#include "mlir/Dialect/Affine/LoopUtils.h"
#include "mlir/Analysis/SliceAnalysis.h"
#include "mlir/Dialect/Affine/Analysis/AffineAnalysis.h"
#include "mlir/Dialect/Affine/Analysis/LoopAnalysis.h"
#include "mlir/Dialect/Affine/Analysis/Utils.h"
#include "mlir/Dialect/Affine/IR/AffineOps.h"
#include "mlir/Dialect/Affine/IR/AffineValueMap.h"
#include "mlir/Dialect/Affine/Utils.h"
#include "mlir/Dialect/Func/IR/FuncOps.h"
#include "mlir/Dialect/MemRef/IR/MemRef.h"
#include "mlir/Dialect/SCF/IR/SCF.h"
#include "mlir/IR/IRMapping.h"
#include "mlir/IR/IntegerSet.h"
#include "mlir/Support/MathExtras.h"
#include "mlir/Transforms/GreedyPatternRewriteDriver.h"
#include "llvm/ADT/MapVector.h"
#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/raw_ostream.h"
#include <optional>

Classes
struct	JamBlockGatherer

struct	StrideInfo

Macros
#define	DEBUG_TYPE "loop-utils"

Functions
static void	getCleanupLoopLowerBound (AffineForOp forOp, unsigned unrollFactor, AffineMap &cleanupLbMap, SmallVectorImpl< Value > &cleanupLbOperands)
	Computes the cleanup loop lower bound of the loop being unrolled with the specified unroll factor; this bound will also be upper bound of the main part of the unrolled loop. More...

static void	replaceIterArgsAndYieldResults (AffineForOp forOp)
	Helper to replace uses of loop carried values (iter_args) and loop yield values while promoting single iteration affine.for ops. More...

static AffineForOp	generateShiftedLoop (AffineMap lbMap, AffineMap ubMap, const std::vector< std::pair< uint64_t, ArrayRef< Operation * >>> &opGroupQueue, unsigned offset, AffineForOp srcForOp, OpBuilder b)
	Generates an affine.for op with the specified lower and upper bounds while generating the right IV remappings to realize shifts for operations in its body. More...

static LogicalResult	checkIfHyperRectangular (MutableArrayRef< AffineForOp > input)
	Checks whether a loop nest is hyper-rectangular or not. More...

template<typename t >
static LogicalResult	performPreTilingChecks (MutableArrayRef< AffineForOp > input, ArrayRef< t > tileSizes)
	Check if the input nest is supported for tiling and whether tiling would be legal or not. More...

static void	moveLoopBodyImpl (AffineForOp src, AffineForOp dest, Block::iterator loc)
	Move the loop body of AffineForOp 'src' from 'src' into the specified location in destination's body, ignoring the terminator. More...

static void	moveLoopBody (AffineForOp src, AffineForOp dest)
	Move the loop body of AffineForOp 'src' from 'src' to the start of dest body. More...

static void	constructTiledLoopNest (MutableArrayRef< AffineForOp > origLoops, AffineForOp rootAffineForOp, unsigned width, MutableArrayRef< AffineForOp > tiledLoops)
	Constructs tiled loop nest, without setting the loop bounds and move the body of the original loop nest to the tiled loop nest. More...

static void	setIntraTileBoundsParametric (OpBuilder &b, AffineForOp origLoop, AffineForOp newInterTileLoop, AffineForOp newIntraTileLoop, Value tileSize)
	Set lower and upper bounds of intra-tile loops for parametric tiling. More...

static void	setInterTileBoundsParametric (OpBuilder &b, AffineForOp origLoop, AffineForOp newLoop, Value tileSize)
	Set lower and upper bounds of inter-tile loops for parametric tiling. More...

static void	constructParametricallyTiledIndexSetHyperRect (MutableArrayRef< AffineForOp > origLoops, MutableArrayRef< AffineForOp > newLoops, ArrayRef< Value > tileSizes)
	Constructs and sets new loop bounds after tiling for the case of hyper-rectangular index sets, where the bounds of one dimension do not depend on other dimensions and tiling parameters are captured from SSA values. More...

