Skip to content

Use cadence replace ops passes in turing partitioner. #12788

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.

Sign up for GitHub

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

Open
wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Use cadence replace ops passes in turing partitioner. #12788

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
61 changes: 33 additions & 28 deletions backends/cadence/aot/replace_ops.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2407,47 +2407,52 @@ def call_operator(self, op, args, kwargs, meta):
)


# This class encapsulates all the functions that replace/switch one op in the
# graph with another.
class CadenceReplaceOpsInGraph:
# Encapsulates the generic graph improvement passes.
class GenericReplaceOpsInGraph:
passes = [
ReplaceEmptyTensorsWithFullPass,
ReplaceFunctionallyEquivalentOpTargets,
ReplaceSqueezeAndUnsqueezeWithViewPass,
ReplaceSplitWithSlicePass,
ReplaceSelectWithViewOpPass,
ReplaceMMWithAddMMPass,
ReplaceRepeatWithCatPass,
ReplaceFullLikeWithFullPass,
ReplaceFunctionallyEquivalentOpTargets,
ReplaceConvolutionOptionalArgsWithConcreteArgsPass,
ReplaceAddMMWithLinearPass,
ReplacePadWithCatPass,
ReplaceConstantPadNdWithSlicePass,
ReplaceTrivialConvWithLinear,
ReplaceScalarTensorWithFullPass,
ReplaceInfArgInFullWithValuePass,
ReplaceLogicalNotBooleanWhereWithWherePass,
ReplaceAdaptiveAvgPoolWithAtenAvgPoolPass,
ReplaceAtenApproxGeluWithApproxGeluPass,
ReplaceAtenConvolutionWithJarvisConvolutionPass,
ReplacePT2QuantWithCadenceQuantPass,
ReplacePT2DequantWithCadenceDequantPass,
ReplacePowWithMulPass,
ReplaceEmptyTensorsWithFullPass,
ReplaceScalarWithTensorArgPass,
RemoveNopSelectOpPass,
ReplaceNopTransposeOrPermuteWithViewPass,
]

# Encapsulates all the passes which replace ops in graph including Sadence specific ones.
class CadenceReplaceOpsInGraph:
cadence_specific_passes = [
ReplacePermuteWithTransposePass,
ReplaceScalarWithTensorArgPass,
ReplaceConvolutionOptionalArgsWithConcreteArgsPass,
ReplaceMMWithAddMMPass,
ReplaceSqueezeAndUnsqueezeWithViewPass,
ReplaceAddMMWithLinearPass,
RemoveNopSelectOpPass,
ReplaceSelectWithViewOpPass,
ReplaceRepeatWithCatPass,
ReplacePadWithCatPass,
ReplaceConstantPadNdWithSlicePass,
ReplaceConvWithChannelLastConvPass,
ReplaceAtenConvolutionWithJarvisConvolutionPass,
ForceChannelLastForConvPass,
ReplaceTrivialConvWithLinear,
ReplaceConvWithIm2RowAndLinear,
ReplaceTransposedConvWithLinearPass,
# This pass should be after passes that replace conv -> im2row + linear.
ReplaceIm2RowWithViewPass,
MakeSliceAndCatDimOutermostPass,
ReplaceMatmulWithTransposedMatmulPass,
ReplaceNopTransposeOrPermuteWithViewPass,
ReplaceLinearWithFullyConnectedOpPass,
ReplaceScalarTensorWithFullPass,
ReplaceFullLikeWithFullPass,
ReplaceInfArgInFullWithValuePass,
ReplaceLogicalNotBooleanWhereWithWherePass,
ReplacePT2QuantWithCadenceQuantPass,
ReplacePT2DequantWithCadenceDequantPass,
ReplaceSingleElementTensorArgumentsFromFullOpWithScalarPass,
ReplaceAdaptiveAvgPoolWithAtenAvgPoolPass,
ReplaceAtenAvgPoolWithJarvisAvgPoolPass,
ReplaceWhereWithFullArgsWithWhereScalar,
ReplaceAtenApproxGeluWithApproxGeluPass,
ReplaceSplitWithSlicePass,
ReplacePowWithMulPass,
ReplaceMulTensorWithMulAndFullOpsPass,
]
passes = GenericReplaceOpsInGraph.passes + cadence_specific_passes
Loading