TVM Monthly - August 2021

小咪咪 2022-09-09 14:59 9阅读 0赞

# [TVM Monthly - August 2021][] #

As discussed by the TVM PMC, our goal is to provide a monthly summary of the project so users and developers can get a better understanding of the goings on of the TVM community.

Feedback and suggestions are welcomed so that we can further improve these updates.

# Community #

During August of 2021 we welcomed many new contributors to the project. Importantly we welcomed [@manupa-arm][manupa-arm] as a new committer, [@electriclilies][electriclilies], [@Mousius][Mousius], [@gromero][gromero], [@Lunderberg][Lunderberg], and [@mdw-octoml][mdw-octoml] as new reviewers. Thanks to everyone for the hard work and contributions!

We continue to improve TOPI and frontend support, especially on the ONNX importer and new frontends ([PaddlePaddle 2][PaddlePaddle_2], [OneFlow 1][OneFlow_1]). [TensorIR][] is in steady progress, several schedule primitives have been added. We started adding features for [Meta Schedule (AutoTIR) 1][Meta Schedule _AutoTIR_1], the new auto-scheduling system on top of TensorIR. We improved Relay with better profilers, executors and mixed-precision support. We landed [Project API][], an infrastructure for MicroTVM platforms. The community has also made various improvements to CI and documentation.

This forum got 122k pageviews, 2.8k user visits in the last month.

# Pull Requests #

The below is high-level summary of the PRs closed in the last month grouped by area.

# TensorIR #

*  Fix a typo in include/tvm/ir/function.h [\#8617 1][8617_1]
 *  Add `from_legacy_te_schdule` attr to TE PrimFuncs [\#8641][8641]
 *  LowerWarpMemory: remove unneeded shuffle when accessing from the same thread [\#8681][8681]
 *  Storage Align [\#8693][8693]
 *  Improve the error message in module.cc [\#8694][8694]
 *  Parallel, Vectorize, Bind & Unroll [\#8716][8716]
 *  Reorder [\#8767][8767]
 *  CacheRead/Write [\#8863][8863]
 *  enhance tir signed-unsigned cast [\#8706 1][8706_1]
 *  Change Integer Implicit Conversion Rule to C Standard Way [\#8733][8733]
 *  Support fold constants in specialize process [\#8803][8803]
 *  Fix buffer scope in structural equal [\#8768][8768]
 *  Add LowerTEPass, and convert calls to LowerTE to application of LowerTEPass [\#8802][8802]
 *  GetBlockReadWriteRegion [\#8875][8875]
 *  Bug fix for a floormod rewrite simplify rule [\#8852][8852]
 *  Fix opaque access in buffer locator pass and match\_buffer in region detector [\#8855][8855]
 *  Fix printing ForNode annotations [\#8891][8891]

# Relay #

*  Extend FakeQuantizationToInteger to more ops [\#8241][8241]
 *  Change Default "opt\_level" of Sequential from 2 to 0 [\#8634][8634]
 *  Support for non scalar zero points in qnn.conv2d [\#8620][8620]
 *  Remove redundant cuda kernels caused by fusion of less & logical or [\#8618][8618]
 *  Replace compile engine with TE compiler in the VM [\#8501][8501]
 *  Dense alter layout fixed for packed input [\#8669][8669]
 *  Refactor Interpreter to treat lowering as IRModule->IRModule rewrite. [\#8597][8597]
 *  Extract dataflow matcher data structure into header [\#8774][8774]
 *  Support of depthwise conv2d NHWC for Mali/Bifrost. [\#8584][8584]
 *  Avoid Override Generic Op Strategy in "hls.py" [\#8614][8614]
 *  Add batch\_matmul convertion to FQ2I pass [\#8635][8635]
 *  Expose FTVMInferCorrectLayout Python interface [\#8755][8755]
 *  ToBasicBlockNormalForm immutability [\#8778][8778]
 *  Disallow fp16 conversion for summation-like ops [\#8810][8810]
 *  Add an option to rewrite the graph only once [\#8843][8843]

