features/bladedisc rebase 20220830 #20

tanyokwok · 2022-09-21T09:58:08Z

Change to notifyMatchFailure (Change to notifyMatchFailure llvm/torch-mlir#1073)
[MLIR][TORCH] Add canonicalization pattern for prim.ListUnpack op
[MLIR][TORCH] Add negative dim support for aten.cat and aten.slice op
bump llvm tag to 3580daa (bump llvm tag to 3580daa llvm/torch-mlir#1078)
Add support for index.Tensor on dimensions other than the first
Emit underscore version of aten.sqrt (Emit underscore version of aten.sqrt llvm/torch-mlir#1072)
torch: [nfc] use WalkResult::isInterrupted() instead of booleans (torch: [nfc] use WalkResult::isInterrupted() instead of booleans llvm/torch-mlir#1081)
Allow expanding and collapsing in aten::view (Allow expanding and collapsing in aten::view llvm/torch-mlir#1082)
[MHLO] Init MHLO integration. ([MHLO] Init MHLO integration. llvm/torch-mlir#1083)
python: trim registration and loading of dialects and passes (python: trim registration and loading of dialects and passes llvm/torch-mlir#1084)
Revert "python: trim registration and loading of dialects and passes" (Revert "python: trim registration and loading of dialects and passes" llvm/torch-mlir#1093)
Improve error message
Add linux and macOS source builds in CI (Add linux source builds in CI llvm/torch-mlir#1070)
Add verifiers for ToBuiltinTensorOp and FromBuiltinTensorOp (Add verifiers for ToBuiltinTensorOp and FromBuiltinTensorOp llvm/torch-mlir#1089)
[MHLO] Init MHLO slice like op patterns ([MHLO] Init MHLO slice like op patterns llvm/torch-mlir#1091)
[MHLO] Init MHLO view like op patterns ([MHLO] Init MHLO view like op patterns llvm/torch-mlir#1090)
Add option to expose custom PyTorch repo/branch (Add option to expose custom PyTorch repo/branch llvm/torch-mlir#1103)
[MHLO] Add [un]squeeze op patterns ([MHLO] Add [un]squeeze op patterns llvm/torch-mlir#1099)
[MHLO] Init MHLO gather op patterns ([MHLO] Init MHLO gather op patterns llvm/torch-mlir#1104)
Add lowering to linalg for softplus and log1p
Update development.md with source builds (Update development.md with source builds llvm/torch-mlir#1105)
[MHLO] Init MHLO basic op conversion ([MHLO] Init MHLO basic op conversion llvm/torch-mlir#1092)
Add e2e support for aten.expm1
README: Add op office hours
[MLIR][TORCH] Add support for multiple indexing tensors for aten.index.Tensor ([MLIR][TORCH] Add support for multiple indexing tensors for aten.index.Tensor llvm/torch-mlir#1097)
[MLIR][TORCH] Add E2E support for aten.ge.int op
[MLIR][TORCH] Add decomposition for aten.var.correction op
[MLIR][TORCH} Fix empty dim cases for the .dim ops
Add mhlo to bazel build (Add mhlo to bazel build llvm/torch-mlir#1120)
Fix CI failure due to upstream PyTorch change in aten.mean.dim op
buildAndTest.yml for matrix builds (buildAndTest.yml for matrix builds llvm/torch-mlir#1098)
[MLIR][TORCH] Add decomposition for aten.std.dim Op
Add initial LTC backend (Add initial LTC backend llvm/torch-mlir#610)
Torch-MLIR LTC Backend Lowering Codegen (Torch-MLIR LTC Backend Lowering Codegen llvm/torch-mlir#621)
Got LTC working until compile (Got LTC working until compile llvm/torch-mlir#689)
Fix Torch-MLIR LTC Backend based off latest PyTorch master ( Decouple MLIR Backend from TS Backend llvm/torch-mlir#723)
Added JIT to MLIR lowering (Added JIT to MLIR lowering llvm/torch-mlir#724)
Add example Torch MLIR LTC Backend (Added example Torch MLIR LTC Backend llvm/torch-mlir#725)
Generate MLIR with shape information via LTC frontend (Generate MLIR with shape information via LTC frontend llvm/torch-mlir#742)
Enable support for LTC Input/Output Mapping (Enable support for LTC Input/Output Mapping llvm/torch-mlir#764)
Fix LTC Decoupling (Fix LTC Decoupling llvm/torch-mlir#815)
Use JIT GraphExecutor for execution in example backend (Use JIT GraphExecutor for execution in example backend llvm/torch-mlir#830)
Bert example and relevant shape inference functions (Enable BERT support for torch-mlir via LTC llvm/torch-mlir#831)
Add static shape for scalar tensors (Add static shape for scalar tensors llvm/torch-mlir#833)
E2E HuggingFace Bert using LTC Backend (E2E HuggingFace Bert using LTC Backend llvm/torch-mlir#912)
Refactor Node Lowering (Refactor Node Lowering llvm/torch-mlir#914)
Added e2e LTC tests (Added e2e LTC tests llvm/torch-mlir#916)
Refactor autogen (Refactor autogen llvm/torch-mlir#925)
Only import the LTC backend that's used (Only import the LTC backend that's used in example models llvm/torch-mlir#939)
Upstream native_batch_norm and native_batch_norm_backward shape inference functions (Upstream native_batch_norm and native_batch_norm_backward shape inference functions llvm/torch-mlir#978)
