Namespaces
	detail

Classes
class	LinalgOpToLibraryCallRewrite

class	Aliases
	A very primitive alias analysis which just records for each view, either: More...

class	LinalgDependenceGraph
	Data structure for holding a dependence graph that operates on LinalgOp and views as SSA values. More...

struct	TiledLinalgOp
	Perform standalone tiling of a single LinalgOp by `tileSizes`. More...

struct	LinalgPromotionOptions

struct	PromotionInfo
	Create a new buffer using the `allocationFn` provided. More...

struct	MultiSizeSpecification
	A description of a multi-size tiling comprising tile sizes and numbers of tiles, expressed as Values which may or may not be constant. More...

struct	StaticMultiSizeSpecification

struct	ForeachThreadTilingResult
	Rewrite a TilingInterface `op` to a tiled `scf.foreach_thread`, applying tiling by `numThreads`. More...

struct	ForeachThreadReductionTilingResult
	Transformation information returned after reduction tiling. More...

struct	LinalgPaddingOptions

struct	LinalgTilingAndFusionOptions

struct	LinalgTilingOptions

struct	LinalgPaddingPattern
	Linalg padding pattern. More...

struct	DownscaleSizeOneWindowed2DConvolution
	Rewrites 2-D convolution ops with size-1 window dimensions into 1-D convolution ops. More...

struct	DownscaleDepthwiseConv2DNhwcHwcOp
	Rewrites 2-D depthwise convolution ops with size-1 (w, kw) or (h, kh) dimensions into 1-D depthwise convolution ops. More...

struct	LinalgGeneralizationPattern
	Linalg generalization pattern. More...

struct	CopyVectorizationPattern
	Vectorization pattern for memref::CopyOp. More...

struct	PadOpTransformationPattern
	tensor::PadOp is not canonicalized away yet, so we provide a transformation to `linalg.generic`. More...

struct	GeneralizePadOpPattern
	Rewrite a tensor::PadOp into a sequence of EmptyOp, FillOp and InsertSliceOp. More...

struct	GeneralizeOuterUnitDimsPackOpPattern
	Rewrites a tensor::PackOp into a sequence of tensor.pad + linalg.transpose + tensor.insert_slice ops, where the tensor::PackOp has outer dims being all 1s. More...

struct	GeneralizeOuterUnitDimsUnPackOpPattern
	Rewrites a tensor::UnPackOp into a sequence of rank-reduced extract_slice op. More...

struct	LinalgCopyVTRForwardingPattern
	Match and rewrite for the pattern: More...

struct	LinalgCopyVTWForwardingPattern
	Match and rewrite for the pattern: More...

struct	ExtractSliceOfPadTensorSwapPattern
	Rewrite extract_slice(tensor.pad(x)) into tensor.pad(extract_slice(x)). More...

struct	SplitReductionOptions
	Split Reduction options. More...

struct	SplitReductionResult
	Apply transformation to split the single linalg op reduction into a parallel and reduction dimension. More...

struct	PackTransposeResult
	Struct to hold the result of a `packTranspose` call. More...

struct	SliceParameters
	A struct containg offsets-sizes-strides arguments of the tiled shape. More...

struct	FusionInfo
	A struct containing the Linalg producer before and after fusion. More...

struct	ProcInfo
	Callback function type used to get processor ID, and number of processors used for distribution for all parallel loops generated. More...

struct	LinalgLoopDistributionOptions
	Options that allow distribution of loops generated in Linalg transforms to processors while generating the loops. More...

class	TileLoopNest
	A struct to manage the tile loop nest specific information. More...

struct	RegionMatcher
	A struct containing common matchers over linalg op's region. More...

struct	GenerateLoopNest
	Utility class used to generate nested loops with ranges described by `loopRanges` and loop type described by the `iteratorTypes`. More...

Typedefs
using	LinalgLoops = SmallVector< Operation *, 4 >

using	ControlFusionFn = std::function< bool(OpOperand *fusedOperand)>
	Function type which is used to control when to stop fusion. More...

using	GetCollapsableDimensionsFn = std::function< SmallVector< ReassociationIndices >(linalg::GenericOp)>
	Function type to control generic op dimension collapsing. More...

using	AllocBufferCallbackFn = std::function< std::optional< Value >(OpBuilder &b, memref::SubViewOp subView, ArrayRef< Value > boundingSubViewSize, DataLayout &layout)>
	Callback function type used to perform the allocation for the promoted `subView`. More...

using	DeallocBufferCallbackFn = std::function< LogicalResult(OpBuilder &b, Value buffer)>
	Callback function type used to deallocate the buffers used to hold the promoted subview. More...

using	CopyCallbackFn = std::function< LogicalResult(OpBuilder &b, Value src, Value dst)>
	Callback function type used to insert copy from original subview to subview of the promoted region for the read operands/subview of promoted region to original subview for the results. More...

using	TileSizeComputationFunction = std::function< SmallVector< Value, 4 >(OpBuilder &, Operation *)>

using	LoopIndexToRangeIndexMap = DenseMap< int, int >
	Creates a number of ranges equal to the number of non-zero in `tileSizes`. More...

using	OptimizeCopyFn = std::function< LogicalResult(PatternRewriter &, tensor::PadOp, Value)>

using	ControlSplitReductionFn = std::function< SplitReductionOptions(LinalgOp op)>
	Function signature to control reduction splitting. More...

using	FusableOpDependencesTy = llvm::MapVector< Operation *, SmallVector< LinalgDependenceGraph::LinalgDependenceGraphElem, 1 > >

using	ProcInfoCallBackFn = std::function< SmallVector< ProcInfo >(OpBuilder &b, Location loc, ArrayRef< Range > parallelLoopRanges)>

Enumerations
enum class	LinalgTilingLoopType { Loops = 0 , AffineLoops = 1 , ParallelLoops = 2 }
	The type of loops to be generated during tiling. More...

enum class	DistributionMethod { Cyclic = 0 , CyclicNumProcsGeNumIters = 1 , CyclicNumProcsEqNumIters = 2 , None = 3 }
	Scheme used to distribute loops to processors. More...

Functions
void	populateLinalgToStandardConversionPatterns (RewritePatternSet &patterns)
	Populate the given list with patterns that convert from Linalg to Standard. More...

std::string	generateLibraryCallName (Operation *op)
	Returns the name mangled library call name to disambiguate between different overloads at the C level. More...

SmallVector< AffineExpr, 4 >	makeAffineDimExprs (unsigned num, unsigned &startIdx, MLIRContext *context)
	Returns `num` AffineDimExpr dimensions at positions [startIdx, startIdx + num) and increments `startIdx` to `startIdx + num`. More...

AffineMap	extractOrIdentityMap (std::optional< AffineMap > maybeMap, unsigned rank, MLIRContext *context)
	Returns `maybeMap.get()` if `maybeMap` is set, otherwise returns the symbol-less identity map of `rank`. More...

SmallVector< AffineExpr, 4 >	concat (ArrayRef< AffineExpr > a, ArrayRef< AffineExpr > b)
	Return the vector that is the concatenation of `a` and `b`. More...

bool	isaContractionOpInterface (LinalgOp linalgOp)
	Checks whether `linalgOp` conforms to ContractionOpInterface. More...

void	registerTransformDialectExtension (DialectRegistry &registry)

void	registerBufferizableOpInterfaceExternalModels (DialectRegistry &registry)

void	hoistRedundantVectorTransfers (func::FuncOp func)
	Hoist vector.transfer_read/vector.transfer_write on buffers pairs out of immediately enclosing scf::ForOp iteratively, if the following conditions are true: More...

void	hoistRedundantVectorTransfersOnTensor (func::FuncOp func)
	Same behavior as `hoistRedundantVectorTransfers` but works on tensors instead of buffers. More...

FailureOr< Value >	hoistPaddingOnTensors (tensor::PadOp opToHoist, int numLoops, ArrayRef< int64_t > transposeVector, tensor::PadOp &hoistedOp, SmallVectorImpl< GenericOp > &transposeOps)
	Mechanically hoist padding operations on tensors by `numLoops` into a new, generally larger tensor. More...

void	registerTilingInterfaceExternalModels (DialectRegistry &registry)

void	populatePadTensorTilingPatterns (RewritePatternSet &patterns, const LinalgTilingOptions &options)

void	populateDecomposeLinalgOpsPattern (RewritePatternSet &patterns, bool removeDeadArgsAndResults=true)
	Populate patterns for splitting a `LinalgOp` with multiple statements within its payload into multiple `GenericOp` that have a single statement. More...

void	populateConvertToDestinationStylePatterns (RewritePatternSet &patterns)
	Populate patterns that convert non-destination-style ops to destination style ops. More...

void	populateConvolutionVectorizationPatterns (RewritePatternSet &patterns, PatternBenefit benefit=1)
	Populate patterns for vectorizing low-D convolution ops. More...

void	populateElementwiseToLinalgConversionPatterns (RewritePatternSet &patterns)
	Populate patterns that convert `ElementwiseMappable` ops to linalg parallel loops. More...

void	populateSparseTensorRewriting (RewritePatternSet &patterns)
	Populate patterns that are only useful in the context of sparse tensors. More...

void	populateElementwiseOpsFusionPatterns (RewritePatternSet &patterns, const ControlFusionFn &controlElementwiseOpFusion)
	Patterns for fusing linalg operation on tensors. More...

void	populateDataLayoutPropagationPatterns (RewritePatternSet &patterns)
	Patterns to bubble up or down data layout ops across other operations. More...

void	populateEraseUnusedOperandsAndResultsPatterns (RewritePatternSet &patterns)
	Pattern to remove dead operands and results of `linalg.generic` operations. More...

void	populateEraseUnnecessaryInputsPatterns (RewritePatternSet &patterns)
	Patterns to promote inputs to outputs and remove unused inputs of `linalg.generic` ops. More...

void	populateCollapseDimensions (RewritePatternSet &patterns, const GetCollapsableDimensionsFn &controlCollapseDimensions)
	Pattern to collapse dimensions in a linalg.generic op. More...

void	populateFoldReshapeOpsByExpansionPatterns (RewritePatternSet &patterns, const ControlFusionFn &controlFoldingReshapes)
	Patterns to fold an expanding (collapsing) tensor_reshape operation with its producer (consumer) generic operation by expanding the dimensionality of the loop in the generic op. More...

void	populateFoldReshapeOpsByCollapsingPatterns (RewritePatternSet &patterns, const ControlFusionFn &controlFoldingReshapes)
	Patterns to fold an expanding tensor.expand_shape operation with its producer generic operation by collapsing the dimensions of the generic op. More...

void	populateConstantFoldLinalgOperations (RewritePatternSet &patterns, const ControlFusionFn &controlFn)
	Patterns to constant fold Linalg operations. More...

void	populateFuseTensorPadWithProducerLinalgOpPatterns (RewritePatternSet &patterns)
	Pattern to fuse a `tensor.pad` operation with the producer of its source, if the producer is a `linalg` operation with all parallel iterator types. More...

void	populateLinalgNamedOpConversionPatterns (RewritePatternSet &patterns)
	Patterns to convert from one named op to another. More...

void	populateFoldUnitExtentDimsViaReshapesPatterns (RewritePatternSet &patterns)
	Patterns to fold unit-extent dimensions in operands/results of linalg ops on tensors via reassociative reshape ops. More...

void	populateFoldUnitExtentDimsViaSlicesPatterns (RewritePatternSet &patterns)
	Patterns to fold unit-extent dimensions in operands/results of linalg ops on tensors via rank-reducing slices. More...

void	populateMoveInitOperandsToInputPattern (RewritePatternSet &patterns)
	A pattern that converts init operands to input operands. More...

void	populateInlineConstantOperandsPatterns (RewritePatternSet &patterns)
	Patterns that are used to inline constant operands into linalg generic ops. More...

void	populateBubbleUpExtractSliceOpPatterns (RewritePatternSet &patterns)
	Patterns that are used to bubble up extract slice op above linalg op. More...

void	populateSwapExtractSliceWithFillPatterns (RewritePatternSet &patterns)
	Adds patterns that waps tensor.extract_slice(linalg.fill(cst, init)) into linalg.fill(cst, tensor.extract_slice(init)). More...

bool	areElementwiseOpsFusable (OpOperand *fusedOperand)
	Return true if two `linalg.generic` operations with producer/consumer relationship through `fusedOperand` can be fused using elementwise op fusion. More...

FailureOr< Operation * >	fuseElementwiseOps (RewriterBase &rewriter, OpOperand *fusedOperand)
	Fuse two `linalg.generic` operations that have a producer-consumer relationship captured through `fusedOperand`. More...

std::pair< TilingInterface, TilingInterface >	splitOp (RewriterBase &rewriter, TilingInterface op, unsigned dimension, OpFoldResult splitPoint)
	Split the given `op` into two parts along the given iteration space `dimension` at the specified `splitPoint`, and return the two parts. More...

FailureOr< TiledLinalgOp >	tileLinalgOp (RewriterBase &b, LinalgOp op, const LinalgTilingOptions &options)

SmallVector< Value >	peelLoop (RewriterBase &rewriter, Operation *op)
	Try to peel and canonicalize loop `op` and return the new result. More...

void	peelLoops (RewriterBase &rewriter, ArrayRef< scf::ForOp > loops)
	Peel and canonicalize 'loops'. More...

FailureOr< GenericOp >	interchangeGenericOp (RewriterBase &rewriter, GenericOp genericOp, ArrayRef< unsigned > interchangeVector)
	Interchange the `iterator_types` and `iterator_maps` dimensions and adapts the index accesses of `op`. More...

FailureOr< GenericOp >	generalizeNamedOp (RewriterBase &rewriter, LinalgOp namedOp)
	Create a GenericOp from the given named operation `namedOp` and replace namedOp. More...

FailureOr< PromotionInfo >	promoteSubviewAsNewBuffer (OpBuilder &b, Location loc, memref::SubViewOp subView, const AllocBufferCallbackFn &allocationFn, DataLayout &layout)

FailureOr< LinalgOp >	promoteSubViews (OpBuilder &b, LinalgOp op, const LinalgPromotionOptions &options)
	Promote the `subViews` into a new buffer allocated at the insertion point `b`. More...

LogicalResult	vectorize (RewriterBase &rewriter, LinalgOp linalgOp, ArrayRef< int64_t > inputVectorSizes={}, bool vectorizeNDExtract=false)
	Emit a suitable vector form for a Linalg op. More...