static void	constructTiledIndexSetHyperRect (MutableArrayRef< AffineForOp > origLoops, MutableArrayRef< AffineForOp > newLoops, ArrayRef< unsigned > tileSizes)
	Constructs and sets new loop bounds after tiling for the case of hyper-rectangular index sets, where the bounds of one dimension do not depend on other dimensions. More...

static void	generateUnrolledLoop (Block loopBodyBlock, Value forOpIV, uint64_t unrollFactor, function_ref< Value(unsigned, Value, OpBuilder)> ivRemapFn, function_ref< void(unsigned, Operation , OpBuilder)> annotateFn, ValueRange iterArgs, ValueRange yieldedValues)
	Generates unrolled copies of AffineForOp 'loopBodyBlock', with associated 'forOpIV' by 'unrollFactor', calling 'ivRemapFn' to remap 'forOpIV' for each unrolled body. More...

static LogicalResult	generateCleanupLoopForUnroll (AffineForOp forOp, uint64_t unrollFactor)
	Helper to generate cleanup loop for unroll or unroll-and-jam when the trip count is not a multiple of `unrollFactor`. More...

static bool	areInnerBoundsInvariant (AffineForOp forOp)
	Check if all control operands of all loops are defined outside of `forOp` and return false if not. More...

static bool	checkLoopInterchangeDependences (const std::vector< SmallVector< DependenceComponent, 2 >> &depCompsVec, ArrayRef< AffineForOp > loops, ArrayRef< unsigned > loopPermMap)

static void	augmentMapAndBounds (OpBuilder &b, Value iv, AffineMap map, SmallVector< Value, 4 > operands, int64_t offset=0)

static SmallVector< AffineForOp, 8 >	stripmineSink (AffineForOp forOp, uint64_t factor, ArrayRef< AffineForOp > targets)

template<typename SizeType >
static AffineForOp	stripmineSink (AffineForOp forOp, SizeType factor, AffineForOp target)

static void	findHighestBlockForPlacement (const MemRefRegion &region, Block &block, Block::iterator &begin, Block::iterator &end, Block *copyPlacementBlock, Block::iterator copyInPlacementStart, Block::iterator *copyOutPlacementStart)
	Given a memref region, determine the lowest depth at which transfers can be placed for it, and return the corresponding block, start and end positions in the block for placing incoming (read) and outgoing (write) copies respectively. More...

static void	getMultiLevelStrides (const MemRefRegion &region, ArrayRef< int64_t > bufferShape, SmallVectorImpl< StrideInfo > *strideInfos)
	Returns striding information for a copy/transfer of this region with potentially multiple striding levels from outermost to innermost. More...

static AffineForOp	generatePointWiseCopy (Location loc, Value memref, Value fastMemRef, ArrayRef< AffineMap > lbMaps, ArrayRef< Value > lbOperands, ArrayRef< AffineMap > ubMaps, ArrayRef< Value > ubOperands, ArrayRef< AffineExpr > fastBufOffsets, bool isCopyOut, OpBuilder b)
	Generates a point-wise copy from/to ‘memref’ to/from ‘fastMemRef’ and returns the outermost AffineForOp of the copy loop nest. More...

static InFlightDiagnostic LLVM_ATTRIBUTE_UNUSED	emitRemarkForBlock (Block &block)

static LogicalResult	generateCopy (const MemRefRegion &region, Block block, Block::iterator begin, Block::iterator end, Block copyPlacementBlock, Block::iterator copyInPlacementStart, Block::iterator copyOutPlacementStart, const AffineCopyOptions &copyOptions, DenseMap< Value, Value > &fastBufferMap, DenseSet< Operation * > &copyNests, uint64_t sizeInBytes, Block::iterator nBegin, Block::iterator *nEnd)
	Creates a buffer in the faster memory space for the specified memref region; generates a copy from the lower memory space to this one, and replaces all loads/stores in the block range [‘begin’, ‘end’) of ‘block’ to load/store from that buffer. More...

