Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Added e2e LTC tests #916

Merged
merged 13 commits into from
Jun 9, 2022
Merged
5 changes: 5 additions & 0 deletions .github/workflows/buildAndTest.yml
Original file line number Diff line number Diff line change
Expand Up @@ -56,6 +56,11 @@ jobs:
cd $GITHUB_WORKSPACE
export PYTHONPATH="$GITHUB_WORKSPACE/build/tools/torch-mlir/python_packages/torch_mlir"
python -m e2e_testing.torchscript.main --config=tosa -v
- name: Lazy Tensor Core - TorchScript end-to-end tests
run: |
cd $GITHUB_WORKSPACE
export PYTHONPATH="$GITHUB_WORKSPACE/build/tools/torch-mlir/python_packages/torch_mlir"
python -m e2e_testing.torchscript.main --config=lazy_tensor_core -v

build-out-of-tree:
name: Build out-of-tree (Release Asserts)
Expand Down
10 changes: 7 additions & 3 deletions e2e_testing/torchscript/main.py
Original file line number Diff line number Diff line change
Expand Up @@ -15,20 +15,20 @@

# Available test configs.
from torch_mlir_e2e_test.torchscript.configs import (
LinalgOnTensorsBackendTestConfig, NativeTorchTestConfig, TorchScriptTestConfig, TosaBackendTestConfig, EagerModeTestConfig
LazyTensorCoreTestConfig, LinalgOnTensorsBackendTestConfig, NativeTorchTestConfig, TorchScriptTestConfig, TosaBackendTestConfig, EagerModeTestConfig
)

from torch_mlir_e2e_test.linalg_on_tensors_backends.refbackend import RefBackendLinalgOnTensorsBackend
from torch_mlir_e2e_test.tosa_backends.linalg_on_tensors import LinalgOnTensorsTosaBackend

from .xfail_sets import REFBACKEND_XFAIL_SET, TOSA_PASS_SET, EAGER_MODE_XFAIL_SET
from .xfail_sets import REFBACKEND_XFAIL_SET, TOSA_PASS_SET, EAGER_MODE_XFAIL_SET, LTC_XFAIL_SET

# Import tests to register them in the global registry.
from torch_mlir_e2e_test.test_suite import register_all_tests
register_all_tests()

def _get_argparse():
config_choices = ['native_torch', 'torchscript', 'refbackend', 'tosa', 'eager_mode']
config_choices = ['native_torch', 'torchscript', 'refbackend', 'tosa', 'eager_mode', 'lazy_tensor_core']
parser = argparse.ArgumentParser(description='Run torchscript e2e tests.')
parser.add_argument('-c', '--config',
choices=config_choices,
Expand All @@ -40,6 +40,7 @@ def _get_argparse():
"native_torch": run the torch.nn.Module as-is without compiling (useful for verifying model is deterministic; ALL tests should pass in this configuration).
"torchscript": compile the model to a torch.jit.ScriptModule, and then run that as-is (useful for verifying TorchScript is modeling the program correctly).
"eager_mode": run through torch-mlir's eager mode frontend, using RefBackend for execution.
"lazy_tensor_core": run the model through the Lazy Tensor Core frontend and execute the traced graph.
''')
parser.add_argument('-f', '--filter', default='.*', help='''
Regular expression specifying which tests to include in this run.
Expand Down Expand Up @@ -86,6 +87,9 @@ def main():
elif args.config == 'eager_mode':
config = EagerModeTestConfig()
xfail_set = EAGER_MODE_XFAIL_SET
elif args.config == 'lazy_tensor_core':
config = LazyTensorCoreTestConfig()
xfail_set = LTC_XFAIL_SET

# Find the selected tests, and emit a diagnostic if none are found.
tests = [
Expand Down
309 changes: 309 additions & 0 deletions e2e_testing/torchscript/xfail_sets.py
Original file line number Diff line number Diff line change
Expand Up @@ -154,3 +154,312 @@
"TestMultipleTensorReturn_basic",
"AdaptiveAvgPool2dUnitOutputSizeStaticModule_basic",
}

LTC_XFAIL_SET = {
"AdaptiveAvgPool2dNonUnitOutputSizeDynamicModule_basic",
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Any triage on why this could be failing?