LogicalResult	vectorizeCopy (RewriterBase &builder, memref::CopyOp copyOp)
	Emit a suitable vector form for a Copy op with fully static shape. More...

FailureOr< LinalgLoops >	linalgOpToLoops (PatternRewriter &rewriter, LinalgOp linalgOp)
	Emit a loop nest of `scf.for` with the proper body for `linalgOp`. More...

FailureOr< LinalgLoops >	linalgOpToParallelLoops (PatternRewriter &rewriter, LinalgOp linalgOp)
	Emit a loop nest of `scf.parallel` with the proper body for `linalgOp`. More...

FailureOr< LinalgLoops >	linalgOpToAffineLoops (PatternRewriter &rewriter, LinalgOp linalgOp)
	Emit a loop nest of `affine.for` with the proper body for `linalgOp`. More...

LogicalResult	promoteSubviewsPrecondition (Operation *op, LinalgPromotionOptions options)
	Promote memref.subviews feeding linalg-on-buffers operations. More...

LogicalResult	vectorizeLinalgOpPrecondition (LinalgOp linalgOp, ArrayRef< int64_t > inputVectorSizes={}, bool vectorizeNDExtract=false)
	Return success if the operation can be vectorized. More...

std::tuple< SmallVector< Range, 4 >, LoopIndexToRangeIndexMap >	makeTiledLoopRanges (RewriterBase &b, Location loc, AffineMap map, ArrayRef< OpFoldResult > allShapeSizes, ArrayRef< OpFoldResult > allTileSizes)

FailureOr< MultiSizeSpecification >	computeMultiTileSizes (OpBuilder &builder, LinalgOp op, unsigned dimension, OpFoldResult targetSize, OpFoldResult divisor, bool emitAssertions=true)
	Emits the IR computing the multi-sized tiling specification with two tile sizes not exceeding `targetSize`, each divisible by `sizeDivisor`, such that there exist numbers of tiles with these sizes that fully cover the given iteration space `dimension` of the structured `op`. More...

FailureOr< StaticMultiSizeSpecification >	computeStaticMultiTileSizes (LinalgOp op, unsigned dimension, int64_t targetSize, int64_t divisor)

FailureOr< ForeachThreadTilingResult >	tileToForeachThreadOp (RewriterBase &builder, TilingInterface op, ArrayRef< OpFoldResult > numThreads, std::optional< ArrayAttr > mapping)

FailureOr< ForeachThreadTilingResult >	tileToForeachThreadOpUsingTileSizes (RewriterBase &builder, TilingInterface op, ArrayRef< OpFoldResult > tileSizes, std::optional< ArrayAttr > mapping)
	Same as `tileToForeachThreadOp`, but calculate the number of threads required using the given tileSizes. More...

FailureOr< ForeachThreadReductionTilingResult >	tileReductionUsingForeachThread (RewriterBase &b, PartialReductionOpInterface op, ArrayRef< OpFoldResult > numThreads, ArrayRef< OpFoldResult > tileSizes={}, std::optional< ArrayAttr > mapping=std::nullopt)
	Method to tile a reduction to parallel iterations computing partial reductions. More...

void	transformIndexOps (RewriterBase &b, LinalgOp op, SmallVectorImpl< Value > &ivs, const LoopIndexToRangeIndexMap &loopIndexToRangeIndex)
	All indices returned by IndexOp should be invariant with respect to tiling. More...

RewritePatternSet	getLinalgTilingCanonicalizationPatterns (MLIRContext *ctx)
	Canonicalization patterns relevant to apply after tiling patterns. More...

void	populateLinalgTilingCanonicalizationPatterns (RewritePatternSet &patterns)

std::optional< vector::CombiningKind >	getCombinerOpKind (Operation *combinerOp)
	Return vector::CombiningKind for the given op. More...

void	populateLinalgNamedOpsGeneralizationPatterns (RewritePatternSet &patterns)
	Linalg generalization patterns. More...

void	populateDecomposeConvolutionPatterns (RewritePatternSet &patterns, PatternBenefit benefit=1)
	Linalg decompose convolutions patterns. More...

FailureOr< SmallVector< Value > >	rewriteAsPaddedOp (OpBuilder &b, LinalgOp opToPad, ArrayRef< int64_t > paddingDimensions, ArrayRef< Attribute > paddingValues, ArrayRef< bool > packPaddings, LinalgOp &paddedOp)
	Pad the iterator dimensions `paddingDimensions` of all `opToPad` operands to a static bounding box. More...

void	populatePadOpVectorizationPatterns (RewritePatternSet &patterns, PatternBenefit baseBenefit=1)
	Populates `patterns` with patterns that vectorize tensor.pad. More...

void	populateExtractOpVectorizationPatterns (RewritePatternSet &patterns, PatternBenefit baseBenefit=1)

void	populateSplitReductionPattern (RewritePatternSet &patterns, const ControlSplitReductionFn &controlSplitReductionFn, bool useAlloc=false)
	Patterns to apply `splitReduction` below. More...

FailureOr< SplitReductionResult >	splitReduction (PatternRewriter &b, LinalgOp op, const ControlSplitReductionFn &controlSplitReductionFn, bool useAlloc=false)

FailureOr< SplitReductionResult >	splitReductionByScaling (PatternRewriter &b, LinalgOp op, const ControlSplitReductionFn &controlSplitReductionFn, bool useAlloc=false)
	Scaling-based implementation of the split reduction transformation. More...

FailureOr< SmallVector< Value > >	collapseGenericOpIterationDims (GenericOp genericOp, ArrayRef< ReassociationIndices > foldedIterationDims, RewriterBase &rewriter)
	Collapses dimensions of linalg.generic operation. More...

FailureOr< linalg::LinalgOp >	pack (RewriterBase &rewriter, linalg::LinalgOp linalgOp, ArrayRef< OpFoldResult > packedSizes)
	Implement packing of a single LinalgOp by `packedSizes`. More...

FailureOr< PackTransposeResult >	packTranspose (RewriterBase &rewriter, tensor::PackOp packOp, linalg::LinalgOp linalgOp, tensor::UnPackOp maybeUnPackOp, ArrayRef< int64_t > outerPerm, ArrayRef< int64_t > innerPerm)
	Transpose a single PackOp -> LinalgOp -> UnPackOp chain and return the transposed PackOp -> LinalgOp -> UnPackOp chain after replacements. More...

bool	allIndexingsAreProjectedPermutation (LinalgOp op)
	Check if all indexing maps are projected permutations. More...

bool	hasOnlyScalarElementwiseOp (Region &r)
	Detect whether `r` has only ConstantOp, ElementwiseMappable and YieldOp. More...

bool	isElementwise (LinalgOp op)
	Check if a LinalgOp is an element-wise operation. More...

bool	isParallelIterator (utils::IteratorType iteratorType)
	Check if iterator type has "parallel" semantics. More...

bool	isReductionIterator (utils::IteratorType iteratorType)
	Check if iterator type has "reduction" semantics. More...

Value	createOrFoldDimOp (OpBuilder &b, Location loc, Value source, int64_t dim)
	Helper function that creates a memref::DimOp or tensor::DimOp depending on the type of `source`. More...

OpFoldResult	createFoldedDimOp (OpBuilder &b, Location loc, Value source, int64_t dim)

SmallVector< Value, 4 >	getDynOperands (Location loc, Value val, OpBuilder &b)
	Given an operation, retrieves the value of each dynamic dimension through constructing the necessary DimOp operators. More...

void	getUpperBoundForIndex (Value value, AffineMap &boundMap, SmallVectorImpl< Value > &boundOperands, bool constantRequired=false)
	Computes an upper bound for the result `value` of an index computation. More...

FailureOr< int64_t >	getConstantUpperBoundForIndex (Value value)
	Returns a constant upper bound for the result `value` of an index computation. More...

Value	makeComposedPadHighOp (OpBuilder &b, Location loc, RankedTensorType type, Value source, Value pad, bool nofold)
	Create a tensor::PadOp that pads `source` to the size of the statically sized `type` whose static sizes are assumed to be greater than the dynamic `source` size. More...

GenericOp	makeTransposeOp (OpBuilder &b, Location loc, Value inputTensor, Value outputTensor, ArrayRef< int64_t > transposeVector)
	Returns a GenericOp that tansposes `inputTensor` into `outputTensor` using `transposeVector` to permute the `inputTensor` dimensions. More...

GenericOp	makeMemRefCopyOp (OpBuilder &b, Location loc, Value from, Value to)
	Returns GenericOp that copies an n-D memref. More...

std::optional< SmallVector< ReassociationIndices > >	getReassociationMapForFoldingUnitDims (ArrayRef< OpFoldResult > mixedSizes)
	Get the reassociation maps to fold the result of a extract_slice (or source of a insert_slice) operation with given offsets, and sizes to its rank-reduced version. More...

std::optional< Attribute >	getNeutralElement (Operation *op)
	Return the identity numeric value associated to the give op. More...

bool	isProducerLastWriteOfView (const LinalgDependenceGraph &graph, LinalgOp consumer, Value consumedView, LinalgOp producer)
	Checks whether the specific `producer` is the last write to exactly the whole `consumedView`. More...

bool	isFusableInto (const LinalgDependenceGraph &graph, LinalgOp consumer, Value consumedView, LinalgOp producer)
	Checks whether fusing the specific `producer` of the `consumedView` is feasible. More...

SmallVector< OpFoldResult >	computeTileOffsets (OpBuilder &b, Location loc, ArrayRef< OpFoldResult > ivs, ArrayRef< OpFoldResult > tileSizes)
	Computes tile offsets, given a list of loop `ivs` and `tileSizes`. More...

SmallVector< OpFoldResult >	computeTileSizes (OpBuilder &b, Location loc, ArrayRef< OpFoldResult > tileSizes, ArrayRef< OpFoldResult > sizeBounds)
	Computes tile sizes, given a list of `tileSizes` and dimension sizes (`sizeBounds`). More...

SmallVector< Type >	getTensorOutputTypes (LinalgOp op, ValueRange operands)
	Returns the list of tensor output types produced when the given structured operation `op` is applied to the given `operands`. More...

SmallVector< Value >	insertSlicesBack (OpBuilder &builder, Location loc, LinalgOp op, ValueRange operands, ValueRange results)
	Creates `insert_slice` ops that insert `results` back into larger tensors they were originally extracted from with `extract_slice` before being passed as `operands` to the given structured operation `op` or its clone. More...

SliceParameters	computeSliceParameters (OpBuilder &builder, Location loc, Value valueToTile, ArrayRef< OpFoldResult > tileSizes, AffineMap map, ArrayRef< OpFoldResult > lbs, ArrayRef< OpFoldResult > ubs, ArrayRef< OpFoldResult > subShapeSizes, bool omitPartialTileCheck)
	Computes SliceParameters for a single `valueToTile` assuming that its user is being tiled with the given loop bounds `lbs` and `ubs` and the tile sizes `tileSizes`. More...

SmallVector< std::optional< SliceParameters > >	computeAllSliceParameters (OpBuilder &builder, Location loc, LinalgOp linalgOp, ValueRange valuesToTile, ArrayRef< OpFoldResult > ivs, ArrayRef< OpFoldResult > tileSizes, ArrayRef< OpFoldResult > sizeBounds, bool omitPartialTileCheck)
	Computes SliceParamaters for all `valuesToTile` of the given `linalgOp`, assuming `linalgOp` is being fused into a loop nest. More...

Value	makeTiledShape (OpBuilder &builder, Location loc, Value valueToTile, ArrayRef< OpFoldResult > tileSizes, AffineMap map, ArrayRef< OpFoldResult > lbs, ArrayRef< OpFoldResult > ubs, ArrayRef< OpFoldResult > subShapeSizes, bool omitPartialTileCheck)
	Creates an extract_slice/subview op for a single `valueToTile` with `builder`. More...

SmallVector< Value >	makeTiledShapes (OpBuilder &builder, Location loc, LinalgOp linalgOp, ValueRange valuesToTile, ArrayRef< OpFoldResult > ivs, ArrayRef< OpFoldResult > tileSizes, ArrayRef< OpFoldResult > sizeBounds, bool omitPartialTileCheck)
	Creates extract_slice/subview ops for all `valuesToTile` of the given `linalgOp` with `builder`, assuming `linalgOp` is being fused into a loop nest for tiling with the given induction variables `ivs` and tile sizes `tileSizes`. More...

void	offsetIndices (OpBuilder &b, LinalgOp linalgOp, ArrayRef< OpFoldResult > offests)
	Add the specified offsets to any `linalg.index` ops contained in the given `linalgOp`. More...

void	offsetIndices (RewriterBase &b, LinalgOp linalgOp, ArrayRef< OpFoldResult > offests)

FusableOpDependencesTy	findAllFusableDependences (ArrayRef< LinalgOp > ops, const LinalgDependenceGraph &dependenceGraph)

FailureOr< FusionInfo >	fuseProducerOfBuffer (OpBuilder &b, OpOperand &consumerOpOperand, const LinalgDependenceGraph &graph)
	Fuses producer into consumer if the producer is structurally feasible and the fusion would not violate dependencies. More...

FailureOr< FusionInfo >	fuseProducerOfTensor (OpBuilder &b, OpOperand &consumerOpOperand)
	Tensor counterpart of `fuseProducerOfBuffer`. More...

FailureOr< FusionInfo >	fuseProducerOfTensor (OpBuilder &b, OpResult producerOpResult, OpOperand &consumerOpOperand)
	Tensor counterpart of `fuseProducerOfBuffer`. More...

void	updateBoundsForCyclicDistribution (OpBuilder &builder, Location loc, Value procId, Value nprocs, Value &lb, Value &ub, Value &step)
	Update the `lb`, `ub` and `step` to get per processor `lb`, `ub` and `step`. More...

template<typename OpTy >
SmallVector< NamedAttribute >	getPrunedAttributeList (OpTy op)
	Returns an attribute list that excludes pre-defined attributes. More...

static void	generateParallelLoopNest (OpBuilder &b, Location loc, ValueRange lbs, ValueRange ubs, ValueRange steps, ArrayRef< utils::IteratorType > iteratorTypes, ArrayRef< linalg::ProcInfo > procInfo, function_ref< void(OpBuilder &, Location, ValueRange)> bodyBuilderFn, SmallVectorImpl< Value > &ivStorage)
	Generates a loop nest consisting of scf.parallel and scf.for, depending on the `iteratorTypes.` More...

static Value	materializeTiledShape (OpBuilder &builder, Location loc, Value valueToTile, const SliceParameters &sliceParams)

Typedef Documentation

◆ AllocBufferCallbackFn

using mlir::linalg::AllocBufferCallbackFn = typedef std::function<std::optional<Value>( OpBuilder &b, memref::SubViewOp subView, ArrayRef<Value> boundingSubViewSize, DataLayout &layout)>

Callback function type used to perform the allocation for the promoted subView.

