passed
    ...71%, 0.28 MB, 1227 KB/s, 0 seconds passed
    ...73%, 0.29 MB, 1261 KB/s, 0 seconds passed
    ...75%, 0.30 MB, 1294 KB/s, 0 seconds passed
    ...77%, 0.30 MB, 1324 KB/s, 0 seconds passed
    ...79%, 0.31 MB, 1357 KB/s, 0 seconds passed
    ...81%, 0.32 MB, 1391 KB/s, 0 seconds passed
    ...83%, 0.33 MB,
  1424 KB/s, 0 seconds passed
    ...85%, 0.34 MB, 1456 KB/s, 0 seconds passed
    ...87%, 0.34 MB, 1490 KB/s, 0 seconds passed
    ...89%, 0.35 MB, 1522 KB/s, 0 seconds passed
    ...91%, 0.36 MB, 1555 KB/s, 0 seconds passed
    ...93%, 0.37 MB, 1587 KB/s, 0 seconds passed
    ...95%, 0.38 MB, 1620 KB/s, 0 seconds passed
    ...97%, 0.38 MB, 1652 KB/s, 0 seconds passed
    ...99%, 0.39 MB, 1685 KB/s, 0 seconds passed
    ...100%, 0.40 MB, 1717 KB/s, 0 seconds passed
     Cannot find config for target=cuda -model=unknown, workload=('conv2d_nchw.cuda', ('TENSOR', (1, 64, 56, 56), 'float32'), ('TENSOR', (64, 64, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'float32'). A fallback configuration is used, which may bring great performance regression.
     Cannot find config for target=cuda -model=unknown, workload=('conv2d_nchw.cuda', ('TENSOR', (1, 128, 28, 28), 'float32'), ('TENSOR', (128, 128, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'float32'). A fallback configuration is used, which may bring great performance regression.
     Cannot find config for target=cuda -model=unknown, workload=('conv2d_nchw.cuda', ('TENSOR', (1, 256, 14, 14), 'float32'), ('TENSOR', (256, 256, 3, 3), 'float32'), (1, 1), (1, 1, 1, 1), (1, 1), 'float32'). A fallback configuration is used, which may bring great performance regression.
diff --git a/docs/_sources/tutorials/sg_execution_times.rst.txt b/docs/_sources/tutorials/sg_execution_times.rst.txt
index 3f23fab..3714e98 100644
--- a/docs/_sources/tutorials/sg_execution_times.rst.txt
+++ b/docs/_sources/tutorials/sg_execution_times.rst.txt
@@ -5,8 +5,8 @@
 
 Computation times
 =================
-**00:10.456** total execution time for **tutorials** files:
+**00:09.340** total execution time for **tutorials** files:
 
-- **00:09.989**: :ref:`sphx_glr_tutorials_relay_quick_start.py` (``relay_quick_start.py``)
-- **00:00.325**: :ref:`sphx_glr_tutorials_tensor_expr_get_started.py` (``tensor_expr_get_started.py``)
-- **00:00.142**: :ref:`sphx_glr_tutorials_cross_compilation_and_rpc.py` (``cross_compilation_and_rpc.py``)
+- **00:08.875**: :ref:`sphx_glr_tutorials_relay_quick_start.py` (``relay_quick_start.py``)
+- **00:00.324**: :ref:`sphx_glr_tutorials_tensor_expr_get_started.py` (``tensor_expr_get_started.py``)
+- **00:00.141**: :ref:`sphx_glr_tutorials_cross_compilation_and_rpc.py` (``cross_compilation_and_rpc.py``)
diff --git a/docs/_sources/tutorials/topi/intro_topi.rst.txt b/docs/_sources/tutorials/topi/intro_topi.rst.txt
index 77d1219..8ce6bfd 100644
--- a/docs/_sources/tutorials/topi/intro_topi.rst.txt
+++ b/docs/_sources/tutorials/topi/intro_topi.rst.txt
@@ -72,18 +72,21 @@ and to examine the IR code in human readable format, we can do
 
  .. code-block:: none
 
-    PrimFunc([A]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      // attr [B] storage_scope = "global"
-      allocate B[float32 * n]
-      for (i, 0, n) {
-        B[i] = 0f
-        for (k, 0, m) {
-          B[i] = (B[i] + A[((i*stride) + (k*stride))])
+    primfn(A_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {A: Buffer(A_2: handle, float32, [n: int32, m: int32], [stride: int32, stride_1: int32], type="auto")}
+      buffer_map = {A_1: A} {
+      attr [B: handle] "storage_scope" = "global";
+      allocate(B, float32, [n]);
+      for (i: int32, 0, n) {
+        B[i] = 0f32
+        for (k: int32, 0, m) {
+          B[i] = ((float32*)B[i]) + (float32*)A_2[((i*stride) + (k*stride_1))]))
         }
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -109,18 +112,21 @@ Fortunately, we can replace those two lines with simple :code:`topi.sum` much li
 
  .. code-block:: none
 
-    PrimFunc([A]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      // attr [A_red] storage_scope = "global"
-      allocate A_red[float32 * n]
-      for (ax0, 0, n) {
-        A_red[ax0] = 0f
-        for (k1, 0, m) {
-          A_red[ax0] = (A_red[ax0] + A[((ax0*stride) + (k1*stride))])
+    primfn(A_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {A: Buffer(A_2: handle, float32, [n: int32, m: int32], [stride: int32, stride_1: int32], type="auto")}
+      buffer_map = {A_1: A} {
+      attr [A_red: handle] "storage_scope" = "global";
+      allocate(A_red, float32, [n]);
+      for (ax0: int32, 0, n) {
+        A_red[ax0] = 0f32
+        for (k1: int32, 0, m) {
+          A_red[ax0] = ((float32*)A_red[ax0]) + (float32*)A_2[((ax0*stride) + (k1*stride_1))]))
         }
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -176,28 +182,33 @@ we can schedule the following series of operations ending with :code:`topi.sum`
 
  .. code-block:: none
 
-    PrimFunc([a, b]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      // attr [T_divide_red] storage_scope = "global"
-      allocate T_divide_red[float32 * 1]
-      // attr [iter_var(threadIdx.x, range(min=0, ext=1024), threadIdx.x)] thread_extent = 1024
-      // attr [T_divide_red.rf] storage_scope = "local"
-      allocate T_divide_red.rf[float32 * 1]
-      // attr [reduce_temp0] storage_scope = "local"
-      allocate reduce_temp0[float32 * 1]
-      T_divide_red.rf[0] = 0f
-      for (k0.k1.fused.k2.fused.outer, 0, 10) {
-        if (likely((((((k0.k1.fused.k2.fused.outer*1024) + threadIdx.x) < 10000) && (((k0.k1.fused.k2.fused.outer*1024) + threadIdx.x) < 10000)) && (((k0.k1.fused.k2.fused.outer*1024) + threadIdx.x) < 10000)))) {
-          T_divide_red.rf[0] = (T_divide_red.rf[0] + (((a[((k0.k1.fused.k2.fused.outer*1024) + threadIdx.x)] + b[floormod(((k0.k1.fused.k2.fused.outer*1024) + threadIdx.x), 100)]) + (a[((k0.k1.fused.k2.fused.outer*1024) + threadIdx.x)]*b[floormod(((k0.k1.fused.k2.fused.outer*1024) + threadIdx.x), 100)]))*0.5f))
+    primfn(a_1: handle, b_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {a: Buffer(a_2: handle, float32, [100, 10, 10], []),
+                 b: Buffer(b_2: handle, float32, [10, 10], [])}
+      buffer_map = {b_1: b, a_1: a} {
+      attr [T_divide_red: handle] "storage_scope" = "global";
+      allocate(T_divide_red, float32, [1]);
+      attr [IterVar(threadIdx.x: int32, [0:1024], "ThreadIndex", "threadIdx.x")] "thread_extent" = 1024;
+      attr [T_divide_red.rf: handle] "storage_scope" = "local";
+      allocate(T_divide_red.rf, float32, [1]);
+      attr [reduce_temp0: handle] "storage_scope" = "local";
+      allocate(reduce_temp0, float32, [1]) {
+        T_divide_red.rf[0] = 0f32
+        for (k0.k1.fused.k2.fused.outer: int32, 0, 10) {
+          if @likely((((((k0.k1.fused.k2.fused.outer*1024) + threadIdx.x) < 10000) && (((k0.k1.fused.k2.fused.outer*1024) + threadIdx.x) < 10000)) && (((k0.k1.fused.k2.fused.outer*1024) + threadIdx.x) < 10000)), dtype=bool, type="pure_intrin", index=0) {
+            T_divide_red.rf[0] = ((float32*)T_divide_red.rf[0]) + ((((float32*)a_2[((k0.k1.fused.k2.fused.outer*1024) + threadIdx.x)]) + (float32*)b_2[floormod(((k0.k1.fused.k2.fused.outer*1024) + threadIdx.x), 100)])) + ((float32*)a_2[((k0.k1.fused.k2.fused.outer*1024) + threadIdx.x)])*(float32*)b_2[floormod(((k0.k1.fused.k2.fused.outer*1024) + threadIdx.x), 100)])))*0.5f32))
+          }
+        }
+        attr [meta[CommReducer][0]] "reduce_scope" = @reinterpret(0u64, dtype=handle, type="pure_intrin", index=0);
+        @tvm_thread_allreduce(1u32, (float32*)T_divide_red.rf[0]), True, reduce_temp0, threadIdx.x, dtype=handle, type="intrin", index=0)
+        if (threadIdx.x == 0) {
+          T_divide_red[0] = (float32*)reduce_temp0[0])
         }
-      }
-      // attr [comm_reducer(result=[(x + y)], lhs=[x], rhs=[y], identity_element=[0f])] reduce_scope = reinterpret((uint64)0)
-      tvm_thread_allreduce((uint32)1, T_divide_red.rf[0], (bool)1, reduce_temp0, threadIdx.x)
-      if ((threadIdx.x == 0)) {
-        T_divide_red[0] = reduce_temp0[0]
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -219,7 +230,7 @@ As you can see, scheduled stages of computation have been accumulated and we can
 
  .. code-block:: none
 
-    [stage(a, 0x128bf87c0), stage(b, 0x15056c1b0), stage(T_add, 0x1310db130), stage(T_multiply, 0xbbaf8fc0), stage(T_elemwise_sum, 0x13110d5b0), stage(T_divide, 0x12d92eea0), stage(T_divide_red.rf, 0x128539d90), stage(T_divide_red, 0x12c4df270)]
+    [stage(a, 0x1297b7ee0), stage(b, 0x128ee4530), stage(T_add, 0x12d420070), stage(T_multiply, 0x13ac7b5c0), stage(T_elemwise_sum, 0x130d16410), stage(T_divide, 0xaadccc50), stage(T_divide_red.rf, 0x14003d290), stage(T_divide_red, 0xbfadc4c0)]
 
 
 
@@ -268,43 +279,49 @@ TOPI also provides common neural nets operations such as _softmax_ with optimize
 
  .. code-block:: none
 
-    PrimFunc([tarray]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      // attr [T_softmax_maxelem] storage_scope = "global"
-      allocate T_softmax_maxelem[float32 * 512]
-      // attr [T_softmax_exp] storage_scope = "global"
-      allocate T_softmax_exp[float32 * 262144]
-      // attr [iter_var(blockIdx.x, , blockIdx.x)] thread_extent = 512
-      T_softmax_maxelem[blockIdx.x] = -3.40282e+38f
-      for (k, 0, 512) {
-        T_softmax_maxelem[blockIdx.x] = max(T_softmax_maxelem[blockIdx.x], tarray[((blockIdx.x*512) + k)])
-      }
-      // attr [iter_var(blockIdx.x, , blockIdx.x)] thread_extent = 512
-      for (i1, 0, 512) {
-        T_softmax_exp[((blockIdx.x*512) + i1)] = exp((tarray[((blockIdx.x*512) + i1)] - T_softmax_maxelem[blockIdx.x]))
-      }
-      // attr [iter_var(blockIdx.x, , blockIdx.x)] thread_extent = 512
-      // attr [T_softmax_expsum.rf] storage_scope = "local"
-      allocate T_softmax_expsum.rf[float32 * 1]
-      // attr [reduce_temp0] storage_scope = "local"
-      allocate reduce_temp0[float32 * 1]
-      // attr [iter_var(threadIdx.x, range(min=0, ext=64), threadIdx.x)] thread_extent = 64
-      T_softmax_expsum.rf[0] = 0f
-      for (k.outer, 0, 8) {
-        T_softmax_expsum.rf[0] = (T_softmax_expsum.rf[0] + T_softmax_exp[(((blockIdx.x*512) + (k.outer*64)) + threadIdx.x)])
-      }
-      // attr [comm_reducer(result=[(x + y)], lhs=[x], rhs=[y], identity_element=[0f])] reduce_scope = reinterpret((uint64)0)
-      tvm_thread_allreduce((uint32)1, T_softmax_expsum.rf[0], (bool)1, reduce_temp0, threadIdx.x)
-      if ((threadIdx.x == 0)) {
-        T_softmax_maxelem[blockIdx.x] = reduce_temp0[0]
-      }
-      // attr [iter_var(blockIdx.x, , blockIdx.x)] thread_extent = 512
-      // attr [iter_var(threadIdx.x, range(min=0, ext=64), threadIdx.x)] thread_extent = 64
-      for (i1.inner, 0, 8) {
-        T_softmax_exp[(((blockIdx.x*512) + (threadIdx.x*8)) + i1.inner)] = (T_softmax_exp[(((blockIdx.x*512) + (threadIdx.x*8)) + i1.inner)]/T_softmax_maxelem[blockIdx.x])
+    primfn(tarray_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {tarray: Buffer(tarray_2: handle, float32, [512, 512], [])}
+      buffer_map = {tarray_1: tarray} {
+      attr [T_softmax_maxelem: handle] "storage_scope" = "global";
+      allocate(T_softmax_maxelem, float32, [512]);
+      attr [T_softmax_exp: handle] "storage_scope" = "global";
+      allocate(T_softmax_exp, float32, [262144]) {
+        attr [IterVar(blockIdx.x: int32, (nullptr), "ThreadIndex", "blockIdx.x")] "thread_extent" = 512 {
+          T_softmax_maxelem[blockIdx.x] = -3.40282e+38f32
+          for (k: int32, 0, 512) {
+            T_softmax_maxelem[blockIdx.x] = max((float32*)T_softmax_maxelem[blockIdx.x]), (float32*)tarray_2[((blockIdx.x*512) + k)]))
+          }
+        }
+        attr [IterVar(blockIdx.x, (nullptr), "ThreadIndex", "blockIdx.x")] "thread_extent" = 512;
+        for (i1: int32, 0, 512) {
+          T_softmax_exp[((blockIdx.x*512) + i1)] = @exp(((float32*)tarray_2[((blockIdx.x*512) + i1)]) - (float32*)T_softmax_maxelem[blockIdx.x])), dtype=float32, type="pure_intrin", index=0)
+        }
+        attr [IterVar(blockIdx.x, (nullptr), "ThreadIndex", "blockIdx.x")] "thread_extent" = 512;
+        attr [T_softmax_expsum.rf: handle] "storage_scope" = "local";
+        allocate(T_softmax_expsum.rf, float32, [1]);
+        attr [reduce_temp0: handle] "storage_scope" = "local";
+        allocate(reduce_temp0, float32, [1]);
+        attr [IterVar(threadIdx.x: int32, [0:64], "ThreadIndex", "threadIdx.x")] "thread_extent" = 64 {
+          T_softmax_expsum.rf[0] = 0f32
+          for (k.outer: int32, 0, 8) {
+            T_softmax_expsum.rf[0] = ((float32*)T_softmax_expsum.rf[0]) + (float32*)T_softmax_exp[(((blockIdx.x*512) + (k.outer*64)) + threadIdx.x)]))
+          }
+          attr [meta[CommReducer][0]] "reduce_scope" = @reinterpret(0u64, dtype=handle, type="pure_intrin", index=0);
+          @tvm_thread_allreduce(1u32, (float32*)T_softmax_expsum.rf[0]), True, reduce_temp0, threadIdx.x, dtype=handle, type="intrin", index=0)
+          if (threadIdx.x == 0) {
+            T_softmax_maxelem[blockIdx.x] = (float32*)reduce_temp0[0])
+          }
+        }
+        attr [IterVar(blockIdx.x, (nullptr), "ThreadIndex", "blockIdx.x")] "thread_extent" = 512;
+        attr [IterVar(threadIdx.x, [0:64], "ThreadIndex", "threadIdx.x")] "thread_extent" = 64;
+        for (i1.inner: int32, 0, 8) {
+          T_softmax_exp[(((blockIdx.x*512) + (threadIdx.x*8)) + i1.inner)] = ((float32*)T_softmax_exp[(((blockIdx.x*512) + (threadIdx.x*8)) + i1.inner)]) / (float32*)T_softmax_maxelem[blockIdx.x]))
+        }
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -343,266 +360,276 @@ We can fuse :code:`topi.nn.conv2d` and :code:`topi.nn.relu` together.
 
  .. code-block:: none
 
-    PrimFunc([placeholder, placeholder]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      // attr [compute] storage_scope = "global"
-      allocate compute[float32 * 501760]
-      // attr [iter_var(blockIdx.z, , blockIdx.z)] thread_extent = 5
-      // attr [compute] storage_scope = "local"
-      allocate compute[float32 * 14]
-      // attr [pad_temp.shared] storage_scope = "shared"
-      allocate pad_temp.shared[float32 * 112]
-      // attr [placeholder.shared] storage_scope = "shared"
-      allocate placeholder.shared[float32 * 2]
-      // attr [iter_var(blockIdx.y, , blockIdx.y)] thread_extent = 224
-      // attr [iter_var(blockIdx.x, , blockIdx.x)] thread_extent = 2
-      // attr [iter_var(threadIdx.z, , threadIdx.z)] thread_extent = 1
-      // attr [iter_var(threadIdx.y, , threadIdx.y)] thread_extent = 1
-      // attr [iter_var(threadIdx.x, , threadIdx.x)] thread_extent = 16
-      compute[0] = 0f
-      compute[2] = 0f
-      compute[4] = 0f
-      compute[6] = 0f
-      compute[8] = 0f
-      compute[10] = 0f
-      compute[12] = 0f
-      compute[1] = 0f
-      compute[3] = 0f
-      compute[5] = 0f
-      compute[7] = 0f
-      compute[9] = 0f
-      compute[11] = 0f
-      compute[13] = 0f
-      for (rc.outer, 0, 3) {
-        for (ry.outer, 0, 5) {
-          // attr [iter_var(threadIdx.z, , threadIdx.z)] thread_extent = 1
-          // attr [iter_var(threadIdx.y, , threadIdx.y)] thread_extent = 1
-          // attr [iter_var(threadIdx.x, , threadIdx.x)] thread_extent = 16
-          pad_temp.shared[(threadIdx.x*7)] = tvm_if_then_else((((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)) && (2 <= ((blockIdx.x*112) + (threadIdx.x*7)))), placeholder[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x*7)) - 450)], 0f)
-          pad_temp.shared[((threadIdx.x*7) + 1)] = tvm_if_then_else((((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)) && (1 <= ((blockIdx.x*112) + (threadIdx.x*7)))), placeholder[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x*7)) - 449)], 0f)
-          pad_temp.shared[((threadIdx.x*7) + 2)] = tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x*7)) - 448)], 0f)
-          pad_temp.shared[((threadIdx.x*7) + 3)] = tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x*7)) - 447)], 0f)
-          pad_temp.shared[((threadIdx.x*7) + 4)] = tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x*7)) - 446)], 0f)
-          pad_temp.shared[((threadIdx.x*7) + 5)] = tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x*7)) - 445)], 0f)
-          pad_temp.shared[((threadIdx.x*7) + 6)] = tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x*7)) - 444)], 0f)
-          // attr [iter_var(threadIdx.z, , threadIdx.z)] thread_extent = 1
-          // attr [iter_var(threadIdx.y, , threadIdx.y)] thread_extent = 1
-          // attr [iter_var(threadIdx.x, , threadIdx.x)] thread_extent = 16
-          if (likely((threadIdx.x < 2))) {
-            if (likely((threadIdx.x < 2))) {
-              if (likely((threadIdx.x < 2))) {
-                if (likely((threadIdx.x < 2))) {
-                  if (likely((threadIdx.x < 2))) {
-                    if (likely((threadIdx.x < 2))) {
-                      if (likely((((blockIdx.z*2) + threadIdx.x) < 10))) {
-                        placeholder.shared[threadIdx.x] = placeholder[((((blockIdx.z*150) + (threadIdx.x*75)) + (rc.outer*25)) + (ry.outer*5))]
+    primfn(placeholder_2: handle, placeholder_3: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {placeholder_1: Buffer(placeholder_4: handle, float32, [1, 3, 224, 224], []),
+                 placeholder: Buffer(placeholder_5: handle, float32, [10, 3, 5, 5], [])}
+      buffer_map = {placeholder_3: placeholder, placeholder_2: placeholder_1} {
+      attr [compute: handle] "storage_scope" = "global";
+      allocate(compute, float32, [501760]);
+      attr [IterVar(blockIdx.z: int32, (nullptr), "ThreadIndex", "blockIdx.z")] "thread_extent" = 5;
+      attr [compute_1: handle] "storage_scope" = "local";
+      allocate(compute_1, float32, [14]);
+      attr [pad_temp.shared: handle] "storage_scope" = "shared";
+      allocate(pad_temp.shared, float32, [112]);
+      attr [placeholder.shared: handle] "storage_scope" = "shared";
+      allocate(placeholder.shared, float32, [2]);
+      attr [IterVar(blockIdx.y: int32, (nullptr), "ThreadIndex", "blockIdx.y")] "thread_extent" = 224;
+      attr [IterVar(blockIdx.x: int32, (nullptr), "ThreadIndex", "blockIdx.x")] "thread_extent" = 2;
+      attr [IterVar(threadIdx.z: int32, (nullptr), "ThreadIndex", "threadIdx.z")] "thread_extent" = 1;
+      attr [IterVar(threadIdx.y: int32, (nullptr), "ThreadIndex", "threadIdx.y")] "thread_extent" = 1;
+      attr [IterVar(threadIdx.x: int32, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 16 {
+        compute_1[0] = 0f32
+        compute_1[2] = 0f32
+        compute_1[4] = 0f32
+        compute_1[6] = 0f32
+        compute_1[8] = 0f32
+        compute_1[10] = 0f32
+        compute_1[12] = 0f32
+        compute_1[1] = 0f32
+        compute_1[3] = 0f32
+        compute_1[5] = 0f32
+        compute_1[7] = 0f32
+        compute_1[9] = 0f32
+        compute_1[11] = 0f32
+        compute_1[13] = 0f32
+        for (rc.outer: int32, 0, 3) {
+          for (ry.outer: int32, 0, 5) {
+            attr [IterVar(threadIdx.z_1: int32, (nullptr), "ThreadIndex", "threadIdx.z")] "thread_extent" = 1;
+            attr [IterVar(threadIdx.y_1: int32, (nullptr), "ThreadIndex", "threadIdx.y")] "thread_extent" = 1;
+            attr [IterVar(threadIdx.x_1: int32, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 16 {
+              pad_temp.shared[(threadIdx.x_1*7)] = @tvm_if_then_else((((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)) && (2 <= ((blockIdx.x*112) + (threadIdx.x_1*7)))), (float32*)placeholder_4[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 450)]), 0f32, dtype=float32, type="pure_intrin", index=0)
+              pad_temp.shared[((threadIdx.x_1*7) + 1)] = @tvm_if_then_else((((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)) && (1 <= ((blockIdx.x*112) + (threadIdx.x_1*7)))), (float32*)placeholder_4[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 449)]), 0f32, dtype=float32, type="pure_intrin", index=0)
+              pad_temp.shared[((threadIdx.x_1*7) + 2)] = @tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), (float32*)placeholder_4[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 448)]), 0f32, dtype=float32, type="pure_intrin", index=0)
+              pad_temp.shared[((threadIdx.x_1*7) + 3)] = @tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), (float32*)placeholder_4[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 447)]), 0f32, dtype=float32, type="pure_intrin", index=0)
+              pad_temp.shared[((threadIdx.x_1*7) + 4)] = @tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), (float32*)placeholder_4[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 446)]), 0f32, dtype=float32, type="pure_intrin", index=0)
+              pad_temp.shared[((threadIdx.x_1*7) + 5)] = @tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), (float32*)placeholder_4[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 445)]), 0f32, dtype=float32, type="pure_intrin", index=0)
+              pad_temp.shared[((threadIdx.x_1*7) + 6)] = @tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), (float32*)placeholder_4[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 444)]), 0f32, dtype=float32, type="pure_intrin", index=0)
+            }
+            attr [IterVar(threadIdx.z_2: int32, (nullptr), "ThreadIndex", "threadIdx.z")] "thread_extent" = 1;
+            attr [IterVar(threadIdx.y_2: int32, (nullptr), "ThreadIndex", "threadIdx.y")] "thread_extent" = 1;
+            attr [IterVar(threadIdx.x_2: int32, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 16;
+            if @likely((threadIdx.x_2 < 2), dtype=bool, type="pure_intrin", index=0) {
+              if @likely((threadIdx.x_2 < 2), dtype=bool, type="pure_intrin", index=0) {
+                if @likely((threadIdx.x_2 < 2), dtype=bool, type="pure_intrin", index=0) {
+                  if @likely((threadIdx.x_2 < 2), dtype=bool, type="pure_intrin", index=0) {
+                    if @likely((threadIdx.x_2 < 2), dtype=bool, type="pure_intrin", index=0) {
+                      if @likely((threadIdx.x_2 < 2), dtype=bool, type="pure_intrin", index=0) {
+                        if @likely((((blockIdx.z*2) + threadIdx.x_2) < 10), dtype=bool, type="pure_intrin", index=0) {
+                          placeholder.shared[threadIdx.x_2] = (float32*)placeholder_5[((((blockIdx.z*150) + (threadIdx.x_2*75)) + (rc.outer*25)) + (ry.outer*5))])
+                        }
                       }
                     }
                   }
                 }
               }
             }
-          }
-          compute[0] = (compute[0] + (pad_temp.shared[threadIdx.x]*placeholder.shared[0]))
-          compute[2] = (compute[2] + (pad_temp.shared[(threadIdx.x + 16)]*placeholder.shared[0]))
-          compute[4] = (compute[4] + (pad_temp.shared[(threadIdx.x + 32)]*placeholder.shared[0]))
-          compute[6] = (compute[6] + (pad_temp.shared[(threadIdx.x + 48)]*placeholder.shared[0]))
-          compute[8] = (compute[8] + (pad_temp.shared[(threadIdx.x + 64)]*placeholder.shared[0]))
-          compute[10] = (compute[10] + (pad_temp.shared[(threadIdx.x + 80)]*placeholder.shared[0]))
-          compute[12] = (compute[12] + (pad_temp.shared[(threadIdx.x + 96)]*placeholder.shared[0]))
-          compute[1] = (compute[1] + (pad_temp.shared[threadIdx.x]*placeholder.shared[1]))
-          compute[3] = (compute[3] + (pad_temp.shared[(threadIdx.x + 16)]*placeholder.shared[1]))
-          compute[5] = (compute[5] + (pad_temp.shared[(threadIdx.x + 32)]*placeholder.shared[1]))
-          compute[7] = (compute[7] + (pad_temp.shared[(threadIdx.x + 48)]*placeholder.shared[1]))
-          compute[9] = (compute[9] + (pad_temp.shared[(threadIdx.x + 64)]*placeholder.shared[1]))
-          compute[11] = (compute[11] + (pad_temp.shared[(threadIdx.x + 80)]*placeholder.shared[1]))
-          compute[13] = (compute[13] + (pad_temp.shared[(threadIdx.x + 96)]*placeholder.shared[1]))
-          // attr [iter_var(threadIdx.z, , threadIdx.z)] thread_extent = 1
-          // attr [iter_var(threadIdx.y, , threadIdx.y)] thread_extent = 1
-          // attr [iter_var(threadIdx.x, , threadIdx.x)] thread_extent = 16
-          pad_temp.shared[(threadIdx.x*7)] = tvm_if_then_else((((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)) && (1 <= ((blockIdx.x*112) + (threadIdx.x*7)))), placeholder[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x*7)) - 449)], 0f)
-          pad_temp.shared[((threadIdx.x*7) + 1)] = tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x*7)) - 448)], 0f)
-          pad_temp.shared[((threadIdx.x*7) + 2)] = tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x*7)) - 447)], 0f)
-          pad_temp.shared[((threadIdx.x*7) + 3)] = tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x*7)) - 446)], 0f)
-          pad_temp.shared[((threadIdx.x*7) + 4)] = tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x*7)) - 445)], 0f)
-          pad_temp.shared[((threadIdx.x*7) + 5)] = tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x*7)) - 444)], 0f)
-          pad_temp.shared[((threadIdx.x*7) + 6)] = tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x*7)) - 443)], 0f)
-          // attr [iter_var(threadIdx.z, , threadIdx.z)] thread_extent = 1
-          // attr [iter_var(threadIdx.y, , threadIdx.y)] thread_extent = 1
-          // attr [iter_var(threadIdx.x, , threadIdx.x)] thread_extent = 16
-          if (likely((threadIdx.x < 2))) {
-            if (likely((threadIdx.x < 2))) {
-              if (likely((threadIdx.x < 2))) {
-                if (likely((threadIdx.x < 2))) {
-                  if (likely((threadIdx.x < 2))) {
-                    if (likely((threadIdx.x < 2))) {
-                      if (likely((((blockIdx.z*2) + threadIdx.x) < 10))) {
-                        placeholder.shared[threadIdx.x] = placeholder[(((((blockIdx.z*150) + (threadIdx.x*75)) + (rc.outer*25)) + (ry.outer*5)) + 1)]
+            compute_1[0] = ((float32*)compute_1[0]) + ((float32*)pad_temp.shared[threadIdx.x])*(float32*)placeholder.shared[0])))
+            compute_1[2] = ((float32*)compute_1[2]) + ((float32*)pad_temp.shared[(threadIdx.x + 16)])*(float32*)placeholder.shared[0])))
+            compute_1[4] = ((float32*)compute_1[4]) + ((float32*)pad_temp.shared[(threadIdx.x + 32)])*(float32*)placeholder.shared[0])))
+            compute_1[6] = ((float32*)compute_1[6]) + ((float32*)pad_temp.shared[(threadIdx.x + 48)])*(float32*)placeholder.shared[0])))
+            compute_1[8] = ((float32*)compute_1[8]) + ((float32*)pad_temp.shared[(threadIdx.x + 64)])*(float32*)placeholder.shared[0])))
+            compute_1[10] = ((float32*)compute_1[10]) + ((float32*)pad_temp.shared[(threadIdx.x + 80)])*(float32*)placeholder.shared[0])))
+            compute_1[12] = ((float32*)compute_1[12]) + ((float32*)pad_temp.shared[(threadIdx.x + 96)])*(float32*)placeholder.shared[0])))
+            compute_1[1] = ((float32*)compute_1[1]) + ((float32*)pad_temp.shared[threadIdx.x])*(float32*)placeholder.shared[1])))
+            compute_1[3] = ((float32*)compute_1[3]) + ((float32*)pad_temp.shared[(threadIdx.x + 16)])*(float32*)placeholder.shared[1])))
+            compute_1[5] = ((float32*)compute_1[5]) + ((float32*)pad_temp.shared[(threadIdx.x + 32)])*(float32*)placeholder.shared[1])))
+            compute_1[7] = ((float32*)compute_1[7]) + ((float32*)pad_temp.shared[(threadIdx.x + 48)])*(float32*)placeholder.shared[1])))
+            compute_1[9] = ((float32*)compute_1[9]) + ((float32*)pad_temp.shared[(threadIdx.x + 64)])*(float32*)placeholder.shared[1])))
+            compute_1[11] = ((float32*)compute_1[11]) + ((float32*)pad_temp.shared[(threadIdx.x + 80)])*(float32*)placeholder.shared[1])))
+            compute_1[13] = ((float32*)compute_1[13]) + ((float32*)pad_temp.shared[(threadIdx.x + 96)])*(float32*)placeholder.shared[1])))
+            attr [IterVar(threadIdx.z_1, (nullptr), "ThreadIndex", "threadIdx.z")] "thread_extent" = 1;
+            attr [IterVar(threadIdx.y_1, (nullptr), "ThreadIndex", "threadIdx.y")] "thread_extent" = 1;
+            attr [IterVar(threadIdx.x_1, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 16 {
+              pad_temp.shared[(threadIdx.x_1*7)] = @tvm_if_then_else((((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)) && (1 <= ((blockIdx.x*112) + (threadIdx.x_1*7)))), (float32*)placeholder_4[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 449)]), 0f32, dtype=float32, type="pure_intrin", index=0)
+              pad_temp.shared[((threadIdx.x_1*7) + 1)] = @tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), (float32*)placeholder_4[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 448)]), 0f32, dtype=float32, type="pure_intrin", index=0)
+              pad_temp.shared[((threadIdx.x_1*7) + 2)] = @tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), (float32*)placeholder_4[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 447)]), 0f32, dtype=float32, type="pure_intrin", index=0)
+              pad_temp.shared[((threadIdx.x_1*7) + 3)] = @tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), (float32*)placeholder_4[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 446)]), 0f32, dtype=float32, type="pure_intrin", index=0)
+              pad_temp.shared[((threadIdx.x_1*7) + 4)] = @tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), (float32*)placeholder_4[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 445)]), 0f32, dtype=float32, type="pure_intrin", index=0)
+              pad_temp.shared[((threadIdx.x_1*7) + 5)] = @tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), (float32*)placeholder_4[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 444)]), 0f32, dtype=float32, type="pure_intrin", index=0)
+              pad_temp.shared[((threadIdx.x_1*7) + 6)] = @tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), (float32*)placeholder_4[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 443)]), 0f32, dtype=float32, type="pure_intrin", index=0)
+            }
+            attr [IterVar(threadIdx.z_2, (nullptr), "ThreadIndex", "threadIdx.z")] "thread_extent" = 1;
+            attr [IterVar(threadIdx.y_2, (nullptr), "ThreadIndex", "threadIdx.y")] "thread_extent" = 1;
+            attr [IterVar(threadIdx.x_2, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 16;
+            if @likely((threadIdx.x_2 < 2), dtype=bool, type="pure_intrin", index=0) {
+              if @likely((threadIdx.x_2 < 2), dtype=bool, type="pure_intrin", index=0) {
+                if @likely((threadIdx.x_2 < 2), dtype=bool, type="pure_intrin", index=0) {
+                  if @likely((threadIdx.x_2 < 2), dtype=bool, type="pure_intrin", index=0) {
+                    if @likely((threadIdx.x_2 < 2), dtype=bool, type="pure_intrin", index=0) {
+                      if @likely((threadIdx.x_2 < 2), dtype=bool, type="pure_intrin", index=0) {
+                        if @likely((((blockIdx.z*2) + threadIdx.x_2) < 10), dtype=bool, type="pure_intrin", index=0) {
+                          placeholder.shared[threadIdx.x_2] = (float32*)placeholder_5[(((((blockIdx.z*150) + (threadIdx.x_2*75)) + (rc.outer*25)) + (ry.outer*5)) + 1)])
+                        }
                       }
                     }
                   }
                 }
               }
             }
-          }
-          compute[0] = (compute[0] + (pad_temp.shared[threadIdx.x]*placeholder.shared[0]))
-          compute[2] = (compute[2] + (pad_temp.shared[(threadIdx.x + 16)]*placeholder.shared[0]))
-          compute[4] = (compute[4] + (pad_temp.shared[(threadIdx.x + 32)]*placeholder.shared[0]))
-          compute[6] = (compute[6] + (pad_temp.shared[(threadIdx.x + 48)]*placeholder.shared[0]))
-          compute[8] = (compute[8] + (pad_temp.shared[(threadIdx.x + 64)]*placeholder.shared[0]))
-          compute[10] = (compute[10] + (pad_temp.shared[(threadIdx.x + 80)]*placeholder.shared[0]))
-          compute[12] = (compute[12] + (pad_temp.shared[(threadIdx.x + 96)]*placeholder.shared[0]))
-          compute[1] = (compute[1] + (pad_temp.shared[threadIdx.x]*placeholder.shared[1]))
-          compute[3] = (compute[3] + (pad_temp.shared[(threadIdx.x + 16)]*placeholder.shared[1]))
-          compute[5] = (compute[5] + (pad_temp.shared[(threadIdx.x + 32)]*placeholder.shared[1]))
-          compute[7] = (compute[7] + (pad_temp.shared[(threadIdx.x + 48)]*placeholder.shared[1]))
-          compute[9] = (compute[9] + (pad_temp.shared[(threadIdx.x + 64)]*placeholder.shared[1]))
-          compute[11] = (compute[11] + (pad_temp.shared[(threadIdx.x + 80)]*placeholder.shared[1]))
-          compute[13] = (compute[13] + (pad_temp.shared[(threadIdx.x + 96)]*placeholder.shared[1]))
-          // attr [iter_var(threadIdx.z, , threadIdx.z)] thread_extent = 1
-          // attr [iter_var(threadIdx.y, , threadIdx.y)] thread_extent = 1
-          // attr [iter_var(threadIdx.x, , threadIdx.x)] thread_extent = 16
-          pad_temp.shared[(threadIdx.x*7)] = tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x*7)) - 448)], 0f)
-          pad_temp.shared[((threadIdx.x*7) + 1)] = tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x*7)) - 447)], 0f)
-          pad_temp.shared[((threadIdx.x*7) + 2)] = tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x*7)) - 446)], 0f)
-          pad_temp.shared[((threadIdx.x*7) + 3)] = tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x*7)) - 445)], 0f)
-          pad_temp.shared[((threadIdx.x*7) + 4)] = tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x*7)) - 444)], 0f)
-          pad_temp.shared[((threadIdx.x*7) + 5)] = tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x*7)) - 443)], 0f)
-          pad_temp.shared[((threadIdx.x*7) + 6)] = tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x*7)) - 442)], 0f)
-          // attr [iter_var(threadIdx.z, , threadIdx.z)] thread_extent = 1
-          // attr [iter_var(threadIdx.y, , threadIdx.y)] thread_extent = 1
-          // attr [iter_var(threadIdx.x, , threadIdx.x)] thread_extent = 16
-          if (likely((threadIdx.x < 2))) {
-            if (likely((threadIdx.x < 2))) {
-              if (likely((threadIdx.x < 2))) {
-                if (likely((threadIdx.x < 2))) {
-                  if (likely((threadIdx.x < 2))) {
-                    if (likely((threadIdx.x < 2))) {
-                      if (likely((((blockIdx.z*2) + threadIdx.x) < 10))) {
-                        placeholder.shared[threadIdx.x] = placeholder[(((((blockIdx.z*150) + (threadIdx.x*75)) + (rc.outer*25)) + (ry.outer*5)) + 2)]
+            compute_1[0] = ((float32*)compute_1[0]) + ((float32*)pad_temp.shared[threadIdx.x])*(float32*)placeholder.shared[0])))
+            compute_1[2] = ((float32*)compute_1[2]) + ((float32*)pad_temp.shared[(threadIdx.x + 16)])*(float32*)placeholder.shared[0])))
+            compute_1[4] = ((float32*)compute_1[4]) + ((float32*)pad_temp.shared[(threadIdx.x + 32)])*(float32*)placeholder.shared[0])))
+            compute_1[6] = ((float32*)compute_1[6]) + ((float32*)pad_temp.shared[(threadIdx.x + 48)])*(float32*)placeholder.shared[0])))
+            compute_1[8] = ((float32*)compute_1[8]) + ((float32*)pad_temp.shared[(threadIdx.x + 64)])*(float32*)placeholder.shared[0])))
+            compute_1[10] = ((float32*)compute_1[10]) + ((float32*)pad_temp.shared[(threadIdx.x + 80)])*(float32*)placeholder.shared[0])))
+            compute_1[12] = ((float32*)compute_1[12]) + ((float32*)pad_temp.shared[(threadIdx.x + 96)])*(float32*)placeholder.shared[0])))
+            compute_1[1] = ((float32*)compute_1[1]) + ((float32*)pad_temp.shared[threadIdx.x])*(float32*)placeholder.shared[1])))
+            compute_1[3] = ((float32*)compute_1[3]) + ((float32*)pad_temp.shared[(threadIdx.x + 16)])*(float32*)placeholder.shared[1])))
+            compute_1[5] = ((float32*)compute_1[5]) + ((float32*)pad_temp.shared[(threadIdx.x + 32)])*(float32*)placeholder.shared[1])))
+            compute_1[7] = ((float32*)compute_1[7]) + ((float32*)pad_temp.shared[(threadIdx.x + 48)])*(float32*)placeholder.shared[1])))
+            compute_1[9] = ((float32*)compute_1[9]) + ((float32*)pad_temp.shared[(threadIdx.x + 64)])*(float32*)placeholder.shared[1])))
+            compute_1[11] = ((float32*)compute_1[11]) + ((float32*)pad_temp.shared[(threadIdx.x + 80)])*(float32*)placeholder.shared[1])))
+            compute_1[13] = ((float32*)compute_1[13]) + ((float32*)pad_temp.shared[(threadIdx.x + 96)])*(float32*)placeholder.shared[1])))
+            attr [IterVar(threadIdx.z_1, (nullptr), "ThreadIndex", "threadIdx.z")] "thread_extent" = 1;
+            attr [IterVar(threadIdx.y_1, (nullptr), "ThreadIndex", "threadIdx.y")] "thread_extent" = 1;
+            attr [IterVar(threadIdx.x_1, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 16 {
+              pad_temp.shared[(threadIdx.x_1*7)] = @tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), (float32*)placeholder_4[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 448)]), 0f32, dtype=float32, type="pure_intrin", index=0)
+              pad_temp.shared[((threadIdx.x_1*7) + 1)] = @tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), (float32*)placeholder_4[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 447)]), 0f32, dtype=float32, type="pure_intrin", index=0)
+              pad_temp.shared[((threadIdx.x_1*7) + 2)] = @tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), (float32*)placeholder_4[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 446)]), 0f32, dtype=float32, type="pure_intrin", index=0)
+              pad_temp.shared[((threadIdx.x_1*7) + 3)] = @tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), (float32*)placeholder_4[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 445)]), 0f32, dtype=float32, type="pure_intrin", index=0)
+              pad_temp.shared[((threadIdx.x_1*7) + 4)] = @tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), (float32*)placeholder_4[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 444)]), 0f32, dtype=float32, type="pure_intrin", index=0)
+              pad_temp.shared[((threadIdx.x_1*7) + 5)] = @tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), (float32*)placeholder_4[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 443)]), 0f32, dtype=float32, type="pure_intrin", index=0)
+              pad_temp.shared[((threadIdx.x_1*7) + 6)] = @tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), (float32*)placeholder_4[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 442)]), 0f32, dtype=float32, type="pure_intrin", index=0)
+            }
+            attr [IterVar(threadIdx.z_2, (nullptr), "ThreadIndex", "threadIdx.z")] "thread_extent" = 1;
+            attr [IterVar(threadIdx.y_2, (nullptr), "ThreadIndex", "threadIdx.y")] "thread_extent" = 1;
+            attr [IterVar(threadIdx.x_2, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 16;
+            if @likely((threadIdx.x_2 < 2), dtype=bool, type="pure_intrin", index=0) {
+              if @likely((threadIdx.x_2 < 2), dtype=bool, type="pure_intrin", index=0) {
+                if @likely((threadIdx.x_2 < 2), dtype=bool, type="pure_intrin", index=0) {
+                  if @likely((threadIdx.x_2 < 2), dtype=bool, type="pure_intrin", index=0) {
+                    if @likely((threadIdx.x_2 < 2), dtype=bool, type="pure_intrin", index=0) {
+                      if @likely((threadIdx.x_2 < 2), dtype=bool, type="pure_intrin", index=0) {
+                        if @likely((((blockIdx.z*2) + threadIdx.x_2) < 10), dtype=bool, type="pure_intrin", index=0) {
+                          placeholder.shared[threadIdx.x_2] = (float32*)placeholder_5[(((((blockIdx.z*150) + (threadIdx.x_2*75)) + (rc.outer*25)) + (ry.outer*5)) + 2)])
+                        }
                       }
                     }
                   }
                 }
               }
             }
-          }
-          compute[0] = (compute[0] + (pad_temp.shared[threadIdx.x]*placeholder.shared[0]))
-          compute[2] = (compute[2] + (pad_temp.shared[(threadIdx.x + 16)]*placeholder.shared[0]))
-          compute[4] = (compute[4] + (pad_temp.shared[(threadIdx.x + 32)]*placeholder.shared[0]))
-          compute[6] = (compute[6] + (pad_temp.shared[(threadIdx.x + 48)]*placeholder.shared[0]))
-          compute[8] = (compute[8] + (pad_temp.shared[(threadIdx.x + 64)]*placeholder.shared[0]))
-          compute[10] = (compute[10] + (pad_temp.shared[(threadIdx.x + 80)]*placeholder.shared[0]))
-          compute[12] = (compute[12] + (pad_temp.shared[(threadIdx.x + 96)]*placeholder.shared[0]))
-          compute[1] = (compute[1] + (pad_temp.shared[threadIdx.x]*placeholder.shared[1]))
-          compute[3] = (compute[3] + (pad_temp.shared[(threadIdx.x + 16)]*placeholder.shared[1]))
-          compute[5] = (compute[5] + (pad_temp.shared[(threadIdx.x + 32)]*placeholder.shared[1]))
-          compute[7] = (compute[7] + (pad_temp.shared[(threadIdx.x + 48)]*placeholder.shared[1]))
-          compute[9] = (compute[9] + (pad_temp.shared[(threadIdx.x + 64)]*placeholder.shared[1]))
-          compute[11] = (compute[11] + (pad_temp.shared[(threadIdx.x + 80)]*placeholder.shared[1]))
-          compute[13] = (compute[13] + (pad_temp.shared[(threadIdx.x + 96)]*placeholder.shared[1]))
-          // attr [iter_var(threadIdx.z, , threadIdx.z)] thread_extent = 1
-          // attr [iter_var(threadIdx.y, , threadIdx.y)] thread_extent = 1
-          // attr [iter_var(threadIdx.x, , threadIdx.x)] thread_extent = 16
-          pad_temp.shared[(threadIdx.x*7)] = tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x*7)) - 447)], 0f)
-          pad_temp.shared[((threadIdx.x*7) + 1)] = tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x*7)) - 446)], 0f)
-          pad_temp.shared[((threadIdx.x*7) + 2)] = tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x*7)) - 445)], 0f)
-          pad_temp.shared[((threadIdx.x*7) + 3)] = tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x*7)) - 444)], 0f)
-          pad_temp.shared[((threadIdx.x*7) + 4)] = tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x*7)) - 443)], 0f)
-          pad_temp.shared[((threadIdx.x*7) + 5)] = tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x*7)) - 442)], 0f)
-          pad_temp.shared[((threadIdx.x*7) + 6)] = tvm_if_then_else((((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)) && (((blockIdx.x*112) + (threadIdx.x*7)) < 217)), placeholder[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x*7)) - 441)], 0f)
-          // attr [iter_var(threadIdx.z, , threadIdx.z)] thread_extent = 1
-          // attr [iter_var(threadIdx.y, , threadIdx.y)] thread_extent = 1
-          // attr [iter_var(threadIdx.x, , threadIdx.x)] thread_extent = 16
-          if (likely((threadIdx.x < 2))) {
-            if (likely((threadIdx.x < 2))) {
-              if (likely((threadIdx.x < 2))) {
-                if (likely((threadIdx.x < 2))) {
-                  if (likely((threadIdx.x < 2))) {
-                    if (likely((threadIdx.x < 2))) {
-                      if (likely((((blockIdx.z*2) + threadIdx.x) < 10))) {
-                        placeholder.shared[threadIdx.x] = placeholder[(((((blockIdx.z*150) + (threadIdx.x*75)) + (rc.outer*25)) + (ry.outer*5)) + 3)]
+            compute_1[0] = ((float32*)compute_1[0]) + ((float32*)pad_temp.shared[threadIdx.x])*(float32*)placeholder.shared[0])))
+            compute_1[2] = ((float32*)compute_1[2]) + ((float32*)pad_temp.shared[(threadIdx.x + 16)])*(float32*)placeholder.shared[0])))
+            compute_1[4] = ((float32*)compute_1[4]) + ((float32*)pad_temp.shared[(threadIdx.x + 32)])*(float32*)placeholder.shared[0])))
+            compute_1[6] = ((float32*)compute_1[6]) + ((float32*)pad_temp.shared[(threadIdx.x + 48)])*(float32*)placeholder.shared[0])))
+            compute_1[8] = ((float32*)compute_1[8]) + ((float32*)pad_temp.shared[(threadIdx.x + 64)])*(float32*)placeholder.shared[0])))
+            compute_1[10] = ((float32*)compute_1[10]) + ((float32*)pad_temp.shared[(threadIdx.x + 80)])*(float32*)placeholder.shared[0])))
+            compute_1[12] = ((float32*)compute_1[12]) + ((float32*)pad_temp.shared[(threadIdx.x + 96)])*(float32*)placeholder.shared[0])))
+            compute_1[1] = ((float32*)compute_1[1]) + ((float32*)pad_temp.shared[threadIdx.x])*(float32*)placeholder.shared[1])))
+            compute_1[3] = ((float32*)compute_1[3]) + ((float32*)pad_temp.shared[(threadIdx.x + 16)])*(float32*)placeholder.shared[1])))
+            compute_1[5] = ((float32*)compute_1[5]) + ((float32*)pad_temp.shared[(threadIdx.x + 32)])*(float32*)placeholder.shared[1])))
+            compute_1[7] = ((float32*)compute_1[7]) + ((float32*)pad_temp.shared[(threadIdx.x + 48)])*(float32*)placeholder.shared[1])))
+            compute_1[9] = ((float32*)compute_1[9]) + ((float32*)pad_temp.shared[(threadIdx.x + 64)])*(float32*)placeholder.shared[1])))
+            compute_1[11] = ((float32*)compute_1[11]) + ((float32*)pad_temp.shared[(threadIdx.x + 80)])*(float32*)placeholder.shared[1])))
+            compute_1[13] = ((float32*)compute_1[13]) + ((float32*)pad_temp.shared[(threadIdx.x + 96)])*(float32*)placeholder.shared[1])))
+            attr [IterVar(threadIdx.z_1, (nullptr), "ThreadIndex", "threadIdx.z")] "thread_extent" = 1;
+            attr [IterVar(threadIdx.y_1, (nullptr), "ThreadIndex", "threadIdx.y")] "thread_extent" = 1;
+            attr [IterVar(threadIdx.x_1, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 16 {
+              pad_temp.shared[(threadIdx.x_1*7)] = @tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), (float32*)placeholder_4[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 447)]), 0f32, dtype=float32, type="pure_intrin", index=0)
+              pad_temp.shared[((threadIdx.x_1*7) + 1)] = @tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), (float32*)placeholder_4[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 446)]), 0f32, dtype=float32, type="pure_intrin", index=0)
+              pad_temp.shared[((threadIdx.x_1*7) + 2)] = @tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), (float32*)placeholder_4[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 445)]), 0f32, dtype=float32, type="pure_intrin", index=0)
+              pad_temp.shared[((threadIdx.x_1*7) + 3)] = @tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), (float32*)placeholder_4[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 444)]), 0f32, dtype=float32, type="pure_intrin", index=0)
+              pad_temp.shared[((threadIdx.x_1*7) + 4)] = @tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), (float32*)placeholder_4[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 443)]), 0f32, dtype=float32, type="pure_intrin", index=0)
+              pad_temp.shared[((threadIdx.x_1*7) + 5)] = @tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), (float32*)placeholder_4[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 442)]), 0f32, dtype=float32, type="pure_intrin", index=0)
+              pad_temp.shared[((threadIdx.x_1*7) + 6)] = @tvm_if_then_else((((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)) && (((blockIdx.x*112) + (threadIdx.x_1*7)) < 217)), (float32*)placeholder_4[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 441)]), 0f32, dtype=float32, type="pure_intrin", index=0)
+            }
+            attr [IterVar(threadIdx.z_2, (nullptr), "ThreadIndex", "threadIdx.z")] "thread_extent" = 1;
+            attr [IterVar(threadIdx.y_2, (nullptr), "ThreadIndex", "threadIdx.y")] "thread_extent" = 1;
+            attr [IterVar(threadIdx.x_2, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 16;
+            if @likely((threadIdx.x_2 < 2), dtype=bool, type="pure_intrin", index=0) {
+              if @likely((threadIdx.x_2 < 2), dtype=bool, type="pure_intrin", index=0) {
+                if @likely((threadIdx.x_2 < 2), dtype=bool, type="pure_intrin", index=0) {
+                  if @likely((threadIdx.x_2 < 2), dtype=bool, type="pure_intrin", index=0) {
+                    if @likely((threadIdx.x_2 < 2), dtype=bool, type="pure_intrin", index=0) {
+                      if @likely((threadIdx.x_2 < 2), dtype=bool, type="pure_intrin", index=0) {
+                        if @likely((((blockIdx.z*2) + threadIdx.x_2) < 10), dtype=bool, type="pure_intrin", index=0) {
+                          placeholder.shared[threadIdx.x_2] = (float32*)placeholder_5[(((((blockIdx.z*150) + (threadIdx.x_2*75)) + (rc.outer*25)) + (ry.outer*5)) + 3)])
+                        }
                       }
                     }
                   }
                 }
               }
             }
-          }
-          compute[0] = (compute[0] + (pad_temp.shared[threadIdx.x]*placeholder.shared[0]))
-          compute[2] = (compute[2] + (pad_temp.shared[(threadIdx.x + 16)]*placeholder.shared[0]))
-          compute[4] = (compute[4] + (pad_temp.shared[(threadIdx.x + 32)]*placeholder.shared[0]))
-          compute[6] = (compute[6] + (pad_temp.shared[(threadIdx.x + 48)]*placeholder.shared[0]))
-          compute[8] = (compute[8] + (pad_temp.shared[(threadIdx.x + 64)]*placeholder.shared[0]))
-          compute[10] = (compute[10] + (pad_temp.shared[(threadIdx.x + 80)]*placeholder.shared[0]))
-          compute[12] = (compute[12] + (pad_temp.shared[(threadIdx.x + 96)]*placeholder.shared[0]))
-          compute[1] = (compute[1] + (pad_temp.shared[threadIdx.x]*placeholder.shared[1]))
-          compute[3] = (compute[3] + (pad_temp.shared[(threadIdx.x + 16)]*placeholder.shared[1]))
-          compute[5] = (compute[5] + (pad_temp.shared[(threadIdx.x + 32)]*placeholder.shared[1]))
-          compute[7] = (compute[7] + (pad_temp.shared[(threadIdx.x + 48)]*placeholder.shared[1]))
-          compute[9] = (compute[9] + (pad_temp.shared[(threadIdx.x + 64)]*placeholder.shared[1]))
-          compute[11] = (compute[11] + (pad_temp.shared[(threadIdx.x + 80)]*placeholder.shared[1]))
-          compute[13] = (compute[13] + (pad_temp.shared[(threadIdx.x + 96)]*placeholder.shared[1]))
-          // attr [iter_var(threadIdx.z, , threadIdx.z)] thread_extent = 1
-          // attr [iter_var(threadIdx.y, , threadIdx.y)] thread_extent = 1
-          // attr [iter_var(threadIdx.x, , threadIdx.x)] thread_extent = 16
-          pad_temp.shared[(threadIdx.x*7)] = tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x*7)) - 446)], 0f)
-          pad_temp.shared[((threadIdx.x*7) + 1)] = tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x*7)) - 445)], 0f)
-          pad_temp.shared[((threadIdx.x*7) + 2)] = tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x*7)) - 444)], 0f)
-          pad_temp.shared[((threadIdx.x*7) + 3)] = tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x*7)) - 443)], 0f)
-          pad_temp.shared[((threadIdx.x*7) + 4)] = tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), placeholder[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x*7)) - 442)], 0f)
-          pad_temp.shared[((threadIdx.x*7) + 5)] = tvm_if_then_else((((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)) && (((blockIdx.x*112) + (threadIdx.x*7)) < 217)), placeholder[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x*7)) - 441)], 0f)
-          pad_temp.shared[((threadIdx.x*7) + 6)] = tvm_if_then_else((((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)) && (((blockIdx.x*112) + (threadIdx.x*7)) < 216)), placeholder[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x*7)) - 440)], 0f)
-          // attr [iter_var(threadIdx.z, , threadIdx.z)] thread_extent = 1
-          // attr [iter_var(threadIdx.y, , threadIdx.y)] thread_extent = 1
-          // attr [iter_var(threadIdx.x, , threadIdx.x)] thread_extent = 16
-          if (likely((threadIdx.x < 2))) {
-            if (likely((threadIdx.x < 2))) {
-              if (likely((threadIdx.x < 2))) {
-                if (likely((threadIdx.x < 2))) {
-                  if (likely((threadIdx.x < 2))) {
-                    if (likely((threadIdx.x < 2))) {
-                      if (likely((((blockIdx.z*2) + threadIdx.x) < 10))) {
-                        placeholder.shared[threadIdx.x] = placeholder[(((((blockIdx.z*150) + (threadIdx.x*75)) + (rc.outer*25)) + (ry.outer*5)) + 4)]
+            compute_1[0] = ((float32*)compute_1[0]) + ((float32*)pad_temp.shared[threadIdx.x])*(float32*)placeholder.shared[0])))
+            compute_1[2] = ((float32*)compute_1[2]) + ((float32*)pad_temp.shared[(threadIdx.x + 16)])*(float32*)placeholder.shared[0])))
+            compute_1[4] = ((float32*)compute_1[4]) + ((float32*)pad_temp.shared[(threadIdx.x + 32)])*(float32*)placeholder.shared[0])))
+            compute_1[6] = ((float32*)compute_1[6]) + ((float32*)pad_temp.shared[(threadIdx.x + 48)])*(float32*)placeholder.shared[0])))
+            compute_1[8] = ((float32*)compute_1[8]) + ((float32*)pad_temp.shared[(threadIdx.x + 64)])*(float32*)placeholder.shared[0])))
+            compute_1[10] = ((float32*)compute_1[10]) + ((float32*)pad_temp.shared[(threadIdx.x + 80)])*(float32*)placeholder.shared[0])))
+            compute_1[12] = ((float32*)compute_1[12]) + ((float32*)pad_temp.shared[(threadIdx.x + 96)])*(float32*)placeholder.shared[0])))
+            compute_1[1] = ((float32*)compute_1[1]) + ((float32*)pad_temp.shared[threadIdx.x])*(float32*)placeholder.shared[1])))
+            compute_1[3] = ((float32*)compute_1[3]) + ((float32*)pad_temp.shared[(threadIdx.x + 16)])*(float32*)placeholder.shared[1])))
+            compute_1[5] = ((float32*)compute_1[5]) + ((float32*)pad_temp.shared[(threadIdx.x + 32)])*(float32*)placeholder.shared[1])))
+            compute_1[7] = ((float32*)compute_1[7]) + ((float32*)pad_temp.shared[(threadIdx.x + 48)])*(float32*)placeholder.shared[1])))
+            compute_1[9] = ((float32*)compute_1[9]) + ((float32*)pad_temp.shared[(threadIdx.x + 64)])*(float32*)placeholder.shared[1])))
+            compute_1[11] = ((float32*)compute_1[11]) + ((float32*)pad_temp.shared[(threadIdx.x + 80)])*(float32*)placeholder.shared[1])))
+            compute_1[13] = ((float32*)compute_1[13]) + ((float32*)pad_temp.shared[(threadIdx.x + 96)])*(float32*)placeholder.shared[1])))
+            attr [IterVar(threadIdx.z_1, (nullptr), "ThreadIndex", "threadIdx.z")] "thread_extent" = 1;
+            attr [IterVar(threadIdx.y_1, (nullptr), "ThreadIndex", "threadIdx.y")] "thread_extent" = 1;
+            attr [IterVar(threadIdx.x_1, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 16 {
+              pad_temp.shared[(threadIdx.x_1*7)] = @tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), (float32*)placeholder_4[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 446)]), 0f32, dtype=float32, type="pure_intrin", index=0)
+              pad_temp.shared[((threadIdx.x_1*7) + 1)] = @tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), (float32*)placeholder_4[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 445)]), 0f32, dtype=float32, type="pure_intrin", index=0)
+              pad_temp.shared[((threadIdx.x_1*7) + 2)] = @tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), (float32*)placeholder_4[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 444)]), 0f32, dtype=float32, type="pure_intrin", index=0)
+              pad_temp.shared[((threadIdx.x_1*7) + 3)] = @tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), (float32*)placeholder_4[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 443)]), 0f32, dtype=float32, type="pure_intrin", index=0)
+              pad_temp.shared[((threadIdx.x_1*7) + 4)] = @tvm_if_then_else(((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)), (float32*)placeholder_4[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 442)]), 0f32, dtype=float32, type="pure_intrin", index=0)
+              pad_temp.shared[((threadIdx.x_1*7) + 5)] = @tvm_if_then_else((((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)) && (((blockIdx.x*112) + (threadIdx.x_1*7)) < 217)), (float32*)placeholder_4[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 441)]), 0f32, dtype=float32, type="pure_intrin", index=0)
+              pad_temp.shared[((threadIdx.x_1*7) + 6)] = @tvm_if_then_else((((2 <= (blockIdx.y + ry.outer)) && ((blockIdx.y + ry.outer) < 226)) && (((blockIdx.x*112) + (threadIdx.x_1*7)) < 216)), (float32*)placeholder_4[((((((rc.outer*50176) + (blockIdx.y*224)) + (ry.outer*224)) + (blockIdx.x*112)) + (threadIdx.x_1*7)) - 440)]), 0f32, dtype=float32, type="pure_intrin", index=0)
+            }
+            attr [IterVar(threadIdx.z_2, (nullptr), "ThreadIndex", "threadIdx.z")] "thread_extent" = 1;
+            attr [IterVar(threadIdx.y_2, (nullptr), "ThreadIndex", "threadIdx.y")] "thread_extent" = 1;
+            attr [IterVar(threadIdx.x_2, (nullptr), "ThreadIndex", "threadIdx.x")] "thread_extent" = 16;
+            if @likely((threadIdx.x_2 < 2), dtype=bool, type="pure_intrin", index=0) {
+              if @likely((threadIdx.x_2 < 2), dtype=bool, type="pure_intrin", index=0) {
+                if @likely((threadIdx.x_2 < 2), dtype=bool, type="pure_intrin", index=0) {
+                  if @likely((threadIdx.x_2 < 2), dtype=bool, type="pure_intrin", index=0) {
+                    if @likely((threadIdx.x_2 < 2), dtype=bool, type="pure_intrin", index=0) {
+                      if @likely((threadIdx.x_2 < 2), dtype=bool, type="pure_intrin", index=0) {
+                        if @likely((((blockIdx.z*2) + threadIdx.x_2) < 10), dtype=bool, type="pure_intrin", index=0) {
+                          placeholder.shared[threadIdx.x_2] = (float32*)placeholder_5[(((((blockIdx.z*150) + (threadIdx.x_2*75)) + (rc.outer*25)) + (ry.outer*5)) + 4)])
+                        }
                       }
                     }
                   }
                 }
               }
             }
+            compute_1[0] = ((float32*)compute_1[0]) + ((float32*)pad_temp.shared[threadIdx.x])*(float32*)placeholder.shared[0])))
+            compute_1[2] = ((float32*)compute_1[2]) + ((float32*)pad_temp.shared[(threadIdx.x + 16)])*(float32*)placeholder.shared[0])))
+            compute_1[4] = ((float32*)compute_1[4]) + ((float32*)pad_temp.shared[(threadIdx.x + 32)])*(float32*)placeholder.shared[0])))
+            compute_1[6] = ((float32*)compute_1[6]) + ((float32*)pad_temp.shared[(threadIdx.x + 48)])*(float32*)placeholder.shared[0])))
+            compute_1[8] = ((float32*)compute_1[8]) + ((float32*)pad_temp.shared[(threadIdx.x + 64)])*(float32*)placeholder.shared[0])))
+            compute_1[10] = ((float32*)compute_1[10]) + ((float32*)pad_temp.shared[(threadIdx.x + 80)])*(float32*)placeholder.shared[0])))
+            compute_1[12] = ((float32*)compute_1[12]) + ((float32*)pad_temp.shared[(threadIdx.x + 96)])*(float32*)placeholder.shared[0])))
+            compute_1[1] = ((float32*)compute_1[1]) + ((float32*)pad_temp.shared[threadIdx.x])*(float32*)placeholder.shared[1])))
+            compute_1[3] = ((float32*)compute_1[3]) + ((float32*)pad_temp.shared[(threadIdx.x + 16)])*(float32*)placeholder.shared[1])))
+            compute_1[5] = ((float32*)compute_1[5]) + ((float32*)pad_temp.shared[(threadIdx.x + 32)])*(float32*)placeholder.shared[1])))
+            compute_1[7] = ((float32*)compute_1[7]) + ((float32*)pad_temp.shared[(threadIdx.x + 48)])*(float32*)placeholder.shared[1])))
+            compute_1[9] = ((float32*)compute_1[9]) + ((float32*)pad_temp.shared[(threadIdx.x + 64)])*(float32*)placeholder.shared[1])))
+            compute_1[11] = ((float32*)compute_1[11]) + ((float32*)pad_temp.shared[(threadIdx.x + 80)])*(float32*)placeholder.shared[1])))
+            compute_1[13] = ((float32*)compute_1[13]) + ((float32*)pad_temp.shared[(threadIdx.x + 96)])*(float32*)placeholder.shared[1])))
           }
-          compute[0] = (compute[0] + (pad_temp.shared[threadIdx.x]*placeholder.shared[0]))
-          compute[2] = (compute[2] + (pad_temp.shared[(threadIdx.x + 16)]*placeholder.shared[0]))
-          compute[4] = (compute[4] + (pad_temp.shared[(threadIdx.x + 32)]*placeholder.shared[0]))
-          compute[6] = (compute[6] + (pad_temp.shared[(threadIdx.x + 48)]*placeholder.shared[0]))
-          compute[8] = (compute[8] + (pad_temp.shared[(threadIdx.x + 64)]*placeholder.shared[0]))
-          compute[10] = (compute[10] + (pad_temp.shared[(threadIdx.x + 80)]*placeholder.shared[0]))
-          compute[12] = (compute[12] + (pad_temp.shared[(threadIdx.x + 96)]*placeholder.shared[0]))
-          compute[1] = (compute[1] + (pad_temp.shared[threadIdx.x]*placeholder.shared[1]))
-          compute[3] = (compute[3] + (pad_temp.shared[(threadIdx.x + 16)]*placeholder.shared[1]))
-          compute[5] = (compute[5] + (pad_temp.shared[(threadIdx.x + 32)]*placeholder.shared[1]))
-          compute[7] = (compute[7] + (pad_temp.shared[(threadIdx.x + 48)]*placeholder.shared[1]))
-          compute[9] = (compute[9] + (pad_temp.shared[(threadIdx.x + 64)]*placeholder.shared[1]))
-          compute[11] = (compute[11] + (pad_temp.shared[(threadIdx.x + 80)]*placeholder.shared[1]))
-          compute[13] = (compute[13] + (pad_temp.shared[(threadIdx.x + 96)]*placeholder.shared[1]))
         }
+        compute[((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x)] = max((float32*)compute_1[0]), 0f32)
+        compute[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 16)] = max((float32*)compute_1[2]), 0f32)
+        compute[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 32)] = max((float32*)compute_1[4]), 0f32)
+        compute[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 48)] = max((float32*)compute_1[6]), 0f32)
+        compute[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 64)] = max((float32*)compute_1[8]), 0f32)
+        compute[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 80)] = max((float32*)compute_1[10]), 0f32)
+        compute[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 96)] = max((float32*)compute_1[12]), 0f32)
+        compute[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 50176)] = max((float32*)compute_1[1]), 0f32)
+        compute[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 50192)] = max((float32*)compute_1[3]), 0f32)
+        compute[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 50208)] = max((float32*)compute_1[5]), 0f32)
+        compute[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 50224)] = max((float32*)compute_1[7]), 0f32)
+        compute[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 50240)] = max((float32*)compute_1[9]), 0f32)
+        compute[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 50256)] = max((float32*)compute_1[11]), 0f32)
+        compute[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 50272)] = max((float32*)compute_1[13]), 0f32)
       }
-      compute[((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x)] = max(compute[0], 0f)
-      compute[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 16)] = max(compute[2], 0f)
-      compute[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 32)] = max(compute[4], 0f)
-      compute[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 48)] = max(compute[6], 0f)
-      compute[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 64)] = max(compute[8], 0f)
-      compute[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 80)] = max(compute[10], 0f)
-      compute[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 96)] = max(compute[12], 0f)
-      compute[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 50176)] = max(compute[1], 0f)
-      compute[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 50192)] = max(compute[3], 0f)
-      compute[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 50208)] = max(compute[5], 0f)
-      compute[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 50224)] = max(compute[7], 0f)
-      compute[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 50240)] = max(compute[9], 0f)
-      compute[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 50256)] = max(compute[11], 0f)
-      compute[(((((blockIdx.z*100352) + (blockIdx.y*224)) + (blockIdx.x*112)) + threadIdx.x) + 50272)] = max(compute[13], 0f)
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
diff --git a/docs/_sources/tutorials/topi/sg_execution_times.rst.txt b/docs/_sources/tutorials/topi/sg_execution_times.rst.txt
index 79da3d0..9db192a 100644
--- a/docs/_sources/tutorials/topi/sg_execution_times.rst.txt
+++ b/docs/_sources/tutorials/topi/sg_execution_times.rst.txt
@@ -5,6 +5,6 @@
 
