TVM Monthly - July 2021

喜欢ヅ旅行 2022-09-09 14:59 14阅读 0赞

# [TVM Monthly - July 2021][] #

As discussed by the TVM PPMC, 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 July of 2021 we welcomed many new contributors to the project. Importantly we welcomed [@comaniac][comaniac] , [@junrushao1994][junrushao1994] as new PMC members! Thanks to everyone for the hard work and contributions! Lots of discussion around RFCs happens in the [new RFC process][], like [Automatic Mixed Precision Pass 7][Automatic Mixed Precision Pass_7], [Static Memory Planing 7][Automatic Mixed Precision Pass_7], etc. Welcome to check out。

This forum got 129k pageviews, 2.9k user visits in the last month.

More improvements along with details are listed below.

# Relay #

*  Fix type relation for batch\_matmul [\#8376 1][8376_1]
 *  Resize 1D [\#8346][8346]
 *  Fix index order in conv2d computation for Arm CPU. [\#8361][8361]
 *  Add support of conv2d with NHWC for Mali [\#8422][8422]
 *  Add support of conv2d with NHWC for Bifrost [\#8430][8430]
 *  Modify create\_executor to pass params [\#8418][8418]
 *  Batch\_matmul to dense optimization [\#8440][8440]
 *  Add ConvInteger support. [\#8456][8456]
 *  Add RandomUniform converter and tests to onnx frontend. [\#8426][8426]
 *  Allow importing models with malformed Loop nodes. [\#8475][8475]
 *  Switch from CompileEngine to TECompiler in Interpreter [\#8486][8486]
 *  Support resize in the ONNX conversion [\#8455][8455]
 *  Fix bug in test\_op\_level3 [\#8508][8508]
 *  Extend FakeQuantizationToInteger to more ops [\#8241][8241]
 *  Change Default "opt\_level" of Sequential from 2 to 0 [\#8634][8634]

# Refactor #

*  Remove dead code from depthwise\_conv2d for Intel graphics [\#8381][8381]
 *  Enforce attaching storage scope to PointerType [\#8366][8366]
 *  Remove scope attribute from Buffer class [\#8463][8463]
 *  Remove AttrStmt with storage\_scope key [\#8516][8516]
 *  Unify the shared pass prefix between vm and graph [\#8526][8526]
 *  Avoid Override Generic Op Strategy in "hls.py" [\#8614][8614]

# Docs #

*  Fix for broken link in apps for wasm-standalone dir [\#8045][8045]
 *  Add docs for Pass Instrument [\#8220][8220]
 *  Corrected typo in googletest build instructions. [\#8459][8459]
 *  Fix scipy docs inv [\#8619][8619]
 *  TVM install addenda for M1 Macs [\#8568][8568]

# TIR #

*  fix storage rewrite index remap [\#8338][8338]
 *  Bugfix for zero number arguments tir functions. [\#8515][8515]
 *  cast disparate floating point types for binary ops [\#8517 1][8517_1]
 *  specialize [\#8354][8354]

# Microtvm #

*  Add Nucleo stm32l4r5zi board to zephyr [\#8386 1][8386_1]
 *  Add fixture to zephyr test [\#8393][8393]
 *  Fix Stack Size Issue for Zephyr AOT Demo on Physical Hardware [\#8453][8453]
 *  Fix clock skew on virtualbox [\#8395][8395]
 *  Add zephyr cortex-r5 board to Zephyr [\#8519 1][8519_1]
 *  Set the number of cores based on the VM sizing [\#8624][8624]
 *  Fix platform name in base-box-tool [\#8612][8612]

# Onnx #

*  Wrap 'If' if it has multiple outputs [\#8385][8385]
 *  Parametrize ONNX Unit tests [\#8621][8621]

# Cuda #

*  dense\_tensorcore/batch\_matmul\_tensorcore support int8/int4 [\#8402][8402]
 *  Improve injective schedule to enable half2 [\#8457][8457]
 *  Initial support for dynamic shared memory [\#8466][8466]
 *  Support multiple TIR-level dynamic shared memory allocations [\#8571][8571]