In boundingSubViewsize a best attempt is made to find the smallest constant value for the size of the buffer needed for each dimension. If that is not possible, contains the dynamic size of the subview. The call back should return the buffer to use.

Definition at line 255 of file Transforms.h.

◆ ControlFusionFn

using mlir::linalg::ControlFusionFn = typedef std::function<bool(OpOperand *fusedOperand)>

Function type which is used to control when to stop fusion.

It is expected that OpOperand is not modified in the callback. The OpOperand is not marked as const to allow callers to use non-const methods.

Definition at line 79 of file Transforms.h.

◆ ControlSplitReductionFn

using mlir::linalg::ControlSplitReductionFn = typedef std::function<SplitReductionOptions(LinalgOp op)>

Function signature to control reduction splitting.

This returns SplitReductionOptions.

Definition at line 1027 of file Transforms.h.

◆ CopyCallbackFn

using mlir::linalg::CopyCallbackFn = typedef std::function<LogicalResult(OpBuilder &b, Value src, Value dst)>

Callback function type used to insert copy from original subview to subview of the promoted region for the read operands/subview of promoted region to original subview for the results.

The copy has to happen from src to dst.

Definition at line 268 of file Transforms.h.

◆ DeallocBufferCallbackFn

using mlir::linalg::DeallocBufferCallbackFn = typedef std::function<LogicalResult(OpBuilder &b, Value buffer)>

Callback function type used to deallocate the buffers used to hold the promoted subview.

Definition at line 261 of file Transforms.h.

◆ FusableOpDependencesTy

using mlir::linalg::FusableOpDependencesTy = typedef llvm::MapVector< Operation *, SmallVector<LinalgDependenceGraph::LinalgDependenceGraphElem, 1> >

Definition at line 271 of file Utils.h.

◆ GetCollapsableDimensionsFn

using mlir::linalg::GetCollapsableDimensionsFn = typedef std::function<SmallVector<ReassociationIndices>(linalg::GenericOp)>

Function type to control generic op dimension collapsing.

It is expected to return an array of ReassociationIndices representing dimensions that should be merged.

Definition at line 103 of file Transforms.h.

◆ LinalgLoops

using mlir::linalg::LinalgLoops = typedef SmallVector<Operation *, 4>

Definition at line 45 of file Transforms.h.

◆ LoopIndexToRangeIndexMap

using mlir::linalg::LoopIndexToRangeIndexMap = typedef DenseMap<int, int>

Creates a number of ranges equal to the number of non-zero in tileSizes.

One for each loop of the LinalgOp that is tiled. The tileSizes argument has one entry per surrounding loop. It uses zero as the convention that a particular loop is not tiled. This convention simplifies implementations by avoiding affine map manipulations. The returned ranges correspond to the loop ranges, in the proper order, that are tiled and for which new loops will be created. Also the function returns a map from loop indices of the LinalgOp to the corresponding non-empty range indices of newly created loops.

Definition at line 418 of file Transforms.h.

◆ OptimizeCopyFn

using mlir::linalg::OptimizeCopyFn = typedef std::function<LogicalResult(PatternRewriter &, tensor::PadOp, Value)>

Definition at line 873 of file Transforms.h.

◆ ProcInfoCallBackFn

using mlir::linalg::ProcInfoCallBackFn = typedef std::function<SmallVector<ProcInfo>( OpBuilder &b, Location loc, ArrayRef<Range> parallelLoopRanges)>

Definition at line 366 of file Utils.h.

◆ TileSizeComputationFunction

using mlir::linalg::TileSizeComputationFunction = typedef std::function<SmallVector<Value, 4>(OpBuilder &, Operation *)>

Definition at line 406 of file Transforms.h.

Enumeration Type Documentation

◆ DistributionMethod

enum mlir::linalg::DistributionMethod

strong

Scheme used to distribute loops to processors.

Enumerator
Cyclic	Cyclic distribution where no assumption is made about the dynamic relationship between number of processors and number of iterations of the distributed loop. Distributes the following loop scf.parallel (iv) = (lb) to (ub) step (step) to scf.parallel(iv)= (lb + procId * step) to (ub) step (step * nprocs)
CyclicNumProcsGeNumIters	Cyclic distribution where the number of processors can be assumed to be more than or equal to the number of iterations of the distributed loop. In such cases, a simple in-bounds check is enough (instead of materializing a loop). Distributes the following loop scf.parallel (iv) = (lb) to (ub) step (step) to iv = lb + procId * step cond = arith.cmpi "slt", iv, ub scf.if cond { ... }
CyclicNumProcsEqNumIters	Cyclic distribution where the number of processors can be assumed to be equal to the number of iterations of the distributed loop. In such cases, no bounds check is needed. Distributes the following loop scf.parallel (iv) = (lb) to (ub) step (step) to iv = lb + procId * step
None	No Distribution.

Definition at line 316 of file Utils.h.

◆ LinalgTilingLoopType

enum mlir::linalg::LinalgTilingLoopType

strong

The type of loops to be generated during tiling.

Enumerator
Loops
AffineLoops
ParallelLoops

Definition at line 150 of file Utils.h.

Function Documentation

◆ allIndexingsAreProjectedPermutation()

bool mlir::linalg::allIndexingsAreProjectedPermutation ( LinalgOp op )

Check if all indexing maps are projected permutations.

Definition at line 156 of file Utils.cpp.

Referenced by isElementwise(), and vectorizeLinalgOpPrecondition().

◆ areElementwiseOpsFusable()

bool mlir::linalg::areElementwiseOpsFusable ( OpOperand * fusedOperand )

Return true if two linalg.generic operations with producer/consumer relationship through fusedOperand can be fused using elementwise op fusion.

Conditions for elementwise fusion of generic operations.

Definition at line 75 of file ElementwiseOpFusion.cpp.

References mlir::IROperand< DerivedT, IRValueT >::get(), mlir::Value::getDefiningOp(), getIndexingMapOfProducerOperandsInCoordinatesOfFusedOp(), mlir::AffineMap::getNumResults(), mlir::detail::IROperandBase::getOwner(), mlir::Value::getType(), mlir::Type::isa(), and mlir::AffineMap::isPermutation().

Referenced by fuseElementwiseOps().

◆ collapseGenericOpIterationDims()

FailureOr< SmallVector< Value > > mlir::linalg::collapseGenericOpIterationDims	(	GenericOp	genericOp,
		ArrayRef< ReassociationIndices >	foldedIterationDims,
		RewriterBase &	rewriter
	)

Collapses dimensions of linalg.generic operation.

Implementation of fusion with reshape operation by collapsing dimensions.

It also collapses inputs before the op and expands outputs after the op.

Definition at line 1401 of file ElementwiseOpFusion.cpp.

References mlir::Type::cast(), mlir::OpBuilder::create(), mlir::detail::enumerate(), mlir::failed(), mlir::failure(), mlir::Block::front(), generateCollapsedIndexingRegion(), mlir::Block::getArguments(), getCollapsedOpIteratorTypes(), getCollapsedOpOperand(), getOperandReassociation(), mlir::Value::getType(), mlir::getValueOrCreateConstantIndexOp(), mlir::m_ConstantInt(), mlir::matchPattern(), mlir::RewriterBase::mergeBlocks(), mlir::RewriterBase::notifyMatchFailure(), mlir::Range::offset, mlir::OpBuilder::setInsertionPoint(), mlir::Range::size, and mlir::Range::stride.

◆ computeAllSliceParameters()

SmallVector< std::optional< SliceParameters > > mlir::linalg::computeAllSliceParameters	(	OpBuilder &	builder,
		Location	loc,
		LinalgOp	linalgOp,
		ValueRange	valuesToTile,
		ArrayRef< OpFoldResult >	ivs,
		ArrayRef< OpFoldResult >	tileSizes,
		ArrayRef< OpFoldResult >	sizeBounds,
		bool	omitPartialTileCheck
	)

Computes SliceParamaters for all valuesToTile of the given linalgOp, assuming linalgOp is being fused into a loop nest.

Calls computeSliceParameters for every individual value.

Note that a constant zero in tileSizes means no tiling at that implicit loop. The number of non-zero values in tileSizes should be equal to the number of values in ivs.

Some of the valuesToTile won't be affected by tiling. For these values, std::nullopt will be returned.

Definition at line 934 of file Utils.cpp.

References computeSliceParameters(), computeTileOffsets(), computeTileSizes(), mlir::Type::isa(), and isTiled().

Referenced by makeTiledShapes().

◆ computeMultiTileSizes()

FailureOr< MultiSizeSpecification > mlir::linalg::computeMultiTileSizes	(	OpBuilder &	builder,
		LinalgOp	op,
		unsigned	dimension,
		OpFoldResult	targetSize,
		OpFoldResult	divisor,
		bool	emitAssertions = `true`
	)

Emits the IR computing the multi-sized tiling specification with two tile sizes not exceeding targetSize, each divisible by sizeDivisor, such that there exist numbers of tiles with these sizes that fully cover the given iteration space dimension of the structured op.

The computation is as follows:

b = originalTripCount floordiv sizeDivisor t = (targetSize + sizeDivisor - 1) floordiv sizeDivisor d = (b + t - 1) floordiv t s = (b floordiv d) * sizeDivisor v = b % d u = d - v

where the tile sizes are s and s + sizeDivisor, and the numbers of the corresponding tiles are u and v, respectively. Alternatively,

s * u + (s + sizeDivisor) * v == original size, where s mod sizeDivisor = 0.

Expects all values to be positive. In some cases with the target tile size sufficiently close to the dimension shape and non-unit divisor, it is impossible to compute such sizes. If emitAssertion is set, also emit the assertion that size computation succeeded.

Returns the specification consisting of both tile values and the number of tiles of each size.

Definition at line 147 of file Tiling.cpp.

References mlir::ImplicitLocOpBuilder::create(), emitIsPositiveIndexAssertion(), mlir::failure(), mlir::floorDiv(), mlir::AffineExpr::floorDiv(), mlir::Builder::getAffineSymbolExpr(), mlir::ImplicitLocOpBuilder::getLoc(), mlir::Builder::getStringAttr(), mlir::getValueOrCreateConstantIndexOp(), mlir::linalg::detail::MultiSizeSpecificationBase< T >::lowTileSize, mlir::makeComposedAffineApply(), and mlir::makeComposedFoldedMultiResultAffineApply().

◆ computeSliceParameters()

SliceParameters mlir::linalg::computeSliceParameters	(	OpBuilder &	builder,
		Location	loc,
		Value	valueToTile,
		ArrayRef< OpFoldResult >	tileSizes,
		AffineMap	map,
		ArrayRef< OpFoldResult >	lbs,
		ArrayRef< OpFoldResult >	ubs,
		ArrayRef< OpFoldResult >	subShapeSizes,
		bool	omitPartialTileCheck
	)

Computes SliceParameters for a single valueToTile assuming that its user is being tiled with the given loop bounds lbs and ubs and the tile sizes tileSizes.

omitPartialTileCheck controls whether to omit the partial/boundary tile condition check in cases where we statically know that it is unnecessary.

Definition at line 759 of file Utils.cpp.

References mlir::bindDims(), createFoldedDimOp(), mlir::Type::dyn_cast(), mlir::getAffineSymbolExpr(), mlir::getConstantIntValue(), mlir::Builder::getContext(), mlir::Builder::getIndexAttr(), mlir::AffineMap::getSubMap(), mlir::Value::getType(), mlir::AffineMap::inferFromExprList(), isTiled(), mlir::makeComposedFoldedAffineApply(), mlir::makeComposedFoldedAffineMin(), mlir::linalg::SliceParameters::offsets, mlir::linalg::SliceParameters::sizes, and mlir::linalg::SliceParameters::strides.

Referenced by computeAllSliceParameters(), and makeTiledShape().

◆ computeStaticMultiTileSizes()

FailureOr< StaticMultiSizeSpecification > mlir::linalg::computeStaticMultiTileSizes	(	LinalgOp	op,
		unsigned	dimension,
		int64_t	targetSize,
		int64_t	divisor
	)

Definition at line 121 of file Tiling.cpp.

References mlir::failure(), mlir::linalg::detail::MultiSizeSpecificationBase< T >::highTileSize, mlir::linalg::detail::MultiSizeSpecificationBase< T >::highTripCount, mlir::linalg::detail::MultiSizeSpecificationBase< T >::lowTileSize, and mlir::linalg::detail::MultiSizeSpecificationBase< T >::lowTripCount.

◆ computeTileOffsets()

SmallVector< OpFoldResult > mlir::linalg::computeTileOffsets	(	OpBuilder &	b,
		Location	loc,
		ArrayRef< OpFoldResult >	ivs,
		ArrayRef< OpFoldResult >	tileSizes
	)

Computes tile offsets, given a list of loop ivs and tileSizes.

In case a tile size is zero (i.e., no tiling), the corresponding offset is also zero.

Definition at line 866 of file Utils.cpp.

References mlir::Builder::getIndexAttr(), isTiled(), and isZero().

Referenced by computeAllSliceParameters(), and tilePadOp().

◆ computeTileSizes()

SmallVector< OpFoldResult > mlir::linalg::computeTileSizes	(	OpBuilder &	b,
		Location	loc,
		ArrayRef< OpFoldResult >	tileSizes,
		ArrayRef< OpFoldResult >	sizeBounds
	)

Computes tile sizes, given a list of tileSizes and dimension sizes (sizeBounds).

In case a tile size is zero (i.e., no tiling), the corresponding result size is the corresponding value from sizeBounds. Note: The returned tile sizes are closed intervals.

Definition at line 880 of file Utils.cpp.

References mlir::getAffineDimExpr(), mlir::Builder::getContext(), isTiled(), isZero(), and mlir::makeComposedFoldedAffineApply().

Referenced by computeAllSliceParameters(), and tilePadOp().

◆ concat()

SmallVector< AffineExpr, 4 > mlir::linalg::concat	(	ArrayRef< AffineExpr >	a,
		ArrayRef< AffineExpr >	b
	)

Return the vector that is the concatenation of a and b.

