You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@tvm.apache.org by GitBox <gi...@apache.org> on 2020/09/25 00:45:31 UTC
[GitHub] [incubator-tvm] ZihengJiang commented on issue #6486: TVM v0.7 Release Note Candidate
ZihengJiang commented on issue #6486:
URL: https://github.com/apache/incubator-tvm/issues/6486#issuecomment-698659713
# Feature Improvement
## Accelerator and Microcontroller Support
- Cleanup legacy verilog code ([#4576](https://github.com/apache/incubator-tvm/pull/4576))
- uTVM support for ARM STM32F746XX boards ([#4274](https://github.com/apache/incubator-tvm/pull/4274))
- Add --runtime=c, remove micro_dev target, enable LLVM backend [#6145](https://github.com/apache/incubator-tvm/pull/6145)
## Arithmetic Analysis
- Linear system and equation solver ([#5171](https://github.com/apache/incubator-tvm/pull/5171))
- Inequalities solver [#5618](https://github.com/apache/incubator-tvm/pull/5618)
- Improve IntervalSet's floormod ([#5367](https://github.com/apache/incubator-tvm/pull/5367))
- Remove legacy const pattern functions ([#5387](https://github.com/apache/incubator-tvm/pull/5387))
- Handle likely in IRMutatorWithAnalyzer [#5665](https://github.com/apache/incubator-tvm/pull/5665)
- ExtendedEuclidean merge impl to int_operator [#5625](https://github.com/apache/incubator-tvm/pull/5625)
- Rewrite simplify fix for Vectorized Cooperative Fetching [#5924](https://github.com/apache/incubator-tvm/pull/5924)
## AutoTVM and Graph Tuner
- Adding ROCM schedules for TOPI ([#4507](https://github.com/apache/incubator-tvm/pull/4507))
- NHWC conv2d schedule templates for ARM ([#3859](https://github.com/apache/incubator-tvm/pull/3859))
- Use VM compile to extract autotvm tasks [#4328](https://github.com/apache/incubator-tvm/issues/4328)
- Download fallback schedule file if it does not exist [#4671](https://github.com/apache/incubator-tvm/issues/4671)
- Ignore error when removing tmpdir [#4781](https://github.com/apache/incubator-tvm/issues/4781)
- Fix a bug in generating the search space [#4779](https://github.com/apache/incubator-tvm/issues/4779)
- Minor bug fixes in AutoTVM for QNN graphs [#4797](https://github.com/apache/incubator-tvm/issues/4797)
- Fix autotvm customized template [#5034](https://github.com/apache/incubator-tvm/pull/5034)
- Add opt out operator for has_multiple_inputs for graph tuner [#5000](https://github.com/apache/incubator-tvm/pull/5000)
- Customize SI prefix in logging ([#5411](https://github.com/apache/incubator-tvm/pull/5411))
- Update XGBoost verbosity option [#5649](https://github.com/apache/incubator-tvm/pull/5649)
- Support range in index based tuners [#4870](https://github.com/apache/incubator-tvm/pull/4870)
## BYOC
- [BYOC] Bind constant tuples in graph partitioner ([#5476](https://github.com/apache/incubator-tvm/pull/5476))
- [BYOC] Add support for composite functions in BYOC ([#5261](https://github.com/apache/incubator-tvm/pull/5261))
- [BYOC] Register pattern tables from external codegens ([#5262](https://github.com/apache/incubator-tvm/pull/5262))
- [BYOC] Enhance partitioning and external codegen ([#5310](https://github.com/apache/incubator-tvm/pull/5310))
- [BYOC] Refine AnnotateTarget and MergeCompilerRegion Passes ([#5277](https://github.com/apache/incubator-tvm/pull/5277))
- [BYOC] Use Non-Recursive Visitor/Mutator ([#5410](https://github.com/apache/incubator-tvm/pull/5410))
- [BYOC] Refine DNNL Codegen ([#5288](https://github.com/apache/incubator-tvm/pull/5288))
- [BYOC] Add example of Composite + Annotate for DNNL fused op ([#5272](https://github.com/apache/incubator-tvm/pull/5272))
- [BYOC] Prevent duplicate outputs in subgraph Tuple ([#5320](https://github.com/apache/incubator-tvm/pull/5320))
## Codegen
- Intrinsic dispatching with OCML instead of LLVM for ROCm ([#4499](https://github.com/apache/incubator-tvm/pull/4499))
- Make target codegen take IRModule and PrimFunc. [#5107](https://github.com/apache/incubator-tvm/pull/5107)
- Enhance CUDA codegen for SelectNode [#4983](https://github.com/apache/incubator-tvm/pull/4983)
- Vectorization for intrinsics [#5101](https://github.com/apache/incubator-tvm/pull/5101)
- [LLVM] Do not use x86_vcvtph2ps_256 intrinsic with LLVM 11+ ([#5267](https://github.com/apache/incubator-tvm/pull/5267))
- [LLVM] Use llvm::ElementCount with LLVM 11+ when creating vectors ([#5265](https://github.com/apache/incubator-tvm/pull/5265))
- [LLVM] Use llvm::FunctionCallee in IRBuilder::CreateCall with LLVM 11+ ([#5338](https://github.com/apache/incubator-tvm/pull/5338))
- [LLVM] Include Support/Host.h for declaration of getDefaultTargetTriple ([#5268](https://github.com/apache/incubator-tvm/pull/5268))
- [LLVM] Replace calls to Type::getVectorNumElements ([#5398](https://github.com/apache/incubator-tvm/pull/5398))
- [LLVM] Use ArrayRef in calls to CreateShuffleVector ([#5399](https://github.com/apache/incubator-tvm/pull/5399))
- [LLVM] Use llvm::Align with LLVM 11+ to avoid warnings ([#5264](https://github.com/apache/incubator-tvm/pull/5264))
- [CodeGen] Cleanup generated code (#5424)
- Rename target_id => target_kind [#6199](https://github.com/apache/incubator-tvm/pull/6199)
- 64-bit RPi4b target [#6211](https://github.com/apache/incubator-tvm/pull/6211)
- Creating Target from JSON-like Configuration [#6218](https://github.com/apache/incubator-tvm/pull/6218)
- Add python binding to new JSON target construction [#6315](https://github.com/apache/incubator-tvm/pull/6315)
- Use target class in all codegens [#6347](https://github.com/apache/incubator-tvm/pull/6347)
- Initial support for Hexagon codegen [#6261](https://github.com/apache/incubator-tvm/pull/6261)
- Add --runtime=c, remove micro_dev target, enable LLVM backend [#6145](https://github.com/apache/incubator-tvm/pull/6145)
- Add tvm::support::hexdump() debug utility [#6154](https://github.com/apache/incubator-tvm/pull/6154)
- Adding AMD codegen unit tests ([#4509](https://github.com/apache/incubator-tvm/pull/4509))
- Support cuda tensorcore subbyte int data type in auto tensorcore [#4546](https://github.com/apache/incubator-tvm/pull/4546)
- Handle empty LLVMModule in GetFunction [#5146](https://github.com/apache/incubator-tvm/pull/5146)
- Support int4/int8 conv2d tensor core with HWNC layout [#6121](https://github.com/apache/incubator-tvm/pull/6121)
## Dynamism Support
- Add shape function for zero, zeros_like, ones, ones_like ([#4448](https://github.com/apache/incubator-tvm/pull/4448)), tile ([#4441](https://github.com/apache/incubator-tvm/pull/4441/files))