Integrate Functionalization Pass (Integrate Functionalization Pass llvm/torch-mlir#998)
Update README.md to reflect LTC landing in main branch (Update README.md to reflect LTC landing in main branch llvm/torch-mlir#1000)
LTC Documentation (LTC Documentation llvm/torch-mlir#1021)
Update Torch MLIR readme
Prune xfail e2e LTC tests & fix bugs from functionalization pass (Prune xfail e2e LTC tests & fix bugs from functionalization pass llvm/torch-mlir#1044)
Reference Lazy Backend (Reference Lazy Backend llvm/torch-mlir#1045)
Blacklist _convolution op (Blacklist _convolution op llvm/torch-mlir#1048)
Fix LTC autogen for CI with nightly PyTorch
Default Device Ordinal API (Default Device Ordinal API llvm/torch-mlir#1079)
Add support for lift_fresh op (Add support for lift_fresh op llvm/torch-mlir#1101)
Resolve CI testing failure for Lazy Tensor Core (Resolve CI testing failure for Lazy Tensor Core llvm/torch-mlir#1088)
Export LTC Headers (Export LTC Headers llvm/torch-mlir#1108)
Clean up Autogen (Clean up Autogen llvm/torch-mlir#1112)
Resolve remaining LTC CI failures (Resolve remaining LTC CI failures llvm/torch-mlir#1110)
llvm: update tag to 02b3a35 (llvm: update tag to 02b3a35 llvm/torch-mlir#1124)
Disable LTC on Release builds (Disable LTC on Release builds llvm/torch-mlir#1125)
Implemented a decomposition of aten::narrow
E2E support for AtenEmbeddingBagPaddingIdxOp SUM Mode (Add E2E support for AtenEmbeddingBagPaddingIdxOp llvm/torch-mlir#1066)
Add embedding Bag e2e case in xfail set (Add embedding Bag e2e case in xfail set llvm/torch-mlir#1130)
[NFC][MHLO] move util funcs to MhloLegalizeUtils.h/cpp ([NFC][MHLO] move util funcs to MhloLegalizeUtils.h/cpp llvm/torch-mlir#1128)
[MHLO] Support for dynamic shape in basic op conversion by introducing CHLO dialect ([MHLO] Support for dynamic shape in basic op conversion by introducing CHLO dialect llvm/torch-mlir#1123)
[MHLO] add aten::gelu op pattern ([MHLO] add aten::gelu op pattern llvm/torch-mlir#1127)
add e2e support for aten.atan2 ([MLIR][TORCH] Add e2e support for aten.atan2 llvm/torch-mlir#1117)
Add support for dim=None to AtenMeanDimOp (Add support for dim=None to AtenMeanDimOp llvm/torch-mlir#1129)
Fix mhlo bazel rule name (Fix mhlo bazel rule name llvm/torch-mlir#1136)
[MHLO] support non-constant torch scalar in BasicOps ([MHLO] support non-constant torch scalar in BasicOps llvm/torch-mlir#1134)
[MHLO] Init MHLO reduce-like op conversion ([MHLO] Init MHLO reduce-like op conversion llvm/torch-mlir#1133)
[MLIR][TORCH] Fix CI failure
Fix cache-suffix name bug (Fix cache-suffix name bug llvm/torch-mlir#1138)
Update buildAndTest.yml (Update buildAndTest.yml llvm/torch-mlir#1145)
[Fix bazel] Add recently added torch->mhlo conversion pass to bazel ([bazel build] Add recently added torch->mhlo conversion pass to bazel llvm/torch-mlir#1148)
[MHLO] Init MHLO linear op patterns ([MHLO] Init MHLO linear op patterns llvm/torch-mlir#1132)
[MHLO] Init MHLO pooling-like op conversion ([MHLO] Init MHLO pooling-like op conversion llvm/torch-mlir#1141)
Add non-unit groups support to aten.convolution (Add non-unit groups to aten::convolution llvm/torch-mlir#858)
[MHLO] Add convolution op pattern ([MHLO] Add convolution op pattern llvm/torch-mlir#1152)
Make numerical stability test more perverse
torch_mlir.compile: Allow ignoring traced shapes
[MLIR][TORCH] Add support for dim=None to Aten[Var|Std]DimOp
E2E framework: Report tensor dtype in summary
Update LTC CMake hack documentation (Update LTC CMake hack documentation llvm/torch-mlir#1155)
development.md: How to enable ASan
development.md: Avoid name collisions with PYTORCH_ variables
[MHLO] fix tensor mode aten.div op pattern ([MHLO] fix tensor mode aten.div op pattern llvm/torch-mlir#1160)
importer: add initial support for loading Float16 tensors (importer: add initial support for loading Float16 tensors llvm/torch-mlir#1169)
Reenable LTC in out-of-tree build (Reenable LTC in out-of-tree build llvm/torch-mlir#1177)
Add decomposition and E2E support for Aten_EmbeddingBag (Add decomposition and E2E support for Aten_EmbeddingBag llvm/torch-mlir#1137)
E2E support for AtenRemainderScalarOp (E2E support for AtenRemainderScalarOp llvm/torch-mlir#1119)
Rework how global slot initializers work.
[MHLO] Add transposed convolution conversion pattern ([MHLO] Add transposed convolution conversion pattern llvm/torch-mlir#1171)
Revert "Reenable LTC in out-of-tree build (Reenable LTC in out-of-tree build llvm/torch-mlir#1177)" (Revert "Reenable LTC in out-of-tree build" llvm/torch-mlir#1183)