Definition at line 1790 of file LinalgOps.cpp.

Referenced by mlir::presburger::Simplex::makeProduct().

◆ createFoldedDimOp()

OpFoldResult mlir::linalg::createFoldedDimOp	(	OpBuilder &	b,
		Location	loc,
		Value	source,
		int64_t	dim
	)

Definition at line 209 of file Utils.cpp.

References mlir::Type::cast(), createOrFoldDimOp(), mlir::Builder::getIndexAttr(), and mlir::Value::getType().

Referenced by fuse().

◆ createOrFoldDimOp()

Value mlir::linalg::createOrFoldDimOp	(	OpBuilder &	b,
		Location	loc,
		Value	source,
		int64_t	dim
	)

Helper function that creates a memref::DimOp or tensor::DimOp depending on the type of source.

Definition at line 201 of file Utils.cpp.

References mlir::OpBuilder::createOrFold(), mlir::Value::getType(), and mlir::Type::isa().

Referenced by concatSizesFromInputs(), createFoldedDimOp(), mlir::sparse_tensor::genDenseTensorOrSparseConstantIterLoop(), mlir::sparse_tensor::LoopEmitter::initializeLoopEmit(), and mlir::sparse_tensor::sizesFromSrc().

◆ extractOrIdentityMap()

AffineMap mlir::linalg::extractOrIdentityMap	(	std::optional< AffineMap >	maybeMap,
		unsigned	rank,
		MLIRContext *	context
	)

Returns maybeMap.get() if maybeMap is set, otherwise returns the symbol-less identity map of rank.

Definition at line 1770 of file LinalgOps.cpp.

References mlir::AffineMap::get(), and mlir::AffineMap::getMultiDimIdentityMap().

◆ findAllFusableDependences()

FusableOpDependencesTy mlir::linalg::findAllFusableDependences	(	ArrayRef< LinalgOp >	ops,
		const LinalgDependenceGraph &	dependenceGraph
	)

◆ fuseElementwiseOps()

FailureOr< Operation * > mlir::linalg::fuseElementwiseOps	(	RewriterBase &	rewriter,
		OpOperand *	fusedOperand
	)

Fuse two linalg.generic operations that have a producer-consumer relationship captured through fusedOperand.

The method expects that areElementwiseOpsFusable returns true for the given fusedOperand.

Find the results of the producer that have uses outside of the consumer.

Definition at line 274 of file ElementwiseOpFusion.cpp.

References areElementwiseOpsFusable(), mlir::Value::cast(), mlir::AffineMap::compose(), mlir::OpBuilder::create(), mlir::detail::enumerate(), mlir::RewriterBase::eraseOp(), generateFusedElementwiseOpRegion(), mlir::IROperand< DerivedT, IRValueT >::get(), mlir::Builder::getAffineMapArrayAttr(), getIndexingMapOfProducerOperandsInCoordinatesOfFusedOp(), mlir::detail::IROperandBase::getOwner(), mlir::inversePermutation(), mlir::Type::isa(), and mlir::RewriterBase::notifyMatchFailure().

◆ fuseProducerOfBuffer()

FailureOr< FusionInfo > mlir::linalg::fuseProducerOfBuffer	(	OpBuilder &	b,
		OpOperand &	consumerOpOperand,
		const LinalgDependenceGraph &	graph
	)

Fuses producer into consumer if the producer is structurally feasible and the fusion would not violate dependencies.

Implements the fusion part of the "tileAndFuse on buffers" transformation and thus requires the consumerOpOperand to be a subview op (generally obtained by applying the tiling transformation).

Definition at line 335 of file Fusion.cpp.

References mlir::failure(), findFusableProducer(), fuse(), mlir::IROperand< DerivedT, IRValueT >::get(), mlir::Value::getDefiningOp(), mlir::detail::IROperandBase::getOwner(), mlir::Value::getParentBlock(), and mlir::OpBuilder::setInsertionPoint().

◆ fuseProducerOfTensor() [1/2]

FailureOr< FusionInfo > mlir::linalg::fuseProducerOfTensor	(	OpBuilder &	b,
		OpOperand &	consumerOpOperand
	)

Tensor counterpart of fuseProducerOfBuffer.

This implements the fusion part of the "tileAndFuse on tensors" transformation and thus requires the consumerOpOperand to be a extract_slice op (generally obtained by applying the tiling transformation).

Definition at line 405 of file Fusion.cpp.

References mlir::failure(), mlir::IROperand< DerivedT, IRValueT >::get(), and getProducerOfTensor().

◆ fuseProducerOfTensor() [2/2]

FailureOr< FusionInfo > mlir::linalg::fuseProducerOfTensor	(	OpBuilder &	b,
		OpResult	producerOpResult,
		OpOperand &	consumerOpOperand
	)

Tensor counterpart of fuseProducerOfBuffer.

This implements the fusion part of the "tileAndFuse on tensors" transformation and thus requires the consumerOpOperand to be a extract_slice op (generally obtained by applying the tiling transformation). Assumes producerOfTensor is a Linalg op that produces consumerOpOperand.

Definition at line 417 of file Fusion.cpp.

References mlir::OpBuilder::create(), mlir::failure(), fuse(), mlir::IROperand< DerivedT, IRValueT >::get(), mlir::Value::getDefiningOp(), mlir::detail::IROperandBase::getOwner(), mlir::OpResult::getOwner(), mlir::Value::getParentBlock(), mlir::OpResult::getResultNumber(), mlir::Value::getType(), mlir::IROperand< DerivedT, IRValueT >::set(), and mlir::OpBuilder::setInsertionPoint().

◆ generalizeNamedOp()

FailureOr< GenericOp > mlir::linalg::generalizeNamedOp	(	RewriterBase &	rewriter,
		LinalgOp	namedOp
	)

Create a GenericOp from the given named operation namedOp and replace namedOp.

Return failure if namedOp is a GenericOp or misses a region builder.

Definition at line 48 of file Generalization.cpp.

References mlir::OpBuilder::create(), mlir::failed(), generalizeNamedOpPrecondition(), mlir::RewriterBase::inlineRegionBefore(), mlir::RewriterBase::notifyMatchFailure(), and mlir::RewriterBase::replaceOp().

Referenced by packGemmGreedily(), and mlir::linalg::LinalgGeneralizationPattern::returningMatchAndRewrite().

◆ generateLibraryCallName()

std::string mlir::linalg::generateLibraryCallName ( Operation * op )

Returns the name mangled library call name to disambiguate between different overloads at the C level.

The name mangling scheme is basic and uses MLIR type names:

form a string which is the concatenation of the linalg op name with all the operand type names, separate by underscores;
drop the linalg. prefix, and the <, >, ? symbols from the type. Assumes op is a LinalgOp.

Examples:

linalg.fill(f, A) : f32, memref<f32> name mangles into linalg_fill_f32_viewf32
linalg.dot A, B, C : (memref<?xf32, stride_specification>, memref<?xf32, stride_specification>, memref<f32>) name mangles into linalg_dot_viewxf32_viewxf32_viewf32
linalg.matmul(...) : memref<?x?xf32, stride_specification>, memref<?x?xf32, stride_specification>, memref<?x?xf32, stride_specification> name mangles into linalg_matmul_viewxxf32_viewxxf32_viewxxf32

Definition at line 1824 of file LinalgOps.cpp.

References appendMangledType(), mlir::failed(), mlir::Operation::getName(), mlir::Operation::getOperandTypes(), and mlir::OperationName::getStringRef().

◆ generateParallelLoopNest()

static void mlir::linalg::generateParallelLoopNest	(	OpBuilder &	b,
		Location	loc,
		ValueRange	lbs,
		ValueRange	ubs,
		ValueRange	steps,
		ArrayRef< utils::IteratorType >	iteratorTypes,
		ArrayRef< linalg::ProcInfo >	procInfo,
		function_ref< void(OpBuilder &, Location, ValueRange)>	bodyBuilderFn,
		SmallVectorImpl< Value > &	ivStorage
	)

static

Generates a loop nest consisting of scf.parallel and scf.for, depending on the iteratorTypes.

Consecutive parallel loops create a single scf.parallel operation; each sequential loop creates a new scf.for operation. The body of the innermost loop is populated by bodyBuilderFn that accepts a range of induction variables for all loops. ivStorage is used to store the partial list of induction variables.

Definition at line 565 of file Utils.cpp.

References mlir::ArithBuilder::_and(), mlir::scf::buildLoopNest(), mlir::OpBuilder::create(), mlir::linalg::ProcInfo::distributionMethod, isParallelIterator(), None, and mlir::ArithBuilder::slt().

Referenced by mlir::linalg::GenerateLoopNest< LoopTy >::doit().

◆ getCombinerOpKind()

std::optional< vector::CombiningKind > mlir::linalg::getCombinerOpKind ( Operation * combinerOp )

Return vector::CombiningKind for the given op.

Definition at line 363 of file Vectorization.cpp.

Referenced by buildMultiDimReduce(), and reductionPreconditions().

◆ getConstantUpperBoundForIndex()

FailureOr< int64_t > mlir::linalg::getConstantUpperBoundForIndex ( Value value )

Returns a constant upper bound for the result value of an index computation.

Calls getUpperBoundForIndex and returns a constant upper bound if the result of boundMap is a constant expression and failure otherwise.

Example:

%0 = affine.min affine.map<(d0) -> (40, d0)> (%d0)

%1 = affine.apply affine.map<(d0) -> (d0 + 2)> (%0)

getConstantUpperBoundForIndex(%1) returns 42 (boundsMap = affine.map<() -> (42)>)

Definition at line 329 of file Utils.cpp.

References mlir::failure(), mlir::AffineMap::getResults(), and getUpperBoundForIndex().

Referenced by padOperandToSmallestStaticBoundingBox(), and promoteSubviewAsNewBuffer().

◆ getDynOperands()

SmallVector< Value, 4 > mlir::linalg::getDynOperands	(	Location	loc,
		Value	val,
		OpBuilder &	b
	)

Given an operation, retrieves the value of each dynamic dimension through constructing the necessary DimOp operators.

Definition at line 219 of file Utils.cpp.

References mlir::Type::cast(), createOrFoldDimOp(), mlir::detail::enumerate(), and mlir::Value::getType().

Referenced by getOrCreateOperandsMatchingResultTypes().

◆ getLinalgTilingCanonicalizationPatterns()

RewritePatternSet mlir::linalg::getLinalgTilingCanonicalizationPatterns ( MLIRContext * ctx )

Canonicalization patterns relevant to apply after tiling patterns.

These are applied automatically by the tiling pass but need to be applied manually when tiling is called programmatically.

Definition at line 964 of file Tiling.cpp.

References populateLinalgTilingCanonicalizationPatterns().

◆ getNeutralElement()

std::optional< Attribute > mlir::linalg::getNeutralElement ( Operation * op )

Return the identity numeric value associated to the give op.

Return std::nullopt if there is no known neutral element.

Definition at line 1064 of file Utils.cpp.

References mlir::Type::dyn_cast(), mlir::Operation::getContext(), mlir::Builder::getFloatAttr(), mlir::Builder::getIntegerAttr(), mlir::Operation::getResult(), mlir::Value::getType(), getZero(), max(), and min().

Referenced by splitReduction(), and splitReductionByScaling().

◆ getPrunedAttributeList()

template<typename OpTy >

SmallVector<NamedAttribute> mlir::linalg::getPrunedAttributeList ( OpTy op )

Returns an attribute list that excludes pre-defined attributes.

Definition at line 494 of file Utils.h.

Referenced by matchAndReplaceDepthwiseConv().

◆ getReassociationMapForFoldingUnitDims()

std::optional< SmallVector< ReassociationIndices > > mlir::linalg::getReassociationMapForFoldingUnitDims ( ArrayRef< OpFoldResult > mixedSizes )

Get the reassociation maps to fold the result of a extract_slice (or source of a insert_slice) operation with given offsets, and sizes to its rank-reduced version.

This is only done for the cases where the size is 1 and offset is 0. Strictly speaking the offset 0 is not required in general, but non-zero offsets are not handled by SPIR-V backend at this point (and potentially cannot be handled).

Definition at line 1042 of file Utils.cpp.

References mlir::detail::enumerate().

◆ getTensorOutputTypes()

SmallVector< Type > mlir::linalg::getTensorOutputTypes	(	LinalgOp	op,
		ValueRange	operands
	)

Returns the list of tensor output types produced when the given structured operation op is applied to the given operands.

Note that operands are not necessarily the actual operands of op.

Definition at line 896 of file Utils.cpp.

Referenced by tileLinalgOpImpl().

◆ getUpperBoundForIndex()

void mlir::linalg::getUpperBoundForIndex	(	Value	value,
		AffineMap &	boundMap,
		SmallVectorImpl< Value > &	boundOperands,
		bool	constantRequired = `false`
	)

Computes an upper bound for the result value of an index computation.

Translates AffineMinOps and AffineApplyOps along the use-def chains of the index computation to affine constraints and projects out intermediate values. The method sets boundMap to an affine map that given boundOperands evaluates to an upper bound for the index computation.

If constantRequired is true, only returns the constant bounds (potentially over-approximating) and fails when not possible.

Example:

%dim0 = dim %tensor, %c0
%dim1 = dim %tensor, %c1
%0 = affine.min affine.map<(d0) -> (40, d0)> (%dim0)
%1 = affine.apply affine.map<(d0, d1) -> (d0 + d1)> (%0, %dim1)

getUpperBoundForIndex(%1, boundMap, boundOperands) set the output parameters to:

boundMap = affine.map<(d0) -> (d0 + 40)>
boundOperands = [dim1]

Definition at line 229 of file Utils.cpp.

References mlir::FlatAffineValueConstraints::addBound(), mlir::FlatAffineValueConstraints::appendDimVar(), mlir::canonicalizeMapAndOperands(), mlir::FlatAffineValueConstraints::computeAlignedMap(), mlir::FlatAffineValueConstraints::containsVar(), mlir::presburger::IntegerRelation::EQ, mlir::failed(), mlir::FlatAffineValueConstraints::findVar(), mlir::FlatAffineValueConstraints::getAllValues(), mlir::getBackwardSlice(), mlir::presburger::IntegerRelation::getConstantBound64(), mlir::AffineMap::getConstantMap(), mlir::Value::getContext(), mlir::Value::getDefiningOp(), mlir::AffineMap::getMultiDimIdentityMap(), mlir::presburger::IntegerRelation::getNumDimVars(), mlir::Operation::getOperands(), mlir::Operation::getResults(), mlir::FlatAffineValueConstraints::getSliceBounds(), mlir::Value::getType(), mlir::Type::isIndex(), and mlir::presburger::IntegerRelation::UB.

