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 2023/04/01 19:31:59 UTC
[tvm] branch unity-staging updated (8843e73a6e -> 5c8b7af59c)
This is an automated email from the ASF dual-hosted git repository.
tqchen pushed a change to branch unity-staging
in repository https://gitbox.apache.org/repos/asf/tvm.git
discard 8843e73a6e [Unity] Fix lint during for upstream black
discard 6ffad3374e [Unity] Pattern-based rewriting for dataflow block (#14446)
discard 821a4fb40f [Unity][Graph matching] Clean up undo stack for parent and child nodes properly (#14440)
discard 228ab429c6 [Unity][Op][Docs] Update comment for `call_tir_dyn` (#14441)
discard 17b8c6cbae [Unity][Graph matching] Automatically add `used-by` constraints for `is_op` pattern (#14439)
discard 40e6ca73b1 [Unity] Remove non-deterministic behavior from graph pattern matching (#14417)
discard 8f567b5477 [Unity] Minor updates to DataFlowBlockRewrite (#14431)
discard 053609e64f [Unity][Fix] Annotate TIR op pattern could have no stores. (#14420)
discard 72231d2c34 [Unity] Include constant shapes in the profiler result (#14428)
discard 3163763e1a [Unity] Handle extern func calls in static memory planning (#14419)
discard 5951059dd8 [Unity][Fix] Copy over module attrs in FuseTIR (#14418)
discard 56d8d208a6 [Unity][Hexagon] Enable Relax VM for Hexagon (#14415)
discard 0a6a62f1e6 [Unity][Op] Expose scale in `R.nn.attention` and add its legalize op (#14412)
discard 9d1b2d5b21 [Unity] Fix getting shapes for cutlass BYOC kernels (#14411)
discard a85acb37a6 [Unity][Op] Conv1d (#14388)
discard e797baa915 [Unity][QNN][Hexagon]Support Relax Constants in the QNN TOPI operations (#14386)
discard a7b9347998 [Unity][Transform] Common Subexpression Elimination (#14361)
discard 58b581ee77 [Unity][TVMScript] Fix Shape Var occurrence in Tensor annotation (#14404)
discard af1bf15311 [Unity][Op] Add stop_lift_params (#14368)
discard 4c88e1ef4c [Unity] Support simple dynamic-shape-aware fusion (#14396)
discard 6540bc2f32 [Unity][Transform] SplitCallTIRByPattern and CUTLASS backend (#14274)
discard c43b7042a2 [Unity] Add missing #include <array> (#14383)
discard 63921a926d [Unity][VM] Add CUDA graph vm builtins (#14371)
discard bb961ca5ae [Unity] Also include output dtype in simt MathInstruction (#14372)
discard a80d07f5ce [Unity][Fix] Allow scalar layout initialization (#14370)
discard 83583d82fd [Unity][TVMScript] Update GlobalVar `checked_type_` when `emit_te` (#14367)
discard a1158ab679 [Unity] Add More Ops For FX Translator (#14348)
discard 1377c87dab [Unity][Fix] Infer Layout must support negative axes (#14365)
discard 7a21f00ed0 [Unity][Pass] Fix FuseOps error if there is no output of a given group (#14354)
discard 0dae4870e3 [Unity][WEB] Support async pipeline creation (#14362)
discard 0e2c944d0f [Unity] Add support to append relay op attrs in translator (#14356)
discard 367ba1eb1f [Unity][Transform] Fix AMP tests (#14360)
discard 6699a75247 [Unity][Transform] Introduce data-dependent operation of reshape and its constant folding (#14282)
discard b7e56dfc17 [Unity][Fix] Fix block memory plan to handle bool (#14357)
discard 3c6ec6a4ec [Unity][Transform] AMP out_dtype=float16 testcases (#14358)
discard 8b740442ba [Unity][BYOC] Check leaked intermediate variables in cutlass patterns (#14350)
discard dae16e88b1 [Unity] Support model kwargs in dynamo_capture_subgraph (#14349)
discard 0e5c55f5a0 [Unity][Frontend] FX exp and strided_slice fix (#14338)
discard 8e785f7fba [Unity][BYOC] Update testcases to follow recent changes (#14339)
discard 98846fdbad [Unity] Remove Python interface of RemoveUnusedFunction (#14336)