llvm: update tag to 061e0189 (llvm: update tag to 061e0189 llvm/torch-mlir#1180)
Revert "E2E support for AtenRemainderScalarOp (E2E support for AtenRemainderScalarOp llvm/torch-mlir#1119)" (Revert "E2E support for AtenRemainderScalarOp" llvm/torch-mlir#1190)
[MHLO]fix lowering failed on reduction op with i32 shape ([MHLO]fix lowering failed on reduction op with i32 shape llvm/torch-mlir#1185)
Add CMake dep to Func dialect (Add CMake dep to Func dialect llvm/torch-mlir#1196)
Don't set MLIR_TABLEGEN_EXE (Don't set MLIR_TABLEGEN_EXE llvm/torch-mlir#1197)
Flip to C++17 (Flip to C++17 llvm/torch-mlir#1198)
Enable OOT builds in CI (Include torch-mlir builds in OOT workflow llvm/torch-mlir#1188)
[Bazel] Allow workflow_dispatch manual trigger on bazel workflow ([Bazel] Allow workflow_dispatch manual trigger on bazel workflow llvm/torch-mlir#1203)
[Bazel] Add EraseModuleInitializer to TorchMLIRTorchPasses library ([Bazel] Add EraseModuleInitializer to TorchMLIRTorchPasses library llvm/torch-mlir#1202)
Disable out-of-tree and PyTorch binary (Disable out-of-tree and PyTorch binary llvm/torch-mlir#1206)
[MHLO] Add AtenCatOp conversion pattern to MHLO ([MHLO] Add AtenCatOp conversion pattern to MHLO llvm/torch-mlir#1208)
mac m1 cross compile (mac m1 cross compile llvm/torch-mlir#1204)
Rename TorchToStd pass as TorchToArith (Rename TorchToStd pass as TorchToArith llvm/torch-mlir#1163)
Add note about MLIR compiled outputs in dev docs (Add note about MLIR compiled outputs in dev docs llvm/torch-mlir#1195)
Update TorchToStd to TorchtoArith in bazel files too. (Update TorchToStd to TorchtoArith in bazel files too. llvm/torch-mlir#1210)
Add lowering for aten.to.device (Add lowering for aten.to.device llvm/torch-mlir#1107)
E2E support for AtenRemainderScalarOp (E2E support for AtenRemainderScalarOp llvm/torch-mlir#1200)
remove torch dialect from legal list ([MHLO]remove torch dialect from legal list llvm/torch-mlir#1192)
Add decomposition of aten.masked.tensor op.
Add PyYaml to requirements.txt (Add PyYaml to requirements.txt llvm/torch-mlir#1174)
Simplify matrix configuration for CI workflows (Simplify matrix configuration for CI workflows llvm/torch-mlir#1213)
use pytorch binary for macos-arm64 builds (Use pytorch binary for macos-arm64 workflow llvm/torch-mlir#1215)
Merge matrix runs to fail fast globally (Merge matrix runs to fail fast globally llvm/torch-mlir#1216)
build: fix typo in path (build: fix typo in path llvm/torch-mlir#1218)
[Build] Add concurrency groups to address long queue times ([Build] Add concurrency groups to address long queue times llvm/torch-mlir#1219)
build: streamline options for enabling LTC and MHLO (build: streamline options for enabling LTC and MHLO llvm/torch-mlir#1221)
s/external/externals/g (s/external/externals/g llvm/torch-mlir#1222)
add native_dropout and related ops pattern ([MHLO]add native_dropout and related ops pattern llvm/torch-mlir#1211)
Fix unused-variables warnings about EmbeddingBag ops (Fix unused-variables warnings about EmbeddingBag ops llvm/torch-mlir#1220)
Generalize canonicalization pattern for more aten.sub/div/mul/add op (Generalize canonicalization pattern for more aten.sub/div/mul/add op llvm/torch-mlir#1209)
build: update llvm tag to 2dde4ba6 (build: update llvm tag to 2dde4ba6 llvm/torch-mlir#1229)
Propagate device data names (Propagate device data names llvm/torch-mlir#1157)
Re-enable custom op support
[MHLO] Add AtenRSubScalarOp conversion pattern to MHLO ([MHLO] Add AtenRSubScalarOp conversion pattern to MHLO llvm/torch-mlir#1233)
Bump the shape lib to match the upstream functions currently in PyTorch (Bump the shape lib to match the upstream functions currently in PyTorch llvm/torch-mlir#1236)
Revert "add native_dropout and related ops pattern ([MHLO]add native_dropout and related ops pattern llvm/torch-mlir#1211)" (Revert "[MHLO]add native_dropout and related ops pattern" llvm/torch-mlir#1230)
Dockerize and Cache Bazel {Local, CI} Builds (Dockerize and Cache Bazel {Local, CI} Builds llvm/torch-mlir#1240)
Iteratively run the main simplification pipeline.
[Bazel] Check cache directory exists before changing owners ([Bazel] Check cache directory exists before changing owners llvm/torch-mlir#1241)
[Bazel] Add LowerToBackendContract.cpp to TorchMLIRTorchPasses bazel target ([Bazel] Add LowerToBackendContract.cpp to TorchMLIRTorchPasses bazel target llvm/torch-mlir#1243)
Clean up shape functions that use sum_mean_dim (Clean up shape functions that use sum_mean_dim llvm/torch-mlir#1217)
Add a concept of "backend legal ops".
[docs] Add architecture doc.