Referenced by getConstantUpperBoundForIndex(), and HoistingAnalysis::getPackedTensorSizes().

◆ hasOnlyScalarElementwiseOp()

bool mlir::linalg::hasOnlyScalarElementwiseOp ( Region & r )

Detect whether r has only ConstantOp, ElementwiseMappable and YieldOp.

Definition at line 162 of file Utils.cpp.

References mlir::Region::front(), and mlir::OpTrait::hasElementwiseMappableTraits().

Referenced by isElementwise().

◆ hoistPaddingOnTensors()

FailureOr< Value > mlir::linalg::hoistPaddingOnTensors	(	tensor::PadOp	opToHoist,
		int	numLoops,
		ArrayRef< int64_t >	transposeVector,
		tensor::PadOp &	hoistedOp,
		SmallVectorImpl< GenericOp > &	transposeOps
	)

Mechanically hoist padding operations on tensors by numLoops into a new, generally larger tensor.

This achieves packing of multiple padding ops into a larger tensor. On success, opToHoist is replaced by the cloned version in the packing loop so the caller can continue reasoning about the padding operation. If transposeVector is non-empty, hoist padding introduces a GenericOp to transpose the padded tensor before inserting it into the packed tensor. A transposeVector can change the storage order of the padded tensor but does not change the order of the pack or compute loops.

Example in pseudo-mlir:

If hoistPaddingOnTensors is called with nLoops = 2 on the following IR.

scf.for (%i, %j, %k)
  %st0 = tensor.extract_slice f(%i, %k) : ... to tensor<?x?xf32>
  %0 = tensor.pad %st0 low[0, 0] high[...] {
  ^bb0( ... ):
    linalg.yield %pad
  } : tensor<?x?xf32> to tensor<4x8xf32>
  compute(%0)

IR resembling the following is produced:

scf.for (%i) {
  %packed_init = tensor.empty range(%j) : tensor<?x4x8xf32>
  %packed = scf.for (%k) iter_args(%p : %packed_init) {
    %st0 = tensor.extract_slice f(%i, %k) : ... to tensor<?x?xf32>
    %0 = tensor.pad %st0 low[0, 0] high[...] {
    ^bb0( ... ):
      linalg.yield %pad
    } : tensor<?x?xf32> to tensor<4x8xf32>
    %1 = tensor.insert_slice %0 ...
        : tensor<4x8xf32> to tensor<?x4x8xf32>
    scf.yield %1: tensor<?x4x8xf32>
  } -> tensor<?x4x8xf32>
  scf.for (%j, %k) {
    %st0 = tensor.extract_slice %packed [%k, 0, 0][1, 4, 8][1, 1, 1] :
             tensor<?x4x8xf32> to tensor<4x8xf32>
    compute(%st0)
  }
}

Definition at line 397 of file HoistPadding.cpp.

References HoistingAnalysis::backwardSlice, buildLoopIterationCount(), mlir::OpBuilder::clone(), computeTransposedType(), mlir::ImplicitLocOpBuilder::create(), DBGS, mlir::failed(), mlir::failure(), mlir::scf::getForInductionVarOwner(), mlir::Builder::getIndexAttr(), HoistingAnalysis::getPackedTensorSizes(), HoistingAnalysis::isValid(), mlir::IRMapping::lookup(), mlir::IRMapping::lookupOrDefault(), makeTransposeOp(), mlir::IRMapping::map(), HoistingAnalysis::outermostEnclosingForOp, HoistingAnalysis::packingLoops, mlir::OpBuilder::setInsertionPoint(), mlir::OpBuilder::setInsertionPointToEnd(), and mlir::OpBuilder::setInsertionPointToStart().

Referenced by mlir::linalg::LinalgPaddingPattern::returningMatchAndRewrite().

◆ hoistRedundantVectorTransfers()

void mlir::linalg::hoistRedundantVectorTransfers ( func::FuncOp func )

Hoist vector.transfer_read/vector.transfer_write on buffers pairs out of immediately enclosing scf::ForOp iteratively, if the following conditions are true:

The two ops access the same memref with the same indices.
All operands are invariant under the enclosing scf::ForOp.
No uses of the memref either dominate the transfer_read or are dominated by the transfer_write (i.e. no aliasing between the write and the read across the loop) To improve hoisting opportunities, call the moveLoopInvariantCode helper function on the candidate loop above which to hoist. Hoisting the transfers results in scf::ForOp yielding the value that originally transited through memory.

Definition at line 416 of file Hoisting.cpp.

References mlir::WalkResult::advance(), DBGS, mlir::getForwardSlice(), mlir::WalkResult::interrupt(), mlir::vector::isDisjointTransferSet(), mlir::moveLoopInvariantCode(), mlir::DominanceInfo::properlyDominates(), mlir::replaceForOpWithNewYields(), and mlir::replaceLoopWithNewYields().

◆ hoistRedundantVectorTransfersOnTensor()

void mlir::linalg::hoistRedundantVectorTransfersOnTensor ( func::FuncOp func )

Same behavior as hoistRedundantVectorTransfers but works on tensors instead of buffers.

Definition at line 364 of file Hoisting.cpp.

References mlir::WalkResult::advance(), mlir::applyPatternsAndFoldGreedily(), DBGS, mlir::detail::enumerate(), findMatchingTransferRead(), mlir::RewritePatternSet::getContext(), hoistReadWrite(), mlir::WalkResult::interrupt(), and tensorChunkAccessedByUnknownOp().

◆ insertSlicesBack()

SmallVector< Value > mlir::linalg::insertSlicesBack	(	OpBuilder &	builder,
		Location	loc,
		LinalgOp	op,
		ValueRange	operands,
		ValueRange	results
	)

Creates insert_slice ops that insert results back into larger tensors they were originally extracted from with extract_slice before being passed as operands to the given structured operation op or its clone.

Note that operands are not necessarily the actual operands of op, the operation serves only as metadata container for operand types and positions.

Definition at line 905 of file Utils.cpp.

References mlir::OpBuilder::create(), and mlir::Value::getDefiningOp().

Referenced by tileLinalgOpImpl().

◆ interchangeGenericOp()

FailureOr< GenericOp > mlir::linalg::interchangeGenericOp	(	RewriterBase &	rewriter,
		GenericOp	genericOp,
		ArrayRef< unsigned >	interchangeVector
	)

Interchange the iterator_types and iterator_maps dimensions and adapts the index accesses of op.

This is an in-place transformation controlled by interchangeVector. An empty vector is interpreted as the identity permutation and the transformation returns early.

E.g. the permutation (i,j,k) -> (j,k,i) is expressed with interchangeVector = [1,2,0]. All values in interchangeVector must be integers, in the range 0..op.rank without duplications (i.e. [1,1,2] is an invalid permutation).

Return failure if the permutation is not valid.

Definition at line 51 of file Interchange.cpp.

References mlir::applyPermutationToVector(), mlir::AffineMap::compose(), mlir::failed(), mlir::RewriterBase::finalizeRootUpdate(), mlir::Builder::getAffineMapArrayAttr(), mlir::Builder::getArrayAttr(), mlir::AffineMap::getPermutationMap(), mlir::AffineMap::getSubMap(), interchangeGenericOpPrecondition(), mlir::inversePermutation(), mlir::AffineMap::isEmpty(), mlir::RewriterBase::notifyMatchFailure(), mlir::RewriterBase::replaceOpWithNewOp(), mlir::OpBuilder::setInsertionPoint(), and mlir::RewriterBase::startRootUpdate().

Referenced by packGemmGreedily().

◆ isaContractionOpInterface()

bool mlir::linalg::isaContractionOpInterface ( LinalgOp linalgOp )

Checks whether linalgOp conforms to ContractionOpInterface.

Definition at line 145 of file LinalgInterfaces.cpp.

References isContractionInterfaceImpl(), and Success.

◆ isElementwise()

bool mlir::linalg::isElementwise ( LinalgOp op )

Check if a LinalgOp is an element-wise operation.

Definition at line 176 of file Utils.cpp.

References allIndexingsAreProjectedPermutation(), and hasOnlyScalarElementwiseOp().

Referenced by vectorizeLinalgOpPrecondition().

◆ isFusableInto()

bool mlir::linalg::isFusableInto	(	const LinalgDependenceGraph &	graph,
		LinalgOp	consumer,
		Value	consumedView,
		LinalgOp	producer
	)

Checks whether fusing the specific producer of the consumedView is feasible.

This checks producer is the last write of consumedView and that no interleaved dependence would be violated (RAW, WAR or WAW).

Definition at line 254 of file Fusion.cpp.

References mlir::linalg::LinalgDependenceGraph::findCoveringDependences(), and isProducerLastWriteOfView().

Referenced by findFusableProducer().

◆ isParallelIterator()

bool mlir::linalg::isParallelIterator ( utils::IteratorType iteratorType )

Check if iterator type has "parallel" semantics.

Definition at line 191 of file Utils.cpp.

Referenced by generateParallelLoopNest(), getCollapsableIterationSpaceDims(), isFusableWithReshapeByDimExpansion(), and tileLinalgOpImpl().

◆ isProducerLastWriteOfView()

bool mlir::linalg::isProducerLastWriteOfView	(	const LinalgDependenceGraph &	graph,
		LinalgOp	consumer,
		Value	consumedView,
		LinalgOp	producer
	)

Checks whether the specific producer is the last write to exactly the whole consumedView.

This checks structural dominance, that the dependence is a RAW without any interleaved write to any piece of consumedView.

Definition at line 230 of file Fusion.cpp.

References mlir::linalg::LinalgDependenceGraph::findCoveringWrites(), and isStructurallyFusableProducer().

Referenced by isFusableInto().

◆ isReductionIterator()

bool mlir::linalg::isReductionIterator ( utils::IteratorType iteratorType )

Check if iterator type has "reduction" semantics.

Definition at line 195 of file Utils.cpp.

Referenced by getCollapsableIterationSpaceDims(), getDimsToReduce(), isAdmissibleTensorExp(), reductionPreconditions(), and topSortOptimal().

◆ linalgOpToAffineLoops()

FailureOr< LinalgLoops > mlir::linalg::linalgOpToAffineLoops	(	PatternRewriter &	rewriter,
		LinalgOp	linalgOp
	)

Emit a loop nest of affine.for with the proper body for linalgOp.

Emits a loop nest of affine.for with the proper body for linalgOp.

Definition at line 368 of file Loops.cpp.

◆ linalgOpToLoops()

FailureOr< LinalgLoops > mlir::linalg::linalgOpToLoops	(	PatternRewriter &	rewriter,
		LinalgOp	linalgOp
	)

Emit a loop nest of scf.for with the proper body for linalgOp.

Emits a loop nest of scf.for with the proper body for linalgOp.

Definition at line 374 of file Loops.cpp.

◆ linalgOpToParallelLoops()

FailureOr< LinalgLoops > mlir::linalg::linalgOpToParallelLoops	(	PatternRewriter &	rewriter,
		LinalgOp	linalgOp
	)

Emit a loop nest of scf.parallel with the proper body for linalgOp.

Emits a loop nest of scf.parallel with the proper body for linalgOp.

Definition at line 381 of file Loops.cpp.

◆ makeAffineDimExprs()

SmallVector< AffineExpr, 4 > mlir::linalg::makeAffineDimExprs	(	unsigned	num,
		unsigned &	startIdx,
		MLIRContext *	context
	)

Returns num AffineDimExpr dimensions at positions [startIdx, startIdx + num) and increments startIdx to startIdx + num.

Definition at line 1781 of file LinalgOps.cpp.

References mlir::getAffineDimExpr().

◆ makeComposedPadHighOp()

Value mlir::linalg::makeComposedPadHighOp	(	OpBuilder &	b,
		Location	loc,
		RankedTensorType	type,
		Value	source,
		Value	pad,
		bool	nofold
	)

Create a tensor::PadOp that pads source to the size of the statically sized type whose static sizes are assumed to be greater than the dynamic source size.

The padding introduces trailing pad values until the target size is met. If source is defined by one or more LinalgOps that have been padded with the same value and sizes, return their padded result instead of creating a tensor::PadOp.

Example:

%0 = tensor.extract_slice %arg0 [%iv0, %iv1] [%sz0, %sz1]
%1 = tensor.pad %0 low[0, 0] high[...] { tensor.yield %cst }
%2 = linalg.matmul ins(...) outs(%1)
%3 = tensor.extract_slice %2 [0, 0] [%sz0, %sz1]

makeComposedPadHighOp(source=%3, pad=cst) returns %2 makeComposedPadHighOp(source=%3, pad=other_cst) returns %4

%4 = tensor.pad %3 low[0, 0] high[...] { tensor.yield %other_cst }

Definition at line 348 of file Utils.cpp.

References mlir::Value::cast(), mlir::tensor::createPadHighOp(), mlir::Value::getDefiningOp(), mlir::OpResult::getResultNumber(), mlir::m_Constant(), and mlir::matchPattern().

Referenced by padOperandToSmallestStaticBoundingBox().

◆ makeMemRefCopyOp()

GenericOp mlir::linalg::makeMemRefCopyOp	(	OpBuilder &	b,
		Location	loc,
		Value	from,
		Value	to
	)

Returns GenericOp that copies an n-D memref.

Unlike the current implementation of memref::CopyOp, this op can further tile, lower to loops or vectorize.

Definition at line 445 of file Utils.cpp.

References mlir::Type::cast(), mlir::OpBuilder::create(), mlir::Builder::getContext(), mlir::AffineMap::getMultiDimIdentityMap(), and mlir::Value::getType().

◆ makeTiledLoopRanges()

std::tuple< SmallVector< Range, 4 >, LoopIndexToRangeIndexMap > mlir::linalg::makeTiledLoopRanges	(	RewriterBase &	b,
		Location	loc,
		AffineMap	map,
		ArrayRef< OpFoldResult >	allShapeSizes,
		ArrayRef< OpFoldResult >	allTileSizes
	)

Definition at line 60 of file Tiling.cpp.

References mlir::Builder::getIndexAttr(), mlir::AffineMap::getNumResults(), isZero(), and mlir::makeComposedFoldedMultiResultAffineApply().

Referenced by tileLinalgOpImpl().