discard a2e3826fb5 [Unity][Pass] Reuse prior infra to implement more complete DCE (#14334)
discard e75838d0aa [Unity][Op] Fix Strided Slice Shape Inference (#14324)
discard 90cf347c01 [Unity][Transform] DefaultSchedule pass (#14266)
discard b898379f67 [Unity][Lint] Fix cpplint casting (#14333)
discard 41c7761c40 [Unity][Transform] Automatic Mixed Precision (#14242)
discard 543bc300f7 [Unity][Transform] Simple Dead Code Elimination (#14262)
discard 226745f0d7 [Unity][Transform] Automatic Layout Conversion (#14257)
discard fd34a2f5b6 [Unity][TOPI] fp16 LayerNorm & GroupNorm (#14264)
discard e47cd3634f [Unity][Contrib] Introduce several features of cutlass profiler (#14275)
discard 932e702094 [Unity][Transform] Enhance RewriteDataflowReshape transform (#14265)
discard 10774dc447 [Unity][BYOC] Improve expressiveness of the pattern check function in FuseOpsByPattern (#14310)
discard 0b675407f6 [Unity][BYOC] Support matmul + residual block fusion in CUTLASS BYOC (#14317)
discard 89f54c9d89 [Unity] Support pattern-based rewriting (#14312)
discard a61db82f98 [Unity][Web] WebGPU explicit max buffer size (#14321)
discard 46e78ed951 [Unity][Op] Enable special dimension value 0 in reshape (#14311)
discard 79add7bb73 [Unity][Pass] Add a pass to alter the TIR implementation of an operator (#14215)
discard c2b19be187 [Unity][DEBUG] Add Instrument (#14302)
discard ed01b9d90e [Unity][Op] Cumsum (#14297)
discard 72513b7287 [Unity] Fix StructInfo Infer for `vm.alloc_tensor` (#14283)
discard c513cb9c70 [Unity] Mark tests that need python3.8 compact.
discard 360f7566ac [TVMScript][Unity] Improve PyLint Compatibility (#14276)
discard 59cc3219b2 [Unity][ci] Use CPU-SMALL instances (#14256)
discard 5963846dbf [Unity] Introduce call_dps_packed (#14183)
discard bb9e9d0e55 [Unity] Consider target context for Relay to Relax conversion (#14269)
discard 3e9c8d327b [Unity][Frontend] Import `tanh` and fix `layer_norm` (#14247)
discard 2ca9fa4180 [Unity][BYOC] Add conv2d and residual block patterns for Relax cutlass BYOC (#14252)
discard eeac231d76 [Unity] Allow user defined func attrs in emit_te (#14255)
discard 02bf2439af [Unity][Op] Add repeat, tile, conv2d_transpose, avg_pool2d (#14238)
discard edaa4529b3 [Unity][Op][Tweak] Improve `StructInfo` inference for `shape_of` (#14243)
discard 3069fa0dd0 [Unity][WEB] Improve ndarray cache (#14236)
discard 866e2379c5 [Unity][WEB] Update text prompts for syntactical correctness (#14237)
discard 500042f9ae [Unity][TVMScript] Fix prim_func lost issue in relax.emit_te (#14189)
discard 8bdfac16e0 [Unity][TVMScript] Enable Context-Aware Parsing (#14234)
discard ada8675944 [Unity][Bugfix] Do not include `PrimFunc`s in the dependency graph when checking for recursion (#14228)
discard 451f95554b [Unity][Transform] SimplifyNormInference (#14221)
discard d4a4e81623 [Unity] Improve implementation of FuseOps (#14229)
discard 1bf1dd0578 [Unity] ensure memory.alloc_tensor/storage roundtrippable (#14226)
discard ce22d8edc9 [Unity][WEB] Simplify WebGPU Codegen per spec (#14225)
discard bbe04cb457 [Unity][Transform] Memory plan across the IRModule (#14220)
discard 17837b471c [Unity][BYOC] Add dynamic shape support to CUTLASS matmul (#14216)
discard 1bb59ff0d5 [Unity][Frontend] from_fx keeps parameters in order (#14214)
discard 6d2db12fed [Unity][WEB] Improve webgpu codegen options to skip readonly (#14213)
discard d013e834a9 [Unity][Frontend] FX translator supports unwrapping unit return tuple (#14212)
discard a3a593b492 [Unity][Frontend] Attach imported model weights, deprecate ImporterOutput (#14211)
discard 313faa7292 [Unity] Introduce Default GPU Schedule Pass (#14182)