static bool	getFullMemRefAsRegion (Operation op, unsigned numParamLoopIVs, MemRefRegion region)
	Construct the memref region to just include the entire memref. More...

static void	gatherLoopsInBlock (Block *block, unsigned currLoopDepth, std::vector< SmallVector< AffineForOp, 2 >> &depthToLoops)
	Gathers all AffineForOps in 'block' at 'currLoopDepth' in 'depthToLoops'. More...

static AffineIfOp	createSeparationCondition (MutableArrayRef< AffineForOp > loops, OpBuilder b)
	Creates an AffineIfOp that encodes the conditional to choose between the constant trip count version and an unknown trip count version of this nest of loops. More...

static LogicalResult	createFullTiles (MutableArrayRef< AffineForOp > inputNest, SmallVectorImpl< AffineForOp > &fullTileLoops, OpBuilder b)
	Create the full tile loop nest (along with its body). More...

Macro Definition Documentation

◆ DEBUG_TYPE

#define DEBUG_TYPE "loop-utils"

Definition at line 34 of file LoopUtils.cpp.

Function Documentation

◆ areInnerBoundsInvariant()

static bool areInnerBoundsInvariant ( AffineForOp forOp )

static

Check if all control operands of all loops are defined outside of forOp and return false if not.

Definition at line 1094 of file LoopUtils.cpp.

References mlir::WalkResult::advance().

Referenced by mlir::affine::loopUnrollJamByFactor().

◆ augmentMapAndBounds()

static void augmentMapAndBounds	(	OpBuilder &	b,
		Value	iv,
		AffineMap *	map,
		SmallVector< Value, 4 > *	operands,
		int64_t	offset = `0`
	)

static

Definition at line 1549 of file LoopUtils.cpp.

References mlir::affine::canonicalizeMapAndOperands(), mlir::AffineMap::get(), mlir::Builder::getAffineDimExpr(), mlir::Builder::getContext(), mlir::AffineMap::getNumDims(), mlir::AffineMap::getNumSymbols(), and mlir::AffineMap::getResults().

Referenced by stripmineSink().

◆ checkIfHyperRectangular()

static LogicalResult checkIfHyperRectangular ( MutableArrayRef< AffineForOp > input )

static

Checks whether a loop nest is hyper-rectangular or not.

Definition at line 369 of file LoopUtils.cpp.

References mlir::failed(), mlir::failure(), mlir::affine::getIndexSet(), mlir::presburger::IntegerRelation::isHyperRectangular(), and mlir::success().

◆ checkLoopInterchangeDependences()

static bool checkLoopInterchangeDependences	(	const std::vector< SmallVector< DependenceComponent, 2 >> &	depCompsVec,
		ArrayRef< AffineForOp >	loops,
		ArrayRef< unsigned >	loopPermMap
	)

static

Definition at line 1353 of file LoopUtils.cpp.

Referenced by mlir::affine::isValidLoopInterchangePermutation(), and mlir::affine::sinkSequentialLoops().

◆ constructParametricallyTiledIndexSetHyperRect()

static void constructParametricallyTiledIndexSetHyperRect	(	MutableArrayRef< AffineForOp >	origLoops,
		MutableArrayRef< AffineForOp >	newLoops,
		ArrayRef< Value >	tileSizes
	)

static

Constructs and sets new loop bounds after tiling for the case of hyper-rectangular index sets, where the bounds of one dimension do not depend on other dimensions and tiling parameters are captured from SSA values.

Bounds of each dimension can thus be treated independently, and deriving the new bounds is much simpler and faster than for the case of tiling arbitrary polyhedral shapes.

Definition at line 668 of file LoopUtils.cpp.

References setInterTileBoundsParametric(), and setIntraTileBoundsParametric().

Referenced by mlir::affine::tilePerfectlyNestedParametric().

◆ constructTiledIndexSetHyperRect()

static void constructTiledIndexSetHyperRect	(	MutableArrayRef< AffineForOp >	origLoops,
		MutableArrayRef< AffineForOp >	newLoops,
		ArrayRef< unsigned >	tileSizes
	)

static

Constructs and sets new loop bounds after tiling for the case of hyper-rectangular index sets, where the bounds of one dimension do not depend on other dimensions.