# Topi #

*  Bugfix for topi.prod [\#8416][8416]
 *  Add support for arbitrary dtypes to CSRMV and CSRMM [\#8437][8437]
 *  Parameterize conv2d and depthwise\_conv2d tests [\#8433][8433]
 *  minor change on assert statement in conv2d\_NCHWc\_int8.cuda [\#8554][8554]
 *  Fix `nn.pool*d` issue with 'vectorize' function and add unit tests [\#8541][8541]
 *  Add transpose\_a/b & dynamic shape support for batch matmul [\#8527][8527]
 *  Improve the performance of scatter\_nd [\#8479][8479]

# Frontend #

*  Stridedslice and concat\_v2 fix [\#8483][8483]
 *  Added support for TensorList ops [\#8454][8454]
 *  Check LLVM enabled/installed [\#8414][8414]
 *  Add support for unpack with dim 0 after tensorlist stack [\#8558][8558]
 *  Vc/pytorch lstm [\#8447][8447]

# Fixes #

*  Minimal type checking on TIR schedule [\#8367][8367]
 *  Update the tvmc tutorial with additional requirements [\#8334][8334]
 *  Add pass for splitting kernel with huge number of args [\#8313][8313]
 *  Minor bugfix to arm\_compute\_lib bulid scripts [\#8377][8377]
 *  Add support for log\_softmax [\#8369][8369]
 *  Allow multiprocessing spawn to work (on macOS llvm at least) [\#8363][8363]
 *  Allow tvmc to compile models with AOT executor in MLF [\#8331][8331]
 *  Support QLinearAdd from onnx runtime com.microsoft contrib ops. [\#8305][8305]
 *  Fix np.int and np.float usage in the tree. [\#8389][8389]
 *  Add "operator" style to Model Library Format [\#8072][8072]
 *  macOS is now supported by TVMC [\#8396][8396]
 *  Remove unused conversion [\#8397][8397]
 *  Support aten::flip [\#8398 1][8398_1]
 *  Inverse affine map [\#8384][8384]
 *  Add Compute Library tests to Jenkins for AArch64 CI [\#8394][8394]
 *  add aten::masked\_fill\_ in pytorch frontend [\#8403][8403]
 *  Cleanup more uses of np.bool and np.int. [\#8399][8399]
 *  Revert "Actually add Compute Library tests to the Jenkins File (\#8394)" [\#8400][8400]
 *  TECompiler: Staged refactor and removal of compile engine [\#7518][7518]
 *  fix keras install [\#8391][8391]
 *  Add missing annotation for requires\_gpu in test\_topi\_dense.py [\#8387][8387]
 *  Minor updates to pass pylint locally. [\#8424][8424]
 *  Fix x86 dense schedule extern ops [\#8420][8420]
 *  Fix Relay pattern rewrite [\#8425][8425]
 *  Simplify MatchFusePattern in InverseAffineMap [\#8427][8427]
 *  Improve XGBTuner document [\#8428][8428]
 *  TVMScript Parser support BufferSlice indices [\#8408][8408]
 *  Replace RuntimeError in \_lookup\_task with deferred error. [\#8421][8421]
 *  fix flaky TF crop\_and\_resize [\#8431][8431]
 *  Fix address and port reported by android\_rpc to tracker [\#8405][8405]
 *  Fix undefined symbols by adding library [\#8446][8446]
 *  Extend type checking and annotation for TIR [\#8429][8429]
 *  Add qnn batch\_matmul operator [\#8401][8401]
 *  Use PAPI to collect hardware performance counters on CPU and CUDA [\#7983 1][7983_1]
 *  Fix cpp\_rpc connection to rpc\_tracker [\#8388][8388]
 *  Minor fixes to unit tests for cudnn/vulkan targets [\#8462][8462]
 *  Add default op attribute registration to \_\_**init** \_\_.py [\#8460][8460]
 *  Fix auto-scheduling after 9c6658721 [\#8478][8478]
 *  FoldScaleAxis became non-recursive [\#8325][8325]
 *  Remove compile\_enginer header [\#8471][8471]
 *  DeviceType enums match dlpack [\#8407][8407]
 *  fix typo [\#8484][8484]
 *  Fix the shape function of conv & Add dynamic support for conv2d nhwc [\#8480][8480]
 *  add multi functions support in partition pass [\#8464][8464]
 *  Fix \_get\_yolo\_detections [\#8477][8477]
 *  apps: microtvm: Disable `CONFIG_FPU `for Zephyr runtime [\#8055][8055]
 *  Support tir.abs node in tvm script [\#8488][8488]
 *  Allow serialization of function attrs which are strings [\#8485][8485]
 *  Re-enable `ref_input` [\#8113][8113]
 *  Fix dynamic batching when use\_implicit\_batch=False [\#8461][8461]
 *  fix zero iter bug in arith [\#8494][8494]
 *  Add missing shape functions for relay.nn operations [\#8489][8489]
 *  Better error message for `src/runtime/module.cc` if function cannot be loaded. [\#8496][8496]
 *  Update Docker CI [\#8193][8193]
 *  Re-enabled tests and updated module hashes [\#8498][8498]
 *  Keep CODEOWNERS file up to date. [\#8500][8500]