If we are running the JIT graph directly, I would assume that everything "Just Works".

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

We don't have support for all ops yet. Some are missing some shape inference functions, some ops are blacklisted, etc.

Copy link
Contributor

@silvasean silvasean Jun 9, 2022

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Ah, got it. It's a little annoying that we are duplicating the shape functions for LTC. Could you collaborate with @eellison to reuse their shape stuff?

We have been upstreaming all our shape functions from https://github.com/llvm/torch-mlir/blob/main/python/torch_mlir/dialects/torch/importer/jit_ir/build_tools/shape_lib_gen.py, so this should all be part of a big source of truth upstream. E.g. pytorch/pytorch#76889

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hm our shape inferencing happens in C++, so I don't think there's much we can do as far as reusing that code 😢

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

All of the shapes functions are also stored in C++ fwiw, so I don't think that's a fundamental blocker. but I dont know all the details of what's going on here

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@eellison Can you point me to where that is in the PyTorch repo? I just looked at the PR that Sean linked earlier, which had C++ with some long strings.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It should turn into JIT IR (that's what you see in the strings) that we can then load up and execute from C++.

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@silvasean are you suggesting that we apply this at the JIT graph level, or port this over to work with LTC generally?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If it's the former, I don't think we can use it, because we require shape information at the LTC layer before generating a JIT graph.

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Oh, interesting. Yeah, that makes things more tricky to reuse. @eellison is the shape inference infra tied specifically to JIT graphs?

I suppose we could create a shapeless jit graph just for the purpose of reusing the shape inference infrastructure, but that's quite a few hoops to jump through.