Add support for returning more than one copy of the same tensor (Add support for returning more than one copy of the same tensor llvm/torch-mlir#1228)
Add white background to diagram.
Drop Python 3.7x from the nightly binary builds (Drop Python 3.7x from the nightly binary builds llvm/torch-mlir#1246)
[MHLO] Eliminate explicit dynamic output shape generating in converting AtenSliceTensorOp ([MHLO] Eliminate explicit dynamic output shape generating in converting AtenSliceTensorOp llvm/torch-mlir#1245)
[Bazel] Run buildifier ([Bazel] Run buildifier llvm/torch-mlir#1250)
[MLIR][TORCH] Fix dynamic cases for aten.index.Tensor
Reenable LTC in out-of-tree build (for real this time) (Reenable LTC in out-of-tree build (for real this time) llvm/torch-mlir#1205)
Add folder for ToF64Op and FromF64Op (Add folder for ToF64Op and FromF64Op llvm/torch-mlir#1257)
Add lowering for _convolution.deprecated (Add lowering for _convolution.deprecated llvm/torch-mlir#1259)
[MLIR][TORCH] Fix CI failure due to failing tests
Don't explicitly set MLIR_PDLL_TABLEGEN_EXE (Don't explicitly set MLIR_PDLL_TABLEGEN_EXE llvm/torch-mlir#1262)
Update Torch-MLIR Architecture Diagram (Update Torch-MLIR Architecture Diagram llvm/torch-mlir#1254)
Add a way for backends to control which ops are legal for them.
Add decomposition for aten.flatten.using_ints (Add decomposition for aten.flatten.using_ints llvm/torch-mlir#1161)
[MLIR][TORCH] Add torch.Device type to backend contract scalar types
[MHLO] Init end to end unit tests ([MHLO] Init end-to-end unit tests llvm/torch-mlir#1223)
[MHLO] bert-tiny and resnet18 example from torchscript to mhlo ([MHLO] bert-tiny and resnet18 e2e example from torchscript to mhlo llvm/torch-mlir#1266)
Add decomposition for aten.roll (Add decomposition for aten.roll llvm/torch-mlir#1170)
[cleanup] Change OutputType enum values to strings
Add transposed case for at::convolution (Add transposed case for at::convolution llvm/torch-mlir#917)
Fix LTC build warnings (Fix LTC build warnings llvm/torch-mlir#1272)
mlir: fix replacement of OpaqueElementsAttr (mlir: fix replacement of OpaqueElementsAttr llvm/torch-mlir#1274)
doc: fix instructions for LLVM and MHLO updates (doc: fix instructions for LLVM and MHLO updates llvm/torch-mlir#1273)
Revert updating mlir_native_functions.cpp signature (Revert updating mlir_native_functions.cpp signature llvm/torch-mlir#1281)
Fix LTC lib_torch_mlir_ltc.so import error (Fix LTC lib_torch_mlir_ltc.so import error llvm/torch-mlir#1283)
Add TestUtils.randint + replace torch.randint with tu.randint (Add TestUtils.randint + replace torch.randint with tu.randint llvm/torch-mlir#1276)
Fix a few build warnings
Remove VerifyInvariantsBeforeBackendLowering
Fix symint related functionalization ops (Fix symint related functionalization ops llvm/torch-mlir#1289)
Add LTC architecture diagram (Add LTC architecture diagram llvm/torch-mlir#1291)
Add documentation for adding E2E tests (Add documentation for adding E2E tests llvm/torch-mlir#1269)
Avoid cascading failures when compiler crashes
Add reference to slides and presentation.
Disable LTC by default until upstream revert relands (Disable LTC by default until upstream revert relands llvm/torch-mlir#1303)
Rename an outdated class name
Add CI for generated files.
Update MHLO xfails.
Move development.md to docs/ for consistency
Remove "torchscript" association from the e2e framework.
Sort tests before anything else.
Ensure that tests have unique names
Dockerize CI + Release builds (Dockerize CI + Release builds llvm/torch-mlir#1234)
Add some missing dependencies.
build: update llvm tag to 00d648bd (build: update llvm tag to 00d648bd llvm/torch-mlir#1307)
Add shapelib and Torch ODS gen tests (Add shapelib and Torch ODS gen tests llvm/torch-mlir#1318)
Use pre-compiled headers for PyTorch Source builds (Use pre-compiled headers for PyTorch Source builds llvm/torch-mlir#1327)
Set ccache logging to verbose temporarily (Set ccache logging to verbose temporarily llvm/torch-mlir#1326)
[MHLO] refactor pass configurations ([MHLO] refactor pass configurations llvm/torch-mlir#1315)
Slightly tweak generated file checks
aicompiler rebase 20220830
To comply with the old pytorch versions
add relu6 op
Add decomposition for aten::native_layer_norm (Add decomposition for aten::native_layer_norm #13)
Add native_dropout_backward & native_layer_norm_backward decomposition (Add native_dropout_backward & native_layer_norm_backward decomposition #15)
add native_dropout and related ops pattern ([MHLO]add native_dropout and related ops pattern llvm/torch-mlir#1211)
Fix dot product result type setting (Fix dot product result type setting #19)
fix BatchNormInference