 *  Rename runtime-config to executor-config and add documentation for Model Library Format [\#8270][8270]
 *  Enable ONNX tests that needed onnxruntime 1.7.0 [\#8502][8502]
 *  Fix \#8093, Enhance Buffer Index Simplify [\#8204][8204]
 *  Organize CodeOwners File [\#8512][8512]
 *  Fuse, Split [\#8467][8467]
 *  Fix script printters StructuralEqual check failed [\#8499][8499]
 *  Add json output to profiling reports [\#8503][8503]
 *  Fix the repeatitive cast in scripr printing [\#8531][8531]
 *  Fix TypeKey2Index when for root Object [\#8547 1][8547_1]
 *  Split out libinfo.cc into a separate target. [\#8520][8520]
 *  Mimic the TFLite 2.4 reader's behaviour [\#8538][8538]
 *  Remove unused variable in topi cpp test [\#8549][8549]
 *  Add explicit type cast to print. [\#8524][8524]
 *  Specifically check handle for recursion during shutdown [\#8548][8548]
 *  Add a `--context-path` for build.sh [\#8557][8557]
 *  Handling a corner case in TRT RemoveDropout pass [\#8506][8506]
 *  Re-enable Compute library tests. [\#8573][8573]
 *  Fix AutoScheduler test to cover Conv2D Winograd [\#8539][8539]
 *  Fix Coreml Input Shape Handling [\#8562][8562]
 *  Fix task extraction with TE compiler [\#8560][8560]
 *  add support for softmax and log\_softmax with MIOpen [\#8543][8543]
 *  Added default non-verbose to download\_testdata(), pass to download() [\#8533][8533]
 *  Disable pip cache when creating Docker images [\#8575][8575]
 *  wasm32-standalone app repaired [\#8563][8563]
 *  Bug fix for numpy scalar input in vm [\#8553][8553]
 *  Reduce testing time of LSTM tests [\#8583][8583]
 *  Prioritize discrete GPUs as device\_id=0. [\#8588][8588]
 *  speed up reference resize kernel [\#8592][8592]
 *  Delete pytest-results as part of CI workspace preparation [\#8594][8594]
 *  Use SizeVar instead of Var when convert Any in the GetShape function [\#8555][8555]
 *  Fix storage\_access not visiting else branch [\#8525][8525]
 *  Reduction Factoring (RFactor) [\#8544][8544]
 *  Support for match\_buffer from subregion [\#8585][8585]
 *  Recover rpc server support [\#8604][8604]
 *  Add caching to CMake [\#8373][8373]
 *  Add support for AOT in external code generation tests [\#8591][8591]
 *  Fix global pip cache disable change [\#8590][8590]
 *  Fix Initial Memory Misalignment [\#8487][8487]
 *  Remove QEMU Install [\#8518][8518]
 *  Remove unused parameter. [\#8580][8580]
 *  Docker env for Arm® Ethos™-U55 Port [\#8514][8514]
 *  Instruction and Trace [\#8615][8615]
 *  Introduce --interface-api=\{c,packed\} parameter [\#8280][8280]
 *  Fix test\_external\_codegen, broken by \#8591 [\#8630][8630]
 *  Rewrote PointerValueTypeRewrite transform [\#8528][8528]
 *  Framework for device querying for all targets. [\#8602][8602]
 *  Add graph\_executor get\_input\_index API. [\#8633 1][8633_1]
 *  Disallow fp16 conversion for arange op [\#8644][8644]
 *  Allow spaces in target attributes [\#8587][8587]
 *  Several minor corrections to the device property query [\#8651][8651]
 *  Fix depthwise conv2d on non-cuda GPU platforms [\#8379][8379]
 *  Fix wrong log of tir pass VerifyMemory [\#8445][8445]
 *  Explicitly retain `__hash__` of `StringImm` [\#8449][8449]
 *  Update stale relay.Module API in docs/comments [\#8411][8411]
 *  Remove unused variable in GraphExecutorCodegen [\#8465][8465]
 *  Compiler supports input with a slash [\#8481][8481]
 *  Minor misspelling [\#8476][8476]
 *  Enhance robustness of DefuseOps [\#8564][8564]
 *  Add USE\_PAPI configuration to config.cmake [\#8567][8567]
 *  Fix a typo in include/tvm/ir/function.h [\#8617][8617]
 *  hotfix check\_grad perf regression [\#8581][8581]
 *  Fix broadcast type func with incomplete type [\#8438][8438]
 *  Fix the integer overflow problem of the scatter\_nd op. [\#8415][8415]
 *  do not simplify 'Any() - Any()' to 0 [\#8266][8266]
 *  Visit each input param of the function in ExprVisitor visit\_function [\#8521][8521]
 *  Correct class number in Golang frontend sample [\#8511][8511]
 *  fix android rpc app undefined reference problem [\#8530][8530]
 *  fix illegal memory access bug in reduce op schedule by constriant thread\_y [\#8566][8566]
 *  Preserve IRModule type definition and imports in NameMangleExtFuncs [\#8523][8523]
 *  Fix \#8536 Get Target When Heterogeneous Execution [\#8537][8537]