"AdaptiveAvgPool2dNonUnitOutputSizeStaticModule_basic",
"AdaptiveAvgPool2dUnitOutputSizeDynamicModule_basic",
"AdaptiveAvgPool2dUnitOutputSizeStaticModule_basic",
"AddIntModule_basic",
"AllBoolFalseModule_basic",
"AllBoolTrueModule_basic",
"AnyBoolFalseModule_basic",
"AnyBoolTrueModule_basic",
"ArangeDtypeFloatModule_basic",
"ArangeDtypeIntModule_basic",
"ArangeFalsePinMemoryModule_basic",
"ArangeFloatModule_basic",
"ArangeIntModule_basic",
"ArangeNegativeStartFloatModule_basic",
"ArangeNegativeStartIntModule_basic",
"ArangeStartFloatModule_basic",
"ArangeStartIntModule_basic",
"ArangeStartNegativeStepFloatModule_basic",
"ArangeStartNegativeStepIntModule_basic",
"ArangeStartStepFloatModule_basic",
"ArangeStartStepIntModule_basic",
"ArangeZeroElementOutputModule_basic",
"AvgPool2dCeilModeTrueModule_basic",
"AvgPool2dDivisorOverrideModule_basic",
"AvgPool2dFloatModule_basic",
"AvgPool2dIntModule_basic",
"AvgPool2dStaticModule_basic",
"BernoulliFloatModule_basic",
"BernoulliModule_basic",
"BernoulliOnesModule_basic",
"BernoulliTensorModule_basic",
"BernoulliZerosModule_basic",
"BincountMinlengthModule_basic",
"BincountModule_basic",
"BincountStaticSizeModule_basic",
"BoolFloatConstantModule_basic",
"BoolFloatFalseModule_basic",
"BoolFloatTrueModule_basic",
"BoolIntConstantModule_basic",
"BoolIntFalseModule_basic",
"BoolIntTrueModule_basic",
"CeilFloatModule_basic",
"DivFloatModule_basic",
"DropoutTrainModule_basic",
"ElementwiseAtenLogicalOrOpBrodcastModule_basic",
"ElementwiseAtenLogicalOrOpDiffArgs1Module_basic",
"ElementwiseAtenLogicalOrOpDiffArgs2Module_basic",
"ElementwiseAtenLogicalOrOpDiffArgs3Module_basic",
"ElementwiseAtenLogicalOrOpModule_basic",
"ElementwiseAtenLogicalOrOpNegativeModule_basic",
"ElementwiseAtenLogicalOrOpRandomFloatModule_basic",
"ElementwiseAtenLogicalOrOpRandomModule_basic",
"ElementwiseClampMaxModule_basic",
"ElementwiseClampMinModule_basic",
"ElementwiseClampModule_basic",
"ElementwiseWhereScalarModule_basic",
"ElementwiseWhereScalarOtherModule_basic",
"ElementwiseWhereScalarSelfModule_basic",
"ElementwiseWhereSelfModule_basic",
"EmptyLikeMemoryFormatModule_basic",
"EmptyLikeModule_defaultDtype",
"EmptyLikeModule_falsePinMemory",
"EmptyLikeModule_float",
"EmptyLikeModule_int",
"EmptyModule_contiguous",
"EmptyModule_defaultDtype",
"EmptyModule_falsePinMemory",
"EmptyModule_float",
"EmptyModule_int",
"EqIntModule_basic",
"Fill_TensorFloat64WithFloat32_basic",
"Fill_TensorFloat64WithFloat64_basic",
"Fill_TensorFloat64WithInt64_basic",
"FullLikeModuleDefaultDtype_basic",
"FullLikeModuleFalsePinMemory_basic",
"FullLikeModuleFloat2D_basic",
"FullLikeModuleFloat3DStatic_basic",
"FullLikeModuleFloat3D_basic",
"FullLikeModuleInt2DStatic_basic",
"FullLikeModuleInt2D_basic",
"FullLikeModuleInt3D_basic",
"FullModuleDefaultDtype_basic",
"FullModuleFalsePinMemory_basic",
"FullModuleFloat2D_basic",
"FullModuleFloat3D_basic",
"FullModuleInt2D_basic",
"FullModuleInt3D_basic",
"GeFloatIntModule_basic",
"GeFloatModule_basic",
"GtFloatIntModule_basic",
"GtIntModule_basic",
"HBC_basic",
"HardTanhIntModule_basic",
"HardTanhModule_basic",
"HardswishModule_basic",
"HardswishRandomModule_basic",
"IndexPut1DFloatAccumulateModule_basic",
"IndexPut1DFloatNonAccumulateModule_basic",
"IndexPut1DIntAccumulateModule_basic",
"IndexPut1DIntNonAccumulateModule_basic",
"IndexPut2DFloatAccumulateModule_basic",
"IndexPut2DFloatNonAccumulateModule_basic",
"IndexPut2DIntAccumulateModule_basic",
"IndexPut2DIntNonAccumulateModule_basic",
"IndexPut3DFloatAccumulateModule_basic",