discard 16ca7ded54 [Unity][Frontend] FX translator support torch.baddbmm (#14202)
discard b29a518bee [Unity][TIR][Pass] ForceNarrowIndexToInt32 (#14203)
discard f4f122589a [Unity][Fix] FX translating dtype (#14201)
discard 82de2b24b4 [Unity][Frontend] FX translator returning weights with `keep_params_as_input` (#14197)
discard 040dec513d [Unity][Frontend] FX translator supporting more ops (#14196)
discard 60a23a5e6f [Unity][Op] Legalize `round`, `floor`, `ceil`, `sign` (#14198)
discard a54fcbef49 [Unity][Op] Argmax and argmin (#14195)
discard 30f5c5a14f [Unity][Op] Group normalization (#14194)
discard 87aea68e60 [Unity][Transform] LiftTransformParams handling multiple functions (#14192)
discard 930df87cb5 [Unity][WEBGPU] Codegen improvements and WebRuntime (#14187)
discard 59ec211dd1 [Unity][OP] Add an operator for fused multi head attention (#14150)
discard f3e391ae3a [Unity][Analysis] Restore Python bindings for var analyses (#14180)
discard fa29543f2d [Unity][Op] Full support of Relax op `power` (#14171)
discard 0640642ee6 [Unity][BYOC] Add batch matmul support to Relax CUTLASS BYOC (#14166)
discard d266b3b6ca [Unity][Analysis] Analysis for detecting recursion in Relax (#14149)
discard eec815c584 [Unity] Add bind_constants option to FuseOpsByPattern (#14151)
discard 250cf734bd [Unity][BYOC] Use Relax legalize + CPU build for reference in tests (#14162)
discard a5b0555dc3 [Unity][Analysis] Checking function return struct info in well-formed check (#14155)
discard 6ac387cfea [Unity][Pass] Support Symbolic Shape Deduction during BindParam (#14154)
discard d83d4e52e3 [Unity][Debugging] AST printer (#14152)
discard 56e8114043 [Unity][Pass] Enhance constant folding to fold relax ops by evaluating them. (#14146)
discard 959f5b3a16 [Unity][Legalize] Fix Scalar Constant Legalization (#14127)
discard 008fce3082 [Unity] Add callback to FuseOpsByPattern to check match result is accepted (#14109)
discard 16f81969bd [Unity][BYOC] Assign group to unused bindings and ignroe PrimFunc (#14139)
discard ed777d5097 [Unity][TVMScript] emit_te sugar (#14123)
discard 4c27f82564 [Unity][BYOC] Add transposed matmul support to Relax CUTLASS BYOC (#14128)
discard ff95127cc0 [Unity] Add Global info (#14132)
discard 44f1bfedb5 [Unity][WEB] Relax vm on web runtime (#14131)
discard 209ee04928 [Unity][BlockBuilder] Add `name_hint` argument for `emit` and `emit_output` (#14126)
discard d537bcd977 [Unity][Fix] Fix bug in MergeCompositeFunctions (#14117)
discard c2c84991ae [Unity] Update tests again to adapt to latest TVMScript syntax (#14115)
discard 42a06d878d [Unity][BYOC]Add relax backend pattern registry (#14106)
discard 6a7bdf57ac [Unity] Remove attributes of relax.print, assert and unique (#14101)
discard 9aa6926f97 [Unity][Layout] Add layout transformation analysis for PrimFunc (#14066)
discard 5f8e0aa2ad [Unity] Relax Recursive function (#14092)
discard ea0d012123 [Unity] Lower `shape_of` to a builtin (#14093)
discard 6235c6d652 [Unity] Fix typo in the comment (#14096)
discard a8fd3ffb41 [Unity][Relax] Set Shape Function to Be Host Function (#14090)
discard 9d6c86680c [Unity] Refactor Relax Build JIT UX (#14088)
discard 5aa8547300 [Unity][Fix][Pass] FoldConstant with DCE in dataflow block (#14087)
discard fadf377c83 [Unity][Analysis] TIR pattern kind analysis for multi-buffer write block (#14075)
discard 85b6f66508 [Unity][Op] `log_softmax` and `cross_entropy_with_logits` (#14083)
discard f0399994b5 [Unity][BYOC] Add DNNL backend (#14082)
discard 6d3a9b3032 [Unity][BYOC] Add CUTLASS backend (#14081)
discard f015a97ac8 [Unity] Add testcases for `expr_args_converter` (#14080)
