Google Git
Sign in
fuchsia / third_party / llvm-project / a336055ddb4290b953871d1714159de4b70670a8 / . / mlir / lib / Dialect / MemRef / IR / MemRefOps.cpp
blob: 836dcb8f329e701edc6a689b2c5947e1753f7c3d [file] [log] [blame]
//===----------------------------------------------------------------------===//
//
// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
// See https://llvm.org/LICENSE.txt for license information.
// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
//
//===----------------------------------------------------------------------===//
#include "mlir/Dialect/Arith/IR/Arith.h"
#include "mlir/Dialect/Arith/Utils/Utils.h"
#include "mlir/Dialect/MemRef/IR/MemRef.h"
#include "mlir/Dialect/Utils/StaticValueUtils.h"
#include "mlir/IR/AffineMap.h"
#include "mlir/IR/Builders.h"
#include "mlir/IR/BuiltinTypes.h"
#include "mlir/IR/Matchers.h"
#include "mlir/IR/OpDefinition.h"
#include "mlir/IR/PatternMatch.h"
#include "mlir/IR/TypeUtilities.h"
#include "mlir/Interfaces/InferTypeOpInterface.h"
#include "mlir/Interfaces/SideEffectInterfaces.h"
#include "mlir/Interfaces/ViewLikeInterface.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/ADT/SmallBitVector.h"
using namespace mlir;
using namespace mlir::memref;
/// Materialize a single constant operation from a given attribute value with
/// the desired resultant type.
Operation *MemRefDialect::materializeConstant(OpBuilder &builder,
Attribute value, Type type,
Location loc) {
return arith::ConstantOp::materialize(builder, value, type, loc);
}
//===----------------------------------------------------------------------===//
// Common canonicalization pattern support logic
//===----------------------------------------------------------------------===//
/// This is a common class used for patterns of the form
/// "someop(memrefcast) -> someop". It folds the source of any memref.cast
/// into the root operation directly.
LogicalResult mlir::memref::foldMemRefCast(Operation *op, Value inner) {
bool folded = false;
for (OpOperand &operand : op->getOpOperands()) {
auto cast = operand.get().getDefiningOp<CastOp>();
if (cast && operand.get() != inner &&
!llvm::isa<UnrankedMemRefType>(cast.getOperand().getType())) {
operand.set(cast.getOperand());
folded = true;
}
}
return success(folded);
}
/// Return an unranked/ranked tensor type for the given unranked/ranked memref
/// type.
Type mlir::memref::getTensorTypeFromMemRefType(Type type) {
if (auto memref = llvm::dyn_cast<MemRefType>(type))
return RankedTensorType::get(memref.getShape(), memref.getElementType());
if (auto memref = llvm::dyn_cast<UnrankedMemRefType>(type))
return UnrankedTensorType::get(memref.getElementType());
return NoneType::get(type.getContext());
}
OpFoldResult memref::getMixedSize(OpBuilder &builder, Location loc, Value value,
int64_t dim) {
auto memrefType = llvm::cast<MemRefType>(value.getType());
SmallVector<OpFoldResult> result;
if (memrefType.isDynamicDim(dim))
return builder.createOrFold<memref::DimOp>(loc, value, dim);
return builder.getIndexAttr(memrefType.getDimSize(dim));
}
SmallVector<OpFoldResult> memref::getMixedSizes(OpBuilder &builder,
Location loc, Value value) {
auto memrefType = llvm::cast<MemRefType>(value.getType());
SmallVector<OpFoldResult> result;
for (int64_t i = 0; i < memrefType.getRank(); ++i)
result.push_back(getMixedSize(builder, loc, value, i));
return result;
}
//===----------------------------------------------------------------------===//
// Utility functions for propagating static information
//===----------------------------------------------------------------------===//
/// Helper function that infers the constant values from a list of \p values,
/// a \p memRefTy, and another helper function \p getAttributes.
/// The inferred constant values replace the related `OpFoldResult` in
/// \p values.