- Support symbolic newshape for Reshape #5429
- Support symbolic TopK, Ones, Zeros and Full [#5459](https://github.com/apache/incubator-tvm/pull/5459)
- Add shape_of instruction [#5855](https://github.com/apache/incubator-tvm/pull/5855)
- symbolic max_output_size [#5844](https://github.com/apache/incubator-tvm/pull/5844)
- Dynamic TopK Op [#6008](https://github.com/apache/incubator-tvm/pull/6008)
- Dynamic broadcast_to, zeros, ones [#6007](https://github.com/apache/incubator-tvm/pull/6007)
- Add dynamic reshape grad [#6080](https://github.com/apache/incubator-tvm/pull/6080)
- Keep fixed dim when unifying dynamic shape [#5795](https://github.com/apache/incubator-tvm/pull/5795)
- OneHot operation [#6209](https://github.com/apache/incubator-tvm/pull/6209)
- Add Dynamic Resize Op [#6198](https://github.com/apache/incubator-tvm/pull/6198)
- Dynamic full operator [#6260](https://github.com/apache/incubator-tvm/pull/6260)
- Dynamic upsampling relay op [#6273](https://github.com/apache/incubator-tvm/pull/6273)
- Dynamic Tile Op [#5983](https://github.com/apache/incubator-tvm/pull/5983)
## Frontend and User Interface
- TFLite parser support for transpose_conv ([#4440](https://github.com/apache/incubator-tvm/pull/4440)), unpack ([#4447](https://github.com/apache/incubator-tvm/pull/4447))
- LLDB pretty printers for relay ([#4453](https://github.com/apache/incubator-tvm/pull/4453))
- ONNX to Relay converter op support: expand op ([#4483](https://github.com/apache/incubator-tvm/pull/4483))
- ONNX auto_pad in conv and convtranspose ([#4563](https://github.com/apache/incubator-tvm/pull/4563/))
- TF to Relay converter op support: bilinear and neighbour implementation refactor ([#4504](https://github.com/apache/incubator-tvm/pull/4504)), max_pool3d ([#4551](https://github.com/apache/incubator-tvm/pull/4551)), conv2d_transpose with “same” padding support for larger than 1x1 kernels ([#4484](https://github.com/apache/incubator-tvm/pull/4484))
- Remove unnecessary cast of constants in ONNX converter ([#4573](https://github.com/apache/incubator-tvm/pull/4573))
- Add support for tf.Keras networks in Relay Keras frontend [#4630](https://github.com/apache/incubator-tvm/issues/4630)
- Add conv3d [#4604](https://github.com/apache/incubator-tvm/issues/4604)
- Fix incorrect calculations in tf SLICE [#4518](https://github.com/apache/incubator-tvm/issues/4518)
- Dynamically calculate input_stats of any fake_quant range [#4789](https://github.com/apache/incubator-tvm/pull/4789)
- LSTM Support [#4825](https://github.com/apache/incubator-tvm/pull/4825)
- Add MIRROR_PAD operator [#4822](https://github.com/apache/incubator-tvm/pull/4822)
- use qnn helper function in softmax [#4840](https://github.com/apache/incubator-tvm/pull/4840)
- Add Resize op converter [#4838](https://github.com/apache/incubator-tvm/pull/4838)
- Add support for TFLite_Detection_PostProcess [#4543](https://github.com/apache/incubator-tvm/pull/4543)
- Fix tests for tflite unary elemwise operations [#4913](https://github.com/apache/incubator-tvm/pull/4913)
- GaussianDropout/Noise parsing support [#4928](https://github.com/apache/incubator-tvm/pull/4928)
- Add parser support for 'square' operator [#4915](https://github.com/apache/incubator-tvm/pull/4915)
- make_loss operator support [#4930](https://github.com/apache/incubator-tvm/pull/4930)
- Add parser support for l2_normalization [#4966](https://github.com/apache/incubator-tvm/pull/4966)
- ReadVariableOp operator support [#4952](https://github.com/apache/incubator-tvm/pull/4952)
- Check graph inputs match expected [#4992](https://github.com/apache/incubator-tvm/pull/4992)
- support multiply outputs [#4980](https://github.com/apache/incubator-tvm/pull/4980)
- TFLite: Using real image for QNN testing. [#4816](https://github.com/apache/incubator-tvm/pull/4816)
- TFLite: FLOOR_MOD & FLOOR_DIV support [#4971](https://github.com/apache/incubator-tvm/pull/4971)
- PyTorch: Upsampling op support and enable registering a user defined op conversion map [#4961](https://github.com/apache/incubator-tvm/pull/4961)
- PyTorch: fix unordered dictionary problem for python version under 3.6 [#4982](https://github.com/apache/incubator-tvm/pull/4982)
- Operator support NonZero [#5073](https://github.com/apache/incubator-tvm/pull/5073)
- Upsampling op support and enable registering a user defined op conversion map [#4961](https://github.com/apache/incubator-tvm/pull/4961)
- Check graph inputs match expected [#4992](https://github.com/apache/incubator-tvm/pull/4992)
- Add support for quantized models via QNN [#4977](https://github.com/apache/incubator-tvm/pull/4977)
- Add initial control flow support [#4964](https://github.com/apache/incubator-tvm/pull/4964)
- Remove FP32 piggy back and use QNN add/mul/concatenate [#5061](https://github.com/apache/incubator-tvm/pull/5061)
- Add missing upcast to uint8 avg_pool conversion [#5089](https://github.com/apache/incubator-tvm/pull/5089)
- Add initial 3D op support and test on Resnet 3D [#5075](https://github.com/apache/incubator-tvm/pull/5075)
- Fix conv2d conversion for group conv (group > 1 but != in channels) [#5132](https://github.com/apache/incubator-tvm/pull/5132)
- Add support for max_pool1d [#5142](https://github.com/apache/incubator-tvm/pull/5142)
- Add support for split [#5174](https://github.com/apache/incubator-tvm/pull/5174)
- FLOOR_MOD & FLOOR_DIV support [#4971](https://github.com/apache/incubator-tvm/pull/4971)
- Activation functions support [#4978](https://github.com/apache/incubator-tvm/pull/4978)
- Round op parsing support added [#5022](https://github.com/apache/incubator-tvm/pull/5022)
- DepthToSpace and SpaceToDepth support [#5041](https://github.com/apache/incubator-tvm/pull/5041)
- TOP_K op parser support [#5051](https://github.com/apache/incubator-tvm/pull/5051)
- ReadVariableOp operator support [#4952](https://github.com/apache/incubator-tvm/pull/4952)
- Support multiply outputs [#4980](https://github.com/apache/incubator-tvm/pull/4980)
- Reduce_any op parsing support [#4926](https://github.com/apache/incubator-tvm/pull/4926)
- TensorFlow Parser Control Flow Enhancement [#5020](https://github.com/apache/incubator-tvm/pull/5020)
- TensorFlow Frontend support with shared params [#5042](https://github.com/apache/incubator-tvm/pull/5042)
- Support for AddV2 in Relay Tensorflow frontend converter. [#5046](https://github.com/apache/incubator-tvm/pull/5046)
- conv3d frontend operator support [#5080](https://github.com/apache/incubator-tvm/pull/5080)