discard 2fbf8ad3ba [Unity][Pass] Canonicalize Bindings (#14079)
discard a816838a99 [Unity][BYOC][Pass] RunCodegen and TensorRT (#14078)
discard 8615b16565 [Unity][Transform] Add LiftTransformParams pass (#14069)
discard 9f7a040c0e [Unity][Frontend] Annotate number of non-static input of FX function (#14067)
discard 2994125e3a [Unity][BYOC] Add pass to merge composite functions to offload large subgraphs (#14062)
discard 3f18864f78 [Unity][Pass] Remove Unused Function (#14061)
discard 197d3a8d66 [Unity][Fix][Pass] Fix FuseOps for lack graph edges (#14058)
discard 71e688c952 [Unity] Relax op: collapse sum (#14059)
discard 5a65b79e6f [Unity][BYOC] Add pattern-based partitioning pass (#14054)
discard f02d567791 [Unity][VM] Add per-op profiling support (#14053)
discard eb660f2377 [Unity][TVMScript] Overload `__neg__` for relax expr (#14045)
discard 8016f30e4b [Unity][Pass] FuseOps FuseTIR fixes (#14044)
omit 969047780a [Unity] Statement rewriter for DataflowBlock (#14043)
omit 0525e05aaf [Unity] Relax dataflow pattern language (matching) (#14041)
omit c45b1a6990 [Unity] Update tests to adapt to latest TVMScript syntax (#14039)
omit db1bf6b039 [Unity] Disallow inline prim_func in relax IR (#14040)
omit 9879fbbd0b [Unity][Pass] Block-level static memory planning (#14038)
omit 8d575f2a73 [Unity] Initial PyTorch Frontend (#14037)
omit 317634bc19 [Unity][Op] Add ShapeExpr Tests for Reshape Op (#14035)
omit db588383bf [Unity][Pass] Operator legalization (#14029)
omit bd8fb78ac4 [Unity][TVMScript] Move tir/relax import in script out of __init__.py (#14033)
omit a50cdd06e3 [Unity][Pass] Wellformed Analysis (#14032)
omit 2cc122cd24 [Unity][BlockBuilder] CallTE convert PrimValue args (#14028)
omit d8a6d1d826 [Unity][Pass] Normalize Pass (#14031)
omit 44b636f9be [Unity] Relay -> Relax translator (#14026)
omit 792d7c5eda [Unity][Pass][TuningAPI] Introduce TuningAPI and MetaSchedule pass (#14014)
omit f81e198ed4 [Unity][Pass] BindParams pass, FoldConstant pass (#14016)
omit 5a6579e1b0 [Unity][VM] Supporting "compiled" exec mode. (#14015)
omit 7de9c82626 [Unity][Pass] LambdaLift pass (#14012)
omit 18ade5f8ba [Unity][Pass] Operator Fusion Passes (#14001)
omit 26b4439cf1 [Unity] NestedMsg Support utility (#13995)
omit f64e91c6da [Unity] Relax op: manipulation (#13989)
omit 044f3bbc41 [Unity] Relax op: search (#13992)
omit bf6e2a9ef6 [Unity] Relax op: linear algebra (#13988)
omit 444d420450 [Unity] Relax op: creation (#13984)
omit 5b3239ad4d [Unity] Relax op: neural networks (#13993)
omit ec110c6023 [Unity] Relax op: statistical (#13991)
omit c534c9c7b3 [Unity] Relax op: arithmetic, comparison (#13983)
omit a9a561b472 [Unity] Relax op: image (#13994)
omit 9a9e4a7823 [Unity] Relax op: set (#13990)
omit 4dfa36202b [Unity] Relax op: datatype (#13986)
omit caddedb418 [Unity] Relax op: index (#13987)
omit 4051a69cec [Unity][TVMScript] Use explicit `R.shape` in TVMScript (#13979)
omit dbedbb25ba [Unity] e2e Relax minimum build flow (#13961)
omit afe71010ef [Unity] Relax VM shape lowering pass (#13956)
omit 7f1e1f5528 [Unity] Relax VM codegen (#13954)
omit a2d032494f [Unity] Relax TVMScript Printer (#13944)
omit 63de0dacbd [Unity] Relax TVMScript Parser. (#13932)
omit 23a7cd1a21 [Unity] Relax BlockBuilder and ExprMutator (#13926)
omit a7086616d7 [Unity] Basic StructInfo Analysis and Expr construction (#13916)
omit f6b68ab7fd [Unity][CI] Unity specific jenkins setup (do not upstream to main) (#13910)
omit 2bb2e4bf75 [Unity][IR] First-class StructInfo (#13907)
omit 0117a28d22 [Unity] Relax expressions and types (#13901)
omit a27451755f [Unity] Relax VM (#13878)
new 70d80cc126 [Unity] Relax VM (#13878)