///
/// \note This function shouldn't be used directly, instead, use the
/// `getConstifiedMixedXXX` methods from the related operations.
///
/// \p getAttributes retuns a list of potentially constant values, as determined
/// by \p isDynamic, from the given \p memRefTy. The returned list must have as
/// many elements as \p values or be empty.
///
/// E.g., consider the following example:
/// ```
/// memref.reinterpret_cast %base to <...> strides: [2, %dyn_stride] :
/// memref<f32> to memref<?x?xf32, strided<[?, 1], offset: ?>>
/// ```
/// `ReinterpretCastOp::getMixedStrides()` will return `[2, %dyn_stride]`.
/// Now using this helper function with:
/// - `values == [2, %dyn_stride]`,
/// - `memRefTy == memref<?x?xf32, strided<[?, 1], offset: ?>>`
/// - `getAttributes == getConstantStrides` (i.e., a wrapper around
/// `getStridesAndOffset`), and
/// - `isDynamic == ShapedType::isDynamic`
/// Will yield: `values == [2, 1]`
static void constifyIndexValues(
SmallVectorImpl<OpFoldResult> &values, MemRefType memRefTy,
MLIRContext *ctxt,
llvm::function_ref<SmallVector<int64_t>(MemRefType)> getAttributes,
llvm::function_ref<bool(int64_t)> isDynamic) {
SmallVector<int64_t> constValues = getAttributes(memRefTy);
Builder builder(ctxt);
for (const auto &it : llvm::enumerate(constValues)) {
int64_t constValue = it.value();
if (!isDynamic(constValue))
values[it.index()] = builder.getIndexAttr(constValue);
}
for (OpFoldResult &ofr : values) {
if (ofr.is<Attribute>()) {
// FIXME: We shouldn't need to do that, but right now, the static indices
// are created with the wrong type: `i64` instead of `index`.
// As a result, if we were to keep the attribute as is, we may fail to see
// that two attributes are equal because one would have the i64 type and
// the other the index type.
// The alternative would be to create constant indices with getI64Attr in
// this and the previous loop, but it doesn't logically make sense (we are
// dealing with indices here) and would only strenghten the inconsistency
// around how static indices are created (some places use getI64Attr,
// others use getIndexAttr).
// The workaround here is to stick to the IndexAttr type for all the
// values, hence we recreate the attribute even when it is already static
// to make sure the type is consistent.
ofr = builder.getIndexAttr(
llvm::cast<IntegerAttr>(ofr.get<Attribute>()).getInt());
continue;
}
std::optional<int64_t> maybeConstant =
getConstantIntValue(ofr.get<Value>());
if (maybeConstant)
ofr = builder.getIndexAttr(*maybeConstant);
}
}
/// Wrapper around `getShape` that conforms to the function signature
/// expected for `getAttributes` in `constifyIndexValues`.
static SmallVector<int64_t> getConstantSizes(MemRefType memRefTy) {
ArrayRef<int64_t> sizes = memRefTy.getShape();
return SmallVector<int64_t>(sizes.begin(), sizes.end());
}
/// Wrapper around `getStridesAndOffset` that returns only the offset and
/// conforms to the function signature expected for `getAttributes` in
/// `constifyIndexValues`.
static SmallVector<int64_t> getConstantOffset(MemRefType memrefType) {
SmallVector<int64_t> strides;
int64_t offset;
LogicalResult hasStaticInformation =
getStridesAndOffset(memrefType, strides, offset);
if (failed(hasStaticInformation))
return SmallVector<int64_t>();
return SmallVector<int64_t>(1, offset);
}
/// Wrapper around `getStridesAndOffset` that returns only the strides and
/// conforms to the function signature expected for `getAttributes` in
/// `constifyIndexValues`.