"IndexPut3DFloatNonAccumulateModule_basic",
"IndexPut3DIntAccumulateModule_basic",
"IndexPut3DIntNonAccumulateModule_basic",
"IndexPutHackedTwin1DFloatAccumulateModule_basic",
"IndexPutHackedTwin1DFloatNonAccumulateModule_basic",
"IndexPutHackedTwin1DIntAccumulateModule_basic",
"IndexPutHackedTwin1DIntNonAccumulateModule_basic",
"IndexPutHackedTwin2DFloatAccumulateModule_basic",
"IndexPutHackedTwin2DFloatNonAccumulateModule_basic",
"IndexPutHackedTwin2DIntAccumulateModule_basic",
"IndexPutHackedTwin2DIntNonAccumulateModule_basic",
"IndexPutHackedTwin3DFloatAccumulateModule_basic",
"IndexPutHackedTwin3DFloatNonAccumulateModule_basic",
"IndexPutHackedTwin3DIntAccumulateModule_basic",
"IndexPutHackedTwin3DIntNonAccumulateModule_basic",
"IndexPutImpl1DFloatAccumulateModule_basic",
"IndexPutImpl1DFloatNonAccumulateModule_basic",
"IndexPutImpl1DIntAccumulateModule_basic",
"IndexPutImpl1DIntNonAccumulateModule_basic",
"IndexPutImpl2DFloatAccumulateModule_basic",
"IndexPutImpl2DFloatNonAccumulateModule_basic",
"IndexPutImpl3DFloatAccumulateModule_basic",
"IndexPutImpl3DFloatNonAccumulateModule_basic",
"IndexSelectDynamicIndexSizeModule_basic",
"IndexSelectDynamicInputSizeModule_basic",
"IndexSelectDynamicModulebasic",
"IndexSelectSingleIdxModule_basic",
"IndexSelectTwoIdxModule_basic",
"IndexSelectWholeDimensionModule_basic",
"IndexSelectWholeTensorModule_basic",
"IndexTensorModule_basic",
"MaskedFillScalarDefaultModule_basic",
"MaskedFillScalarFloatValueModule_basic",
"MaskedFillScalarIntValueModule_basic",
"Matmul_dot",
"Matmul_matvec",
"Matmul_vecmat",
"MaxPool2dCeilModeTrueModule_basic",
"MaxPool2dModule_basic",
"MaxPool2dStaticModule_basic",
"MaxPool2dWith3dInputModule_basic",
"MaxPool2dWithIndicesAllNegativeValuesModule_basic",
"MaxPool2dWithIndicesAllOnesModule_basic",
"MaxPool2dWithIndicesBackwardDynamic3DModule_basic",
"MaxPool2dWithIndicesBackwardDynamic4DModule_basic",
"MaxPool2dWithIndicesBackwardStatic3DModule_basic",
"MaxPool2dWithIndicesBackwardStatic4DModule_basic",
"MaxPool2dWithIndicesCeilModeTrueModule_basic",
"MaxPool2dWithIndicesFullSizeKernelModule_basic",
"MaxPool2dWithIndicesModule_basic",
"MaxPool2dWithIndicesNonDefaultDilationModule_basic",
"MaxPool2dWithIndicesNonDefaultPaddingModule_basic",
"MaxPool2dWithIndicesNonDefaultParamsModule_basic",
"MaxPool2dWithIndicesNonDefaultStrideModule_basic",
"MaxPool2dWithIndicesStaticModule_basic",
"MaxPool2dWithIndicesWith3dInputModule_basic",
"MeanDimAllReduceKeepdimModule_basic",
"MeanDimAllReduceModule_basic",
"MeanDimDtypeModule_basic",
"MeanDimKeepdimModule_basic",
"MeanDimModule_basic",
"MeanDimNegativeModule_basic",
"MeanDtypeModule_basic",
"MeanDynamicSizesModule_basic",
"MeanModule_basic",
"MobilenetV3Module_basic",
"MulIntModule_basic",
"NativeBatchNorm1DModule_basic",
"NativeBatchNorm2DModule_basic",
"NativeBatchNorm3DModule_basic",
"NativeBatchNormNoneWeightModule_basic",
"NativeLayerNormDynamicModule_basic",
"NativeLayerNormModule_basic",
"NeFloatIntModule_basic",
"NeIntModule_basic",
"NewEmptyModuleDefaultDtype_basic",
"NewEmptyModuleFalsePinMemory_basic",
"NewEmptyModuleFloat2D_basic",
"NewEmptyModuleFloat3D_basic",
"NewEmptyModuleInt2D_basic",
"NewEmptyModuleInt3D_basic",
"NewEmptyModuleLayoutIntDtype_basic",
"NewEmptyModuleNonDefaultFloatDtype_basic",
"NewEmptyModuleNonDefaultIntDtype_basic",
"NewOnesModuleDefaultDtype_basic",
"NewOnesModuleFalsePinMemory_basic",
"NewOnesModuleFloat2D_basic",
"NewOnesModuleFloat3D_basic",
"NewOnesModuleInt2D_basic",
"NewOnesModuleInt3D_basic",