emitError is intended for error cases and not match failures of patterns. notifyMatchFailure is intended where pattern reports reason for not matching. Op verification should also not happen inside patterns but as part of verify/verification, but left ones that were obviously verification to emitError inside patterns to keep this change small.

This commit adds the canonicalization pattern for the `prim.ListUnpack` op. Signed-Off By: Vivek Khandelwal <vivek@nod-labs.com>

This commit adds the support for negative dim cases for `aten.cat`, `aten.slice.Tensor` and `aten.slice_scatter` op. Signed-Off By: Vivek Khandelwal <vivek@nod-labs.com>

This patch makes some rudimentary changes to torch-mlir's use of MLIR Python bindings to work with the most recent LLVM code. We can perhaps do better by being more selective in what we link against, instead of using `MLIRPythonExtension.RegisterEverything`.

This patch still only supports a single indexing tensor.

…lvm#1081) An upstream MLIR bug (that was recently fixed) caused the result to be ignored for Region- and Block-visitor functions. Now that the bug is fixed, we don't need an auxiliary boolean to track whether the visitor function has succeeded.

- Supports cases where the view op expands and collapses dims simulataneously. This does not handle the case where it is neither expanding nor collapsing (e.g. [2, 3] -> [3, 2]) - Additionally fixes a previous bug with adding 1-sized dims on both sides of a tensor with aten.view

Co-authored-by: Bairen Yi <yibairen.byron@bytedance.com> Co-authored-by: Jiawei Wu <xremold@gmail.com> Co-authored-by: Tianyou Guo <tianyou.gty@alibaba-inc.com> Co-authored-by: Xu Yan <yancey.yx@alibaba-inc.com> Co-authored-by: Ziheng Jiang <ziheng.jiang@bytedance.com>

In the interest of merging upstream LLVM quickly, a previous patch (7f08169) updated the torch-mlir build to register all dialects and passes through Python bindings. This patch limits the dialects and passes to only those that are used in torch-mlir. Key to this change are the removal of `MLIRPythonExtension.RegisterEverything` and the introduction of a new Python module (`_mlir_libs/_site_initialize_0.py`), where we register the dialects and passes used by torch-mlir.