static SmallVector<int64_t> getConstantStrides(MemRefType memrefType) {
SmallVector<int64_t> strides;
int64_t offset;
LogicalResult hasStaticInformation =
getStridesAndOffset(memrefType, strides, offset);
if (failed(hasStaticInformation))
return SmallVector<int64_t>();
return strides;
}
//===----------------------------------------------------------------------===//
// AllocOp / AllocaOp
//===----------------------------------------------------------------------===//
void AllocOp::getAsmResultNames(
function_ref<void(Value, StringRef)> setNameFn) {
setNameFn(getResult(), "alloc");
}
void AllocaOp::getAsmResultNames(
function_ref<void(Value, StringRef)> setNameFn) {
setNameFn(getResult(), "alloca");
}
template <typename AllocLikeOp>
static LogicalResult verifyAllocLikeOp(AllocLikeOp op) {
static_assert(llvm::is_one_of<AllocLikeOp, AllocOp, AllocaOp>::value,
"applies to only alloc or alloca");
auto memRefType = llvm::dyn_cast<MemRefType>(op.getResult().getType());
if (!memRefType)
return op.emitOpError("result must be a memref");
if (static_cast<int64_t>(op.getDynamicSizes().size()) !=
memRefType.getNumDynamicDims())
return op.emitOpError("dimension operand count does not equal memref "
"dynamic dimension count");
unsigned numSymbols = 0;
if (!memRefType.getLayout().isIdentity())
numSymbols = memRefType.getLayout().getAffineMap().getNumSymbols();
if (op.getSymbolOperands().size() != numSymbols)
return op.emitOpError("symbol operand count does not equal memref symbol "
"count: expected ")
<< numSymbols << ", got " << op.getSymbolOperands().size();
return success();
}
LogicalResult AllocOp::verify() { return verifyAllocLikeOp(*this); }
LogicalResult AllocaOp::verify() {
// An alloca op needs to have an ancestor with an allocation scope trait.
if (!(*this)->getParentWithTrait<OpTrait::AutomaticAllocationScope>())
return emitOpError(
"requires an ancestor op with AutomaticAllocationScope trait");
return verifyAllocLikeOp(*this);
}
namespace {
/// Fold constant dimensions into an alloc like operation.
template <typename AllocLikeOp>
struct SimplifyAllocConst : public OpRewritePattern<AllocLikeOp> {
using OpRewritePattern<AllocLikeOp>::OpRewritePattern;
LogicalResult matchAndRewrite(AllocLikeOp alloc,
PatternRewriter &rewriter) const override {
// Check to see if any dimensions operands are constants. If so, we can
// substitute and drop them.
if (llvm::none_of(alloc.getDynamicSizes(), [](Value operand) {
APInt constSizeArg;
if (!matchPattern(operand, m_ConstantInt(&constSizeArg)))
return false;
return constSizeArg.isNonNegative();
}))
return failure();
auto memrefType = alloc.getType();
// Ok, we have one or more constant operands. Collect the non-constant ones
// and keep track of the resultant memref type to build.
SmallVector<int64_t, 4> newShapeConstants;
newShapeConstants.reserve(memrefType.getRank());
SmallVector<Value, 4> dynamicSizes;
unsigned dynamicDimPos = 0;
for (unsigned dim = 0, e = memrefType.getRank(); dim < e; ++dim) {
int64_t dimSize = memrefType.getDimSize(dim);
// If this is already static dimension, keep it.
if (!ShapedType::isDynamic(dimSize)) {
newShapeConstants.push_back(dimSize);
continue;
}
auto dynamicSize = alloc.getDynamicSizes()[dynamicDimPos];
APInt constSizeArg;
if (matchPattern(dynamicSize, m_ConstantInt(&constSizeArg)) &&
constSizeArg.isNonNegative()) {
// Dynamic shape dimension will be folded.
newShapeConstants.push_back(constSizeArg.getZExtValue());
} else {
// Dynamic shape dimension not folded; copy dynamicSize from old memref.
newShapeConstants.push_back(ShapedType::kDynamic);
dynamicSizes.push_back(dynamicSize);
}
dynamicDimPos++;
}
// Create new memref type (which will have fewer dynamic dimensions).