Bounds of each dimension can thus be treated independently, and deriving the new bounds is much simpler and faster than for the case of tiling arbitrary polyhedral shapes.

Definition at line 696 of file LoopUtils.cpp.

References mlir::AffineMap::get(), mlir::Builder::getAffineDimExpr(), mlir::affine::getConstantTripCount(), mlir::Builder::getContext(), mlir::Builder::getDimIdentityMap(), mlir::affine::getLargestDivisorOfTripCount(), mlir::affine::AffineBound::getMap(), mlir::AffineMap::getNumDims(), mlir::affine::AffineBound::getNumOperands(), mlir::AffineMap::getNumResults(), mlir::AffineMap::getNumSymbols(), mlir::affine::AffineBound::getOperand(), mlir::AffineMap::getResults(), and mlir::Builder::getSingleDimShiftAffineMap().

Referenced by mlir::affine::tilePerfectlyNested().

◆ constructTiledLoopNest()

static void constructTiledLoopNest	(	MutableArrayRef< AffineForOp >	origLoops,
		AffineForOp	rootAffineForOp,
		unsigned	width,
		MutableArrayRef< AffineForOp >	tiledLoops
	)

static

Constructs tiled loop nest, without setting the loop bounds and move the body of the original loop nest to the tiled loop nest.

Definition at line 431 of file LoopUtils.cpp.

References mlir::OpBuilder::create(), mlir::Operation::getBlock(), mlir::Block::getOperations(), and moveLoopBody().

Referenced by mlir::affine::tilePerfectlyNested(), and mlir::affine::tilePerfectlyNestedParametric().

◆ createFullTiles()

static LogicalResult createFullTiles	(	MutableArrayRef< AffineForOp >	inputNest,
		SmallVectorImpl< AffineForOp > &	fullTileLoops,
		OpBuilder	b
	)

static

Create the full tile loop nest (along with its body).

Definition at line 2660 of file LoopUtils.cpp.

References mlir::OpBuilder::atBlockTerminator(), mlir::affine::createCanonicalizedAffineForOp(), mlir::detail::enumerate(), mlir::failure(), mlir::affine::AffineValueMap::getAffineMap(), mlir::presburger::IntegerRelation::getConstantBoundOnDimSize(), mlir::Builder::getContext(), mlir::affine::getIndexSet(), mlir::affine::FlatAffineValueConstraints::getIneqAsAffineValueMap(), mlir::presburger::IntegerRelation::getNumDimAndSymbolVars(), mlir::affine::AffineValueMap::getOperands(), mlir::IRMapping::map(), and mlir::presburger::IntegerRelation::setDimSymbolSeparation().

Referenced by mlir::affine::separateFullTiles().

◆ createSeparationCondition()

static AffineIfOp createSeparationCondition	(	MutableArrayRef< AffineForOp >	loops,
		OpBuilder	b
	)

static

Creates an AffineIfOp that encodes the conditional to choose between the constant trip count version and an unknown trip count version of this nest of loops.

This is used to separate partial and full tiles if loops has the intra-tile loops. The affine.if op is inserted at the builder insertion point of b.

Definition at line 2580 of file LoopUtils.cpp.

Referenced by mlir::affine::separateFullTiles().

◆ emitRemarkForBlock()

static InFlightDiagnostic LLVM_ATTRIBUTE_UNUSED emitRemarkForBlock ( Block & block )

static

Definition at line 1953 of file LoopUtils.cpp.

References mlir::Operation::emitRemark(), and mlir::Block::getParentOp().

Referenced by generateCopy().

◆ findHighestBlockForPlacement()

static void findHighestBlockForPlacement	(	const MemRefRegion &	region,
		Block &	block,
		Block::iterator &	begin,
		Block::iterator &	end,
		Block **	copyPlacementBlock,
		Block::iterator *	copyInPlacementStart,
		Block::iterator *	copyOutPlacementStart
	)

static

Given a memref region, determine the lowest depth at which transfers can be placed for it, and return the corresponding block, start and end positions in the block for placing incoming (read) and outgoing (write) copies respectively.