◆ makeTiledShape()

Value mlir::linalg::makeTiledShape	(	OpBuilder &	builder,
		Location	loc,
		Value	valueToTile,
		ArrayRef< OpFoldResult >	tileSizes,
		AffineMap	map,
		ArrayRef< OpFoldResult >	lbs,
		ArrayRef< OpFoldResult >	ubs,
		ArrayRef< OpFoldResult >	subShapeSizes,
		bool	omitPartialTileCheck
	)

Creates an extract_slice/subview op for a single valueToTile with builder.

This new operation extracts a tile of valueToTile, starting at offsets lbs and with sizes subShapeSizes. omitPartialTileCheck controls whether to omit the partial/boundary tile condition check in cases where we statically know that it is unnecessary.

Definition at line 747 of file Utils.cpp.

References computeSliceParameters(), and materializeTiledShape().

Referenced by tilePadOp().

◆ makeTiledShapes()

SmallVector< Value > mlir::linalg::makeTiledShapes	(	OpBuilder &	builder,
		Location	loc,
		LinalgOp	linalgOp,
		ValueRange	valuesToTile,
		ArrayRef< OpFoldResult >	ivs,
		ArrayRef< OpFoldResult >	tileSizes,
		ArrayRef< OpFoldResult >	sizeBounds,
		bool	omitPartialTileCheck
	)

Creates extract_slice/subview ops for all valuesToTile of the given linalgOp with builder, assuming linalgOp is being fused into a loop nest for tiling with the given induction variables ivs and tile sizes tileSizes.

sizeBounds are the iteration space bounds for all the implicit loops in linalgOp. omitPartialTileCheck controls whether to omit the partial/boundary tile condition check in cases where we statically know that it is unnecessary.

Note that a constant zero in tileSizes means no tiling at that implicit loop. The number of non-zero values in tileSizes should be equal to the number of values in ivs.

Definition at line 985 of file Utils.cpp.

References computeAllSliceParameters(), and materializeTiledShape().

Referenced by fuse(), getTiledProducer(), and tileLinalgOpImpl().

◆ makeTransposeOp()

GenericOp mlir::linalg::makeTransposeOp	(	OpBuilder &	b,
		Location	loc,
		Value	inputTensor,
		Value	outputTensor,
		ArrayRef< int64_t >	transposeVector
	)

Returns a GenericOp that tansposes inputTensor into outputTensor using transposeVector to permute the inputTensor dimensions.

Definition at line 408 of file Utils.cpp.

References mlir::Type::cast(), mlir::OpBuilder::create(), mlir::Builder::getContext(), mlir::AffineMap::getMultiDimIdentityMap(), mlir::AffineMap::getPermutationMap(), mlir::Value::getType(), mlir::inversePermutation(), mlir::isPermutationVector(), mlir::Region::push_back(), and mlir::OpBuilder::setInsertionPointToEnd().

Referenced by hoistPaddingOnTensors().

◆ materializeTiledShape()

static Value mlir::linalg::materializeTiledShape	(	OpBuilder &	builder,
		Location	loc,
		Value	valueToTile,
		const SliceParameters &	sliceParams
	)

static

Definition at line 726 of file Utils.cpp.

References mlir::OpBuilder::create(), mlir::Type::dyn_cast(), mlir::Value::getType(), mlir::linalg::SliceParameters::offsets, mlir::linalg::SliceParameters::sizes, and mlir::linalg::SliceParameters::strides.

Referenced by makeTiledShape(), and makeTiledShapes().

◆ offsetIndices() [1/2]

void mlir::linalg::offsetIndices	(	OpBuilder &	b,
		LinalgOp	linalgOp,
		ArrayRef< OpFoldResult >	offests
	)

Add the specified offsets to any linalg.index ops contained in the given linalgOp.

The offsets are provided in the same order as iteration space dimensions. Null offests are assumed to be zero.

Definition at line 1006 of file Utils.cpp.

Referenced by fuse(), getTiledProducer(), and transformIndexOps().

◆ offsetIndices() [2/2]

void mlir::linalg::offsetIndices	(	RewriterBase &	b,
		LinalgOp	linalgOp,
		ArrayRef< OpFoldResult >	offests
	)

Definition at line 1012 of file Utils.cpp.

References mlir::bindDims(), mlir::Builder::getContext(), mlir::Value::getDefiningOp(), mlir::detail::IROperandBase::getOwner(), mlir::getValueOrCreateConstantIndexOp(), mlir::makeComposedFoldedAffineApply(), mlir::RewriterBase::replaceOpWithIf(), and mlir::OpBuilder::setInsertionPointAfter().

◆ pack()

FailureOr< linalg::LinalgOp > mlir::linalg::pack	(	RewriterBase &	rewriter,
		linalg::LinalgOp	linalgOp,
		ArrayRef< OpFoldResult >	packedSizes
	)

Implement packing of a single LinalgOp by packedSizes.

Implement packing of a single LinalgOp by performing packing by packedSizes.

There must be one packedSizes entry per linalgOp iterator. Return the packed Linalg op on success, failure otherwise.

Definition at line 1056 of file Transforms.cpp.

References mlir::OpBuilder::create(), DBGS, DBGSNL, mlir::failed(), mlir::failure(), mlir::getConstantIntValue(), mlir::getElementTypeOrSelf(), mlir::Operation::getRegion(), mlir::Value::getType(), mlir::ValueRange::getTypes(), mlir::Builder::getZeroAttr(), mlir::RewriterBase::notifyMatchFailure(), packLinalgMetadataOnce(), mlir::RewriterBase::replaceOp(), and mlir::Region::takeBody().

Referenced by packGemmGreedily().

◆ packTranspose()

FailureOr< PackTransposeResult > mlir::linalg::packTranspose	(	RewriterBase &	rewriter,
		tensor::PackOp	packOp,
		linalg::LinalgOp	linalgOp,
		tensor::UnPackOp	maybeUnPackOp,
		ArrayRef< int64_t >	outerPerm,
		ArrayRef< int64_t >	innerPerm
	)

Transpose a single PackOp -> LinalgOp -> UnPackOp chain and return the transposed PackOp -> LinalgOp -> UnPackOp chain after replacements.

Return failure if either:

the packOp does not have the linalgOp as its unique use.
the maybeUnPackOp, if specified must be a consumer of the result tied to the unique packOp use.
outerPerm (resp. innerPerm) must be valid permutations of packOp.getOuterDimsPerm (resp. packOp.getInnerDimsPerm) or empty.

Definition at line 1231 of file Transforms.cpp.

References mlir::OpOperand::getOperandNumber(), mlir::detail::IROperandBase::getOwner(), mlir::isPermutationVector(), mlir::RewriterBase::notifyMatchFailure(), mlir::RewriterBase::replaceOp(), mlir::OpBuilder::setInsertionPoint(), and transposeOneLinalgOperandAndReplace().

◆ peelLoop()

SmallVector< Value > mlir::linalg::peelLoop	(	RewriterBase &	rewriter,
		Operation *	op
	)

Try to peel and canonicalize loop op and return the new result.

Try to peel a loop op and return the new result.

Definition at line 224 of file Transforms.cpp.

References mlir::Operation::getResults(), mlir::scf::peelAndCanonicalizeForLoop(), and mlir::succeeded().

Referenced by peelLoops().

◆ peelLoops()

void mlir::linalg::peelLoops	(	RewriterBase &	rewriter,
		ArrayRef< scf::ForOp >	loops
	)

Peel and canonicalize 'loops'.

Definition at line 239 of file Transforms.cpp.

References peelLoop().

◆ populateBubbleUpExtractSliceOpPatterns()

void mlir::linalg::populateBubbleUpExtractSliceOpPatterns ( RewritePatternSet & patterns )

Patterns that are used to bubble up extract slice op above linalg op.

Definition at line 134 of file BubbleUpExtractSlice.cpp.

References mlir::RewritePatternSet::add(), and mlir::RewritePatternSet::getContext().

◆ populateCollapseDimensions()

void mlir::linalg::populateCollapseDimensions	(	RewritePatternSet &	patterns,
		const GetCollapsableDimensionsFn &	controlCollapseDimensions
	)

Pattern to collapse dimensions in a linalg.generic op.

This will collapse tensor operands when needed and expand back the result tensors.

Definition at line 1819 of file ElementwiseOpFusion.cpp.

References mlir::RewritePatternSet::add(), and mlir::RewritePatternSet::getContext().

◆ populateConstantFoldLinalgOperations()

void mlir::linalg::populateConstantFoldLinalgOperations	(	RewritePatternSet &	patterns,
		const ControlFusionFn &	controlFn
	)

Patterns to constant fold Linalg operations.

Definition at line 304 of file ConstantFold.cpp.

References mlir::RewritePatternSet::getContext(), and mlir::RewritePatternSet::insert().

◆ populateConvertToDestinationStylePatterns()

void mlir::linalg::populateConvertToDestinationStylePatterns ( RewritePatternSet & patterns )

Populate patterns that convert non-destination-style ops to destination style ops.

Definition at line 237 of file ConvertToDestinationStyle.cpp.

References mlir::RewritePatternSet::getContext(), and mlir::RewritePatternSet::insert().

◆ populateConvolutionVectorizationPatterns()

void mlir::linalg::populateConvolutionVectorizationPatterns	(	RewritePatternSet &	patterns,
		PatternBenefit	benefit = `1`
	)

Populate patterns for vectorizing low-D convolution ops.

This is a step in progressive lowering for convolution ops, it assume high-D convolution ops were decomposed previously.

Definition at line 2494 of file Vectorization.cpp.

References mlir::RewritePatternSet::add(), and mlir::RewritePatternSet::getContext().

◆ populateDataLayoutPropagationPatterns()

void mlir::linalg::populateDataLayoutPropagationPatterns ( RewritePatternSet & patterns )

Patterns to bubble up or down data layout ops across other operations.

Definition at line 333 of file DataLayoutPropagation.cpp.

References mlir::RewritePatternSet::getContext(), and mlir::RewritePatternSet::insert().

◆ populateDecomposeConvolutionPatterns()

void mlir::linalg::populateDecomposeConvolutionPatterns	(	RewritePatternSet &	patterns,
		PatternBenefit	benefit = `1`
	)

Linalg decompose convolutions patterns.

Populates patterns to decompose high-D convolution ops into low-D ones. This is a step in progressive lowering for convolution ops, afterwards we can vectorize the low-D convolution ops.

Definition at line 871 of file Transforms.cpp.

References mlir::RewritePatternSet::add(), and mlir::RewritePatternSet::getContext().

◆ populateDecomposeLinalgOpsPattern()

void mlir::linalg::populateDecomposeLinalgOpsPattern	(	RewritePatternSet &	patterns,
		bool	removeDeadArgsAndResults = `true`
	)

Populate patterns for splitting a LinalgOp with multiple statements within its payload into multiple GenericOp that have a single statement.

The option removeDeadArgsAndResults adds patterns to remove dead arguments and results from the generated decomposed ops. This is default true since the core decomposition patterns relies on these clean up patterns. It is set to false only for testing purposes.

Definition at line 379 of file DecomposeLinalgOps.cpp.

References mlir::RewritePatternSet::getContext(), mlir::RewritePatternSet::insert(), and populateEraseUnusedOperandsAndResultsPatterns().

◆ populateElementwiseOpsFusionPatterns()

void mlir::linalg::populateElementwiseOpsFusionPatterns	(	RewritePatternSet &	patterns,
		const ControlFusionFn &	controlElementwiseOpFusion
	)

Patterns for fusing linalg operation on tensors.

Pattern to fuse linalg.generic -> linalg.generic operations when both operations are fusable elementwise operations.

Definition at line 1808 of file ElementwiseOpFusion.cpp.

References mlir::RewritePatternSet::add(), mlir::RewritePatternSet::getContext(), and populateEraseUnusedOperandsAndResultsPatterns().

◆ populateElementwiseToLinalgConversionPatterns()

void mlir::linalg::populateElementwiseToLinalgConversionPatterns ( RewritePatternSet & patterns )

Populate patterns that convert ElementwiseMappable ops to linalg parallel loops.

Definition at line 120 of file ElementwiseToLinalg.cpp.

References mlir::RewritePatternSet::add(), and mlir::RewritePatternSet::getContext().

◆ populateEraseUnnecessaryInputsPatterns()

void mlir::linalg::populateEraseUnnecessaryInputsPatterns ( RewritePatternSet & patterns )

Patterns to promote inputs to outputs and remove unused inputs of linalg.generic ops.

Definition at line 421 of file EraseUnusedOperandsAndResults.cpp.

References mlir::RewritePatternSet::getContext(), and mlir::RewritePatternSet::insert().

◆ populateEraseUnusedOperandsAndResultsPatterns()

void mlir::linalg::populateEraseUnusedOperandsAndResultsPatterns ( RewritePatternSet & patterns )

Pattern to remove dead operands and results of linalg.generic operations.

This is effectively DCE for a linalg op.

Definition at line 414 of file EraseUnusedOperandsAndResults.cpp.

References mlir::RewritePatternSet::getContext(), and mlir::RewritePatternSet::insert().

Referenced by populateDecomposeLinalgOpsPattern(), and populateElementwiseOpsFusionPatterns().

◆ populateExtractOpVectorizationPatterns()

void mlir::linalg::populateExtractOpVectorizationPatterns	(	RewritePatternSet &	patterns,
		PatternBenefit	baseBenefit = `1`
	)

◆ populateFoldReshapeOpsByCollapsingPatterns()

void mlir::linalg::populateFoldReshapeOpsByCollapsingPatterns	(	RewritePatternSet &	patterns,
		const ControlFusionFn &	controlFoldingReshapes
	)

Patterns to fold an expanding tensor.expand_shape operation with its producer generic operation by collapsing the dimensions of the generic op.

Definition at line 1801 of file ElementwiseOpFusion.cpp.

References mlir::RewritePatternSet::add(), and mlir::RewritePatternSet::getContext().

◆ populateFoldReshapeOpsByExpansionPatterns()

void mlir::linalg::populateFoldReshapeOpsByExpansionPatterns	(	RewritePatternSet &	patterns,
		const ControlFusionFn &	controlFoldingReshapes
	)

Patterns to fold an expanding (collapsing) tensor_reshape operation with its producer (consumer) generic operation by expanding the dimensionality of the loop in the generic op.

Definition at line 1792 of file ElementwiseOpFusion.cpp.

References mlir::RewritePatternSet::add(), and mlir::RewritePatternSet::getContext().

