mlir.dialects._tensor_transform_ops_gen¶

Attributes¶

_ods_ir

Classes¶

`ApplyBubbleUpExtractSlicePatternsOp`	Indicates that producers of tensor.extract_slice should swap and operate on
`ApplyDecomposeTensorConcatPatternsOp`	Indicates that tensor.concat ops should be decomposed into a chain of
`ApplyDropRedundantInsertSliceRankExpansionPatternsOp`	Indicates that redundant tensor.insert_slice rank reductions should be
`ApplyFoldTensorEmptyPatternsOp`	Indicates that tensor.extract_slice and reassociative reshapes should be
`ApplyFoldTensorSubsetOpsIntoVectorTransfersPatternsOp`	Indicates that tensor.extract_slice -> vector.transfer_read and
`ApplyFoldTensorSubsetOpsPatternsOp`	Indicates that tensor.empty should be folded with tensor.extract_slice,
`ApplyMergeConsecutiveInsertExtractSlicePatternsOp`	Indicates that consecutive tensor.extract_slice/tensor.insert_slice ops
`ApplyReassociativeReshapeFoldingPatternsOp`	Indicates that reassociative reshapes (tensor.collapse_shape /
`ApplyRewriteTensorOpsAsConstantPatternsOp`	Indicates that tensor ops (such as tensor.generate) should be replaced with
`MakeLoopIndependentOp`	Rewrite the targeted ops such that their index-typed operands no longer
`TypeConversionCastShapeDynamicDimsOp`	Populates a type converter with conversion materialization functions that

Functions¶

`apply_patterns_tensor_bubble_up_extract_slice`(...)
`apply_patterns_tensor_decompose_concat`(...)
`apply_patterns_tensor_drop_redundant_insert_slice_rank_expansion`(...)
`apply_patterns_tensor_fold_tensor_empty`(...)
`apply_patterns_tensor_fold_tensor_subset_ops_into_vector_transfers`(...)
`apply_patterns_tensor_fold_tensor_subset_ops`(...)
`apply_patterns_tensor_merge_consecutive_insert_extract_slice`(...)
`apply_patterns_tensor_reassociative_reshape_folding`(...)
`apply_patterns_tensor_rewrite_as_constant`(...)
`tensor_make_loop_independent`(→ _ods_ir)
`type_conversion_tensor_cast_shape_dynamic_dims`(...)

Module Contents¶

mlir.dialects._tensor_transform_ops_gen._ods_ir¶

class mlir.dialects._tensor_transform_ops_gen.ApplyBubbleUpExtractSlicePatternsOp(*, loc=None, ip=None)¶

Bases: _ods_ir

Indicates that producers of tensor.extract_slice should swap and operate on the result of the slice.

OPERATION_NAME = 'transform.apply_patterns.tensor.bubble_up_extract_slice'¶

_ODS_REGIONS = (0, True)¶

mlir.dialects._tensor_transform_ops_gen.apply_patterns_tensor_bubble_up_extract_slice(*, loc=None, ip=None) → ApplyBubbleUpExtractSlicePatternsOp¶

class mlir.dialects._tensor_transform_ops_gen.ApplyDecomposeTensorConcatPatternsOp(*, loc=None, ip=None)¶

Bases: _ods_ir

Indicates that tensor.concat ops should be decomposed into a chain of tensor.insert_slice operations inserting into a materialized destination.

OPERATION_NAME = 'transform.apply_patterns.tensor.decompose_concat'¶

_ODS_REGIONS = (0, True)¶

mlir.dialects._tensor_transform_ops_gen.apply_patterns_tensor_decompose_concat(*, loc=None, ip=None) → ApplyDecomposeTensorConcatPatternsOp¶

class mlir.dialects._tensor_transform_ops_gen.ApplyDropRedundantInsertSliceRankExpansionPatternsOp(*, loc=None, ip=None)¶

Bases: _ods_ir

Indicates that redundant tensor.insert_slice rank reductions should be dropped. E.g., cases where a tensor.extract_slice rank reduction immediately follows an inverse tensor.insert_slice rank expansion.

OPERATION_NAME = 'transform.apply_patterns.tensor.drop_redundant_insert_slice_rank_expansion'¶

_ODS_REGIONS = (0, True)¶

mlir.dialects._tensor_transform_ops_gen.apply_patterns_tensor_drop_redundant_insert_slice_rank_expansion(*, loc=None, ip=None) → ApplyDropRedundantInsertSliceRankExpansionPatternsOp¶

class mlir.dialects._tensor_transform_ops_gen.ApplyFoldTensorEmptyPatternsOp(*, fold_single_use_only=None, loc=None, ip=None)¶

Bases: _ods_ir

Indicates that tensor.extract_slice and reassociative reshapes should be folded into tensor.empty.

If fold_single_use_only is set to “true”, only tensor.empty that have a single use are folded.

OPERATION_NAME = 'transform.apply_patterns.tensor.fold_tensor_empty'¶

_ODS_REGIONS = (0, True)¶

fold_single_use_only() → _ods_ir¶

mlir.dialects._tensor_transform_ops_gen.apply_patterns_tensor_fold_tensor_empty(*, fold_single_use_only=None, loc=None, ip=None) → ApplyFoldTensorEmptyPatternsOp¶

class mlir.dialects._tensor_transform_ops_gen.ApplyFoldTensorSubsetOpsIntoVectorTransfersPatternsOp(*, loc=None, ip=None)¶

Bases: _ods_ir

Indicates that tensor.extract_slice -> vector.transfer_read and vector.transfer_write -> tensor.insert_slice op chains should be folded into vector tranfer read and write ops

OPERATION_NAME = 'transform.apply_patterns.tensor.fold_tensor_subset_ops_into_vector_transfers'¶

_ODS_REGIONS = (0, True)¶

mlir.dialects._tensor_transform_ops_gen.apply_patterns_tensor_fold_tensor_subset_ops_into_vector_transfers(*, loc=None, ip=None) → ApplyFoldTensorSubsetOpsIntoVectorTransfersPatternsOp¶

class mlir.dialects._tensor_transform_ops_gen.ApplyFoldTensorSubsetOpsPatternsOp(*, loc=None, ip=None)¶

Bases: _ods_ir

Indicates that tensor.empty should be folded with tensor.extract_slice, tensor.expand_shape and tensor.collapse_shape.