 Computation times
 =================
-**00:00.481** total execution time for **tutorials_topi** files:
+**00:00.500** total execution time for **tutorials_topi** files:
 
-- **00:00.481**: :ref:`sphx_glr_tutorials_topi_intro_topi.py` (``intro_topi.py``)
+- **00:00.500**: :ref:`sphx_glr_tutorials_topi_intro_topi.py` (``intro_topi.py``)
diff --git a/docs/_sources/vta/install.rst.txt b/docs/_sources/vta/install.rst.txt
index 900a014..b68fab7 100644
--- a/docs/_sources/vta/install.rst.txt
+++ b/docs/_sources/vta/install.rst.txt
@@ -19,16 +19,16 @@ VTA Installation Guide
 ======================
 
 We present three installation guides, each extending on the previous one:
-1. ref:`VTA Simulator Installation`
-2. ref:`VTA Test Setup for Xilinx Pynq FPGA`
-3. ref:`VTA Test Setup for Intel FPGA`
-4. ref:`VTA FPGA Toolchain Installation`
 
+1. `Simulator Installation`_
+2. `Xilinx Pynq FPGA Setup`_
+3. `Intel DE10 FPGA Setup`_
+4. `Bitstream Generation with Xilinx Toolchains`_
+5. `Bitstream Generation with Intel Toolchains`_
 
-.. _VTA Simulator Installation:
 
-VTA Simulator Installation
---------------------------
+Simulator Installation
+----------------------
 
 You need `TVM installed <https://tvm.apache.org/docs/install/index.html>`_ on your machine.
 For a quick and easy start, checkout the `Docker Guide <https://tvm.apache.org/docs/install/docker.html>`_.
@@ -65,20 +65,19 @@ To ensure that you've properly installed the VTA python package, run the followi
 
    python <tvm root>/vta/tests/python/integration/test_benchmark_topi_conv2d.py
 
-> Note: You'll notice that for every convolution layer, the throughput gets reported in GOPS. These numbers are actually the computational throughput that the simulator achieves, by evaluating the convolutions in software.
-
 You are invited to try out our `VTA programming tutorials <https://tvm.apache.org/docs/vta/tutorials/index.html>`_.
 
+   **Note**: You'll notice that for every convolution layer, the throughput gets reported in GOPS. These numbers are actually the computational throughput that the simulator achieves, by evaluating the convolutions in software.
 
 Advanced Configuration (optional)
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
 VTA is a generic configurable deep learning accelerator.
-The configuration is specified by `vta_config.json` under `3rdparty/vta-hw/config`.
+The configuration is specified by ``vta_config.json`` under ``3rdparty/vta-hw/config``.
 This file provides an architectural specification of the VTA accelerator to parameterize the TVM compiler stack and the VTA hardware stack.
 
 The VTA configuration file also specifies the TVM compiler target.
-When `TARGET` is set to `sim`, all TVM workloads execute on the VTA simulator.
+When ``TARGET`` is set to ``sim``, all TVM workloads execute on the VTA simulator.
 You can modify the content of the configuration file to rebuild VTA to a different parameterization.
 To do so,
 
@@ -90,19 +89,20 @@ To do so,
    make
 
 
-.. _VTA Test Setup for Xilinx Pynq FPGA:
 
-VTA Test Setup for Xilinx Pynq FPGA
------------------------------------
+Xilinx Pynq FPGA Setup
+----------------------
 
 This second guide extends the *VTA Simulator Installation* guide above to run FPGA hardware tests of the complete TVM and VTA software-hardware stack.
 In terms of hardware components you'll need:
+
 * The `Pynq <http://www.pynq.io/>`_ FPGA development board which can be acquired for $200, or $150 for academics from `Digilent <https://store.digilentinc.com/pynq-z1-python-productivity-for-zynq/>`_.
 * An Ethernet-to-USB adapter to connect the Pynq board to your development machine.
 * An 8+GB micro SD card.
 * An AC to DC 12V 3A power adapter.
 
 This guide covers the following themes:
+
 1. Pynq board setup instructions.
 2. Pynq-side RPC server build and deployment.
 3. Revisiting the test examples from the *VTA Simulator Installation* guide, this time executing on the Pynq board.
@@ -111,8 +111,10 @@ Pynq Board Setup
 ^^^^^^^^^^^^^^^^
 
 Setup your Pynq board based on the `Pynq board getting started tutorial <http://pynq.readthedocs.io/en/latest/getting_started.html>`_.
+
 You should follow the instructions up to and including the *Turning On the PYNQ-Z1* step (no need to pursue the tutorial beyond this point).
-* Make sure that you've downloaded the latest Pynq image, `PYNQ-Z1 v2.4 <http://www.pynq.io/board.html>`_(released February 22rd 2019), and have imaged your SD card with it (we recommend the free `Etcher <https://etcher.io/>`_ program).
+
+* Make sure that you've downloaded the latest Pynq image, `PYNQ-Z1 v2.4 <http://www.pynq.io/board.html>`_ (released February 22rd 2019), and have imaged your SD card with it (we recommend the free `Etcher <https://etcher.io/>`_ program).
 * For this test setup, follow the `"Connect to a Computer" <http://pynq.readthedocs.io/en/latest/getting_started.html#connect-to-a-computer>`_ Ethernet setup instructions. To be able to talk to the board, make sure to `assign your computer a static IP address <http://pynq.readthedocs.io/en/latest/appendix.html#assign-your-computer-a-static-ip>`_
 
 Once the board is powered on and connected to your development machine, try connecting to it to make sure you've properly set up your Pynq board:
@@ -159,14 +161,17 @@ The build process should take roughly 5 minutes.
    sudo ./apps/vta_rpc/start_rpc_server.sh # pw is 'xilinx'
 
 
-You should see the following being displayed when starting the RPC server. In order to run the next examples, you'll need to leave the RPC server running in an `ssh` session.
-```
-INFO:root:RPCServer: bind to 0.0.0.0:9091
-```
+You should see the following being displayed when starting the RPC server. In order to run the next examples, you'll need to leave the RPC server running in an ``ssh`` session.
+
+.. code:: bash
+
+   INFO:root:RPCServer: bind to 0.0.0.0:9091
+
 
 Tips regarding the Pynq RPC Server:
-* The RPC server should be listening on port `9091`. If not, an earlier process might have terminated unexpectedly and it's recommended in this case to just reboot the Pynq, and re-run the RPC server.
-* To kill the RPC server, just send the `Ctrl + c` command. You can re-run it with `sudo ./apps/pynq_rpc/start_rpc_server.sh`.
+
+* The RPC server should be listening on port ``9091``. If not, an earlier process might have terminated unexpectedly and it's recommended in this case to just reboot the Pynq, and re-run the RPC server.
+* To kill the RPC server, just send the ``Ctrl + c`` command. You can re-run it with ``sudo ./apps/pynq_rpc/start_rpc_server.sh``.
 * If unresponsive, the board can be rebooted by power-cycling it with the physical power switch.
 
 Testing your Pynq-based Hardware Setup
@@ -181,7 +186,7 @@ Before running the examples on your development machine, you'll need to configur
    export VTA_RPC_PORT=9091
 
 
-In addition, you'll need to edit the `vta_config.json` file on the host to indicate that we are targeting the Pynq platform, by setting the `TARGET` field to `"pynq"`.
+In addition, you'll need to edit the ``vta_config.json`` file on the host to indicate that we are targeting the Pynq platform, by setting the ``TARGET`` field to ``"pynq"``.
 > Note: in contrast to our simulation setup, there are no libraries to compile on the host side since the host offloads all of the computation to the Pynq board.
 
 .. code:: bash
@@ -193,16 +198,15 @@ In addition, you'll need to edit the `vta_config.json` file on the host to indic
 
 This time again, we will run the 2D convolution testbench.
 Beforehand, we need to program the Pynq board FPGA with a VTA bitstream, and build the VTA runtime via RPC.
-The following `test_program_rpc.py` script will perform two operations:
-* FPGA programming, by downloading a pre-compiled bitstream from a `VTA bitstream repository <https://github.com/uwsaml/vta-distro>`_ that matches the default `vta_config.json` configuration set by the host, and sending it over to the Pynq via RPC to program the Pynq's FPGA.
-* Runtime building on the Pynq, which needs to be run every time the `vta_config.json` configuration is modified. This ensures that the VTA software runtime that generates the accelerator's executable via just-in-time (JIT) compilation matches the specifications of the VTA design that is programmed on the FPGA. The build process takes about 30 seconds to complete so be patient!
+The following ``test_program_rpc.py`` script will perform two operations:
+* FPGA programming, by downloading a pre-compiled bitstream from a `VTA bitstream repository <https://github.com/uwsaml/vta-distro>`_ that matches the default ``vta_config.json`` configuration set by the host, and sending it over to the Pynq via RPC to program the Pynq's FPGA.
+* Runtime building on the Pynq, which needs to be run every time the ``vta_config.json`` configuration is modified. This ensures that the VTA software runtime that generates the accelerator's executable via just-in-time (JIT) compilation matches the specifications of the VTA design that is programmed on the FPGA. The build process takes about 30 seconds to complete so be patient!
 
 .. code:: bash
 
    # On the Host-side
    python <tvm root>/vta/tests/python/pynq/test_program_rpc.py
 
-> Tip: You can track progress of the FPGA programming and the runtime rebuilding steps by looking at the RPC server's logging messages in your Pynq `ssh` session.
 
 We are now ready to run the 2D convolution testbench in hardware.
 
@@ -213,13 +217,14 @@ We are now ready to run the 2D convolution testbench in hardware.
 
 The performance metrics measured on the Pynq board will be reported for each convolutional layer.
 
+**Tip**: You can track progress of the FPGA programming and the runtime rebuilding steps by looking at the RPC server's logging messages in your Pynq ``ssh`` session.
+
 You can also try out our `VTA programming tutorials <https://tvm.apache.org/docs/vta/tutorials/index.html>`_.
 
 
-.. _VTA Test Setup for Intel FPGA:
 
-VTA Test Setup for Intel FPGA
------------------------------
+Intel DE10 FPGA Setup
+---------------------
 
 Similar to the PYNQ side setup steps, this third guide bring us the details on how can we setup up the Linux environment for Intel FPGA boards like DE10-Nano.
 
@@ -249,9 +254,9 @@ First, extract the gzipped archive file.
 
    tar xf de10-nano-image-Angstrom-v2016.12.socfpga-sdimg.2017.03.31.tgz
 
-This would produce a single SD card image named `de10-nano-image-Angstrom-v2016.12.socfpga-sdimg` (approx. 2.4 GB), it contains all the file systems to boot Angstrom Linux.
+This would produce a single SD card image named ``de10-nano-image-Angstrom-v2016.12.socfpga-sdimg`` (approx. 2.4 GB), it contains all the file systems to boot Angstrom Linux.
 
-Second, plugin a SD card that is ready to flash in your PC, and identify the device id for the disk with `fdisk -l`, or `gparted` if you feel better to use GUI. The typical device id for your disk would likely to be `/dev/sdb`.
+Second, plugin a SD card that is ready to flash in your PC, and identify the device id for the disk with ``fdisk -l``, or ``gparted`` if you feel better to use GUI. The typical device id for your disk would likely to be ``/dev/sdb``.
 
 Then, flash the disk image into your physical SD card with the following command:
 
@@ -264,25 +269,25 @@ This would take a few minutes for your PC to write the whole file systems into t
 After this process completes, you are ready to unmount the SD card and insert it into your DE10-Nano board.
 Now you can connect the power cable and serial port to boot the Angstrom Linux.
 
-> Note: When boot up from the microSD card, you might notice the incompatibility of the linux kernel `zImage` in the microSD card.
-> In this case, you might need to build the `zImage` file of your own from `socfpga-4.9.78-ltsi <https://github.com/altera-opensource/linux-socfpga/tree/socfpga-4.9.78-ltsi>`_ branch of the `linux-socfpga <https://github.com/altera-opensource/linux-socfpga>`_ repository.
-> For a quick fix, you can also download a prebuilt version of the `zImage` file `from this link <https://raw.githubusercontent.com/liangfu/de10-nano-supplement/master/zImage>`_.
+   **Note**: When boot up from the microSD card, you might notice the incompatibility of the linux kernel ``zImage`` in the microSD card.
+   In this case, you might need to build the ``zImage`` file of your own from `socfpga-4.9.78-ltsi <https://github.com/altera-opensource/linux-socfpga/tree/socfpga-4.9.78-ltsi>`_ branch of the `linux-socfpga <https://github.com/altera-opensource/linux-socfpga>`_ repository.
+   For a quick fix, you can also download a prebuilt version of the ``zImage`` file `from this link <https://raw.githubusercontent.com/liangfu/de10-nano-supplement/master/zImage>`_.
 
-After connecting the usb cables to the DE10-Nano board, power on the board by connecting the power cable. You may then connect to the serial port of the device by using `minicom` on your host PC:
+After connecting the usb cables to the DE10-Nano board, power on the board by connecting the power cable. You may then connect to the serial port of the device by using ``minicom`` on your host PC:
 
 .. code:: bash
 
    # NOTE: root privilege is typically required to run the following command.
    minicom -D /dev/ttyUSB0
 
-The default user name for the device would be `root`, and the password is empty for the default user.
+The default user name for the device would be ``root``, and the password is empty for the default user.
 
-You may now start to install supporting Python3 packages (TVM has dropped the support for Python2), specifically, they are `numpy`, `attrs` and `decorator`.
+You may now start to install supporting Python3 packages (TVM has dropped the support for Python2), specifically, they are ``numpy``, ``attrs`` and ``decorator``.
 
-> Note: You might fail to install `numpy` by using `pip3` on the DE10-Nano device.
-> In that case, you have the option to either build your own filesystem image for the board from `meta-de10-nano <https://github.com/intel/meta-de10-nano>`_ repository;
-> an alternative option is to download prebuilt packages from existing Linux distributions, e.g. Debian.
-> For a quick fix, we have concatenated the supplementary binary files `here <https://raw.githubusercontent.com/liangfu/de10-nano-supplement/master/rootfs_supplement.tgz>`_, and you can extract the files into the root filesystem.
+   **Note**: You might fail to install ``numpy`` by using ``pip3`` on the DE10-Nano device.
+   In that case, you have the option to either build your own filesystem image for the board from `meta-de10-nano <https://github.com/intel/meta-de10-nano>`_ repository;
+   an alternative option is to download prebuilt packages from existing Linux distributions, e.g. Debian.
+   For a quick fix, we have concatenated the supplementary binary files `here <https://raw.githubusercontent.com/liangfu/de10-nano-supplement/master/rootfs_supplement.tgz>`_, and you can extract the files into the root filesystem.
 
 Install Required Python Packages
 """"""""""""""""""""""""""""""""
@@ -293,20 +298,18 @@ Build Additional Components to Use VTA Bitstream
 """"""""""""""""""""""""""""""""""""""""""""""""
 
 To use the above built bitstream on DE10-Nano hardware, several additional components need to be compiled for the system.
-Specifically, to compile application executables for the system, you need to download and install `SoCEDS <http://fpgasoftware.intel.com/soceds/18.1/?edition=standard&download_manager=dlm3&platform=linux>`_ (recommended), or alternatively install the `g++-arm-linux-gnueabihf` package on your host machine. You would also need a `cma` kernel module to allocate contigous memory, and a driver for communicating with the VTA subsystem.
-
+Specifically, to compile application executables for the system, you need to download and install `SoCEDS <http://fpgasoftware.intel.com/soceds/18.1/?edition=standard&download_manager=dlm3&platform=linux>`_ (recommended), or alternatively install the ``g++-arm-linux-gnueabihf`` package on your host machine. You would also need a ``cma`` kernel module to allocate contigous memory, and a driver for communicating with the VTA subsystem.
 
-.. _VTA FPGA Toolchain Installation:
 
-VTA FPGA Toolchain Installation
--------------------------------
+Bitstream Generation with Xilinx Toolchains
+-------------------------------------------
 
-This last guide allows users to generate custom VTA bitstreams using free-to-use Xilinx or Intel compilation toolchains.
+If you're interested in generating the Xilinx FPGA bitstream on your own instead of using the pre-built VTA bistreams, follow the instructions below.
 
 Xilinx Toolchain Installation
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
-We recommend using `Vivado 2018.3` since our scripts have been tested to work on this version of the Xilinx toolchains.
+We recommend using Vivado 2018.3 since our scripts have been tested to work on this version of the Xilinx toolchains.
 Our guide is written for Linux (Ubuntu) installation.
 
 You’ll need to install Xilinx’ FPGA compilation toolchain, `Vivado HL WebPACK 2018.3 <https://www.xilinx.com/products/design-tools/vivado.html>`_, which a license-free version of the Vivado HLx toolchain.
@@ -316,8 +319,8 @@ Obtaining and Launching the Vivado GUI Installer
 
 1. Go to the `download webpage <https://www.xilinx.com/support/download/index.html/content/xilinx/en/downloadNav/vivado-design-tools/2018-3.html>`_, and download the Linux Self Extracting Web Installer for Vivado HLx 2018.3: WebPACK and Editions.
 2. You’ll have to sign in with a Xilinx account. This requires a Xilinx account creation that will take 2 minutes.
-3. Complete the Name and Address Verification by clicking “Next”, and you will get the opportunity to download a binary file, called `Xilinx_Vivado_SDK_Web_2018.3_1207_2324_Lin64.bin`.
-4. Now that the file is downloaded, go to your `Downloads` directory, and change the file permissions so it can be executed:
+3. Complete the Name and Address Verification by clicking “Next”, and you will get the opportunity to download a binary file, called ``Xilinx_Vivado_SDK_Web_2018.3_1207_2324_Lin64.bin``.
+4. Now that the file is downloaded, go to your ``Downloads`` directory, and change the file permissions so it can be executed:
 
 .. code:: bash
 
@@ -334,26 +337,26 @@ Xilinx Vivado GUI Installer Steps
 
 At this point you've launched the Vivado 2018.3 Installer GUI program.
 
-1. Click “Next” on the *Welcome* screen.
-2. On the *Select Install Type* screen, enter your Xilinx user credentials under the “User Authentication” box and select the “Download and Install Now” option before clicking “Next” .
-3. On the *Accept License Agreements* screen, accept all terms before clicking “Next”.
-4. On the *Select Edition to Install* screen, select the “Vivado HL WebPACK” before clicking “Next” .
-5. Under the *Vivado HL WebPACK* screen, before hitting “Next", check the following options (the rest should be unchecked):
+1. Click “Next” on the "Welcome" screen.
+2. On the "Select Install Type" screen, enter your Xilinx user credentials under the “User Authentication” box and select the “Download and Install Now” option before clicking “Next”.
+3. On the "Accept License Agreements" screen, accept all terms before clicking “Next”.
+4. On the "Select Edition to Install" screen, select the “Vivado HL WebPACK” before clicking “Next”.
+5. Under the "Vivado HL WebPACK" screen, before hitting “Next", check the following options (the rest should be unchecked):
    * Design Tools -> Vivado Design Suite -> Vivado
    * Devices -> Production Devices -> SoCs -> Zynq-7000 (if you are targeting the Pynq board)
    * Devices -> Production Devices -> SoCs -> UltraScale+ MPSoC (if you are targeting the Ultra-96 board)
 6. Your total download size should be about 5GB and the amount of Disk Space Required 23GB.
-7. On the *Select Destination Directory* screen, set the installation directory before clicking “Next”. It might highlight some paths as red - that’s because the installer doesn’t have the permission to write to the directory. In that case select a path that doesn’t require special write permissions (e.g. your home directory).
-8. On the *Installation Summary* screen, hit “Install”.
-9. An *Installation Progress* window will pop-up to track progress of the download and the installation.
+7. On the "Select Destination Directory" screen, set the installation directory before clicking “Next”. It might highlight some paths as red - that’s because the installer doesn’t have the permission to write to the directory. In that case select a path that doesn’t require special write permissions (e.g. your home directory).
+8. On the "Installation Summary" screen, hit “Install”.
+9. An "Installation Progress" window will pop-up to track progress of the download and the installation.
 10. This process will take about 20-30 minutes depending on your connection speed.
 11. A pop-up window will inform you that the installation completed successfully. Click "OK".
-12. Finally the *Vivado License Manager* will launch. Select "Get Free ISE WebPACK, ISE/Vivado IP or PetaLinux License" and click "Connect Now" to complete the license registration process.
+12. Finally the "Vivado License Manager" will launch. Select "Get Free ISE WebPACK, ISE/Vivado IP or PetaLinux License" and click "Connect Now" to complete the license registration process.
 
 Environment Setup
 """""""""""""""""
 
-The last step is to update your `~/.bashrc` with the following lines. This will include all of the Xilinx binary paths so you can launch compilation scripts from the command line.
+The last step is to update your ``~/.bashrc`` with the following lines. This will include all of the Xilinx binary paths so you can launch compilation scripts from the command line.
 
 .. code:: bash
 
@@ -361,91 +364,108 @@ The last step is to update your `~/.bashrc` with the following lines. This will
    export XILINX_VIVADO=${XILINX_PATH}/Vivado/2018.3
    export PATH=${XILINX_VIVADO}/bin:${PATH}
 
-Intel Toolchain Installation
-^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+HLS-based Custom VTA Bitstream Compilation for PYNQ
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
-It is recommended to use `Intel Quartus Prime 18.1`, since the test scripts contained in this document have been tested on this version.
+High-level hardware parameters are listed in the VTA configuration file and can be customized by the user.
+For this custom VTA bitstream compilation exercise, we'll change the frequency of our design, so it can be clocked a little faster.
 
-You would need to install Intel's FPGA compilation toolchain, `Quartus Prime Lite <http://fpgasoftware.intel.com/?edition=lite>`_, which is a license-free version of the Intel Quartus Prime software.
+* Set the ``HW_FREQ`` field to ``142``. The Pynq board supports 100, 142, 167 and 200MHz clocks. Note that the higher the frequency, the harder it will be to close timing. Increasing the frequency can lead to timing violation and thus faulty hardware execution.
+* Set the ``HW_CLK_TARGET`` to ``6``. This parameters refers to the target clock period in nano seconds for HLS - a lower clock period leads to more aggressive pipelining to achieve timing closure at higher frequencies. Technically a 142MHz clock would require a 7ns target, but we intentionally lower the clock target to 6ns to more aggressively pipeline our design.
 
-Obtaining and Launching the Quartus GUI Installer
-"""""""""""""""""""""""""""""""""""""""""""""""""
+Bitstream generation is driven by a top-level ``Makefile`` under ``<tvm root>/3rdparty/vta-hw/hardware/xilinx/``.
 
-1. Go to the `download center <http://fpgasoftware.intel.com/?edition=lite>`_, and download the linux version of `Quartus Prime (include Nios II EDS)` and `Cyclone V device support` files in the `Separate file` tab. This avoid downloading unused device support files.
-2. Sign in the form if you have an account, or register on the right side of the web page to create an account.
-3. After signed in, you are able to download the installer and the device support files.
-4. Now that the files are downloaded, go to your `Downloads` directory, and change the file permissions:
+If you just want to simulate the VTA design in software emulation to make sure that it is functional, enter:
 
 .. code:: bash
 
-   chmod u+x QuartusLiteSetup-18.1.0.625-linux.run
+   cd <tvm root>/3rdparty/vta-hw/hardware/xilinx
+   make ip MODE=sim
 
-5. Now ensure both the installer and device support files are in the same directory, and you can run the install with:
+
+If you just want to generate the HLS-based VTA IP cores without launching the entire design place and route, enter:
 
 .. code:: bash
 
-   ./QuartusLiteSetup-18.1.0.625-linux.run
+   make ip
 
-6. Follow the instructions on the pop-up GUI form, and install all the content in the `/usr/local` directory. After installation, `/usr/local/intelFPGA_lite/18.1` would be created and the Quartus program along with other programs would be available in the folder.
+You'll be able to view the HLS synthesis reports under ``<tvm root>/3rdparty/vta-hw/build/hardware/xilinx/hls/<configuration>/<block>/solution0/syn/report/<block>_csynth.rpt``
 
-Environment Setup
-"""""""""""""""""
+   **Note**: The ``<configuration>`` name is a string that summarizes the VTA configuration parameters listed in the ``vta_config.json``. The ``<block>`` name refers to the specific module (or HLS function) that compose the high-level VTA pipeline.
 
-Similar to what should be done for Xilinx toolchain, the following line should be added to your `~/.bashrc`.
+Finally to run the full hardware compilation and generate the VTA bitstream, run ``make``.
 
-.. code:: bash
+This process is lengthy, and can take around up to an hour to complete depending on your machine's specs.
+We recommend setting the ``VTA_HW_COMP_THREADS`` variable in the Makefile to take full advantage of all the cores on your development machine.
 
-   # Intel Quartus 18.1 environment
-   export QUARTUS_ROOTDIR="/usr/local/intelFPGA_lite/18.1/quartus"
-   export PATH=${QUARTUS_ROOTDIR}/bin:${PATH}
-   export PATH=${QUARTUS_ROOTDIR}/sopc_builder/bin:${PATH}
+Once the compilation completes, the generated bitstream can be found under ``<tvm root>/3rdparty/vta-hw/build/hardware/xilinx/vivado/<configuration>/export/vta.bit``.
 
-This would add quartus binary path into your `PATH` environment variable, so you can launch compilation scripts from the command line.
+Using A Custom Bitstream
+^^^^^^^^^^^^^^^^^^^^^^^^
 
-HLS-based Custom VTA Bitstream Compilation for PYNQ
-^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
+We can program the new VTA FPGA bitstream by setting the bitstream path of the ``vta.program_fpga()`` function in the tutorial examples, or in the ``test_program_rpc.py`` script.
 
-High-level hardware parameters are listed in the VTA configuration file and can be customized by the user.
-For this custom VTA bitstream compilation exercise, we'll change the frequency of our design, so it can be clocked a little faster.
-* Set the `HW_FREQ` field to `142`. The Pynq board supports 100, 142, 167 and 200MHz clocks. Note that the higher the frequency, the harder it will be to close timing. Increasing the frequency can lead to timing violation and thus faulty hardware execution.
-* Set the `HW_CLK_TARGET` to `6`. This parameters refers to the target clock period in nano seconds for HLS - a lower clock period leads to more aggressive pipelining to achieve timing closure at higher frequencies. Technically a 142MHz clock would require a 7ns target, but we intentionally lower the clock target to 6ns to more aggressively pipeline our design.
+.. code:: python
 
-Bitstream generation is driven by a top-level `Makefile` under `<tvm root>/3rdparty/vta-hw/hardware/xilinx/`.
+   vta.program_fpga(remote, bitstream="<tvm root>/3rdparty/vta-hw/build/hardware/xilinx/vivado/<configuration>/export/vta.bit")
 
-If you just want to simulate the VTA design in software emulation to make sure that it is functional, enter:
+Instead of downloading a pre-built bitstream from the VTA bitstream repository, TVM will instead use the new bitstream you just generated, which is a VTA design clocked at a higher frequency.
+Do you observe a noticeable performance increase on the ImageNet classification example?
 
-.. code:: bash
 
-   cd <tvm root>/3rdparty/vta-hw/hardware/xilinx
-   make ip MODE=sim
 
+Bitstream Generation with Intel Toolchains
+-------------------------------------------
 
-If you just want to generate the HLS-based VTA IP cores without launching the entire design place and route, enter:
+If you're interested in generating the Xilinx FPGA bitstream on your own instead of using the pre-built VTA bistreams, follow the instructions below.
 
-.. code:: bash
+Intel Toolchain Installation
+^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
-   make ip
+It is recommended to use ``Intel Quartus Prime 18.1``, since the test scripts contained in this document have been tested on this version.
 
-You'll be able to view the HLS synthesis reports under `<tvm root>/3rdparty/vta-hw/build/hardware/xilinx/hls/` `<configuration>/<block>/solution0/syn/report/<block>_csynth.rpt`
-> Note: The `<configuration>` name is a string that summarizes the VTA configuration parameters listed in the `vta_config.json`. The `<block>` name refers to the specific module (or HLS function) that compose the high-level VTA pipeline.
+You would need to install Intel's FPGA compilation toolchain, `Quartus Prime Lite <http://fpgasoftware.intel.com/?edition=lite>`_, which is a license-free version of the Intel Quartus Prime software.
 
-Finally to run the full hardware compilation and generate the VTA bitstream, run:
+Obtaining and Launching the Quartus GUI Installer
+"""""""""""""""""""""""""""""""""""""""""""""""""
+
+1. Go to the `download center <http://fpgasoftware.intel.com/?edition=lite>`_, and download the linux version of "Quartus Prime (include Nios II EDS)" and "Cyclone V device support" files in the "Separate file" tab. This avoid downloading unused device support files.
+2. Sign in the form if you have an account, or register on the right side of the web page to create an account.
+3. After signed in, you are able to download the installer and the device support files.
+4. Now that the files are downloaded, go to your ``Downloads`` directory, and change the file permissions:
 
 .. code:: bash
 
-   make
+   chmod u+x QuartusLiteSetup-18.1.0.625-linux.run
 
-This process is lengthy, and can take around up to an hour to complete depending on your machine's specs.
-We recommend setting the `VTA_HW_COMP_THREADS` variable in the Makefile to take full advantage of all the cores on your development machine.
+5. Now ensure both the installer and device support files are in the same directory, and you can run the install with:
+
+.. code:: bash
 
-Once the compilation completes, the generated bitstream can be found under `<tvm root>/3rdparty/vta-hw/build/hardware/xilinx/vivado/<configuration>/export/vta.bit`.
+   ./QuartusLiteSetup-18.1.0.625-linux.run
+
+6. Follow the instructions on the pop-up GUI form, and install all the content in the ``/usr/local`` directory. After installation, ``/usr/local/intelFPGA_lite/18.1`` would be created and the Quartus program along with other programs would be available in the folder.
+
+Environment Setup
+"""""""""""""""""
+
+Similar to what should be done for Xilinx toolchain, the following line should be added to your ``~/.bashrc``.
+
+.. code:: bash
+
+   # Intel Quartus 18.1 environment
+   export QUARTUS_ROOTDIR="/usr/local/intelFPGA_lite/18.1/quartus"
+   export PATH=${QUARTUS_ROOTDIR}/bin:${PATH}
+   export PATH=${QUARTUS_ROOTDIR}/sopc_builder/bin:${PATH}
+
+This would add quartus binary path into your ``PATH`` environment variable, so you can launch compilation scripts from the command line.
 