MemRefType newMemRefType =
MemRefType::Builder(memrefType).setShape(newShapeConstants);
assert(static_cast<int64_t>(dynamicSizes.size()) ==
newMemRefType.getNumDynamicDims());
// Create and insert the alloc op for the new memref.
auto newAlloc = rewriter.create<AllocLikeOp>(
alloc.getLoc(), newMemRefType, dynamicSizes, alloc.getSymbolOperands(),
alloc.getAlignmentAttr());
// Insert a cast so we have the same type as the old alloc.
rewriter.replaceOpWithNewOp<CastOp>(alloc, alloc.getType(), newAlloc);
return success();
}
};
/// Fold alloc operations with no users or only store and dealloc uses.
template <typename T>
struct SimplifyDeadAlloc : public OpRewritePattern<T> {
using OpRewritePattern<T>::OpRewritePattern;
LogicalResult matchAndRewrite(T alloc,
PatternRewriter &rewriter) const override {
if (llvm::any_of(alloc->getUsers(), [&](Operation *op) {
if (auto storeOp = dyn_cast<StoreOp>(op))
return storeOp.getValue() == alloc;
return !isa<DeallocOp>(op);
}))
return failure();
for (Operation *user : llvm::make_early_inc_range(alloc->getUsers()))
rewriter.eraseOp(user);
rewriter.eraseOp(alloc);
return success();
}
};
} // namespace
void AllocOp::getCanonicalizationPatterns(RewritePatternSet &results,
MLIRContext *context) {
results.add<SimplifyAllocConst<AllocOp>, SimplifyDeadAlloc<AllocOp>>(context);
}
void AllocaOp::getCanonicalizationPatterns(RewritePatternSet &results,
MLIRContext *context) {
results.add<SimplifyAllocConst<AllocaOp>, SimplifyDeadAlloc<AllocaOp>>(
context);
}
//===----------------------------------------------------------------------===//
// ReallocOp
//===----------------------------------------------------------------------===//
LogicalResult ReallocOp::verify() {
auto sourceType = llvm::cast<MemRefType>(getOperand(0).getType());
MemRefType resultType = getType();
// The source memref should have identity layout (or none).
if (!sourceType.getLayout().isIdentity())
return emitError("unsupported layout for source memref type ")
<< sourceType;
// The result memref should have identity layout (or none).
if (!resultType.getLayout().isIdentity())
return emitError("unsupported layout for result memref type ")
<< resultType;
// The source memref and the result memref should be in the same memory space.
if (sourceType.getMemorySpace() != resultType.getMemorySpace())
return emitError("different memory spaces specified for source memref "
"type ")
<< sourceType << " and result memref type " << resultType;
// The source memref and the result memref should have the same element type.
if (sourceType.getElementType() != resultType.getElementType())
return emitError("different element types specified for source memref "
"type ")
<< sourceType << " and result memref type " << resultType;
// Verify that we have the dynamic dimension operand when it is needed.
if (resultType.getNumDynamicDims() && !getDynamicResultSize())
return emitError("missing dimension operand for result type ")
<< resultType;
if (!resultType.getNumDynamicDims() && getDynamicResultSize())
return emitError("unnecessary dimension operand for result type ")
<< resultType;
return success();
}
void ReallocOp::getCanonicalizationPatterns(RewritePatternSet &results,
MLIRContext *context) {
results.add<SimplifyDeadAlloc<ReallocOp>>(context);
}
//===----------------------------------------------------------------------===//
// AllocaScopeOp
//===----------------------------------------------------------------------===//
void AllocaScopeOp::print(OpAsmPrinter &p) {
bool printBlockTerminators = false;
p << ' ';
if (!getResults().empty()) {
p << " -> (" << getResultTypes() << ")";
printBlockTerminators = true;
}
p << ' ';
p.printRegion(getBodyRegion(),
/*printEntryBlockArgs=*/false,
/*printBlockTerminators=*/printBlockTerminators);
p.printOptionalAttrDict((*this)->getAttrs());
}
ParseResult AllocaScopeOp::parse(OpAsmParser &parser, OperationState &result) {
// Create a region for the body.