[TVM Monthly - July 2021]: https://discuss.tvm.apache.org/t/tvm-monthly-july-2021/10701
[comaniac]: https://discuss.tvm.apache.org/u/comaniac
[junrushao1994]: https://discuss.tvm.apache.org/u/junrushao1994
[new RFC process]: https://github.com/apache/tvm-rfcs
[Automatic Mixed Precision Pass_7]: https://github.com/apache/tvm-rfcs/pull/6
[8376_1]: https://github.com/apache/tvm/pull/8376
[8346]: https://github.com/apache/tvm/pull/8346
[8361]: https://github.com/apache/tvm/pull/8361
[8422]: https://github.com/apache/tvm/pull/8422
[8430]: https://github.com/apache/tvm/pull/8430
[8418]: https://github.com/apache/tvm/pull/8418
[8440]: https://github.com/apache/tvm/pull/8440
[8456]: https://github.com/apache/tvm/pull/8456
[8426]: https://github.com/apache/tvm/pull/8426
[8475]: https://github.com/apache/tvm/pull/8475
[8486]: https://github.com/apache/tvm/pull/8486
[8455]: https://github.com/apache/tvm/pull/8455
[8508]: https://github.com/apache/tvm/pull/8508
[8241]: https://github.com/apache/tvm/pull/8241
[8634]: https://github.com/apache/tvm/pull/8634
[8381]: https://github.com/apache/tvm/pull/8381
[8366]: https://github.com/apache/tvm/pull/8366
[8463]: https://github.com/apache/tvm/pull/8463
[8516]: https://github.com/apache/tvm/pull/8516
[8526]: https://github.com/apache/tvm/pull/8526
[8614]: https://github.com/apache/tvm/pull/8614
[8045]: https://github.com/apache/tvm/pull/8045
[8220]: https://github.com/apache/tvm/pull/8220
[8459]: https://github.com/apache/tvm/pull/8459
[8619]: https://github.com/apache/tvm/pull/8619
[8568]: https://github.com/apache/tvm/pull/8568
[8338]: https://github.com/apache/tvm/pull/8338
[8515]: https://github.com/apache/tvm/pull/8515
[8517_1]: https://github.com/apache/tvm/pull/8517
[8354]: https://github.com/apache/tvm/pull/8354
[8386_1]: https://github.com/apache/tvm/pull/8386
[8393]: https://github.com/apache/tvm/pull/8393
[8453]: https://github.com/apache/tvm/pull/8453
[8395]: https://github.com/apache/tvm/pull/8395
[8519_1]: https://github.com/apache/tvm/pull/8519
[8624]: https://github.com/apache/tvm/pull/8624
[8612]: https://github.com/apache/tvm/pull/8612
[8385]: https://github.com/apache/tvm/pull/8385
[8621]: https://github.com/apache/tvm/pull/8621
[8402]: https://github.com/apache/tvm/pull/8402
[8457]: https://github.com/apache/tvm/pull/8457
[8466]: https://github.com/apache/tvm/pull/8466
[8571]: https://github.com/apache/tvm/pull/8571
[8416]: https://github.com/apache/tvm/pull/8416
[8437]: https://github.com/apache/tvm/pull/8437
[8433]: https://github.com/apache/tvm/pull/8433
[8554]: https://github.com/apache/tvm/pull/8554
[8541]: https://github.com/apache/tvm/pull/8541
[8527]: https://github.com/apache/tvm/pull/8527
[8479]: https://github.com/apache/tvm/pull/8479
[8483]: https://github.com/apache/tvm/pull/8483
[8454]: https://github.com/apache/tvm/pull/8454
[8414]: https://github.com/apache/tvm/pull/8414
[8558]: https://github.com/apache/tvm/pull/8558
[8447]: https://github.com/apache/tvm/pull/8447
[8367]: https://github.com/apache/tvm/pull/8367