new 25e6174770 [Unity] Relax expressions and types (#13901)
new 51cf110c0c [Unity][IR] First-class StructInfo (#13907)
new 40c43108c2 [Unity][CI] Unity specific jenkins setup (do not upstream to main) (#13910)
new e68ef58c84 [Unity] Basic StructInfo Analysis and Expr construction (#13916)
new aa59d86ac7 [Unity] Relax BlockBuilder and ExprMutator (#13926)
new 540ba28f5c [Unity] Relax TVMScript Parser. (#13932)
new 334fd8be51 [Unity] Relax TVMScript Printer (#13944)
new 574084d340 [Unity] Relax VM codegen (#13954)
new 518a7bdd96 [Unity] Relax VM shape lowering pass (#13956)
new b3d913a737 [Unity] e2e Relax minimum build flow (#13961)
new caad7cdf48 [Unity][TVMScript] Use explicit `R.shape` in TVMScript (#13979)
new 903628f106 [Unity] Relax op: index (#13987)
new 0319939e47 [Unity] Relax op: datatype (#13986)
new 11e5e1d6d4 [Unity] Relax op: set (#13990)
new 128042204c [Unity] Relax op: image (#13994)
new 64f281749a [Unity] Relax op: arithmetic, comparison (#13983)
new be4373a1d2 [Unity] Relax op: statistical (#13991)
new fa3cfbd60e [Unity] Relax op: neural networks (#13993)
new 45e6d0bade [Unity] Relax op: creation (#13984)
new 206762f0e6 [Unity] Relax op: linear algebra (#13988)
new 088afc5d54 [Unity] Relax op: search (#13992)
new 4430d26394 [Unity] Relax op: manipulation (#13989)
new 471ff84280 [Unity] NestedMsg Support utility (#13995)
new 604754a6c8 [Unity][Pass] Operator Fusion Passes (#14001)
new 50f616c95c [Unity][Pass] LambdaLift pass (#14012)
new 0a0166e81f [Unity][VM] Supporting "compiled" exec mode. (#14015)
new 660a1f361c [Unity][Pass] BindParams pass, FoldConstant pass (#14016)
new 30a5669b8a [Unity][Pass][TuningAPI] Introduce TuningAPI and MetaSchedule pass (#14014)
new b792751aaa [Unity] Relay -> Relax translator (#14026)
new 554e609aca [Unity][Pass] Normalize Pass (#14031)
new 477d011093 [Unity][BlockBuilder] CallTE convert PrimValue args (#14028)
new f61c330ec6 [Unity][Pass] Wellformed Analysis (#14032)
new 4df587aba6 [Unity][TVMScript] Move tir/relax import in script out of __init__.py (#14033)
new d2b8ea330d [Unity][Pass] Operator legalization (#14029)
new e0c6ca0383 [Unity][Op] Add ShapeExpr Tests for Reshape Op (#14035)
new e15f59bcf0 [Unity] Initial PyTorch Frontend (#14037)
new e8ebd68e92 [Unity][Pass] Block-level static memory planning (#14038)
new 2e1d85149b [Unity] Disallow inline prim_func in relax IR (#14040)
new 60d3796e20 [Unity] Update tests to adapt to latest TVMScript syntax (#14039)
new 7d70af3723 [Unity] Relax dataflow pattern language (matching) (#14041)
new 44b05d68bb [Unity] Statement rewriter for DataflowBlock (#14043)
new 7a71e53b69 [Unity][Pass] FuseOps FuseTIR fixes (#14044)
new e7cc2876ec [Unity][TVMScript] Overload `__neg__` for relax expr (#14045)
new 5239022d97 [Unity][VM] Add per-op profiling support (#14053)
new 38178a3298 [Unity][BYOC] Add pattern-based partitioning pass (#14054)
new a7e936fca1 [Unity] Relax op: collapse sum (#14059)
new 7ce0a519e7 [Unity][Fix][Pass] Fix FuseOps for lack graph edges (#14058)
new 34d5944135 [Unity][Pass] Remove Unused Function (#14061)
new 2c5ccfa7a6 [Unity][BYOC] Add pass to merge composite functions to offload large subgraphs (#14062)
new 7f5f816084 [Unity][Frontend] Annotate number of non-static input of FX function (#14067)
new 73fe88d04a [Unity][Transform] Add LiftTransformParams pass (#14069)
new b137d22ed4 [Unity][BYOC][Pass] RunCodegen and TensorRT (#14078)
new 38f920d480 [Unity][Pass] Canonicalize Bindings (#14079)
new 2a51dcae46 [Unity] Add testcases for `expr_args_converter` (#14080)
new 026217d998 [Unity][BYOC] Add CUTLASS backend (#14081)
new 4365d1e29d [Unity][BYOC] Add DNNL backend (#14082)