result.regions.reserve(1);
Region *bodyRegion = result.addRegion();
// Parse optional results type list.
if (parser.parseOptionalArrowTypeList(result.types))
return failure();
// Parse the body region.
if (parser.parseRegion(*bodyRegion, /*arguments=*/{}))
return failure();
AllocaScopeOp::ensureTerminator(*bodyRegion, parser.getBuilder(),
result.location);
// Parse the optional attribute list.
if (parser.parseOptionalAttrDict(result.attributes))
return failure();
return success();
}
void AllocaScopeOp::getSuccessorRegions(
RegionBranchPoint point, SmallVectorImpl<RegionSuccessor> &regions) {
if (!point.isParent()) {
regions.push_back(RegionSuccessor(getResults()));
return;
}
regions.push_back(RegionSuccessor(&getBodyRegion()));
}
/// Given an operation, return whether this op is guaranteed to
/// allocate an AutomaticAllocationScopeResource
static bool isGuaranteedAutomaticAllocation(Operation *op) {
MemoryEffectOpInterface interface = dyn_cast<MemoryEffectOpInterface>(op);
if (!interface)
return false;
for (auto res : op->getResults()) {
if (auto effect =
interface.getEffectOnValue<MemoryEffects::Allocate>(res)) {
if (isa<SideEffects::AutomaticAllocationScopeResource>(
effect->getResource()))
return true;
}
}
return false;
}
/// Given an operation, return whether this op itself could
/// allocate an AutomaticAllocationScopeResource. Note that
/// this will not check whether an operation contained within
/// the op can allocate.
static bool isOpItselfPotentialAutomaticAllocation(Operation *op) {
// This op itself doesn't create a stack allocation,
// the inner allocation should be handled separately.
if (op->hasTrait<OpTrait::HasRecursiveMemoryEffects>())
return false;
MemoryEffectOpInterface interface = dyn_cast<MemoryEffectOpInterface>(op);
if (!interface)
return true;
for (auto res : op->getResults()) {
if (auto effect =
interface.getEffectOnValue<MemoryEffects::Allocate>(res)) {
if (isa<SideEffects::AutomaticAllocationScopeResource>(
effect->getResource()))
return true;
}
}
return false;
}
/// Return whether this op is the last non terminating op
/// in a region. That is to say, it is in a one-block region
/// and is only followed by a terminator. This prevents
/// extending the lifetime of allocations.
static bool lastNonTerminatorInRegion(Operation *op) {
return op->getNextNode() == op->getBlock()->getTerminator() &&
op->getParentRegion()->getBlocks().size() == 1;
}
/// Inline an AllocaScopeOp if either the direct parent is an allocation scope
/// or it contains no allocation.
struct AllocaScopeInliner : public OpRewritePattern<AllocaScopeOp> {
using OpRewritePattern<AllocaScopeOp>::OpRewritePattern;
LogicalResult matchAndRewrite(AllocaScopeOp op,
PatternRewriter &rewriter) const override {
bool hasPotentialAlloca =
op->walk<WalkOrder::PreOrder>([&](Operation *alloc) {
if (alloc == op)
return WalkResult::advance();
if (isOpItselfPotentialAutomaticAllocation(alloc))
return WalkResult::interrupt();
if (alloc->hasTrait<OpTrait::AutomaticAllocationScope>())
return WalkResult::skip();
return WalkResult::advance();
}).wasInterrupted();
// If this contains no potential allocation, it is always legal to
// inline. Otherwise, consider two conditions:
if (hasPotentialAlloca) {
// If the parent isn't an allocation scope, or we are not the last
// non-terminator op in the parent, we will extend the lifetime.