# Frontend #

*  add suppport for 'aten::upsample\_bicubic2d' [\#8648][8648]
 *  Support for nn.SiLU added [\#8753][8753]
 *  Implement fake quant [\#8780][8780]
 *  GRU layer [\#8781][8781]
 *  Unified LSTM cell [\#8599][8599]
 *  Add onnx opset v13 support for softmax, logsoftmax [\#8625][8625]
 *  Add a PaddlePaddle Frontend [\#8645 2][PaddlePaddle_2]
 *  Support TensorFlow < 1.13 for test\_sparse\_add [\#8647][8647]
 *  add support for half\_pixel\_centers in resize [\#8689][8689]
 *  in-place methods (sigmoid\_ and tanh\_) used by Tacotron2 were added [\#8692][8692]
 *  Fix ELU conversion [\#8699][8699]
 *  chunk and unsafe chunk [\#8718][8718]
 *  Make from\_tensorflow.py more GPU memory friendly. [\#8763][8763]
 *  Add support for QLinearMul ONNX op [\#8773][8773]
 *  Increased tolerance on onnx test\_forward::test\_aten [\#8798][8798]
 *  extend repeat\_interleave op for relay.Expr [\#8839][8839]
 *  Simplify onnx input since name accesses are not reliable. [\#8867][8867]

# Topi & Operators #

*  Improve the performance of scatter\_nd [\#8479][8479]
 *  Float16 unittests for dense, conv2d, depthwise conv2d [\#8529][8529]
 *  Sparse Conv2d Implementation for 3x3 kernels [\#8605 1][8605_1]
 *  Add transpose\_a/b for TensorRT batch\_matmul [\#8607][8607]
 *  minor bugs [\#8622][8622]
 *  CMSIS-NN graph partitioner for softmax [\#8653][8653]
 *  remove wrong fix in x86's dense\_nopack operator [\#8687][8687]
 *  densenet implementation fix [\#8704][8704]
 *  Celu [\#8741][8741]
 *  Bug fix for batch\_matmul parameters mismatch [\#8785][8785]
 *  Support select\_last\_index for argmin/max [\#8816][8816]

# Executor & AOT #

*  add set\_output\_zero\_copy [\#8497][8497]
 *  Remove unused parameter. [\#8580][8580]
 *  Add get\_input\_index support. [\#8661][8661]
 *  Add graph\_executor get\_input\_index API. [\#8633][8633]
 *  Remove unused variables in AOT tests [\#8686][8686]
 *  Refactor AOT Test Utils parameters into object [\#8650][8650]
 *  Convert AOT to TECompiler [\#8697][8697]
 *  Run AOT tests against reference system [\#8744][8744]
 *  Remove old AOT Executor code [\#8758][8758]
 *  Change AOT from ExprVisitor to MixedModeVisitor [\#8856][8856]
 *  Better reflect allocator names in CRT tests [\#8828][8828]
 *  Remove unused allocated memory in crt initialization [\#8819][8819]
 *  Switch profile flag to use new profiler [\#8710][8710]
 *  Add benchmarking function to graph executor and vm [\#8807][8807]
 *  Add end to end benchmarking of models [\#8858][8858]
 *  Correctly link to PAPI [\#8691][8691]

# AutoTVM & AutoScheduler & MetaSchedule #

*  Fix deserization of workload registry entry [\#8662][8662]
 *  Fix FLOPS estimation [\#8695 1][8695_1]
 *  Use PopenPool instead of multiprocessing.pool [\#8492][8492]
 *  Update AutoScheduler Docs – Units for cooldown\_interval [\#8736][8736]
 *  Fix exception handling in measure.py [\#8754][8754]
 *  Configurable workload keys [\#8862][8862]
 *  Fix use of fallback AutoTVM knobs in default scheduling [\#8707][8707]
 *  Updated tolerances to avoid flaky unit test. [\#8723][8723]
 *  Add parameter to allow caller to supply a Runner [\#8747][8747]
 *  Extend tune\_relay\_x86 tutorial to measure default and kernel level tune [\#8794][8794]
 *  Use PopenPool in XGBoostCostModel [\#8820][8820]
 *  Traced Schedule [\#8623][8623]
 *  Linear Congruential Random Engine [\#8642][8642]
 *  Add Sampling Primitive SampleCategorical. [\#8817][8817]
 *  Instruction and Trace [\#8615][8615]