- Max_pool3d and Averagepool3d operator support [#5085](https://github.com/apache/incubator-tvm/pull/5085)
- Support for Atan/Atan2 in Relay Tensorflow frontend converter. [#5104](https://github.com/apache/incubator-tvm/pull/5104)
- Use leaky by default for LeakyReLU [#5192](https://github.com/apache/incubator-tvm/pull/5192)
- Conv3D ONNX support and conv3D_ncdhw x86 schedules [#4949](https://github.com/apache/incubator-tvm/pull/4949)
- Add support for FusedBatchNormV3 [#5065](https://github.com/apache/incubator-tvm/pull/5065)
- Activations for pytorch [#5194](https://github.com/apache/incubator-tvm/pull/5194)
- Dropouts And InstanceNorm support added [#5203](https://github.com/apache/incubator-tvm/pull/5203)
- [Frontend] Asymmetric padding of convolution support ([#4803](https://github.com/apache/incubator-tvm/pull/4803))
- [ONNX]Pool3d & upsample3d op support ([#5135](https://github.com/apache/incubator-tvm/pull/5135))
- Add TopK to ONNX Frontend ([#5441](https://github.com/apache/incubator-tvm/pull/5441))
- Add RoiAlign to Onnx frontend ([#5454](https://github.com/apache/incubator-tvm/pull/5454))
- [PYTORCH]AvgPool3d, MaxPool3d and Squeeze op support ([#5220](https://github.com/apache/incubator-tvm/pull/5220))
- [PYTORCH]celu, gelu, selu activations ([#5263](https://github.com/apache/incubator-tvm/pull/5263))
- [Pytorch]layernorm bug fix and testcase updated ([#5257](https://github.com/apache/incubator-tvm/pull/5257))
- [PYTORCH]LayerNorm support added ([#5249](https://github.com/apache/incubator-tvm/pull/5249))
- [RELAY-OP][PYTORCH]GroupNorm op support added ([#5358](https://github.com/apache/incubator-tvm/pull/5358))
- [TOPI][PYTORCH]Logical & Bitwise operator support ([#5341](https://github.com/apache/incubator-tvm/pull/5341))
- [PYTORCH]Tensor creation ops support ([#5347](https://github.com/apache/incubator-tvm/pull/5347))
- [RELAY][PYTORCH]cosh,sinh,log2,log10,log1p op support ([#5395](https://github.com/apache/incubator-tvm/pull/5395))
- [PYTORCH]Rsub, Embedded, OneHot ops support ([#5434](https://github.com/apache/incubator-tvm/pull/5434))
- [PYTORCH]Abs, Arange, Softplus ops ([#5295](https://github.com/apache/incubator-tvm/pull/5295))
- [RELAY][PYTORCH]isNan, isinf, isfinite, ceil, clamp, round ops ([#5316](https://github.com/apache/incubator-tvm/pull/5316))
- [PYTORCH]Activations for pytorch ([#5194](https://github.com/apache/incubator-tvm/pull/5194))
- [PYTORCH]Repeat, Reciprocal & Reshape Op support ([#5280](https://github.com/apache/incubator-tvm/pull/5280))
- [PYTORCH]Reduce_ops support added ([#5308](https://github.com/apache/incubator-tvm/pull/5308))
- [PYTORCH]Take, Topk op support ([#5332](https://github.com/apache/incubator-tvm/pull/5332))
- [PYTORCH]Dropouts And InstanceNorm support added ([#5203](https://github.com/apache/incubator-tvm/pull/5203))
- [PYTORCH]Unary Ops frontend support. ([#5378](https://github.com/apache/incubator-tvm/pull/5378))
- [Torch] Support Python list, more realistic recurrent networks ([#5306](https://github.com/apache/incubator-tvm/pull/5306))
- [PYTORCH]where, addcdiv, addcmul op support ([#5383](https://github.com/apache/incubator-tvm/pull/5383))
- [Torch] Add support for split ([#5174](https://github.com/apache/incubator-tvm/pull/5174))
- [Frontend][Torch] Fix up graph input handling ([#5204](https://github.com/apache/incubator-tvm/pull/5204))
- [FRONTEND][TFLITE]Logical not op support ([#5475](https://github.com/apache/incubator-tvm/pull/5475))
- [TFLITE]Hard Swish & MobilnetV3 model testing ([#5239](https://github.com/apache/incubator-tvm/pull/5239))
- [FRONTEND][TFLITE]Gather, StridedSlice op support added ([#4788](https://github.com/apache/incubator-tvm/pull/4788))
- [TFLITE] Match TFLite shape for SSD custom op ([#5473](https://github.com/apache/incubator-tvm/pull/5473))
- Factor out import of common tflite.Operator in tflite frontend. ([#5355](https://github.com/apache/incubator-tvm/pull/5355))
- [Frontend][TFLite] support for FILL and SPLIT_V operators ([#5330](https://github.com/apache/incubator-tvm/pull/5330))
- [Frontend][TFLite] L2_POOL_2D operator ([#5452](https://github.com/apache/incubator-tvm/pull/5452))
- [TFLite] Add config option to specify FlatBuffers location ([#5425](https://github.com/apache/incubator-tvm/pull/5425))
- [FRONTEND][TFLITE]Logical not op support ([#5475](https://github.com/apache/incubator-tvm/pull/5475))
- [TENSORFLOW]reduce ops updated ([#5180](https://github.com/apache/incubator-tvm/pull/5180))
- [FRONTEND][TENSORFLOW] Fix gather_nd indices ([#5279](https://github.com/apache/incubator-tvm/pull/5279))
- [Frontend][TensorFlow]Improve TensorFlow Static Shape Tensor Array ([#5243](https://github.com/apache/incubator-tvm/pull/5243))
- [KERAS]Minimum & AlphaDropout op support ([#5380](https://github.com/apache/incubator-tvm/pull/5380))
- [KERAS]Embedding layer ([#5444](https://github.com/apache/incubator-tvm/pull/5444))
- [FRONTEND][KERAS]Max_pool3d and Averagepool3d operator support ([#5085](https://github.com/apache/incubator-tvm/pull/5085))
- [RELAY][FRONTEND][CAFFE2] add Mul and ConvTranspose operator ([#5302](https://github.com/apache/incubator-tvm/pull/5302))
- [MXNET]DepthToSpace & SpaceToDepth Operator ([#5408](https://github.com/apache/incubator-tvm/pull/5408))
- [MXNET]broadcast and logical op support ([#5461](https://github.com/apache/incubator-tvm/pull/5461))
- [FRONTEND][MXNET] Use leaky by default for LeakyReLU ([#5192](https://github.com/apache/incubator-tvm/pull/5192))
- [FRONTEND][MXNET] support elemwise logic ops ([#5361](https://github.com/apache/incubator-tvm/pull/5361))
- [Frontend|MXNet] SwapAxis operator support ([#5246](https://github.com/apache/incubator-tvm/pull/5246))
- [RELAY] Move frontend utils (#5345)
- [Fontend][Pytorch] Fix translation of transpose when axis argument is as a list (#5451)
- LpPool Support added [#5696](https://github.com/apache/incubator-tvm/pull/5696)
- Skip ADD inside Gemm op when vector is zero [#5697](https://github.com/apache/incubator-tvm/pull/5697)
- ReduceL1, ReduceL2, ReduceSumSquare, ReduceLogSum ops added [#5721](https://github.com/apache/incubator-tvm/pull/5721)
- MaxRoiPool, Mod & Xor op support added [#5729](https://github.com/apache/incubator-tvm/pull/5729)
- Skip multiply with 1.0f constant for GEMM import [#5800](https://github.com/apache/incubator-tvm/pull/5800)
- StatefulPartitionedCall/PartitionedCall Ops support added [#5617](https://github.com/apache/incubator-tvm/pull/5617)
- Don't add cast for batch norm when type isn't changing [#5731](https://github.com/apache/incubator-tvm/pull/5731)