new 2409be1934 [Unity][Op] `log_softmax` and `cross_entropy_with_logits` (#14083)
new 9caab4d9d8 [Unity][Analysis] TIR pattern kind analysis for multi-buffer write block (#14075)
new 872c9f23c6 [Unity][Fix][Pass] FoldConstant with DCE in dataflow block (#14087)
new 6b87e35617 [Unity] Refactor Relax Build JIT UX (#14088)
new 3959707f19 [Unity][Relax] Set Shape Function to Be Host Function (#14090)
new 79fe0a22aa [Unity] Fix typo in the comment (#14096)
new 779c54d9ad [Unity] Lower `shape_of` to a builtin (#14093)
new 3173f2855d [Unity] Relax Recursive function (#14092)
new 62a4c735fb [Unity][Layout] Add layout transformation analysis for PrimFunc (#14066)
new 1bf8b63648 [Unity] Remove attributes of relax.print, assert and unique (#14101)
new 30d204bb4a [Unity][BYOC]Add relax backend pattern registry (#14106)
new a1f8d72178 [Unity] Update tests again to adapt to latest TVMScript syntax (#14115)
new 6a9caf40e7 [Unity][Fix] Fix bug in MergeCompositeFunctions (#14117)
new 3515f95ba0 [Unity][BlockBuilder] Add `name_hint` argument for `emit` and `emit_output` (#14126)
new c1f304497e [Unity][WEB] Relax vm on web runtime (#14131)
new aaa457d304 [Unity] Add Global info (#14132)
new 837dd6c228 [Unity][BYOC] Add transposed matmul support to Relax CUTLASS BYOC (#14128)
new a4b1ec097c [Unity][TVMScript] emit_te sugar (#14123)
new be170bb3d2 [Unity][BYOC] Assign group to unused bindings and ignroe PrimFunc (#14139)
new 77f4c719f3 [Unity] Add callback to FuseOpsByPattern to check match result is accepted (#14109)
new 1b6d104f68 [Unity][Legalize] Fix Scalar Constant Legalization (#14127)
new 788c12f84f [Unity][Pass] Enhance constant folding to fold relax ops by evaluating them. (#14146)
new f17a7285bb [Unity][Debugging] AST printer (#14152)
new c78e68bf4d [Unity][Pass] Support Symbolic Shape Deduction during BindParam (#14154)
new ab5aaa0d1b [Unity][Analysis] Checking function return struct info in well-formed check (#14155)
new 89cff055d3 [Unity][BYOC] Use Relax legalize + CPU build for reference in tests (#14162)
new 2ca340a30b [Unity] Add bind_constants option to FuseOpsByPattern (#14151)
new 4fe3a6d1f4 [Unity][Analysis] Analysis for detecting recursion in Relax (#14149)
new 2de852f203 [Unity][BYOC] Add batch matmul support to Relax CUTLASS BYOC (#14166)
new 8de0ff78f5 [Unity][Op] Full support of Relax op `power` (#14171)
new 9797924d54 [Unity][Analysis] Restore Python bindings for var analyses (#14180)
new 30dc2365e7 [Unity][OP] Add an operator for fused multi head attention (#14150)
new 279317d4cd [Unity][WEBGPU] Codegen improvements and WebRuntime (#14187)
new 3f66edc8ed [Unity][Transform] LiftTransformParams handling multiple functions (#14192)
new 283a3db675 [Unity][Op] Group normalization (#14194)
new c25659d88f [Unity][Op] Argmax and argmin (#14195)
new 2f114d28b3 [Unity][Op] Legalize `round`, `floor`, `ceil`, `sign` (#14198)
new 012923572a [Unity][Frontend] FX translator supporting more ops (#14196)
new a115d03e76 [Unity][Frontend] FX translator returning weights with `keep_params_as_input` (#14197)
new 27b6fa7cb4 [Unity][Fix] FX translating dtype (#14201)
new 6ab0abc08d [Unity][TIR][Pass] ForceNarrowIndexToInt32 (#14203)
new 3f4417c6db [Unity][Frontend] FX translator support torch.baddbmm (#14202)
new 2d562f8926 [Unity] Introduce Default GPU Schedule Pass (#14182)
new 97dc05ee18 [Unity][Frontend] Attach imported model weights, deprecate ImporterOutput (#14211)
new 033fc0e34b [Unity][Frontend] FX translator supports unwrapping unit return tuple (#14212)
new 9f9cf2967c [Unity][WEB] Improve webgpu codegen options to skip readonly (#14213)