◆ populateFoldUnitExtentDimsViaReshapesPatterns()

void mlir::linalg::populateFoldUnitExtentDimsViaReshapesPatterns ( RewritePatternSet & patterns )

Patterns to fold unit-extent dimensions in operands/results of linalg ops on tensors via reassociative reshape ops.

Patterns that are used to canonicalize the use of unit-extent dims for broadcasting.

Definition at line 664 of file DropUnitDims.cpp.

References mlir::RewritePatternSet::add(), mlir::RewritePatternSet::getContext(), mlir::tensor::populateFoldTensorEmptyPatterns(), mlir::memref::populateResolveRankedShapeTypeResultDimsPatterns(), and mlir::memref::populateResolveShapedTypeResultDimsPatterns().

◆ populateFoldUnitExtentDimsViaSlicesPatterns()

void mlir::linalg::populateFoldUnitExtentDimsViaSlicesPatterns ( RewritePatternSet & patterns )

Patterns to fold unit-extent dimensions in operands/results of linalg ops on tensors via rank-reducing slices.

Definition at line 683 of file DropUnitDims.cpp.

References mlir::RewritePatternSet::add(), mlir::RewritePatternSet::getContext(), mlir::tensor::populateFoldTensorEmptyPatterns(), mlir::memref::populateResolveRankedShapeTypeResultDimsPatterns(), and mlir::memref::populateResolveShapedTypeResultDimsPatterns().

◆ populateFuseTensorPadWithProducerLinalgOpPatterns()

void mlir::linalg::populateFuseTensorPadWithProducerLinalgOpPatterns ( RewritePatternSet & patterns )

Pattern to fuse a tensor.pad operation with the producer of its source, if the producer is a linalg operation with all parallel iterator types.

Definition at line 124 of file FusePadOpWithLinalgProducer.cpp.

References mlir::RewritePatternSet::add(), and mlir::RewritePatternSet::getContext().

◆ populateInlineConstantOperandsPatterns()

void mlir::linalg::populateInlineConstantOperandsPatterns ( RewritePatternSet & patterns )

Patterns that are used to inline constant operands into linalg generic ops.

Definition at line 94 of file InlineScalarOperands.cpp.

References mlir::RewritePatternSet::add(), and mlir::RewritePatternSet::getContext().

◆ populateLinalgNamedOpConversionPatterns()

void mlir::linalg::populateLinalgNamedOpConversionPatterns ( RewritePatternSet & patterns )

Patterns to convert from one named op to another.

These can be seen as canonicalizations of named ops into another named op.

Definition at line 160 of file NamedOpConversions.cpp.

References mlir::RewritePatternSet::add(), and mlir::RewritePatternSet::getContext().

◆ populateLinalgNamedOpsGeneralizationPatterns()

void mlir::linalg::populateLinalgNamedOpsGeneralizationPatterns ( RewritePatternSet & patterns )

Linalg generalization patterns.

Populates patterns with patterns to convert spec-generated named ops to linalg.generic ops.

Definition at line 88 of file Generalization.cpp.

References mlir::RewritePatternSet::add(), and mlir::RewritePatternSet::getContext().

◆ populateLinalgTilingCanonicalizationPatterns()

void mlir::linalg::populateLinalgTilingCanonicalizationPatterns ( RewritePatternSet & patterns )

Definition at line 970 of file Tiling.cpp.

References mlir::RewritePatternSet::getContext().

Referenced by getLinalgTilingCanonicalizationPatterns().

◆ populateLinalgToStandardConversionPatterns()

void mlir::linalg::populateLinalgToStandardConversionPatterns ( RewritePatternSet & patterns )

Populate the given list with patterns that convert from Linalg to Standard.

Definition at line 127 of file LinalgToStandard.cpp.

References mlir::RewritePatternSet::add(), and mlir::RewritePatternSet::getContext().

◆ populateMoveInitOperandsToInputPattern()

void mlir::linalg::populateMoveInitOperandsToInputPattern ( RewritePatternSet & patterns )

A pattern that converts init operands to input operands.

Definition at line 697 of file DropUnitDims.cpp.

References mlir::RewritePatternSet::add(), and mlir::RewritePatternSet::getContext().

◆ populatePadOpVectorizationPatterns()

void mlir::linalg::populatePadOpVectorizationPatterns	(	RewritePatternSet &	patterns,
		PatternBenefit	baseBenefit = `1`
	)

Populates patterns with patterns that vectorize tensor.pad.

These patterns are meant to apply in a complementary fashion. Benefits are used to encode a certain ordering of pattern application. To avoid scattering magic constants throughout the code base, the patterns must be added with this function. baseBenefit can be used to offset the benefit of all tensor::PadOp vectorization patterns by a certain value.

Definition at line 1592 of file Vectorization.cpp.

References mlir::RewritePatternSet::add(), mlir::PatternBenefit::getBenefit(), and mlir::RewritePatternSet::getContext().

◆ populatePadTensorTilingPatterns()

void mlir::linalg::populatePadTensorTilingPatterns	(	RewritePatternSet &	patterns,
		const LinalgTilingOptions &	options
	)

Definition at line 998 of file Tiling.cpp.

References mlir::RewritePatternSet::add(), mlir::RewritePatternSet::getContext(), and options.

◆ populateSparseTensorRewriting()

void mlir::linalg::populateSparseTensorRewriting ( RewritePatternSet & patterns )

Populate patterns that are only useful in the context of sparse tensors.

◆ populateSplitReductionPattern()

void mlir::linalg::populateSplitReductionPattern	(	RewritePatternSet &	patterns,
		const ControlSplitReductionFn &	controlSplitReductionFn,
		bool	useAlloc = `false`
	)

Patterns to apply splitReduction below.

Definition at line 442 of file SplitReduction.cpp.

References mlir::RewritePatternSet::add(), and mlir::RewritePatternSet::getContext().

◆ populateSwapExtractSliceWithFillPatterns()

void mlir::linalg::populateSwapExtractSliceWithFillPatterns ( RewritePatternSet & patterns )

Adds patterns that waps tensor.extract_slice(linalg.fill(cst, init)) into linalg.fill(cst, tensor.extract_slice(init)).

Definition at line 38 of file SwapExtractSliceWithFillPatterns.cpp.

References mlir::RewritePatternSet::add(), and mlir::RewritePatternSet::getContext().

◆ promoteSubviewAsNewBuffer()

FailureOr< PromotionInfo > mlir::linalg::promoteSubviewAsNewBuffer	(	OpBuilder &	b,
		Location	loc,
		memref::SubViewOp	subView,
		const AllocBufferCallbackFn &	allocationFn,
		DataLayout &	layout
	)

Definition at line 210 of file Promotion.cpp.

References mlir::OpBuilder::create(), mlir::OpBuilder::createOrFold(), mlir::detail::enumerate(), mlir::failed(), mlir::failure(), getConstantUpperBoundForIndex(), mlir::Builder::getIndexAttr(), and mlir::getValueOrCreateConstantIndexOp().

Referenced by promoteSubViews().

◆ promoteSubViews()

FailureOr< LinalgOp > mlir::linalg::promoteSubViews	(	OpBuilder &	b,
		LinalgOp	op,
		const LinalgPromotionOptions &	options
	)

Promote the subViews into a new buffer allocated at the insertion point b.

Promotion occurs in 3 steps:

Create a new buffer for a full tile (i.e. not clipped at the boundary).
Take a full view on the buffer.
Take a partial slice of the full view in step 2. and copy into it.

Return the modified linalg op (the modification happens in place) as well as all the copy ops created.

Definition at line 390 of file Promotion.cpp.

References mlir::failed(), mlir::failure(), options, and promoteSubViews().

◆ promoteSubviewsPrecondition()

LogicalResult mlir::linalg::promoteSubviewsPrecondition	(	Operation *	op,
		LinalgPromotionOptions	options
	)

Promote memref.subviews feeding linalg-on-buffers operations.

Definition at line 368 of file Promotion.cpp.

References mlir::failure(), options, and mlir::success().

◆ registerBufferizableOpInterfaceExternalModels()

void mlir::linalg::registerBufferizableOpInterfaceExternalModels ( DialectRegistry & registry )

Definition at line 126 of file BufferizableOpInterfaceImpl.cpp.

References mlir::DialectRegistry::addExtension().

Referenced by mlir::registerAllDialects().

◆ registerTilingInterfaceExternalModels()

void mlir::linalg::registerTilingInterfaceExternalModels ( DialectRegistry & registry )

Definition at line 417 of file TilingInterfaceImpl.cpp.

References mlir::DialectRegistry::addExtension(), and registerAll().

Referenced by mlir::registerAllDialects().

◆ registerTransformDialectExtension()

void mlir::linalg::registerTransformDialectExtension ( DialectRegistry & registry )

Definition at line 2765 of file LinalgTransformOps.cpp.

References mlir::DialectRegistry::addExtensions().

Referenced by mlir::registerAllDialects().

◆ rewriteAsPaddedOp()

FailureOr< SmallVector< Value > > mlir::linalg::rewriteAsPaddedOp	(	OpBuilder &	b,
		LinalgOp	opToPad,
		ArrayRef< int64_t >	paddingDimensions,
		ArrayRef< Attribute >	paddingValues,
		ArrayRef< bool >	packPaddings,
		LinalgOp &	paddedOp
	)

Pad the iterator dimensions paddingDimensions of all opToPad operands to a static bounding box.

Use paddingValues and packPaddings to set padding value and nofold attribute of the created tensor::PadOps, respectively. Update paddedOp to the cloned operation with statically shaped paddingDimensions and return the extracted dynamically shaped results. If padding fails, return failure.

Definition at line 166 of file Transforms.cpp.

References mlir::Type::cast(), mlir::clone(), mlir::OpBuilder::create(), mlir::detail::enumerate(), mlir::failed(), mlir::failure(), mlir::getAsOpFoldResult(), mlir::Builder::getIndexAttr(), mlir::Value::getType(), mlir::ValueRange::getTypes(), padOperandToSmallestStaticBoundingBox(), and mlir::OpBuilder::setInsertionPointAfter().

Referenced by mlir::linalg::LinalgPaddingPattern::returningMatchAndRewrite().

◆ splitOp()

std::pair< TilingInterface, TilingInterface > mlir::linalg::splitOp	(	RewriterBase &	rewriter,
		TilingInterface	op,
		unsigned	dimension,
		OpFoldResult	splitPoint
	)

Split the given op into two parts along the given iteration space dimension at the specified splitPoint, and return the two parts.

If the second part is statically known to be empty, do not create it and return nullptr instead. Error state is signalled by returning a pair of nullptrs.

For example, the following op:

linalg.matmul ins(%0, %1 : tensor<128x32xf32>, tensor<32x64xf32>) outs(%2 : tensor<128x64xf32>)

split along the first dimension at position 42 will result in:

%3 = tensor.extract_slice %0[0, 0][42, 32][1, 1] %4 = tensor.extract_slice %2[0, 0][42, 64][1, 1] %5 = linalg.matmul ins(%3, %1 : tensor<42x32xf32>, tensor<32x64xf32>) outs(%5 : tensor<42x64xf32>) %6 = tensor.insert_slice %5 into %2[0, 0][42, 64][1, 1]

%7 = tensor.extract_slice %0[42, 0][86, 32][1, 1] %8 = tensor.extract_slice %6[42, 0][86, 64][1, 1] %9 = linalg.matmul ins(%7, %1 : tensor<86x32xf32>, tensor<32x64xf32>) outs(%8 : tensor<86x64xf32>) tensor.insert_slice %5 into %6[42, 0][86, 64][1, 1]

Note that there is no simplification other than constant propagation applied to slice extraction and insertion.

Definition at line 67 of file Split.cpp.

References mlir::bindDims(), createSplitPart(), mlir::Builder::getContext(), mlir::tensor::getOrCreateDestinations(), mlir::AffineMap::inferFromExprList(), mlir::makeComposedFoldedAffineApply(), mlir::makeComposedFoldedAffineMin(), mlir::Range::offset, mlir::RewriterBase::replaceOp(), mlir::Range::size, mlir::succeeded(), and mlir::RewriterBase::updateRootInPlace().

◆ splitReduction()

FailureOr< SplitReductionResult > mlir::linalg::splitReduction	(	PatternRewriter &	b,
		LinalgOp	op,
		const ControlSplitReductionFn &	controlSplitReductionFn,
		bool	useAlloc = `false`
	)

◆ splitReductionByScaling()

FailureOr< SplitReductionResult > mlir::linalg::splitReductionByScaling	(	PatternRewriter &	b,
		LinalgOp	op,
		const ControlSplitReductionFn &	controlSplitReductionFn,
		bool	useAlloc = `false`
	)

Scaling-based implementation of the split reduction transformation.

Core rewrite implementation.

Instead of introducing an ExpandShapeOp, this rewrites a reduction dimension k into k * scale + kk.