- Conv3d Transpose OP added [#5775](https://github.com/apache/incubator-tvm/pull/5775)
- expand bug fix [#5576](https://github.com/apache/incubator-tvm/pull/5576)
- Support max_pool2d_with_indices [#5549](https://github.com/apache/incubator-tvm/pull/5549)
- Add prim::device op [#5584](https://github.com/apache/incubator-tvm/pull/5584)
- ImplicitTensorToNum support added [#5603](https://github.com/apache/incubator-tvm/pull/5603)
- Matmul fix for batch_matmul [#5604](https://github.com/apache/incubator-tvm/pull/5604)
- ReflectionPad2d op [#5624](https://github.com/apache/incubator-tvm/pull/5624)
- Padding op support [#5638](https://github.com/apache/incubator-tvm/pull/5638)
- Minor bug fixes [#5683](https://github.com/apache/incubator-tvm/pull/5683)
- floor_divide support for squeezenet [#5702](https://github.com/apache/incubator-tvm/pull/5702)
- ReplicationPad support added [#5708](https://github.com/apache/incubator-tvm/pull/5708)
- aten::norm support added [#5776](https://github.com/apache/incubator-tvm/pull/5776)
- broadcast and logical op support [#5461](https://github.com/apache/incubator-tvm/pull/5461)
- MaxPool3d and AvgPool3d Ops support added [#5614](https://github.com/apache/incubator-tvm/pull/5614)
- Softmin, trunc op support added [#5715](https://github.com/apache/incubator-tvm/pull/5715)
- conv3d and conv3d_transpose addedx [#5814](https://github.com/apache/incubator-tvm/pull/5814)
- Model importer to be compatible with tflite 2.1.0 [#5497](https://github.com/apache/incubator-tvm/pull/5497)
- Nit: Function names made consistent [#5515](https://github.com/apache/incubator-tvm/pull/5515)
- Select op support for tflite frontend [#5486](https://github.com/apache/incubator-tvm/pull/5486)
- GATHER_ND [#5508](https://github.com/apache/incubator-tvm/pull/5508)
- Quantize & Dequantize op [#5394](https://github.com/apache/incubator-tvm/pull/5394)
- Fully connected op conversion made in sync with TFLite [#5510](https://github.com/apache/incubator-tvm/pull/5510)
- ADD_N operator [#5474](https://github.com/apache/incubator-tvm/pull/5474)
- onnx, mxnet, pytorch mathops added [#5561](https://github.com/apache/incubator-tvm/pull/5561)
- abs, round, reciprocal, sign, softsign, hard_sigmoid ops support [#5587](https://github.com/apache/incubator-tvm/pull/5587)
- Gather nd bug fix for one dim support in tensorflow [#5588](https://github.com/apache/incubator-tvm/pull/5588)
- Add parser support for shape and range [#5329](https://github.com/apache/incubator-tvm/pull/5329)
- Darknet support batch size for yolo [#5688](https://github.com/apache/incubator-tvm/pull/5688)
- Improve Control Flow and TensorArray [#5699](https://github.com/apache/incubator-tvm/pull/5699)
- MXNet: Softmin, trunc op support added [#5715](https://github.com/apache/incubator-tvm/pull/5715)
- MXNet: conv3d and conv3d_transpose addedx [#5814](https://github.com/apache/incubator-tvm/pull/5814)
- MXNet: Add parser for contrib.box_decode [#5967](https://github.com/apache/incubator-tvm/pull/5967)
- Onnx: ReduceL1, ReduceL2, ReduceSumSquare, ReduceLogSum ops added [#5721](https://github.com/apache/incubator-tvm/pull/5721)
- Onnx: MaxRoiPool, Mod & Xor op support added [#5729](https://github.com/apache/incubator-tvm/pull/5729)
- Onnx: Skip multiply with 1.0f constant for GEMM import [#5800](https://github.com/apache/incubator-tvm/pull/5800)
- Onnx: Fix an issue with #5755 and add Batch norm unit tests. [#5845](https://github.com/apache/incubator-tvm/pull/5845)
- TensorFlow: StatefulPartitionedCall/PartitionedCall Ops support added [#5617](https://github.com/apache/incubator-tvm/pull/5617)
- TensorFlow: Don’t add cast for batch norm when type isn’t changing [#5731](https://github.com/apache/incubator-tvm/pull/5731)
- TensorFlow: Conv3d Transpose OP added [#5775](https://github.com/apache/incubator-tvm/pull/5775)
- Add parser support for shape and range [#5329](https://github.com/apache/incubator-tvm/pull/5329)
- Darknet support batch size for yolo [#5688](https://github.com/apache/incubator-tvm/pull/5688)
- Improve Control Flow and TensorArray [#5699](https://github.com/apache/incubator-tvm/pull/5699)
- Improve TF Parser to keep output nodes for saved_model [#5794](https://github.com/apache/incubator-tvm/pull/5794)
- Add parser support for relu6, leaky_relu, relu_n1_to_1, log_softmax [#4805](https://github.com/apache/incubator-tvm/pull/4805)
- Fix TF Dynamic input shape [#5825](https://github.com/apache/incubator-tvm/pull/5825)
- Support a few contrib ops in mxnet [#5819](https://github.com/apache/incubator-tvm/pull/5819)
- Improve TF Parser to keep output nodes for saved_model [#5794](https://github.com/apache/incubator-tvm/pull/5794)
- Add parser support for relu6, leaky_relu, relu_n1_to_1, log_softmax [#4805](https://github.com/apache/incubator-tvm/pull/4805)
- Check all unsupported ops before raising an exception [#5929](https://github.com/apache/incubator-tvm/pull/5929)
- Add Pytorch advanced indexing [#6318](https://github.com/apache/incubator-tvm/pull/6318)
- Support index_select [#6295](https://github.com/apache/incubator-tvm/pull/6295)
- Fix cast to long [#6301](https://github.com/apache/incubator-tvm/pull/6301)
- Fix dtype handling for modules with integer parameters [#6311](https://github.com/apache/incubator-tvm/pull/6311)
- pytorch frontend support conv1d [#6203](https://github.com/apache/incubator-tvm/pull/6203)
- Add cast to double, fix flatten conversion [#6357](https://github.com/apache/incubator-tvm/pull/6357)
- Fix aten::max and aten::min conversion [#6372](https://github.com/apache/incubator-tvm/pull/6372)
- Match pytorch 1.6 googlenet pretrained model (#6201) [#6212](https://github.com/apache/incubator-tvm/pull/6212)Add unbiased variance op and corresponding support in pytorch frontend [#6232](https://github.com/apache/incubator-tvm/pull/6232)
- Implemented PADV2 Operator for TFLite and added support for constant values in PAD. [#6167](https://github.com/apache/incubator-tvm/pull/6167)
- Implemented ONE_HOT Operator for TFLite. [#6223](https://github.com/apache/incubator-tvm/pull/6223)
- Implemented EXPAND_DIMS Operator for TFLite. [#6243](https://github.com/apache/incubator-tvm/pull/6243)
- Implemented REVERSE_V2 Operator for TFLite. [#6304](https://github.com/apache/incubator-tvm/pull/6304)
- Implemented MATRIX_SET_DIAG Operator for Relay/TOPI and TFLite Frontend. [#6303](https://github.com/apache/incubator-tvm/pull/6303)
- RESHAPE with dynamic shape arg in TFLite frontend [#6208](https://github.com/apache/incubator-tvm/pull/6208)