new 8f2153cfb4 [Unity][Frontend] from_fx keeps parameters in order (#14214)
new 30848742cb [Unity][BYOC] Add dynamic shape support to CUTLASS matmul (#14216)
new 73b03dadb9 [Unity][Transform] Memory plan across the IRModule (#14220)
new 780a579c72 [Unity][WEB] Simplify WebGPU Codegen per spec (#14225)
new 3345aeda4a [Unity] ensure memory.alloc_tensor/storage roundtrippable (#14226)
new 30a9403729 [Unity] Improve implementation of FuseOps (#14229)
new 9bc1ecc672 [Unity][Transform] SimplifyNormInference (#14221)
new a6b37b80ce [Unity][Bugfix] Do not include `PrimFunc`s in the dependency graph when checking for recursion (#14228)
new 3e03ca5abe [Unity][TVMScript] Enable Context-Aware Parsing (#14234)
new cea0161bcb [Unity][TVMScript] Fix prim_func lost issue in relax.emit_te (#14189)
new 14a7de57b3 [Unity][WEB] Update text prompts for syntactical correctness (#14237)
new 663c8211a4 [Unity][WEB] Improve ndarray cache (#14236)
new c7c4dee36b [Unity][Op][Tweak] Improve `StructInfo` inference for `shape_of` (#14243)
new 633c9002e3 [Unity][Op] Add repeat, tile, conv2d_transpose, avg_pool2d (#14238)
new 07f7fc0a9c [Unity] Allow user defined func attrs in emit_te (#14255)
new 04d6d15984 [Unity][BYOC] Add conv2d and residual block patterns for Relax cutlass BYOC (#14252)
new df9d084553 [Unity][Frontend] Import `tanh` and fix `layer_norm` (#14247)
new 71c23b7a11 [Unity] Consider target context for Relay to Relax conversion (#14269)
new 70386ecc91 [Unity] Introduce call_dps_packed (#14183)
new c676d1392f [Unity][ci] Use CPU-SMALL instances (#14256)
new 121e1e7a03 [TVMScript][Unity] Improve PyLint Compatibility (#14276)
new 6be5c19c9c [Unity] Mark tests that need python3.8 compact.
new 68827fa1d6 [Unity] Fix StructInfo Infer for `vm.alloc_tensor` (#14283)
new e3f975543f [Unity][Op] Cumsum (#14297)
new 67e183c33f [Unity][DEBUG] Add Instrument (#14302)
new f96e09a9ce [Unity][Pass] Add a pass to alter the TIR implementation of an operator (#14215)
new a46fd90478 [Unity][Op] Enable special dimension value 0 in reshape (#14311)
new e0c4456f50 [Unity][Web] WebGPU explicit max buffer size (#14321)
new 4bfbd3a703 [Unity] Support pattern-based rewriting (#14312)
new 2f1642f128 [Unity][BYOC] Support matmul + residual block fusion in CUTLASS BYOC (#14317)
new 778b343ed9 [Unity][BYOC] Improve expressiveness of the pattern check function in FuseOpsByPattern (#14310)
new 7c5a570dc1 [Unity][Transform] Enhance RewriteDataflowReshape transform (#14265)
new b10e2a2a31 [Unity][Contrib] Introduce several features of cutlass profiler (#14275)
new f67657fe09 [Unity][TOPI] fp16 LayerNorm & GroupNorm (#14264)
new 5cd30b00ef [Unity][Transform] Automatic Layout Conversion (#14257)
new 59746ea9b0 [Unity][Transform] Simple Dead Code Elimination (#14262)
new 371c3a2f2f [Unity][Transform] Automatic Mixed Precision (#14242)
new 691983096c [Unity][Lint] Fix cpplint casting (#14333)
new 5229827888 [Unity][Transform] DefaultSchedule pass (#14266)
new 375a2921ea [Unity][Op] Fix Strided Slice Shape Inference (#14324)
new 920f98d8f9 [Unity][Pass] Reuse prior infra to implement more complete DCE (#14334)
new f4a451d273 [Unity] Remove Python interface of RemoveUnusedFunction (#14336)
new 050562bfe9 [Unity][BYOC] Update testcases to follow recent changes (#14339)
new 7534531a6e [Unity][Frontend] FX exp and strided_slice fix (#14338)
new 021ce24d0c [Unity] Support model kwargs in dynamo_capture_subgraph (#14349)
new fe34a55bee [Unity][BYOC] Check leaked intermediate variables in cutlass patterns (#14350)
new 9ae5895aae [Unity][Transform] AMP out_dtype=float16 testcases (#14358)
new cd48ba9e33 [Unity][Fix] Fix block memory plan to handle bool (#14357)