 Chisel-based Custom VTA Bitstream Compilation for DE10-Nano
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 
 Similar to the HLS-based design, high-level hardware parameters in Chisel-based design are listed in the VTA configuration file `Configs.scala <https://github.com/apache/incubator-tvm/blob/master/3rdparty/vta-hw/hardware/chisel/src/main/scala/core/Configs.scala>`_, and they can be customized by the user.
 
-For Intel FPGA, bitstream generation is driven by a top-level `Makefile` under `<tvm root>/3rdparty/vta-hw/hardware/intel`.
+For Intel FPGA, bitstream generation is driven by a top-level ``Makefile`` under ``<tvm root>/3rdparty/vta-hw/hardware/intel``.
 
 If you just want to generate the Chisel-based VTA IP core for the DE10-Nano board without compiling the design for the FPGA hardware, enter:
 
@@ -454,9 +474,9 @@ If you just want to generate the Chisel-based VTA IP core for the DE10-Nano boar
    cd <tvm root>/3rdparty/vta-hw/hardware/intel
    make ip
 
-Then you'll be able to locate the generated verilog file at `<tvm root>/3rdparty/vta-hw/build/hardware/intel/chisel/<configuration>/VTA.DefaultDe10Config.v`.
+Then you'll be able to locate the generated verilog file at ``<tvm root>/3rdparty/vta-hw/build/hardware/intel/chisel/<configuration>/VTA.DefaultDe10Config.v``.
 
-If you would like to run the full hardware compilation for the `de10nano` board:
+If you would like to run the full hardware compilation for the ``de10nano`` board:
 
 .. code:: bash
 
@@ -464,16 +484,5 @@ If you would like to run the full hardware compilation for the `de10nano` board:
 
 This process might be a bit lengthy, and might take up to half an hour to complete depending on the performance of your PC. The Quartus Prime software would automatically detect the number of cores available on your PC and try to utilize all of them to perform such process.
 
-Once the compilation completes, the generated bistream can be found under `<tvm root>/3rdparty/vta-hw/build/hardware/intel/quartus/<configuration>/export/vta.rbf`. You can also open the Quartus project file (.qpf) available at `<tvm root>/3rdparty/vta-hw/build/hardware/intel/quartus/<configuration>/de10_nano_top.qpf` to look around the generated reports.
-
-Use the Custom Bitstream
-^^^^^^^^^^^^^^^^^^^^^^^^
-
-We can program the new VTA FPGA bitstream by setting the bitstream path of the `vta.program_fpga()` function in the tutorial examples, or in the `test_program_rpc.py` script.
+Once the compilation completes, the generated bistream can be found under ``<tvm root>/3rdparty/vta-hw/build/hardware/intel/quartus/<configuration>/export/vta.rbf``. You can also open the Quartus project file (.qpf) available at ``<tvm root>/3rdparty/vta-hw/build/hardware/intel/quartus/<configuration>/de10_nano_top.qpf`` to look around the generated reports.
 
-.. code:: python
-
-   vta.program_fpga(remote, bitstream="<tvm root>/3rdparty/vta-hw/build/hardware/xilinx/vivado/<configuration>/export/vta.bit")
-
-Instead of downloading a pre-built bitstream from the VTA bitstream repository, TVM will instead use the new bitstream you just generated, which is a VTA design clocked at a higher frequency.
-Do you observe a noticeable performance increase on the ImageNet classification example?
diff --git a/docs/_sources/vta/tutorials/autotvm/sg_execution_times.rst.txt b/docs/_sources/vta/tutorials/autotvm/sg_execution_times.rst.txt
index 1620675..1b536f2 100644
--- a/docs/_sources/vta/tutorials/autotvm/sg_execution_times.rst.txt
+++ b/docs/_sources/vta/tutorials/autotvm/sg_execution_times.rst.txt
@@ -5,6 +5,6 @@
 
 Computation times
 =================
-**00:03.386** total execution time for **vta_tutorials_autotvm** files:
+**00:03.660** total execution time for **vta_tutorials_autotvm** files:
 
-- **00:03.386**: :ref:`sphx_glr_vta_tutorials_autotvm_tune_relay_vta.py` (``tune_relay_vta.py``)
+- **00:03.660**: :ref:`sphx_glr_vta_tutorials_autotvm_tune_relay_vta.py` (``tune_relay_vta.py``)
diff --git a/docs/_sources/vta/tutorials/autotvm/tune_relay_vta.rst.txt b/docs/_sources/vta/tutorials/autotvm/tune_relay_vta.rst.txt
index 33456d3..8c5834e 100644
--- a/docs/_sources/vta/tutorials/autotvm/tune_relay_vta.rst.txt
+++ b/docs/_sources/vta/tutorials/autotvm/tune_relay_vta.rst.txt
@@ -481,7 +481,7 @@ Finally, we launch tuning jobs and evaluate the end-to-end performance.
  .. code-block:: none
 
     Extract tasks...
-
    ...1%, 0.01 MB, 50 KB/s, 0 seconds passed
    ...2%, 0.02 MB, 100 KB/s, 0 seconds passed
    ...3%, 0.02 MB, 150 KB/s, 0 seconds passed
    ...4%, 0.03 MB, 199 KB/s, 0 seconds passed
    ...5%, 0.04 MB, 248 KB/s, 0 seconds passed
    ...6%, 0.05 MB, 297 KB/s, 0 seconds passed
    ...7%, 0.05 MB, 347 KB/s, 0 seconds passed
    ...8%, 0.06 MB, 395 KB/s, 0 seconds passed
    ...9%, 0.07 MB, 444 KB/s, 0 seconds passed
    ...10%, 0.08 MB, 490 KB/s, 0 seconds passed
    ...11%, 0.09 MB, 539 KB/s, 0 seconds passed
    ...13%, 0.09 MB, 587 KB/s, 0 seconds passed
    ...14%, 0.10 MB, 636 KB/s, 0 seconds passed
    ...15%, 0.11 MB, 681 KB/s, 0 seconds passed
    ...16%, 0.12 MB, 730 KB/s, 0 seconds passed
    ...17%, 0.12 MB, 778 KB/s, 0 seconds passed
    ...18%, 0.13 MB, 826 KB/s, 0 seconds passed
    ...19%, 0.14 MB, 870 KB/s, 0 seconds passed
    ...20%, 0.15 MB, 918 KB/s, 0 seconds passed
    ...21%, 0.16 MB, 962 KB/s, 0 seconds passed
    ...22%, 0.16 MB, 1010 KB/s, 0 seconds pass
 ed
    ...23%, 0.17 MB, 1056 KB/s, 0 seconds passed
    ...25%, 0.18 MB, 1104 KB/s, 0 seconds passed
    ...26%, 0.19 MB, 1150 KB/s, 0 seconds passed
    ...27%, 0.20 MB, 1197 KB/s, 0 seconds passed
    ...28%, 0.20 MB, 1240 KB/s, 0 seconds passed
    ...29%, 0.21 MB, 1287 KB/s, 0 seconds passed
    ...30%, 0.22 MB, 1332 KB/s, 0 seconds passed
    ...31%, 0.23 MB, 1379 KB/s, 0 seconds passed
    ...32%, 0.23 MB, 1420 KB/s, 0 seconds passed
    ...33%, 0.24 MB, 1466 KB/s, 0 seconds passed
    ...34%, 0.25 MB, 1511 KB/s, 0 seconds passed
    ...35%, 0.26 MB, 1558 KB/s, 0 seconds passed
    ...36%, 0.27 MB, 1599 KB/s, 0 seconds passed
    ...38%, 0.27 MB, 1645 KB/s, 0 seconds passed
    ...39%, 0.28 MB, 1691 KB/s, 0 seconds passed
    ...40%, 0.29 MB, 1737 KB/s, 0 seconds passed
    ...41%, 0.30 MB, 1776 KB/s, 0 seconds passed
    ...42%, 0.30 MB, 1821 KB/s, 0 seconds passed
    ...43%, 0.31 MB, 1867 KB/s, 0 seconds passed
    ...44%, 0.32 MB, 1912 KB/s, 0 seconds passed
    ...45%, 0.
 33 MB, 1956 KB/s, 0 seconds passed
    ...46%, 0.34 MB, 2002 KB/s, 0 seconds passed
    ...47%, 0.34 MB, 2020 KB/s, 0 seconds passed
    ...48%, 0.35 MB, 2065 KB/s, 0 seconds passed
    ...50%, 0.36 MB, 2105 KB/s, 0 seconds passed
    ...51%, 0.37 MB, 2149 KB/s, 0 seconds passed
    ...52%, 0.38 MB, 2187 KB/s, 0 seconds passed
    ...53%, 0.38 MB, 2232 KB/s, 0 seconds passed
    ...54%, 0.39 MB, 2274 KB/s, 0 seconds passed
    ...55%, 0.40 MB, 2318 KB/s, 0 seconds passed
    ...56%, 0.41 MB, 2352 KB/s, 0 seconds passed
    ...57%, 0.41 MB, 2396 KB/s, 0 seconds passed
    ...58%, 0.42 MB, 2439 KB/s, 0 seconds passed
    ...59%, 0.43 MB, 2483 KB/s, 0 seconds passed
    ...60%, 0.44 MB, 2522 KB/s, 0 seconds passed
    ...62%, 0.45 MB, 2566 KB/s, 0 seconds passed
    ...63%, 0.45 MB, 2603 KB/s, 0 seconds passed
    ...64%, 0.46 MB, 2647 KB/s, 0 seconds passed
    ...65%, 0.47 MB, 2688 KB/s, 0 seconds passed
    ...66%, 0.48 MB, 2731 KB/s, 0 seconds passed
    ...67%, 0.48 MB, 2774 KB/s,
  0 seconds passed
    ...68%, 0.49 MB, 2817 KB/s, 0 seconds passed
    ...69%, 0.50 MB, 2854 KB/s, 0 seconds passed
    ...70%, 0.51 MB, 2897 KB/s, 0 seconds passed
    ...71%, 0.52 MB, 2937 KB/s, 0 seconds passed
    ...72%, 0.52 MB, 2980 KB/s, 0 seconds passed
    ...73%, 0.53 MB, 3022 KB/s, 0 seconds passed
    ...75%, 0.54 MB, 3065 KB/s, 0 seconds passed
    ...76%, 0.55 MB, 3101 KB/s, 0 seconds passed
    ...77%, 0.55 MB, 3144 KB/s, 0 seconds passed
    ...78%, 0.56 MB, 3180 KB/s, 0 seconds passed
    ...79%, 0.57 MB, 3222 KB/s, 0 seconds passed
    ...80%, 0.58 MB, 3264 KB/s, 0 seconds passed
    ...81%, 0.59 MB, 3307 KB/s, 0 seconds passed
    ...82%, 0.59 MB, 3338 KB/s, 0 seconds passed
    ...83%, 0.60 MB, 3381 KB/s, 0 seconds passed
    ...84%, 0.61 MB, 3422 KB/s, 0 seconds passed
    ...85%, 0.62 MB, 3464 KB/s, 0 seconds passed
    ...87%, 0.62 MB, 3506 KB/s, 0 seconds passed
    ...88%, 0.63 MB, 3548 KB/s, 0 seconds passed
    ...89%, 0.64 MB, 3583 KB/s, 0 seconds passed
 
    ...90%, 0.65 MB, 3625 KB/s, 0 seconds passed
    ...91%, 0.66 MB, 3666 KB/s, 0 seconds passed
    ...92%, 0.66 MB, 3708 KB/s, 0 seconds passed
    ...93%, 0.67 MB, 3749 KB/s, 0 seconds passed
    ...94%, 0.68 MB, 3791 KB/s, 0 seconds passed
    ...95%, 0.69 MB, 3829 KB/s, 0 seconds passed
    ...96%, 0.70 MB, 3870 KB/s, 0 seconds passed
    ...97%, 0.70 MB, 3903 KB/s, 0 seconds passed
    ...99%, 0.71 MB, 3944 KB/s, 0 seconds passed
    ...100%, 0.72 MB, 3983 KB/s, 0 seconds passed
+
    ...1%, 0.01 MB, 25 KB/s, 0 seconds passed
    ...2%, 0.02 MB, 50 KB/s, 0 seconds passed
    ...3%, 0.02 MB, 75 KB/s, 0 seconds passed
    ...4%, 0.03 MB, 101 KB/s, 0 seconds passed
    ...5%, 0.04 MB, 126 KB/s, 0 seconds passed
    ...6%, 0.05 MB, 151 KB/s, 0 seconds passed
    ...7%, 0.05 MB, 176 KB/s, 0 seconds passed
    ...8%, 0.06 MB, 201 KB/s, 0 seconds passed
    ...9%, 0.07 MB, 226 KB/s, 0 seconds passed
    ...10%, 0.08 MB, 250 KB/s, 0 seconds passed
    ...11%, 0.09 MB, 275 KB/s, 0 seconds passed
    ...13%, 0.09 MB, 300 KB/s, 0 seconds passed
    ...14%, 0.10 MB, 324 KB/s, 0 seconds passed
    ...15%, 0.11 MB, 349 KB/s, 0 seconds passed
    ...16%, 0.12 MB, 374 KB/s, 0 seconds passed
    ...17%, 0.12 MB, 398 KB/s, 0 seconds passed
    ...18%, 0.13 MB, 423 KB/s, 0 seconds passed
    ...19%, 0.14 MB, 448 KB/s, 0 seconds passed
    ...20%, 0.15 MB, 472 KB/s, 0 seconds passed
    ...21%, 0.16 MB, 497 KB/s, 0 seconds passed
    ...22%, 0.16 MB, 521 KB/s, 0 seconds passed
     ...23%, 0.17 MB, 546 KB/s, 0 seconds passed
    ...25%, 0.18 MB, 570 KB/s, 0 seconds passed
    ...26%, 0.19 MB, 594 KB/s, 0 seconds passed
    ...27%, 0.20 MB, 619 KB/s, 0 seconds passed
    ...28%, 0.20 MB, 643 KB/s, 0 seconds passed
    ...29%, 0.21 MB, 667 KB/s, 0 seconds passed
    ...30%, 0.22 MB, 692 KB/s, 0 seconds passed
    ...31%, 0.23 MB, 716 KB/s, 0 seconds passed
    ...32%, 0.23 MB, 740 KB/s, 0 seconds passed
    ...33%, 0.24 MB, 764 KB/s, 0 seconds passed
    ...34%, 0.25 MB, 788 KB/s, 0 seconds passed
    ...35%, 0.26 MB, 812 KB/s, 0 seconds passed
    ...36%, 0.27 MB, 836 KB/s, 0 seconds passed
    ...38%, 0.27 MB, 860 KB/s, 0 seconds passed
    ...39%, 0.28 MB, 883 KB/s, 0 seconds passed
    ...40%, 0.29 MB, 907 KB/s, 0 seconds passed
    ...41%, 0.30 MB, 931 KB/s, 0 seconds passed
    ...42%, 0.30 MB, 955 KB/s, 0 seconds passed
    ...43%, 0.31 MB, 979 KB/s, 0 seconds passed
    ...44%, 0.32 MB, 1002 KB/s, 0 seconds passed
    ...45%, 0.33 MB, 1026 KB/s, 0 se
 conds passed
    ...46%, 0.34 MB, 1048 KB/s, 0 seconds passed
    ...47%, 0.34 MB, 1072 KB/s, 0 seconds passed
    ...48%, 0.35 MB, 1096 KB/s, 0 seconds passed
    ...50%, 0.36 MB, 1120 KB/s, 0 seconds passed
    ...51%, 0.37 MB, 1143 KB/s, 0 seconds passed
    ...52%, 0.38 MB, 1167 KB/s, 0 seconds passed
    ...53%, 0.38 MB, 1191 KB/s, 0 seconds passed
    ...54%, 0.39 MB, 1214 KB/s, 0 seconds passed
    ...55%, 0.40 MB, 1238 KB/s, 0 seconds passed
    ...56%, 0.41 MB, 1261 KB/s, 0 seconds passed
    ...57%, 0.41 MB, 1284 KB/s, 0 seconds passed
    ...58%, 0.42 MB, 1308 KB/s, 0 seconds passed
    ...59%, 0.43 MB, 1331 KB/s, 0 seconds passed
    ...60%, 0.44 MB, 1355 KB/s, 0 seconds passed
    ...62%, 0.45 MB, 1378 KB/s, 0 seconds passed
    ...63%, 0.45 MB, 1401 KB/s, 0 seconds passed
    ...64%, 0.46 MB, 1424 KB/s, 0 seconds passed
    ...65%, 0.47 MB, 1447 KB/s, 0 seconds passed
    ...66%, 0.48 MB, 1471 KB/s, 0 seconds passed
    ...67%, 0.48 MB, 1494 KB/s, 0 seconds passed
    
 ...68%, 0.49 MB, 1517 KB/s, 0 seconds passed
    ...69%, 0.50 MB, 1541 KB/s, 0 seconds passed
    ...70%, 0.51 MB, 1565 KB/s, 0 seconds passed
    ...71%, 0.52 MB, 1586 KB/s, 0 seconds passed
    ...72%, 0.52 MB, 1610 KB/s, 0 seconds passed
    ...73%, 0.53 MB, 1633 KB/s, 0 seconds passed
    ...75%, 0.54 MB, 1657 KB/s, 0 seconds passed
    ...76%, 0.55 MB, 1677 KB/s, 0 seconds passed
    ...77%, 0.55 MB, 1700 KB/s, 0 seconds passed
    ...78%, 0.56 MB, 1724 KB/s, 0 seconds passed
    ...79%, 0.57 MB, 1747 KB/s, 0 seconds passed
    ...80%, 0.58 MB, 1770 KB/s, 0 seconds passed
    ...81%, 0.59 MB, 1794 KB/s, 0 seconds passed
    ...82%, 0.59 MB, 1817 KB/s, 0 seconds passed
    ...83%, 0.60 MB, 1841 KB/s, 0 seconds passed
    ...84%, 0.61 MB, 1863 KB/s, 0 seconds passed
    ...85%, 0.62 MB, 1887 KB/s, 0 seconds passed
    ...87%, 0.62 MB, 1910 KB/s, 0 seconds passed
    ...88%, 0.63 MB, 1933 KB/s, 0 seconds passed
    ...89%, 0.64 MB, 1956 KB/s, 0 seconds passed
    ...90%, 0.65 MB, 
 1979 KB/s, 0 seconds passed
    ...91%, 0.66 MB, 2002 KB/s, 0 seconds passed
    ...92%, 0.66 MB, 2024 KB/s, 0 seconds passed
    ...93%, 0.67 MB, 2047 KB/s, 0 seconds passed
    ...94%, 0.68 MB, 2070 KB/s, 0 seconds passed
    ...95%, 0.69 MB, 2093 KB/s, 0 seconds passed
    ...96%, 0.70 MB, 2116 KB/s, 0 seconds passed
    ...97%, 0.70 MB, 2140 KB/s, 0 seconds passed
    ...99%, 0.71 MB, 2163 KB/s, 0 seconds passed
    ...100%, 0.72 MB, 2185 KB/s, 0 seconds passed
     Exception in thread Thread-9:
     Traceback (most recent call last):
       File "/usr/lib/python3.6/threading.py", line 916, in _bootstrap_inner
@@ -492,17 +492,17 @@ Finally, we launch tuning jobs and evaluate the end-to-end performance.
         grc.codegen(mod["main"])
       File "/workspace/docs/../python/tvm/relay/backend/graph_runtime_codegen.py", line 88, in codegen
         arr = self._get_param_by_name(key)
-      File "tvm/_ffi/_cython/./packed_func.pxi", line 315, in tvm._ffi._cy3.core.PackedFuncBase.__call__
-      File "tvm/_ffi/_cython/./packed_func.pxi", line 250, in tvm._ffi._cy3.core.FuncCall
-      File "tvm/_ffi/_cython/./packed_func.pxi", line 239, in tvm._ffi._cy3.core.FuncCall3
+      File "tvm/_ffi/_cython/./packed_func.pxi", line 321, in tvm._ffi._cy3.core.PackedFuncBase.__call__
+      File "tvm/_ffi/_cython/./packed_func.pxi", line 256, in tvm._ffi._cy3.core.FuncCall
+      File "tvm/_ffi/_cython/./packed_func.pxi", line 245, in tvm._ffi._cy3.core.FuncCall3
       File "tvm/_ffi/_cython/./base.pxi", line 160, in tvm._ffi._cy3.core.CALL
     tvm._ffi.base.TVMError: Traceback (most recent call last):
-      [bt] (4) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7ff84dca8c61]
-      [bt] (3) /workspace/build/libtvm.so(std::_Function_handler<void (tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*), tvm::relay::backend::GraphRuntimeCodegenModule::GetFunction(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, tvm::runtime::ObjectPtr<tvm::runtime::Object> const&)::{lambda(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*)#5}>::_M_invoke(std::_Any_data const&, tvm::runtime::TVMArgs&&, tvm::runtime::TVMRetValue*&&)+0x17) [0x7ff84 [...]
-      [bt] (2) /workspace/build/libtvm.so(tvm::relay::backend::GraphRuntimeCodegenModule::GetFunction(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, tvm::runtime::ObjectPtr<tvm::runtime::Object> const&)::{lambda(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*)#5}::operator()(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const+0x58) [0x7ff84db3f668]
-      [bt] (1) /workspace/build/libtvm.so(tvm::runtime::TVMArgValue::operator std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >[abi:cxx11]() const+0x13b) [0x7ff84d33237b]
-      [bt] (0) /workspace/build/libtvm.so(dmlc::LogMessageFatal::~LogMessageFatal()+0x32) [0x7ff84d31a7e2]
-      File "/workspace/include/tvm/runtime/packed_func.h", line 507
+      [bt] (4) /workspace/build/libtvm.so(TVMFuncCall+0x61) [0x7fb02147c191]
+      [bt] (3) /workspace/build/libtvm.so(std::_Function_handler<void (tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*), tvm::relay::backend::GraphRuntimeCodegenModule::GetFunction(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, tvm::runtime::ObjectPtr<tvm::runtime::Object> const&)::{lambda(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*)#5}>::_M_invoke(std::_Any_data const&, tvm::runtime::TVMArgs&&, tvm::runtime::TVMRetValue*&&)+0x17) [0x7fb02 [...]
+      [bt] (2) /workspace/build/libtvm.so(tvm::relay::backend::GraphRuntimeCodegenModule::GetFunction(std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&, tvm::runtime::ObjectPtr<tvm::runtime::Object> const&)::{lambda(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*)#5}::operator()(tvm::runtime::TVMArgs, tvm::runtime::TVMRetValue*) const+0x58) [0x7fb0212e7368]
+      [bt] (1) /workspace/build/libtvm.so(tvm::runtime::TVMArgValue::operator std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >[abi:cxx11]() const+0x13b) [0x7fb020ac464b]
+      [bt] (0) /workspace/build/libtvm.so(dmlc::LogMessageFatal::~LogMessageFatal()+0x64) [0x7fb020aac1f4]
+      File "/workspace/include/tvm/runtime/packed_func.h", line 484
     TVMError: Check failed: type_code_ == kTVMStr (8 vs. 11) : expected str but get Object
 
     Extracted 10 conv2d tasks:
diff --git a/docs/_sources/vta/tutorials/frontend/deploy_classification.rst.txt b/docs/_sources/vta/tutorials/frontend/deploy_classification.rst.txt
index 6b75f5d..fb3c5a5 100644
--- a/docs/_sources/vta/tutorials/frontend/deploy_classification.rst.txt
+++ b/docs/_sources/vta/tutorials/frontend/deploy_classification.rst.txt
@@ -243,8 +243,8 @@ The compilation steps are:
 
  .. code-block:: none
 
-
    ...12%, 0.01 MB, 44 KB/s, 0 seconds passed
    ...25%, 0.02 MB, 88 KB/s, 0 seconds passed
    ...38%, 0.02 MB, 131 KB/s, 0 seconds passed
    ...51%, 0.03 MB, 174 KB/s, 0 seconds passed
    ...64%, 0.04 MB, 218 KB/s, 0 seconds passed
    ...77%, 0.05 MB, 260 KB/s, 0 seconds passed
    ...90%, 0.05 MB, 304 KB/s, 0 seconds passed
    ...100%, 0.06 MB, 347 KB/s, 0 seconds passed
-    resnet18_v1 inference graph built in 3.67s!
+
    ...12%, 0.01 MB, 44 KB/s, 0 seconds passed
    ...25%, 0.02 MB, 88 KB/s, 0 seconds passed
    ...38%, 0.02 MB, 132 KB/s, 0 seconds passed
    ...51%, 0.03 MB, 176 KB/s, 0 seconds passed
    ...64%, 0.04 MB, 219 KB/s, 0 seconds passed
    ...77%, 0.05 MB, 262 KB/s, 0 seconds passed
    ...90%, 0.05 MB, 306 KB/s, 0 seconds passed
    ...100%, 0.06 MB, 349 KB/s, 0 seconds passed
+    resnet18_v1 inference graph built in 3.82s!
 
 
 
diff --git a/docs/_sources/vta/tutorials/frontend/deploy_detection.rst.txt b/docs/_sources/vta/tutorials/frontend/deploy_detection.rst.txt
index 9d2f9b2..9830a0d 100644
--- a/docs/_sources/vta/tutorials/frontend/deploy_detection.rst.txt
+++ b/docs/_sources/vta/tutorials/frontend/deploy_detection.rst.txt
@@ -315,7 +315,7 @@ The compilation steps are:
 
  .. code-block:: none
 
-    yolov3-tiny inference graph built in 4.72s!
+    yolov3-tiny inference graph built in 5.28s!
 
 
 
diff --git a/docs/_sources/vta/tutorials/frontend/sg_execution_times.rst.txt b/docs/_sources/vta/tutorials/frontend/sg_execution_times.rst.txt
index e178511..00ddd3d 100644
--- a/docs/_sources/vta/tutorials/frontend/sg_execution_times.rst.txt
+++ b/docs/_sources/vta/tutorials/frontend/sg_execution_times.rst.txt
@@ -5,7 +5,7 @@
 
 Computation times
 =================
-**00:49.003** total execution time for **vta_tutorials_frontend** files:
+**00:54.020** total execution time for **vta_tutorials_frontend** files:
 
-- **00:29.319**: :ref:`sphx_glr_vta_tutorials_frontend_deploy_detection.py` (``deploy_detection.py``)
-- **00:19.684**: :ref:`sphx_glr_vta_tutorials_frontend_deploy_classification.py` (``deploy_classification.py``)
+- **00:31.582**: :ref:`sphx_glr_vta_tutorials_frontend_deploy_detection.py` (``deploy_detection.py``)
+- **00:22.438**: :ref:`sphx_glr_vta_tutorials_frontend_deploy_classification.py` (``deploy_classification.py``)
diff --git a/docs/_sources/vta/tutorials/matrix_multiply.rst.txt b/docs/_sources/vta/tutorials/matrix_multiply.rst.txt
index c1908be..053abb5 100644
--- a/docs/_sources/vta/tutorials/matrix_multiply.rst.txt
+++ b/docs/_sources/vta/tutorials/matrix_multiply.rst.txt
@@ -302,45 +302,51 @@ After we construct the schedule, by default the schedule computes
 
  .. code-block:: none
 
-    PrimFunc([A, B, C]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      // attr [A_buf] storage_scope = "global"
-      allocate A_buf[int8 * 256]
-      // attr [B_buf] storage_scope = "global"
-      allocate B_buf[int8 * 65536]
-      // attr [C_buf] storage_scope = "global"
-      allocate C_buf[int32 * 256]
-      for (i1, 0, 16) {
-        for (i3, 0, 16) {
-          A_buf[((i1*16) + i3)] = A[((i1*16) + i3)]
+    primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {A: Buffer(A_2: handle, int8, [1, 16, 1, 16], []),
+                 B: Buffer(B_2: handle, int8, [16, 16, 16, 16], []),
+                 C: Buffer(C_2: handle, int8, [1, 16, 1, 16], [])}
+      buffer_map = {C_1: C, B_1: B, A_1: A} {
+      attr [A_buf: handle] "storage_scope" = "global";
+      allocate(A_buf, int8, [256]);
+      attr [B_buf: handle] "storage_scope" = "global";
+      allocate(B_buf, int8, [65536]);
+      attr [C_buf: handle] "storage_scope" = "global";
+      allocate(C_buf, int32, [256]) {
+        for (i1: int32, 0, 16) {
+          for (i3: int32, 0, 16) {
+            A_buf[((i1*16) + i3)] = (int8*)A_2[((i1*16) + i3)])
+          }
         }
-      }
-      for (i0, 0, 16) {
-        for (i1, 0, 16) {
-          for (i2, 0, 16) {
-            for (i3, 0, 16) {
-              B_buf[((((i0*4096) + (i1*256)) + (i2*16)) + i3)] = B[((((i0*4096) + (i1*256)) + (i2*16)) + i3)]
+        for (i0: int32, 0, 16) {
+          for (i1_1: int32, 0, 16) {
+            for (i2: int32, 0, 16) {
+              for (i3_1: int32, 0, 16) {
+                B_buf[((((i0*4096) + (i1_1*256)) + (i2*16)) + i3_1)] = (int8*)B_2[((((i0*4096) + (i1_1*256)) + (i2*16)) + i3_1)])
+              }
             }
           }
         }
-      }
-      for (co, 0, 16) {
-        for (ci, 0, 16) {
-          C_buf[((co*16) + ci)] = 0
-          for (ko, 0, 16) {
-            for (ki, 0, 16) {
-              C_buf[((co*16) + ci)] = (C_buf[((co*16) + ci)] + (int32(A_buf[((ko*16) + ki)])*int32(B_buf[((((co*4096) + (ko*256)) + (ci*16)) + ki)])))
+        for (co: int32, 0, 16) {
+          for (ci: int32, 0, 16) {
+            C_buf[((co*16) + ci)] = 0
+            for (ko: int32, 0, 16) {
+              for (ki: int32, 0, 16) {
+                C_buf[((co*16) + ci)] = ((int32*)C_buf[((co*16) + ci)]) + (cast(int32, (int8*)A_buf[((ko*16) + ki)]))*cast(int32, (int8*)B_buf[((((co*4096) + (ko*256)) + (ci*16)) + ki)]))))
+              }
             }
           }
         }
-      }
-      for (i1, 0, 16) {
-        for (i3, 0, 16) {
-          C[((i1*16) + i3)] = int8(C_buf[((i1*16) + i3)])
+        for (i1_2: int32, 0, 16) {
+          for (i3_2: int32, 0, 16) {
+            C_2[((i1_2*16) + i3_2)] = cast(int8, (int32*)C_buf[((i1_2*16) + i3_2)]))
+          }
         }
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -442,40 +448,46 @@ moving the copy operations into the matrix multiplication loop.
 
  .. code-block:: none
 
-    PrimFunc([A, B, C]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      // attr [C_buf] storage_scope = "local.acc_buffer"
-      allocate C_buf[int32 * 256]
-      // attr [A_buf] storage_scope = "local.inp_buffer"
-      allocate A_buf[int8 * 16]
-      // attr [B_buf] storage_scope = "local.wgt_buffer"
-      allocate B_buf[int8 * 16]
-      for (co, 0, 16) {
-        for (ci, 0, 16) {
-          C_buf[((co*16) + ci)] = 0
-          for (ko, 0, 16) {
-            // attr [iter_var(i0, )] pragma_dma_copy = 1
-            for (i3, 0, 16) {
-              A_buf[i3] = A[((ko*16) + i3)]
-            }
-            // attr [iter_var(i0, )] pragma_dma_copy = 1
-            for (i3, 0, 16) {
-              B_buf[i3] = B[((((co*4096) + (ko*256)) + (ci*16)) + i3)]
-            }
-            for (ki, 0, 16) {
-              C_buf[((co*16) + ci)] = (C_buf[((co*16) + ci)] + (int32(A_buf[ki])*int32(B_buf[ki])))
+    primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {A: Buffer(A_2: handle, int8, [1, 16, 1, 16], []),
+                 B: Buffer(B_2: handle, int8, [16, 16, 16, 16], []),
+                 C: Buffer(C_2: handle, int8, [1, 16, 1, 16], [])}
+      buffer_map = {C_1: C, B_1: B, A_1: A} {
+      attr [C_buf: handle] "storage_scope" = "local.acc_buffer";
+      allocate(C_buf, int32, [256]);
+      attr [A_buf: handle] "storage_scope" = "local.inp_buffer";
+      allocate(A_buf, int8, [16]);
+      attr [B_buf: handle] "storage_scope" = "local.wgt_buffer";
+      allocate(B_buf, int8, [16]) {
+        for (co: int32, 0, 16) {
+          for (ci: int32, 0, 16) {
+            C_buf[((co*16) + ci)] = 0
+            for (ko: int32, 0, 16) {
+              attr [IterVar(i0: int32, (nullptr), "DataPar", "")] "pragma_dma_copy" = 1;
+              for (i3: int32, 0, 16) {
+                A_buf[i3] = (int8*)A_2[((ko*16) + i3)])
+              }
+              attr [IterVar(i0_1: int32, (nullptr), "DataPar", "")] "pragma_dma_copy" = 1;
+              for (i3_1: int32, 0, 16) {
+                B_buf[i3_1] = (int8*)B_2[((((co*4096) + (ko*256)) + (ci*16)) + i3_1)])
+              }
+              for (ki: int32, 0, 16) {
+                C_buf[((co*16) + ci)] = ((int32*)C_buf[((co*16) + ci)]) + (cast(int32, (int8*)A_buf[ki]))*cast(int32, (int8*)B_buf[ki]))))
+              }
             }
           }
         }
-      }
-      // attr [iter_var(i0, )] pragma_dma_copy = 1
-      for (i1, 0, 16) {
-        for (i3, 0, 16) {
-          C[((i1*16) + i3)] = int8(C_buf[((i1*16) + i3)])
+        attr [IterVar(i0_2: int32, (nullptr), "DataPar", "")] "pragma_dma_copy" = 1;
+        for (i1: int32, 0, 16) {
+          for (i3_2: int32, 0, 16) {
+            C_2[((i1*16) + i3_2)] = cast(int8, (int32*)C_buf[((i1*16) + i3_2)]))
+          }
         }
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -525,39 +537,51 @@ by the VTA runtime JIT compiler.
 
  .. code-block:: none
 
-    PrimFunc([A, B, C]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      // attr [C_buf] storage_scope = "local.acc_buffer"
-      // attr [A_buf] storage_scope = "local.inp_buffer"
-      // attr [B_buf] storage_scope = "local.wgt_buffer"
-      // attr [iter_var(vta, , vta)] coproc_scope = 2
-      // attr [iter_var(vta, , vta)] coproc_uop_scope = "VTAPushGEMMOp"
-      VTAUopLoopBegin(16, 1, 0, 0)
-      VTAUopPush(0, 1, 0, 0, 0, 0, 0, 0)
-      VTAUopLoopEnd()
-      vta.coproc_dep_push(2, 1)
-      for (ko, 0, 16) {
-        // attr [iter_var(vta, , vta)] coproc_scope = 1
-        vta.coproc_dep_pop(2, 1)
-        VTALoadBuffer2D(tvm_thread_context(VTATLSCommandHandle()), A, ko, 1, 1, 1, 0, 0, 0, 0, 0, 2)
-        VTALoadBuffer2D(tvm_thread_context(VTATLSCommandHandle()), B, ko, 1, 16, 16, 0, 0, 0, 0, 0, 1)
-        vta.coproc_dep_push(1, 2)
-        // attr [iter_var(vta, , vta)] coproc_scope = 2
-        vta.coproc_dep_pop(1, 2)
-        // attr [iter_var(vta, , vta)] coproc_uop_scope = "VTAPushGEMMOp"
-        VTAUopLoopBegin(16, 1, 0, 1)
-        VTAUopPush(0, 0, 0, 0, 0, 0, 0, 0)
-        VTAUopLoopEnd()
-        vta.coproc_dep_push(2, 1)
+    primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {A: Buffer(A_2: handle, int8, [1, 16, 1, 16], []),
+                 B: Buffer(B_2: handle, int8, [16, 16, 16, 16], []),
+                 C: Buffer(C_2: handle, int8, [1, 16, 1, 16], [])}
+      buffer_map = {C_1: C, B_1: B, A_1: A} {
+      attr [C_buf: handle] "storage_scope" = "local.acc_buffer";
+      attr [A_buf: handle] "storage_scope" = "local.inp_buffer";
+      attr [B_buf: handle] "storage_scope" = "local.wgt_buffer" {
+        attr [IterVar(vta: int32, (nullptr), "ThreadIndex", "vta")] "coproc_scope" = 2 {
+          attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_uop_scope" = "VTAPushGEMMOp" {
+            @VTAUopLoopBegin(16, 1, 0, 0, dtype=int32, type="extern", index=0)
+            @VTAUopPush(0, 1, 0, 0, 0, 0, 0, 0, dtype=int32, type="extern", index=0)
+            @VTAUopLoopEnd(, dtype=int32, type="extern", index=0)
+          }
+          @vta.coproc_dep_push(2, 1, dtype=int32, type="intrin", index=0)
+        }
+        for (ko: int32, 0, 16) {
+          attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_scope" = 1 {
+            @vta.coproc_dep_pop(2, 1, dtype=int32, type="intrin", index=0)
+            @VTALoadBuffer2D(@tvm_thread_context(@VTATLSCommandHandle(, dtype=handle, type="extern", index=0), dtype=handle, type="intrin", index=0), A_2, ko, 1, 1, 1, 0, 0, 0, 0, 0, 2, dtype=int32, type="extern", index=0)
+            @VTALoadBuffer2D(@tvm_thread_context(@VTATLSCommandHandle(, dtype=handle, type="extern", index=0), dtype=handle, type="intrin", index=0), B_2, ko, 1, 16, 16, 0, 0, 0, 0, 0, 1, dtype=int32, type="extern", index=0)
+            @vta.coproc_dep_push(1, 2, dtype=int32, type="intrin", index=0)
+          }
+          attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_scope" = 2 {
+            @vta.coproc_dep_pop(1, 2, dtype=int32, type="intrin", index=0)
+            attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_uop_scope" = "VTAPushGEMMOp" {
+              @VTAUopLoopBegin(16, 1, 0, 1, dtype=int32, type="extern", index=0)
+              @VTAUopPush(0, 0, 0, 0, 0, 0, 0, 0, dtype=int32, type="extern", index=0)
+              @VTAUopLoopEnd(, dtype=int32, type="extern", index=0)
+            }
+            @vta.coproc_dep_push(2, 1, dtype=int32, type="intrin", index=0)
+          }
+        }
+        @vta.coproc_dep_push(2, 3, dtype=int32, type="intrin", index=0)
+        @vta.coproc_dep_pop(2, 1, dtype=int32, type="intrin", index=0)
+        attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_scope" = 3 {
+          @vta.coproc_dep_pop(2, 3, dtype=int32, type="intrin", index=0)
+          @VTAStoreBuffer2D(@tvm_thread_context(@VTATLSCommandHandle(, dtype=handle, type="extern", index=0), dtype=handle, type="intrin", index=0), 0, 4, C_2, 0, 16, 1, 16, dtype=int32, type="extern", index=0)
+        }
+        @vta.coproc_sync(, dtype=int32, type="intrin", index=0)
       }
-      vta.coproc_dep_push(2, 3)
-      vta.coproc_dep_pop(2, 1)
-      // attr [iter_var(vta, , vta)] coproc_scope = 3
-      vta.coproc_dep_pop(2, 3)
-      VTAStoreBuffer2D(tvm_thread_context(VTATLSCommandHandle()), 0, 4, C, 0, 16, 1, 16)
-      vta.coproc_sync()
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
diff --git a/docs/_sources/vta/tutorials/optimize/convolution_opt.rst.txt b/docs/_sources/vta/tutorials/optimize/convolution_opt.rst.txt
index 8588f2d..530a346 100644
--- a/docs/_sources/vta/tutorials/optimize/convolution_opt.rst.txt
+++ b/docs/_sources/vta/tutorials/optimize/convolution_opt.rst.txt
@@ -255,45 +255,51 @@ Those include:
 
  .. code-block:: none
 
-    PrimFunc([data, kernel, res]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      // attr [data_buf] storage_scope = "global"
-      allocate data_buf[int8 * 65536]
-      // attr [kernel_buf] storage_scope = "global"
-      allocate kernel_buf[int8 * 589824]
-      // attr [res_conv] storage_scope = "global"
-      allocate res_conv[int32 * 50176]
-      for (i1, 0, 16) {
-        for (i2, 0, 16) {
-          for (i3, 0, 16) {
-            for (i5, 0, 16) {
-              data_buf[((((i1*4096) + (i2*256)) + (i3*16)) + i5)] = tvm_if_then_else(((((1 <= i2) && (i2 < 15)) && (1 <= i3)) && (i3 < 15)), data[(((((i1*3136) + (i2*224)) + (i3*16)) + i5) - 240)], (int8)0)
+    primfn(data_1: handle, kernel_1: handle, res_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {data: Buffer(data_2: handle, int8, [1, 16, 14, 14, 1, 16], []),
+                 res: Buffer(res_2: handle, int8, [1, 16, 14, 14, 1, 16], []),
+                 kernel: Buffer(kernel_2: handle, int8, [16, 16, 3, 3, 16, 16], [])}
+      buffer_map = {kernel_1: kernel, res_1: res, data_1: data} {
+      attr [data_buf: handle] "storage_scope" = "global";
+      allocate(data_buf, int8, [65536]);
+      attr [kernel_buf: handle] "storage_scope" = "global";
+      allocate(kernel_buf, int8, [589824]);
+      attr [res_conv: handle] "storage_scope" = "global";
+      allocate(res_conv, int32, [50176]) {
+        for (i1: int32, 0, 16) {
+          for (i2: int32, 0, 16) {
+            for (i3: int32, 0, 16) {
+              for (i5: int32, 0, 16) {
+                data_buf[((((i1*4096) + (i2*256)) + (i3*16)) + i5)] = @tvm_if_then_else(((((1 <= i2) && (i2 < 15)) && (1 <= i3)) && (i3 < 15)), (int8*)data_2[(((((i1*3136) + (i2*224)) + (i3*16)) + i5) - 240)]), 0i8, dtype=int8, type="pure_intrin", index=0)
+              }
             }
           }
         }
-      }
-      for (i0, 0, 16) {
-        for (i1, 0, 16) {
-          for (i2, 0, 3) {
-            for (i3, 0, 3) {
-              for (i4, 0, 16) {
-                for (i5, 0, 16) {
-                  kernel_buf[((((((i0*36864) + (i1*2304)) + (i2*768)) + (i3*256)) + (i4*16)) + i5)] = kernel[((((((i0*36864) + (i1*2304)) + (i2*768)) + (i3*256)) + (i4*16)) + i5)]
+        for (i0: int32, 0, 16) {
+          for (i1_1: int32, 0, 16) {
+            for (i2_1: int32, 0, 3) {
+              for (i3_1: int32, 0, 3) {
+                for (i4: int32, 0, 16) {
+                  for (i5_1: int32, 0, 16) {
+                    kernel_buf[((((((i0*36864) + (i1_1*2304)) + (i2_1*768)) + (i3_1*256)) + (i4*16)) + i5_1)] = (int8*)kernel_2[((((((i0*36864) + (i1_1*2304)) + (i2_1*768)) + (i3_1*256)) + (i4*16)) + i5_1)])
+                  }
                 }
               }
             }
           }
         }
-      }
-      for (co, 0, 16) {
-        for (i, 0, 14) {
-          for (j, 0, 14) {
-            for (ci, 0, 16) {
-              res_conv[((((co*3136) + (i*224)) + (j*16)) + ci)] = 0
-              for (ic, 0, 16) {
-                for (dy, 0, 3) {
-                  for (dx, 0, 3) {
-                    for (ic_tns, 0, 16) {
-                      res_conv[((((co*3136) + (i*224)) + (j*16)) + ci)] = (res_conv[((((co*3136) + (i*224)) + (j*16)) + ci)] + (int32(data_buf[((((((ic*4096) + (i*256)) + (dy*256)) + (j*16)) + (dx*16)) + ic_tns)])*int32(kernel_buf[((((((co*36864) + (ic*2304)) + (dy*768)) + (dx*256)) + (ci*16)) + ic_tns)])))
+        for (co: int32, 0, 16) {
+          for (i: int32, 0, 14) {
+            for (j: int32, 0, 14) {
+              for (ci: int32, 0, 16) {
+                res_conv[((((co*3136) + (i*224)) + (j*16)) + ci)] = 0
+                for (ic: int32, 0, 16) {
+                  for (dy: int32, 0, 3) {
+                    for (dx: int32, 0, 3) {
+                      for (ic_tns: int32, 0, 16) {
+                        res_conv[((((co*3136) + (i*224)) + (j*16)) + ci)] = ((int32*)res_conv[((((co*3136) + (i*224)) + (j*16)) + ci)]) + (cast(int32, (int8*)data_buf[((((((ic*4096) + (i*256)) + (dy*256)) + (j*16)) + (dx*16)) + ic_tns)]))*cast(int32, (int8*)kernel_buf[((((((co*36864) + (ic*2304)) + (dy*768)) + (dx*256)) + (ci*16)) + ic_tns)]))))
+                      }
                     }
                   }
                 }
@@ -301,46 +307,46 @@ Those include:
             }
           }
         }
-      }
-      for (i1, 0, 16) {
-        for (i2, 0, 14) {
-          for (i3, 0, 14) {
-            for (i5, 0, 16) {
-              res_conv[((((i1*3136) + (i2*224)) + (i3*16)) + i5)] = shift_right(res_conv[((((i1*3136) + (i2*224)) + (i3*16)) + i5)], 8)
+        for (i1_2: int32, 0, 16) {
+          for (i2_2: int32, 0, 14) {
+            for (i3_2: int32, 0, 14) {
+              for (i5_2: int32, 0, 16) {
+                res_conv[((((i1_2*3136) + (i2_2*224)) + (i3_2*16)) + i5_2)] = @shift_right((int32*)res_conv[((((i1_2*3136) + (i2_2*224)) + (i3_2*16)) + i5_2)]), 8, dtype=int32, type="pure_intrin", index=0)
+              }
             }
           }
         }
-      }
-      for (i1, 0, 16) {
-        for (i2, 0, 14) {
-          for (i3, 0, 14) {
-            for (i5, 0, 16) {
-              res_conv[((((i1*3136) + (i2*224)) + (i3*16)) + i5)] = max(res_conv[((((i1*3136) + (i2*224)) + (i3*16)) + i5)], 0)
+        for (i1_3: int32, 0, 16) {
+          for (i2_3: int32, 0, 14) {
+            for (i3_3: int32, 0, 14) {
+              for (i5_3: int32, 0, 16) {
+                res_conv[((((i1_3*3136) + (i2_3*224)) + (i3_3*16)) + i5_3)] = max((int32*)res_conv[((((i1_3*3136) + (i2_3*224)) + (i3_3*16)) + i5_3)]), 0)
+              }
             }
           }
         }
-      }
-      for (i1, 0, 16) {
-        for (i2, 0, 14) {
-          for (i3, 0, 14) {
-            for (i5, 0, 16) {
-              res_conv[((((i1*3136) + (i2*224)) + (i3*16)) + i5)] = min(res_conv[((((i1*3136) + (i2*224)) + (i3*16)) + i5)], 127)
+        for (i1_4: int32, 0, 16) {
+          for (i2_4: int32, 0, 14) {
+            for (i3_4: int32, 0, 14) {
+              for (i5_4: int32, 0, 16) {
+                res_conv[((((i1_4*3136) + (i2_4*224)) + (i3_4*16)) + i5_4)] = min((int32*)res_conv[((((i1_4*3136) + (i2_4*224)) + (i3_4*16)) + i5_4)]), 127)
+              }
             }
           }
         }
-      }
-      for (i1, 0, 16) {
-        for (i2, 0, 14) {
-          for (i3, 0, 14) {
-            for (i5, 0, 16) {
-              res[((((i1*3136) + (i2*224)) + (i3*16)) + i5)] = int8(res_conv[((((i1*3136) + (i2*224)) + (i3*16)) + i5)])
+        for (i1_5: int32, 0, 16) {
+          for (i2_5: int32, 0, 14) {
+            for (i3_5: int32, 0, 14) {
+              for (i5_5: int32, 0, 16) {
+                res_2[((((i1_5*3136) + (i2_5*224)) + (i3_5*16)) + i5_5)] = cast(int8, (int32*)res_conv[((((i1_5*3136) + (i2_5*224)) + (i3_5*16)) + i5_5)]))
+              }
             }
           }
         }
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -445,56 +451,62 @@ below.
 