- Constant input attr added to fully connected operation in TFLite frontend [#6228](https://github.com/apache/incubator-tvm/pull/6228)
- Gather operation with indices as tensor expr in TFLite frontend [#6168](https://github.com/apache/incubator-tvm/pull/6168)
- Added support for tflite quantized maximum and minimum [#6018](https://github.com/apache/incubator-tvm/pull/6018)
- Unary ops support added in frontend [#6196](https://github.com/apache/incubator-tvm/pull/6196)
- Introduce caffe frontend for tvm [#6206](https://github.com/apache/incubator-tvm/pull/6206)
- Keras softmax and prelu fix under NHWC [#6278](https://github.com/apache/incubator-tvm/pull/6278)
- add support for MXNET numpy operators [#6054](https://github.com/apache/incubator-tvm/pull/6054)
- Refine tensorflow frontend 1.x & 2.x compatibility [#6240](https://github.com/apache/incubator-tvm/pull/6240)
- Reduceops support added to frontend [#6252](https://github.com/apache/incubator-tvm/pull/6252)
- Update precision in the ONNX strided_slice, update precision of ToScalar [#6272](https://github.com/apache/incubator-tvm/pull/6272)
- NHWC import support. [#4899](https://github.com/apache/incubator-tvm/pull/4899)
- Refine tensorflow frontend 1.x & 2.x compatibility [#6240](https://github.com/apache/incubator-tvm/pull/6240)
- Fix node indices attribute error for tensorflow 2.3 [#6288](https://github.com/apache/incubator-tvm/pull/6288)
- Support NMSv4 [#6085](https://github.com/apache/incubator-tvm/pull/6085)
- Support for PyTorch Non-Maximum Suppression [#6314](https://github.com/apache/incubator-tvm/pull/6314)
- ReplicationPad support added [#5708](https://github.com/apache/incubator-tvm/pull/5708)
- MXNet pre-quantized BERT [#6039](https://github.com/apache/incubator-tvm/pull/6039)
- Keep parameter names from PyTorch [#5887](https://github.com/apache/incubator-tvm/pull/5887)
- Refine LSTMBlockCell to support dynamic rnn [#5963](https://github.com/apache/incubator-tvm/pull/5963)
## Relay
- Add function attributes to IR hash ([#4479](https://github.com/apache/incubator-tvm/pull/4479/files))
- Relay passes lookup overhead optimization ([#4594](https://github.com/apache/incubator-tvm/pull/4594))
- Add half_pixel option to Resize op [#4610](https://github.com/apache/incubator-tvm/issues/4610)
- Skip example json runtime test when config is not set [#4614](https://github.com/apache/incubator-tvm/issues/4614)
- Test tensor_array in vm [#4608](https://github.com/apache/incubator-tvm/issues/4608)
- Improve memory_allocation pass to support multiple i/o dynamic kernels [#4595](https://github.com/apache/incubator-tvm/issues/4595)
- Add unit test for tensor_array_split [#4619](https://github.com/apache/incubator-tvm/issues/4619)
- Add parses support for unary elemwise ops [#4634](https://github.com/apache/incubator-tvm/issues/4634)
- Add parses support for SLICE [#4502](https://github.com/apache/incubator-tvm/issues/4502)
- Added pool autopadding and simplified converters. [#4672](https://github.com/apache/incubator-tvm/issues/4672)
- Fix meaning of conv2d_transpose output_padding parameter [#4318](https://github.com/apache/incubator-tvm/issues/4318)
- Use packed func macro for external codegen [#4710](https://github.com/apache/incubator-tvm/issues/4710)
- Fix _parse_param bug [#4711](https://github.com/apache/incubator-tvm/issues/4711)
- Add constant input support for elemwise ops [#4666](https://github.com/apache/incubator-tvm/issues/4666)
- Add parser support for squared difference [#4652](https://github.com/apache/incubator-tvm/issues/4652)
- Add type check to dense [#4724](https://github.com/apache/incubator-tvm/issues/4724)
- Invoke tvm::build from relay compile_engine and interpreter [#4723](https://github.com/apache/incubator-tvm/issues/4723)
- Broadcast condition, x, and y for Where op [#4774](https://github.com/apache/incubator-tvm/issues/4774)
- Add parser support for relational ops [#4695](https://github.com/apache/incubator-tvm/issues/4695)
- Remove duplicated BindParamByName function in VM compiler [#4793](https://github.com/apache/incubator-tvm/issues/4793)
- Use SimplifyInference for L2 Normalization. [#4795](https://github.com/apache/incubator-tvm/issues/4795)
- Expose vm OptimizeModule to Python [#4800](https://github.com/apache/incubator-tvm/issues/4800)
- Add parser support for logical operators [#4642](https://github.com/apache/incubator-tvm/pull/4642)
- Conv2D padding representation [#4787](https://github.com/apache/incubator-tvm/pull/4787)
- Add support for quantized LOGISTIC [#4696](https://github.com/apache/incubator-tvm/pull/4696)
- Fix VM compiler for while loop with free vars [#4889](https://github.com/apache/incubator-tvm/pull/4889)
- Fix bug in re-processing call node in MergeComposite pass [#4879](https://github.com/apache/incubator-tvm/pull/4879)
- Expose FunctionGetAttr to Python [#4905](https://github.com/apache/incubator-tvm/pull/4905)
- Add a PyTorch to Relay Parser [#4497](https://github.com/apache/incubator-tvm/pull/4497)
- Support data types for CSourceModuleCodegen args and output [#4934](https://github.com/apache/incubator-tvm/pull/4934)
- Clean up and refactor PyTorch frontend [#4944](https://github.com/apache/incubator-tvm/pull/4944)
- Relay pass to use fast exp/tanh [#4873](https://github.com/apache/incubator-tvm/pull/4873)
- BatchNorm support with run-time mean and variance calculation [#4990](https://github.com/apache/incubator-tvm/pull/4990)
- Reduce plevel of conv2d winograd implementation on cuda [#4987](https://github.com/apache/incubator-tvm/pull/4987)
- Add operation tan to TVM [#4938](https://github.com/apache/incubator-tvm/pull/4938)
- Outline and inline lifted functions for external codegen [#4996](https://github.com/apache/incubator-tvm/pull/4996)
- Remove primitive attribute from composite function [#5014](https://github.com/apache/incubator-tvm/pull/5014)
- Refactor Relay Python to use new FFI [#5077](https://github.com/apache/incubator-tvm/pull/5077)
- Fix relay node registration after refactor [#5083](https://github.com/apache/incubator-tvm/pull/5083)
- Codegen_c.h should include relay.function [#5093](https://github.com/apache/incubator-tvm/pull/5093)
- Move expr.Function to function.py [#5087](https://github.com/apache/incubator-tvm/pull/5087)
- Propagate constant to subgraphs [#5094](https://github.com/apache/incubator-tvm/pull/5094)
- Adjust strategy plevel to achieve expected performance by default [#5118](https://github.com/apache/incubator-tvm/pull/5118)