"NewZerosModuleDefaultDtype_basic",
"NewZerosModuleFalsePinMemory_basic",
"NewZerosModuleFloat2D_basic",
"NewZerosModuleFloat3D_basic",
"NewZerosModuleInt2D_basic",
"NewZerosModuleInt3D_basic",
"NllLossModuleBackward1DMeanWeight_basic",
"NllLossModuleBackward1DMean_basic",
"NllLossModuleBackward1DSumWeight_basic",
"NllLossModuleBackward1DSum_basic",
"NllLossModuleBackward1DWeight_basic",
"NllLossModuleBackward1D_basic",
"NllLossModuleBackwardMeanWeight_basic",
"NllLossModuleBackwardMean_basic",
"NllLossModuleBackwardSumWeight_basic",
"NllLossModuleBackwardSum_basic",
"NllLossModuleBackwardWeight_basic",
"NllLossModuleBackward_basic",
"NllLossModuleBackward_ignore_index",
"NllLossModule_1D_basic",
"NllLossModule_basic",
"NllLossModule_ignore_index_out_of_bounds_basic",
"NllLossModule_mean_basic",
"NllLossModule_sum_basic",
"NumelModule_basic",
"NumelZeroRankModule_basic",
"OnesLikeModule_defaultDtype",
"OnesLikeModule_falsePinMemory",
"OnesLikeModule_float",
"OnesLikeModule_int",
"OnesModuleDefaultDtype_basic",
"OnesModuleFalsePinMemory_basic",
"OnesModuleFloat_basic",
"OnesModuleInt_basic",
"QuantizedMLP_basic",
"RandLikeDtypeModule_basic",
"RandLikeModule_basic",
"ReduceMaxKeepDimReturnBoth_basic",
"ReduceMaxNegativeDim_basic",
"ReshapeAliasCollapseModule_basic",
"ReshapeAliasExpandModule_basic",
"ReturnThreeTensorFloat32_basic",
"ReturnTwoTensorF32I64_basic",
"ScalarImplicitFloatModule_basic",
"ScalarImplicitIntModule_basic",
"SelectIntModule_basic",
"SliceEndSleStartModule_basic",
"SliceNegIdxModule_basic",
"SliceOutOfLowerBoundEndIndexModule_basic",
"SliceOutOfLowerBoundStartIndexModule_basic",
"SliceOutOfUpperBoundIndexModule_basic",
"SliceSingleIdxModule_basic",
"SliceSizeTwoStepModule_basic",
"SliceStartEqEndModule_basic",
"SliceWholeTensorModule_basic",
"SqrtIntConstantModule_basic",
"SqrtIntModule_basic",
"StdBiasedModule_basic",
"StdUnbiasedModule_basic",
"SubFloatModule_basic",
"SubIntModule_basic",
"TModuleRank0_basic",
"TModuleRank1_basic",
"TableBatchEmbeddingModule_basic",
"TensorToBoolZeroRank_basic",
"TensorToBool_basic",
"TensorToFloatZeroRank_basic",
"TensorToFloat_basic",
"TensorToIntZeroRank_basic",
"TensorToInt_basic",
"TensorsConcatModule_basic",
"TestMultipleTensorAndPrimitiveTypesReturn_basic",
"TestMultipleTensorReturn_basic",
"Threshold1dFloatModule_basic",
"Threshold1dIntI32Module_basic",
"Threshold1dIntModule_basic",
"Threshold2dFloatModule_basic",
"Threshold2dIntModule_basic",
"Threshold3dFloatModule_basic",
"Threshold3dIntModule_basic",
"ThresholdBackward1dFloatModule_basic",
"ThresholdBackward1dIntModule_basic",
"ThresholdBackward1dMixedModule_basic",
"ThresholdBackward2dFloatModule_basic",
"ThresholdBackward2dIntModule_basic",
"ThresholdBackward2dMixedModule_basic",
"ThresholdBackward3dFloatModule_basic",
"ThresholdBackward3dIntModule_basic",
"ThresholdBackward3dMixedModule_basic",
"TorchPrimLoopForLikeModule_basic",
"TorchPrimLoopWhileLikeModule_basic",
"UniformModule_basic",
"UniformStaticModule_basic",
"UnsafeViewCollapseDynamicWithAtenSizeIntModule_basic",
"VarBiasedModule_basic",
"VarUnbiasedModule_basic",
"ViewCollapseDynamicWithAtenSizeIntModule_basic",
"ZeroFloat32Module_basic",
"ZeroInt32Module_basic",
"ZeroInt64Module_basic",
"ZerosLikeModule_defaultDtype",
"ZerosLikeModule_falsePinMemory",
"ZerosLikeModule_float",
"ZerosLikeModule_int",
"ZerosModuleDefaultDtype_basic",
"ZerosModuleFalsePinMemory_basic",
"ZerosModuleFloat2D_basic",
"ZerosModuleFloat3D_basic",
"ZerosModuleInt2D_basic",
"ZerosModuleInt3D_basic",
}
Loading