  .. code-block:: none
 
-    PrimFunc([data, kernel, res]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      // attr [data_buf] storage_scope = "global"
-      allocate data_buf[int8 * 65536]
-      // attr [kernel_buf] storage_scope = "global"
-      allocate kernel_buf[int8 * 589824]
-      // attr [res_conv] storage_scope = "global"
-      allocate res_conv[int32 * 25088]
-      for (i1, 0, 16) {
-        for (i2, 0, 16) {
-          for (i3, 0, 16) {
-            for (i5, 0, 16) {
-              data_buf[((((i1*4096) + (i2*256)) + (i3*16)) + i5)] = tvm_if_then_else(((((1 <= i2) && (i2 < 15)) && (1 <= i3)) && (i3 < 15)), data[(((((i1*3136) + (i2*224)) + (i3*16)) + i5) - 240)], (int8)0)
+    primfn(data_1: handle, kernel_1: handle, res_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {kernel: Buffer(kernel_2: handle, int8, [16, 16, 3, 3, 16, 16], []),
+                 data: Buffer(data_2: handle, int8, [1, 16, 14, 14, 1, 16], []),
+                 res: Buffer(res_2: handle, int8, [1, 16, 14, 14, 1, 16], [])}
+      buffer_map = {res_1: res, kernel_1: kernel, data_1: data} {
+      attr [data_buf: handle] "storage_scope" = "global";
+      allocate(data_buf, int8, [65536]);
+      attr [kernel_buf: handle] "storage_scope" = "global";
+      allocate(kernel_buf, int8, [589824]);
+      attr [res_conv: handle] "storage_scope" = "global";
+      allocate(res_conv, int32, [25088]) {
+        for (i1: int32, 0, 16) {
+          for (i2: int32, 0, 16) {
+            for (i3: int32, 0, 16) {
+              for (i5: int32, 0, 16) {
+                data_buf[((((i1*4096) + (i2*256)) + (i3*16)) + i5)] = @tvm_if_then_else(((((1 <= i2) && (i2 < 15)) && (1 <= i3)) && (i3 < 15)), (int8*)data_2[(((((i1*3136) + (i2*224)) + (i3*16)) + i5) - 240)]), 0i8, dtype=int8, type="pure_intrin", index=0)
+              }
             }
           }
         }
-      }
-      for (i0, 0, 16) {
-        for (i1, 0, 16) {
-          for (i2, 0, 3) {
-            for (i3, 0, 3) {
-              for (i4, 0, 16) {
-                for (i5, 0, 16) {
-                  kernel_buf[((((((i0*36864) + (i1*2304)) + (i2*768)) + (i3*256)) + (i4*16)) + i5)] = kernel[((((((i0*36864) + (i1*2304)) + (i2*768)) + (i3*256)) + (i4*16)) + i5)]
+        for (i0: int32, 0, 16) {
+          for (i1_1: int32, 0, 16) {
+            for (i2_1: int32, 0, 3) {
+              for (i3_1: int32, 0, 3) {
+                for (i4: int32, 0, 16) {
+                  for (i5_1: int32, 0, 16) {
+                    kernel_buf[((((((i0*36864) + (i1_1*2304)) + (i2_1*768)) + (i3_1*256)) + (i4*16)) + i5_1)] = (int8*)kernel_2[((((((i0*36864) + (i1_1*2304)) + (i2_1*768)) + (i3_1*256)) + (i4*16)) + i5_1)])
+                  }
                 }
               }
             }
           }
         }
-      }
-      for (i2.outer, 0, 2) {
-        for (co.init, 0, 8) {
-          for (i.init, 0, 7) {
-            for (j.init, 0, 14) {
-              for (ci.init, 0, 16) {
-                res_conv[((((co.init*1568) + (i.init*224)) + (j.init*16)) + ci.init)] = 0
-                res_conv[(((((co.init*1568) + (i.init*224)) + (j.init*16)) + ci.init) + 12544)] = 0
+        for (i2.outer: int32, 0, 2) {
+          for (co.init: int32, 0, 8) {
+            for (i.init: int32, 0, 7) {
+              for (j.init: int32, 0, 14) {
+                for (ci.init: int32, 0, 16) {
+                  res_conv[((((co.init*1568) + (i.init*224)) + (j.init*16)) + ci.init)] = 0
+                  res_conv[(((((co.init*1568) + (i.init*224)) + (j.init*16)) + ci.init) + 12544)] = 0
+                }
               }
             }
           }
-        }
-        for (ic.outer, 0, 16) {
-          for (co, 0, 8) {
-            for (i, 0, 7) {
-              for (dy, 0, 3) {
-                for (dx, 0, 3) {
-                  for (j, 0, 14) {
-                    for (ci, 0, 16) {
-                      for (ic_tns, 0, 16) {
-                        res_conv[((((co*1568) + (i*224)) + (j*16)) + ci)] = (res_conv[((((co*1568) + (i*224)) + (j*16)) + ci)] + (int32(data_buf[(((((((ic.outer*4096) + (i2.outer*1792)) + (i*256)) + (dy*256)) + (j*16)) + (dx*16)) + ic_tns)])*int32(kernel_buf[((((((co*36864) + (ic.outer*2304)) + (dy*768)) + (dx*256)) + (ci*16)) + ic_tns)])))
-                        res_conv[(((((co*1568) + (i*224)) + (j*16)) + ci) + 12544)] = (res_conv[(((((co*1568) + (i*224)) + (j*16)) + ci) + 12544)] + (int32(data_buf[(((((((ic.outer*4096) + (i2.outer*1792)) + (i*256)) + (dy*256)) + (j*16)) + (dx*16)) + ic_tns)])*int32(kernel_buf[(((((((co*36864) + (ic.outer*2304)) + (dy*768)) + (dx*256)) + (ci*16)) + ic_tns) + 294912)])))
+          for (ic.outer: int32, 0, 16) {
+            for (co: int32, 0, 8) {
+              for (i: int32, 0, 7) {
+                for (dy: int32, 0, 3) {
+                  for (dx: int32, 0, 3) {
+                    for (j: int32, 0, 14) {
+                      for (ci: int32, 0, 16) {
+                        for (ic_tns: int32, 0, 16) {
+                          res_conv[((((co*1568) + (i*224)) + (j*16)) + ci)] = ((int32*)res_conv[((((co*1568) + (i*224)) + (j*16)) + ci)]) + (cast(int32, (int8*)data_buf[(((((((ic.outer*4096) + (i2.outer*1792)) + (i*256)) + (dy*256)) + (j*16)) + (dx*16)) + ic_tns)]))*cast(int32, (int8*)kernel_buf[((((((co*36864) + (ic.outer*2304)) + (dy*768)) + (dx*256)) + (ci*16)) + ic_tns)]))))
+                          res_conv[(((((co*1568) + (i*224)) + (j*16)) + ci) + 12544)] = ((int32*)res_conv[(((((co*1568) + (i*224)) + (j*16)) + ci) + 12544)]) + (cast(int32, (int8*)data_buf[(((((((ic.outer*4096) + (i2.outer*1792)) + (i*256)) + (dy*256)) + (j*16)) + (dx*16)) + ic_tns)]))*cast(int32, (int8*)kernel_buf[(((((((co*36864) + (ic.outer*2304)) + (dy*768)) + (dx*256)) + (ci*16)) + ic_tns) + 294912)]))))
+                        }
                       }
                     }
                   }
@@ -502,43 +514,43 @@ below.
               }
             }
           }
-        }
-        for (i1, 0, 8) {
-          for (i2, 0, 7) {
-            for (i3, 0, 14) {
-              for (i5, 0, 16) {
-                res_conv[((((i1*1568) + (i2*224)) + (i3*16)) + i5)] = shift_right(res_conv[((((i1*1568) + (i2*224)) + (i3*16)) + i5)], 8)
-                res_conv[(((((i1*1568) + (i2*224)) + (i3*16)) + i5) + 12544)] = shift_right(res_conv[(((((i1*1568) + (i2*224)) + (i3*16)) + i5) + 12544)], 8)
+          for (i1_2: int32, 0, 8) {
+            for (i2_2: int32, 0, 7) {
+              for (i3_2: int32, 0, 14) {
+                for (i5_2: int32, 0, 16) {
+                  res_conv[((((i1_2*1568) + (i2_2*224)) + (i3_2*16)) + i5_2)] = @shift_right((int32*)res_conv[((((i1_2*1568) + (i2_2*224)) + (i3_2*16)) + i5_2)]), 8, dtype=int32, type="pure_intrin", index=0)
+                  res_conv[(((((i1_2*1568) + (i2_2*224)) + (i3_2*16)) + i5_2) + 12544)] = @shift_right((int32*)res_conv[(((((i1_2*1568) + (i2_2*224)) + (i3_2*16)) + i5_2) + 12544)]), 8, dtype=int32, type="pure_intrin", index=0)
+                }
               }
             }
           }
-        }
-        for (i1, 0, 8) {
-          for (i2, 0, 7) {
-            for (i3, 0, 14) {
-              for (i5, 0, 16) {
-                res_conv[((((i1*1568) + (i2*224)) + (i3*16)) + i5)] = max(res_conv[((((i1*1568) + (i2*224)) + (i3*16)) + i5)], 0)
-                res_conv[(((((i1*1568) + (i2*224)) + (i3*16)) + i5) + 12544)] = max(res_conv[(((((i1*1568) + (i2*224)) + (i3*16)) + i5) + 12544)], 0)
+          for (i1_3: int32, 0, 8) {
+            for (i2_3: int32, 0, 7) {
+              for (i3_3: int32, 0, 14) {
+                for (i5_3: int32, 0, 16) {
+                  res_conv[((((i1_3*1568) + (i2_3*224)) + (i3_3*16)) + i5_3)] = max((int32*)res_conv[((((i1_3*1568) + (i2_3*224)) + (i3_3*16)) + i5_3)]), 0)
+                  res_conv[(((((i1_3*1568) + (i2_3*224)) + (i3_3*16)) + i5_3) + 12544)] = max((int32*)res_conv[(((((i1_3*1568) + (i2_3*224)) + (i3_3*16)) + i5_3) + 12544)]), 0)
+                }
               }
             }
           }
-        }
-        for (i1, 0, 8) {
-          for (i2, 0, 7) {
-            for (i3, 0, 14) {
-              for (i5, 0, 16) {
-                res_conv[((((i1*1568) + (i2*224)) + (i3*16)) + i5)] = min(res_conv[((((i1*1568) + (i2*224)) + (i3*16)) + i5)], 127)
-                res_conv[(((((i1*1568) + (i2*224)) + (i3*16)) + i5) + 12544)] = min(res_conv[(((((i1*1568) + (i2*224)) + (i3*16)) + i5) + 12544)], 127)
+          for (i1_4: int32, 0, 8) {
+            for (i2_4: int32, 0, 7) {
+              for (i3_4: int32, 0, 14) {
+                for (i5_4: int32, 0, 16) {
+                  res_conv[((((i1_4*1568) + (i2_4*224)) + (i3_4*16)) + i5_4)] = min((int32*)res_conv[((((i1_4*1568) + (i2_4*224)) + (i3_4*16)) + i5_4)]), 127)
+                  res_conv[(((((i1_4*1568) + (i2_4*224)) + (i3_4*16)) + i5_4) + 12544)] = min((int32*)res_conv[(((((i1_4*1568) + (i2_4*224)) + (i3_4*16)) + i5_4) + 12544)]), 127)
+                }
               }
             }
           }
-        }
-        for (i1.inner, 0, 8) {
-          for (i2.inner, 0, 7) {
-            for (i3.inner, 0, 14) {
-              for (i5, 0, 16) {
-                res[(((((i1.inner*3136) + (i2.outer*1568)) + (i2.inner*224)) + (i3.inner*16)) + i5)] = int8(res_conv[((((i1.inner*1568) + (i2.inner*224)) + (i3.inner*16)) + i5)])
-                res[((((((i1.inner*3136) + (i2.outer*1568)) + (i2.inner*224)) + (i3.inner*16)) + i5) + 25088)] = int8(res_conv[(((((i1.inner*1568) + (i2.inner*224)) + (i3.inner*16)) + i5) + 12544)])
+          for (i1.inner: int32, 0, 8) {
+            for (i2.inner: int32, 0, 7) {
+              for (i3.inner: int32, 0, 14) {
+                for (i5_5: int32, 0, 16) {
+                  res_2[(((((i1.inner*3136) + (i2.outer*1568)) + (i2.inner*224)) + (i3.inner*16)) + i5_5)] = cast(int8, (int32*)res_conv[((((i1.inner*1568) + (i2.inner*224)) + (i3.inner*16)) + i5_5)]))
+                  res_2[((((((i1.inner*3136) + (i2.outer*1568)) + (i2.inner*224)) + (i3.inner*16)) + i5_5) + 25088)] = cast(int8, (int32*)res_conv[(((((i1.inner*1568) + (i2.inner*224)) + (i3.inner*16)) + i5_5) + 12544)]))
+                }
               }
             }
           }
@@ -546,7 +558,7 @@ below.
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -622,92 +634,109 @@ and mapping the shift, and clipping computation to the vector ALU.
 
  .. code-block:: none
 
-    PrimFunc([data, kernel, res]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      // attr [res_conv] storage_scope = "local.acc_buffer"
-      // attr [data_buf] storage_scope = "local.inp_buffer"
-      // attr [kernel_buf] storage_scope = "local.wgt_buffer"
-      vta.coproc_dep_push(3, 2)
-      vta.coproc_dep_push(3, 2)
-      for (i2.outer, 0, 2) {
-        for (cthread.s, 0, 2) {
-          // attr [iter_var(vta, , vta)] coproc_scope = 2
-          vta.coproc_dep_pop(3, 2)
-          // attr [iter_var(vta, , vta)] coproc_uop_scope = "VTAPushGEMMOp"
-          VTAUopLoopBegin(8, 98, 0, 0)
-          VTAUopLoopBegin(7, 14, 0, 0)
-          for (j.init, 0, 14) {
-            VTAUopPush(0, 1, ((cthread.s*784) + j.init), 0, 0, 0, 0, 0)
+    primfn(data_1: handle, kernel_1: handle, res_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {kernel: Buffer(kernel_2: handle, int8, [16, 16, 3, 3, 16, 16], []),
+                 data: Buffer(data_2: handle, int8, [1, 16, 14, 14, 1, 16], []),
+                 res: Buffer(res_2: handle, int8, [1, 16, 14, 14, 1, 16], [])}
+      buffer_map = {res_1: res, kernel_1: kernel, data_1: data} {
+      attr [res_conv: handle] "storage_scope" = "local.acc_buffer";
+      attr [data_buf: handle] "storage_scope" = "local.inp_buffer";
+      attr [kernel_buf: handle] "storage_scope" = "local.wgt_buffer" {
+        @vta.coproc_dep_push(3, 2, dtype=int32, type="intrin", index=0)
+        @vta.coproc_dep_push(3, 2, dtype=int32, type="intrin", index=0)
+        for (i2.outer: int32, 0, 2) {
+          for (cthread.s: int32, 0, 2) {
+            attr [IterVar(vta: int32, (nullptr), "ThreadIndex", "vta")] "coproc_scope" = 2 {
+              @vta.coproc_dep_pop(3, 2, dtype=int32, type="intrin", index=0)
+              attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_uop_scope" = "VTAPushGEMMOp" {
+                @VTAUopLoopBegin(8, 98, 0, 0, dtype=int32, type="extern", index=0)
+                @VTAUopLoopBegin(7, 14, 0, 0, dtype=int32, type="extern", index=0)
+                for (j.init: int32, 0, 14) {
+                  @VTAUopPush(0, 1, ((cthread.s*784) + j.init), 0, 0, 0, 0, 0, dtype=int32, type="extern", index=0)
+                }
+                @VTAUopLoopEnd(, dtype=int32, type="extern", index=0)
+                @VTAUopLoopEnd(, dtype=int32, type="extern", index=0)
+              }
+              @vta.coproc_dep_push(2, 1, dtype=int32, type="intrin", index=0)
+            }
           }
-          VTAUopLoopEnd()
-          VTAUopLoopEnd()
-          vta.coproc_dep_push(2, 1)
-        }
-        for (ic.outer, 0, 16) {
-          // attr [iter_var(vta, , vta)] coproc_scope = 1
-          vta.coproc_dep_pop(2, 1)
-          VTALoadBuffer2D(tvm_thread_context(VTATLSCommandHandle()), data, ((((ic.outer*196) + (i2.outer*98)) + (max((1 - (i2.outer*7)), 0)*14)) - 14), 14, ((9 - max((1 - (i2.outer*7)), 0)) - max(((i2.outer*7) - 6), 0)), 14, 1, max((1 - (i2.outer*7)), 0), 1, max(((i2.outer*7) - 6), 0), 0, 2)
-          VTALoadBuffer2D(tvm_thread_context(VTATLSCommandHandle()), kernel, (ic.outer*9), 9, 8, 144, 0, 0, 0, 0, 0, 1)
-          vta.coproc_dep_push(1, 2)
-          // attr [iter_var(vta, , vta)] coproc_scope = 1
-          vta.coproc_dep_pop(2, 1)
-          VTALoadBuffer2D(tvm_thread_context(VTATLSCommandHandle()), data, ((((ic.outer*196) + (i2.outer*98)) + (max((1 - (i2.outer*7)), 0)*14)) - 14), 14, ((9 - max((1 - (i2.outer*7)), 0)) - max(((i2.outer*7) - 6), 0)), 14, 1, max((1 - (i2.outer*7)), 0), 1, max(((i2.outer*7) - 6), 0), 144, 2)
-          VTALoadBuffer2D(tvm_thread_context(VTATLSCommandHandle()), kernel, ((ic.outer*9) + 1152), 9, 8, 144, 0, 0, 0, 0, 72, 1)
-          vta.coproc_dep_push(1, 2)
-          for (cthread.s, 0, 2) {
-            // attr [iter_var(vta, , vta)] coproc_scope = 2
-            vta.coproc_dep_pop(1, 2)
-            // attr [iter_var(vta, , vta)] coproc_uop_scope = "VTAPushGEMMOp"
-            VTAUopLoopBegin(8, 98, 0, 9)
-            VTAUopLoopBegin(7, 14, 16, 0)
-            for (dy, 0, 3) {
-              for (dx, 0, 3) {
-                for (j, 0, 14) {
-                  VTAUopPush(0, 0, ((cthread.s*784) + j), ((((cthread.s*144) + (dy*16)) + j) + dx), (((cthread.s*72) + (dy*3)) + dx), 0, 0, 0)
+          for (ic.outer: int32, 0, 16) {
+            attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_scope" = 1 {
+              @vta.coproc_dep_pop(2, 1, dtype=int32, type="intrin", index=0)
+              @VTALoadBuffer2D(@tvm_thread_context(@VTATLSCommandHandle(, dtype=handle, type="extern", index=0), dtype=handle, type="intrin", index=0), data_2, ((((ic.outer*196) + (i2.outer*98)) + (max((1 - (i2.outer*7)), 0)*14)) - 14), 14, ((9 - max((1 - (i2.outer*7)), 0)) - max(((i2.outer*7) - 6), 0)), 14, 1, max((1 - (i2.outer*7)), 0), 1, max(((i2.outer*7) - 6), 0), 0, 2, dtype=int32, type="extern", index=0)
+              @VTALoadBuffer2D(@tvm_thread_context(@VTATLSCommandHandle(, dtype=handle, type="extern", index=0), dtype=handle, type="intrin", index=0), kernel_2, (ic.outer*9), 9, 8, 144, 0, 0, 0, 0, 0, 1, dtype=int32, type="extern", index=0)
+              @vta.coproc_dep_push(1, 2, dtype=int32, type="intrin", index=0)
+            }
+            attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_scope" = 1 {
+              @vta.coproc_dep_pop(2, 1, dtype=int32, type="intrin", index=0)
+              @VTALoadBuffer2D(@tvm_thread_context(@VTATLSCommandHandle(, dtype=handle, type="extern", index=0), dtype=handle, type="intrin", index=0), data_2, ((((ic.outer*196) + (i2.outer*98)) + (max((1 - (i2.outer*7)), 0)*14)) - 14), 14, ((9 - max((1 - (i2.outer*7)), 0)) - max(((i2.outer*7) - 6), 0)), 14, 1, max((1 - (i2.outer*7)), 0), 1, max(((i2.outer*7) - 6), 0), 144, 2, dtype=int32, type="extern", index=0)
+              @VTALoadBuffer2D(@tvm_thread_context(@VTATLSCommandHandle(, dtype=handle, type="extern", index=0), dtype=handle, type="intrin", index=0), kernel_2, ((ic.outer*9) + 1152), 9, 8, 144, 0, 0, 0, 0, 72, 1, dtype=int32, type="extern", index=0)
+              @vta.coproc_dep_push(1, 2, dtype=int32, type="intrin", index=0)
+            }
+            for (cthread.s_1: int32, 0, 2) {
+              attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_scope" = 2 {
+                @vta.coproc_dep_pop(1, 2, dtype=int32, type="intrin", index=0)
+                attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_uop_scope" = "VTAPushGEMMOp" {
+                  @VTAUopLoopBegin(8, 98, 0, 9, dtype=int32, type="extern", index=0)
+                  @VTAUopLoopBegin(7, 14, 16, 0, dtype=int32, type="extern", index=0)
+                  for (dy: int32, 0, 3) {
+                    for (dx: int32, 0, 3) {
+                      for (j: int32, 0, 14) {
+                        @VTAUopPush(0, 0, ((cthread.s_1*784) + j), ((((cthread.s_1*144) + (dy*16)) + j) + dx), (((cthread.s_1*72) + (dy*3)) + dx), 0, 0, 0, dtype=int32, type="extern", index=0)
+                      }
+                    }
+                  }
+                  @VTAUopLoopEnd(, dtype=int32, type="extern", index=0)
+                  @VTAUopLoopEnd(, dtype=int32, type="extern", index=0)
                 }
+                @vta.coproc_dep_push(2, 1, dtype=int32, type="intrin", index=0)
               }
             }
-            VTAUopLoopEnd()
-            VTAUopLoopEnd()
-            vta.coproc_dep_push(2, 1)
           }
-        }
-        vta.coproc_dep_pop(2, 1)
-        vta.coproc_dep_pop(2, 1)
-        for (cthread.s, 0, 2) {
-          // attr [iter_var(vta, , vta)] coproc_scope = 2
-          // attr [iter_var(vta, , vta)] coproc_uop_scope = "VTAPushALUOp"
-          VTAUopLoopBegin(784, 1, 1, 0)
-          VTAUopPush(1, 0, (cthread.s*784), (cthread.s*784), 0, 3, 1, 8)
-          VTAUopLoopEnd()
-          // attr [iter_var(vta, , vta)] coproc_uop_scope = "VTAPushALUOp"
-          VTAUopLoopBegin(784, 1, 1, 0)
-          VTAUopPush(1, 0, (cthread.s*784), (cthread.s*784), 0, 1, 1, 0)
-          VTAUopLoopEnd()
-          // attr [iter_var(vta, , vta)] coproc_uop_scope = "VTAPushALUOp"
-          VTAUopLoopBegin(784, 1, 1, 0)
-          VTAUopPush(1, 0, (cthread.s*784), (cthread.s*784), 0, 0, 1, 127)
-          VTAUopLoopEnd()
-          vta.coproc_dep_push(2, 3)
-        }
-        for (cthread.s, 0, 2) {
-          // attr [iter_var(vta, , vta)] coproc_scope = 3
-          vta.coproc_dep_pop(2, 3)
-          for (i1.inner, 0, 8) {
-            for (i2.inner, 0, 7) {
-              for (i3.inner, 0, 14) {
-                VTAStoreBuffer2D(tvm_thread_context(VTATLSCommandHandle()), ((((cthread.s*784) + (i1.inner*98)) + (i2.inner*14)) + i3.inner), 4, res, (((((cthread.s*1568) + (i1.inner*196)) + (i2.outer*98)) + (i2.inner*14)) + i3.inner), 1, 1, 1)
+          @vta.coproc_dep_pop(2, 1, dtype=int32, type="intrin", index=0)
+          @vta.coproc_dep_pop(2, 1, dtype=int32, type="intrin", index=0)
+          for (cthread.s_2: int32, 0, 2) {
+            attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_scope" = 2 {
+              attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_uop_scope" = "VTAPushALUOp" {
+                @VTAUopLoopBegin(784, 1, 1, 0, dtype=int32, type="extern", index=0)
+                @VTAUopPush(1, 0, (cthread.s_2*784), (cthread.s_2*784), 0, 3, 1, 8, dtype=int32, type="extern", index=0)
+                @VTAUopLoopEnd(, dtype=int32, type="extern", index=0)
+              }
+              attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_uop_scope" = "VTAPushALUOp" {
+                @VTAUopLoopBegin(784, 1, 1, 0, dtype=int32, type="extern", index=0)
+                @VTAUopPush(1, 0, (cthread.s_2*784), (cthread.s_2*784), 0, 1, 1, 0, dtype=int32, type="extern", index=0)
+                @VTAUopLoopEnd(, dtype=int32, type="extern", index=0)
+              }
+              attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_uop_scope" = "VTAPushALUOp" {
+                @VTAUopLoopBegin(784, 1, 1, 0, dtype=int32, type="extern", index=0)
+                @VTAUopPush(1, 0, (cthread.s_2*784), (cthread.s_2*784), 0, 0, 1, 127, dtype=int32, type="extern", index=0)
+                @VTAUopLoopEnd(, dtype=int32, type="extern", index=0)
               }
+              @vta.coproc_dep_push(2, 3, dtype=int32, type="intrin", index=0)
+            }
+          }
+          for (cthread.s_3: int32, 0, 2) {
+            attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_scope" = 3 {
+              @vta.coproc_dep_pop(2, 3, dtype=int32, type="intrin", index=0)
+              for (i1.inner: int32, 0, 8) {
+                for (i2.inner: int32, 0, 7) {
+                  for (i3.inner: int32, 0, 14) {
+                    @VTAStoreBuffer2D(@tvm_thread_context(@VTATLSCommandHandle(, dtype=handle, type="extern", index=0), dtype=handle, type="intrin", index=0), ((((cthread.s_3*784) + (i1.inner*98)) + (i2.inner*14)) + i3.inner), 4, res_2, (((((cthread.s_3*1568) + (i1.inner*196)) + (i2.outer*98)) + (i2.inner*14)) + i3.inner), 1, 1, 1, dtype=int32, type="extern", index=0)
+                  }
+                }
+              }
+              @vta.coproc_dep_push(3, 2, dtype=int32, type="intrin", index=0)
             }
           }
-          vta.coproc_dep_push(3, 2)
         }
+        @vta.coproc_dep_pop(3, 2, dtype=int32, type="intrin", index=0)
+        @vta.coproc_dep_pop(3, 2, dtype=int32, type="intrin", index=0)
+        @vta.coproc_sync(, dtype=int32, type="intrin", index=0)
       }
-      vta.coproc_dep_pop(3, 2)
-      vta.coproc_dep_pop(3, 2)
-      vta.coproc_sync()
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
diff --git a/docs/_sources/vta/tutorials/optimize/matrix_multiply_opt.rst.txt b/docs/_sources/vta/tutorials/optimize/matrix_multiply_opt.rst.txt
index 7a43616..3328747 100644
--- a/docs/_sources/vta/tutorials/optimize/matrix_multiply_opt.rst.txt
+++ b/docs/_sources/vta/tutorials/optimize/matrix_multiply_opt.rst.txt
@@ -200,60 +200,66 @@ Those include:
 
  .. code-block:: none
 
-    PrimFunc([data, weight, res]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      // attr [data_buf] storage_scope = "global"
-      allocate data_buf[int8 * 1024]
-      // attr [weight_buf] storage_scope = "global"
-      allocate weight_buf[int8 * 1048576]
-      // attr [res_gem] storage_scope = "global"
-      allocate res_gem[int32 * 1024]
-      for (i1, 0, 64) {
-        for (i3, 0, 16) {
-          data_buf[((i1*16) + i3)] = data[((i1*16) + i3)]
+    primfn(data_1: handle, weight_1: handle, res_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {data: Buffer(data_2: handle, int8, [1, 64, 1, 16], []),
+                 weight: Buffer(weight_2: handle, int8, [64, 64, 16, 16], []),
+                 res: Buffer(res_2: handle, int8, [1, 64, 1, 16], [])}
+      buffer_map = {res_1: res, weight_1: weight, data_1: data} {
+      attr [data_buf: handle] "storage_scope" = "global";
+      allocate(data_buf, int8, [1024]);
+      attr [weight_buf: handle] "storage_scope" = "global";
+      allocate(weight_buf, int8, [1048576]);
+      attr [res_gem: handle] "storage_scope" = "global";
+      allocate(res_gem, int32, [1024]) {
+        for (i1: int32, 0, 64) {
+          for (i3: int32, 0, 16) {
+            data_buf[((i1*16) + i3)] = (int8*)data_2[((i1*16) + i3)])
+          }
         }
-      }
-      for (i0, 0, 64) {
-        for (i1, 0, 64) {
-          for (i2, 0, 16) {
-            for (i3, 0, 16) {
-              weight_buf[((((i0*16384) + (i1*256)) + (i2*16)) + i3)] = weight[((((i0*16384) + (i1*256)) + (i2*16)) + i3)]
+        for (i0: int32, 0, 64) {
+          for (i1_1: int32, 0, 64) {
+            for (i2: int32, 0, 16) {
+              for (i3_1: int32, 0, 16) {
+                weight_buf[((((i0*16384) + (i1_1*256)) + (i2*16)) + i3_1)] = (int8*)weight_2[((((i0*16384) + (i1_1*256)) + (i2*16)) + i3_1)])
+              }
             }
           }
         }
-      }
-      for (co, 0, 64) {
-        for (ci, 0, 16) {
-          res_gem[((co*16) + ci)] = 0
-          for (ic, 0, 64) {
-            for (ic_tns, 0, 16) {
-              res_gem[((co*16) + ci)] = (res_gem[((co*16) + ci)] + (int32(data_buf[((ic*16) + ic_tns)])*int32(weight_buf[((((co*16384) + (ic*256)) + (ci*16)) + ic_tns)])))
+        for (co: int32, 0, 64) {
+          for (ci: int32, 0, 16) {
+            res_gem[((co*16) + ci)] = 0
+            for (ic: int32, 0, 64) {
+              for (ic_tns: int32, 0, 16) {
+                res_gem[((co*16) + ci)] = ((int32*)res_gem[((co*16) + ci)]) + (cast(int32, (int8*)data_buf[((ic*16) + ic_tns)]))*cast(int32, (int8*)weight_buf[((((co*16384) + (ic*256)) + (ci*16)) + ic_tns)]))))
+              }
             }
           }
         }
-      }
-      for (i1, 0, 64) {
-        for (i3, 0, 16) {
-          res_gem[((i1*16) + i3)] = shift_right(res_gem[((i1*16) + i3)], 8)
+        for (i1_2: int32, 0, 64) {
+          for (i3_2: int32, 0, 16) {
+            res_gem[((i1_2*16) + i3_2)] = @shift_right((int32*)res_gem[((i1_2*16) + i3_2)]), 8, dtype=int32, type="pure_intrin", index=0)
+          }
         }
-      }
-      for (i1, 0, 64) {
-        for (i3, 0, 16) {
-          res_gem[((i1*16) + i3)] = max(res_gem[((i1*16) + i3)], 0)
+        for (i1_3: int32, 0, 64) {
+          for (i3_3: int32, 0, 16) {
+            res_gem[((i1_3*16) + i3_3)] = max((int32*)res_gem[((i1_3*16) + i3_3)]), 0)
+          }
         }
-      }
-      for (i1, 0, 64) {
-        for (i3, 0, 16) {
-          res_gem[((i1*16) + i3)] = min(res_gem[((i1*16) + i3)], 127)
+        for (i1_4: int32, 0, 64) {
+          for (i3_4: int32, 0, 16) {
+            res_gem[((i1_4*16) + i3_4)] = min((int32*)res_gem[((i1_4*16) + i3_4)]), 127)
+          }
         }
-      }
-      for (i1, 0, 64) {
-        for (i3, 0, 16) {
-          res[((i1*16) + i3)] = int8(res_gem[((i1*16) + i3)])
+        for (i1_5: int32, 0, 64) {
+          for (i3_5: int32, 0, 16) {
+            res_2[((i1_5*16) + i3_5)] = cast(int8, (int32*)res_gem[((i1_5*16) + i3_5)]))
+          }
         }
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -356,68 +362,74 @@ below:
 
  .. code-block:: none
 
-    PrimFunc([data, weight, res]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      // attr [data_buf] storage_scope = "global"
-      allocate data_buf[int8 * 1024]
-      // attr [weight_buf] storage_scope = "global"
-      allocate weight_buf[int8 * 1048576]
-      // attr [res_gem] storage_scope = "global"
-      allocate res_gem[int32 * 256]
-      for (i1, 0, 64) {
-        for (i3, 0, 16) {
-          data_buf[((i1*16) + i3)] = data[((i1*16) + i3)]
+    primfn(data_1: handle, weight_1: handle, res_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {data: Buffer(data_2: handle, int8, [1, 64, 1, 16], []),
+                 weight: Buffer(weight_2: handle, int8, [64, 64, 16, 16], []),
+                 res: Buffer(res_2: handle, int8, [1, 64, 1, 16], [])}
+      buffer_map = {res_1: res, weight_1: weight, data_1: data} {
+      attr [data_buf: handle] "storage_scope" = "global";
+      allocate(data_buf, int8, [1024]);
+      attr [weight_buf: handle] "storage_scope" = "global";
+      allocate(weight_buf, int8, [1048576]);
+      attr [res_gem: handle] "storage_scope" = "global";
+      allocate(res_gem, int32, [256]) {
+        for (i1: int32, 0, 64) {
+          for (i3: int32, 0, 16) {
+            data_buf[((i1*16) + i3)] = (int8*)data_2[((i1*16) + i3)])
+          }
         }
-      }
-      for (i0, 0, 64) {
-        for (i1, 0, 64) {
-          for (i2, 0, 16) {
-            for (i3, 0, 16) {
-              weight_buf[((((i0*16384) + (i1*256)) + (i2*16)) + i3)] = weight[((((i0*16384) + (i1*256)) + (i2*16)) + i3)]
+        for (i0: int32, 0, 64) {
+          for (i1_1: int32, 0, 64) {
+            for (i2: int32, 0, 16) {
+              for (i3_1: int32, 0, 16) {
+                weight_buf[((((i0*16384) + (i1_1*256)) + (i2*16)) + i3_1)] = (int8*)weight_2[((((i0*16384) + (i1_1*256)) + (i2*16)) + i3_1)])
+              }
             }
           }
         }
-      }
-      for (i1.outer, 0, 4) {
-        for (co.init, 0, 16) {
-          for (ci.init, 0, 16) {
-            res_gem[((co.init*16) + ci.init)] = 0
+        for (i1.outer: int32, 0, 4) {
+          for (co.init: int32, 0, 16) {
+            for (ci.init: int32, 0, 16) {
+              res_gem[((co.init*16) + ci.init)] = 0
+            }
           }
-        }
-        for (ic.outer, 0, 4) {
-          for (co, 0, 16) {
-            for (ic.inner, 0, 16) {
-              for (ci, 0, 16) {
-                for (ic_tns, 0, 16) {
-                  res_gem[((co*16) + ci)] = (res_gem[((co*16) + ci)] + (int32(data_buf[(((ic.outer*256) + (ic.inner*16)) + ic_tns)])*int32(weight_buf[((((((i1.outer*262144) + (co*16384)) + (ic.outer*4096)) + (ic.inner*256)) + (ci*16)) + ic_tns)])))
+          for (ic.outer: int32, 0, 4) {
+            for (co: int32, 0, 16) {
+              for (ic.inner: int32, 0, 16) {
+                for (ci: int32, 0, 16) {
+                  for (ic_tns: int32, 0, 16) {
+                    res_gem[((co*16) + ci)] = ((int32*)res_gem[((co*16) + ci)]) + (cast(int32, (int8*)data_buf[(((ic.outer*256) + (ic.inner*16)) + ic_tns)]))*cast(int32, (int8*)weight_buf[((((((i1.outer*262144) + (co*16384)) + (ic.outer*4096)) + (ic.inner*256)) + (ci*16)) + ic_tns)]))))
+                  }
                 }
               }
             }
           }
-        }
-        for (i1, 0, 16) {
-          for (i3, 0, 16) {
-            res_gem[((i1*16) + i3)] = shift_right(res_gem[((i1*16) + i3)], 8)
+          for (i1_2: int32, 0, 16) {
+            for (i3_2: int32, 0, 16) {
+              res_gem[((i1_2*16) + i3_2)] = @shift_right((int32*)res_gem[((i1_2*16) + i3_2)]), 8, dtype=int32, type="pure_intrin", index=0)
+            }
           }
-        }
-        for (i1, 0, 16) {
-          for (i3, 0, 16) {
-            res_gem[((i1*16) + i3)] = max(res_gem[((i1*16) + i3)], 0)
+          for (i1_3: int32, 0, 16) {
+            for (i3_3: int32, 0, 16) {
+              res_gem[((i1_3*16) + i3_3)] = max((int32*)res_gem[((i1_3*16) + i3_3)]), 0)
+            }
           }
-        }
-        for (i1, 0, 16) {
-          for (i3, 0, 16) {
-            res_gem[((i1*16) + i3)] = min(res_gem[((i1*16) + i3)], 127)
+          for (i1_4: int32, 0, 16) {
+            for (i3_4: int32, 0, 16) {
+              res_gem[((i1_4*16) + i3_4)] = min((int32*)res_gem[((i1_4*16) + i3_4)]), 127)
+            }
           }
-        }
-        for (i1.inner, 0, 16) {
-          for (i3, 0, 16) {
-            res[(((i1.outer*256) + (i1.inner*16)) + i3)] = int8(res_gem[((i1.inner*16) + i3)])
+          for (i1.inner: int32, 0, 16) {
+            for (i3_5: int32, 0, 16) {
+              res_2[(((i1.outer*256) + (i1.inner*16)) + i3_5)] = cast(int8, (int32*)res_gem[((i1.inner*16) + i3_5)]))
+            }
           }
         }
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -493,62 +505,78 @@ and mapping the shift, and clipping computation to the vector ALU.
 
  .. code-block:: none
 
-    PrimFunc([data, weight, res]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      // attr [res_gem] storage_scope = "local.acc_buffer"
-      // attr [data_buf] storage_scope = "local.inp_buffer"
-      // attr [weight_buf] storage_scope = "local.wgt_buffer"
-      vta.coproc_dep_push(3, 2)
-      for (i1.outer, 0, 4) {
-        // attr [iter_var(vta, , vta)] coproc_scope = 2
-        vta.coproc_dep_pop(3, 2)
-        // attr [iter_var(vta, , vta)] coproc_uop_scope = "VTAPushGEMMOp"
-        VTAUopLoopBegin(16, 1, 0, 0)
-        VTAUopPush(0, 1, 0, 0, 0, 0, 0, 0)
-        VTAUopLoopEnd()
-        vta.coproc_dep_push(2, 1)
-        for (ic.outer, 0, 4) {
-          // attr [iter_var(vta, , vta)] coproc_scope = 1
-          vta.coproc_dep_pop(2, 1)
-          VTALoadBuffer2D(tvm_thread_context(VTATLSCommandHandle()), data, (ic.outer*16), 16, 1, 16, 0, 0, 0, 0, 0, 2)
-          VTALoadBuffer2D(tvm_thread_context(VTATLSCommandHandle()), weight, ((i1.outer*1024) + (ic.outer*16)), 16, 16, 64, 0, 0, 0, 0, 0, 1)
-          vta.coproc_dep_push(1, 2)
-          // attr [iter_var(vta, , vta)] coproc_scope = 2
-          vta.coproc_dep_pop(1, 2)
-          // attr [iter_var(vta, , vta)] coproc_uop_scope = "VTAPushGEMMOp"
-          VTAUopLoopBegin(16, 1, 0, 16)
-          VTAUopLoopBegin(16, 0, 1, 1)
-          VTAUopPush(0, 0, 0, 0, 0, 0, 0, 0)
-          VTAUopLoopEnd()
-          VTAUopLoopEnd()
-          vta.coproc_dep_push(2, 1)
-        }
-        vta.coproc_dep_pop(2, 1)
-        // attr [iter_var(vta, , vta)] coproc_scope = 2
-        // attr [iter_var(vta, , vta)] coproc_uop_scope = "VTAPushALUOp"
-        VTAUopLoopBegin(16, 1, 1, 0)
-        VTAUopPush(1, 0, 0, 0, 0, 3, 1, 8)
-        VTAUopLoopEnd()
-        // attr [iter_var(vta, , vta)] coproc_uop_scope = "VTAPushALUOp"
-        VTAUopLoopBegin(16, 1, 1, 0)
-        VTAUopPush(1, 0, 0, 0, 0, 1, 1, 0)
-        VTAUopLoopEnd()
-        // attr [iter_var(vta, , vta)] coproc_uop_scope = "VTAPushALUOp"
-        VTAUopLoopBegin(16, 1, 1, 0)
-        VTAUopPush(1, 0, 0, 0, 0, 0, 1, 127)
-        VTAUopLoopEnd()
-        vta.coproc_dep_push(2, 3)
-        // attr [iter_var(vta, , vta)] coproc_scope = 3
-        vta.coproc_dep_pop(2, 3)
-        for (i1.inner, 0, 16) {
-          VTAStoreBuffer2D(tvm_thread_context(VTATLSCommandHandle()), i1.inner, 4, res, ((i1.outer*16) + i1.inner), 1, 1, 1)
+    primfn(data_1: handle, weight_1: handle, res_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {weight: Buffer(weight_2: handle, int8, [64, 64, 16, 16], []),
+                 data: Buffer(data_2: handle, int8, [1, 64, 1, 16], []),
+                 res: Buffer(res_2: handle, int8, [1, 64, 1, 16], [])}
+      buffer_map = {res_1: res, weight_1: weight, data_1: data} {
+      attr [res_gem: handle] "storage_scope" = "local.acc_buffer";
+      attr [data_buf: handle] "storage_scope" = "local.inp_buffer";
+      attr [weight_buf: handle] "storage_scope" = "local.wgt_buffer" {
+        @vta.coproc_dep_push(3, 2, dtype=int32, type="intrin", index=0)
+        for (i1.outer: int32, 0, 4) {
+          attr [IterVar(vta: int32, (nullptr), "ThreadIndex", "vta")] "coproc_scope" = 2 {
+            @vta.coproc_dep_pop(3, 2, dtype=int32, type="intrin", index=0)
+            attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_uop_scope" = "VTAPushGEMMOp" {
+              @VTAUopLoopBegin(16, 1, 0, 0, dtype=int32, type="extern", index=0)
+              @VTAUopPush(0, 1, 0, 0, 0, 0, 0, 0, dtype=int32, type="extern", index=0)
+              @VTAUopLoopEnd(, dtype=int32, type="extern", index=0)
+            }
+            @vta.coproc_dep_push(2, 1, dtype=int32, type="intrin", index=0)
+          }
+          for (ic.outer: int32, 0, 4) {
+            attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_scope" = 1 {
+              @vta.coproc_dep_pop(2, 1, dtype=int32, type="intrin", index=0)
+              @VTALoadBuffer2D(@tvm_thread_context(@VTATLSCommandHandle(, dtype=handle, type="extern", index=0), dtype=handle, type="intrin", index=0), data_2, (ic.outer*16), 16, 1, 16, 0, 0, 0, 0, 0, 2, dtype=int32, type="extern", index=0)
+              @VTALoadBuffer2D(@tvm_thread_context(@VTATLSCommandHandle(, dtype=handle, type="extern", index=0), dtype=handle, type="intrin", index=0), weight_2, ((i1.outer*1024) + (ic.outer*16)), 16, 16, 64, 0, 0, 0, 0, 0, 1, dtype=int32, type="extern", index=0)
+              @vta.coproc_dep_push(1, 2, dtype=int32, type="intrin", index=0)
+            }
+            attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_scope" = 2 {
+              @vta.coproc_dep_pop(1, 2, dtype=int32, type="intrin", index=0)
+              attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_uop_scope" = "VTAPushGEMMOp" {
+                @VTAUopLoopBegin(16, 1, 0, 16, dtype=int32, type="extern", index=0)
+                @VTAUopLoopBegin(16, 0, 1, 1, dtype=int32, type="extern", index=0)
+                @VTAUopPush(0, 0, 0, 0, 0, 0, 0, 0, dtype=int32, type="extern", index=0)
+                @VTAUopLoopEnd(, dtype=int32, type="extern", index=0)
+                @VTAUopLoopEnd(, dtype=int32, type="extern", index=0)
+              }
+              @vta.coproc_dep_push(2, 1, dtype=int32, type="intrin", index=0)
+            }
+          }
+          @vta.coproc_dep_pop(2, 1, dtype=int32, type="intrin", index=0)
+          attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_scope" = 2 {
+            attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_uop_scope" = "VTAPushALUOp" {
+              @VTAUopLoopBegin(16, 1, 1, 0, dtype=int32, type="extern", index=0)
+              @VTAUopPush(1, 0, 0, 0, 0, 3, 1, 8, dtype=int32, type="extern", index=0)
+              @VTAUopLoopEnd(, dtype=int32, type="extern", index=0)
+            }
+            attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_uop_scope" = "VTAPushALUOp" {
+              @VTAUopLoopBegin(16, 1, 1, 0, dtype=int32, type="extern", index=0)
+              @VTAUopPush(1, 0, 0, 0, 0, 1, 1, 0, dtype=int32, type="extern", index=0)
+              @VTAUopLoopEnd(, dtype=int32, type="extern", index=0)
+            }
+            attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_uop_scope" = "VTAPushALUOp" {
+              @VTAUopLoopBegin(16, 1, 1, 0, dtype=int32, type="extern", index=0)
+              @VTAUopPush(1, 0, 0, 0, 0, 0, 1, 127, dtype=int32, type="extern", index=0)
+              @VTAUopLoopEnd(, dtype=int32, type="extern", index=0)
+            }
+            @vta.coproc_dep_push(2, 3, dtype=int32, type="intrin", index=0)
+          }
+          attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_scope" = 3 {
+            @vta.coproc_dep_pop(2, 3, dtype=int32, type="intrin", index=0)
+            for (i1.inner: int32, 0, 16) {
+              @VTAStoreBuffer2D(@tvm_thread_context(@VTATLSCommandHandle(, dtype=handle, type="extern", index=0), dtype=handle, type="intrin", index=0), i1.inner, 4, res_2, ((i1.outer*16) + i1.inner), 1, 1, 1, dtype=int32, type="extern", index=0)
+            }
+            @vta.coproc_dep_push(3, 2, dtype=int32, type="intrin", index=0)
+          }
         }
-        vta.coproc_dep_push(3, 2)
+        @vta.coproc_sync(, dtype=int32, type="intrin", index=0)
+        @vta.coproc_dep_pop(3, 2, dtype=int32, type="intrin", index=0)
       }
-      vta.coproc_sync()
-      vta.coproc_dep_pop(3, 2)
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
diff --git a/docs/_sources/vta/tutorials/optimize/sg_execution_times.rst.txt b/docs/_sources/vta/tutorials/optimize/sg_execution_times.rst.txt
index 4849591..6fae073 100644
--- a/docs/_sources/vta/tutorials/optimize/sg_execution_times.rst.txt
+++ b/docs/_sources/vta/tutorials/optimize/sg_execution_times.rst.txt
@@ -5,7 +5,7 @@
 
 Computation times
 =================
-**00:02.910** total execution time for **vta_tutorials_optimize** files:
+**00:02.917** total execution time for **vta_tutorials_optimize** files:
 
-- **00:02.543**: :ref:`sphx_glr_vta_tutorials_optimize_convolution_opt.py` (``convolution_opt.py``)
-- **00:00.367**: :ref:`sphx_glr_vta_tutorials_optimize_matrix_multiply_opt.py` (``matrix_multiply_opt.py``)
+- **00:02.541**: :ref:`sphx_glr_vta_tutorials_optimize_convolution_opt.py` (``convolution_opt.py``)
+- **00:00.376**: :ref:`sphx_glr_vta_tutorials_optimize_matrix_multiply_opt.py` (``matrix_multiply_opt.py``)
diff --git a/docs/_sources/vta/tutorials/sg_execution_times.rst.txt b/docs/_sources/vta/tutorials/sg_execution_times.rst.txt
index ca5cb48..5f53aae 100644
--- a/docs/_sources/vta/tutorials/sg_execution_times.rst.txt
+++ b/docs/_sources/vta/tutorials/sg_execution_times.rst.txt
@@ -5,7 +5,7 @@
 
 Computation times
 =================
-**00:00.674** total execution time for **vta_tutorials** files:
+**00:00.682** total execution time for **vta_tutorials** files:
 
-- **00:00.340**: :ref:`sphx_glr_vta_tutorials_matrix_multiply.py` (``matrix_multiply.py``)
-- **00:00.333**: :ref:`sphx_glr_vta_tutorials_vta_get_started.py` (``vta_get_started.py``)
+- **00:00.344**: :ref:`sphx_glr_vta_tutorials_matrix_multiply.py` (``matrix_multiply.py``)
+- **00:00.339**: :ref:`sphx_glr_vta_tutorials_vta_get_started.py` (``vta_get_started.py``)
diff --git a/docs/_sources/vta/tutorials/vta_get_started.rst.txt b/docs/_sources/vta/tutorials/vta_get_started.rst.txt
index e0694ef..420c2d9 100644
--- a/docs/_sources/vta/tutorials/vta_get_started.rst.txt
+++ b/docs/_sources/vta/tutorials/vta_get_started.rst.txt
@@ -298,34 +298,40 @@ After we construct the schedule, by default the schedule computes
 
  .. code-block:: none
 
-    PrimFunc([A, B, C]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      // attr [A_buf] storage_scope = "global"
-      allocate A_buf[int32 * 1024]
-      // attr [B_buf] storage_scope = "global"
-      allocate B_buf[int32 * 1024]
-      for (i1, 0, 64) {
-        for (i3, 0, 16) {
-          A_buf[((i1*16) + i3)] = A[((i1*16) + i3)]
+    primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {B: Buffer(B_2: handle, int32, [1, 64, 1, 16], []),
+                 A: Buffer(A_2: handle, int32, [1, 64, 1, 16], []),
+                 C: Buffer(C_2: handle, int8, [1, 64, 1, 16], [])}
+      buffer_map = {C_1: C, B_1: B, A_1: A} {
+      attr [A_buf: handle] "storage_scope" = "global";
+      allocate(A_buf, int32, [1024]);
+      attr [B_buf: handle] "storage_scope" = "global";
+      allocate(B_buf, int32, [1024]) {
+        for (i1: int32, 0, 64) {
+          for (i3: int32, 0, 16) {
+            A_buf[((i1*16) + i3)] = (int32*)A_2[((i1*16) + i3)])
+          }
         }
-      }
-      for (i1, 0, 64) {
-        for (i3, 0, 16) {
-          B_buf[((i1*16) + i3)] = B[((i1*16) + i3)]
+        for (i1_1: int32, 0, 64) {
+          for (i3_1: int32, 0, 16) {
+            B_buf[((i1_1*16) + i3_1)] = (int32*)B_2[((i1_1*16) + i3_1)])
+          }
         }
-      }
-      for (i1, 0, 64) {
-        for (i3, 0, 16) {
-          A_buf[((i1*16) + i3)] = (A_buf[((i1*16) + i3)] + B_buf[((i1*16) + i3)])
+        for (i1_2: int32, 0, 64) {
+          for (i3_2: int32, 0, 16) {
+            A_buf[((i1_2*16) + i3_2)] = ((int32*)A_buf[((i1_2*16) + i3_2)]) + (int32*)B_buf[((i1_2*16) + i3_2)]))
+          }
         }
-      }
-      for (i1, 0, 64) {
-        for (i3, 0, 16) {
-          C[((i1*16) + i3)] = int8(A_buf[((i1*16) + i3)])
+        for (i1_3: int32, 0, 64) {
+          for (i3_3: int32, 0, 16) {
+            C_2[((i1_3*16) + i3_3)] = cast(int8, (int32*)A_buf[((i1_3*16) + i3_3)]))
+          }
         }
       }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
@@ -415,23 +421,32 @@ with an :code:`env.alu` pragma.
 