The lowest depth depends on whether the region being accessed is hoistable with respect to one or more immediately surrounding loops.

Definition at line 1801 of file LoopUtils.cpp.

References mlir::Block::begin(), mlir::affine::getAffineForIVs(), and mlir::affine::MemRefRegion::getConstraints().

Referenced by mlir::affine::affineDataCopyGenerate().

◆ gatherLoopsInBlock()

static void gatherLoopsInBlock	(	Block *	block,
		unsigned	currLoopDepth,
		std::vector< SmallVector< AffineForOp, 2 >> &	depthToLoops
	)

static

Gathers all AffineForOps in 'block' at 'currLoopDepth' in 'depthToLoops'.

Definition at line 2530 of file LoopUtils.cpp.

Referenced by mlir::affine::gatherLoops().

◆ generateCleanupLoopForUnroll()

static LogicalResult generateCleanupLoopForUnroll	(	AffineForOp	forOp,
		uint64_t	unrollFactor
	)

static

Helper to generate cleanup loop for unroll or unroll-and-jam when the trip count is not a multiple of unrollFactor.

Definition at line 981 of file LoopUtils.cpp.

References mlir::OpBuilder::clone(), mlir::failure(), getCleanupLoopLowerBound(), mlir::affine::promoteIfSingleIteration(), and mlir::success().

Referenced by mlir::affine::loopUnrollByFactor(), and mlir::affine::loopUnrollJamByFactor().

◆ generateCopy()

static LogicalResult generateCopy	(	const MemRefRegion &	region,
		Block *	block,
		Block::iterator	begin,
		Block::iterator	end,
		Block *	copyPlacementBlock,
		Block::iterator	copyInPlacementStart,
		Block::iterator	copyOutPlacementStart,
		const AffineCopyOptions &	copyOptions,
		DenseMap< Value, Value > &	fastBufferMap,
		DenseSet< Operation * > &	copyNests,
		uint64_t *	sizeInBytes,
		Block::iterator *	nBegin,
		Block::iterator *	nEnd
	)

static

Creates a buffer in the faster memory space for the specified memref region; generates a copy from the lower memory space to this one, and replaces all loads/stores in the block range [‘begin’, ‘end’) of ‘block’ to load/store from that buffer.

Returns failure if copies could not be generated due to yet unimplemented cases. copyInPlacementStart and copyOutPlacementStart in copyPlacementBlock specify the insertion points where the incoming copies and outgoing copies, respectively, should be inserted (the insertion happens right before the insertion point). Since begin can itself be invalidated due to the memref rewriting done from this method, the output argument nBegin is set to its replacement (set to begin if no invalidation happens). Since outgoing copies could have been inserted at end, the output argument nEnd is set to the new end. sizeInBytes is set to the size of the fast buffer allocated.

Definition at line 1970 of file LoopUtils.cpp.

References mlir::OpBuilder::create(), mlir::detail::divideCeil(), emitRemarkForBlock(), mlir::failure(), mlir::affine::AffineCopyOptions::fastMemorySpace, mlir::floorDiv(), mlir::affine::fullyComposeAffineMapAndOperands(), mlir::affine::AffineCopyOptions::generateDma, generatePointWiseCopy(), mlir::AffineMap::get(), mlir::get(), mlir::affine::MemRefRegion::getConstantBoundingSizeAndShape(), mlir::affine::MemRefRegion::getConstraints(), mlir::affine::getIntOrFloatMemRefSizeInBytes(), mlir::affine::MemRefRegion::getLowerAndUpperBound(), mlir::Builder::getMultiDimIdentityMap(), getMultiLevelStrides(), mlir::presburger::IntegerRelation::getNumCols(), mlir::presburger::IntegerRelation::getNumDimVars(), mlir::presburger::IntegerRelation::getNumSymbolVars(), mlir::presburger::IntegerRelation::getNumVars(), mlir::Block::getParent(), mlir::Region::getParentOfType(), mlir::FlatLinearValueConstraints::getValues(), mlir::affine::MemRefRegion::isWrite(), mlir::affine::MemRefRegion::loc, mlir::affine::MemRefRegion::memref, mlir::success(), and mlir::affine::AffineCopyOptions::tagMemorySpace.