- Added a AnnotatedRegion utility class [#5030](https://github.com/apache/incubator-tvm/pull/5030)
- Support TupleGetItem in body of pattern [#5106](https://github.com/apache/incubator-tvm/pull/5106)
- Partition graph codestyle fixes [#5202](https://github.com/apache/incubator-tvm/pull/5202)
- Re-wrote the Graph Partitioner to support multiple outputs [#5143](https://github.com/apache/incubator-tvm/pull/5143)
- Fixes to MergeCompilerRegions [#5195](https://github.com/apache/incubator-tvm/pull/5195)
- Refactor build module to take IRModule [#4988](https://github.com/apache/incubator-tvm/pull/4988)
- Separate analysis and transform passes [#5035](https://github.com/apache/incubator-tvm/pull/5035)
- Relay Node::make to constructor [#5128](https://github.com/apache/incubator-tvm/pull/5128)
- relay::StructuralHash to tvm::StructuralHash [#5166](https://github.com/apache/incubator-tvm/pull/5166)
- Conditions updated to cover better user scenarios [#5043](https://github.com/apache/incubator-tvm/pull/5043)
- Replace UseDefaultCompiler with GetAttr [#5088](https://github.com/apache/incubator-tvm/pull/5088)
- Return empty CSourceModule when no lowered_funcs exists in Relay mod [#4847](https://github.com/apache/incubator-tvm/pull/4847)
- [Runtime][Relay][Cleanup] Clean up for memory pass to enable heterogenous execution support. (#5324)
- [RELAY] Remove re-exports of tvm.transform (#5337)
- [Refactor] Add memoized expr translator for use by backend codegen (#5325)
- Legalize - Use Non-recursive Rewriter. (#5296)
- Add additional check before re-using the cached match [#5552](https://github.com/apache/incubator-tvm/pull/5552)
- Remove kCompiler attr from external functions [#5615](https://github.com/apache/incubator-tvm/pull/5615)
- Pattern Language MergeComposite [#5656](https://github.com/apache/incubator-tvm/pull/5656)
- Support Tuple Output in C/DNNL Codegen [#5701](https://github.com/apache/incubator-tvm/pull/5701)
- Infer types in MergeComposite [#5766](https://github.com/apache/incubator-tvm/pull/5766)
- Convert PatternGrouper to do pre-order, non-recursive analysis [#5653](https://github.com/apache/incubator-tvm/pull/5653)
- Remove constants from partitioned functions [#5663](https://github.com/apache/incubator-tvm/pull/5663)
- Add a check for null function attributes [#5674](https://github.com/apache/incubator-tvm/pull/5674)
- Add ConstantPattern [#5689](https://github.com/apache/incubator-tvm/pull/5689)
- Conditionally Embedding Constants in Partitioned Functions [#5693](https://github.com/apache/incubator-tvm/pull/5693)
- Simplify Pattern API Implementations [#5703](https://github.com/apache/incubator-tvm/pull/5703)
- Add ShapePattern and DataTypePattern [#5760](https://github.com/apache/incubator-tvm/pull/5760)
- Remove unnecessary print [#5642](https://github.com/apache/incubator-tvm/pull/5642)
- Improve Shape Func handling for Tuple inputs [#5467](https://github.com/apache/incubator-tvm/pull/5467)
- Relay updated with String [#5578](https://github.com/apache/incubator-tvm/pull/5578)
- Fix the creation of tuple of tuples in PartitionGraph [#5616](https://github.com/apache/incubator-tvm/pull/5616)
- Preserve type information in Merge Composite [#5640](https://github.com/apache/incubator-tvm/pull/5640)
- Move compiler_begin/end_op to local static objects [#5622](https://github.com/apache/incubator-tvm/pull/5622)
- Fix dataflow_pattern.rewrite() hang if Match in IR [#5680](https://github.com/apache/incubator-tvm/pull/5680)
- Fix segfault in pretty print when ObjectRef is null [#5681](https://github.com/apache/incubator-tvm/pull/5681)
- Move fallback_device to config [#5690](https://github.com/apache/incubator-tvm/pull/5690)
- Replace build_config with PassContext [#5698](https://github.com/apache/incubator-tvm/pull/5698)
- Clear compile engine after task extraction [#5724](https://github.com/apache/incubator-tvm/pull/5724)
- Add storage_order ignore in pooling layer. [#5781](https://github.com/apache/incubator-tvm/pull/5781)
- Tweak cublas/cudnn priority level [#5820](https://github.com/apache/incubator-tvm/pull/5820)
- Skip Unknown Function Symbols [#5888](https://github.com/apache/incubator-tvm/pull/5888)
- Allow every runtime module to handle constants [#5885](https://github.com/apache/incubator-tvm/pull/5885)
- handle Tuple/TupleGetItem in first order gradient [#5946](https://github.com/apache/incubator-tvm/pull/5946)
- Add resnet-3d & Update network definitions for NHWC layout [#5945](https://github.com/apache/incubator-tvm/pull/5945)
- Use TargetNode::attrs for Target serialization [#5993](https://github.com/apache/incubator-tvm/pull/5993)
- each option of target str should only contain one ‘=’ [#5988](https://github.com/apache/incubator-tvm/pull/5988)
- Rename target_id => target_kind [#6199](https://github.com/apache/incubator-tvm/pull/6199)
- 64-bit RPi4b target [#6211](https://github.com/apache/incubator-tvm/pull/6211)
- Add resnet-3d & Update network definitions for NHWC layout [#5945](https://github.com/apache/incubator-tvm/pull/5945)
- Small bug fix for Conv1D imports. [#5995](https://github.com/apache/incubator-tvm/pull/5995)
- Move invoke_tvm_op and shape_func to vm dialect [#5958](https://github.com/apache/incubator-tvm/pull/5958)
- GRU Layer Support [#6020](https://github.com/apache/incubator-tvm/pull/6020)
- Add pass for getting calibration data from a relay module [#5997](https://github.com/apache/incubator-tvm/pull/5997)
- Merge two consecutive reshape ops [#6052](https://github.com/apache/incubator-tvm/pull/6052)
- Add operation scatter_add to relay, based on scatter implementation. [#6030](https://github.com/apache/incubator-tvm/pull/6030)
- i64 indices [#5235](https://github.com/apache/incubator-tvm/pull/5235)
- Port eliminate_common_subexpr to non-recursive form [#6134](https://github.com/apache/incubator-tvm/pull/6134)
- Fix interpreter for dyanmic shape input of ndarray_size [#6086](https://github.com/apache/incubator-tvm/pull/6086)
- Allow to config allocator type and refactor vm code structure [#6105](https://github.com/apache/incubator-tvm/pull/6105)
- Handle ndarray_size in FoldConstant [#6156](https://github.com/apache/incubator-tvm/pull/6156)
- when converting constant nodes with types of int64 or float64 [#6159](https://github.com/apache/incubator-tvm/pull/6159)
- Add ReshapeTensor instruction in the VM to replace the reshape op [#6089](https://github.com/apache/incubator-tvm/pull/6089)
- Support combine multiple dense op just into dense [#6062](https://github.com/apache/incubator-tvm/pull/6062)