[8334]: https://github.com/apache/tvm/pull/8334
[8313]: https://github.com/apache/tvm/pull/8313
[8377]: https://github.com/apache/tvm/pull/8377
[8369]: https://github.com/apache/tvm/pull/8369
[8363]: https://github.com/apache/tvm/pull/8363
[8331]: https://github.com/apache/tvm/pull/8331
[8305]: https://github.com/apache/tvm/pull/8305
[8389]: https://github.com/apache/tvm/pull/8389
[8072]: https://github.com/apache/tvm/pull/8072
[8396]: https://github.com/apache/tvm/pull/8396
[8397]: https://github.com/apache/tvm/pull/8397
[8398_1]: https://github.com/apache/tvm/pull/8398
[8384]: https://github.com/apache/tvm/pull/8384
[8394]: https://github.com/apache/tvm/pull/8394
[8403]: https://github.com/apache/tvm/pull/8403
[8399]: https://github.com/apache/tvm/pull/8399
[8400]: https://github.com/apache/tvm/pull/8400
[7518]: https://github.com/apache/tvm/pull/7518
[8391]: https://github.com/apache/tvm/pull/8391
[8387]: https://github.com/apache/tvm/pull/8387
[8424]: https://github.com/apache/tvm/pull/8424
[8420]: https://github.com/apache/tvm/pull/8420
[8425]: https://github.com/apache/tvm/pull/8425
[8427]: https://github.com/apache/tvm/pull/8427
[8428]: https://github.com/apache/tvm/pull/8428
[8408]: https://github.com/apache/tvm/pull/8408
[8421]: https://github.com/apache/tvm/pull/8421
[8431]: https://github.com/apache/tvm/pull/8431
[8405]: https://github.com/apache/tvm/pull/8405
[8446]: https://github.com/apache/tvm/pull/8446
[8429]: https://github.com/apache/tvm/pull/8429
[8401]: https://github.com/apache/tvm/pull/8401
[7983_1]: https://github.com/apache/tvm/pull/7983
[8388]: https://github.com/apache/tvm/pull/8388
[8462]: https://github.com/apache/tvm/pull/8462
[8460]: https://github.com/apache/tvm/pull/8460
[8478]: https://github.com/apache/tvm/pull/8478
[8325]: https://github.com/apache/tvm/pull/8325
[8471]: https://github.com/apache/tvm/pull/8471
[8407]: https://github.com/apache/tvm/pull/8407
[8484]: https://github.com/apache/tvm/pull/8484
[8480]: https://github.com/apache/tvm/pull/8480
[8464]: https://github.com/apache/tvm/pull/8464
[8477]: https://github.com/apache/tvm/pull/8477
[8055]: https://github.com/apache/tvm/pull/8055
[8488]: https://github.com/apache/tvm/pull/8488
[8485]: https://github.com/apache/tvm/pull/8485
[8113]: https://github.com/apache/tvm/pull/8113
[8461]: https://github.com/apache/tvm/pull/8461
[8494]: https://github.com/apache/tvm/pull/8494
[8489]: https://github.com/apache/tvm/pull/8489
[8496]: https://github.com/apache/tvm/pull/8496
[8193]: https://github.com/apache/tvm/pull/8193
[8498]: https://github.com/apache/tvm/pull/8498
[8500]: https://github.com/apache/tvm/pull/8500
[8270]: https://github.com/apache/tvm/pull/8270
[8502]: https://github.com/apache/tvm/pull/8502
[8204]: https://github.com/apache/tvm/pull/8204
[8512]: https://github.com/apache/tvm/pull/8512
[8467]: https://github.com/apache/tvm/pull/8467
[8499]: https://github.com/apache/tvm/pull/8499
[8503]: https://github.com/apache/tvm/pull/8503