  .. code-block:: none
 
-    PrimFunc([A, B, C]) attrs={"tir.noalias": (bool)1, "global_symbol": "main"} {
-      // attr [A_buf] storage_scope = "local.acc_buffer"
-      // attr [iter_var(vta, , vta)] coproc_scope = 2
-      VTALoadBuffer2D(tvm_thread_context(VTATLSCommandHandle()), A, 0, 64, 1, 64, 0, 0, 0, 0, 0, 3)
-      VTALoadBuffer2D(tvm_thread_context(VTATLSCommandHandle()), B, 0, 64, 1, 64, 0, 0, 0, 0, 64, 3)
-      // attr [iter_var(vta, , vta)] coproc_uop_scope = "VTAPushALUOp"
-      VTAUopLoopBegin(64, 1, 1, 0)
-      VTAUopPush(1, 0, 0, 64, 0, 2, 0, 0)
-      VTAUopLoopEnd()
-      vta.coproc_dep_push(2, 3)
-      // attr [iter_var(vta, , vta)] coproc_scope = 3
-      vta.coproc_dep_pop(2, 3)
-      VTAStoreBuffer2D(tvm_thread_context(VTATLSCommandHandle()), 0, 4, C, 0, 64, 1, 64)
-      vta.coproc_sync()
+    primfn(A_1: handle, B_1: handle, C_1: handle) -> ()
+      attr = {"tir.noalias": True, "global_symbol": "main"}
+      buffers = {B: Buffer(B_2: handle, int32, [1, 64, 1, 16], []),
+                 A: Buffer(A_2: handle, int32, [1, 64, 1, 16], []),
+                 C: Buffer(C_2: handle, int8, [1, 64, 1, 16], [])}
+      buffer_map = {C_1: C, B_1: B, A_1: A} {
+      attr [A_buf: handle] "storage_scope" = "local.acc_buffer" {
+        attr [IterVar(vta: int32, (nullptr), "ThreadIndex", "vta")] "coproc_scope" = 2 {
+          @VTALoadBuffer2D(@tvm_thread_context(@VTATLSCommandHandle(, dtype=handle, type="extern", index=0), dtype=handle, type="intrin", index=0), A_2, 0, 64, 1, 64, 0, 0, 0, 0, 0, 3, dtype=int32, type="extern", index=0)
+          @VTALoadBuffer2D(@tvm_thread_context(@VTATLSCommandHandle(, dtype=handle, type="extern", index=0), dtype=handle, type="intrin", index=0), B_2, 0, 64, 1, 64, 0, 0, 0, 0, 64, 3, dtype=int32, type="extern", index=0)
+          attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_uop_scope" = "VTAPushALUOp" {
+            @VTAUopLoopBegin(64, 1, 1, 0, dtype=int32, type="extern", index=0)
+            @VTAUopPush(1, 0, 0, 64, 0, 2, 0, 0, dtype=int32, type="extern", index=0)
+            @VTAUopLoopEnd(, dtype=int32, type="extern", index=0)
+          }
+          @vta.coproc_dep_push(2, 3, dtype=int32, type="intrin", index=0)
+        }
+        attr [IterVar(vta, (nullptr), "ThreadIndex", "vta")] "coproc_scope" = 3 {
+          @vta.coproc_dep_pop(2, 3, dtype=int32, type="intrin", index=0)
+          @VTAStoreBuffer2D(@tvm_thread_context(@VTATLSCommandHandle(, dtype=handle, type="extern", index=0), dtype=handle, type="intrin", index=0), 0, 4, C_2, 0, 64, 1, 64, dtype=int32, type="extern", index=0)
+        }
+        @vta.coproc_sync(, dtype=int32, type="intrin", index=0)
+      }
     }
 
-
+    // meta data omitted. you can use show_meta_data=True to include meta data
 
 
 
diff --git a/docs/api/links.html b/docs/api/links.html
new file mode 100644
index 0000000..033dc3b
--- /dev/null
+++ b/docs/api/links.html
@@ -0,0 +1,251 @@
+
+
+<!DOCTYPE html>
+<!--[if IE 8]><html class="no-js lt-ie9" lang="en" > <![endif]-->
+<!--[if gt IE 8]><!--> <html class="no-js" lang="en" > <!--<![endif]-->
+<head>
+  <meta charset="utf-8">
+  
+  <meta name="viewport" content="width=device-width, initial-scale=1.0">
+  
+  <title>Links to API References &mdash; tvm 0.7.dev1 documentation</title>
+  
+
+  
+  
+    <link rel="shortcut icon" href="../_static/tvm-logo-square.png"/>
+  
+  
+  
+
+  
+  <script type="text/javascript" src="../_static/js/modernizr.min.js"></script>
+  
+    
+      <script type="text/javascript" id="documentation_options" data-url_root="../" src="../_static/documentation_options.js"></script>
+        <script src="../_static/jquery.js"></script>
+        <script src="../_static/underscore.js"></script>
+        <script src="../_static/doctools.js"></script>
+        <script src="../_static/language_data.js"></script>
+        <script async="async" src="https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.5/latest.js?config=TeX-AMS-MML_HTMLorMML"></script>
+    
+    <script type="text/javascript" src="../_static/js/theme.js"></script>
+
+    
+
+  
+  <link rel="stylesheet" href="../_static/css/theme.css" type="text/css" />
+  <link rel="stylesheet" href="../_static/pygments.css" type="text/css" />
+  <link rel="stylesheet" href="../_static/gallery.css" type="text/css" />
+  <link rel="stylesheet" href="../_static/css/tvm_theme.css" type="text/css" />
+    <link rel="index" title="Index" href="../genindex.html" />
+    <link rel="search" title="Search" href="../search.html" />
+    <link rel="next" title="Design and Developer Guide" href="../dev/index.html" />
+    <link rel="prev" title="topi" href="python/topi.html" /> 
+</head>
+
+<body class="wy-body-for-nav">
+
+   
+  <div class="wy-grid-for-nav">
+    
+    <nav data-toggle="wy-nav-shift" class="wy-nav-side">
+      <div class="wy-side-scroll">
+        <div class="wy-side-nav-search" >
+          
+
+          
+            <a href="../index.html">
+          
+
+          
+            
+            <img src="../_static/tvm-logo-small.png" class="logo" alt="Logo"/>
+          
+          </a>
+
+          
+            
+            
+              <div class="version">
+                0.7.dev1
+              </div>
+            
+          
+
+          
+<div role="search">
+  <form id="rtd-search-form" class="wy-form" action="../search.html" method="get">
+    <input type="text" name="q" placeholder="Search docs" />
+    <input type="hidden" name="check_keywords" value="yes" />
+    <input type="hidden" name="area" value="default" />
+  </form>
+</div>
+
+          
+        </div>
+
+        <div class="wy-menu wy-menu-vertical" data-spy="affix" role="navigation" aria-label="main navigation">
+          
+            
+            
+              
+            
+            
+              <ul>
+<li class="toctree-l1"><a class="reference internal" href="../install/index.html">Installation</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../tutorials/index.html">Tutorials</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../vta/index.html">VTA: Deep Learning Accelerator Stack</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../deploy/index.html">Deploy and Integration</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../contribute/index.html">Contribute to TVM</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../faq.html">Frequently Asked Questions</a></li>
+</ul>
+<ul class="current">
+<li class="toctree-l1"><a class="reference internal" href="../langref/index.html">Language Reference</a></li>
+<li class="toctree-l1"><a class="reference internal" href="python/index.html">Python API</a></li>
+<li class="toctree-l1 current"><a class="current reference internal" href="#">Links to API References</a></li>
+</ul>
+<ul>
+<li class="toctree-l1"><a class="reference internal" href="../dev/index.html">Design and Developer Guide</a></li>
+</ul>
+<ul>
+<li class="toctree-l1"><a class="reference internal" href="../frontend/tensorflow.html">TensorFlow Frontend</a></li>
+</ul>
+<ul>
+<li class="toctree-l1"><a class="reference internal" href="../genindex.html">Index</a></li>
+</ul>
+
+            
+          
+        </div>
+      </div>
+    </nav>
+
+    <section data-toggle="wy-nav-shift" class="wy-nav-content-wrap">
+
+      
+      <nav class="wy-nav-top" aria-label="top navigation">
+        
+          <i data-toggle="wy-nav-top" class="fa fa-bars"></i>
+          <a href="../index.html">tvm</a>
+        
+      </nav>
+
+
+      <div class="wy-nav-content">
+        
+        <div class="rst-content">
+        
+          
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+<div role="navigation" aria-label="breadcrumbs navigation">
+
+  <ul class="wy-breadcrumbs">
+    
+      <li><a href="../index.html">Docs</a> &raquo;</li>
+        
+      <li>Links to API References</li>
+    
+    
+      <li class="wy-breadcrumbs-aside">
+        
+            
+            <a href="../_sources/api/links.rst.txt" rel="nofollow"> View page source</a>
+          
+        
+      </li>
+    
+  </ul>
+
+  
+  <hr/>
+</div>
+          <div role="main" class="document" itemscope="itemscope" itemtype="http://schema.org/Article">
+           <div itemprop="articleBody">
+            
+  <div class="section" id="links-to-api-references">
+<h1>Links to API References<a class="headerlink" href="#links-to-api-references" title="Permalink to this headline">¶</a></h1>
+<p>This page contains links to API references that are build with different doc build system.</p>
+<ul class="simple">
+<li><p><a class="reference external" href="doxygen/index.html">C++ doyxgen API</a></p></li>
+<li><p><a class="reference external" href="typedoc/index.html">Typescript typedoc API</a></p></li>
+<li><p><a class="reference external" href="javadoc/index.html">Java Javadoc API</a></p></li>
+</ul>
+</div>
+
+
+           </div>
+           
+          </div>
+          <footer>
+  
+    <div class="rst-footer-buttons" role="navigation" aria-label="footer navigation">
+      
+        <a href="../dev/index.html" class="btn btn-neutral float-right" title="Design and Developer Guide" accesskey="n" rel="next">Next <span class="fa fa-arrow-circle-right"></span></a>
+      
+      
+        <a href="python/topi.html" class="btn btn-neutral float-left" title="topi" accesskey="p" rel="prev"><span class="fa fa-arrow-circle-left"></span> Previous</a>
+      
+    </div>
+  
+
+  <hr/>
+
+  <div role="contentinfo">
+    <p>
+        &copy; Copyright 2020, Apache Software Foundation
+
+    </p>
+  </div>
+  Built with <a href="http://sphinx-doc.org/">Sphinx</a> using a <a href="https://github.com/rtfd/sphinx_rtd_theme">theme</a> provided by <a href="https://readthedocs.org">Read the Docs</a>. 
+
+</footer>
+
+        </div>
+      </div>
+
+    </section>
+
+  </div>
+  
+
+
+  <script type="text/javascript">
+      jQuery(function () {
+          SphinxRtdTheme.Navigation.enable(true);
+      });
+  </script>
+
+  
+  
+    
+    <!-- Theme Analytics -->
+    <script>
+    (function(i,s,o,g,r,a,m){i['GoogleAnalyticsObject']=r;i[r]=i[r]||function(){
+      (i[r].q=i[r].q||[]).push(arguments)},i[r].l=1*new Date();a=s.createElement(o),
+      m=s.getElementsByTagName(o)[0];a.async=1;a.src=g;m.parentNode.insertBefore(a,m)
+    })(window,document,'script','https://www.google-analytics.com/analytics.js','ga');
+
+    ga('create', 'UA-75982049-2', 'auto');
+    ga('send', 'pageview');
+    </script>
+
+    
+   
+
+</body>
+</html>
\ No newline at end of file
diff --git a/docs/api/python/autotvm.html b/docs/api/python/autotvm.html
index 9e5738f..47612f1 100644
--- a/docs/api/python/autotvm.html
+++ b/docs/api/python/autotvm.html
@@ -139,7 +139,7 @@
 <li class="toctree-l2"><a class="reference internal" href="topi.html">topi</a></li>
 </ul>
 </li>
-<li class="toctree-l1"><a class="reference internal" href="../../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../../dev/index.html">Design and Developer Guide</a></li>
diff --git a/docs/api/python/contrib.html b/docs/api/python/contrib.html
index b7d195f..97bd7d3 100644
--- a/docs/api/python/contrib.html
+++ b/docs/api/python/contrib.html
@@ -133,7 +133,7 @@
 <li class="toctree-l3"><a class="reference internal" href="#module-tvm.contrib.cc">tvm.contrib.cc</a></li>
 <li class="toctree-l3"><a class="reference internal" href="#module-tvm.contrib.cublas">tvm.contrib.cublas</a></li>
 <li class="toctree-l3"><a class="reference internal" href="#module-tvm.contrib.dlpack">tvm.contrib.dlpack</a></li>
-<li class="toctree-l3"><a class="reference internal" href="#module-tvm.contrib.emscripten">tvm.contrib.emscripten</a></li>
+<li class="toctree-l3"><a class="reference internal" href="#module-tvm.contrib.emcc">tvm.contrib.emcc</a></li>
 <li class="toctree-l3"><a class="reference internal" href="#module-tvm.contrib.miopen">tvm.contrib.miopen</a></li>
 <li class="toctree-l3"><a class="reference internal" href="#module-tvm.contrib.mxnet">tvm.contrib.mxnet</a></li>
 <li class="toctree-l3"><a class="reference internal" href="#module-tvm.contrib.ndk">tvm.contrib.ndk</a></li>
@@ -155,7 +155,7 @@
 <li class="toctree-l2"><a class="reference internal" href="topi.html">topi</a></li>
 </ul>
 </li>
-<li class="toctree-l1"><a class="reference internal" href="../../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../../dev/index.html">Design and Developer Guide</a></li>
@@ -376,7 +376,7 @@ we will try to guess the matched clang version.</p></li>
 
 <dl class="py function">
 <dt id="tvm.contrib.cc.cross_compiler">
-<code class="sig-prename descclassname">tvm.contrib.cc.</code><code class="sig-name descname">cross_compiler</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">compile_func</span></em>, <em class="sig-param"><span class="n">options</span><span class="o">=</span><span class="default_value">None</span></em>, <em class="sig-param"><span class="n">output_format</span><span class="o">=</span><span class="default_value">None</span></em>, <em class="sig-param"><span cla [...]
+<code class="sig-prename descclassname">tvm.contrib.cc.</code><code class="sig-name descname">cross_compiler</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">compile_func</span></em>, <em class="sig-param"><span class="n">options</span><span class="o">=</span><span class="default_value">None</span></em>, <em class="sig-param"><span class="n">output_format</span><span class="o">=</span><span class="default_value">None</span></em>, <em class="sig-param"><span cla [...]
 <dd><p>Create a cross compiler function by specializing compile_func with options.</p>
 <p>This function can be used to construct compile functions that
 can be passed to AutoTVM measure or export_library.</p>
@@ -387,6 +387,8 @@ can be passed to AutoTVM measure or export_library.</p>
 <li><p><strong>options</strong> (<em>Optional</em><em>[</em><em>List</em><em>[</em><a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.8)"><em>str</em></a><em>]</em><em>]</em>) – List of additional optional string.</p></li>
 <li><p><strong>output_format</strong> (<em>Optional</em><em>[</em><a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.8)"><em>str</em></a><em>]</em>) – Library output format.</p></li>
 <li><p><strong>get_target_triple</strong> (<em>Optional</em><em>[</em><em>Callable</em><em>]</em>) – Function that can target triple according to dumpmachine option of compiler.</p></li>
+<li><p><strong>add_files</strong> (<em>Optional</em><em>[</em><em>List</em><em>[</em><a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.8)"><em>str</em></a><em>]</em><em>]</em>) – List of paths to additional object, source, library files
+to pass as part of the compilation.</p></li>
 </ul>
 </dd>
 <dt class="field-even">Returns</dt>
@@ -499,13 +501,13 @@ can be passed to AutoTVM measure or export_library.</p>
 </dd></dl>
 
 </div>
-<div class="section" id="module-tvm.contrib.emscripten">
-<span id="tvm-contrib-emscripten"></span><h2>tvm.contrib.emscripten<a class="headerlink" href="#module-tvm.contrib.emscripten" title="Permalink to this headline">¶</a></h2>
+<div class="section" id="module-tvm.contrib.emcc">
+<span id="tvm-contrib-emcc"></span><h2>tvm.contrib.emcc<a class="headerlink" href="#module-tvm.contrib.emcc" title="Permalink to this headline">¶</a></h2>
 <p>Util to invoke emscripten compilers in the system.</p>
 <dl class="py function">
-<dt id="tvm.contrib.emscripten.create_js">
-<code class="sig-prename descclassname">tvm.contrib.emscripten.</code><code class="sig-name descname">create_js</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">output</span></em>, <em class="sig-param"><span class="n">objects</span></em>, <em class="sig-param"><span class="n">options</span><span class="o">=</span><span class="default_value">None</span></em>, <em class="sig-param"><span class="n">side_module</span><span class="o">=</span><span class="default_va [...]
-<dd><p>Create emscripten javascript library.</p>
+<dt id="tvm.contrib.emcc.create_tvmjs_wasm">
+<code class="sig-prename descclassname">tvm.contrib.emcc.</code><code class="sig-name descname">create_tvmjs_wasm</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">output</span></em>, <em class="sig-param"><span class="n">objects</span></em>, <em class="sig-param"><span class="n">options</span><span class="o">=</span><span class="default_value">None</span></em>, <em class="sig-param"><span class="n">cc</span><span class="o">=</span><span class="default_value">'e [...]
+<dd><p>Create wasm that is supposed to run with the tvmjs.</p>
 <dl class="field-list simple">
 <dt class="field-odd">Parameters</dt>
 <dd class="field-odd"><ul class="simple">
@@ -1121,12 +1123,24 @@ we will try to guess the matched clang version.</p></li>
 <div class="section" id="module-tvm.contrib.util">
 <span id="tvm-contrib-util"></span><h2>tvm.contrib.util<a class="headerlink" href="#module-tvm.contrib.util" title="Permalink to this headline">¶</a></h2>
 <p>Common system utilities</p>
+<dl class="py exception">
+<dt id="tvm.contrib.util.DirectoryCreatedPastAtExit">
+<em class="property">exception </em><code class="sig-prename descclassname">tvm.contrib.util.</code><code class="sig-name descname">DirectoryCreatedPastAtExit</code><a class="headerlink" href="#tvm.contrib.util.DirectoryCreatedPastAtExit" title="Permalink to this definition">¶</a></dt>
+<dd><p>Raised when a TempDirectory is created after the atexit hook runs.</p>
+</dd></dl>
+
 <dl class="py class">
 <dt id="tvm.contrib.util.TempDirectory">
 <em class="property">class </em><code class="sig-prename descclassname">tvm.contrib.util.</code><code class="sig-name descname">TempDirectory</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">custom_path</span><span class="o">=</span><span class="default_value">None</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.contrib.util.TempDirectory" title="Permalink to this definition">¶</a></dt>
 <dd><p>Helper object to manage temp directory during testing.</p>
 <p>Automatically removes the directory when it went out of scope.</p>
 <dl class="py method">
+<dt id="tvm.contrib.util.TempDirectory.set_keep_for_debug">
+<em class="property">classmethod </em><code class="sig-name descname">set_keep_for_debug</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">set_to</span><span class="o">=</span><span class="default_value">True</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.contrib.util.TempDirectory.set_keep_for_debug" title="Permalink to this definition">¶</a></dt>
+<dd><p>Keep temporary directories past program exit for debugging.</p>
+</dd></dl>
+
+<dl class="py method">
 <dt id="tvm.contrib.util.TempDirectory.remove">
 <code class="sig-name descname">remove</code><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="headerlink" href="#tvm.contrib.util.TempDirectory.remove" title="Permalink to this definition">¶</a></dt>
 <dd><p>Remote the tmp dir</p>
diff --git a/docs/api/python/driver.html b/docs/api/python/driver.html
index 9558ce7..9041e6e 100644
--- a/docs/api/python/driver.html
+++ b/docs/api/python/driver.html
@@ -133,7 +133,7 @@
 <li class="toctree-l2"><a class="reference internal" href="topi.html">topi</a></li>
 </ul>
 </li>
-<li class="toctree-l1"><a class="reference internal" href="../../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../../dev/index.html">Design and Developer Guide</a></li>
diff --git a/docs/api/python/error.html b/docs/api/python/error.html
index fced61f..a7e0acc 100644
--- a/docs/api/python/error.html
+++ b/docs/api/python/error.html
@@ -133,7 +133,7 @@
 <li class="toctree-l2"><a class="reference internal" href="topi.html">topi</a></li>
 </ul>
 </li>
-<li class="toctree-l1"><a class="reference internal" href="../../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../../dev/index.html">Design and Developer Guide</a></li>
@@ -245,7 +245,10 @@ copy the examples and raise errors with the same message convention.</p>
 <tr class="row-even"><td><p><a class="reference internal" href="#tvm.error.OpNotImplemented" title="tvm.error.OpNotImplemented"><code class="xref py py-obj docutils literal notranslate"><span class="pre">OpNotImplemented</span></code></a></p></td>
 <td><p>Operator is not implemented.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.error.TVMError" title="tvm.error.TVMError"><code class="xref py py-obj docutils literal notranslate"><span class="pre">TVMError</span></code></a></p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.error.RPCError" title="tvm.error.RPCError"><code class="xref py py-obj docutils literal notranslate"><span class="pre">RPCError</span></code></a></p></td>
+<td><p>Error thrown by the remote server handling the RPC call.</p></td>
+</tr>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.error.TVMError" title="tvm.error.TVMError"><code class="xref py py-obj docutils literal notranslate"><span class="pre">TVMError</span></code></a></p></td>
 <td><p>Default error thrown by TVM functions.</p></td>
 </tr>
 </tbody>
@@ -314,6 +317,12 @@ copy the examples and raise errors with the same message convention.</p>
 </dd></dl>
 
 <dl class="py exception">
+<dt id="tvm.error.RPCError">
+<em class="property">exception </em><code class="sig-prename descclassname">tvm.error.</code><code class="sig-name descname">RPCError</code><a class="headerlink" href="#tvm.error.RPCError" title="Permalink to this definition">¶</a></dt>
+<dd><p>Error thrown by the remote server handling the RPC call.</p>
+</dd></dl>
+
+<dl class="py exception">
 <dt id="tvm.error.OpError">
 <em class="property">exception </em><code class="sig-prename descclassname">tvm.error.</code><code class="sig-name descname">OpError</code><a class="headerlink" href="#tvm.error.OpError" title="Permalink to this definition">¶</a></dt>
 <dd><p>Base class of all operator errors in frontends.</p>
diff --git a/docs/api/python/graph_runtime.html b/docs/api/python/graph_runtime.html
index b1756e9..8ca9615 100644
--- a/docs/api/python/graph_runtime.html
+++ b/docs/api/python/graph_runtime.html
@@ -133,7 +133,7 @@
 <li class="toctree-l2"><a class="reference internal" href="topi.html">topi</a></li>
 </ul>
 </li>
-<li class="toctree-l1"><a class="reference internal" href="../../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../../dev/index.html">Design and Developer Guide</a></li>
diff --git a/docs/api/python/index.html b/docs/api/python/index.html
index 9474299..9a9163b 100644
--- a/docs/api/python/index.html
+++ b/docs/api/python/index.html
@@ -133,7 +133,7 @@
 <li class="toctree-l2"><a class="reference internal" href="topi.html">topi</a></li>
 </ul>
 </li>
-<li class="toctree-l1"><a class="reference internal" href="../../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../../dev/index.html">Design and Developer Guide</a></li>
@@ -247,7 +247,7 @@
 <li class="toctree-l2"><a class="reference internal" href="contrib.html#module-tvm.contrib.cc">tvm.contrib.cc</a></li>
 <li class="toctree-l2"><a class="reference internal" href="contrib.html#module-tvm.contrib.cublas">tvm.contrib.cublas</a></li>
 <li class="toctree-l2"><a class="reference internal" href="contrib.html#module-tvm.contrib.dlpack">tvm.contrib.dlpack</a></li>
-<li class="toctree-l2"><a class="reference internal" href="contrib.html#module-tvm.contrib.emscripten">tvm.contrib.emscripten</a></li>
+<li class="toctree-l2"><a class="reference internal" href="contrib.html#module-tvm.contrib.emcc">tvm.contrib.emcc</a></li>
 <li class="toctree-l2"><a class="reference internal" href="contrib.html#module-tvm.contrib.miopen">tvm.contrib.miopen</a></li>
 <li class="toctree-l2"><a class="reference internal" href="contrib.html#module-tvm.contrib.mxnet">tvm.contrib.mxnet</a></li>
 <li class="toctree-l2"><a class="reference internal" href="contrib.html#module-tvm.contrib.ndk">tvm.contrib.ndk</a></li>
diff --git a/docs/api/python/ir.html b/docs/api/python/ir.html
index 9476999..8cfdef5 100644
--- a/docs/api/python/ir.html
+++ b/docs/api/python/ir.html
@@ -133,7 +133,7 @@
 <li class="toctree-l2"><a class="reference internal" href="topi.html">topi</a></li>
 </ul>
 </li>
-<li class="toctree-l1"><a class="reference internal" href="../../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../../dev/index.html">Design and Developer Guide</a></li>
diff --git a/docs/api/python/ndarray.html b/docs/api/python/ndarray.html
index c55efa8..0e2de09 100644
--- a/docs/api/python/ndarray.html
+++ b/docs/api/python/ndarray.html
@@ -133,7 +133,7 @@
 <li class="toctree-l2"><a class="reference internal" href="topi.html">topi</a></li>
 </ul>
 </li>
-<li class="toctree-l1"><a class="reference internal" href="../../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../../dev/index.html">Design and Developer Guide</a></li>
diff --git a/docs/api/python/relay/analysis.html b/docs/api/python/relay/analysis.html
index 2624873..795cfb7 100644
--- a/docs/api/python/relay/analysis.html
+++ b/docs/api/python/relay/analysis.html
@@ -133,7 +133,7 @@
 <li class="toctree-l2"><a class="reference internal" href="../topi.html">topi</a></li>
 </ul>
 </li>
-<li class="toctree-l1"><a class="reference internal" href="../../../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../../links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../../../dev/index.html">Design and Developer Guide</a></li>
diff --git a/docs/api/python/relay/backend.html b/docs/api/python/relay/backend.html
index f358467..c6a4da7 100644
--- a/docs/api/python/relay/backend.html
+++ b/docs/api/python/relay/backend.html
@@ -133,7 +133,7 @@
 <li class="toctree-l2"><a class="reference internal" href="../topi.html">topi</a></li>
 </ul>
 </li>
-<li class="toctree-l1"><a class="reference internal" href="../../../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../../links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../../../dev/index.html">Design and Developer Guide</a></li>
diff --git a/docs/api/python/relay/frontend.html b/docs/api/python/relay/frontend.html
index 7785574..daea393 100644
--- a/docs/api/python/relay/frontend.html
+++ b/docs/api/python/relay/frontend.html
@@ -133,7 +133,7 @@
 <li class="toctree-l2"><a class="reference internal" href="../topi.html">topi</a></li>
 </ul>
 </li>
-<li class="toctree-l1"><a class="reference internal" href="../../../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../../links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../../../dev/index.html">Design and Developer Guide</a></li>
@@ -335,7 +335,7 @@ This can be helpful for some testing.</p></li>
 <dl class="field-list simple">
 <dt class="field-odd">Parameters</dt>
 <dd class="field-odd"><ul class="simple">
-<li><p><strong>model</strong> – tflite.Model.Model</p></li>
+<li><p><strong>model</strong> – tflite.Model or tflite.Model.Model (depending on tflite version)</p></li>
 <li><p><strong>shape_dict</strong> (<em>dict of str to int list/tuple</em>) – Input shapes of the model.</p></li>
 <li><p><strong>dtype_dict</strong> (<em>dict of str to str</em>) – Input types of the model.</p></li>
 </ul>
diff --git a/docs/api/python/relay/image.html b/docs/api/python/relay/image.html
index ad27190..07f5d2e 100644
--- a/docs/api/python/relay/image.html
+++ b/docs/api/python/relay/image.html
@@ -133,7 +133,7 @@
 <li class="toctree-l2"><a class="reference internal" href="../topi.html">topi</a></li>
 </ul>
 </li>
-<li class="toctree-l1"><a class="reference internal" href="../../../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../../links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../../../dev/index.html">Design and Developer Guide</a></li>
diff --git a/docs/api/python/relay/index.html b/docs/api/python/relay/index.html
index d56e8fd..52d9515 100644
--- a/docs/api/python/relay/index.html
+++ b/docs/api/python/relay/index.html
@@ -133,7 +133,7 @@
 <li class="toctree-l2"><a class="reference internal" href="../topi.html">topi</a></li>
 </ul>
 </li>
-<li class="toctree-l1"><a class="reference internal" href="../../../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../../links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../../../dev/index.html">Design and Developer Guide</a></li>
@@ -321,6 +321,12 @@
 <tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.abs" title="tvm.relay.abs"><code class="xref py py-obj docutils literal notranslate"><span class="pre">abs</span></code></a>(data)</p></td>
 <td><p>Compute element-wise absolute of data.</p></td>
 </tr>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.acos" title="tvm.relay.acos"><code class="xref py py-obj docutils literal notranslate"><span class="pre">acos</span></code></a>(data)</p></td>
+<td><p>Compute elementwise acos of data.</p></td>
+</tr>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.acosh" title="tvm.relay.acosh"><code class="xref py py-obj docutils literal notranslate"><span class="pre">acosh</span></code></a>(data)</p></td>
+<td><p>Compute elementwise acosh of data.</p></td>
+</tr>
 <tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.add" title="tvm.relay.add"><code class="xref py py-obj docutils literal notranslate"><span class="pre">add</span></code></a>(lhs, rhs)</p></td>
 <td><p>Addition with numpy-style broadcasting.</p></td>
 </tr>
@@ -345,156 +351,168 @@
 <tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.argwhere" title="tvm.relay.argwhere"><code class="xref py py-obj docutils literal notranslate"><span class="pre">argwhere</span></code></a>(condition)</p></td>
 <td><p>Find the indices of elements of a tensor that are non-zero.</p></td>
 </tr>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.asin" title="tvm.relay.asin"><code class="xref py py-obj docutils literal notranslate"><span class="pre">asin</span></code></a>(data)</p></td>
+<td><p>Compute elementwise asin of data.</p></td>
+</tr>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.asinh" title="tvm.relay.asinh"><code class="xref py py-obj docutils literal notranslate"><span class="pre">asinh</span></code></a>(data)</p></td>
+<td><p>Compute elementwise asinh of data.</p></td>
+</tr>
 <tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.atan" title="tvm.relay.atan"><code class="xref py py-obj docutils literal notranslate"><span class="pre">atan</span></code></a>(data)</p></td>
 <td><p>Compute elementwise atan of data.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.bind" title="tvm.relay.bind"><code class="xref py py-obj docutils literal notranslate"><span class="pre">bind</span></code></a>(expr, binds)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.atanh" title="tvm.relay.atanh"><code class="xref py py-obj docutils literal notranslate"><span class="pre">atanh</span></code></a>(data)</p></td>
+<td><p>Compute elementwise atanh of data.</p></td>
+</tr>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.bind" title="tvm.relay.bind"><code class="xref py py-obj docutils literal notranslate"><span class="pre">bind</span></code></a>(expr, binds)</p></td>
 <td><p>Bind an free variables in expr or function arguments.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.bitwise_and" title="tvm.relay.bitwise_and"><code class="xref py py-obj docutils literal notranslate"><span class="pre">bitwise_and</span></code></a>(lhs, rhs)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.bitwise_and" title="tvm.relay.bitwise_and"><code class="xref py py-obj docutils literal notranslate"><span class="pre">bitwise_and</span></code></a>(lhs, rhs)</p></td>
 <td><p>bitwise AND with numpy-style broadcasting.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.bitwise_not" title="tvm.relay.bitwise_not"><code class="xref py py-obj docutils literal notranslate"><span class="pre">bitwise_not</span></code></a>(data)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.bitwise_not" title="tvm.relay.bitwise_not"><code class="xref py py-obj docutils literal notranslate"><span class="pre">bitwise_not</span></code></a>(data)</p></td>
 <td><p>Compute element-wise bitwise not of data.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.bitwise_or" title="tvm.relay.bitwise_or"><code class="xref py py-obj docutils literal notranslate"><span class="pre">bitwise_or</span></code></a>(lhs, rhs)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.bitwise_or" title="tvm.relay.bitwise_or"><code class="xref py py-obj docutils literal notranslate"><span class="pre">bitwise_or</span></code></a>(lhs, rhs)</p></td>
 <td><p>bitwise OR with numpy-style broadcasting.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.bitwise_xor" title="tvm.relay.bitwise_xor"><code class="xref py py-obj docutils literal notranslate"><span class="pre">bitwise_xor</span></code></a>(lhs, rhs)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.bitwise_xor" title="tvm.relay.bitwise_xor"><code class="xref py py-obj docutils literal notranslate"><span class="pre">bitwise_xor</span></code></a>(lhs, rhs)</p></td>
 <td><p>bitwise XOR with numpy-style broadcasting.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.broadcast_to" title="tvm.relay.broadcast_to"><code class="xref py py-obj docutils literal notranslate"><span class="pre">broadcast_to</span></code></a>(data, shape)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.broadcast_to" title="tvm.relay.broadcast_to"><code class="xref py py-obj docutils literal notranslate"><span class="pre">broadcast_to</span></code></a>(data, shape)</p></td>
 <td><p>Return a scalar value array with the same type, broadcast to the provided shape.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.broadcast_to_like" title="tvm.relay.broadcast_to_like"><code class="xref py py-obj docutils literal notranslate"><span class="pre">broadcast_to_like</span></code></a>(data, broadcast_type)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.broadcast_to_like" title="tvm.relay.broadcast_to_like"><code class="xref py py-obj docutils literal notranslate"><span class="pre">broadcast_to_like</span></code></a>(data, broadcast_type)</p></td>
 <td><p>Return a scalar value array with the same shape and type as the input array.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.build" title="tvm.relay.build"><code class="xref py py-obj docutils literal notranslate"><span class="pre">build</span></code></a>(mod[, target, target_host, params])</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.build" title="tvm.relay.build"><code class="xref py py-obj docutils literal notranslate"><span class="pre">build</span></code></a>(mod[, target, target_host, params])</p></td>
 <td><p>Helper function that builds a Relay function to run on TVM graph runtime.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.build_config" title="tvm.relay.build_config"><code class="xref py py-obj docutils literal notranslate"><span class="pre">build_config</span></code></a>([opt_level, fallback_device, …])</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.build_config" title="tvm.relay.build_config"><code class="xref py py-obj docutils literal notranslate"><span class="pre">build_config</span></code></a>([opt_level, fallback_device, …])</p></td>
 <td><p>Configure the build behavior by setting config variables.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.cast" title="tvm.relay.cast"><code class="xref py py-obj docutils literal notranslate"><span class="pre">cast</span></code></a>(data, dtype)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.cast" title="tvm.relay.cast"><code class="xref py py-obj docutils literal notranslate"><span class="pre">cast</span></code></a>(data, dtype)</p></td>
 <td><p>Cast input tensor to data type.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.cast_like" title="tvm.relay.cast_like"><code class="xref py py-obj docutils literal notranslate"><span class="pre">cast_like</span></code></a>(data, dtype_like)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.cast_like" title="tvm.relay.cast_like"><code class="xref py py-obj docutils literal notranslate"><span class="pre">cast_like</span></code></a>(data, dtype_like)</p></td>
 <td><p>Cast input tensor to data type of another tensor.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.ceil" title="tvm.relay.ceil"><code class="xref py py-obj docutils literal notranslate"><span class="pre">ceil</span></code></a>(data)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.ceil" title="tvm.relay.ceil"><code class="xref py py-obj docutils literal notranslate"><span class="pre">ceil</span></code></a>(data)</p></td>
 <td><p>Compute element-wise ceil of data.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.clip" title="tvm.relay.clip"><code class="xref py py-obj docutils literal notranslate"><span class="pre">clip</span></code></a>(a, a_min, a_max)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.clip" title="tvm.relay.clip"><code class="xref py py-obj docutils literal notranslate"><span class="pre">clip</span></code></a>(a, a_min, a_max)</p></td>
 <td><p>Clip the elements in <cite>a</cite> between <cite>a_min</cite> and <cite>a_max</cite>.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.collapse_sum_like" title="tvm.relay.collapse_sum_like"><code class="xref py py-obj docutils literal notranslate"><span class="pre">collapse_sum_like</span></code></a>(data, collapse_type)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.collapse_sum_like" title="tvm.relay.collapse_sum_like"><code class="xref py py-obj docutils literal notranslate"><span class="pre">collapse_sum_like</span></code></a>(data, collapse_type)</p></td>
 <td><p>Return a scalar value array with the same shape and type as the input array.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.concatenate" title="tvm.relay.concatenate"><code class="xref py py-obj docutils literal notranslate"><span class="pre">concatenate</span></code></a>(data, axis)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.concatenate" title="tvm.relay.concatenate"><code class="xref py py-obj docutils literal notranslate"><span class="pre">concatenate</span></code></a>(data, axis)</p></td>
 <td><p>Concatenate the input tensors along the given axis.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.const" title="tvm.relay.const"><code class="xref py py-obj docutils literal notranslate"><span class="pre">const</span></code></a>(value[, dtype])</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.const" title="tvm.relay.const"><code class="xref py py-obj docutils literal notranslate"><span class="pre">const</span></code></a>(value[, dtype])</p></td>
 <td><p>Create a constant value.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.copy" title="tvm.relay.copy"><code class="xref py py-obj docutils literal notranslate"><span class="pre">copy</span></code></a>(data)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.copy" title="tvm.relay.copy"><code class="xref py py-obj docutils literal notranslate"><span class="pre">copy</span></code></a>(data)</p></td>
 <td><p>Copy a tensor.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.cos" title="tvm.relay.cos"><code class="xref py py-obj docutils literal notranslate"><span class="pre">cos</span></code></a>(data)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.cos" title="tvm.relay.cos"><code class="xref py py-obj docutils literal notranslate"><span class="pre">cos</span></code></a>(data)</p></td>
 <td><p>Compute elementwise cos of data.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.cosh" title="tvm.relay.cosh"><code class="xref py py-obj docutils literal notranslate"><span class="pre">cosh</span></code></a>(data)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.cosh" title="tvm.relay.cosh"><code class="xref py py-obj docutils literal notranslate"><span class="pre">cosh</span></code></a>(data)</p></td>
 <td><p>Compute elementwise cosh of data.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.create_executor" title="tvm.relay.create_executor"><code class="xref py py-obj docutils literal notranslate"><span class="pre">create_executor</span></code></a>([kind, mod, ctx, target])</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.create_executor" title="tvm.relay.create_executor"><code class="xref py py-obj docutils literal notranslate"><span class="pre">create_executor</span></code></a>([kind, mod, ctx, target])</p></td>
 <td><p>Factory function to create an executor.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.device_copy" title="tvm.relay.device_copy"><code class="xref py py-obj docutils literal notranslate"><span class="pre">device_copy</span></code></a>(data, src_dev, dst_dev)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.device_copy" title="tvm.relay.device_copy"><code class="xref py py-obj docutils literal notranslate"><span class="pre">device_copy</span></code></a>(data, src_dev, dst_dev)</p></td>
 <td><p>Copy data from the source device to the destination device.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.divide" title="tvm.relay.divide"><code class="xref py py-obj docutils literal notranslate"><span class="pre">divide</span></code></a>(lhs, rhs)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.divide" title="tvm.relay.divide"><code class="xref py py-obj docutils literal notranslate"><span class="pre">divide</span></code></a>(lhs, rhs)</p></td>
 <td><p>Division with numpy-style broadcasting.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.equal" title="tvm.relay.equal"><code class="xref py py-obj docutils literal notranslate"><span class="pre">equal</span></code></a>(lhs, rhs)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.equal" title="tvm.relay.equal"><code class="xref py py-obj docutils literal notranslate"><span class="pre">equal</span></code></a>(lhs, rhs)</p></td>
 <td><p>Broadcasted elementwise test for (lhs == rhs).</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.erf" title="tvm.relay.erf"><code class="xref py py-obj docutils literal notranslate"><span class="pre">erf</span></code></a>(data)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.erf" title="tvm.relay.erf"><code class="xref py py-obj docutils literal notranslate"><span class="pre">erf</span></code></a>(data)</p></td>
 <td><p>Compute elementwise error function of data.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.exp" title="tvm.relay.exp"><code class="xref py py-obj docutils literal notranslate"><span class="pre">exp</span></code></a>(data)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.exp" title="tvm.relay.exp"><code class="xref py py-obj docutils literal notranslate"><span class="pre">exp</span></code></a>(data)</p></td>
 <td><p>Compute elementwise exp of data.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.expand_dims" title="tvm.relay.expand_dims"><code class="xref py py-obj docutils literal notranslate"><span class="pre">expand_dims</span></code></a>(data, axis[, num_newaxis])</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.expand_dims" title="tvm.relay.expand_dims"><code class="xref py py-obj docutils literal notranslate"><span class="pre">expand_dims</span></code></a>(data, axis[, num_newaxis])</p></td>
 <td><p>Insert <cite>num_newaxis</cite> axises at the position given by <cite>axis</cite>.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.floor" title="tvm.relay.floor"><code class="xref py py-obj docutils literal notranslate"><span class="pre">floor</span></code></a>(data)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.floor" title="tvm.relay.floor"><code class="xref py py-obj docutils literal notranslate"><span class="pre">floor</span></code></a>(data)</p></td>
 <td><p>Compute element-wise floor of data.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.floor_divide" title="tvm.relay.floor_divide"><code class="xref py py-obj docutils literal notranslate"><span class="pre">floor_divide</span></code></a>(lhs, rhs)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.floor_divide" title="tvm.relay.floor_divide"><code class="xref py py-obj docutils literal notranslate"><span class="pre">floor_divide</span></code></a>(lhs, rhs)</p></td>
 <td><p>Floor division with numpy-style broadcasting.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.floor_mod" title="tvm.relay.floor_mod"><code class="xref py py-obj docutils literal notranslate"><span class="pre">floor_mod</span></code></a>(lhs, rhs)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.floor_mod" title="tvm.relay.floor_mod"><code class="xref py py-obj docutils literal notranslate"><span class="pre">floor_mod</span></code></a>(lhs, rhs)</p></td>
 <td><p>Floor mod with numpy-style broadcasting.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.full" title="tvm.relay.full"><code class="xref py py-obj docutils literal notranslate"><span class="pre">full</span></code></a>(fill_value[, shape, dtype])</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.full" title="tvm.relay.full"><code class="xref py py-obj docutils literal notranslate"><span class="pre">full</span></code></a>(fill_value[, shape, dtype])</p></td>
 <td><p>Fill array with scalar value.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.full_like" title="tvm.relay.full_like"><code class="xref py py-obj docutils literal notranslate"><span class="pre">full_like</span></code></a>(data, fill_value)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.full_like" title="tvm.relay.full_like"><code class="xref py py-obj docutils literal notranslate"><span class="pre">full_like</span></code></a>(data, fill_value)</p></td>
 <td><p>Return a scalar value array with the same shape and type as the input array.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.gather_nd" title="tvm.relay.gather_nd"><code class="xref py py-obj docutils literal notranslate"><span class="pre">gather_nd</span></code></a>(data, indices)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.gather_nd" title="tvm.relay.gather_nd"><code class="xref py py-obj docutils literal notranslate"><span class="pre">gather_nd</span></code></a>(data, indices)</p></td>
 <td><p>Gather elements or slices from data and store to a tensor whose shape is defined by indices.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.greater" title="tvm.relay.greater"><code class="xref py py-obj docutils literal notranslate"><span class="pre">greater</span></code></a>(lhs, rhs)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.greater" title="tvm.relay.greater"><code class="xref py py-obj docutils literal notranslate"><span class="pre">greater</span></code></a>(lhs, rhs)</p></td>
 <td><p>Broadcasted elementwise test for (lhs &gt; rhs).</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.greater_equal" title="tvm.relay.greater_equal"><code class="xref py py-obj docutils literal notranslate"><span class="pre">greater_equal</span></code></a>(lhs, rhs)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.greater_equal" title="tvm.relay.greater_equal"><code class="xref py py-obj docutils literal notranslate"><span class="pre">greater_equal</span></code></a>(lhs, rhs)</p></td>
 <td><p>Broadcasted elementwise test for (lhs &gt;= rhs).</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.isfinite" title="tvm.relay.isfinite"><code class="xref py py-obj docutils literal notranslate"><span class="pre">isfinite</span></code></a>(data)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.isfinite" title="tvm.relay.isfinite"><code class="xref py py-obj docutils literal notranslate"><span class="pre">isfinite</span></code></a>(data)</p></td>
 <td><p>Compute element-wise finiteness of data.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.isinf" title="tvm.relay.isinf"><code class="xref py py-obj docutils literal notranslate"><span class="pre">isinf</span></code></a>(data)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.isinf" title="tvm.relay.isinf"><code class="xref py py-obj docutils literal notranslate"><span class="pre">isinf</span></code></a>(data)</p></td>
 <td><p>Compute element-wise infiniteness of data.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.isnan" title="tvm.relay.isnan"><code class="xref py py-obj docutils literal notranslate"><span class="pre">isnan</span></code></a>(data)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.isnan" title="tvm.relay.isnan"><code class="xref py py-obj docutils literal notranslate"><span class="pre">isnan</span></code></a>(data)</p></td>
 <td><p>Check nan in input data element-wise.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.layout_transform" title="tvm.relay.layout_transform"><code class="xref py py-obj docutils literal notranslate"><span class="pre">layout_transform</span></code></a>(data, src_layout, dst_layout)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.layout_transform" title="tvm.relay.layout_transform"><code class="xref py py-obj docutils literal notranslate"><span class="pre">layout_transform</span></code></a>(data, src_layout, dst_layout)</p></td>
 <td><p>Transform the layout of a tensor</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.left_shift" title="tvm.relay.left_shift"><code class="xref py py-obj docutils literal notranslate"><span class="pre">left_shift</span></code></a>(lhs, rhs)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.left_shift" title="tvm.relay.left_shift"><code class="xref py py-obj docutils literal notranslate"><span class="pre">left_shift</span></code></a>(lhs, rhs)</p></td>
 <td><p>Left shift with numpy-style broadcasting.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.less" title="tvm.relay.less"><code class="xref py py-obj docutils literal notranslate"><span class="pre">less</span></code></a>(lhs, rhs)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.less" title="tvm.relay.less"><code class="xref py py-obj docutils literal notranslate"><span class="pre">less</span></code></a>(lhs, rhs)</p></td>
 <td><p>Broadcasted elementwise test for (lhs &lt; rhs).</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.less_equal" title="tvm.relay.less_equal"><code class="xref py py-obj docutils literal notranslate"><span class="pre">less_equal</span></code></a>(lhs, rhs)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.less_equal" title="tvm.relay.less_equal"><code class="xref py py-obj docutils literal notranslate"><span class="pre">less_equal</span></code></a>(lhs, rhs)</p></td>
 <td><p>Broadcasted elementwise test for (lhs &lt;= rhs).</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.load_param_dict" title="tvm.relay.load_param_dict"><code class="xref py py-obj docutils literal notranslate"><span class="pre">load_param_dict</span></code></a>(param_bytes)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.load_param_dict" title="tvm.relay.load_param_dict"><code class="xref py py-obj docutils literal notranslate"><span class="pre">load_param_dict</span></code></a>(param_bytes)</p></td>
 <td><p>Load parameter dictionary to binary bytes.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.log" title="tvm.relay.log"><code class="xref py py-obj docutils literal notranslate"><span class="pre">log</span></code></a>(data)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.log" title="tvm.relay.log"><code class="xref py py-obj docutils literal notranslate"><span class="pre">log</span></code></a>(data)</p></td>
 <td><p>Compute elementwise log of data.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.log10" title="tvm.relay.log10"><code class="xref py py-obj docutils literal notranslate"><span class="pre">log10</span></code></a>(data)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.log10" title="tvm.relay.log10"><code class="xref py py-obj docutils literal notranslate"><span class="pre">log10</span></code></a>(data)</p></td>
 <td><p>Compute elementwise log to the base 10 of data.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.log2" title="tvm.relay.log2"><code class="xref py py-obj docutils literal notranslate"><span class="pre">log2</span></code></a>(data)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.log2" title="tvm.relay.log2"><code class="xref py py-obj docutils literal notranslate"><span class="pre">log2</span></code></a>(data)</p></td>
 <td><p>Compute elementwise log to the base 2 of data.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.logical_and" title="tvm.relay.logical_and"><code class="xref py py-obj docutils literal notranslate"><span class="pre">logical_and</span></code></a>(lhs, rhs)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.logical_and" title="tvm.relay.logical_and"><code class="xref py py-obj docutils literal notranslate"><span class="pre">logical_and</span></code></a>(lhs, rhs)</p></td>
 <td><p>logical AND with numpy-style broadcasting.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.logical_not" title="tvm.relay.logical_not"><code class="xref py py-obj docutils literal notranslate"><span class="pre">logical_not</span></code></a>(data)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.logical_not" title="tvm.relay.logical_not"><code class="xref py py-obj docutils literal notranslate"><span class="pre">logical_not</span></code></a>(data)</p></td>
 <td><p>Compute element-wise logical not of data.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.logical_or" title="tvm.relay.logical_or"><code class="xref py py-obj docutils literal notranslate"><span class="pre">logical_or</span></code></a>(lhs, rhs)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.logical_or" title="tvm.relay.logical_or"><code class="xref py py-obj docutils literal notranslate"><span class="pre">logical_or</span></code></a>(lhs, rhs)</p></td>
 <td><p>logical OR with numpy-style broadcasting.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.logical_xor" title="tvm.relay.logical_xor"><code class="xref py py-obj docutils literal notranslate"><span class="pre">logical_xor</span></code></a>(lhs, rhs)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.logical_xor" title="tvm.relay.logical_xor"><code class="xref py py-obj docutils literal notranslate"><span class="pre">logical_xor</span></code></a>(lhs, rhs)</p></td>
 <td><p>logical XOR with numpy-style broadcasting.</p></td>
 </tr>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.logsumexp" title="tvm.relay.logsumexp"><code class="xref py py-obj docutils literal notranslate"><span class="pre">logsumexp</span></code></a>(data[, axis, keepdims])</p></td>
+<td><p>Compute the log of the sum of exponentials of input elements over given axes.</p></td>
+</tr>
 <tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.max" title="tvm.relay.max"><code class="xref py py-obj docutils literal notranslate"><span class="pre">max</span></code></a>(data[, axis, keepdims, exclude])</p></td>
 <td><p>Computes the max of array elements over given axes.</p></td>
 </tr>
@@ -1497,6 +1515,40 @@ as an argument to an FFI function.</p>
 </dd></dl>
 