if (!op->getParentOp()->hasTrait<OpTrait::AutomaticAllocationScope>())
return failure();
if (!lastNonTerminatorInRegion(op))
return failure();
}
Block *block = &op.getRegion().front();
Operation *terminator = block->getTerminator();
ValueRange results = terminator->getOperands();
rewriter.inlineBlockBefore(block, op);
rewriter.replaceOp(op, results);
rewriter.eraseOp(terminator);
return success();
}
};
/// Move allocations into an allocation scope, if it is legal to
/// move them (e.g. their operands are available at the location
/// the op would be moved to).
struct AllocaScopeHoister : public OpRewritePattern<AllocaScopeOp> {
using OpRewritePattern<AllocaScopeOp>::OpRewritePattern;
LogicalResult matchAndRewrite(AllocaScopeOp op,
PatternRewriter &rewriter) const override {
if (!op->getParentWithTrait<OpTrait::AutomaticAllocationScope>())
return failure();
Operation *lastParentWithoutScope = op->getParentOp();
if (!lastParentWithoutScope ||
lastParentWithoutScope->hasTrait<OpTrait::AutomaticAllocationScope>())
return failure();
// Only apply to if this is this last non-terminator
// op in the block (lest lifetime be extended) of a one
// block region
if (!lastNonTerminatorInRegion(op) ||
!lastNonTerminatorInRegion(lastParentWithoutScope))
return failure();
while (!lastParentWithoutScope->getParentOp()
->hasTrait<OpTrait::AutomaticAllocationScope>()) {
lastParentWithoutScope = lastParentWithoutScope->getParentOp();
if (!lastParentWithoutScope ||
!lastNonTerminatorInRegion(lastParentWithoutScope))
return failure();
}
assert(lastParentWithoutScope->getParentOp()
->hasTrait<OpTrait::AutomaticAllocationScope>());
Region *containingRegion = nullptr;
for (auto &r : lastParentWithoutScope->getRegions()) {
if (r.isAncestor(op->getParentRegion())) {
assert(containingRegion == nullptr &&
"only one region can contain the op");
containingRegion = &r;
}
}
assert(containingRegion && "op must be contained in a region");
SmallVector<Operation *> toHoist;
op->walk([&](Operation *alloc) {
if (!isGuaranteedAutomaticAllocation(alloc))
return WalkResult::skip();
// If any operand is not defined before the location of
// lastParentWithoutScope (i.e. where we would hoist to), skip.
if (llvm::any_of(alloc->getOperands(), [&](Value v) {
return containingRegion->isAncestor(v.getParentRegion());
}))
return WalkResult::skip();
toHoist.push_back(alloc);
return WalkResult::advance();
});
if (toHoist.empty())
return failure();
rewriter.setInsertionPoint(lastParentWithoutScope);
for (auto *op : toHoist) {
auto *cloned = rewriter.clone(*op);
rewriter.replaceOp(op, cloned->getResults());
}
return success();
}
};
void AllocaScopeOp::getCanonicalizationPatterns(RewritePatternSet &results,
MLIRContext *context) {
results.add<AllocaScopeInliner, AllocaScopeHoister>(context);
}
//===----------------------------------------------------------------------===//
// AssumeAlignmentOp
//===----------------------------------------------------------------------===//
LogicalResult AssumeAlignmentOp::verify() {
if (!llvm::isPowerOf2_32(getAlignment()))
return emitOpError("alignment must be power of 2");
return success();
}
//===----------------------------------------------------------------------===//
// CastOp
//===----------------------------------------------------------------------===//
void CastOp::getAsmResultNames(function_ref<void(Value, StringRef)> setNameFn) {
setNameFn(getResult(), "cast");
}
/// Determines whether MemRef_CastOp casts to a more dynamic version of the
/// source memref. This is useful to fold a memref.cast into a consuming op
/// and implement canonicalization patterns for ops in different dialects that
/// may consume the results of memref.cast operations. Such foldable memref.cast
/// operations are typically inserted as `view` and `subview` ops are
/// canonicalized, to preserve the type compatibility of their uses.