# Target & Codegen #

*  \[Texture support\] TIR lowering and OpenCL support [\#7686][7686]
 *  Allow spaces in target attributes [\#8587][8587]
 *  Add support for AOT in external code generation tests [\#8591][8591]
 *  Framework for device querying for all targets. [\#8602][8602]
 *  Fix test\_external\_codegen, broken by \#8591 [\#8630][8630]
 *  Add \_\_launch\_bounds\_\_ directive as part of the CUDA code generation [\#8678][8678]
 *  Disallow fp16 conversion for arange op [\#8644][8644]
 *  Several minor corrections to the device property query [\#8651][8651]
 *  Correct passing of target-queried bool/int parameters [\#8660][8660]
 *  Support fp16 input in cpu sort [\#8672][8672]
 *  Fix builtin\_fp16.h path according to: [https://discuss.tvm.apache.org/…][https_discuss.tvm.apache.org] [\#8705][8705]
 *  fix tir.erf codegen to opencl directly [\#8756][8756]
 *  Check at codegen if the shader is within shared memory limits. [\#8746][8746]
 *  Fix Vulkan runtime support [\#8791][8791]
 *  Remote target.h \#include [\#8813][8813]
 *  Remove uses of LLVM from simulator runtime [\#8821][8821]
 *  Reuse Hexagon SDK analysis across cmake files [\#8822][8822]
 *  Rework tvm.target.hexagon() interface [\#8823][8823]
 *  Change target string to Target object in the TE compiler and interpreter [\#8835][8835]
 *  Add support for llvm parameter -mabi (aka -target-abi) [\#8860][8860]
 *  Added the driver name to the vulkan target string. [\#8882][8882]

# MicroTVM #

*  Introduce --interface-api=\{c,packed\} parameter [\#8280][8280]
 *  Set the number of cores based on the VM sizing [\#8624][8624]
 *  Fix platform name in base-box-tool [\#8612][8612]
 *  Add skip for AOT test [\#8628][8628]
 *  Project API infrastructure [\#8380][8380]
 *  Add Arduino CLI support to ci-qemu [\#8504][8504]
 *  Rev ci-qemu to 0.07 (add arduino-cli to ci-qemu) [\#8698][8698]
 *  Zephyr Test Refactor [\#8713 1][8713_1]
 *  Remove QEMU installation from RVM [\#8701][8701]
 *  Fix warnings on Zephyr tests [\#8740][8740]
 *  Fix ci-qemu Arduino install dir [\#8766][8766]
 *  Project API Arduino support [\#8708][8708]
 *  Fix base-box-tool command in README.md [\#8613][8613]
 *  Fix: Test fails on hardware because of short timeout [\#8677][8677]
 *  Fix platform name for qemu\_x86 in Zephyr AOT tests [\#8762][8762]
 *  skip aot checks when USE\_MICRO=OFF [\#8772][8772]
 *  Increase timeout to fix flaky tests [\#8846][8846]
 *  Add Arduino RVM [\#8748][8748]
 *  Update QemuTransport\#write() to match new write API contract. [\#8761][8761]
 *  Remove AOT Executor header from Arduino project [\#8857][8857]

# VTA #

*  Fix vta rpc server, refactor launch cond to not depend on sys.argv [\#8671][8671]
 *  Make vta graph\_pack compatible with latest TVM, and bring back object detection tutorials. [\#8731][8731]
 *  VTA cmake change to include Verilator header for building tsim library [\#8797][8797]

# Rust #

*  Fix rust rt link [\#8631 1][8631_1]
 *  Allow rust tvm build configuration through cargo features [\#8665][8665]
 *  Memory leak [\#8714 1][8714_1]
 *  Fix memory leak \#2 [\#8725 1][8725_1]

# Docs #

*  Fix scipy docs inv [\#8619][8619]
 *  Fix the usage of executors in tutorials [\#8586][8586]
 *  TVM install addenda for M1 Macs [\#8568][8568]
 *  Added documentation on pytest target parametrization. [\#8638][8638]
 *  Updated target parametrization documentation [\#8724][8724]
 *  Moved the generated tutorials folders into a \_staging folder. [\#8735][8735]
 *  refactor optimize GEMM on CPU tutorial [\#8825][8825]
 *  Correct function signatures for CreateXPass functions in docs [\#8829][8829]
 *  Add link to docs and tutorials in the README. [\#8832][8832]

