Commit Graph

1097 Commits (2a0c5de3633ec2df1c939eac15fbafcd64a3c968)
 

Author SHA1 Message Date
Ramiro Leal-Cavazos 3aef3a9e30 Remove duplicate example + fix README typo 2021-09-30 13:23:25 -07:00
Sean Silva 05a01b10b3
Add example deps for torchscript_resnet18_e2e.py 2021-09-30 12:48:35 -07:00
Sean Silva bc62a7fbf3 Update to new name of torchscript-module-to-linalg-on-tensors-backend-pipeline 2021-09-30 19:46:05 +00:00
Yi Zhang 89225b0cd8 Add BertSequenceClassification model to e2e
Use torch tracing to get the module because the original model is not
TorchScriptable out of box.
2021-09-30 13:30:29 -04:00
powderluv 649d6e4f28 Update README.md 2021-09-29 10:14:20 -07:00
Sean Silva 4a64fa28e7
Add MLP1Layer IR sample. 2021-09-29 09:20:04 -07:00
powderluv 36ac87dd83
Update Readme with examples (#336) 2021-09-28 21:43:38 -07:00
Sean Silva 8b2c099914 Update llvm-project to 204d301bb1921431a853c0bfba32007c018df1d5
This brings in the fix for the obscure RefBackend bug we were hitting.
2021-09-28 17:38:10 -07:00
Ramiro Leal-Cavazos b59f2cb673
Implement the lazytensor package (#331)
Implement the `lazytensor` python package for converting
lazy computations captured by the Lazy Tensor Core into MLIR.
This PR also fixes a few things with `torchfx` and its example
2021-09-28 17:25:06 -07:00
Sean Silva 2b99c8b990
Wordsmith readme 2021-09-28 13:53:33 -07:00
powderluv b55baf508a
Updates to Readme.md (#334) 2021-09-28 13:50:25 -07:00
powderluv 2e6c423a65
Caught by grammar police. PyTorch it is, not Pytorch. (#333) 2021-09-27 21:44:23 -07:00
Sean Silva de1a8c92f7 First pass cleaning the README 2021-09-27 17:20:50 -07:00
Sean Silva 64ce5d54d3 Update examples.
TorchFX example has been simplified, since it seems to be hitting that
weird RefBackend bug. Will dig into that.
2021-09-27 17:20:50 -07:00
Sean Silva e687d39074 Update buildAndTest.yml 2021-09-27 17:11:08 -07:00
Sean Silva 4fad753073 Move external/torch-mlir to the root of the repo. 2021-09-27 17:11:08 -07:00
Sean Silva d8f603a4e5 Remove old stuff in prep for move-to-root. 2021-09-27 17:11:08 -07:00
Sean Silva 404bd74ddf Port the bulk of the remaining code to torch-mlir
This leaves no real code outside torch-mlir.

This also renames the "npcomp backend contract" to "linalg on tensors
backend contract" as the name of the abstraction layer that RefBackend
(IREE too) accepts.
2021-09-27 12:48:33 -07:00
Yi Zhang aa10ec66a7 Fix torchscript_resnet18_e2e.py and resnet_inference.ipynb
Fix the tests to run with refbackend.
2021-09-27 13:03:54 -04:00
Yi Zhang cd7053dfde Add runtime check 2021-09-24 12:01:36 -04:00
Yi Zhang c9cc4cb2e9 Add i64 tensor argument support and bring back GatherModule_basic 2021-09-24 12:01:36 -04:00
Sean Silva 12d0fe7c85 Emit a proper error.
This assertion was reachable from user code (e.g. !torch.qint8 dtype),
which should never be possible for an assertion.

Instead, emit a diagnostic.
2021-09-23 21:39:31 -07:00
Sean Silva a99cbeeb7e Move TorchConversion dialect and TorchTo* into torch-mlir 2021-09-23 21:39:31 -07:00
Sean Silva 35fa1a34cd npcomp-lsp-server -> torch-mlir-lsp-server 2021-09-23 21:39:31 -07:00
Sean Silva 01c6c54dd8 Fix dependency. 2021-09-23 21:39:31 -07:00
Sean Silva 2213584c4f VerifyBackendContract -> VerifyLinalgOnTensorsBackendContract
This moves it into TorchConversion since it is only needed there.