///
/// Returns true when all conditions are met:
/// 1. source and result are ranked memrefs with strided semantics and same
/// element type and rank.
/// 2. each of the source's size, offset or stride has more static information
/// than the corresponding result's size, offset or stride.
///
/// Example 1:
/// ```mlir
/// %1 = memref.cast %0 : memref<8x16xf32> to memref<?x?xf32>
/// %2 = consumer %1 ... : memref<?x?xf32> ...
/// ```
///
/// may fold into:
///
/// ```mlir
/// %2 = consumer %0 ... : memref<8x16xf32> ...
/// ```
///
/// Example 2:
/// ```
/// %1 = memref.cast %0 : memref<?x16xf32, affine_map<(i, j)->(16 * i + j)>>
/// to memref<?x?xf32>
/// consumer %1 : memref<?x?xf32> ...
/// ```
///
/// may fold into:
///
/// ```
/// consumer %0 ... : memref<?x16xf32, affine_map<(i, j)->(16 * i + j)>>
/// ```
bool CastOp::canFoldIntoConsumerOp(CastOp castOp) {
MemRefType sourceType =
llvm::dyn_cast<MemRefType>(castOp.getSource().getType());
MemRefType resultType = llvm::dyn_cast<MemRefType>(castOp.getType());
// Requires ranked MemRefType.
if (!sourceType || !resultType)
return false;
// Requires same elemental type.
if (sourceType.getElementType() != resultType.getElementType())
return false;
// Requires same rank.
if (sourceType.getRank() != resultType.getRank())
return false;
// Only fold casts between strided memref forms.
int64_t sourceOffset, resultOffset;
SmallVector<int64_t, 4> sourceStrides, resultStrides;
if (failed(getStridesAndOffset(sourceType, sourceStrides, sourceOffset)) ||
failed(getStridesAndOffset(resultType, resultStrides, resultOffset)))
return false;
// If cast is towards more static sizes along any dimension, don't fold.
for (auto it : llvm::zip(sourceType.getShape(), resultType.getShape())) {
auto ss = std::get<0>(it), st = std::get<1>(it);
if (ss != st)
if (ShapedType::isDynamic(ss) && !ShapedType::isDynamic(st))
return false;
}
// If cast is towards more static offset along any dimension, don't fold.
if (sourceOffset != resultOffset)
if (ShapedType::isDynamic(sourceOffset) &&
!ShapedType::isDynamic(resultOffset))
return false;
// If cast is towards more static strides along any dimension, don't fold.
for (auto it : llvm::zip(sourceStrides, resultStrides)) {
auto ss = std::get<0>(it), st = std::get<1>(it);
if (ss != st)
if (ShapedType::isDynamic(ss) && !ShapedType::isDynamic(st))
return false;
}
return true;
}
bool CastOp::areCastCompatible(TypeRange inputs, TypeRange outputs) {
if (inputs.size() != 1 || outputs.size() != 1)
return false;
Type a = inputs.front(), b = outputs.front();
auto aT = llvm::dyn_cast<MemRefType>(a);
auto bT = llvm::dyn_cast<MemRefType>(b);
auto uaT = llvm::dyn_cast<UnrankedMemRefType>(a);
auto ubT = llvm::dyn_cast<UnrankedMemRefType>(b);
if (aT && bT) {
if (aT.getElementType() != bT.getElementType())
return false;
if (aT.getLayout() != bT.getLayout()) {
int64_t aOffset, bOffset;
SmallVector<int64_t, 4> aStrides, bStrides;
if (failed(getStridesAndOffset(aT, aStrides, aOffset)) ||
failed(getStridesAndOffset(bT, bStrides, bOffset)) ||
aStrides.size() != bStrides.size())
return false;
// Strides along a dimension/offset are compatible if the value in the
// source memref is static and the value in the target memref is the
// same. They are also compatible if either one is dynamic (see
// description of MemRefCastOp for details).