# CI & Build #

*  Add caching to CMake [\#8373][8373]
 *  Add pre-commit configuration to perform minimal checks locally [\#8382][8382]
 *  Docker env for Arm® Ethos™-U55 Port [\#8514][8514]
 *  Add USE\_PAPI configuration to config.cmake [\#8567][8567]
 *  Fix global pip cache disable change [\#8590][8590]
 *  Move flake8 to ci\_lint [\#8652][8652]
 *  Refactor RPC test to isolate runs into a sub-function [\#8656][8656]
 *  Restore the Rust CI testing after Docker image update [\#8657][8657]
 *  Refactor/clean-up of docker/bash.sh [\#8670][8670]
 *  Fix error when compile tvm with latest llvm14git [\#8682][8682]
 *  Increase atol for CI [\#8712][8712]
 *  Add Arm Compute Library to Arm CI unit test pipeline [\#8734][8734]
 *  Enable custom images to be set in TVM Jenkinsfile [\#8721][8721]
 *  Add PaddlePaddle dependency in docker file [\#8742][8742]
 *  Rev ci-cpu to v0.76 [\#8786][8786]
 *  Move Rust Format Script [\#8726][8726]
 *  Install rust in ci-lint so cargo fmt can move to lint stage [\#8727][8727]
 *  Allow Linker script files to be committed [\#8745][8745]
 *  Add params.\* to Jenkins file parameters [\#8771][8771]
 *  Rev ci-qemu to v0.08 [\#8776][8776]
 *  Allow Vulkan GPU access in docker container. [\#8784][8784]
 *  Remove leftover instances of USE\_GRAPH\_EXECUTOR\_DEBUG [\#8796][8796]
 *  Update CPU and GPU Image [\#8853][8853]
 *  Add synr==0.3.0 dependency for Docker images and Python dependency. [\#8801][8801]
 *  A small bug fix on the CmakeLists [\#8826][8826]
 *  Support for CMSIS-NN in Corstone300 Makefile [\#8831][8831]
 *  Force CMake targets in top-level Makefile to run [\#8840][8840]
 *  Update CI Lint Image Version [\#8841][8841]
 *  make pre-commit hooks to run on every push instead of every commit [\#8888][8888]

# Unit tests #

*  Added cuDNN to default test targets [\#8383][8383]
 *  Expose TVM pytest helpers as plugin [\#8532][8532]
 *  Apply correct requires\_gpu() pytest marks for parametrized target [\#8542][8542]
 *  Parametrize ONNX Unit tests [\#8621][8621]
 *  Use CTest for C++ tests [\#8809][8809]
 *  Apply CPPLint to C++ Unit Tests [\#8827][8827]
 *  Apply CPPLint to CRT Tests [\#8844][8844]
 *  Bump up tolerance on flaky test [\#8850][8850]
 *  Require cached fixtures to be copy-able, with opt-in. [\#8451][8451]
 *  Remove duplicated PackedFunc C++ test [\#8812][8812]

# Misc #

*  Rename .asnumpy() to .numpy() [\#8659][8659]
 *  Add DictAttrs to IRModule and refactor DictAttrs utility functions [\#8750][8750]
 *  Force a gc between sphinx-gallery items to reclaim GPU memory. [\#8722][8722]
 *  Restore License [\#8779][8779]
 *  Remove reference to Apache Incubator status. [\#8837][8837]
 *  Allow customized initializer in PopenPool [\#8789][8789]
 *  Fix typos [\#8787][8787]
 *  Remove unnecessary memset in TVMMutableFuncRegistry initialization [\#8818][8818]
 *  Fix threadpool reset by killing threads before destroying their shared queue [\#8658][8658]
 *  Fix ios\_rpc build [\#8864][8864]
 *  Change declaration order of unique\_ptr objects to fix crash [\#8859][8859]