Referenced by mlir::affine::affineDataCopyGenerate(), and mlir::affine::generateCopyForMemRegion().

◆ generatePointWiseCopy()

static AffineForOp generatePointWiseCopy	(	Location	loc,
		Value	memref,
		Value	fastMemRef,
		ArrayRef< AffineMap >	lbMaps,
		ArrayRef< Value >	lbOperands,
		ArrayRef< AffineMap >	ubMaps,
		ArrayRef< Value >	ubOperands,
		ArrayRef< AffineExpr >	fastBufOffsets,
		bool	isCopyOut,
		OpBuilder	b
	)

static

Generates a point-wise copy from/to ‘memref’ to/from ‘fastMemRef’ and returns the outermost AffineForOp of the copy loop nest.

lbMaps and ubMaps along with lbOperands and ubOperands hold the lower and upper bound information for the copy loop nest. fastBufOffsets contain the expressions to be subtracted out from the respective copy loop iterators in order to index the fast buffer. If ‘copyOut’ is true, generates a copy-out; otherwise a copy-in. Builder b should be set to the point the copy nest is inserted. The copy-in nest is generated as follows as an example for a 2-d region: for x = ... for y = ... fast_buf[x - offset_x][y - offset_y] = memref[x][y]

Definition at line 1881 of file LoopUtils.cpp.

References mlir::OpBuilder::atBlockTerminator(), mlir::affine::canonicalizeMapAndOperands(), mlir::OpBuilder::create(), mlir::affine::createCanonicalizedAffineForOp(), mlir::affine::fullyComposeAffineMapAndOperands(), mlir::AffineMap::get(), mlir::Builder::getAffineDimExpr(), mlir::Builder::getContext(), mlir::AffineMap::getNumInputs(), mlir::Value::getType(), and mlir::simplifyAffineMap().

Referenced by generateCopy().

◆ generateShiftedLoop()

static AffineForOp generateShiftedLoop	(	AffineMap	lbMap,
		AffineMap	ubMap,
		const std::vector< std::pair< uint64_t, ArrayRef< Operation * >>> &	opGroupQueue,
		unsigned	offset,
		AffineForOp	srcForOp,
		OpBuilder	b
	)

static

Generates an affine.for op with the specified lower and upper bounds while generating the right IV remappings to realize shifts for operations in its body.

The operations that go into the loop body are specified in opGroupQueue starting from the specified offset, and in that order. The first element of the pair specifies the shift applied to that group of operations; the shift is multiplied by the loop step before being applied. Returns nullptr if the generated loop simplifies to a single iteration one.

Definition at line 182 of file LoopUtils.cpp.

References mlir::OpBuilder::atBlockTerminator(), mlir::OpBuilder::create(), mlir::AffineMap::getNumInputs(), and mlir::IRMapping::map().

◆ generateUnrolledLoop()

static void generateUnrolledLoop	(	Block *	loopBodyBlock,
		Value	forOpIV,
		uint64_t	unrollFactor,
		function_ref< Value(unsigned, Value, OpBuilder)>	ivRemapFn,
		function_ref< void(unsigned, Operation *, OpBuilder)>	annotateFn,
		ValueRange	iterArgs,
		ValueRange	yieldedValues
	)

static

Generates unrolled copies of AffineForOp 'loopBodyBlock', with associated 'forOpIV' by 'unrollFactor', calling 'ivRemapFn' to remap 'forOpIV' for each unrolled body.

If specified, annotates the Ops in each unrolled iteration using annotateFn.

Definition at line 921 of file LoopUtils.cpp.