 <dl class="py function">
+<dt id="tvm.relay.acos">
+<code class="sig-prename descclassname">tvm.relay.</code><code class="sig-name descname">acos</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">data</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.relay.acos" title="Permalink to this definition">¶</a></dt>
+<dd><p>Compute elementwise acos of data.</p>
+<dl class="field-list simple">
+<dt class="field-odd">Parameters</dt>
+<dd class="field-odd"><p><strong>data</strong> (<em>relay.Expr</em>) – The input data</p>
+</dd>
+<dt class="field-even">Returns</dt>
+<dd class="field-even"><p><strong>result</strong> – The computed result.</p>
+</dd>
+<dt class="field-odd">Return type</dt>
+<dd class="field-odd"><p>relay.Expr</p>
+</dd>
+</dl>
+</dd></dl>
+
+<dl class="py function">
+<dt id="tvm.relay.acosh">
+<code class="sig-prename descclassname">tvm.relay.</code><code class="sig-name descname">acosh</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">data</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.relay.acosh" title="Permalink to this definition">¶</a></dt>
+<dd><p>Compute elementwise acosh of data.</p>
+<dl class="field-list simple">
+<dt class="field-odd">Parameters</dt>
+<dd class="field-odd"><p><strong>data</strong> (<em>relay.Expr</em>) – The input data</p>
+</dd>
+<dt class="field-even">Returns</dt>
+<dd class="field-even"><p><strong>result</strong> – The computed result.</p>
+</dd>
+<dt class="field-odd">Return type</dt>
+<dd class="field-odd"><p>relay.Expr</p>
+</dd>
+</dl>
+</dd></dl>
+
+<dl class="py function">
 <dt id="tvm.relay.add">
 <code class="sig-prename descclassname">tvm.relay.</code><code class="sig-name descname">add</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">lhs</span></em>, <em class="sig-param"><span class="n">rhs</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.relay.add" title="Permalink to this definition">¶</a></dt>
 <dd><p>Addition with numpy-style broadcasting.</p>
@@ -1750,6 +1802,40 @@ non-zero.</p>
 </dd></dl>
 
 <dl class="py function">
+<dt id="tvm.relay.asin">
+<code class="sig-prename descclassname">tvm.relay.</code><code class="sig-name descname">asin</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">data</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.relay.asin" title="Permalink to this definition">¶</a></dt>
+<dd><p>Compute elementwise asin of data.</p>
+<dl class="field-list simple">
+<dt class="field-odd">Parameters</dt>
+<dd class="field-odd"><p><strong>data</strong> (<em>relay.Expr</em>) – The input data</p>
+</dd>
+<dt class="field-even">Returns</dt>
+<dd class="field-even"><p><strong>result</strong> – The computed result.</p>
+</dd>
+<dt class="field-odd">Return type</dt>
+<dd class="field-odd"><p>relay.Expr</p>
+</dd>
+</dl>
+</dd></dl>
+
+<dl class="py function">
+<dt id="tvm.relay.asinh">
+<code class="sig-prename descclassname">tvm.relay.</code><code class="sig-name descname">asinh</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">data</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.relay.asinh" title="Permalink to this definition">¶</a></dt>
+<dd><p>Compute elementwise asinh of data.</p>
+<dl class="field-list simple">
+<dt class="field-odd">Parameters</dt>
+<dd class="field-odd"><p><strong>data</strong> (<em>relay.Expr</em>) – The input data</p>
+</dd>
+<dt class="field-even">Returns</dt>
+<dd class="field-even"><p><strong>result</strong> – The computed result.</p>
+</dd>
+<dt class="field-odd">Return type</dt>
+<dd class="field-odd"><p>relay.Expr</p>
+</dd>
+</dl>
+</dd></dl>
+
+<dl class="py function">
 <dt id="tvm.relay.atan">
 <code class="sig-prename descclassname">tvm.relay.</code><code class="sig-name descname">atan</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">data</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.relay.atan" title="Permalink to this definition">¶</a></dt>
 <dd><p>Compute elementwise atan of data.</p>
@@ -1767,6 +1853,23 @@ non-zero.</p>
 </dd></dl>
 
 <dl class="py function">
+<dt id="tvm.relay.atanh">
+<code class="sig-prename descclassname">tvm.relay.</code><code class="sig-name descname">atanh</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">data</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.relay.atanh" title="Permalink to this definition">¶</a></dt>
+<dd><p>Compute elementwise atanh of data.</p>
+<dl class="field-list simple">
+<dt class="field-odd">Parameters</dt>
+<dd class="field-odd"><p><strong>data</strong> (<em>relay.Expr</em>) – The input data</p>
+</dd>
+<dt class="field-even">Returns</dt>
+<dd class="field-even"><p><strong>result</strong> – The computed result.</p>
+</dd>
+<dt class="field-odd">Return type</dt>
+<dd class="field-odd"><p>relay.Expr</p>
+</dd>
+</dl>
+</dd></dl>
+
+<dl class="py function">
 <dt id="tvm.relay.bitwise_and">
 <code class="sig-prename descclassname">tvm.relay.</code><code class="sig-name descname">bitwise_and</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">lhs</span></em>, <em class="sig-param"><span class="n">rhs</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.relay.bitwise_and" title="Permalink to this definition">¶</a></dt>
 <dd><p>bitwise AND with numpy-style broadcasting.</p>
@@ -2609,6 +2712,35 @@ defined by indices.</p>
 </dd></dl>
 
 <dl class="py function">
+<dt id="tvm.relay.logsumexp">
+<code class="sig-prename descclassname">tvm.relay.</code><code class="sig-name descname">logsumexp</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">data</span></em>, <em class="sig-param"><span class="n">axis</span><span class="o">=</span><span class="default_value">None</span></em>, <em class="sig-param"><span class="n">keepdims</span><span class="o">=</span><span class="default_value">False</span></em><span class="sig-paren">)</span><a class="headerlink" href [...]
+<dd><p>Compute the log of the sum of exponentials of input elements over given axes.</p>
+<blockquote>
+<div><p>This function is more numerically stable than log(sum(exp(input))).
+It avoids overflows caused by taking the exp of large inputs and underflows
+caused by taking the log of small inputs.</p>
+</div></blockquote>
+<dl class="field-list simple">
+<dt class="field-odd">Parameters</dt>
+<dd class="field-odd"><ul class="simple">
+<li><p><strong>data</strong> (<em>relay.Expr</em>) – The input data</p></li>
+<li><p><strong>axis</strong> (<a class="reference external" href="https://docs.python.org/3/library/constants.html#None" title="(in Python v3.8)"><em>None</em></a><em> or </em><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.8)"><em>int</em></a><em> or </em><em>tuple of int</em>) – Axis or axes along which a standard deviation operation is performed.
+The default, axis=None, will compute the log of the sum of exponentials of all elements
+in the input array. If axis is negative it counts from the last to the first axis.</p></li>
+<li><p><strong>keepdims</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#bool" title="(in Python v3.8)"><em>bool</em></a>) – If this is set to True, the axes which are reduced are left in the result as dimensions
+with size one.</p></li>
+</ul>
+</dd>
+<dt class="field-even">Returns</dt>
+<dd class="field-even"><p><strong>result</strong> – The computed result.</p>
+</dd>
+<dt class="field-odd">Return type</dt>
+<dd class="field-odd"><p>relay.Expr</p>
+</dd>
+</dl>
+</dd></dl>
+
+<dl class="py function">
 <dt id="tvm.relay.max">
 <code class="sig-prename descclassname">tvm.relay.</code><code class="sig-name descname">max</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">data</span></em>, <em class="sig-param"><span class="n">axis</span><span class="o">=</span><span class="default_value">None</span></em>, <em class="sig-param"><span class="n">keepdims</span><span class="o">=</span><span class="default_value">False</span></em>, <em class="sig-param"><span class="n">exclude</span><span clas [...]
 <dd><p>Computes the max of array elements over given axes.</p>
@@ -3105,7 +3237,7 @@ to -4 in shape (can contain -1).</p>
 <dt class="field-odd">Parameters</dt>
 <dd class="field-odd"><ul class="simple">
 <li><p><strong>data</strong> (<em>relay.Expr</em>) – The input data to the operator.</p></li>
-<li><p><strong>newshape</strong> (<em>Union</em><em>[</em><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.8)"><em>int</em></a><em>, </em><a class="reference internal" href="#tvm.relay.Tuple" title="tvm.relay.Tuple"><em>Tuple</em></a><em>[</em><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.8)"><em>int</em></a><em>]</em><em>, </em><em>List</em><em>[</em><a class="r [...]
+<li><p><strong>newshape</strong> (<em>Union</em><em>[</em><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.8)"><em>int</em></a><em>, </em><a class="reference internal" href="#tvm.relay.Tuple" title="tvm.relay.Tuple"><em>Tuple</em></a><em>[</em><a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.8)"><em>int</em></a><em>]</em><em>, </em><em>List</em><em>[</em><a class="r [...]
 </ul>
 </dd>
 <dt class="field-even">Returns</dt>
diff --git a/docs/api/python/relay/nn.html b/docs/api/python/relay/nn.html
index 5aef914..78da283 100644
--- a/docs/api/python/relay/nn.html
+++ b/docs/api/python/relay/nn.html
@@ -133,7 +133,7 @@
 <li class="toctree-l2"><a class="reference internal" href="../topi.html">topi</a></li>
 </ul>
 </li>
-<li class="toctree-l1"><a class="reference internal" href="../../../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../../links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../../../dev/index.html">Design and Developer Guide</a></li>
@@ -327,79 +327,88 @@
 <tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.nn.fifo_buffer" title="tvm.relay.nn.fifo_buffer"><code class="xref py py-obj docutils literal notranslate"><span class="pre">fifo_buffer</span></code></a>(data, buffer, axis)</p></td>
 <td><p>FIFO buffer to enable computation reuse in CNNs with sliding indow input</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.nn.get_pad_tuple2d" title="tvm.relay.nn.get_pad_tuple2d"><code class="xref py py-obj docutils literal notranslate"><span class="pre">get_pad_tuple2d</span></code></a>(padding)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.nn.get_pad_tuple1d" title="tvm.relay.nn.get_pad_tuple1d"><code class="xref py py-obj docutils literal notranslate"><span class="pre">get_pad_tuple1d</span></code></a>(padding)</p></td>
+<td><p>Common code to get the 1 dimensional pad option</p></td>
+</tr>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.nn.get_pad_tuple2d" title="tvm.relay.nn.get_pad_tuple2d"><code class="xref py py-obj docutils literal notranslate"><span class="pre">get_pad_tuple2d</span></code></a>(padding)</p></td>
 <td><p>Common code to get the pad option</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.nn.get_pad_tuple3d" title="tvm.relay.nn.get_pad_tuple3d"><code class="xref py py-obj docutils literal notranslate"><span class="pre">get_pad_tuple3d</span></code></a>(padding)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.nn.get_pad_tuple3d" title="tvm.relay.nn.get_pad_tuple3d"><code class="xref py py-obj docutils literal notranslate"><span class="pre">get_pad_tuple3d</span></code></a>(padding)</p></td>
 <td><p>Common code to get the pad option</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.nn.global_avg_pool2d" title="tvm.relay.nn.global_avg_pool2d"><code class="xref py py-obj docutils literal notranslate"><span class="pre">global_avg_pool2d</span></code></a>(data[, layout])</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.nn.global_avg_pool2d" title="tvm.relay.nn.global_avg_pool2d"><code class="xref py py-obj docutils literal notranslate"><span class="pre">global_avg_pool2d</span></code></a>(data[, layout])</p></td>
 <td><p>2D global average pooling operator.</p></td>
 </tr>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.nn.global_avg_pool3d" title="tvm.relay.nn.global_avg_pool3d"><code class="xref py py-obj docutils literal notranslate"><span class="pre">global_avg_pool3d</span></code></a>(data[, layout])</p></td>
+<td><p>3D global average pooling operator.</p></td>
+</tr>
 <tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.nn.global_max_pool2d" title="tvm.relay.nn.global_max_pool2d"><code class="xref py py-obj docutils literal notranslate"><span class="pre">global_max_pool2d</span></code></a>(data[, layout])</p></td>
 <td><p>2D global maximum pooling operator.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.nn.group_norm" title="tvm.relay.nn.group_norm"><code class="xref py py-obj docutils literal notranslate"><span class="pre">group_norm</span></code></a>(data, gamma, beta, num_groups[, …])</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.nn.global_max_pool3d" title="tvm.relay.nn.global_max_pool3d"><code class="xref py py-obj docutils literal notranslate"><span class="pre">global_max_pool3d</span></code></a>(data[, layout])</p></td>
+<td><p>3D global maximum pooling operator.</p></td>
+</tr>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.nn.group_norm" title="tvm.relay.nn.group_norm"><code class="xref py py-obj docutils literal notranslate"><span class="pre">group_norm</span></code></a>(data, gamma, beta, num_groups[, …])</p></td>
 <td><p>Group normalization normalizes over group of channels for each training examples.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.nn.instance_norm" title="tvm.relay.nn.instance_norm"><code class="xref py py-obj docutils literal notranslate"><span class="pre">instance_norm</span></code></a>(data, gamma, beta[, axis, …])</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.nn.instance_norm" title="tvm.relay.nn.instance_norm"><code class="xref py py-obj docutils literal notranslate"><span class="pre">instance_norm</span></code></a>(data, gamma, beta[, axis, …])</p></td>
 <td><p>Instance Normalization (Ulyanov and et al., 2016) Applies instance normalization to the n-dimensional input array.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.nn.l2_normalize" title="tvm.relay.nn.l2_normalize"><code class="xref py py-obj docutils literal notranslate"><span class="pre">l2_normalize</span></code></a>(data, eps[, axis])</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.nn.l2_normalize" title="tvm.relay.nn.l2_normalize"><code class="xref py py-obj docutils literal notranslate"><span class="pre">l2_normalize</span></code></a>(data, eps[, axis])</p></td>
 <td><p>Perform L2 normalization on the input data</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.nn.layer_norm" title="tvm.relay.nn.layer_norm"><code class="xref py py-obj docutils literal notranslate"><span class="pre">layer_norm</span></code></a>(data, gamma, beta[, axis, …])</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.nn.layer_norm" title="tvm.relay.nn.layer_norm"><code class="xref py py-obj docutils literal notranslate"><span class="pre">layer_norm</span></code></a>(data, gamma, beta[, axis, …])</p></td>
 <td><p>Layer normalization (Lei Ba and et al., 2016).</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.nn.leaky_relu" title="tvm.relay.nn.leaky_relu"><code class="xref py py-obj docutils literal notranslate"><span class="pre">leaky_relu</span></code></a>(data, alpha)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.nn.leaky_relu" title="tvm.relay.nn.leaky_relu"><code class="xref py py-obj docutils literal notranslate"><span class="pre">leaky_relu</span></code></a>(data, alpha)</p></td>
 <td><p>This operator takes data as input and does Leaky version of a Rectified Linear Unit.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.nn.log_softmax" title="tvm.relay.nn.log_softmax"><code class="xref py py-obj docutils literal notranslate"><span class="pre">log_softmax</span></code></a>(data[, axis])</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.nn.log_softmax" title="tvm.relay.nn.log_softmax"><code class="xref py py-obj docutils literal notranslate"><span class="pre">log_softmax</span></code></a>(data[, axis])</p></td>
 <td><p>Computes log softmax.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.nn.lrn" title="tvm.relay.nn.lrn"><code class="xref py py-obj docutils literal notranslate"><span class="pre">lrn</span></code></a>(data[, size, axis, bias, alpha, beta])</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.nn.lrn" title="tvm.relay.nn.lrn"><code class="xref py py-obj docutils literal notranslate"><span class="pre">lrn</span></code></a>(data[, size, axis, bias, alpha, beta])</p></td>
 <td><p>This operator takes data as input and does local response normalization.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.nn.max_pool1d" title="tvm.relay.nn.max_pool1d"><code class="xref py py-obj docutils literal notranslate"><span class="pre">max_pool1d</span></code></a>(data[, pool_size, strides, …])</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.nn.max_pool1d" title="tvm.relay.nn.max_pool1d"><code class="xref py py-obj docutils literal notranslate"><span class="pre">max_pool1d</span></code></a>(data[, pool_size, strides, …])</p></td>
 <td><p>1D maximum pooling operator.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.nn.max_pool2d" title="tvm.relay.nn.max_pool2d"><code class="xref py py-obj docutils literal notranslate"><span class="pre">max_pool2d</span></code></a>(data[, pool_size, strides, …])</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.nn.max_pool2d" title="tvm.relay.nn.max_pool2d"><code class="xref py py-obj docutils literal notranslate"><span class="pre">max_pool2d</span></code></a>(data[, pool_size, strides, …])</p></td>
 <td><p>2D maximum pooling operator.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.nn.max_pool2d_grad" title="tvm.relay.nn.max_pool2d_grad"><code class="xref py py-obj docutils literal notranslate"><span class="pre">max_pool2d_grad</span></code></a>(out_grad, data[, pool_size, …])</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.nn.max_pool2d_grad" title="tvm.relay.nn.max_pool2d_grad"><code class="xref py py-obj docutils literal notranslate"><span class="pre">max_pool2d_grad</span></code></a>(out_grad, data[, pool_size, …])</p></td>
 <td><p>Gradient of 2D maximum pooling operator.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.nn.max_pool3d" title="tvm.relay.nn.max_pool3d"><code class="xref py py-obj docutils literal notranslate"><span class="pre">max_pool3d</span></code></a>(data[, pool_size, strides, …])</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.nn.max_pool3d" title="tvm.relay.nn.max_pool3d"><code class="xref py py-obj docutils literal notranslate"><span class="pre">max_pool3d</span></code></a>(data[, pool_size, strides, …])</p></td>
 <td><p>3D maximum pooling operator.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.nn.mirror_pad" title="tvm.relay.nn.mirror_pad"><code class="xref py py-obj docutils literal notranslate"><span class="pre">mirror_pad</span></code></a>(data, pad_width[, mode])</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.nn.mirror_pad" title="tvm.relay.nn.mirror_pad"><code class="xref py py-obj docutils literal notranslate"><span class="pre">mirror_pad</span></code></a>(data, pad_width[, mode])</p></td>
 <td><p>MirrorPadding</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.nn.pad" title="tvm.relay.nn.pad"><code class="xref py py-obj docutils literal notranslate"><span class="pre">pad</span></code></a>(data, pad_width[, pad_value, pad_mode])</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.nn.pad" title="tvm.relay.nn.pad"><code class="xref py py-obj docutils literal notranslate"><span class="pre">pad</span></code></a>(data, pad_width[, pad_value, pad_mode])</p></td>
 <td><p>Padding</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.nn.prelu" title="tvm.relay.nn.prelu"><code class="xref py py-obj docutils literal notranslate"><span class="pre">prelu</span></code></a>(data, alpha[, axis])</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.nn.prelu" title="tvm.relay.nn.prelu"><code class="xref py py-obj docutils literal notranslate"><span class="pre">prelu</span></code></a>(data, alpha[, axis])</p></td>
 <td><p>This operator takes data as input and does Leaky version of a Rectified Linear Unit.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.nn.relu" title="tvm.relay.nn.relu"><code class="xref py py-obj docutils literal notranslate"><span class="pre">relu</span></code></a>(data)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.nn.relu" title="tvm.relay.nn.relu"><code class="xref py py-obj docutils literal notranslate"><span class="pre">relu</span></code></a>(data)</p></td>
 <td><p>Rectified linear unit.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.nn.softmax" title="tvm.relay.nn.softmax"><code class="xref py py-obj docutils literal notranslate"><span class="pre">softmax</span></code></a>(data[, axis])</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.nn.softmax" title="tvm.relay.nn.softmax"><code class="xref py py-obj docutils literal notranslate"><span class="pre">softmax</span></code></a>(data[, axis])</p></td>
 <td><p>Computes softmax.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.nn.space_to_depth" title="tvm.relay.nn.space_to_depth"><code class="xref py py-obj docutils literal notranslate"><span class="pre">space_to_depth</span></code></a>(data, block_size[, layout])</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.nn.space_to_depth" title="tvm.relay.nn.space_to_depth"><code class="xref py py-obj docutils literal notranslate"><span class="pre">space_to_depth</span></code></a>(data, block_size[, layout])</p></td>
 <td><p>Convert spatial blocks into channels.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.nn.sparse_dense" title="tvm.relay.nn.sparse_dense"><code class="xref py py-obj docutils literal notranslate"><span class="pre">sparse_dense</span></code></a>(data, weight)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.nn.sparse_dense" title="tvm.relay.nn.sparse_dense"><code class="xref py py-obj docutils literal notranslate"><span class="pre">sparse_dense</span></code></a>(data, weight)</p></td>
 <td><p>Computes the matrix multiplication of <cite>data</cite> and <cite>weight</cite>, where <cite>data</cite> is a dense matrix and <cite>weight</cite> is a sparse (either BSR or CSR) namedtuple with fields <cite>data</cite>, <cite>indices</cite>, and <cite>indptr</cite>.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.nn.sparse_transpose" title="tvm.relay.nn.sparse_transpose"><code class="xref py py-obj docutils literal notranslate"><span class="pre">sparse_transpose</span></code></a>(x)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.nn.sparse_transpose" title="tvm.relay.nn.sparse_transpose"><code class="xref py py-obj docutils literal notranslate"><span class="pre">sparse_transpose</span></code></a>(x)</p></td>
 <td><p>Computes the fast matrix transpose of x, where x is a sparse tensor in CSR format (represented as a namedtuple with fields <cite>data</cite>, <cite>indices</cite>, and <cite>indptr</cite>).</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.nn.upsampling" title="tvm.relay.nn.upsampling"><code class="xref py py-obj docutils literal notranslate"><span class="pre">upsampling</span></code></a>(data[, scale_h, scale_w, layout, …])</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.nn.upsampling" title="tvm.relay.nn.upsampling"><code class="xref py py-obj docutils literal notranslate"><span class="pre">upsampling</span></code></a>(data[, scale_h, scale_w, layout, …])</p></td>
 <td><p>Upsampling.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.nn.upsampling3d" title="tvm.relay.nn.upsampling3d"><code class="xref py py-obj docutils literal notranslate"><span class="pre">upsampling3d</span></code></a>(data[, scale_d, scale_h, …])</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.nn.upsampling3d" title="tvm.relay.nn.upsampling3d"><code class="xref py py-obj docutils literal notranslate"><span class="pre">upsampling3d</span></code></a>(data[, scale_d, scale_h, …])</p></td>
 <td><p>3D Upsampling.</p></td>
 </tr>
 </tbody>
@@ -1533,6 +1542,23 @@ to keep the expected sum of the input unchanged.</p>
 </dd></dl>
 
 <dl class="py function">
+<dt id="tvm.relay.nn.get_pad_tuple1d">
+<code class="sig-prename descclassname">tvm.relay.nn.</code><code class="sig-name descname">get_pad_tuple1d</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">padding</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.relay.nn.get_pad_tuple1d" title="Permalink to this definition">¶</a></dt>
+<dd><p>Common code to get the 1 dimensional pad option
+:param padding: Padding size
+:type padding: Union[int, Tuple[int, …]]</p>
+<dl class="field-list simple">
+<dt class="field-odd">Returns</dt>
+<dd class="field-odd"><p><ul class="simple">
+<li><p><strong>pad_left</strong> (<em>int</em>) – Padding size on left</p></li>
+<li><p><strong>pad_right</strong> (<em>int</em>) – Padding size on right.</p></li>
+</ul>
+</p>
+</dd>
+</dl>
+</dd></dl>
+
+<dl class="py function">
 <dt id="tvm.relay.nn.get_pad_tuple2d">
 <code class="sig-prename descclassname">tvm.relay.nn.</code><code class="sig-name descname">get_pad_tuple2d</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">padding</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.relay.nn.get_pad_tuple2d" title="Permalink to this definition">¶</a></dt>
 <dd><p>Common code to get the pad option
@@ -1602,6 +1628,35 @@ to produce an output Tensor with the following rule:</p>
 </dd></dl>
 
 <dl class="py function">
+<dt id="tvm.relay.nn.global_avg_pool3d">
+<code class="sig-prename descclassname">tvm.relay.nn.</code><code class="sig-name descname">global_avg_pool3d</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">data</span></em>, <em class="sig-param"><span class="n">layout</span><span class="o">=</span><span class="default_value">'NCDHW'</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.relay.nn.global_avg_pool3d" title="Permalink to this definition">¶</a></dt>
+<dd><p>3D global average pooling operator.</p>
+<p>This operator takes data as input and does 3D average value calculation
+across each window represented by DxWxH.</p>
+<p>In the default case, where the data_layout is <cite>NCDHW</cite>
+a data Tensor with shape <cite>(batch_size, in_channels, depth, height, width)</cite>,
+to produce an output Tensor with the following rule:</p>
+<p>with data of shape (b, c, d, h, w)</p>
+<div class="math notranslate nohighlight">
+\[\mbox{out}(b, c, 1, 1, 1)  = \frac{1}{d * h * w} \sum_{l=0}^{d-1}  \sum_{m=0}^{h-1}
+     \sum_{n=0}^{w-1} \mbox{data}(b, c, l, m, n)\]</div>
+<dl class="field-list simple">
+<dt class="field-odd">Parameters</dt>
+<dd class="field-odd"><ul class="simple">
+<li><p><strong>data</strong> (<em>tvm.relay.Expr</em>) – The input data to the operator.</p></li>
+<li><p><strong>layout</strong> (<a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.8)"><em>str</em></a><em>, </em><em>optional</em>) – Layout of the input.</p></li>
+</ul>
+</dd>
+<dt class="field-even">Returns</dt>
+<dd class="field-even"><p><strong>result</strong> – The computed result.</p>
+</dd>
+<dt class="field-odd">Return type</dt>
+<dd class="field-odd"><p>tvm.relay.Expr</p>
+</dd>
+</dl>
+</dd></dl>
+
+<dl class="py function">
 <dt id="tvm.relay.nn.global_max_pool2d">
 <code class="sig-prename descclassname">tvm.relay.nn.</code><code class="sig-name descname">global_max_pool2d</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">data</span></em>, <em class="sig-param"><span class="n">layout</span><span class="o">=</span><span class="default_value">'NCHW'</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.relay.nn.global_max_pool2d" title="Permalink to this definition">¶</a></dt>
 <dd><p>2D global maximum pooling operator.</p>
@@ -1631,6 +1686,37 @@ to produce an output Tensor with the following rule:</p>
 </dd></dl>
 
 <dl class="py function">
+<dt id="tvm.relay.nn.global_max_pool3d">
+<code class="sig-prename descclassname">tvm.relay.nn.</code><code class="sig-name descname">global_max_pool3d</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">data</span></em>, <em class="sig-param"><span class="n">layout</span><span class="o">=</span><span class="default_value">'NCDHW'</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.relay.nn.global_max_pool3d" title="Permalink to this definition">¶</a></dt>
+<dd><p>3D global maximum pooling operator.</p>
+<p>This operator takes data as input and does 3D max value calculation
+across each window represented by DxWxH.</p>
+<p>In the default case, where the data_layout is <cite>NCDHW</cite>
+a data Tensor with shape <cite>(batch_size, in_channels, depth, height, width)</cite>,
+to produce an output Tensor with the following rule:</p>
+<p>with data of shape (b, c, d, h, w)
+.. math:</p>
+<div class="highlight-default notranslate"><div class="highlight"><pre>\<span class="n">mbox</span><span class="p">{</span><span class="n">out</span><span class="p">}(</span><span class="n">b</span><span class="p">,</span> <span class="n">c</span><span class="p">,</span> <span class="mi">1</span><span class="p">,</span> <span class="mi">1</span><span class="p">,</span> <span class="mi">1</span><span class="p">)</span>  <span class="o">=</span>  \<span class="n">max_</span><span class="p" [...]
+     \<span class="n">max_</span><span class="p">{</span><span class="n">n</span><span class="o">=</span><span class="mi">0</span><span class="p">,</span> \<span class="n">ldots</span><span class="p">,</span> <span class="n">w</span><span class="p">}</span> \<span class="n">mbox</span><span class="p">{</span><span class="n">data</span><span class="p">}(</span><span class="n">b</span><span class="p">,</span> <span class="n">c</span><span class="p">,</span> <span class="n">l</span><span cl [...]
+</pre></div>
+</div>
+<dl class="field-list simple">
+<dt class="field-odd">Parameters</dt>
+<dd class="field-odd"><ul class="simple">
+<li><p><strong>data</strong> (<em>tvm.relay.Expr</em>) – The input data to the operator.</p></li>
+<li><p><strong>layout</strong> (<a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.8)"><em>str</em></a><em>, </em><em>optional</em>) – Layout of the input.</p></li>
+</ul>
+</dd>
+<dt class="field-even">Returns</dt>
+<dd class="field-even"><p><strong>result</strong> – The computed result.</p>
+</dd>
+<dt class="field-odd">Return type</dt>
+<dd class="field-odd"><p>tvm.relay.Expr</p>
+</dd>
+</dl>
+</dd></dl>
+
+<dl class="py function">
 <dt id="tvm.relay.nn.group_norm">
 <code class="sig-prename descclassname">tvm.relay.nn.</code><code class="sig-name descname">group_norm</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">data</span></em>, <em class="sig-param"><span class="n">gamma</span></em>, <em class="sig-param"><span class="n">beta</span></em>, <em class="sig-param"><span class="n">num_groups</span></em>, <em class="sig-param"><span class="n">axis</span><span class="o">=</span><span class="default_value">1</span></em>, <em  [...]
 <dd><p>Group normalization normalizes over group of channels for each training examples.
diff --git a/docs/api/python/relay/testing.html b/docs/api/python/relay/testing.html
index 48a8959..2fda996 100644
--- a/docs/api/python/relay/testing.html
+++ b/docs/api/python/relay/testing.html
@@ -133,7 +133,7 @@
 <li class="toctree-l2"><a class="reference internal" href="../topi.html">topi</a></li>
 </ul>
 </li>
-<li class="toctree-l1"><a class="reference internal" href="../../../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../../links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../../../dev/index.html">Design and Developer Guide</a></li>
diff --git a/docs/api/python/relay/transform.html b/docs/api/python/relay/transform.html
index b8e1fe1..eaa2542 100644
--- a/docs/api/python/relay/transform.html
+++ b/docs/api/python/relay/transform.html
@@ -133,7 +133,7 @@
 <li class="toctree-l2"><a class="reference internal" href="../topi.html">topi</a></li>
 </ul>
 </li>
-<li class="toctree-l1"><a class="reference internal" href="../../../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../../links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../../../dev/index.html">Design and Developer Guide</a></li>
@@ -240,7 +240,7 @@
 <tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.transform.CombineParallelDense" title="tvm.relay.transform.CombineParallelDense"><code class="xref py py-obj docutils literal notranslate"><span class="pre">CombineParallelDense</span></code></a>([min_num_branches])</p></td>
 <td><p>Combine multiple dense operators into one.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.transform.ConvertLayout" title="tvm.relay.transform.ConvertLayout"><code class="xref py py-obj docutils literal notranslate"><span class="pre">ConvertLayout</span></code></a>(desired_layout)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.relay.transform.ConvertLayout" title="tvm.relay.transform.ConvertLayout"><code class="xref py py-obj docutils literal notranslate"><span class="pre">ConvertLayout</span></code></a>(desired_layouts)</p></td>
 <td><p>Given a dest layout, this pass transforms the expr such that most of the ops input data layout is changed to the dest layout.</p></td>
 </tr>
 <tr class="row-odd"><td><p><a class="reference internal" href="#tvm.relay.transform.DeadCodeElimination" title="tvm.relay.transform.DeadCodeElimination"><code class="xref py py-obj docutils literal notranslate"><span class="pre">DeadCodeElimination</span></code></a>([inline_once])</p></td>
@@ -496,7 +496,7 @@ optimization.</p>
 
 <dl class="py function">
 <dt id="tvm.relay.transform.ConvertLayout">
-<code class="sig-prename descclassname">tvm.relay.transform.</code><code class="sig-name descname">ConvertLayout</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">desired_layout</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.relay.transform.ConvertLayout" title="Permalink to this definition">¶</a></dt>
+<code class="sig-prename descclassname">tvm.relay.transform.</code><code class="sig-name descname">ConvertLayout</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">desired_layouts</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.relay.transform.ConvertLayout" title="Permalink to this definition">¶</a></dt>
 <dd><p>Given a dest layout, this pass transforms the expr such that most of the ops input data
 layout is changed to the dest layout. In ideal situation, there are only 2 layout transforms,
 one at the start and one at the end.</p>
@@ -509,7 +509,10 @@ new layouts for conv2d ops for now. Most of the other operators try to adapt to
 layout using the InferCorrectLayout infrastructure.</p>
 <dl class="field-list simple">
 <dt class="field-odd">Parameters</dt>
-<dd class="field-odd"><p><strong>desired_layout</strong> (<a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.8)"><em>str</em></a>) – The desired layout for the transformed expr.</p>
+<dd class="field-odd"><p><strong>desired_layouts</strong> (<em>map of op_name to list of layouts</em>) – Specify a mapping of operator names to a list of layouts to convert to, in the order
+defined by the operator. An example for nn.conv2d could be: {“nn.conv2d”, [“NHWC”, “OHWI]},
+where the first item in the list specifies the data layout and the second specifies the
+kernel layout.</p>
 </dd>
 <dt class="field-even">Returns</dt>
 <dd class="field-even"><p><strong>pass</strong> – The pass.</p>
diff --git a/docs/api/python/relay/vision.html b/docs/api/python/relay/vision.html
index 19903d8..bc00fa1 100644
--- a/docs/api/python/relay/vision.html
+++ b/docs/api/python/relay/vision.html
@@ -133,7 +133,7 @@
 <li class="toctree-l2"><a class="reference internal" href="../topi.html">topi</a></li>
 </ul>
 </li>
-<li class="toctree-l1"><a class="reference internal" href="../../../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../../links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../../../dev/index.html">Design and Developer Guide</a></li>
diff --git a/docs/api/python/rpc.html b/docs/api/python/rpc.html
index f46c4f3..2cb302e 100644
--- a/docs/api/python/rpc.html
+++ b/docs/api/python/rpc.html
@@ -133,7 +133,7 @@
 <li class="toctree-l2"><a class="reference internal" href="topi.html">topi</a></li>
 </ul>
 </li>
-<li class="toctree-l1"><a class="reference internal" href="../../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../../dev/index.html">Design and Developer Guide</a></li>
@@ -228,13 +228,16 @@ upload and run remote RPC server, get the result back to verify correctness.</p>
 <tr class="row-odd"><td><p><a class="reference internal" href="#tvm.rpc.LocalSession" title="tvm.rpc.LocalSession"><code class="xref py py-obj docutils literal notranslate"><span class="pre">LocalSession</span></code></a>()</p></td>
 <td><p>RPCSession interface backed by local environment.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.rpc.RPCSession" title="tvm.rpc.RPCSession"><code class="xref py py-obj docutils literal notranslate"><span class="pre">RPCSession</span></code></a>(sess)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.rpc.PopenSession" title="tvm.rpc.PopenSession"><code class="xref py py-obj docutils literal notranslate"><span class="pre">PopenSession</span></code></a>(binary)</p></td>
+<td><p>RPCSession interface backed by popen.</p></td>
+</tr>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.rpc.RPCSession" title="tvm.rpc.RPCSession"><code class="xref py py-obj docutils literal notranslate"><span class="pre">RPCSession</span></code></a>(sess)</p></td>
 <td><p>RPC Client session module</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.rpc.Server" title="tvm.rpc.Server"><code class="xref py py-obj docutils literal notranslate"><span class="pre">Server</span></code></a>(host[, port, port_end, is_proxy, …])</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.rpc.Server" title="tvm.rpc.Server"><code class="xref py py-obj docutils literal notranslate"><span class="pre">Server</span></code></a>(host[, port, port_end, is_proxy, …])</p></td>
 <td><p>Start RPC server on a separate process.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.rpc.TrackerSession" title="tvm.rpc.TrackerSession"><code class="xref py py-obj docutils literal notranslate"><span class="pre">TrackerSession</span></code></a>(addr)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.rpc.TrackerSession" title="tvm.rpc.TrackerSession"><code class="xref py py-obj docutils literal notranslate"><span class="pre">TrackerSession</span></code></a>(addr)</p></td>
 <td><p>Tracker client session.</p></td>
 </tr>
 </tbody>
@@ -246,12 +249,15 @@ upload and run remote RPC server, get the result back to verify correctness.</p>
 <col style="width: 90%" />
 </colgroup>
 <tbody>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.rpc.connect" title="tvm.rpc.connect"><code class="xref py py-obj docutils literal notranslate"><span class="pre">connect</span></code></a>(url, port[, key, session_timeout])</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.rpc.connect" title="tvm.rpc.connect"><code class="xref py py-obj docutils literal notranslate"><span class="pre">connect</span></code></a>(url, port[, key, session_timeout, …])</p></td>
 <td><p>Connect to RPC Server</p></td>
 </tr>
 <tr class="row-even"><td><p><a class="reference internal" href="#tvm.rpc.connect_tracker" title="tvm.rpc.connect_tracker"><code class="xref py py-obj docutils literal notranslate"><span class="pre">connect_tracker</span></code></a>(url, port)</p></td>
 <td><p>Connect to a RPC tracker</p></td>
 </tr>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.rpc.with_minrpc" title="tvm.rpc.with_minrpc"><code class="xref py py-obj docutils literal notranslate"><span class="pre">with_minrpc</span></code></a>(compile_func[, server, runtime])</p></td>
+<td><p>Attach the compiler function with minrpc related options.</p></td>
+</tr>
 </tbody>
 </table>
 <dl class="py class">
@@ -302,6 +308,69 @@ If is not None, the server will register itself to the tracker.</p></li>
 
 </dd></dl>
 
+<dl class="py function">
+<dt id="tvm.rpc.connect">
+<code class="sig-prename descclassname">tvm.rpc.</code><code class="sig-name descname">connect</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">url</span></em>, <em class="sig-param"><span class="n">port</span></em>, <em class="sig-param"><span class="n">key</span><span class="o">=</span><span class="default_value">''</span></em>, <em class="sig-param"><span class="n">session_timeout</span><span class="o">=</span><span class="default_value">0</span></em>, <em c [...]
+<dd><p>Connect to RPC Server</p>
+<dl class="field-list simple">
+<dt class="field-odd">Parameters</dt>
+<dd class="field-odd"><ul class="simple">
+<li><p><strong>url</strong> (<a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.8)"><em>str</em></a>) – The url of the host</p></li>
+<li><p><strong>port</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.8)"><em>int</em></a>) – The port to connect to</p></li>
+<li><p><strong>key</strong> (<a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.8)"><em>str</em></a><em>, </em><em>optional</em>) – Additional key to match server</p></li>
+<li><p><strong>session_timeout</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#float" title="(in Python v3.8)"><em>float</em></a><em>, </em><em>optional</em>) – The duration of the session, allows server to kill
+the connection when duration is longer than this value.
+When duration is zero, it means the request must always be kept alive.</p></li>
+<li><p><strong>session_constructor_args</strong> (<em>List</em>) – List of additional arguments to passed as the remote session constructor.
+The first element of the list is always a string specifying the name of
+the session constructor, the following args are the positional args to that function.</p></li>
+</ul>
+</dd>
+<dt class="field-even">Returns</dt>
+<dd class="field-even"><p><strong>sess</strong> – The connected session.</p>
+</dd>
+<dt class="field-odd">Return type</dt>
+<dd class="field-odd"><p><a class="reference internal" href="#tvm.rpc.RPCSession" title="tvm.rpc.RPCSession">RPCSession</a></p>
+</dd>
+</dl>
+<p class="rubric">Examples</p>
+<p>Normal usage
+.. code-block:: python</p>
+<blockquote>
+<div><p>client = rpc.connect(server_url, server_port, server_key)</p>
+</div></blockquote>
+<p>Session_constructor can be used to customize the session in the remote
+The following code connects to a remote internal server via a proxy
+by constructing another RPCClientSession on the proxy machine and use that
+as the serving session of the proxy endpoint.</p>
+<div class="highlight-python notranslate"><div class="highlight"><pre><span class="n">client_via_proxy</span> <span class="o">=</span> <span class="n">rpc</span><span class="o">.</span><span class="n">connect</span><span class="p">(</span>
+    <span class="n">proxy_server_url</span><span class="p">,</span> <span class="n">proxy_server_port</span><span class="p">,</span> <span class="n">proxy_server_key</span><span class="p">,</span>
+    <span class="n">session_constructor_args</span><span class="o">=</span><span class="p">[</span>
+        <span class="s2">&quot;rpc.Connect&quot;</span><span class="p">,</span> <span class="n">internal_url</span><span class="p">,</span> <span class="n">internal_port</span><span class="p">,</span> <span class="n">internal_key</span><span class="p">])</span>
+</pre></div>
+</div>
+</dd></dl>
+
+<dl class="py function">
+<dt id="tvm.rpc.connect_tracker">
+<code class="sig-prename descclassname">tvm.rpc.</code><code class="sig-name descname">connect_tracker</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">url</span></em>, <em class="sig-param"><span class="n">port</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.rpc.connect_tracker" title="Permalink to this definition">¶</a></dt>
+<dd><p>Connect to a RPC tracker</p>
+<dl class="field-list simple">
+<dt class="field-odd">Parameters</dt>
+<dd class="field-odd"><ul class="simple">
+<li><p><strong>url</strong> (<a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.8)"><em>str</em></a>) – The url of the host</p></li>
+<li><p><strong>port</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.8)"><em>int</em></a>) – The port to connect to</p></li>
+</ul>
+</dd>
+<dt class="field-even">Returns</dt>
+<dd class="field-even"><p><strong>sess</strong> – The connected tracker session.</p>
+</dd>
+<dt class="field-odd">Return type</dt>
+<dd class="field-odd"><p><a class="reference internal" href="#tvm.rpc.TrackerSession" title="tvm.rpc.TrackerSession">TrackerSession</a></p>
+</dd>
+</dl>
+</dd></dl>
+
 <dl class="py class">
 <dt id="tvm.rpc.RPCSession">
 <em class="property">class </em><code class="sig-prename descclassname">tvm.rpc.</code><code class="sig-name descname">RPCSession</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">sess</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.rpc.RPCSession" title="Permalink to this definition">¶</a></dt>
@@ -347,15 +416,39 @@ If is not None, the server will register itself to the tracker.</p></li>
 <tr class="row-odd"><td><p><a class="reference internal" href="#tvm.rpc.RPCSession.remove" title="tvm.rpc.RPCSession.remove"><code class="xref py py-obj docutils literal notranslate"><span class="pre">remove</span></code></a>(path)</p></td>
 <td><p>Remove file from remote temp folder.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.rpc.RPCSession.upload" title="tvm.rpc.RPCSession.upload"><code class="xref py py-obj docutils literal notranslate"><span class="pre">upload</span></code></a>(data[, target])</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.rpc.RPCSession.system_lib" title="tvm.rpc.RPCSession.system_lib"><code class="xref py py-obj docutils literal notranslate"><span class="pre">system_lib</span></code></a>()</p></td>
+<td><p>Get system-wide library module.</p></td>
+</tr>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.rpc.RPCSession.upload" title="tvm.rpc.RPCSession.upload"><code class="xref py py-obj docutils literal notranslate"><span class="pre">upload</span></code></a>(data[, target])</p></td>
 <td><p>Upload file to remote runtime temp folder</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.rpc.RPCSession.vulkan" title="tvm.rpc.RPCSession.vulkan"><code class="xref py py-obj docutils literal notranslate"><span class="pre">vulkan</span></code></a>([dev_id])</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.rpc.RPCSession.vulkan" title="tvm.rpc.RPCSession.vulkan"><code class="xref py py-obj docutils literal notranslate"><span class="pre">vulkan</span></code></a>([dev_id])</p></td>
 <td><p>Construct Vulkan device.</p></td>
 </tr>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.rpc.RPCSession.webgpu" title="tvm.rpc.RPCSession.webgpu"><code class="xref py py-obj docutils literal notranslate"><span class="pre">webgpu</span></code></a>([dev_id])</p></td>
+<td><p>Construct WebGPU device.</p></td>
+</tr>
 </tbody>
 </table>
 <dl class="py method">
+<dt id="tvm.rpc.RPCSession.system_lib">
+<code class="sig-name descname">system_lib</code><span class="sig-paren">(</span><span class="sig-paren">)</span><a class="headerlink" href="#tvm.rpc.RPCSession.system_lib" title="Permalink to this definition">¶</a></dt>
+<dd><p>Get system-wide library module.</p>
+<dl class="field-list simple">
+<dt class="field-odd">Returns</dt>
+<dd class="field-odd"><p><strong>module</strong> – The system-wide library module.</p>
+</dd>
+<dt class="field-even">Return type</dt>
+<dd class="field-even"><p><a class="reference internal" href="runtime.html#tvm.runtime.Module" title="tvm.runtime.Module">runtime.Module</a></p>
+</dd>
+</dl>
+<div class="admonition seealso">
+<p class="admonition-title">See also</p>
+<p><a class="reference internal" href="runtime.html#tvm.runtime.system_lib" title="tvm.runtime.system_lib"><code class="xref py py-meth docutils literal notranslate"><span class="pre">tvm.runtime.system_lib()</span></code></a></p>
+</div>
+</dd></dl>
+
+<dl class="py method">
 <dt id="tvm.rpc.RPCSession.get_function">
 <code class="sig-name descname">get_function</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">name</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.rpc.RPCSession.get_function" title="Permalink to this definition">¶</a></dt>
 <dd><p>Get function from the session.</p>
@@ -493,6 +586,12 @@ If is not None, the server will register itself to the tracker.</p></li>
 <dd><p>Construct extension device.</p>
 </dd></dl>
 