[TVM Monthly - August 2021]: https://discuss.tvm.apache.org/t/tvm-monthly-august-2021/10929
[manupa-arm]: https://discuss.tvm.apache.org/u/manupa-arm
[electriclilies]: https://discuss.tvm.apache.org/u/electriclilies
[Mousius]: https://discuss.tvm.apache.org/u/mousius
[gromero]: https://discuss.tvm.apache.org/u/gromero
[Lunderberg]: https://discuss.tvm.apache.org/u/lunderberg
[mdw-octoml]: https://discuss.tvm.apache.org/u/mdw-octoml
[PaddlePaddle_2]: https://github.com/apache/tvm/pull/8645
[OneFlow_1]: https://github.com/apache/tvm-rfcs/pull/24
[TensorIR]: https://github.com/apache/tvm/issues/7527
[Meta Schedule _AutoTIR_1]: https://github.com/apache/tvm/issues/8473
[Project API]: https://github.com/apache/tvm-rfcs/blob/main/rfcs/0008-microtvm-project-api.md
[8617_1]: https://github.com/apache/tvm/pull/8617
[8641]: https://github.com/apache/tvm/pull/8641
[8681]: https://github.com/apache/tvm/pull/8681
[8693]: https://github.com/apache/tvm/pull/8693
[8694]: https://github.com/apache/tvm/pull/8694
[8716]: https://github.com/apache/tvm/pull/8716
[8767]: https://github.com/apache/tvm/pull/8767
[8863]: https://github.com/apache/tvm/pull/8863
[8706_1]: https://github.com/apache/tvm/pull/8706
[8733]: https://github.com/apache/tvm/pull/8733
[8803]: https://github.com/apache/tvm/pull/8803
[8768]: https://github.com/apache/tvm/pull/8768
[8802]: https://github.com/apache/tvm/pull/8802
[8875]: https://github.com/apache/tvm/pull/8875
[8852]: https://github.com/apache/tvm/pull/8852
[8855]: https://github.com/apache/tvm/pull/8855
[8891]: https://github.com/apache/tvm/pull/8891
[8241]: https://github.com/apache/tvm/pull/8241
[8634]: https://github.com/apache/tvm/pull/8634
[8620]: https://github.com/apache/tvm/pull/8620
[8618]: https://github.com/apache/tvm/pull/8618
[8501]: https://github.com/apache/tvm/pull/8501
[8669]: https://github.com/apache/tvm/pull/8669
[8597]: https://github.com/apache/tvm/pull/8597
[8774]: https://github.com/apache/tvm/pull/8774
[8584]: https://github.com/apache/tvm/pull/8584
[8614]: https://github.com/apache/tvm/pull/8614
[8635]: https://github.com/apache/tvm/pull/8635
[8755]: https://github.com/apache/tvm/pull/8755
[8778]: https://github.com/apache/tvm/pull/8778
[8810]: https://github.com/apache/tvm/pull/8810
[8843]: https://github.com/apache/tvm/pull/8843
[8648]: https://github.com/apache/tvm/pull/8648
[8753]: https://github.com/apache/tvm/pull/8753
[8780]: https://github.com/apache/tvm/pull/8780
[8781]: https://github.com/apache/tvm/pull/8781
[8599]: https://github.com/apache/tvm/pull/8599
[8625]: https://github.com/apache/tvm/pull/8625
[8647]: https://github.com/apache/tvm/pull/8647
[8689]: https://github.com/apache/tvm/pull/8689
[8692]: https://github.com/apache/tvm/pull/8692
[8699]: https://github.com/apache/tvm/pull/8699
[8718]: https://github.com/apache/tvm/pull/8718
[8763]: https://github.com/apache/tvm/pull/8763
[8773]: https://github.com/apache/tvm/pull/8773
[8798]: https://github.com/apache/tvm/pull/8798
[8839]: https://github.com/apache/tvm/pull/8839
[8867]: https://github.com/apache/tvm/pull/8867
[8479]: https://github.com/apache/tvm/pull/8479
[8529]: https://github.com/apache/tvm/pull/8529
[8605_1]: https://github.com/apache/tvm/pull/8605
[8607]: https://github.com/apache/tvm/pull/8607
[8622]: https://github.com/apache/tvm/pull/8622
[8653]: https://github.com/apache/tvm/pull/8653
[8687]: https://github.com/apache/tvm/pull/8687