…llvm#1093) This reverts commit ad283c1, since it's causing nightly build failures for all platforms.

The unknown dtype case can come from RefineTypes.

This enables building Pytorch from source in the CI. The build should mostly hit the ccache. Release builds will follow once we have some runtime on the CI.

This commit adds verifiers to the ops `ToBuiltinTensorOp` and `FromBuiltinTensorOp` that make sure that the input and output have the same shape and data type.

See RFC: llvm#999 Co-authored-by: Bairen Yi yibairen.byron@bytedance.com Co-authored-by: Jiawei Wu xremold@gmail.com Co-authored-by: Tianyou Guo tianyou.gty@alibaba-inc.com Co-authored-by: Xu Yan yancey.yx@alibaba-inc.com Co-authored-by: Ziheng Jiang ziheng.jiang@bytedance.com

* [MHLO] Init MHLO view like op patterns See RFC: llvm#999 Co-authored-by: Bairen Yi yibairen.byron@bytedance.com Co-authored-by: Jiawei Wu xremold@gmail.com Co-authored-by: Tianyou Guo tianyou.gty@alibaba-inc.com Co-authored-by: Xu Yan yancey.yx@alibaba-inc.com Co-authored-by: Ziheng Jiang ziheng.jiang@bytedance.com * update filecheck test cases * rebase, remove chlo and clang-format

* [MHLO] Add [un]squeeze op patterns * Conform to llvm coding standard * minor update

See RFC llvm#999 Co-authored-by: Bairen Yi yibairen.byron@bytedance.com Co-authored-by: Jiawei Wu xremold@gmail.com Co-authored-by: Tianyou Guo tianyou.gty@alibaba-inc.com Co-authored-by: Xu Yan yancey.yx@alibaba-inc.com Co-authored-by: Ziheng Jiang ziheng.jiang@bytedance.com

Follows existing conventions for unary operators.

Add source builds and remove deprecated libtorch information.

@llvm-project

* [MHLO] Init MHLO basic Op Conversion Co-authored-by: Bairen Yi <yibairen.byron@bytedance.com> Co-authored-by: Jiawei Wu <xremold@gmail.com> Co-authored-by: Tianyou Guo <tianyou.gty@alibaba-inc.com> Co-authored-by: Xu Yan <yancey.yx@alibaba-inc.com> Co-authored-by: Ziheng Jiang <ziheng.jiang@bytedance.com> * [NFC] Remove 'from @llvm-project' annotation Co-authored-by: wujiawei.jw <wujiawei.jw@bytedance.com>

…x.Tensor (llvm#1097) - Includes a canonicalizer for `aten.add.t`needed for successfully lowering the shape function - Only offers support for statically sized index tensors when there is more than one - Dynamic shape support remains for single indexing tensors

This commit adds lowering of `aten.ge.int` op. Signed-Off By: Vivek Khandelwal <vivek@nod-labs.com>

This commit adds the decomposition for `aten.var.correction` op. Signed-Off By: Vivek Khandelwal <vivek@nod-labs.com

This commit fixes the shape calculation for: 1.) aten.mean.dim 2.) aten.var.dim 3.) aten.sum.dim_IntList op Also, it fixes the lowering of `aten.mean.dim` and `aten.sum.dim_IntList` for handling the cases of empty dim list. Signed-Off By: Vivek Khandelwal <vivek@nod-labs.com

Fixes llvm#1121 Signed-Off By: Vivek Khandelwal<vivek@nod-labs.com>

* Fix symint related functionalization ops * Remove zeros xfail from LTC tests

* Add LTC architecture diagram * Use PNG for diagrams * Update diagram

This commit adds a file explaining the structure of an E2E test, as well as useful things to keep in mind when adding new tests.

Change logic so that we never run the multiprocessing codepath with only 1 worker. That configuration was causing all subsequent tests to spuriously fail if one test failed with a crash (this was easy to see after sorting the tests). That configuration was the one used by the CI. Also, sort tests to make output nicer. Also, make verbose mode more verbose so that it is easy to see in `-s` mode which test is crashing.

* Disable LTC by default until upstream revert relands Tracked with the WIP llvm#1292 * Disable LTC e2e tests temporarily * Update setup.py Disable LTC in setup.py temporarily until upstream is fixed.

We used to not have "value-semantic" tensors but rather "immutable" tensors

This ensures that they are always up to date. This also updates the shape lib to make the new CI actually pass :)

We use it for more than TorchScript testing now. This is a purely mechanical change to adjust some file paths to remove "torchscript". The most perceptible change here is that now e2e tests are run with ``` ./tools/e2e_test.sh instead of: ./tools/torchscript_e2e_test.sh ```

In the sequential case we weren't sorting, which was confusing.