+<dl class="py method">
+<dt id="tvm.rpc.RPCSession.webgpu">
+<code class="sig-name descname">webgpu</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">dev_id</span><span class="o">=</span><span class="default_value">0</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.rpc.RPCSession.webgpu" title="Permalink to this definition">¶</a></dt>
+<dd><p>Construct WebGPU device.</p>
+</dd></dl>
+
 </dd></dl>
 
 <dl class="py class">
@@ -501,74 +600,19 @@ If is not None, the server will register itself to the tracker.</p></li>
 <dd><p>RPCSession interface backed by local environment.</p>
 <p>This class can be used to implement functions that
 need to be ran both locally and remotely.</p>
-<p><strong>Methods</strong></p>
-<table class="longtable docutils align-default">
-<colgroup>
-<col style="width: 10%" />
-<col style="width: 90%" />
-</colgroup>
-<tbody>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.rpc.LocalSession.download" title="tvm.rpc.LocalSession.download"><code class="xref py py-obj docutils literal notranslate"><span class="pre">download</span></code></a>(path)</p></td>
-<td><p>Download file from remote temp folder.</p></td>
-</tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.rpc.LocalSession.load_module" title="tvm.rpc.LocalSession.load_module"><code class="xref py py-obj docutils literal notranslate"><span class="pre">load_module</span></code></a>(path)</p></td>
-<td><p>Load a remote module, the file need to be uploaded first.</p></td>
-</tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.rpc.LocalSession.upload" title="tvm.rpc.LocalSession.upload"><code class="xref py py-obj docutils literal notranslate"><span class="pre">upload</span></code></a>(data[, target])</p></td>
-<td><p>Upload file to remote runtime temp folder</p></td>
-</tr>
-</tbody>
-</table>
-<dl class="py method">
-<dt id="tvm.rpc.LocalSession.upload">
-<code class="sig-name descname">upload</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">data</span></em>, <em class="sig-param"><span class="n">target</span><span class="o">=</span><span class="default_value">None</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.rpc.LocalSession.upload" title="Permalink to this definition">¶</a></dt>
-<dd><p>Upload file to remote runtime temp folder</p>
-<dl class="field-list simple">
-<dt class="field-odd">Parameters</dt>
-<dd class="field-odd"><ul class="simple">
-<li><p><strong>data</strong> (<a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.8)"><em>str</em></a><em> or </em><a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#bytearray" title="(in Python v3.8)"><em>bytearray</em></a>) – The file name or binary in local to upload.</p></li>
-<li><p><strong>target</strong> (<a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.8)"><em>str</em></a><em>, </em><em>optional</em>) – The path in remote</p></li>
-</ul>
-</dd>
-</dl>
 </dd></dl>
 
-<dl class="py method">
-<dt id="tvm.rpc.LocalSession.download">
-<code class="sig-name descname">download</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">path</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.rpc.LocalSession.download" title="Permalink to this definition">¶</a></dt>
-<dd><p>Download file from remote temp folder.</p>
-<dl class="field-list simple">
-<dt class="field-odd">Parameters</dt>
-<dd class="field-odd"><p><strong>path</strong> (<a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.8)"><em>str</em></a>) – The relative location to remote temp folder.</p>
-</dd>
-<dt class="field-even">Returns</dt>
-<dd class="field-even"><p><strong>blob</strong> – The result blob from the file.</p>
-</dd>
-<dt class="field-odd">Return type</dt>
-<dd class="field-odd"><p><a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#bytearray" title="(in Python v3.8)">bytearray</a></p>
-</dd>
-</dl>
-</dd></dl>
-
-<dl class="py method">
-<dt id="tvm.rpc.LocalSession.load_module">
-<code class="sig-name descname">load_module</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">path</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.rpc.LocalSession.load_module" title="Permalink to this definition">¶</a></dt>
-<dd><p>Load a remote module, the file need to be uploaded first.</p>
+<dl class="py class">
+<dt id="tvm.rpc.PopenSession">
+<em class="property">class </em><code class="sig-prename descclassname">tvm.rpc.</code><code class="sig-name descname">PopenSession</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">binary</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.rpc.PopenSession" title="Permalink to this definition">¶</a></dt>
+<dd><p>RPCSession interface backed by popen.</p>
 <dl class="field-list simple">
 <dt class="field-odd">Parameters</dt>
-<dd class="field-odd"><p><strong>path</strong> (<a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.8)"><em>str</em></a>) – The relative location to remote temp folder.</p>
-</dd>
-<dt class="field-even">Returns</dt>
-<dd class="field-even"><p><strong>m</strong> – The remote module containing remote function.</p>
-</dd>
-<dt class="field-odd">Return type</dt>
-<dd class="field-odd"><p><a class="reference internal" href="runtime.html#tvm.runtime.Module" title="tvm.runtime.Module">Module</a></p>
+<dd class="field-odd"><p><strong>binary</strong> (<em>List</em><em>[</em><em>Union</em><em>[</em><a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.8)"><em>str</em></a><em>, </em><a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#bytes" title="(in Python v3.8)"><em>bytes</em></a><em>]</em><em>]</em>) – The binary to be executed.</p>
 </dd>
 </dl>
 </dd></dl>
 
-</dd></dl>
-
 <dl class="py class">
 <dt id="tvm.rpc.TrackerSession">
 <em class="property">class </em><code class="sig-prename descclassname">tvm.rpc.</code><code class="sig-name descname">TrackerSession</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">addr</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.rpc.TrackerSession" title="Permalink to this definition">¶</a></dt>
@@ -662,45 +706,22 @@ When duration is zero, it means the request must always be kept alive.</p></li>
 </dd></dl>
 
 <dl class="py function">
-<dt id="tvm.rpc.connect">
-<code class="sig-prename descclassname">tvm.rpc.</code><code class="sig-name descname">connect</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">url</span></em>, <em class="sig-param"><span class="n">port</span></em>, <em class="sig-param"><span class="n">key</span><span class="o">=</span><span class="default_value">''</span></em>, <em class="sig-param"><span class="n">session_timeout</span><span class="o">=</span><span class="default_value">0</span></em><span c [...]
-<dd><p>Connect to RPC Server</p>
+<dt id="tvm.rpc.with_minrpc">
+<code class="sig-prename descclassname">tvm.rpc.</code><code class="sig-name descname">with_minrpc</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">compile_func</span></em>, <em class="sig-param"><span class="n">server</span><span class="o">=</span><span class="default_value">'posix_popen_server'</span></em>, <em class="sig-param"><span class="n">runtime</span><span class="o">=</span><span class="default_value">'libtvm'</span></em><span class="sig-paren">)</spa [...]
+<dd><p>Attach the compiler function with minrpc related options.</p>
 <dl class="field-list simple">
 <dt class="field-odd">Parameters</dt>
 <dd class="field-odd"><ul class="simple">
-<li><p><strong>url</strong> (<a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.8)"><em>str</em></a>) – The url of the host</p></li>
-<li><p><strong>port</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.8)"><em>int</em></a>) – The port to connect to</p></li>
-<li><p><strong>key</strong> (<a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.8)"><em>str</em></a><em>, </em><em>optional</em>) – Additional key to match server</p></li>
-<li><p><strong>session_timeout</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#float" title="(in Python v3.8)"><em>float</em></a><em>, </em><em>optional</em>) – The duration of the session, allows server to kill
-the connection when duration is longer than this value.
-When duration is zero, it means the request must always be kept alive.</p></li>
+<li><p><strong>compile_func</strong> (<em>Union</em><em>[</em><a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.8)"><em>str</em></a><em>, </em><em>Callable</em><em>[</em><em>[</em><a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.8)"><em>str</em></a><em>, </em><a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.8 [...]
+<li><p><strong>server</strong> (<a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.8)"><em>str</em></a>) – The server type.</p></li>
+<li><p><strong>runtime</strong> (<a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.8)"><em>str</em></a>) – The runtime library.</p></li>
 </ul>
 </dd>
 <dt class="field-even">Returns</dt>
-<dd class="field-even"><p><strong>sess</strong> – The connected session.</p>
+<dd class="field-even"><p><strong>fcompile</strong> – The return compilation.</p>
 </dd>
 <dt class="field-odd">Return type</dt>
-<dd class="field-odd"><p><a class="reference internal" href="#tvm.rpc.RPCSession" title="tvm.rpc.RPCSession">RPCSession</a></p>
-</dd>
-</dl>
-</dd></dl>
-
-<dl class="py function">
-<dt id="tvm.rpc.connect_tracker">
-<code class="sig-prename descclassname">tvm.rpc.</code><code class="sig-name descname">connect_tracker</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">url</span></em>, <em class="sig-param"><span class="n">port</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.rpc.connect_tracker" title="Permalink to this definition">¶</a></dt>
-<dd><p>Connect to a RPC tracker</p>
-<dl class="field-list simple">
-<dt class="field-odd">Parameters</dt>
-<dd class="field-odd"><ul class="simple">
-<li><p><strong>url</strong> (<a class="reference external" href="https://docs.python.org/3/library/stdtypes.html#str" title="(in Python v3.8)"><em>str</em></a>) – The url of the host</p></li>
-<li><p><strong>port</strong> (<a class="reference external" href="https://docs.python.org/3/library/functions.html#int" title="(in Python v3.8)"><em>int</em></a>) – The port to connect to</p></li>
-</ul>
-</dd>
-<dt class="field-even">Returns</dt>
-<dd class="field-even"><p><strong>sess</strong> – The connected tracker session.</p>
-</dd>
-<dt class="field-odd">Return type</dt>
-<dd class="field-odd"><p><a class="reference internal" href="#tvm.rpc.TrackerSession" title="tvm.rpc.TrackerSession">TrackerSession</a></p>
+<dd class="field-odd"><p>function</p>
 </dd>
 </dl>
 </dd></dl>
diff --git a/docs/api/python/runtime.html b/docs/api/python/runtime.html
index d4a1004..b31e653 100644
--- a/docs/api/python/runtime.html
+++ b/docs/api/python/runtime.html
@@ -133,7 +133,7 @@
 <li class="toctree-l2"><a class="reference internal" href="topi.html">topi</a></li>
 </ul>
 </li>
-<li class="toctree-l1"><a class="reference internal" href="../../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../../dev/index.html">Design and Developer Guide</a></li>
@@ -539,7 +539,7 @@ TVM backend also registers and exposes its API as Functions.</p>
 <col style="width: 90%" />
 </colgroup>
 <tbody>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.runtime.Module.export_library" title="tvm.runtime.Module.export_library"><code class="xref py py-obj docutils literal notranslate"><span class="pre">export_library</span></code></a>(file_name[, fcompile])</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.runtime.Module.export_library" title="tvm.runtime.Module.export_library"><code class="xref py py-obj docutils literal notranslate"><span class="pre">export_library</span></code></a>(file_name[, fcompile, addons])</p></td>
 <td><p>Export the module and its imported device code one library.</p></td>
 </tr>
 <tr class="row-even"><td><p><a class="reference internal" href="#tvm.runtime.Module.get_function" title="tvm.runtime.Module.get_function"><code class="xref py py-obj docutils literal notranslate"><span class="pre">get_function</span></code></a>(name[, query_imports])</p></td>
@@ -707,7 +707,7 @@ The ProfileResult reports <cite>repeat</cite> time costs in seconds.</p>
 
 <dl class="py method">
 <dt id="tvm.runtime.Module.export_library">
-<code class="sig-name descname">export_library</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">file_name</span></em>, <em class="sig-param"><span class="n">fcompile</span><span class="o">=</span><span class="default_value">None</span></em>, <em class="sig-param"><span class="o">**</span><span class="n">kwargs</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.runtime.Module.export_library" title="Permalink to this definition">¶</a></dt>
+<code class="sig-name descname">export_library</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">file_name</span></em>, <em class="sig-param"><span class="n">fcompile</span><span class="o">=</span><span class="default_value">None</span></em>, <em class="sig-param"><span class="n">addons</span><span class="o">=</span><span class="default_value">None</span></em>, <em class="sig-param"><span class="o">**</span><span class="n">kwargs</span></em><span class="sig-pare [...]
 <dd><p>Export the module and its imported device code one library.</p>
 <p>This function only works on host llvm modules.
 It will pack all the imported modules</p>
diff --git a/docs/api/python/target.html b/docs/api/python/target.html
index 74cfabd..2ecfd5e 100644
--- a/docs/api/python/target.html
+++ b/docs/api/python/target.html
@@ -133,7 +133,7 @@
 <li class="toctree-l2"><a class="reference internal" href="topi.html">topi</a></li>
 </ul>
 </li>
-<li class="toctree-l1"><a class="reference internal" href="../../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../../dev/index.html">Design and Developer Guide</a></li>
diff --git a/docs/api/python/te.html b/docs/api/python/te.html
index e464262..b6c3093 100644
--- a/docs/api/python/te.html
+++ b/docs/api/python/te.html
@@ -133,7 +133,7 @@
 <li class="toctree-l2"><a class="reference internal" href="topi.html">topi</a></li>
 </ul>
 </li>
-<li class="toctree-l1"><a class="reference internal" href="../../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../../dev/index.html">Design and Developer Guide</a></li>
@@ -264,169 +264,184 @@
 <tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.abs" title="tvm.te.abs"><code class="xref py py-obj docutils literal notranslate"><span class="pre">abs</span></code></a>(x)</p></td>
 <td><p>Get absolute value of the input element-wise.</p></td>
 </tr>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.acos" title="tvm.te.acos"><code class="xref py py-obj docutils literal notranslate"><span class="pre">acos</span></code></a>(x)</p></td>
+<td><p>Take acos of input x.</p></td>
+</tr>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.acosh" title="tvm.te.acosh"><code class="xref py py-obj docutils literal notranslate"><span class="pre">acosh</span></code></a>(x)</p></td>
+<td><p>Take acos of input x.</p></td>
+</tr>
 <tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.all" title="tvm.te.all"><code class="xref py py-obj docutils literal notranslate"><span class="pre">all</span></code></a>(*args)</p></td>
 <td><p>Create a new experssion of the intersection of all conditions in the</p></td>
 </tr>
 <tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.any" title="tvm.te.any"><code class="xref py py-obj docutils literal notranslate"><span class="pre">any</span></code></a>(*args)</p></td>
 <td><p>Create a new experssion of the union of all conditions in the arguments</p></td>
 </tr>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.asin" title="tvm.te.asin"><code class="xref py py-obj docutils literal notranslate"><span class="pre">asin</span></code></a>(x)</p></td>
+<td><p>Take asin of input x.</p></td>
+</tr>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.asinh" title="tvm.te.asinh"><code class="xref py py-obj docutils literal notranslate"><span class="pre">asinh</span></code></a>(x)</p></td>
+<td><p>Take asinh of input x.</p></td>
+</tr>
 <tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.atan" title="tvm.te.atan"><code class="xref py py-obj docutils literal notranslate"><span class="pre">atan</span></code></a>(x)</p></td>
 <td><p>Take atan of input x.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.ceil" title="tvm.te.ceil"><code class="xref py py-obj docutils literal notranslate"><span class="pre">ceil</span></code></a>(x)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.atanh" title="tvm.te.atanh"><code class="xref py py-obj docutils literal notranslate"><span class="pre">atanh</span></code></a>(x)</p></td>
+<td><p>Take atanh of input x.</p></td>
+</tr>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.ceil" title="tvm.te.ceil"><code class="xref py py-obj docutils literal notranslate"><span class="pre">ceil</span></code></a>(x)</p></td>
 <td><p>Take ceil of float input x.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.comm_reducer" title="tvm.te.comm_reducer"><code class="xref py py-obj docutils literal notranslate"><span class="pre">comm_reducer</span></code></a>(fcombine, fidentity[, name])</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.comm_reducer" title="tvm.te.comm_reducer"><code class="xref py py-obj docutils literal notranslate"><span class="pre">comm_reducer</span></code></a>(fcombine, fidentity[, name])</p></td>
 <td><p>Create a commutative reducer for reduction.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.compute" title="tvm.te.compute"><code class="xref py py-obj docutils literal notranslate"><span class="pre">compute</span></code></a>(shape, fcompute[, name, tag, attrs])</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.compute" title="tvm.te.compute"><code class="xref py py-obj docutils literal notranslate"><span class="pre">compute</span></code></a>(shape, fcompute[, name, tag, attrs])</p></td>
 <td><p>Construct a new tensor by computing over the shape domain.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.cos" title="tvm.te.cos"><code class="xref py py-obj docutils literal notranslate"><span class="pre">cos</span></code></a>(x)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.cos" title="tvm.te.cos"><code class="xref py py-obj docutils literal notranslate"><span class="pre">cos</span></code></a>(x)</p></td>
 <td><p>Take cos of input x.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.cosh" title="tvm.te.cosh"><code class="xref py py-obj docutils literal notranslate"><span class="pre">cosh</span></code></a>(x)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.cosh" title="tvm.te.cosh"><code class="xref py py-obj docutils literal notranslate"><span class="pre">cosh</span></code></a>(x)</p></td>
 <td><p>Take cosh of input x.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.create_schedule" title="tvm.te.create_schedule"><code class="xref py py-obj docutils literal notranslate"><span class="pre">create_schedule</span></code></a>(ops)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.create_schedule" title="tvm.te.create_schedule"><code class="xref py py-obj docutils literal notranslate"><span class="pre">create_schedule</span></code></a>(ops)</p></td>
 <td><p>Create a schedule for list of ops</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.decl_tensor_intrin" title="tvm.te.decl_tensor_intrin"><code class="xref py py-obj docutils literal notranslate"><span class="pre">decl_tensor_intrin</span></code></a>(op, fcompute[, name, …])</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.decl_tensor_intrin" title="tvm.te.decl_tensor_intrin"><code class="xref py py-obj docutils literal notranslate"><span class="pre">decl_tensor_intrin</span></code></a>(op, fcompute[, name, …])</p></td>
 <td><p>Declare a tensor intrinsic function.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.div" title="tvm.te.div"><code class="xref py py-obj docutils literal notranslate"><span class="pre">div</span></code></a>(a, b)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.div" title="tvm.te.div"><code class="xref py py-obj docutils literal notranslate"><span class="pre">div</span></code></a>(a, b)</p></td>
 <td><p>Compute a / b as in C/C++ semantics.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.erf" title="tvm.te.erf"><code class="xref py py-obj docutils literal notranslate"><span class="pre">erf</span></code></a>(x)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.erf" title="tvm.te.erf"><code class="xref py py-obj docutils literal notranslate"><span class="pre">erf</span></code></a>(x)</p></td>
 <td><p>Take gauss error function of the input x.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.exp" title="tvm.te.exp"><code class="xref py py-obj docutils literal notranslate"><span class="pre">exp</span></code></a>(x)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.exp" title="tvm.te.exp"><code class="xref py py-obj docutils literal notranslate"><span class="pre">exp</span></code></a>(x)</p></td>
 <td><p>Take exponetial of input x.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.extern" title="tvm.te.extern"><code class="xref py py-obj docutils literal notranslate"><span class="pre">extern</span></code></a>(shape, inputs, fcompute[, name, …])</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.extern" title="tvm.te.extern"><code class="xref py py-obj docutils literal notranslate"><span class="pre">extern</span></code></a>(shape, inputs, fcompute[, name, …])</p></td>
 <td><p>Compute several tensor via extern function.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.floor" title="tvm.te.floor"><code class="xref py py-obj docutils literal notranslate"><span class="pre">floor</span></code></a>(x)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.floor" title="tvm.te.floor"><code class="xref py py-obj docutils literal notranslate"><span class="pre">floor</span></code></a>(x)</p></td>
 <td><p>Take floor of float input x.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.floordiv" title="tvm.te.floordiv"><code class="xref py py-obj docutils literal notranslate"><span class="pre">floordiv</span></code></a>(a, b)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.floordiv" title="tvm.te.floordiv"><code class="xref py py-obj docutils literal notranslate"><span class="pre">floordiv</span></code></a>(a, b)</p></td>
 <td><p>Compute the floordiv of two expressions.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.floormod" title="tvm.te.floormod"><code class="xref py py-obj docutils literal notranslate"><span class="pre">floormod</span></code></a>(a, b)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.floormod" title="tvm.te.floormod"><code class="xref py py-obj docutils literal notranslate"><span class="pre">floormod</span></code></a>(a, b)</p></td>
 <td><p>Compute the floormod of two expressions.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.fmod" title="tvm.te.fmod"><code class="xref py py-obj docutils literal notranslate"><span class="pre">fmod</span></code></a>(x, y)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.fmod" title="tvm.te.fmod"><code class="xref py py-obj docutils literal notranslate"><span class="pre">fmod</span></code></a>(x, y)</p></td>
 <td><p>Return the remainder of x divided by y with the same sign as x.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.gradient" title="tvm.te.gradient"><code class="xref py py-obj docutils literal notranslate"><span class="pre">gradient</span></code></a>(output, inputs[, head])</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.gradient" title="tvm.te.gradient"><code class="xref py py-obj docutils literal notranslate"><span class="pre">gradient</span></code></a>(output, inputs[, head])</p></td>
 <td><p>Perform reverse-mode automatic differentiation.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.if_then_else" title="tvm.te.if_then_else"><code class="xref py py-obj docutils literal notranslate"><span class="pre">if_then_else</span></code></a>(cond, t, f)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.if_then_else" title="tvm.te.if_then_else"><code class="xref py py-obj docutils literal notranslate"><span class="pre">if_then_else</span></code></a>(cond, t, f)</p></td>
 <td><p>Conditional selection expression.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.indexdiv" title="tvm.te.indexdiv"><code class="xref py py-obj docutils literal notranslate"><span class="pre">indexdiv</span></code></a>(a, b)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.indexdiv" title="tvm.te.indexdiv"><code class="xref py py-obj docutils literal notranslate"><span class="pre">indexdiv</span></code></a>(a, b)</p></td>
 <td><p>Compute floor(a / b) where a and b are non-negative.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.indexmod" title="tvm.te.indexmod"><code class="xref py py-obj docutils literal notranslate"><span class="pre">indexmod</span></code></a>(a, b)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.indexmod" title="tvm.te.indexmod"><code class="xref py py-obj docutils literal notranslate"><span class="pre">indexmod</span></code></a>(a, b)</p></td>
 <td><p>Compute the remainder of indexdiv.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.isfinite" title="tvm.te.isfinite"><code class="xref py py-obj docutils literal notranslate"><span class="pre">isfinite</span></code></a>(x)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.isfinite" title="tvm.te.isfinite"><code class="xref py py-obj docutils literal notranslate"><span class="pre">isfinite</span></code></a>(x)</p></td>
 <td><p>Check if input value is finite.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.isinf" title="tvm.te.isinf"><code class="xref py py-obj docutils literal notranslate"><span class="pre">isinf</span></code></a>(x)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.isinf" title="tvm.te.isinf"><code class="xref py py-obj docutils literal notranslate"><span class="pre">isinf</span></code></a>(x)</p></td>
 <td><p>Check if input value is infinite.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.isnan" title="tvm.te.isnan"><code class="xref py py-obj docutils literal notranslate"><span class="pre">isnan</span></code></a>(x)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.isnan" title="tvm.te.isnan"><code class="xref py py-obj docutils literal notranslate"><span class="pre">isnan</span></code></a>(x)</p></td>
 <td><p>Check if input value is Nan.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.log" title="tvm.te.log"><code class="xref py py-obj docutils literal notranslate"><span class="pre">log</span></code></a>(x)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.log" title="tvm.te.log"><code class="xref py py-obj docutils literal notranslate"><span class="pre">log</span></code></a>(x)</p></td>
 <td><p>Take log of input x.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.log10" title="tvm.te.log10"><code class="xref py py-obj docutils literal notranslate"><span class="pre">log10</span></code></a>(x)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.log10" title="tvm.te.log10"><code class="xref py py-obj docutils literal notranslate"><span class="pre">log10</span></code></a>(x)</p></td>
 <td><p>Take log10 of input x.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.log2" title="tvm.te.log2"><code class="xref py py-obj docutils literal notranslate"><span class="pre">log2</span></code></a>(x)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.log2" title="tvm.te.log2"><code class="xref py py-obj docutils literal notranslate"><span class="pre">log2</span></code></a>(x)</p></td>
 <td><p>Take log2 of input x.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.max" title="tvm.te.max"><code class="xref py py-obj docutils literal notranslate"><span class="pre">max</span></code></a>(expr, axis[, where])</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.max" title="tvm.te.max"><code class="xref py py-obj docutils literal notranslate"><span class="pre">max</span></code></a>(expr, axis[, where])</p></td>
 <td><p>Create a max expression over axis.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.max_value" title="tvm.te.max_value"><code class="xref py py-obj docutils literal notranslate"><span class="pre">max_value</span></code></a>(dtype)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.max_value" title="tvm.te.max_value"><code class="xref py py-obj docutils literal notranslate"><span class="pre">max_value</span></code></a>(dtype)</p></td>
 <td><p>maximum value of dtype</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.min" title="tvm.te.min"><code class="xref py py-obj docutils literal notranslate"><span class="pre">min</span></code></a>(expr, axis[, where])</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.min" title="tvm.te.min"><code class="xref py py-obj docutils literal notranslate"><span class="pre">min</span></code></a>(expr, axis[, where])</p></td>
 <td><p>Create a min expression over axis.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.min_value" title="tvm.te.min_value"><code class="xref py py-obj docutils literal notranslate"><span class="pre">min_value</span></code></a>(dtype)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.min_value" title="tvm.te.min_value"><code class="xref py py-obj docutils literal notranslate"><span class="pre">min_value</span></code></a>(dtype)</p></td>
 <td><p>minimum value of dtype</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.nearbyint" title="tvm.te.nearbyint"><code class="xref py py-obj docutils literal notranslate"><span class="pre">nearbyint</span></code></a>(x)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.nearbyint" title="tvm.te.nearbyint"><code class="xref py py-obj docutils literal notranslate"><span class="pre">nearbyint</span></code></a>(x)</p></td>
 <td><p>Round elements of the array to the nearest integer.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.placeholder" title="tvm.te.placeholder"><code class="xref py py-obj docutils literal notranslate"><span class="pre">placeholder</span></code></a>(shape[, dtype, name])</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.placeholder" title="tvm.te.placeholder"><code class="xref py py-obj docutils literal notranslate"><span class="pre">placeholder</span></code></a>(shape[, dtype, name])</p></td>
 <td><p>Construct an empty tensor object.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.popcount" title="tvm.te.popcount"><code class="xref py py-obj docutils literal notranslate"><span class="pre">popcount</span></code></a>(x)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.popcount" title="tvm.te.popcount"><code class="xref py py-obj docutils literal notranslate"><span class="pre">popcount</span></code></a>(x)</p></td>
 <td><p>Count the number of set bits in input x.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.power" title="tvm.te.power"><code class="xref py py-obj docutils literal notranslate"><span class="pre">power</span></code></a>(x, y)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.power" title="tvm.te.power"><code class="xref py py-obj docutils literal notranslate"><span class="pre">power</span></code></a>(x, y)</p></td>
 <td><p>x power y</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.reduce_axis" title="tvm.te.reduce_axis"><code class="xref py py-obj docutils literal notranslate"><span class="pre">reduce_axis</span></code></a>(dom[, name])</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.reduce_axis" title="tvm.te.reduce_axis"><code class="xref py py-obj docutils literal notranslate"><span class="pre">reduce_axis</span></code></a>(dom[, name])</p></td>
 <td><p>Create a new IterVar for reduction.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.round" title="tvm.te.round"><code class="xref py py-obj docutils literal notranslate"><span class="pre">round</span></code></a>(x)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.round" title="tvm.te.round"><code class="xref py py-obj docutils literal notranslate"><span class="pre">round</span></code></a>(x)</p></td>
 <td><p>Round elements of the array to the nearest integer.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.rsqrt" title="tvm.te.rsqrt"><code class="xref py py-obj docutils literal notranslate"><span class="pre">rsqrt</span></code></a>(x)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.rsqrt" title="tvm.te.rsqrt"><code class="xref py py-obj docutils literal notranslate"><span class="pre">rsqrt</span></code></a>(x)</p></td>
 <td><p>Take reciprocal of square root of input x.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.scan" title="tvm.te.scan"><code class="xref py py-obj docutils literal notranslate"><span class="pre">scan</span></code></a>(init, update, state_placeholder[, …])</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.scan" title="tvm.te.scan"><code class="xref py py-obj docutils literal notranslate"><span class="pre">scan</span></code></a>(init, update, state_placeholder[, …])</p></td>
 <td><p>Construct new tensors by scanning over axis.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.sigmoid" title="tvm.te.sigmoid"><code class="xref py py-obj docutils literal notranslate"><span class="pre">sigmoid</span></code></a>(x)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.sigmoid" title="tvm.te.sigmoid"><code class="xref py py-obj docutils literal notranslate"><span class="pre">sigmoid</span></code></a>(x)</p></td>
 <td><p>Quick function to get sigmoid</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.sin" title="tvm.te.sin"><code class="xref py py-obj docutils literal notranslate"><span class="pre">sin</span></code></a>(x)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.sin" title="tvm.te.sin"><code class="xref py py-obj docutils literal notranslate"><span class="pre">sin</span></code></a>(x)</p></td>
 <td><p>Take sin of input x.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.sinh" title="tvm.te.sinh"><code class="xref py py-obj docutils literal notranslate"><span class="pre">sinh</span></code></a>(x)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.sinh" title="tvm.te.sinh"><code class="xref py py-obj docutils literal notranslate"><span class="pre">sinh</span></code></a>(x)</p></td>
 <td><p>Take sinh of input x.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.size_var" title="tvm.te.size_var"><code class="xref py py-obj docutils literal notranslate"><span class="pre">size_var</span></code></a>([name, dtype])</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.size_var" title="tvm.te.size_var"><code class="xref py py-obj docutils literal notranslate"><span class="pre">size_var</span></code></a>([name, dtype])</p></td>
 <td><p>Create a new variable represents a tensor shape size, which is non-negative.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.sqrt" title="tvm.te.sqrt"><code class="xref py py-obj docutils literal notranslate"><span class="pre">sqrt</span></code></a>(x)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.sqrt" title="tvm.te.sqrt"><code class="xref py py-obj docutils literal notranslate"><span class="pre">sqrt</span></code></a>(x)</p></td>
 <td><p>Take square root of input x.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.sum" title="tvm.te.sum"><code class="xref py py-obj docutils literal notranslate"><span class="pre">sum</span></code></a>(expr, axis[, where])</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.sum" title="tvm.te.sum"><code class="xref py py-obj docutils literal notranslate"><span class="pre">sum</span></code></a>(expr, axis[, where])</p></td>
 <td><p>Create a sum expression over axis.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.tag_scope" title="tvm.te.tag_scope"><code class="xref py py-obj docutils literal notranslate"><span class="pre">tag_scope</span></code></a>(tag)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.tag_scope" title="tvm.te.tag_scope"><code class="xref py py-obj docutils literal notranslate"><span class="pre">tag_scope</span></code></a>(tag)</p></td>
 <td><p>The operator tag scope.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.tan" title="tvm.te.tan"><code class="xref py py-obj docutils literal notranslate"><span class="pre">tan</span></code></a>(x)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.tan" title="tvm.te.tan"><code class="xref py py-obj docutils literal notranslate"><span class="pre">tan</span></code></a>(x)</p></td>
 <td><p>Take tan of input x.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.tanh" title="tvm.te.tanh"><code class="xref py py-obj docutils literal notranslate"><span class="pre">tanh</span></code></a>(x)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.tanh" title="tvm.te.tanh"><code class="xref py py-obj docutils literal notranslate"><span class="pre">tanh</span></code></a>(x)</p></td>
 <td><p>Take hyperbolic tanh of input x.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.thread_axis" title="tvm.te.thread_axis"><code class="xref py py-obj docutils literal notranslate"><span class="pre">thread_axis</span></code></a>([dom, tag, name])</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.thread_axis" title="tvm.te.thread_axis"><code class="xref py py-obj docutils literal notranslate"><span class="pre">thread_axis</span></code></a>([dom, tag, name])</p></td>
 <td><p>Create a new IterVar to represent thread index.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.trace" title="tvm.te.trace"><code class="xref py py-obj docutils literal notranslate"><span class="pre">trace</span></code></a>(args[, trace_action])</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.trace" title="tvm.te.trace"><code class="xref py py-obj docutils literal notranslate"><span class="pre">trace</span></code></a>(args[, trace_action])</p></td>
 <td><p>Trace tensor data at the runtime.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.trunc" title="tvm.te.trunc"><code class="xref py py-obj docutils literal notranslate"><span class="pre">trunc</span></code></a>(x)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.trunc" title="tvm.te.trunc"><code class="xref py py-obj docutils literal notranslate"><span class="pre">trunc</span></code></a>(x)</p></td>
 <td><p>Get truncated value of the input.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.truncdiv" title="tvm.te.truncdiv"><code class="xref py py-obj docutils literal notranslate"><span class="pre">truncdiv</span></code></a>(a, b)</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.truncdiv" title="tvm.te.truncdiv"><code class="xref py py-obj docutils literal notranslate"><span class="pre">truncdiv</span></code></a>(a, b)</p></td>
 <td><p>Compute the truncdiv of two expressions.</p></td>
 </tr>
-<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.truncmod" title="tvm.te.truncmod"><code class="xref py py-obj docutils literal notranslate"><span class="pre">truncmod</span></code></a>(a, b)</p></td>
+<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.truncmod" title="tvm.te.truncmod"><code class="xref py py-obj docutils literal notranslate"><span class="pre">truncmod</span></code></a>(a, b)</p></td>
 <td><p>Compute the truncmod of two expressions.</p></td>
 </tr>
-<tr class="row-even"><td><p><a class="reference internal" href="#tvm.te.var" title="tvm.te.var"><code class="xref py py-obj docutils literal notranslate"><span class="pre">var</span></code></a>([name, dtype])</p></td>
+<tr class="row-odd"><td><p><a class="reference internal" href="#tvm.te.var" title="tvm.te.var"><code class="xref py py-obj docutils literal notranslate"><span class="pre">var</span></code></a>([name, dtype])</p></td>
 <td><p>Create a new variable with specified name and dtype</p></td>
 </tr>
 </tbody>
@@ -683,24 +698,6 @@ tvm.default_trace_action is used.</p>
 </dd></dl>
 
 <dl class="py function">
-<dt id="tvm.te.atan">
-<code class="sig-prename descclassname">tvm.te.</code><code class="sig-name descname">atan</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">x</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.te.atan" title="Permalink to this definition">¶</a></dt>
-<dd><p>Take atan of input x.</p>
-<dl class="field-list simple">
-<dt class="field-odd">Parameters</dt>
-<dd class="field-odd"><p><strong>x</strong> (<a class="reference internal" href="ir.html#tvm.ir.PrimExpr" title="tvm.ir.PrimExpr"><em>PrimExpr</em></a>) – Input argument.</p>
-</dd>
-<dt class="field-even">Returns</dt>
-<dd class="field-even"><p><strong>y</strong> – The result.</p>
-</dd>
-<dt class="field-odd">Return type</dt>
-<dd class="field-odd"><p><a class="reference internal" href="ir.html#tvm.ir.PrimExpr" title="tvm.ir.PrimExpr">PrimExpr</a></p>
-</dd>
-</dl>
-<p class="rubric">Alias of <a class="reference internal" href="tir.html#tvm.tir.atan" title="tvm.tir.atan"><code class="xref py py-func docutils literal notranslate"><span class="pre">tvm.tir.atan()</span></code></a></p>
-</dd></dl>
-
-<dl class="py function">
 <dt id="tvm.te.sqrt">
 <code class="sig-prename descclassname">tvm.te.</code><code class="sig-name descname">sqrt</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">x</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.te.sqrt" title="Permalink to this definition">¶</a></dt>
 <dd><p>Take square root of input x.</p>
@@ -845,6 +842,114 @@ tvm.default_trace_action is used.</p>
 </dd></dl>
 
 <dl class="py function">
+<dt id="tvm.te.asin">
+<code class="sig-prename descclassname">tvm.te.</code><code class="sig-name descname">asin</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">x</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.te.asin" title="Permalink to this definition">¶</a></dt>
+<dd><p>Take asin of input x.</p>
+<dl class="field-list simple">
+<dt class="field-odd">Parameters</dt>
+<dd class="field-odd"><p><strong>x</strong> (<a class="reference internal" href="ir.html#tvm.ir.PrimExpr" title="tvm.ir.PrimExpr"><em>PrimExpr</em></a>) – Input argument.</p>
+</dd>
+<dt class="field-even">Returns</dt>
+<dd class="field-even"><p><strong>y</strong> – The result.</p>
+</dd>
+<dt class="field-odd">Return type</dt>
+<dd class="field-odd"><p><a class="reference internal" href="ir.html#tvm.ir.PrimExpr" title="tvm.ir.PrimExpr">PrimExpr</a></p>
+</dd>
+</dl>
+<p class="rubric">Alias of <a class="reference internal" href="tir.html#tvm.tir.asin" title="tvm.tir.asin"><code class="xref py py-func docutils literal notranslate"><span class="pre">tvm.tir.asin()</span></code></a></p>
+</dd></dl>
+
+<dl class="py function">
+<dt id="tvm.te.asinh">
+<code class="sig-prename descclassname">tvm.te.</code><code class="sig-name descname">asinh</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">x</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.te.asinh" title="Permalink to this definition">¶</a></dt>
+<dd><p>Take asinh of input x.</p>
+<dl class="field-list simple">
+<dt class="field-odd">Parameters</dt>
+<dd class="field-odd"><p><strong>x</strong> (<a class="reference internal" href="ir.html#tvm.ir.PrimExpr" title="tvm.ir.PrimExpr"><em>PrimExpr</em></a>) – Input argument.</p>
+</dd>
+<dt class="field-even">Returns</dt>
+<dd class="field-even"><p><strong>y</strong> – The result.</p>
+</dd>
+<dt class="field-odd">Return type</dt>
+<dd class="field-odd"><p><a class="reference internal" href="ir.html#tvm.ir.PrimExpr" title="tvm.ir.PrimExpr">PrimExpr</a></p>
+</dd>
+</dl>
+<p class="rubric">Alias of <a class="reference internal" href="tir.html#tvm.tir.asinh" title="tvm.tir.asinh"><code class="xref py py-func docutils literal notranslate"><span class="pre">tvm.tir.asinh()</span></code></a></p>
+</dd></dl>
+
+<dl class="py function">
+<dt id="tvm.te.acos">
+<code class="sig-prename descclassname">tvm.te.</code><code class="sig-name descname">acos</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">x</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.te.acos" title="Permalink to this definition">¶</a></dt>
+<dd><p>Take acos of input x.</p>
+<dl class="field-list simple">
+<dt class="field-odd">Parameters</dt>
+<dd class="field-odd"><p><strong>x</strong> (<a class="reference internal" href="ir.html#tvm.ir.PrimExpr" title="tvm.ir.PrimExpr"><em>PrimExpr</em></a>) – Input argument.</p>
+</dd>
+<dt class="field-even">Returns</dt>
+<dd class="field-even"><p><strong>y</strong> – The result.</p>
+</dd>
+<dt class="field-odd">Return type</dt>
+<dd class="field-odd"><p><a class="reference internal" href="ir.html#tvm.ir.PrimExpr" title="tvm.ir.PrimExpr">PrimExpr</a></p>
+</dd>
+</dl>
+<p class="rubric">Alias of <a class="reference internal" href="tir.html#tvm.tir.acos" title="tvm.tir.acos"><code class="xref py py-func docutils literal notranslate"><span class="pre">tvm.tir.acos()</span></code></a></p>
+</dd></dl>
+
+<dl class="py function">
+<dt id="tvm.te.acosh">
+<code class="sig-prename descclassname">tvm.te.</code><code class="sig-name descname">acosh</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">x</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.te.acosh" title="Permalink to this definition">¶</a></dt>
+<dd><p>Take acos of input x.</p>
+<dl class="field-list simple">
+<dt class="field-odd">Parameters</dt>
+<dd class="field-odd"><p><strong>x</strong> (<a class="reference internal" href="ir.html#tvm.ir.PrimExpr" title="tvm.ir.PrimExpr"><em>PrimExpr</em></a>) – Input argument.</p>
+</dd>
+<dt class="field-even">Returns</dt>
+<dd class="field-even"><p><strong>y</strong> – The result.</p>
+</dd>
+<dt class="field-odd">Return type</dt>
+<dd class="field-odd"><p><a class="reference internal" href="ir.html#tvm.ir.PrimExpr" title="tvm.ir.PrimExpr">PrimExpr</a></p>
+</dd>
+</dl>
+<p class="rubric">Alias of <a class="reference internal" href="tir.html#tvm.tir.acosh" title="tvm.tir.acosh"><code class="xref py py-func docutils literal notranslate"><span class="pre">tvm.tir.acosh()</span></code></a></p>
+</dd></dl>
+
+<dl class="py function">
+<dt id="tvm.te.atan">
+<code class="sig-prename descclassname">tvm.te.</code><code class="sig-name descname">atan</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">x</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.te.atan" title="Permalink to this definition">¶</a></dt>
+<dd><p>Take atan of input x.</p>
+<dl class="field-list simple">
+<dt class="field-odd">Parameters</dt>
+<dd class="field-odd"><p><strong>x</strong> (<a class="reference internal" href="ir.html#tvm.ir.PrimExpr" title="tvm.ir.PrimExpr"><em>PrimExpr</em></a>) – Input argument.</p>
+</dd>
+<dt class="field-even">Returns</dt>
+<dd class="field-even"><p><strong>y</strong> – The result.</p>
+</dd>
+<dt class="field-odd">Return type</dt>
+<dd class="field-odd"><p><a class="reference internal" href="ir.html#tvm.ir.PrimExpr" title="tvm.ir.PrimExpr">PrimExpr</a></p>
+</dd>
+</dl>
+<p class="rubric">Alias of <a class="reference internal" href="tir.html#tvm.tir.atan" title="tvm.tir.atan"><code class="xref py py-func docutils literal notranslate"><span class="pre">tvm.tir.atan()</span></code></a></p>
+</dd></dl>
+
+<dl class="py function">
+<dt id="tvm.te.atanh">
+<code class="sig-prename descclassname">tvm.te.</code><code class="sig-name descname">atanh</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">x</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.te.atanh" title="Permalink to this definition">¶</a></dt>
+<dd><p>Take atanh of input x.</p>
+<dl class="field-list simple">
+<dt class="field-odd">Parameters</dt>
+<dd class="field-odd"><p><strong>x</strong> (<a class="reference internal" href="ir.html#tvm.ir.PrimExpr" title="tvm.ir.PrimExpr"><em>PrimExpr</em></a>) – Input argument.</p>
+</dd>
+<dt class="field-even">Returns</dt>
+<dd class="field-even"><p><strong>y</strong> – The result.</p>
+</dd>
+<dt class="field-odd">Return type</dt>
+<dd class="field-odd"><p><a class="reference internal" href="ir.html#tvm.ir.PrimExpr" title="tvm.ir.PrimExpr">PrimExpr</a></p>
+</dd>
+</dl>
+<p class="rubric">Alias of <a class="reference internal" href="tir.html#tvm.tir.atanh" title="tvm.tir.atanh"><code class="xref py py-func docutils literal notranslate"><span class="pre">tvm.tir.atanh()</span></code></a></p>
+</dd></dl>
+
+<dl class="py function">
 <dt id="tvm.te.trunc">
 <code class="sig-prename descclassname">tvm.te.</code><code class="sig-name descname">trunc</code><span class="sig-paren">(</span><em class="sig-param"><span class="n">x</span></em><span class="sig-paren">)</span><a class="headerlink" href="#tvm.te.trunc" title="Permalink to this definition">¶</a></dt>
 <dd><p>Get truncated value of the input.</p>
diff --git a/docs/api/python/tir.html b/docs/api/python/tir.html
index edb0348..5f340d1 100644
--- a/docs/api/python/tir.html
+++ b/docs/api/python/tir.html
@@ -133,7 +133,7 @@
 <li class="toctree-l2"><a class="reference internal" href="topi.html">topi</a></li>
 </ul>
 </li>
-<li class="toctree-l1"><a class="reference internal" href="../../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../../dev/index.html">Design and Developer Guide</a></li>
diff --git a/docs/api/python/topi.html b/docs/api/python/topi.html
index 545dbce..4e1bce8 100644
--- a/docs/api/python/topi.html
+++ b/docs/api/python/topi.html
@@ -40,7 +40,7 @@
   <link rel="stylesheet" href="../../_static/css/tvm_theme.css" type="text/css" />
     <link rel="index" title="Index" href="../../genindex.html" />
     <link rel="search" title="Search" href="../../search.html" />
-    <link rel="next" title="Links to API References" href="../../api_links.html" />
+    <link rel="next" title="Links to API References" href="../links.html" />
     <link rel="prev" title="vta" href="vta/index.html" /> 
 </head>
 
@@ -142,7 +142,7 @@
 </li>
 </ul>
 </li>
-<li class="toctree-l1"><a class="reference internal" href="../../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../../dev/index.html">Design and Developer Guide</a></li>
@@ -2567,7 +2567,7 @@ of an array of tensors.</p>
   
     <div class="rst-footer-buttons" role="navigation" aria-label="footer navigation">
       
-        <a href="../../api_links.html" class="btn btn-neutral float-right" title="Links to API References" accesskey="n" rel="next">Next <span class="fa fa-arrow-circle-right"></span></a>
+        <a href="../links.html" class="btn btn-neutral float-right" title="Links to API References" accesskey="n" rel="next">Next <span class="fa fa-arrow-circle-right"></span></a>
       