[8704]: https://github.com/apache/tvm/pull/8704
[8741]: https://github.com/apache/tvm/pull/8741
[8785]: https://github.com/apache/tvm/pull/8785
[8816]: https://github.com/apache/tvm/pull/8816
[8497]: https://github.com/apache/tvm/pull/8497
[8580]: https://github.com/apache/tvm/pull/8580
[8661]: https://github.com/apache/tvm/pull/8661
[8633]: https://github.com/apache/tvm/pull/8633
[8686]: https://github.com/apache/tvm/pull/8686
[8650]: https://github.com/apache/tvm/pull/8650
[8697]: https://github.com/apache/tvm/pull/8697
[8744]: https://github.com/apache/tvm/pull/8744
[8758]: https://github.com/apache/tvm/pull/8758
[8856]: https://github.com/apache/tvm/pull/8856
[8828]: https://github.com/apache/tvm/pull/8828
[8819]: https://github.com/apache/tvm/pull/8819
[8710]: https://github.com/apache/tvm/pull/8710
[8807]: https://github.com/apache/tvm/pull/8807
[8858]: https://github.com/apache/tvm/pull/8858
[8691]: https://github.com/apache/tvm/pull/8691
[8662]: https://github.com/apache/tvm/pull/8662
[8695_1]: https://github.com/apache/tvm/pull/8695
[8492]: https://github.com/apache/tvm/pull/8492
[8736]: https://github.com/apache/tvm/pull/8736
[8754]: https://github.com/apache/tvm/pull/8754
[8862]: https://github.com/apache/tvm/pull/8862
[8707]: https://github.com/apache/tvm/pull/8707
[8723]: https://github.com/apache/tvm/pull/8723
[8747]: https://github.com/apache/tvm/pull/8747
[8794]: https://github.com/apache/tvm/pull/8794
[8820]: https://github.com/apache/tvm/pull/8820
[8623]: https://github.com/apache/tvm/pull/8623
[8642]: https://github.com/apache/tvm/pull/8642
[8817]: https://github.com/apache/tvm/pull/8817
[8615]: https://github.com/apache/tvm/pull/8615
[7686]: https://github.com/apache/tvm/pull/7686
[8587]: https://github.com/apache/tvm/pull/8587
[8591]: https://github.com/apache/tvm/pull/8591
[8602]: https://github.com/apache/tvm/pull/8602
[8630]: https://github.com/apache/tvm/pull/8630
[8678]: https://github.com/apache/tvm/pull/8678
[8644]: https://github.com/apache/tvm/pull/8644
[8651]: https://github.com/apache/tvm/pull/8651
[8660]: https://github.com/apache/tvm/pull/8660
[8672]: https://github.com/apache/tvm/pull/8672
[https_discuss.tvm.apache.org]: https://discuss.tvm.apache.org/%E2%80%A6
[8705]: https://github.com/apache/tvm/pull/8705
[8756]: https://github.com/apache/tvm/pull/8756
[8746]: https://github.com/apache/tvm/pull/8746
[8791]: https://github.com/apache/tvm/pull/8791
[8813]: https://github.com/apache/tvm/pull/8813
[8821]: https://github.com/apache/tvm/pull/8821
[8822]: https://github.com/apache/tvm/pull/8822
[8823]: https://github.com/apache/tvm/pull/8823
[8835]: https://github.com/apache/tvm/pull/8835
[8860]: https://github.com/apache/tvm/pull/8860
[8882]: https://github.com/apache/tvm/pull/8882
[8280]: https://github.com/apache/tvm/pull/8280
[8624]: https://github.com/apache/tvm/pull/8624
[8612]: https://github.com/apache/tvm/pull/8612
[8628]: https://github.com/apache/tvm/pull/8628
[8380]: https://github.com/apache/tvm/pull/8380
[8504]: https://github.com/apache/tvm/pull/8504
[8698]: https://github.com/apache/tvm/pull/8698
[8713_1]: https://github.com/apache/tvm/pull/8713
[8701]: https://github.com/apache/tvm/pull/8701
[8740]: https://github.com/apache/tvm/pull/8740
[8766]: https://github.com/apache/tvm/pull/8766
[8708]: https://github.com/apache/tvm/pull/8708
[8613]: https://github.com/apache/tvm/pull/8613
[8677]: https://github.com/apache/tvm/pull/8677