Gets both CI and Release builds integrated in one workflow. Mount ccache and pip cache as required for fast iterative builds Current Release docker builds still run with root perms, fix it in the future to run as the same user. There may be some corner cases left especially when switching build types etc. Docker build TEST plan: tl;dr: Build everythin: Releases (Python 3.8, 3.9, 3.10) and CIs. TM_PACKAGES="torch-mlir out-of-tree in-tree" 2.57s user 2.49s system 0% cpu 30:33.11 total Out of Tree + PyTorch binaries: Fresh build (purged cache): TM_PACKAGES="out-of-tree" 0.47s user 0.51s system 0% cpu 5:24.99 total Incremental with ccache: TM_PACKAGES="out-of-tree" 0.09s user 0.08s system 0% cpu 34.817 total Out of Tree + PyTorch from source Incremental TM_PACKAGES="out-of-tree" TM_USE_PYTORCH_BINARY=OFF 1.58s user 1.81s system 2% cpu 1:59.61 total In-Tree + PyTorch binaries: Fresh build and tests: (purge ccache) TM_PACKAGES="in-tree" 0.53s user 0.49s system 0% cpu 6:23.35 total Fresh build/ but with prior ccache TM_PACKAGES="in-tree" 0.45s user 0.66s system 0% cpu 3:57.47 total Incremental in-tree with all tests and regression tests TM_PACKAGES="in-tree" 0.16s user 0.09s system 0% cpu 2:18.52 total In-Tree + PyTorch from source Fresh build and tests: (purge ccache) TM_PACKAGES="in-tree" TM_USE_PYTORCH_BINARY=OFF 2.03s user 2.28s system 0% cpu 11:11.86 total Fresh build/ but with prior ccache TM_PACKAGES="in-tree" TM_USE_PYTORCH_BINARY=OFF 1.58s user 1.88s system 1% cpu 4:53.15 total Incremental in-tree with all tests and regression tests TM_PACKAGES="in-tree" TM_USE_PYTORCH_BINARY=OFF 1.09s user 1.10s system 1% cpu 3:29.84 total Incremental without tests TM_PACKAGES="in-tree" TM_USE_PYTORCH_BINARY=OFF TM_SKIP_TESTS=ON 1.52s user 1.42s system 3% cpu 1:15.82 total In-tree+out-of-tree + Pytorch Binaries TM_PACKAGES="out-of-tree in-tree" 0.25s user 0.18s system 0% cpu 3:01.91 total To clear all artifacts: rm -rf build build_oot llvm-build libtorch docker_venv externals/pytorch/build

Caught in the wild here: https://github.com/llvm/torch-mlir/runs/8046660640?check_suite_focus=true It is common for a missing dependency to only surface as an issue on the CI machines since they have fewer cores which prevents a "race" that happens to cause the dependency to be built before the dependent.

- Update MHLO commit to build with LLVM commit hash 00d648bd - Update TorchToMhlo code to work with Stablehlo - Re-enabled two failing TOSA tests, thus resolving Github Issue llvm#1231

This should speed up source builds and ccache. May cause issues on macOS (pytorch/pytorch#80018)

This is to debug what is causing the exactly ccache look up failures etc.

Related to llvm#1227 1. Reduce MHLO #ifdefs 2. Dismiss compilation warnings

The new logic has the following benefits: 1. It does not clobber the working tree state. We expect testing to not change the work tree. 2. It correctly handles the case where a user has changes to the generated files, but hasn't checked them in yet (this happens frequently when adding new ops).

fix float width fix divide_floor & export promoteTypes api (#9)

#15)

jpienaar and others added 30 commits July 17, 2022 18:39

[MLIR][TORCH] Add canonicalization pattern for prim.ListUnpack op

4c25878

This commit adds the canonicalization pattern for the `prim.ListUnpack` op. Signed-Off By: Vivek Khandelwal <vivek@nod-labs.com>

[MLIR][TORCH] Add negative dim support for aten.cat and aten.slice op

df0b1e7

This commit adds the support for negative dim cases for `aten.cat`, `aten.slice.Tensor` and `aten.slice_scatter` op. Signed-Off By: Vivek Khandelwal <vivek@nod-labs.com>

Add support for index.Tensor on dimensions other than the first

c73a39e

This patch still only supports a single indexing tensor.

Emit underscore version of aten.sqrt (llvm#1072)

21f905a

Revert "python: trim registration and loading of dialects and passes" (…

72dd04c

…llvm#1093) This reverts commit ad283c1, since it's causing nightly build failures for all platforms.

Improve error message

c0ef192

The unknown dtype case can come from RefineTypes.