References mlir::OpBuilder::atBlockTerminator(), mlir::Block::begin(), mlir::Operation::clone(), mlir::Block::end(), mlir::Operation::getBlock(), mlir::Block::getTerminator(), mlir::IRMapping::lookup(), mlir::IRMapping::map(), mlir::Operation::setOperands(), and mlir::Value::use_empty().

Referenced by mlir::affine::loopUnrollByFactor().

◆ getCleanupLoopLowerBound()

static void getCleanupLoopLowerBound	(	AffineForOp	forOp,
		unsigned	unrollFactor,
		AffineMap &	cleanupLbMap,
		SmallVectorImpl< Value > &	cleanupLbOperands
	)

static

Computes the cleanup loop lower bound of the loop being unrolled with the specified unroll factor; this bound will also be upper bound of the main part of the unrolled loop.

Computes the bound as an AffineMap with its operands or a null map when the trip count can't be expressed as an affine expression.

Definition at line 57 of file LoopUtils.cpp.

References mlir::affine::canonicalizeMapAndOperands(), mlir::OpBuilder::create(), mlir::affine::fullyComposeAffineMapAndOperands(), mlir::AffineMap::get(), mlir::Builder::getAffineDimExpr(), mlir::Builder::getContext(), mlir::AffineMap::getNumDims(), mlir::AffineMap::getNumResults(), mlir::AffineMap::getNumSymbols(), mlir::AffineMap::getResult(), mlir::affine::getTripCountMapAndOperands(), and mlir::simplifyAffineMap().

Referenced by generateCleanupLoopForUnroll().

◆ getFullMemRefAsRegion()

static bool getFullMemRefAsRegion	(	Operation *	op,
		unsigned	numParamLoopIVs,
		MemRefRegion *	region
	)

static

Construct the memref region to just include the entire memref.

Returns false dynamic shaped memref's for now. numParamLoopIVs is the number of enclosing loop IVs of op (starting from the outermost) that the region is parametric on.

Definition at line 2263 of file LoopUtils.cpp.

Referenced by mlir::affine::affineDataCopyGenerate().

◆ getMultiLevelStrides()

static void getMultiLevelStrides	(	const MemRefRegion &	region,
		ArrayRef< int64_t >	bufferShape,
		SmallVectorImpl< StrideInfo > *	strideInfos
	)

static

Returns striding information for a copy/transfer of this region with potentially multiple striding levels from outermost to innermost.

For an n-dimensional region, there can be at most n-1 levels of striding successively nested.

Definition at line 1845 of file LoopUtils.cpp.

References mlir::Value::getType(), and mlir::affine::MemRefRegion::memref.

Referenced by generateCopy().

◆ moveLoopBody()

static void moveLoopBody	(	AffineForOp	src,
		AffineForOp	dest
	)

static

Move the loop body of AffineForOp 'src' from 'src' to the start of dest body.

Definition at line 425 of file LoopUtils.cpp.

References moveLoopBodyImpl().

Referenced by constructTiledLoopNest().

◆ moveLoopBodyImpl()

static void moveLoopBodyImpl	(	AffineForOp	src,
		AffineForOp	dest,
		Block::iterator	loc
	)

static

Move the loop body of AffineForOp 'src' from 'src' into the specified location in destination's body, ignoring the terminator.

Definition at line 416 of file LoopUtils.cpp.

Referenced by moveLoopBody().

◆ performPreTilingChecks()

template<typename t >

static LogicalResult performPreTilingChecks	(	MutableArrayRef< AffineForOp >	input,
		ArrayRef< t >	tileSizes
	)

static

Check if the input nest is supported for tiling and whether tiling would be legal or not.

Definition at line 390 of file LoopUtils.cpp.

Referenced by mlir::affine::tilePerfectlyNested(), and mlir::affine::tilePerfectlyNestedParametric().

◆ replaceIterArgsAndYieldResults()

static void replaceIterArgsAndYieldResults ( AffineForOp forOp )

static

Helper to replace uses of loop carried values (iter_args) and loop yield values while promoting single iteration affine.for ops.

Definition at line 115 of file LoopUtils.cpp.