[8762]: https://github.com/apache/tvm/pull/8762
[8772]: https://github.com/apache/tvm/pull/8772
[8846]: https://github.com/apache/tvm/pull/8846
[8748]: https://github.com/apache/tvm/pull/8748
[8761]: https://github.com/apache/tvm/pull/8761
[8857]: https://github.com/apache/tvm/pull/8857
[8671]: https://github.com/apache/tvm/pull/8671
[8731]: https://github.com/apache/tvm/pull/8731
[8797]: https://github.com/apache/tvm/pull/8797
[8631_1]: https://github.com/apache/tvm/pull/8631
[8665]: https://github.com/apache/tvm/pull/8665
[8714_1]: https://github.com/apache/tvm/pull/8714
[8725_1]: https://github.com/apache/tvm/pull/8725
[8619]: https://github.com/apache/tvm/pull/8619
[8586]: https://github.com/apache/tvm/pull/8586
[8568]: https://github.com/apache/tvm/pull/8568
[8638]: https://github.com/apache/tvm/pull/8638
[8724]: https://github.com/apache/tvm/pull/8724
[8735]: https://github.com/apache/tvm/pull/8735
[8825]: https://github.com/apache/tvm/pull/8825
[8829]: https://github.com/apache/tvm/pull/8829
[8832]: https://github.com/apache/tvm/pull/8832
[8373]: https://github.com/apache/tvm/pull/8373
[8382]: https://github.com/apache/tvm/pull/8382
[8514]: https://github.com/apache/tvm/pull/8514
[8567]: https://github.com/apache/tvm/pull/8567
[8590]: https://github.com/apache/tvm/pull/8590
[8652]: https://github.com/apache/tvm/pull/8652
[8656]: https://github.com/apache/tvm/pull/8656
[8657]: https://github.com/apache/tvm/pull/8657
[8670]: https://github.com/apache/tvm/pull/8670
[8682]: https://github.com/apache/tvm/pull/8682
[8712]: https://github.com/apache/tvm/pull/8712
[8734]: https://github.com/apache/tvm/pull/8734
[8721]: https://github.com/apache/tvm/pull/8721
[8742]: https://github.com/apache/tvm/pull/8742
[8786]: https://github.com/apache/tvm/pull/8786
[8726]: https://github.com/apache/tvm/pull/8726
[8727]: https://github.com/apache/tvm/pull/8727
[8745]: https://github.com/apache/tvm/pull/8745
[8771]: https://github.com/apache/tvm/pull/8771
[8776]: https://github.com/apache/tvm/pull/8776
[8784]: https://github.com/apache/tvm/pull/8784
[8796]: https://github.com/apache/tvm/pull/8796
[8853]: https://github.com/apache/tvm/pull/8853
[8801]: https://github.com/apache/tvm/pull/8801
[8826]: https://github.com/apache/tvm/pull/8826
[8831]: https://github.com/apache/tvm/pull/8831
[8840]: https://github.com/apache/tvm/pull/8840
[8841]: https://github.com/apache/tvm/pull/8841
[8888]: https://github.com/apache/tvm/pull/8888
[8383]: https://github.com/apache/tvm/pull/8383
[8532]: https://github.com/apache/tvm/pull/8532
[8542]: https://github.com/apache/tvm/pull/8542
[8621]: https://github.com/apache/tvm/pull/8621
[8809]: https://github.com/apache/tvm/pull/8809
[8827]: https://github.com/apache/tvm/pull/8827
[8844]: https://github.com/apache/tvm/pull/8844
[8850]: https://github.com/apache/tvm/pull/8850
[8451]: https://github.com/apache/tvm/pull/8451
[8812]: https://github.com/apache/tvm/pull/8812
[8659]: https://github.com/apache/tvm/pull/8659
[8750]: https://github.com/apache/tvm/pull/8750
[8722]: https://github.com/apache/tvm/pull/8722
[8779]: https://github.com/apache/tvm/pull/8779
[8837]: https://github.com/apache/tvm/pull/8837
[8789]: https://github.com/apache/tvm/pull/8789
[8787]: https://github.com/apache/tvm/pull/8787
[8818]: https://github.com/apache/tvm/pull/8818
[8658]: https://github.com/apache/tvm/pull/8658
[8864]: https://github.com/apache/tvm/pull/8864
[8859]: https://github.com/apache/tvm/pull/8859