- Add unbiased variance op and corresponding support in pytorch frontend [#6232](https://github.com/apache/incubator-tvm/pull/6232)
- Specify additional layouts in convert layout pass [#5422](https://github.com/apache/incubator-tvm/pull/5422)
- Safe check added for Merge Composite Call Node [#5562](https://github.com/apache/incubator-tvm/pull/5562)
- Non recursive partitioning [#5493](https://github.com/apache/incubator-tvm/pull/5493)
- Support combine multiple dense op just into dense [#6062](https://github.com/apache/incubator-tvm/pull/6062)
- Make the max number of fused ops configurable [#6327](https://github.com/apache/incubator-tvm/pull/6327)
- Implementation of the dynamic pad operator [#6284](https://github.com/apache/incubator-tvm/pull/6284)
- change device annotation from post DFS to recursive [#6124](https://github.com/apache/incubator-tvm/pull/6124)
- Make check stricter: disallow inserting function with free vars into module [#6313](https://github.com/apache/incubator-tvm/pull/6313)
- Make check stricter by using Feature. Fixed multiple bugs [#6326](https://github.com/apache/incubator-tvm/pull/6326)
- Resize support for NCHW-convertible layouts [#6293](https://github.com/apache/incubator-tvm/pull/6293)
- Make AutoDiff thread through global function [#6336](https://github.com/apache/incubator-tvm/pull/6336)
- Create Interpreter for each constant subgraph [#6195](https://github.com/apache/incubator-tvm/pull/6195)
- Add Dynamic reshape to a dynamic namespace and add DynamicToStatic Pass [#5826](https://github.com/apache/incubator-tvm/pull/5826)
- Expose relay BindParamsByName to Python [#4751](https://github.com/apache/incubator-tvm/issues/4751)
- Implement pass manager tracing API [#4782](https://github.com/apache/incubator-tvm/issues/4782)
- Move Ops in relay.op.contrib.* [#4942](https://github.com/apache/incubator-tvm/pull/4942)
- Conditions updated to cover better user scenarios [#4951](https://github.com/apache/incubator-tvm/pull/4951)
- [External codegen] Add test cases for fused ops with manual annotation ([#4741](https://github.com/apache/incubator-tvm/pull/4741))
- Multiple output support, reshape, split ops added [#6296](https://github.com/apache/incubator-tvm/pull/6296)
## Operator Coverage
- Allow empty tensor for reshape, tile and strided_slice [#4618](https://github.com/apache/incubator-tvm/issues/4618)
- Fix meaning of conv2d_transpose output_padding parameter"; [#4708](https://github.com/apache/incubator-tvm/issues/4708)
- Remove cpp upsampling and resize op [#4769](https://github.com/apache/incubator-tvm/issues/4769)
- upsample operator 'NCHWinic' format support. [#4791](https://github.com/apache/incubator-tvm/pull/4791)
- Injective schedule improvement [#4786](https://github.com/apache/incubator-tvm/pull/4786)
- Enable vectorization on fp16 type [#4867](https://github.com/apache/incubator-tvm/pull/4867)
- Support for Int8 schedules - CUDA/x86 [#5031](https://github.com/apache/incubator-tvm/pull/5031)
- New PR to re-add tan to TVM [#5025](https://github.com/apache/incubator-tvm/pull/5025)
- Register topi schedule for Relay fast_exp and fast_tanh [#5131](https://github.com/apache/incubator-tvm/pull/5131)
- Move Dilation2d from nn to image namespace [#5110](https://github.com/apache/incubator-tvm/pull/5110)
- Use Thrust sort for argsort and topk [#5097](https://github.com/apache/incubator-tvm/pull/5097)
- Conv2d and Dense ops support on Tensor Core [#5099](https://github.com/apache/incubator-tvm/pull/5099)
- Setting workload correctly for Depthwise Spatial conv ARM. [#5182](https://github.com/apache/incubator-tvm/pull/5182)
- Adding a few missing math intrin [#5011](https://github.com/apache/incubator-tvm/pull/5011)
- Missing vectorize for depthwise conv2d. [#5196](https://github.com/apache/incubator-tvm/pull/5196)
- [TOPI] Using x86 schedules for ARM conv2d ([#5334](https://github.com/apache/incubator-tvm/pull/5334))
- [TOPI-ARM] Do not alter layout if layout is NHWC ([#5350](https://github.com/apache/incubator-tvm/pull/5350))
- [TOPI] Setting workload correctly for Depthwise Spatial conv ARM. ([#5182](https://github.com/apache/incubator-tvm/pull/5182))
- [Relay][OP] Add fast_erf implementation ([#5241](https://github.com/apache/incubator-tvm/pull/5241))
- [Topi] Tensorcore support for Conv3D ([#5284](https://github.com/apache/incubator-tvm/pull/5284))
- [intrin] a few more math functions ([#5468](https://github.com/apache/incubator-tvm/pull/5468))
- [Intrinsic] Add log1p, ldexp, atan2, hypot, nextafter, copysign ([#5312](https://github.com/apache/incubator-tvm/pull/5312))
- [relay][topi] Add operation relay.nn.dilate() which calls topi.nn.dilate() ([#5331](https://github.com/apache/incubator-tvm/pull/5331))
- [Topi x86] Missing vectorize for depthwise conv2d. ([#5196](https://github.com/apache/incubator-tvm/pull/5196))
- [TOPI x86] Adding unroll_kw config option for depthwise conv2d. (#5197)
- [Topi] Breakdown [topi.cc](http://topi.cc/) into smaller files ([#5253](https://github.com/apache/incubator-tvm/pull/5253))
- ReduceLogSumExp Operator support [#5453](https://github.com/apache/incubator-tvm/pull/5453)
- Math ops added [#5502](https://github.com/apache/incubator-tvm/pull/5502)
- Enable blocking format in x86 conv2d and fold scale axis [#5357](https://github.com/apache/incubator-tvm/pull/5357)
- Add operation gather to relay. [#5716](https://github.com/apache/incubator-tvm/pull/5716)
- Add storage_order ignore in pooling layer. [#5781](https://github.com/apache/incubator-tvm/pull/5781)
- Fix bifrost spatial packing conv2d auto tune [#5684](https://github.com/apache/incubator-tvm/pull/5684)
- Fix reshape usage in ARM schedule [#5732](https://github.com/apache/incubator-tvm/pull/5732)
- Block sparse dense on cuda [#5746](https://github.com/apache/incubator-tvm/pull/5746)
- Improve CUDA softmax scheduling [#5600](https://github.com/apache/incubator-tvm/pull/5600)
- block sparse dense on cuda [#5746](https://github.com/apache/incubator-tvm/pull/5746)
- pass-by-value -> pass-by-const-reference [#5783](https://github.com/apache/incubator-tvm/pull/5783)
- Using MKL blas for quantized dense [#6115](https://github.com/apache/incubator-tvm/pull/6115)
- topi -> tvm/topi [#6186](https://github.com/apache/incubator-tvm/pull/6186)
- Use auto-tuner to improve conv2d_gemm performance [#6117](https://github.com/apache/incubator-tvm/pull/6117)
- Improve CUDA conv2d_transpose_nchw [#4762](https://github.com/apache/incubator-tvm/issues/4762)