Referenced by mlir::affine::promoteIfSingleIteration().

◆ setInterTileBoundsParametric()

static void setInterTileBoundsParametric	(	OpBuilder &	b,
		AffineForOp	origLoop,
		AffineForOp	newLoop,
		Value	tileSize
	)

static

Set lower and upper bounds of inter-tile loops for parametric tiling.

Definition at line 567 of file LoopUtils.cpp.

References mlir::ceilDiv(), mlir::AffineMap::get(), mlir::Builder::getAffineConstantExpr(), mlir::Builder::getAffineSymbolExpr(), mlir::Builder::getContext(), mlir::affine::AffineBound::getMap(), mlir::AffineMap::getNumDims(), mlir::affine::AffineBound::getNumOperands(), mlir::AffineMap::getNumResults(), mlir::AffineMap::getNumSymbols(), mlir::affine::AffineBound::getOperand(), and mlir::AffineMap::getResults().

Referenced by constructParametricallyTiledIndexSetHyperRect().

◆ setIntraTileBoundsParametric()

static void setIntraTileBoundsParametric	(	OpBuilder &	b,
		AffineForOp	origLoop,
		AffineForOp	newInterTileLoop,
		AffineForOp	newIntraTileLoop,
		Value	tileSize
	)

static

Set lower and upper bounds of intra-tile loops for parametric tiling.

Definition at line 472 of file LoopUtils.cpp.

References mlir::AffineMap::get(), mlir::Builder::getAffineConstantExpr(), mlir::Builder::getAffineDimExpr(), mlir::Builder::getAffineSymbolExpr(), mlir::Builder::getContext(), mlir::affine::AffineBound::getMap(), mlir::AffineMap::getNumDims(), mlir::affine::AffineBound::getNumOperands(), mlir::AffineMap::getNumResults(), mlir::AffineMap::getNumSymbols(), mlir::affine::AffineBound::getOperand(), and mlir::AffineMap::getResults().

Referenced by constructParametricallyTiledIndexSetHyperRect().

◆ stripmineSink() [1/2]

template<typename SizeType >

static AffineForOp stripmineSink	(	AffineForOp	forOp,
		SizeType	factor,
		AffineForOp	target
	)

static

Definition at line 1612 of file LoopUtils.cpp.

References stripmineSink().

◆ stripmineSink() [2/2]

static SmallVector<AffineForOp, 8> stripmineSink	(	AffineForOp	forOp,
		uint64_t	factor,
		ArrayRef< AffineForOp >	targets
	)

static

Definition at line 1569 of file LoopUtils.cpp.

References mlir::OpBuilder::atBlockTerminator(), augmentMapAndBounds(), mlir::OpBuilder::create(), and mlir::replaceAllUsesInRegionWith().

Referenced by stripmineSink(), and mlir::affine::tile().

Classes

Macros

Functions

Macro Definition Documentation

◆ DEBUG_TYPE

Function Documentation

◆ areInnerBoundsInvariant()

◆ augmentMapAndBounds()

◆ checkIfHyperRectangular()

◆ checkLoopInterchangeDependences()

◆ constructParametricallyTiledIndexSetHyperRect()

◆ constructTiledIndexSetHyperRect()

◆ constructTiledLoopNest()

◆ createFullTiles()

◆ createSeparationCondition()

◆ emitRemarkForBlock()

◆ findHighestBlockForPlacement()

◆ gatherLoopsInBlock()

◆ generateCleanupLoopForUnroll()

◆ generateCopy()

◆ generatePointWiseCopy()

◆ generateShiftedLoop()

◆ generateUnrolledLoop()

◆ getCleanupLoopLowerBound()

◆ getFullMemRefAsRegion()

◆ getMultiLevelStrides()

◆ moveLoopBody()

◆ moveLoopBodyImpl()

◆ performPreTilingChecks()

◆ replaceIterArgsAndYieldResults()

◆ setInterTileBoundsParametric()

◆ setIntraTileBoundsParametric()

◆ stripmineSink() [1/2]

◆ stripmineSink() [2/2]