This removes the Backend/ directory.
2021-09-23 21:39:31 -07:00
Anush Elangovan 06559efe94 Update diagram and Readme.md 2021-09-23 14:34:02 -07:00
Sean Silva e291aa688f Bring back tools/torchscript_e2e_test.sh
Accidentally deleted it in the last commit.
2021-09-23 19:01:48 +00:00
powderluv 81a9cf7800 Update README.md 2021-09-23 11:50:02 -07:00
anush elangovan cbafe55b3a Add Torch-mlir.png Architecture Diagram 2021-09-23 11:50:02 -07:00
Ramiro Leal-Cavazos 2b18aad807 Removed import typo in torchfx example 2021-09-22 16:49:46 -07:00
Yi Zhang 603e068e45 E2e implementation for `aten.cat`,`aten.gather`, `aten.bmm`
Also contains the following changes:
- Remove derefineOp canonicalizer because it's not safe.
- Support for optional tensor and list tensors in reduceOpVariant. This
only works for some special detected and easy to handle cases. For list,
it covers the case list is got from a `ListConstruct`. For optional, it
covers the case optional is constructed from a `DerefineOp`.
- Remove the `inferReturnTypes` for `FromBuiltinTensorOp` because it's
not safe to deduce types from the input. For example, a built-in tensor
of i8 could be converted to si8 or ui8. It's better to let the user
specify the return type explicitly.
2021-09-22 19:15:01 -04:00
Sean Silva 3dc9b4ee2f Remove some more old stray files. 2021-09-22 16:13:03 -07:00
Sean Silva a0ccfcf581 Remove some older docs. 2021-09-22 16:13:03 -07:00
Sean Silva 12dc2fa933 Remove docker stuff.
It's not used in CI and doesn't seem to be up to date.
2021-09-22 16:13:03 -07:00
Sean Silva 699159f440 Remove frontends/ directory.
It still had a bit of stale stuff.
2021-09-22 16:13:03 -07:00
Sean Silva 1a0b953ea7 Eliminate almost all mentions of IREE.
A few remain in examples/docs that will be naturally be updated in due
time.

This regresses the list support and the general direction of more widely
supported control flow, lists/dicts/globals that we were going for with
the TorchScript path. The idea is that we are deferring that work to
make torch-mlir a very clean standalone thing. We will reboot it,
probably using some of the tools of iree_pydm to make it simpler, and in
a more natural place (such as an iree-torch repo that depends on IREE and
torch-mlir to build a working PyTorch frontend solution for IREE -- it
was really weird that npcomp depended on IREE).
2021-09-22 16:06:38 -07:00
Sean Silva 8779d920b2 Remove "refjit" terminology.
We now use RefBackend/refbackend consistently.
2021-09-22 15:41:23 -07:00
Sean Silva a25163fbfa Remove old RefBackend
It is superceded by the new one.
2021-09-22 15:33:28 -07:00
Sean Silva f9c48d0b89 Bring up new RefBackend.
`tools/torchscript_e2e_test.sh` is all green.

This needs a few passes I put into torch-mlir/lib/RefBackend (not to be
confused with `npcomp/lib/RefBackend`, which will soon be deleted).

For the sake of review, since this brings together a lot of things, I
split this into its own commit. I temporarily commented out some "list"
stuff that we are going to remove as part of the torch-mlir refocus.
2021-09-22 14:20:22 -07:00
Ramiro Leal-Cavazos 1f00f95d2e
WIP implementation of torchfx (#304)
Implements a python package for taking a `torch.fx.GraphModule`
and turning it into MLIR in the `torch` dialect that can then
be further compiled by `npcomp`. This is a WIP, so the coverage
of PyTorch operations is very small.
2021-09-22 10:27:55 -07:00
Ramiro Leal-Cavazos 19feb58ef8 Update update_torch_ods.sh to use new torch_mlir python layout 2021-09-22 09:38:51 -07:00
Sean Silva 6d8e7f1bb1 Implement Python relayout from #311
Fixes https://github.com/llvm/mlir-npcomp/issues/311

The key change is that TorchPlugin is folded into
`torch_mlir.dialects.torch.importer.jit_ir` (it imports the PyTorch
JIT's IR, so that's a good, scoped name for it).
The CMake option `-DTORCH_MLIR_ENABLE_JIT_IR_IMPORTER=OFF` disables it,
which allows building without a PyTorch native dependency.
2021-09-21 09:29:40 -07:00
George Petterson ecc334123c Added transpose lowering 2021-09-19 20:28:27 -04:00
Stella Laurenzo c24ca5d639
Build fix: Include the builtin and std dialects in the torch_mlir python build. (#310)
Should fix issue noted in #304
2021-09-19 12:26:33 -07:00
Sean Silva 9e2442d6b0 Factor out definition of the _torch_mlir.get_registered_ops function.
It didn't make sense in the main registration file.
2021-09-17 09:28:15 -07:00
Sean Silva b738db34cd Move files out of TorchPlugin/csrc/builder/ directory.
It was an unnecessary layer of indirection -- there was nothing outside
of it, and it was just harder to follow the structure.
2021-09-17 09:28:15 -07:00
Sean Silva 8d27c41f21 Fix file comment filename 2021-09-17 09:28:15 -07:00
Sean Silva 3e3459690c Remove TypeMapper 2021-09-17 09:28:15 -07:00
Sean Silva 6f710bbc47 Remove op_builder.h/cpp 2021-09-17 09:28:15 -07:00