OPERATION_NAME = 'transform.apply_patterns.tensor.fold_tensor_subset_ops'¶

_ODS_REGIONS = (0, True)¶

mlir.dialects._tensor_transform_ops_gen.apply_patterns_tensor_fold_tensor_subset_ops(*, loc=None, ip=None) → ApplyFoldTensorSubsetOpsPatternsOp¶

class mlir.dialects._tensor_transform_ops_gen.ApplyMergeConsecutiveInsertExtractSlicePatternsOp(*, loc=None, ip=None)¶

Bases: _ods_ir

Indicates that consecutive tensor.extract_slice/tensor.insert_slice ops should be merged into a single op. These patterns are not canonicalizations because the bufferization is sensitive to IR structure.

OPERATION_NAME = 'transform.apply_patterns.tensor.merge_consecutive_insert_extract_slice'¶

_ODS_REGIONS = (0, True)¶

mlir.dialects._tensor_transform_ops_gen.apply_patterns_tensor_merge_consecutive_insert_extract_slice(*, loc=None, ip=None) → ApplyMergeConsecutiveInsertExtractSlicePatternsOp¶

class mlir.dialects._tensor_transform_ops_gen.ApplyReassociativeReshapeFoldingPatternsOp(*, loc=None, ip=None)¶

Bases: _ods_ir

Indicates that reassociative reshapes (tensor.collapse_shape / tensor.expand_shape) should be folded with inverse rank expansions / rank reductions (via tensor.insert_slice / tensor.extract_slice).

OPERATION_NAME = 'transform.apply_patterns.tensor.reassociative_reshape_folding'¶

_ODS_REGIONS = (0, True)¶

mlir.dialects._tensor_transform_ops_gen.apply_patterns_tensor_reassociative_reshape_folding(*, loc=None, ip=None) → ApplyReassociativeReshapeFoldingPatternsOp¶

class mlir.dialects._tensor_transform_ops_gen.ApplyRewriteTensorOpsAsConstantPatternsOp(*, aggressive=None, loc=None, ip=None)¶

Bases: _ods_ir

Indicates that tensor ops (such as tensor.generate) should be replaced with constants (arith.constant) when possible.

OPERATION_NAME = 'transform.apply_patterns.tensor.rewrite_as_constant'¶

_ODS_REGIONS = (0, True)¶

aggressive() → bool¶

mlir.dialects._tensor_transform_ops_gen.apply_patterns_tensor_rewrite_as_constant(*, aggressive=None, loc=None, ip=None) → ApplyRewriteTensorOpsAsConstantPatternsOp¶

class mlir.dialects._tensor_transform_ops_gen.MakeLoopIndependentOp(transformed, target, num_loops, *, loc=None, ip=None)¶

Bases: _ods_ir

Rewrite the targeted ops such that their index-typed operands no longer depend on any loop induction variable of the num_loop enclosing scf.for loops. I.e., compute an upper bound that is independent of any such loop IV for every tensor dimension. The transformed op could then be hoisted from the num_loop enclosing loops. To preserve the original semantics, place a tensor.extract_slice inside the loop.

Currently supported operations are:

tensor.empty: Replaced with a new tensor.empty with upper bound sizes,

followed by a tensor.extract_slice. * tensor.pad: Replaced by an upper bound padding, followed by a tensor.extract_slice.

Return modes¶

This operation fails if at least one induction variable could not be eliminated. In case the targeted op is already independent of induction variables, this transform succeeds and returns the unmodified target op.

Otherwise, the returned handle points to a subset of the produced ops:

tensor.empty: The returned handle points to the tensor.extract_slice op.
tensor.pad: The returned handle points to the tensor.extract_slice op.

This transform op consumes the target handle and produces a result handle.

OPERATION_NAME = 'transform.tensor.make_loop_independent'¶

_ODS_REGIONS = (0, True)¶

target() → _ods_ir¶

num_loops() → _ods_ir¶

transformed() → _ods_ir¶

mlir.dialects._tensor_transform_ops_gen.tensor_make_loop_independent(transformed, target, num_loops, *, loc=None, ip=None) → _ods_ir¶

class mlir.dialects._tensor_transform_ops_gen.TypeConversionCastShapeDynamicDimsOp(*, ignore_dynamic_info=None, loc=None, ip=None)¶

Bases: _ods_ir

Populates a type converter with conversion materialization functions that cast a tensor value between two cast-compatible tensors. See tensor.cast for more information on cast compatibility between tensors.

If ignore_dynamic_info is not set, this will set an additional constraint that source materializations do not cast dynamic dimensions to static ones.

OPERATION_NAME = 'transform.type_conversion.tensor.cast_shape_dynamic_dims'¶

_ODS_REGIONS = (0, True)¶

ignore_dynamic_info() → bool¶

mlir.dialects._tensor_transform_ops_gen.type_conversion_tensor_cast_shape_dynamic_dims(*, ignore_dynamic_info=None, loc=None, ip=None) → TypeConversionCastShapeDynamicDimsOp¶