Example: ``` %0 = linalg.matmul ins(A, B: tensor<16x256xf32>, tensor<256x32xf32>) outs(C: tensor<16x32xf32>) -> tensor<16x32xf32> ```

Is transformed to:

``` #map0 = affine_map<(d0, d1, d2, d3) -> (d0, d2 * 4 + d3)> #map1 = affine_map<(d0, d1, d2, d3) -> (d2 * 4 + d3, d1)> #map2 = affine_map<(d0, d1, d2, d3) -> (d2, d3)> #map3 = affine_map<(d0, d1, d2, d3) -> (d0, d1, d2)> #map4 = affine_map<(d0, d1, d2) -> (d0, d1, d2)> #map5 = affine_map<(d0, d1, d2) -> (d0, d1)> %0 = tensor.empty [16, 32, 64] : tensor<16x32x64xf32> cst = arith.constant 0.000000e+00 : f32 %1 = linalg.fill ins(cst : f32) outs(%0 : tensor<16x32x64xf32>) -> tensor<16x32x64xf32> %2 = tensor.empty [64, 4] : tensor<64x4xi1>

%3 = linalg.generic {indexing_maps = [#map0, #map1, #map2, #map3], iterator_types = ["parallel", "parallel", "parallel", "reduction"]} ins(A, B, %2 : tensor<16x256xf32>, tensor<256x32xf32>, tensor<64x4xi1>) outs(%1 : tensor<16x32x64xf32>) { ^bb0(arg3: f32, arg4: f32, arg5: i1, arg6: f32): %5 = arith.mulf arg3, arg4 : f32 %6 = arith.addf arg6, %5 : f32 linalg.yield %6 : f32 } -> tensor<16x32x64xf32>

%4 = linalg.generic {indexing_maps = [#map4, #map5], iterator_types = ["parallel", "parallel", "reduction"]} */ // ins(%3 : tensor<16x32x64xf32>) /** outs(C : tensor<16x32xf32>) { ^bb0(arg3: f32, arg4: f32): %5 = arith.addf arg3, arg4 : f32 linalg.yield %5 : f32 } -> tensor<16x32xf32>

return %4 : tensor<16x32xf32> ```

Definition at line 236 of file SplitReduction.cpp.

References mlir::Type::cast(), mlir::Operation::clone(), mlir::OpBuilder::create(), mlir::tensor::createDynamicDimValues(), mlir::AffineMap::dropResult(), mlir::AffineMap::get(), mlir::getAffineDimExpr(), mlir::Value::getDefiningOp(), mlir::Builder::getIntegerType(), mlir::Builder::getMultiDimIdentityMap(), getNeutralElement(), mlir::Operation::getResult(), mlir::Value::getType(), mlir::ValueRange::getTypes(), mlir::linalg::SplitReductionOptions::index, mlir::RewriterBase::inlineRegionBefore(), mlir::linalg::SplitReductionOptions::innerParallel, mlir::RankedTensorType::Builder::insertDim(), insertParallelDim(), mlir::matchReduction(), mlir::RewriterBase::notifyMatchFailure(), mlir::linalg::SplitReductionOptions::ratio, mlir::RewriterBase::replaceOp(), scaleReductionDim(), mlir::OpBuilder::setInsertionPoint(), and mlir::Operation::setOperand().

◆ tileLinalgOp()

FailureOr< TiledLinalgOp > mlir::linalg::tileLinalgOp	(	RewriterBase &	b,
		LinalgOp	op,
		const LinalgTilingOptions &	options
	)

Definition at line 843 of file Tiling.cpp.

References mlir::failure(), Loops, options, and ParallelLoops.

Referenced by mlir::linalg::TileLoopNest::tileRootOp().

◆ tileReductionUsingForeachThread()

FailureOr< linalg::ForeachThreadReductionTilingResult > mlir::linalg::tileReductionUsingForeachThread	(	RewriterBase &	b,
		PartialReductionOpInterface	op,
		ArrayRef< OpFoldResult >	numThreads,
		ArrayRef< OpFoldResult >	tileSizes = `{}`,
		std::optional< ArrayAttr >	mapping = `std::nullopt`
	)

Method to tile a reduction to parallel iterations computing partial reductions.

After the loop all the partial reduction are merged into a final reduction. For example for the following sequence

%0 = linalg.generic %in ["parallel", "reduction"]

: tensor<7x9xf32> -> tensor<7xf32>

into:

%0 = linalg.fill ... : tensor<7x4xf32>
%1 = scf.foreach_thread (%iv) in (%c4) shared_outs(%arg0 = %0)
  -> (tensor<7x4xf32>) {
  %2 = tensor.extract_slice %arg3 : tensor<7x4xf32> to tensor<7xf32>
  %3 = tensor.extract_slice %in : tensor<7x9xf32> -> tensor<7x?xf32>
  %4 = linalg.generic %2, %3 ["parallel", "reduction"]
    : tensor<7x?xf32> -> tensor<7xf32>
  %5 = tensor.insert_slice %3, %arg0[0, %iv] : tensor<7x4xf32>
}
%6 = linalg.generic %1 ["parallel", "reduction"]
  : tensor<7x4xf32> -> tensor<7xf32>

Definition at line 611 of file Tiling.cpp.

References calculateTileOffsetsAndSizes(), mlir::Attribute::cast(), mlir::OpBuilder::clone(), mlir::OpBuilder::create(), mlir::RewriterBase::eraseOp(), mlir::failed(), mlir::getAsValues(), mlir::Builder::getIndexAttr(), mlir::tensor::getOrCreateDestinations(), mlir::Operation::getResults(), mlir::linalg::ForeachThreadReductionTilingResult::initialOp, mlir::isConstantIntValue(), mlir::linalg::ForeachThreadReductionTilingResult::loops, mlir::mapLoopToProcessorIds(), mlir::linalg::ForeachThreadReductionTilingResult::mergeOp, mlir::RewriterBase::notifyMatchFailure(), options, mlir::linalg::ForeachThreadReductionTilingResult::parallelTiledOp, mlir::RewriterBase::replaceOp(), mlir::OpBuilder::setInsertionPoint(), mlir::OpBuilder::setInsertionPointAfter(), mlir::OpBuilder::setInsertionPointToEnd(), and mlir::RewriterBase::updateRootInPlace().

◆ tileToForeachThreadOp()

FailureOr< ForeachThreadTilingResult > mlir::linalg::tileToForeachThreadOp	(	RewriterBase &	builder,
		TilingInterface	op,
		ArrayRef< OpFoldResult >	numThreads,
		std::optional< ArrayAttr >	mapping
	)

Definition at line 426 of file Tiling.cpp.

References tileToForeachThreadOpImpl().

Referenced by mlir::transform::tileToForeachThreadOpImpl().

◆ tileToForeachThreadOpUsingTileSizes()

FailureOr< ForeachThreadTilingResult > mlir::linalg::tileToForeachThreadOpUsingTileSizes	(	RewriterBase &	builder,
		TilingInterface	op,
		ArrayRef< OpFoldResult >	tileSizes,
		std::optional< ArrayAttr >	mapping
	)

Same as tileToForeachThreadOp, but calculate the number of threads required using the given tileSizes.

Definition at line 435 of file Tiling.cpp.

References mlir::bindSymbols(), mlir::AffineExpr::ceilDiv(), mlir::Builder::getContext(), mlir::isConstantIntValue(), mlir::makeComposedFoldedAffineApply(), and tileToForeachThreadOpImpl().

Referenced by mlir::transform::tileToForeachThreadOpImpl().

◆ transformIndexOps()

void mlir::linalg::transformIndexOps	(	RewriterBase &	b,
		LinalgOp	op,
		SmallVectorImpl< Value > &	ivs,
		const LoopIndexToRangeIndexMap &	loopIndexToRangeIndex
	)

All indices returned by IndexOp should be invariant with respect to tiling.

Therefore, if an operation is tiled, we have to transform the indices accordingly, i.e. offset them by the values of the corresponding induction variables that are captured implicitly in the body of the op.

Example. linalg.generic before tiling:

#id_2d = (i, j) -> (i, j) #pointwise_2d_trait = { indexing_maps = [#id_2d, #id_2d], iterator_types = ["parallel", "parallel"] } linalg.generic #pointwise_2d_trait operand, result { ^bb0(operand_in: f32, result_in: f32): i = linalg.index 0 : index j = linalg.index 1 : index <some operations that use i, j> }: memref<50x100xf32>, memref<50x100xf32>

After tiling pass with tiles sizes 10 and 25:

#strided = (i, j)[s0, s1, s2] -> (i * s1 + s0 + j * s2)

c1 = arith.constant 1 : index c0 = arith.constant 0 : index c25 = arith.constant 25 : index c10 = arith.constant 10 : index operand_dim_0 = dim operand, 0 : memref<50x100xf32> operand_dim_1 = dim operand, 1 : memref<50x100xf32> scf.for k = c0 to operand_dim_0 step c10 { scf.for l = c0 to operand_dim_1 step c25 { %4 = memref.subview operand[k, l][c10, c25][c1, c1] : memref<50x100xf32> to memref<?x?xf32, #strided> %5 = memref.subview result[k, l][c10, c25][c1, c1] : memref<50x100xf32> to memref<?x?xf32, #strided> linalg.generic pointwise_2d_trait %4, %5 { ^bb0(operand_in: f32, result_in: f32): i = linalg.index 0 : index j = linalg.index 1 : index // Indices k and l are implicitly captured in the body. transformed_i = arith.addi i, k : index // index i is offset by k transformed_j = arith.addi j, l : index // index j is offset by l // Every use of i, j is replaced with transformed_i, transformed_j <some operations that use transformed_i, transformed_j> }: memref<?x?xf32, #strided>, memref<?x?xf32, #strided> } }

TODO: Investigate whether mixing implicit and explicit indices does not lead to losing information.

Definition at line 88 of file Tiling.cpp.

References mlir::detail::enumerate(), mlir::getAsOpFoldResult(), and offsetIndices().

Referenced by tileLinalgOpImpl().

◆ updateBoundsForCyclicDistribution()

void mlir::linalg::updateBoundsForCyclicDistribution	(	OpBuilder &	builder,
		Location	loc,
		Value	procId,
		Value	nprocs,
		Value &	lb,
		Value &	ub,
		Value &	step
	)

Update the lb, ub and step to get per processor lb, ub and step.

Definition at line 545 of file Utils.cpp.

References mlir::bindDims(), mlir::getAffineSymbolExpr(), mlir::Builder::getContext(), and mlir::makeComposedAffineApply().

Referenced by mlir::linalg::GenerateLoopNest< LoopTy >::doit().

◆ vectorize()

LogicalResult mlir::linalg::vectorize	(	RewriterBase &	rewriter,
		LinalgOp	linalgOp,
		ArrayRef< int64_t >	inputVectorSizes = `{}`,
		bool	vectorizeNDExtract = `false`
	)

Emit a suitable vector form for a Linalg op.

If provided, inputVectorSizes are used to vectorize this operation. inputVectorSizes must match the rank of the iteration space of the operation and the sizes must be smaller or equal than their counterpart interation space sizes, if static. inputVectorShapes also allows the vectorization of operations with dynamic shapes.

If provided, inputVectorSizes are used to vectorize this operation. inputVectorSizes must match the rank of the iteration space of the operation and the input vector sizes must be greater than or equal to their counterpart iteration space sizes, if static. inputVectorShapes also allows the vectorization of operations with dynamic shapes.

Definition at line 1074 of file Vectorization.cpp.

References convertAffineApply(), mlir::RewriterBase::eraseOp(), mlir::failed(), mlir::failure(), VectorizationState::initState(), LDBG, mlir::RewriterBase::replaceOp(), mlir::succeeded(), mlir::success(), vectorizeAsLinalgGeneric(), vectorizeConvolution(), and vectorizeLinalgOpPrecondition().

◆ vectorizeCopy()

LogicalResult mlir::linalg::vectorizeCopy	(	RewriterBase &	builder,
		memref::CopyOp	copyOp
	)

Emit a suitable vector form for a Copy op with fully static shape.

Definition at line 1125 of file Vectorization.cpp.

References mlir::Type::cast(), mlir::OpBuilder::create(), mlir::failure(), mlir::getElementTypeOrSelf(), mlir::Builder::getMultiDimIdentityMap(), mlir::Operation::getResults(), mlir::Value::getType(), mlir::RewriterBase::replaceOp(), and mlir::success().

Referenced by mlir::linalg::CopyVectorizationPattern::matchAndRewrite().

◆ vectorizeLinalgOpPrecondition()

LogicalResult mlir::linalg::vectorizeLinalgOpPrecondition	(	LinalgOp	linalgOp,
		ArrayRef< int64_t >	inputVectorSizes = `{}`,
		bool	vectorizeNDExtract = `false`
	)

Return success if the operation can be vectorized.

Definition at line 974 of file Vectorization.cpp.

References allIndexingsAreProjectedPermutation(), mlir::failed(), mlir::failure(), mlir::Region::front(), mlir::Operation::getRegion(), isElementwise(), LDBG, reductionPreconditions(), mlir::succeeded(), mlir::success(), tensorExtractVectorizationPrecondition(), and vectorizeDynamicLinalgOpPrecondition().

Referenced by vectorize().

Namespaces

Classes

Typedefs

Enumerations

Functions

Typedef Documentation

◆ AllocBufferCallbackFn

◆ ControlFusionFn

◆ ControlSplitReductionFn

◆ CopyCallbackFn

◆ DeallocBufferCallbackFn

◆ FusableOpDependencesTy

◆ GetCollapsableDimensionsFn

◆ LinalgLoops

◆ LoopIndexToRangeIndexMap

◆ OptimizeCopyFn

◆ ProcInfoCallBackFn

◆ TileSizeComputationFunction

Enumeration Type Documentation

◆ DistributionMethod

◆ LinalgTilingLoopType

Function Documentation

◆ allIndexingsAreProjectedPermutation()

◆ areElementwiseOpsFusable()

◆ collapseGenericOpIterationDims()

◆ computeAllSliceParameters()

◆ computeMultiTileSizes()

◆ computeSliceParameters()

◆ computeStaticMultiTileSizes()

◆ computeTileOffsets()

◆ computeTileSizes()

◆ concat()

◆ createFoldedDimOp()

◆ createOrFoldDimOp()

◆ extractOrIdentityMap()

◆ findAllFusableDependences()

◆ fuseElementwiseOps()

◆ fuseProducerOfBuffer()

◆ fuseProducerOfTensor() [1/2]

◆ fuseProducerOfTensor() [2/2]

◆ generalizeNamedOp()

◆ generateLibraryCallName()

◆ generateParallelLoopNest()

◆ getCombinerOpKind()

◆ getConstantUpperBoundForIndex()

◆ getDynOperands()

◆ getLinalgTilingCanonicalizationPatterns()

◆ getNeutralElement()

◆ getPrunedAttributeList()

◆ getReassociationMapForFoldingUnitDims()

◆ getTensorOutputTypes()

◆ getUpperBoundForIndex()

◆ hasOnlyScalarElementwiseOp()

◆ hoistPaddingOnTensors()

Example in pseudo-mlir:

◆ hoistRedundantVectorTransfers()

◆ hoistRedundantVectorTransfersOnTensor()

◆ insertSlicesBack()

◆ interchangeGenericOp()

◆ isaContractionOpInterface()

◆ isElementwise()

◆ isFusableInto()

◆ isParallelIterator()

◆ isProducerLastWriteOfView()

◆ isReductionIterator()

◆ linalgOpToAffineLoops()

◆ linalgOpToLoops()

◆ linalgOpToParallelLoops()

◆ makeAffineDimExprs()

◆ makeComposedPadHighOp()

◆ makeMemRefCopyOp()

◆ makeTiledLoopRanges()

◆ makeTiledShape()

◆ makeTiledShapes()

◆ makeTransposeOp()

◆ materializeTiledShape()

◆ offsetIndices() [1/2]

◆ offsetIndices() [2/2]

◆ pack()

◆ packTranspose()