Add linux and macOS source builds in CI (llvm#1070)

31fd812

This enables building Pytorch from source in the CI. The build should mostly hit the ccache. Release builds will follow once we have some runtime on the CI.

Add verifiers for ToBuiltinTensorOp and FromBuiltinTensorOp (llvm#1089)

f271e6a

This commit adds verifiers to the ops `ToBuiltinTensorOp` and `FromBuiltinTensorOp` that make sure that the input and output have the same shape and data type.

Add option to expose custom PyTorch repo/branch (llvm#1103)

f424930

[MHLO] Add [un]squeeze op patterns (llvm#1099)

f50d701

* [MHLO] Add [un]squeeze op patterns * Conform to llvm coding standard * minor update

Add lowering to linalg for softplus and log1p

e8f327c

Follows existing conventions for unary operators.

Update development.md with source builds (llvm#1105)

e23fbc8

Add source builds and remove deprecated libtorch information.

Add e2e support for aten.expm1

3c9addf

README: Add op office hours

b36a17c

[MLIR][TORCH] Add E2E support for aten.ge.int op

7247c6a

This commit adds lowering of `aten.ge.int` op. Signed-Off By: Vivek Khandelwal <vivek@nod-labs.com>

[MLIR][TORCH] Add decomposition for aten.var.correction op

d386b8f

This commit adds the decomposition for `aten.var.correction` op. Signed-Off By: Vivek Khandelwal <vivek@nod-labs.com

Add mhlo to bazel build (llvm#1120)

b389053

Fix CI failure due to upstream PyTorch change in aten.mean.dim op

9a1203c

Fixes llvm#1121 Signed-Off By: Vivek Khandelwal<vivek@nod-labs.com>

antoniojkim and others added 29 commits August 26, 2022 16:13

Fix symint related functionalization ops (llvm#1289)

8e880a2

* Fix symint related functionalization ops * Remove zeros xfail from LTC tests

Add LTC architecture diagram (llvm#1291)

883c6b4

* Add LTC architecture diagram * Use PNG for diagrams * Update diagram

Add documentation for adding E2E tests (llvm#1269)

493f45f

This commit adds a file explaining the structure of an E2E test, as well as useful things to keep in mind when adding new tests.

Add reference to slides and presentation.

f245613

Disable LTC by default until upstream revert relands (llvm#1303)

c0630da

* Disable LTC by default until upstream revert relands Tracked with the WIP llvm#1292 * Disable LTC e2e tests temporarily * Update setup.py Disable LTC in setup.py temporarily until upstream is fixed.

Rename an outdated class name

2623185

We used to not have "value-semantic" tensors but rather "immutable" tensors

Add CI for generated files.

bcccf41

This ensures that they are always up to date. This also updates the shape lib to make the new CI actually pass :)

Update MHLO xfails.

15fca6e

Move development.md to docs/ for consistency

f402eb2

Sort tests before anything else.

079bff3

In the sequential case we weren't sorting, which was confusing.

Ensure that tests have unique names

0f40d98

build: update llvm tag to 00d648bd (llvm#1307)

e52e886

- Update MHLO commit to build with LLVM commit hash 00d648bd - Update TorchToMhlo code to work with Stablehlo - Re-enabled two failing TOSA tests, thus resolving Github Issue llvm#1231

Add shapelib and Torch ODS gen tests (llvm#1318)

928c815

Use pre-compiled headers for PyTorch Source builds (llvm#1327)

3704363

This should speed up source builds and ccache. May cause issues on macOS (pytorch/pytorch#80018)

Set ccache logging to verbose temporarily (llvm#1326)

7769eb8

This is to debug what is causing the exactly ccache look up failures etc.

[MHLO] refactor pass configurations (llvm#1315)

29cafdb

Related to llvm#1227 1. Reduce MHLO #ifdefs 2. Dismiss compilation warnings

aicompiler rebase 20220830

f5a8a93

fix float width fix divide_floor & export promoteTypes api (#9)

To comply with the old pytorch versions

1465814

add relu6 op

5d5504d

Add decomposition for aten::native_layer_norm (#13)

5d6b854

Add native_dropout_backward & native_layer_norm_backward decomposition (

3bc47c3

#15)

add native_dropout and related ops pattern (llvm#1211)

c1b556a

Fix dot product result type setting (#19)

207c7ad

fix BatchNormInference

8f29de9

tanyokwok closed this Sep 21, 2022

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

features/bladedisc rebase 20220830 #20

features/bladedisc rebase 20220830 #20

tanyokwok commented Sep 21, 2022

features/bladedisc rebase 20220830 #20

features/bladedisc rebase 20220830 #20

Conversation

tanyokwok commented Sep 21, 2022