new 57b90700d2 [Unity][Transform] Introduce data-dependent operation of reshape and its constant folding (#14282)
new 27f0ad3230 [Unity][Transform] Fix AMP tests (#14360)
new 20cc2ff300 [Unity] Add support to append relay op attrs in translator (#14356)
new 79492b81d9 [Unity][WEB] Support async pipeline creation (#14362)
new 7974e0f816 [Unity][Pass] Fix FuseOps error if there is no output of a given group (#14354)
new b5cdcec5d6 [Unity][Fix] Infer Layout must support negative axes (#14365)
new eb7b020c22 [Unity] Add More Ops For FX Translator (#14348)
new 14f7b5af61 [Unity][TVMScript] Update GlobalVar `checked_type_` when `emit_te` (#14367)
new a0b28c486e [Unity][Fix] Allow scalar layout initialization (#14370)
new 9253f0fef1 [Unity] Also include output dtype in simt MathInstruction (#14372)
new 8d8ced72a1 [Unity][VM] Add CUDA graph vm builtins (#14371)
new 983b569325 [Unity] Add missing #include <array> (#14383)
new d6efe9f042 [Unity][Transform] SplitCallTIRByPattern and CUTLASS backend (#14274)
new f7f45dcc8e [Unity] Support simple dynamic-shape-aware fusion (#14396)
new 34b32cdf68 [Unity][Op] Add stop_lift_params (#14368)
new 6eb12ff85a [Unity][TVMScript] Fix Shape Var occurrence in Tensor annotation (#14404)
new ae3cf5f5f1 [Unity][Transform] Common Subexpression Elimination (#14361)
new 6818799e74 [Unity][QNN][Hexagon]Support Relax Constants in the QNN TOPI operations (#14386)
new a8205ed0c1 [Unity][Op] Conv1d (#14388)
new 906f69056d [Unity] Fix getting shapes for cutlass BYOC kernels (#14411)
new 5426085988 [Unity][Op] Expose scale in `R.nn.attention` and add its legalize op (#14412)
new 7f44914317 [Unity][Hexagon] Enable Relax VM for Hexagon (#14415)
new 94b9bbc6bc [Unity][Fix] Copy over module attrs in FuseTIR (#14418)
new 4d29a7cff8 [Unity] Handle extern func calls in static memory planning (#14419)
new e424eea3dd [Unity] Include constant shapes in the profiler result (#14428)
new b1346b97a9 [Unity][Fix] Annotate TIR op pattern could have no stores. (#14420)
new beb25a701c [Unity] Minor updates to DataFlowBlockRewrite (#14431)
new 22c363a72a [Unity] Remove non-deterministic behavior from graph pattern matching (#14417)
new 61f7edc1af [Unity][Graph matching] Automatically add `used-by` constraints for `is_op` pattern (#14439)
new ad4bfbc5f9 [Unity][Op][Docs] Update comment for `call_tir_dyn` (#14441)
new 0a4ce4dfc3 [Unity][Graph matching] Clean up undo stack for parent and child nodes properly (#14440)
new 52f6e639de [Unity] Pattern-based rewriting for dataflow block (#14446)
new 5c8b7af59c [Unity] Fix lint during for upstream black
This update added new revisions after undoing existing revisions.
That is to say, some revisions that were in the old version of the
branch are not in the new version. This situation occurs
when a user --force pushes a change and generates a repository
containing something like this:
* -- * -- B -- O -- O -- O (8843e73a6e)
\
N -- N -- N refs/heads/unity-staging (5c8b7af59c)
You should already have received notification emails for all of the O
revisions, and so the following emails describe only the N revisions
from the common base, B.
Any revisions marked "omit" are not gone; other references still
refer to them. Any revisions marked "discard" are gone forever.
The 184 revisions listed above as "new" are entirely new to this
repository and will be described in separate emails. The revisions
listed as "add" were already present in the repository and have only
been added to this reference.
Summary of changes:
include/tvm/runtime/relax_vm/executable.h | 4 ++++
web/emcc/webgpu_runtime.cc | 2 ++
2 files changed, 6 insertions(+)