[8531]: https://github.com/apache/tvm/pull/8531
[8547_1]: https://github.com/apache/tvm/pull/8547
[8520]: https://github.com/apache/tvm/pull/8520
[8538]: https://github.com/apache/tvm/pull/8538
[8549]: https://github.com/apache/tvm/pull/8549
[8524]: https://github.com/apache/tvm/pull/8524
[8548]: https://github.com/apache/tvm/pull/8548
[8557]: https://github.com/apache/tvm/pull/8557
[8506]: https://github.com/apache/tvm/pull/8506
[8573]: https://github.com/apache/tvm/pull/8573
[8539]: https://github.com/apache/tvm/pull/8539
[8562]: https://github.com/apache/tvm/pull/8562
[8560]: https://github.com/apache/tvm/pull/8560
[8543]: https://github.com/apache/tvm/pull/8543
[8533]: https://github.com/apache/tvm/pull/8533
[8575]: https://github.com/apache/tvm/pull/8575
[8563]: https://github.com/apache/tvm/pull/8563
[8553]: https://github.com/apache/tvm/pull/8553
[8583]: https://github.com/apache/tvm/pull/8583
[8588]: https://github.com/apache/tvm/pull/8588
[8592]: https://github.com/apache/tvm/pull/8592
[8594]: https://github.com/apache/tvm/pull/8594
[8555]: https://github.com/apache/tvm/pull/8555
[8525]: https://github.com/apache/tvm/pull/8525
[8544]: https://github.com/apache/tvm/pull/8544
[8585]: https://github.com/apache/tvm/pull/8585
[8604]: https://github.com/apache/tvm/pull/8604
[8373]: https://github.com/apache/tvm/pull/8373
[8591]: https://github.com/apache/tvm/pull/8591
[8590]: https://github.com/apache/tvm/pull/8590
[8487]: https://github.com/apache/tvm/pull/8487
[8518]: https://github.com/apache/tvm/pull/8518
[8580]: https://github.com/apache/tvm/pull/8580
[8514]: https://github.com/apache/tvm/pull/8514
[8615]: https://github.com/apache/tvm/pull/8615
[8280]: https://github.com/apache/tvm/pull/8280
[8630]: https://github.com/apache/tvm/pull/8630
[8528]: https://github.com/apache/tvm/pull/8528
[8602]: https://github.com/apache/tvm/pull/8602
[8633_1]: https://github.com/apache/tvm/pull/8633
[8644]: https://github.com/apache/tvm/pull/8644
[8587]: https://github.com/apache/tvm/pull/8587
[8651]: https://github.com/apache/tvm/pull/8651
[8379]: https://github.com/apache/tvm/pull/8379
[8445]: https://github.com/apache/tvm/pull/8445
[8449]: https://github.com/apache/tvm/pull/8449
[8411]: https://github.com/apache/tvm/pull/8411
[8465]: https://github.com/apache/tvm/pull/8465
[8481]: https://github.com/apache/tvm/pull/8481
[8476]: https://github.com/apache/tvm/pull/8476
[8564]: https://github.com/apache/tvm/pull/8564
[8567]: https://github.com/apache/tvm/pull/8567
[8617]: https://github.com/apache/tvm/pull/8617
[8581]: https://github.com/apache/tvm/pull/8581
[8438]: https://github.com/apache/tvm/pull/8438
[8415]: https://github.com/apache/tvm/pull/8415
[8266]: https://github.com/apache/tvm/pull/8266
[8521]: https://github.com/apache/tvm/pull/8521
[8511]: https://github.com/apache/tvm/pull/8511
[8530]: https://github.com/apache/tvm/pull/8530
[8566]: https://github.com/apache/tvm/pull/8566
[8523]: https://github.com/apache/tvm/pull/8523
[8537]: https://github.com/apache/tvm/pull/8537