- Add CUDA conv2d for NHWC layout [#4737](https://github.com/apache/incubator-tvm/issues/4737)
- conv3d_ndhwc schedule [#4775](https://github.com/apache/incubator-tvm/issues/4775)
- Fast exponent [#4790](https://github.com/apache/incubator-tvm/pull/4790)
- Add Scatter to Topi/Relay/ONNX via hybrid script [#5619](https://github.com/apache/incubator-tvm/pull/5619)
- Split MKL from BLAS. [#6182](https://github.com/apache/incubator-tvm/pull/6182)
- Change the meaning of conv3d_transpose output_padding to match conv{1,2}d_transpose [#6065](https://github.com/apache/incubator-tvm/pull/6065)
- Gather op support added [#6013](https://github.com/apache/incubator-tvm/pull/6013)
## Runtime and Backend
- Cythonize NDArray.copyto ([#4549](https://github.com/apache/incubator-tvm/pull/4549))
- Unified Object System runtime refactor ([#4578](https://github.com/apache/incubator-tvm/pull/4578), [#4581](https://github.com/apache/incubator-tvm/pull/4581), [#4603](https://github.com/apache/incubator-tvm/pull/4603))
- VM profiler: sort VM stats by time ([#4601](https://github.com/apache/incubator-tvm/pull/4601))
- Update RPC runtime to allow remote module as arg ([#4462](https://github.com/apache/incubator-tvm/pull/4462))
- Refactorying system lib and dso lib into library module ([#4481](https://github.com/apache/incubator-tvm/pull/4481))
- Improve TSIM virtual memory mapping ([#4545](https://github.com/apache/incubator-tvm/pull/4545))
- make adt tag signed [#4605](https://github.com/apache/incubator-tvm/issues/4605)
- Improve TVMBackendPackedCFunc to allow return val [#4637](https://github.com/apache/incubator-tvm/issues/4637)
- EdgeTPU runtime for Coral Boards [#4698](https://github.com/apache/incubator-tvm/issues/4698)
- Fix memory leak when using openMP [#4811](https://github.com/apache/incubator-tvm/issues/4811)
- Fix memory leakage of TVMByteArray [#4856](https://github.com/apache/incubator-tvm/pull/4856)
- Fix TVM_DLL_EXPORT_TYPED_FUNC to work on Windows [#4955](https://github.com/apache/incubator-tvm/pull/4955)
- Fix memory leak when using openMP [#4811](https://github.com/apache/incubator-tvm/pull/4811)
- Export GraphRuntime in tvm_runtime.dll [#5002](https://github.com/apache/incubator-tvm/pull/5002)
- MISRA-C compliant TVM runtime [#3934](https://github.com/apache/incubator-tvm/pull/3934)
- Update the type_keys to reflect the code-org [#5074](https://github.com/apache/incubator-tvm/pull/5074)
- Fix AttrEqual for Array and StrMap, double [#5054](https://github.com/apache/incubator-tvm/pull/5054)
- Export GraphRuntime in tvm_runtime.dll [#5002](https://github.com/apache/incubator-tvm/pull/5002)
- Fix unused-value warning [#5140](https://github.com/apache/incubator-tvm/pull/5140)
- crt error handling [#5147](https://github.com/apache/incubator-tvm/pull/5147)
- Bundle deployment with static linking [#5158](https://github.com/apache/incubator-tvm/pull/5158)
- Implemented kDLCPUPinned (cudaMallocHost) [#4985](https://github.com/apache/incubator-tvm/pull/4985)
- Explicitly cast min/max operands [#5090](https://github.com/apache/incubator-tvm/pull/5090)
- ref_counter -> ref_counter_ [#5184](https://github.com/apache/incubator-tvm/pull/5184)
- Expose runtime::String to Python ([#5212](https://github.com/apache/incubator-tvm/pull/5212))
- [PY][FFI] Refactor runtime.String to subclass str ([#5426](https://github.com/apache/incubator-tvm/pull/5426))
- [RUNTIME] Auto conversion from str to runtime::String in PackedFUnc ([#5251](https://github.com/apache/incubator-tvm/pull/5251))
- [RUNTIME] Improved Packed FFI for optional. ([#5478](https://github.com/apache/incubator-tvm/pull/5478))
- [Hexagon] Add hexagon_posix.cc to TVM/RT sources in the right place (#5346)
- [PY][FFI] Refactor runtime.String to subclass str (#5426)
- Fix workspace [#5503](https://github.com/apache/incubator-tvm/pull/5503)
- Store nullptr PackedFunc as nullptr for better error propagation [#5540](https://github.com/apache/incubator-tvm/pull/5540)
- Improve PackedFunc robustness [#5517](https://github.com/apache/incubator-tvm/pull/5517)
- Seg fault in WorkspacePool's destructor (#5632) [#5636](https://github.com/apache/incubator-tvm/pull/5636)
- Resolve constexpr issue in debug mode. [#5651](https://github.com/apache/incubator-tvm/pull/5651)
- Add compile_shared option to linux compile utility fn [#5751](https://github.com/apache/incubator-tvm/pull/5751)
- Call sync in CopyFromRemote and CopyToRemote [#5512](https://github.com/apache/incubator-tvm/pull/5512)
- Fix the multihop cpu case [#5522](https://github.com/apache/incubator-tvm/pull/5522)
- Improve RPCServer AsyncIO support. [#5544](https://github.com/apache/incubator-tvm/pull/5544)
- Modularize the RPC infra [#5484](https://github.com/apache/incubator-tvm/pull/5484)
- Add compile_shared option to linux compile utility fn [#5751](https://github.com/apache/incubator-tvm/pull/5751)
- Overload string operators [#5806](https://github.com/apache/incubator-tvm/pull/5806)
- Only initialize required module [#5926](https://github.com/apache/incubator-tvm/pull/5926)
- if a param not in input, we should still consume it’s data [#5990](https://github.com/apache/incubator-tvm/pull/5990)
- init TVMPackedFunc’s name [#6044](https://github.com/apache/incubator-tvm/pull/6044)
- Enable auto conversion String->DLDataType [#6214](https://github.com/apache/incubator-tvm/pull/6214)
- Support random fill [#5913](https://github.com/apache/incubator-tvm/pull/5913)
- Use new to avoid exit-time de-allocation order [#6292](https://github.com/apache/incubator-tvm/pull/6292)
- Add parallel_for support to run a loop in parallel [#6275](https://github.com/apache/incubator-tvm/pull/6275)
- Solve ARM BIG.LITTLE heterogeneous multicores [#4747](https://github.com/apache/incubator-tvm/issues/4747)
- [RUNTIME] Quick fix PackedFunc String passing ([#5266](https://github.com/apache/incubator-tvm/pull/5266))
- Introduce runtime::String::CanConvertFrom [#5718](https://github.com/apache/incubator-tvm/pull/5718)
- Restore the StrMap behavior in JSON/SHash/SEqual [#5719](https://github.com/apache/incubator-tvm/pull/5719)
- Support overriding RPCWatchdog termination behavior on Android and other platforms [#6216](https://github.com/apache/incubator-tvm/pull/6216)
- Set NDArray::Container.shape_ in NDArray::FromDLPack ([#5301](https://github.com/apache/incubator-tvm/pull/5301))
- Enable x86 cpu cache flush [#5914](https://github.com/apache/incubator-tvm/pull/5914)
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
users@infra.apache.org