       
         <a href="vta/index.html" class="btn btn-neutral float-left" title="vta" accesskey="p" rel="prev"><span class="fa fa-arrow-circle-left"></span> Previous</a>
diff --git a/docs/api/python/vta/index.html b/docs/api/python/vta/index.html
index c348c3c..d9b085c 100644
--- a/docs/api/python/vta/index.html
+++ b/docs/api/python/vta/index.html
@@ -138,7 +138,7 @@
 <li class="toctree-l2"><a class="reference internal" href="../topi.html">topi</a></li>
 </ul>
 </li>
-<li class="toctree-l1"><a class="reference internal" href="../../../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../../links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../../../dev/index.html">Design and Developer Guide</a></li>
diff --git a/docs/api_links.html b/docs/api_links.html
deleted file mode 100644
index f581b2d..0000000
--- a/docs/api_links.html
+++ /dev/null
@@ -1,251 +0,0 @@
-
-
-<!DOCTYPE html>
-<!--[if IE 8]><html class="no-js lt-ie9" lang="en" > <![endif]-->
-<!--[if gt IE 8]><!--> <html class="no-js" lang="en" > <!--<![endif]-->
-<head>
-  <meta charset="utf-8">
-  
-  <meta name="viewport" content="width=device-width, initial-scale=1.0">
-  
-  <title>Links to API References &mdash; tvm 0.7.dev1 documentation</title>
-  
-
-  
-  
-    <link rel="shortcut icon" href="_static/tvm-logo-square.png"/>
-  
-  
-  
-
-  
-  <script type="text/javascript" src="_static/js/modernizr.min.js"></script>
-  
-    
-      <script type="text/javascript" id="documentation_options" data-url_root="./" src="_static/documentation_options.js"></script>
-        <script src="_static/jquery.js"></script>
-        <script src="_static/underscore.js"></script>
-        <script src="_static/doctools.js"></script>
-        <script src="_static/language_data.js"></script>
-        <script async="async" src="https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.5/latest.js?config=TeX-AMS-MML_HTMLorMML"></script>
-    
-    <script type="text/javascript" src="_static/js/theme.js"></script>
-
-    
-
-  
-  <link rel="stylesheet" href="_static/css/theme.css" type="text/css" />
-  <link rel="stylesheet" href="_static/pygments.css" type="text/css" />
-  <link rel="stylesheet" href="_static/gallery.css" type="text/css" />
-  <link rel="stylesheet" href="_static/css/tvm_theme.css" type="text/css" />
-    <link rel="index" title="Index" href="genindex.html" />
-    <link rel="search" title="Search" href="search.html" />
-    <link rel="next" title="Design and Developer Guide" href="dev/index.html" />
-    <link rel="prev" title="topi" href="api/python/topi.html" /> 
-</head>
-
-<body class="wy-body-for-nav">
-
-   
-  <div class="wy-grid-for-nav">
-    
-    <nav data-toggle="wy-nav-shift" class="wy-nav-side">
-      <div class="wy-side-scroll">
-        <div class="wy-side-nav-search" >
-          
-
-          
-            <a href="index.html">
-          
-
-          
-            
-            <img src="_static/tvm-logo-small.png" class="logo" alt="Logo"/>
-          
-          </a>
-
-          
-            
-            
-              <div class="version">
-                0.7.dev1
-              </div>
-            
-          
-
-          
-<div role="search">
-  <form id="rtd-search-form" class="wy-form" action="search.html" method="get">
-    <input type="text" name="q" placeholder="Search docs" />
-    <input type="hidden" name="check_keywords" value="yes" />
-    <input type="hidden" name="area" value="default" />
-  </form>
-</div>
-
-          
-        </div>
-
-        <div class="wy-menu wy-menu-vertical" data-spy="affix" role="navigation" aria-label="main navigation">
-          
-            
-            
-              
-            
-            
-              <ul>
-<li class="toctree-l1"><a class="reference internal" href="install/index.html">Installation</a></li>
-<li class="toctree-l1"><a class="reference internal" href="tutorials/index.html">Tutorials</a></li>
-<li class="toctree-l1"><a class="reference internal" href="vta/index.html">VTA: Deep Learning Accelerator Stack</a></li>
-<li class="toctree-l1"><a class="reference internal" href="deploy/index.html">Deploy and Integration</a></li>
-<li class="toctree-l1"><a class="reference internal" href="contribute/index.html">Contribute to TVM</a></li>
-<li class="toctree-l1"><a class="reference internal" href="faq.html">Frequently Asked Questions</a></li>
-</ul>
-<ul class="current">
-<li class="toctree-l1"><a class="reference internal" href="langref/index.html">Language Reference</a></li>
-<li class="toctree-l1"><a class="reference internal" href="api/python/index.html">Python API</a></li>
-<li class="toctree-l1 current"><a class="current reference internal" href="#">Links to API References</a></li>
-</ul>
-<ul>
-<li class="toctree-l1"><a class="reference internal" href="dev/index.html">Design and Developer Guide</a></li>
-</ul>
-<ul>
-<li class="toctree-l1"><a class="reference internal" href="frontend/tensorflow.html">TensorFlow Frontend</a></li>
-</ul>
-<ul>
-<li class="toctree-l1"><a class="reference internal" href="genindex.html">Index</a></li>
-</ul>
-
-            
-          
-        </div>
-      </div>
-    </nav>
-
-    <section data-toggle="wy-nav-shift" class="wy-nav-content-wrap">
-
-      
-      <nav class="wy-nav-top" aria-label="top navigation">
-        
-          <i data-toggle="wy-nav-top" class="fa fa-bars"></i>
-          <a href="index.html">tvm</a>
-        
-      </nav>
-
-
-      <div class="wy-nav-content">
-        
-        <div class="rst-content">
-        
-          
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-<div role="navigation" aria-label="breadcrumbs navigation">
-
-  <ul class="wy-breadcrumbs">
-    
-      <li><a href="index.html">Docs</a> &raquo;</li>
-        
-      <li>Links to API References</li>
-    
-    
-      <li class="wy-breadcrumbs-aside">
-        
-            
-            <a href="_sources/api_links.rst.txt" rel="nofollow"> View page source</a>
-          
-        
-      </li>
-    
-  </ul>
-
-  
-  <hr/>
-</div>
-          <div role="main" class="document" itemscope="itemscope" itemtype="http://schema.org/Article">
-           <div itemprop="articleBody">
-            
-  <div class="section" id="links-to-api-references">
-<h1>Links to API References<a class="headerlink" href="#links-to-api-references" title="Permalink to this headline">¶</a></h1>
-<p>This page contains links to API references that are build with different doc build system.</p>
-<ul class="simple">
-<li><p><a class="reference external" href="doxygen/index.html">C++ doyxgen API</a></p></li>
-<li><p><a class="reference external" href="jsdoc/index.html">Javascript jsdoc API</a></p></li>
-<li><p><a class="reference external" href="javadoc/index.html">Java Javadoc API</a></p></li>
-</ul>
-</div>
-
-
-           </div>
-           
-          </div>
-          <footer>
-  
-    <div class="rst-footer-buttons" role="navigation" aria-label="footer navigation">
-      
-        <a href="dev/index.html" class="btn btn-neutral float-right" title="Design and Developer Guide" accesskey="n" rel="next">Next <span class="fa fa-arrow-circle-right"></span></a>
-      
-      
-        <a href="api/python/topi.html" class="btn btn-neutral float-left" title="topi" accesskey="p" rel="prev"><span class="fa fa-arrow-circle-left"></span> Previous</a>
-      
-    </div>
-  
-
-  <hr/>
-
-  <div role="contentinfo">
-    <p>
-        &copy; Copyright 2020, Apache Software Foundation
-
-    </p>
-  </div>
-  Built with <a href="http://sphinx-doc.org/">Sphinx</a> using a <a href="https://github.com/rtfd/sphinx_rtd_theme">theme</a> provided by <a href="https://readthedocs.org">Read the Docs</a>. 
-
-</footer>
-
-        </div>
-      </div>
-
-    </section>
-
-  </div>
-  
-
-
-  <script type="text/javascript">
-      jQuery(function () {
-          SphinxRtdTheme.Navigation.enable(true);
-      });
-  </script>
-
-  
-  
-    
-    <!-- Theme Analytics -->
-    <script>
-    (function(i,s,o,g,r,a,m){i['GoogleAnalyticsObject']=r;i[r]=i[r]||function(){
-      (i[r].q=i[r].q||[]).push(arguments)},i[r].l=1*new Date();a=s.createElement(o),
-      m=s.getElementsByTagName(o)[0];a.async=1;a.src=g;m.parentNode.insertBefore(a,m)
-    })(window,document,'script','https://www.google-analytics.com/analytics.js','ga');
-
-    ga('create', 'UA-75982049-2', 'auto');
-    ga('send', 'pageview');
-    </script>
-
-    
-   
-
-</body>
-</html>
\ No newline at end of file
diff --git a/docs/contribute/code_guide.html b/docs/contribute/code_guide.html
index cb7fa11..3faf87a 100644
--- a/docs/contribute/code_guide.html
+++ b/docs/contribute/code_guide.html
@@ -119,7 +119,7 @@
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../langref/index.html">Language Reference</a></li>
 <li class="toctree-l1"><a class="reference internal" href="../api/python/index.html">Python API</a></li>
-<li class="toctree-l1"><a class="reference internal" href="../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../api/links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../dev/index.html">Design and Developer Guide</a></li>
@@ -210,6 +210,37 @@ Except when the function consumes the value by copy constructor or move,
 pass by value is better than pass by const reference in such cases.</p></li>
 <li><p>Favor <code class="docutils literal notranslate"><span class="pre">const</span></code> member function when possible.</p></li>
 </ul>
+<p>We use <cite>clang-format</cite> to enforce the code style. Because different version
+of clang-format might change by its version, it is recommended to use the same
+version of the clang-format as the master.
+You can also use the following command via docker.</p>
+<div class="highlight-bash notranslate"><div class="highlight"><pre>docker/bash.sh tvmai/ci-lint clang-format-10 <span class="o">[</span>path-to-file<span class="o">]</span>
+</pre></div>
+</div>
+<p>clang-format is also not perfect, when necessary, you can use disble clang-format on certain code regions.</p>
+<div class="highlight-c notranslate"><div class="highlight"><pre><span class="c1">// clang-format off</span>
+<span class="kt">void</span> <span class="nf">Test</span><span class="p">()</span> <span class="p">{</span>
+   <span class="c1">// clang-format will be disabled in this region.</span>
+<span class="p">}</span>
+<span class="c1">// clang-format on</span>
+</pre></div>
+</div>
+<p>Because clang-format may not recognize macros, it is recommended to use macro like normal function styles.</p>
+<div class="highlight-c notranslate"><div class="highlight"><pre><span class="cp">#define MACRO_IMPL { custom impl; }</span>
+<span class="cp">#define MACRO_FUNC(x)</span>
+
+<span class="c1">// not preferred, because clang-format might recognize it as types.</span>
+<span class="n">virtual</span> <span class="kt">void</span> <span class="nf">Func1</span><span class="p">()</span> <span class="n">MACRO_IMPL</span>
+
+<span class="c1">// preferred</span>
+<span class="n">virtual</span> <span class="kt">void</span> <span class="n">Func2</span><span class="p">()</span> <span class="n">MACRO_IMPL</span><span class="p">;</span>
+
+<span class="kt">void</span> <span class="nf">Func3</span><span class="p">()</span> <span class="p">{</span>
+  <span class="c1">// preferred</span>
+  <span class="n">MACRO_FUNC</span><span class="p">(</span><span class="n">xyz</span><span class="p">);</span>
+<span class="p">}</span>
+</pre></div>
+</div>
 </div>
 <div class="section" id="python-code-styles">
 <h2>Python Code Styles<a class="headerlink" href="#python-code-styles" title="Permalink to this headline">¶</a></h2>
diff --git a/docs/contribute/code_review.html b/docs/contribute/code_review.html
index ac00139..82d014d 100644
--- a/docs/contribute/code_review.html
+++ b/docs/contribute/code_review.html
@@ -127,7 +127,7 @@
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../langref/index.html">Language Reference</a></li>
 <li class="toctree-l1"><a class="reference internal" href="../api/python/index.html">Python API</a></li>
-<li class="toctree-l1"><a class="reference internal" href="../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../api/links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../dev/index.html">Design and Developer Guide</a></li>
diff --git a/docs/contribute/committer_guide.html b/docs/contribute/committer_guide.html
index dddb489..f2dd72c 100644
--- a/docs/contribute/committer_guide.html
+++ b/docs/contribute/committer_guide.html
@@ -121,7 +121,7 @@
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../langref/index.html">Language Reference</a></li>
 <li class="toctree-l1"><a class="reference internal" href="../api/python/index.html">Python API</a></li>
-<li class="toctree-l1"><a class="reference internal" href="../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../api/links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../dev/index.html">Design and Developer Guide</a></li>
diff --git a/docs/contribute/community.html b/docs/contribute/community.html
index 2746288..1cec818 100644
--- a/docs/contribute/community.html
+++ b/docs/contribute/community.html
@@ -119,7 +119,7 @@
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../langref/index.html">Language Reference</a></li>
 <li class="toctree-l1"><a class="reference internal" href="../api/python/index.html">Python API</a></li>
-<li class="toctree-l1"><a class="reference internal" href="../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../api/links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../dev/index.html">Design and Developer Guide</a></li>
diff --git a/docs/contribute/document.html b/docs/contribute/document.html
index 330b289..c6daaf3 100644
--- a/docs/contribute/document.html
+++ b/docs/contribute/document.html
@@ -120,7 +120,7 @@
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../langref/index.html">Language Reference</a></li>
 <li class="toctree-l1"><a class="reference internal" href="../api/python/index.html">Python API</a></li>
-<li class="toctree-l1"><a class="reference internal" href="../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../api/links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../dev/index.html">Design and Developer Guide</a></li>
diff --git a/docs/contribute/error_handling.html b/docs/contribute/error_handling.html
index 1e95e16..7edcde2 100644
--- a/docs/contribute/error_handling.html
+++ b/docs/contribute/error_handling.html
@@ -118,7 +118,7 @@
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../langref/index.html">Language Reference</a></li>
 <li class="toctree-l1"><a class="reference internal" href="../api/python/index.html">Python API</a></li>
-<li class="toctree-l1"><a class="reference internal" href="../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../api/links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../dev/index.html">Design and Developer Guide</a></li>
diff --git a/docs/contribute/git_howto.html b/docs/contribute/git_howto.html
index 5158aea..be164b3 100644
--- a/docs/contribute/git_howto.html
+++ b/docs/contribute/git_howto.html
@@ -121,7 +121,7 @@
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../langref/index.html">Language Reference</a></li>
 <li class="toctree-l1"><a class="reference internal" href="../api/python/index.html">Python API</a></li>
-<li class="toctree-l1"><a class="reference internal" href="../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../api/links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../dev/index.html">Design and Developer Guide</a></li>
diff --git a/docs/contribute/index.html b/docs/contribute/index.html
index 234fb10..2609f89 100644
--- a/docs/contribute/index.html
+++ b/docs/contribute/index.html
@@ -114,7 +114,7 @@
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../langref/index.html">Language Reference</a></li>
 <li class="toctree-l1"><a class="reference internal" href="../api/python/index.html">Python API</a></li>
-<li class="toctree-l1"><a class="reference internal" href="../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../api/links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../dev/index.html">Design and Developer Guide</a></li>
diff --git a/docs/contribute/pull_request.html b/docs/contribute/pull_request.html
index 077531c..7c8f1cc 100644
--- a/docs/contribute/pull_request.html
+++ b/docs/contribute/pull_request.html
@@ -122,7 +122,7 @@
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../langref/index.html">Language Reference</a></li>
 <li class="toctree-l1"><a class="reference internal" href="../api/python/index.html">Python API</a></li>
-<li class="toctree-l1"><a class="reference internal" href="../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../api/links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../dev/index.html">Design and Developer Guide</a></li>
@@ -209,8 +209,14 @@ git rebase upstream/master
 </pre></div>
 </div>
 </li>
-<li><p>Make sure code style check pass by typing the following command, and all the existing test-cases pass.</p></li>
-<li><p><code class="docutils literal notranslate"><span class="pre">docker/bash.sh</span> <span class="pre">tvmai/ci-lint</span> <span class="pre">./tests/scripts/task_lint.sh</span></code>. (Note: You must install docker beforehand so you can run a docker image.)</p></li>
+<li><p>Make sure code style check pass by typing the following command, and all the existing test-cases pass.</p>
+<div class="highlight-bash notranslate"><div class="highlight"><pre><span class="c1"># Reproduce the lint procedure in the CI.</span>
+docker/bash.sh tvmai/ci-lint ./tests/scripts/task_lint.sh
+<span class="c1"># Run clang-format check for all the files that changed since upstream/master</span>
+docker/bash.sh tvmai/ci-lint ./tests/lint/git-clang-format.sh upstream/master
+</pre></div>
+</div>
+</li>
 <li><p>Add test-cases to cover the new features or bugfix the patch introduces.</p></li>
 <li><p>Document the code you wrote, see more at <a class="reference internal" href="document.html#doc-guide"><span class="std std-ref">Write Document and Tutorials</span></a></p></li>
 <li><p>Send the pull request and fix the problems reported by automatic checks.</p></li>
@@ -229,7 +235,6 @@ We highly value patches that can get in without extensive reviews.</p></li>
 <h2>CI Environment<a class="headerlink" href="#ci-environment" title="Permalink to this headline">¶</a></h2>
 <p>We use docker container to create stable CI environments
 that can be deployed to multiple machines.
-You can find the prebuilt images in <a class="reference external" href="https://hub.docker.com/r/tvmai/">https://hub.docker.com/r/tvmai/</a> .
 Because we want a relatively stable CI environment and make use of pre-cached image,
 all of the CI images are built and maintained by committers.</p>
 <p>Upgrade of CI images can cause problems and need fixes to accommodate the new env.
diff --git a/docs/contribute/release_process.html b/docs/contribute/release_process.html
index dd7fd5c..6aaf09f 100644
--- a/docs/contribute/release_process.html
+++ b/docs/contribute/release_process.html
@@ -123,7 +123,7 @@
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../langref/index.html">Language Reference</a></li>
 <li class="toctree-l1"><a class="reference internal" href="../api/python/index.html">Python API</a></li>
-<li class="toctree-l1"><a class="reference internal" href="../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../api/links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../dev/index.html">Design and Developer Guide</a></li>
diff --git a/docs/deploy/android.html b/docs/deploy/android.html
index d22c969..0fb4db2 100644
--- a/docs/deploy/android.html
+++ b/docs/deploy/android.html
@@ -113,7 +113,7 @@
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../langref/index.html">Language Reference</a></li>
 <li class="toctree-l1"><a class="reference internal" href="../api/python/index.html">Python API</a></li>
-<li class="toctree-l1"><a class="reference internal" href="../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../api/links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../dev/index.html">Design and Developer Guide</a></li>
diff --git a/docs/deploy/cpp_deploy.html b/docs/deploy/cpp_deploy.html
index 540ca18..1ca97d9 100644
--- a/docs/deploy/cpp_deploy.html
+++ b/docs/deploy/cpp_deploy.html
@@ -113,7 +113,7 @@
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../langref/index.html">Language Reference</a></li>
 <li class="toctree-l1"><a class="reference internal" href="../api/python/index.html">Python API</a></li>
-<li class="toctree-l1"><a class="reference internal" href="../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../api/links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../dev/index.html">Design and Developer Guide</a></li>
diff --git a/docs/deploy/hls.html b/docs/deploy/hls.html
index 4a7b23c..3725aef 100644
--- a/docs/deploy/hls.html
+++ b/docs/deploy/hls.html
@@ -115,7 +115,7 @@
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../langref/index.html">Language Reference</a></li>
 <li class="toctree-l1"><a class="reference internal" href="../api/python/index.html">Python API</a></li>
-<li class="toctree-l1"><a class="reference internal" href="../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../api/links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../dev/index.html">Design and Developer Guide</a></li>
diff --git a/docs/deploy/index.html b/docs/deploy/index.html
index 97c66c6..1275c84 100644
--- a/docs/deploy/index.html
+++ b/docs/deploy/index.html
@@ -109,7 +109,7 @@
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../langref/index.html">Language Reference</a></li>
 <li class="toctree-l1"><a class="reference internal" href="../api/python/index.html">Python API</a></li>
-<li class="toctree-l1"><a class="reference internal" href="../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../api/links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../dev/index.html">Design and Developer Guide</a></li>
diff --git a/docs/deploy/integrate.html b/docs/deploy/integrate.html
index a8c8930..36a360e 100644
--- a/docs/deploy/integrate.html
+++ b/docs/deploy/integrate.html
@@ -113,7 +113,7 @@
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../langref/index.html">Language Reference</a></li>
 <li class="toctree-l1"><a class="reference internal" href="../api/python/index.html">Python API</a></li>
-<li class="toctree-l1"><a class="reference internal" href="../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../api/links.html">Links to API References</a></li>
 </ul>
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../dev/index.html">Design and Developer Guide</a></li>
diff --git a/docs/dev/benchmark.html b/docs/dev/benchmark.html
index 529f906..33f080a 100644
--- a/docs/dev/benchmark.html
+++ b/docs/dev/benchmark.html
@@ -103,7 +103,7 @@
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../langref/index.html">Language Reference</a></li>
 <li class="toctree-l1"><a class="reference internal" href="../api/python/index.html">Python API</a></li>
-<li class="toctree-l1"><a class="reference internal" href="../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../api/links.html">Links to API References</a></li>
 </ul>
 <ul class="current">
 <li class="toctree-l1 current"><a class="reference internal" href="index.html">Design and Developer Guide</a><ul class="current">
diff --git a/docs/dev/codebase_walkthrough.html b/docs/dev/codebase_walkthrough.html
index 327ef2b..1237e74 100644
--- a/docs/dev/codebase_walkthrough.html
+++ b/docs/dev/codebase_walkthrough.html
@@ -103,7 +103,7 @@
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../langref/index.html">Language Reference</a></li>
 <li class="toctree-l1"><a class="reference internal" href="../api/python/index.html">Python API</a></li>
-<li class="toctree-l1"><a class="reference internal" href="../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../api/links.html">Links to API References</a></li>
 </ul>
 <ul class="current">
 <li class="toctree-l1 current"><a class="reference internal" href="index.html">Design and Developer Guide</a><ul class="current">
diff --git a/docs/dev/convert_layout.html b/docs/dev/convert_layout.html
index ab5266c..def16a4 100644
--- a/docs/dev/convert_layout.html
+++ b/docs/dev/convert_layout.html
@@ -103,7 +103,7 @@
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../langref/index.html">Language Reference</a></li>
 <li class="toctree-l1"><a class="reference internal" href="../api/python/index.html">Python API</a></li>
-<li class="toctree-l1"><a class="reference internal" href="../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../api/links.html">Links to API References</a></li>
 </ul>
 <ul class="current">
 <li class="toctree-l1 current"><a class="reference internal" href="index.html">Design and Developer Guide</a><ul class="current">
@@ -263,7 +263,7 @@
 <p>These steps happen for each operator in sequence, where ConvertLayout pass keeps on passing the new layouts to the next operator properties, finally resulting in modifying the whole graph operator-by-operator. Now, let’s look at a couple of examples of how to define the two properties.</p>
 <p><strong>FTVMConvertLayout - Python callback for layout alteration</strong> - This is used for <em>heavily-layout sensitive</em> operators. For example, one can return a new convolution operator with new data and kernel layout. The other 2 components will infer layout and insert layout transforms if needed. One example for convolution operator is as follows where we are converting to NCHW layout.</p>
 <div class="highlight-python notranslate"><div class="highlight"><pre><span class="nd">@reg.register_convert_op_layout</span><span class="p">(</span><span class="s2">&quot;nn.conv2d&quot;</span><span class="p">)</span>
-<span class="k">def</span> <span class="nf">convert_conv2d</span><span class="p">(</span><span class="n">attrs</span><span class="p">,</span> <span class="n">inputs</span><span class="p">,</span> <span class="n">tinfos</span><span class="p">,</span> <span class="n">desired_layout</span><span class="p">):</span>
+<span class="k">def</span> <span class="nf">convert_conv2d</span><span class="p">(</span><span class="n">attrs</span><span class="p">,</span> <span class="n">inputs</span><span class="p">,</span> <span class="n">tinfos</span><span class="p">,</span> <span class="n">desired_layouts</span><span class="p">):</span>
     <span class="sd">&quot;&quot;&quot;Convert Layout pass registration for conv2d op.</span>
 
 <span class="sd">    Parameters</span>
@@ -274,8 +274,9 @@
 <span class="sd">        The args of the Relay expr to be legalized</span>
 <span class="sd">    tinfos : list of types</span>
 <span class="sd">        List of input and output types</span>
-<span class="sd">    desired_layout : str</span>
-<span class="sd">        The desired layout</span>
+<span class="sd">    desired_layouts : list of layout strings</span>
+<span class="sd">            List of layouts defining our desired</span>
+<span class="sd">            layout for the data and kernel inputs respectively.</span>
 
 <span class="sd">    Returns</span>
 <span class="sd">    -------</span>
@@ -284,19 +285,30 @@
 <span class="sd">    &quot;&quot;&quot;</span>
 
     <span class="kn">from</span> <span class="nn">tvm</span> <span class="kn">import</span> <span class="n">relay</span>
-    <span class="n">data_layout</span> <span class="o">=</span> <span class="n">attrs</span><span class="p">[</span><span class="s1">&#39;data_layout&#39;</span><span class="p">]</span>
-    <span class="n">kernel_layout</span> <span class="o">=</span> <span class="n">attrs</span><span class="p">[</span><span class="s1">&#39;kernel_layout&#39;</span><span class="p">]</span>
     <span class="n">data</span><span class="p">,</span> <span class="n">weight</span> <span class="o">=</span> <span class="n">inputs</span>
-    <span class="k">assert</span> <span class="n">desired_layout</span> <span class="o">==</span> <span class="s1">&#39;NCHW&#39;</span><span class="p">,</span> \
-            <span class="s2">&quot;Currently only transformation to NCHW layout is supported.&quot;</span>
-    <span class="k">if</span> <span class="n">desired_layout</span> <span class="o">==</span> <span class="s1">&#39;NCHW&#39;</span><span class="p">:</span>
-        <span class="n">new_attrs</span> <span class="o">=</span> <span class="nb">dict</span><span class="p">(</span><span class="n">attrs</span><span class="p">)</span>
-        <span class="n">new_attrs</span><span class="p">[</span><span class="s1">&#39;data_layout&#39;</span><span class="p">]</span> <span class="o">=</span> <span class="n">desired_layout</span>
-        <span class="n">new_attrs</span><span class="p">[</span><span class="s1">&#39;kernel_layout&#39;</span><span class="p">]</span> <span class="o">=</span> <span class="s1">&#39;OIHW&#39;</span>
+    <span class="n">new_attrs</span> <span class="o">=</span> <span class="nb">dict</span><span class="p">(</span><span class="n">attrs</span><span class="p">)</span>
+
+    <span class="c1"># We expect 2 desired layouts to be specified, one for the data and one for the kernel.</span>
+    <span class="k">assert</span> <span class="nb">len</span><span class="p">(</span><span class="n">desired_layouts</span><span class="p">)</span> <span class="o">==</span> <span class="mi">2</span><span class="p">,</span> <span class="s2">&quot;A desired layout is expected for both of nn.conv2d&#39;s inputs&quot;</span>
+
+    <span class="c1"># Use the first entry in desired layouts which specifies the data layout.</span>
+    <span class="c1"># The expected ordering of layouts for this operator is defined by this function.</span>
+    <span class="n">desired_data_layout</span><span class="p">,</span> <span class="n">desired_kernel_layout</span> <span class="o">=</span> <span class="nb">map</span><span class="p">(</span><span class="nb">str</span><span class="p">,</span> <span class="n">desired_layouts</span><span class="p">)</span>
+
+    <span class="k">assert</span> <span class="n">desired_data_layout</span> <span class="o">!=</span> <span class="s2">&quot;default&quot;</span><span class="p">,</span> <span class="s2">&quot;Data layout cannot be default&quot;</span>
+
+    <span class="n">new_attrs</span><span class="p">[</span><span class="s1">&#39;data_layout&#39;</span><span class="p">]</span> <span class="o">=</span> <span class="n">desired_data_layout</span>
+
+    <span class="k">if</span> <span class="n">desired_data_layout</span> <span class="o">==</span> <span class="s1">&#39;NCHW&#39;</span><span class="p">:</span>
+        <span class="k">if</span> <span class="n">desired_kernel_layout</span> <span class="o">!=</span> <span class="s1">&#39;default&#39;</span><span class="p">:</span>
+            <span class="n">new_attrs</span><span class="p">[</span><span class="s1">&#39;kernel_layout&#39;</span><span class="p">]</span> <span class="o">=</span> <span class="n">desired_kernel_layout</span>
+        <span class="k">else</span><span class="p">:</span>
+            <span class="n">new_attrs</span><span class="p">[</span><span class="s1">&#39;kernel_layout&#39;</span><span class="p">]</span> <span class="o">=</span> <span class="s1">&#39;OIHW&#39;</span>
         <span class="c1"># Actual insertion of layout transforms is taken care internally</span>
         <span class="c1"># by ConvertLayout pass.</span>
         <span class="k">return</span> <span class="n">relay</span><span class="o">.</span><span class="n">nn</span><span class="o">.</span><span class="n">conv2d</span><span class="p">(</span><span class="n">data</span><span class="p">,</span> <span class="n">weight</span><span class="p">,</span> <span class="o">**</span><span class="n">new_attrs</span><span class="p">)</span>
-    <span class="k">return</span> <span class="bp">None</span>
+
+    <span class="k">raise</span> <span class="ne">ValueError</span><span class="p">(</span><span class="s1">&#39;Layout </span><span class="si">%s</span><span class="s1"> is not yet supported&#39;</span> <span class="o">%</span> <span class="n">desired_data_layout</span><span class="p">)</span>
 </pre></div>
 </div>
 <p><strong>FInferCorrectLayout - Layout inference</strong> - Currently, this attribute is exposed only in C++. This function takes original input layouts and the new input layouts (passed from the previous operator or from the python callback for layout alteration), and infers the final data layouts. Layout inference is called for each operator. The usage might vary for different operator categories. For layout agnostic operators, we just want to return the new data layouts in this funct [...]
@@ -379,15 +391,19 @@
 <div class="section" id="usage">
 <h2>4. Usage<a class="headerlink" href="#usage" title="Permalink to this headline">¶</a></h2>
 <p>ConvertLayout pass is extremely easy to use. The pass is not a part of default relay.build pipeline. The intended usage is to call it between the framework-to-relay parser and relay.build module call.</p>
+<p>In order to specify the layouts to convert to, we create a mapping of heavily-layout sensitive operators to a list of the desired layouts for that operator. The first example below specifies data layout, we allow the kernel layout to be automatically converted to one that is supported by TVM (for that particular data layout and operator). This is specified by the use of the “default” keyword. The second example shows how we could have also converted to a specific kernel layout of our  [...]
 <div class="highlight-python notranslate"><div class="highlight"><pre><span class="c1"># TFlite framework to Relay parser - Default layout is NHWC</span>
 <span class="n">mod</span><span class="p">,</span> <span class="n">params</span> <span class="o">=</span> <span class="n">relay</span><span class="o">.</span><span class="n">frontend</span><span class="o">.</span><span class="n">from_tflite</span><span class="p">(</span><span class="n">tflite_model</span><span class="p">,</span>
                                          <span class="n">shape_dict</span><span class="o">=</span><span class="n">shape_dict</span><span class="p">,</span>
                                          <span class="n">dtype_dict</span><span class="o">=</span><span class="n">dtype_dict</span><span class="p">)</span>
 
+<span class="c1"># We assume our model&#39;s heavily-layout sensitive operators only consist of nn.conv2d</span>
+<span class="n">desired_layouts</span> <span class="o">=</span> <span class="p">{</span><span class="s1">&#39;nn.conv2d&#39;</span><span class="p">:</span> <span class="p">[</span><span class="s1">&#39;NCHW&#39;</span><span class="p">,</span> <span class="s1">&#39;default&#39;</span><span class="p">]}</span>
+
 <span class="c1"># Convert the layout to NCHW</span>
 <span class="c1"># RemoveUnunsedFunctions is used to clean up the graph.</span>
 <span class="n">seq</span> <span class="o">=</span> <span class="n">tvm</span><span class="o">.</span><span class="n">transform</span><span class="o">.</span><span class="n">Sequential</span><span class="p">([</span><span class="n">relay</span><span class="o">.</span><span class="n">transform</span><span class="o">.</span><span class="n">RemoveUnusedFunctions</span><span class="p">(),</span>
-                                  <span class="n">relay</span><span class="o">.</span><span class="n">transform</span><span class="o">.</span><span class="n">ConvertLayout</span><span class="p">(</span><span class="s1">&#39;NCHW&#39;</span><span class="p">)])</span>
+                                <span class="n">relay</span><span class="o">.</span><span class="n">transform</span><span class="o">.</span><span class="n">ConvertLayout</span><span class="p">(</span><span class="n">desired_layouts</span><span class="p">)])</span>
 <span class="k">with</span> <span class="n">relay</span><span class="o">.</span><span class="n">transform</span><span class="o">.</span><span class="n">PassContext</span><span class="p">(</span><span class="n">opt_level</span><span class="o">=</span><span class="mi">3</span><span class="p">):</span>
     <span class="n">mod</span> <span class="o">=</span> <span class="n">seq</span><span class="p">(</span><span class="n">mod</span><span class="p">)</span>
 
@@ -396,6 +412,11 @@
      <span class="n">graph</span><span class="p">,</span> <span class="n">lib</span><span class="p">,</span> <span class="n">params</span> <span class="o">=</span> <span class="n">relay</span><span class="o">.</span><span class="n">build</span><span class="p">(</span><span class="n">mod</span><span class="p">,</span> <span class="n">target</span><span class="p">,</span> <span class="n">params</span><span class="o">=</span><span class="n">params</span><span class="p">)</span>
 </pre></div>
 </div>
+<div class="highlight-python notranslate"><div class="highlight"><pre><span class="n">desired_layouts</span> <span class="o">=</span> <span class="p">{</span><span class="s1">&#39;nn.conv2d&#39;</span><span class="p">:</span> <span class="p">[</span><span class="s1">&#39;NCHW&#39;</span><span class="p">,</span> <span class="s1">&#39;OIHW&#39;</span><span class="p">]}</span>
+<span class="k">pass</span> <span class="o">=</span> <span class="n">relay</span><span class="o">.</span><span class="n">transform</span><span class="o">.</span><span class="n">ConvertLayout</span><span class="p">(</span><span class="n">desired_layouts</span><span class="p">)</span>
+</pre></div>
+</div>
+<p>The ordering of the layouts is defined by the implementation of <cite>register_convert_op_layout(“OPNAME”)</cite>, you can refer to the docstring which should explicitly state the expected layout. In the examples above it’s [data_layout, kernel_layout].</p>
 <p>Current implementation has support for almost all the operators commonly used in image classification models. However, if one encounters too many data layout transforms in the graph, it is highly likely that there is an operator whose layouts need special handling as described in Section 3. Some pull requests that can help in such a situation are</p>
 <ul class="simple">
 <li><p>Layout inference for <a class="reference external" href="https://github.com/apache/incubator-tvm/pull/4600">Batch Norm</a> - Batch normalization falls into the category of lightly-sensitive operator. The PR shows how to handle the layout inference for batch norm.</p></li>
diff --git a/docs/dev/debugger.html b/docs/dev/debugger.html
index c9c1e76..38ebdd6 100644
--- a/docs/dev/debugger.html
+++ b/docs/dev/debugger.html
@@ -103,7 +103,7 @@
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../langref/index.html">Language Reference</a></li>
 <li class="toctree-l1"><a class="reference internal" href="../api/python/index.html">Python API</a></li>
-<li class="toctree-l1"><a class="reference internal" href="../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../api/links.html">Links to API References</a></li>
 </ul>
 <ul class="current">
 <li class="toctree-l1 current"><a class="reference internal" href="index.html">Design and Developer Guide</a><ul class="current">
diff --git a/docs/dev/hybrid_script.html b/docs/dev/hybrid_script.html
index 9a2acbe..28b3e6a 100644
--- a/docs/dev/hybrid_script.html
+++ b/docs/dev/hybrid_script.html
@@ -103,7 +103,7 @@
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../langref/index.html">Language Reference</a></li>
 <li class="toctree-l1"><a class="reference internal" href="../api/python/index.html">Python API</a></li>
-<li class="toctree-l1"><a class="reference internal" href="../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../api/links.html">Links to API References</a></li>
 </ul>
 <ul class="current">
 <li class="toctree-l1 current"><a class="reference internal" href="index.html">Design and Developer Guide</a><ul class="current">
diff --git a/docs/dev/index.html b/docs/dev/index.html
index 570f2fe..aa49fbc 100644
--- a/docs/dev/index.html
+++ b/docs/dev/index.html
@@ -41,7 +41,7 @@
     <link rel="index" title="Index" href="../genindex.html" />
     <link rel="search" title="Search" href="../search.html" />
     <link rel="next" title="TVM Runtime System" href="runtime.html" />
-    <link rel="prev" title="Links to API References" href="../api_links.html" /> 
+    <link rel="prev" title="Links to API References" href="../api/links.html" /> 
 </head>
 
 <body class="wy-body-for-nav">
@@ -103,7 +103,7 @@
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../langref/index.html">Language Reference</a></li>
 <li class="toctree-l1"><a class="reference internal" href="../api/python/index.html">Python API</a></li>
-<li class="toctree-l1"><a class="reference internal" href="../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../api/links.html">Links to API References</a></li>
 </ul>
 <ul class="current">
 <li class="toctree-l1 current"><a class="current reference internal" href="#">Design and Developer Guide</a><ul>
@@ -313,7 +313,7 @@ In this part of documentation, we share the rationale for the specific choices m
         <a href="runtime.html" class="btn btn-neutral float-right" title="TVM Runtime System" accesskey="n" rel="next">Next <span class="fa fa-arrow-circle-right"></span></a>
       
       
-        <a href="../api_links.html" class="btn btn-neutral float-left" title="Links to API References" accesskey="p" rel="prev"><span class="fa fa-arrow-circle-left"></span> Previous</a>
+        <a href="../api/links.html" class="btn btn-neutral float-left" title="Links to API References" accesskey="p" rel="prev"><span class="fa fa-arrow-circle-left"></span> Previous</a>
       
     </div>
   
diff --git a/docs/dev/inferbound.html b/docs/dev/inferbound.html
index 2826af3..dfe5d31 100644
--- a/docs/dev/inferbound.html
+++ b/docs/dev/inferbound.html
@@ -103,7 +103,7 @@
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../langref/index.html">Language Reference</a></li>
 <li class="toctree-l1"><a class="reference internal" href="../api/python/index.html">Python API</a></li>
-<li class="toctree-l1"><a class="reference internal" href="../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../api/links.html">Links to API References</a></li>
 </ul>
 <ul class="current">
 <li class="toctree-l1 current"><a class="reference internal" href="index.html">Design and Developer Guide</a><ul class="current">
diff --git a/docs/dev/introduction_to_module_serialization.html b/docs/dev/introduction_to_module_serialization.html
index 6f772d4..7f4a398 100644
--- a/docs/dev/introduction_to_module_serialization.html
+++ b/docs/dev/introduction_to_module_serialization.html
@@ -103,7 +103,7 @@
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../langref/index.html">Language Reference</a></li>
 <li class="toctree-l1"><a class="reference internal" href="../api/python/index.html">Python API</a></li>
-<li class="toctree-l1"><a class="reference internal" href="../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../api/links.html">Links to API References</a></li>
 </ul>
 <ul class="current">
 <li class="toctree-l1 current"><a class="reference internal" href="index.html">Design and Developer Guide</a><ul class="current">
diff --git a/docs/dev/relay_add_op.html b/docs/dev/relay_add_op.html
index e074770..89fce72 100644
--- a/docs/dev/relay_add_op.html
+++ b/docs/dev/relay_add_op.html
@@ -103,7 +103,7 @@
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../langref/index.html">Language Reference</a></li>
 <li class="toctree-l1"><a class="reference internal" href="../api/python/index.html">Python API</a></li>
-<li class="toctree-l1"><a class="reference internal" href="../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../api/links.html">Links to API References</a></li>
 </ul>
 <ul class="current">
 <li class="toctree-l1 current"><a class="reference internal" href="index.html">Design and Developer Guide</a><ul class="current">
diff --git a/docs/dev/relay_add_pass.html b/docs/dev/relay_add_pass.html
index a1bc794..6d8960c 100644
--- a/docs/dev/relay_add_pass.html
+++ b/docs/dev/relay_add_pass.html
@@ -103,7 +103,7 @@
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../langref/index.html">Language Reference</a></li>
 <li class="toctree-l1"><a class="reference internal" href="../api/python/index.html">Python API</a></li>
-<li class="toctree-l1"><a class="reference internal" href="../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../api/links.html">Links to API References</a></li>
 </ul>
 <ul class="current">
 <li class="toctree-l1 current"><a class="reference internal" href="index.html">Design and Developer Guide</a><ul class="current">
@@ -444,7 +444,7 @@ default implementation.</p>
 pointed to by <code class="docutils literal notranslate"><span class="pre">op-&gt;index</span></code>. The reason we need to check is because
 <code class="docutils literal notranslate"><span class="pre">op-&gt;tuple</span></code> might evaluate to a tuple, without itself being a tuple.</p>
 <div class="highlight-c notranslate"><div class="highlight"><pre><span class="n">Expr</span> <span class="nf">VisitExpr_</span><span class="p">(</span><span class="k">const</span> <span class="n">CallNode</span><span class="o">*</span> <span class="n">call</span><span class="p">)</span> <span class="n">final</span> <span class="p">{</span>
-  <span class="k">static</span> <span class="k">auto</span> <span class="n">op_stateful</span> <span class="o">=</span> <span class="n">Op</span><span class="o">::</span><span class="n">GetAttr</span><span class="o">&lt;</span><span class="n">TOpIsStateful</span><span class="o">&gt;</span><span class="p">(</span><span class="s">&quot;TOpIsStateful&quot;</span><span class="p">);</span>
+  <span class="k">static</span> <span class="k">auto</span> <span class="n">op_stateful</span> <span class="o">=</span> <span class="n">Op</span><span class="o">::</span><span class="n">GetAttrMap</span><span class="o">&lt;</span><span class="n">TOpIsStateful</span><span class="o">&gt;</span><span class="p">(</span><span class="s">&quot;TOpIsStateful&quot;</span><span class="p">);</span>
   <span class="n">Expr</span> <span class="n">res</span> <span class="o">=</span> <span class="n">ExprMutator</span><span class="o">::</span><span class="n">VisitExpr_</span><span class="p">(</span><span class="n">call</span><span class="p">);</span>
   <span class="n">call</span> <span class="o">=</span> <span class="n">res</span><span class="p">.</span><span class="n">as</span><span class="o">&lt;</span><span class="n">CallNode</span><span class="o">&gt;</span><span class="p">();</span>
   <span class="c1">// We don&#39;t constant fold function with zero arguments.</span>
diff --git a/docs/dev/relay_bring_your_own_codegen.html b/docs/dev/relay_bring_your_own_codegen.html
index a8f6886..af4d6fc 100644
--- a/docs/dev/relay_bring_your_own_codegen.html
+++ b/docs/dev/relay_bring_your_own_codegen.html
@@ -103,7 +103,7 @@
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../langref/index.html">Language Reference</a></li>
 <li class="toctree-l1"><a class="reference internal" href="../api/python/index.html">Python API</a></li>
-<li class="toctree-l1"><a class="reference internal" href="../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../api/links.html">Links to API References</a></li>
 </ul>
 <ul class="current">
 <li class="toctree-l1 current"><a class="reference internal" href="index.html">Design and Developer Guide</a><ul class="current">
diff --git a/docs/dev/relay_intro.html b/docs/dev/relay_intro.html
index c0c0778..ba6d717 100644
--- a/docs/dev/relay_intro.html
+++ b/docs/dev/relay_intro.html
@@ -103,7 +103,7 @@
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../langref/index.html">Language Reference</a></li>
 <li class="toctree-l1"><a class="reference internal" href="../api/python/index.html">Python API</a></li>
-<li class="toctree-l1"><a class="reference internal" href="../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../api/links.html">Links to API References</a></li>
 </ul>
 <ul class="current">
 <li class="toctree-l1 current"><a class="reference internal" href="index.html">Design and Developer Guide</a><ul class="current">
diff --git a/docs/dev/relay_op_strategy.html b/docs/dev/relay_op_strategy.html
index 030858d..485dcec 100644
--- a/docs/dev/relay_op_strategy.html
+++ b/docs/dev/relay_op_strategy.html
@@ -103,7 +103,7 @@
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../langref/index.html">Language Reference</a></li>
 <li class="toctree-l1"><a class="reference internal" href="../api/python/index.html">Python API</a></li>
-<li class="toctree-l1"><a class="reference internal" href="../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../api/links.html">Links to API References</a></li>
 </ul>
 <ul class="current">
 <li class="toctree-l1 current"><a class="reference internal" href="index.html">Design and Developer Guide</a><ul class="current">
diff --git a/docs/dev/relay_pass_infra.html b/docs/dev/relay_pass_infra.html
index 187722f..48918bb 100644
--- a/docs/dev/relay_pass_infra.html
+++ b/docs/dev/relay_pass_infra.html
@@ -103,7 +103,7 @@
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../langref/index.html">Language Reference</a></li>
 <li class="toctree-l1"><a class="reference internal" href="../api/python/index.html">Python API</a></li>
-<li class="toctree-l1"><a class="reference internal" href="../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../api/links.html">Links to API References</a></li>
 </ul>
 <ul class="current">
 <li class="toctree-l1 current"><a class="reference internal" href="index.html">Design and Developer Guide</a><ul class="current">
diff --git a/docs/dev/runtime.html b/docs/dev/runtime.html
index 1ba4eb8..f09f8b4 100644
--- a/docs/dev/runtime.html
+++ b/docs/dev/runtime.html
@@ -103,7 +103,7 @@
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../langref/index.html">Language Reference</a></li>
 <li class="toctree-l1"><a class="reference internal" href="../api/python/index.html">Python API</a></li>
-<li class="toctree-l1"><a class="reference internal" href="../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../api/links.html">Links to API References</a></li>
 </ul>
 <ul class="current">
 <li class="toctree-l1 current"><a class="reference internal" href="index.html">Design and Developer Guide</a><ul class="current">
diff --git a/docs/dev/security.html b/docs/dev/security.html
index 1c7deae..73f73ba 100644
--- a/docs/dev/security.html
+++ b/docs/dev/security.html
@@ -103,7 +103,7 @@
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../langref/index.html">Language Reference</a></li>
 <li class="toctree-l1"><a class="reference internal" href="../api/python/index.html">Python API</a></li>
-<li class="toctree-l1"><a class="reference internal" href="../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../api/links.html">Links to API References</a></li>
 </ul>
 <ul class="current">
 <li class="toctree-l1 current"><a class="reference internal" href="index.html">Design and Developer Guide</a><ul class="current">
diff --git a/docs/dev/virtual_machine.html b/docs/dev/virtual_machine.html
index b8e3256..e85c7f3 100644
--- a/docs/dev/virtual_machine.html
+++ b/docs/dev/virtual_machine.html
@@ -103,7 +103,7 @@
 <ul>
 <li class="toctree-l1"><a class="reference internal" href="../langref/index.html">Language Reference</a></li>
 <li class="toctree-l1"><a class="reference internal" href="../api/python/index.html">Python API</a></li>
-<li class="toctree-l1"><a class="reference internal" href="../api_links.html">Links to API References</a></li>
+<li class="toctree-l1"><a class="reference internal" href="../api/links.html">Links to API References</a></li>
 </ul>
 <ul class="current">
 <li class="toctree-l1 current"><a class="reference internal" href="index.html">Design and Developer Guide</a><ul class="current">
diff --git a/docs/doxygen/algorithm_8h.html b/docs/doxygen/algorithm_8h.html
index d5b5394..9860ed1 100644
--- a/docs/doxygen/algorithm_8h.html
+++ b/docs/doxygen/algorithm_8h.html
@@ -98,7 +98,7 @@ var searchBox = new SearchBox("searchBox", "search",false,'Search');
 </div><div class="textblock"><div class="dynheader">
 Include dependency graph for algorithm.h:</div>
 <div class="dyncontent">
-<div class="center"><iframe scrolling="no" frameborder="0" src="algorithm_8h__incl.svg" width="3952" height="1440"><p><b>This browser is not able to show SVG: try Firefox, Chrome, Safari, or Opera instead.</b></p></iframe>
+<div class="center"><iframe scrolling="no" frameborder="0" src="algorithm_8h__incl.svg" width="4070" height="1440"><p><b>This browser is not able to show SVG: try Firefox, Chrome, Safari, or Opera instead.</b></p></iframe>
 </div>
 </div>
 </div>
diff --git a/docs/doxygen/algorithm_8h__incl.svg b/docs/doxygen/algorithm_8h__incl.svg
index 94a3df0..f70af28 100644
--- a/docs/doxygen/algorithm_8h__incl.svg
+++ b/docs/doxygen/algorithm_8h__incl.svg
@@ -4,1167 +4,1172 @@
 <!-- Generated by graphviz version 2.38.0 (20140413.2041)
  -->
 <!-- Title: include/tvm/relay/attrs/algorithm.h Pages: 1 -->
-<svg width="2964pt" height="1080pt"
- viewBox="0.00 0.00 2964.00 1080.00" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink">
+<svg width="3052pt" height="1080pt"
+ viewBox="0.00 0.00 3052.00 1080.00" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink">
 <g id="graph0" class="graph" transform="scale(1 1) rotate(0) translate(4 1076)">
 <title>include/tvm/relay/attrs/algorithm.h</title>
-<polygon fill="white" stroke="none" points="-4,4 -4,-1076 2960,-1076 2960,4 -4,4"/>
+<polygon fill="white" stroke="none" points="-4,4 -4,-1076 3048,-1076 3048,4 -4,4"/>
 <!-- Node1 -->
 <g id="node1" class="node"><title>Node1</title>
-<polygon fill="#bfbfbf" stroke="black" points="128,-1041.5 128,-1071.5 252,-1071.5 252,-1041.5 128,-1041.5"/>
-<text text-anchor="start" x="136" y="-1059.5" font-family="Helvetica,sans-Serif" font-size="10.00">include/tvm/relay/attrs</text>
-<text text-anchor="middle" x="190" y="-1048.5" font-family="Helvetica,sans-Serif" font-size="10.00">/algorithm.h</text>
+<polygon fill="#bfbfbf" stroke="black" points="2499,-1041.5 2499,-1071.5 2623,-1071.5 2623,-1041.5 2499,-1041.5"/>
+<text text-anchor="start" x="2507" y="-1059.5" font-family="Helvetica,sans-Serif" font-size="10.00">include/tvm/relay/attrs</text>
+<text text-anchor="middle" x="2561" y="-1048.5" font-family="Helvetica,sans-Serif" font-size="10.00">/algorithm.h</text>
 </g>
 <!-- Node2 -->
 <g id="node2" class="node"><title>Node2</title>
 <g id="a_node2"><a xlink:href="ir_2attrs_8h.html" target="_top" xlink:title="Helpers for attribute objects. ">
-<polygon fill="white" stroke="black" points="814.5,-817.5 814.5,-836.5 893.5,-836.5 893.5,-817.5 814.5,-817.5"/>
-<text text-anchor="middle" x="854" y="-824.5" font-family="Helvetica,sans-Serif" font-size="10.00">tvm/ir/attrs.h</text>
+<polygon fill="white" stroke="black" points="651.5,-817.5 651.5,-836.5 730.5,-836.5 730.5,-817.5 651.5,-817.5"/>
+<text text-anchor="middle" x="691" y="-824.5" font-family="Helvetica,sans-Serif" font-size="10.00">tvm/ir/attrs.h</text>
 </a>
 </g>
 </g>
 <!-- Node1&#45;&gt;Node2 -->
 <g id="edge1" class="edge"><title>Node1&#45;&gt;Node2</title>
-<path fill="none" stroke="midnightblue" d="M231.035,-1041.44C352.211,-999.923 707.002,-878.365 819.416,-839.849"/>
-<polygon fill="midnightblue" stroke="midnightblue" points="820.593,-843.146 828.919,-836.593 818.324,-836.524 820.593,-843.146"/>
+<path fill="none" stroke="midnightblue" d="M2498.89,-1055.18C2127.21,-1052.97 210.534,-1037.31 126,-949 102.643,-924.599 103.083,-897.815 126,-873 160.572,-835.565 509.108,-829.248 640.891,-828.201"/>
+<polygon fill="midnightblue" stroke="midnightblue" points="641.136,-831.699 651.11,-828.127 641.085,-824.7 641.136,-831.699"/>
 </g>
-<!-- Node12 -->
-<g id="node12" class="node"><title>Node12</title>
-<polygon fill="white" stroke="#bfbfbf" points="480,-62 480,-81 524,-81 524,-62 480,-62"/>
-<text text-anchor="middle" x="502" y="-69" font-family="Helvetica,sans-Serif" font-size="10.00">string</text>
+<!-- Node16 -->
+<g id="node16" class="node"><title>Node16</title>
+<polygon fill="white" stroke="#bfbfbf" points="2642,-62 2642,-81 2686,-81 2686,-62 2642,-62"/>
+<text text-anchor="middle" x="2664" y="-69" font-family="Helvetica,sans-Serif" font-size="10.00">string</text>
 </g>
-<!-- Node1&#45;&gt;Node12 -->
-<g id="edge169" class="edge"><title>Node1&#45;&gt;Node12</title>
... 462149 lines suppressed ...