auto checkCompatible = [](int64_t a, int64_t b) {
return (ShapedType::isDynamic(a) || ShapedType::isDynamic(b) || a == b);
};
if (!checkCompatible(aOffset, bOffset))
return false;
for (const auto &aStride : enumerate(aStrides))
if (!checkCompatible(aStride.value(), bStrides[aStride.index()]))
return false;
}
if (aT.getMemorySpace() != bT.getMemorySpace())
return false;
// They must have the same rank, and any specified dimensions must match.
if (aT.getRank() != bT.getRank())
return false;
for (unsigned i = 0, e = aT.getRank(); i != e; ++i) {
int64_t aDim = aT.getDimSize(i), bDim = bT.getDimSize(i);
if (!ShapedType::isDynamic(aDim) && !ShapedType::isDynamic(bDim) &&
aDim != bDim)
return false;
}
return true;
} else {
if (!aT && !uaT)
return false;
if (!bT && !ubT)
return false;
// Unranked to unranked casting is unsupported
if (uaT && ubT)
return false;
auto aEltType = (aT) ? aT.getElementType() : uaT.getElementType();
auto bEltType = (bT) ? bT.getElementType() : ubT.getElementType();
if (aEltType != bEltType)
return false;
auto aMemSpace = (aT) ? aT.getMemorySpace() : uaT.getMemorySpace();
auto bMemSpace = (bT) ? bT.getMemorySpace() : ubT.getMemorySpace();
return aMemSpace == bMemSpace;
}
return false;
}
OpFoldResult CastOp::fold(FoldAdaptor adaptor) {
return succeeded(foldMemRefCast(*this)) ? getResult() : Value();
}
//===----------------------------------------------------------------------===//
// CopyOp
//===----------------------------------------------------------------------===//
namespace {
/// If the source/target of a CopyOp is a CastOp that does not modify the shape
/// and element type, the cast can be skipped. Such CastOps only cast the layout
/// of the type.
struct FoldCopyOfCast : public OpRewritePattern<CopyOp> {
using OpRewritePattern<CopyOp>::OpRewritePattern;
LogicalResult matchAndRewrite(CopyOp copyOp,
PatternRewriter &rewriter) const override {
bool modified = false;
// Check source.
if (auto castOp = copyOp.getSource().getDefiningOp<CastOp>()) {
auto fromType = llvm::dyn_cast<MemRefType>(castOp.getSource().getType());
auto toType = llvm::dyn_cast<MemRefType>(castOp.getSource().getType());
if (fromType && toType) {
if (fromType.getShape() == toType.getShape() &&
fromType.getElementType() == toType.getElementType()) {
rewriter.modifyOpInPlace(copyOp, [&] {
copyOp.getSourceMutable().assign(castOp.getSource());
});
modified = true;
}
}
}
// Check target.
if (auto castOp = copyOp.getTarget().getDefiningOp<CastOp>()) {
auto fromType = llvm::dyn_cast<MemRefType>(castOp.getSource().getType());
auto toType = llvm::dyn_cast<MemRefType>(castOp.getSource().getType());
if (fromType && toType) {
if (fromType.getShape() == toType.getShape() &&
fromType.getElementType() == toType.getElementType()) {
rewriter.modifyOpInPlace(copyOp, [&] {
copyOp.getTargetMutable().assign(castOp.getSource());
});
modified = true;
}
}
}
return success(modified);
}
};
/// Fold memref.copy(%x, %x).
struct FoldSelfCopy : public OpRewritePattern<CopyOp> {
using OpRewritePattern<CopyOp>::OpRewritePattern;
LogicalResult matchAndRewrite(CopyOp copyOp,
PatternRewriter &rewriter) const override {
if (copyOp.getSource() != copyOp.getTarget())
return failure();
rewriter.eraseOp(copyOp);
return success();
}
};
} // namespace
void CopyOp::getCanonicalizationPatterns(RewritePatternSet &results,
MLIRContext *context) {
results.add<FoldCopyOfCast, FoldSelfCopy>(context);
}
LogicalResult CopyOp::fold(FoldAdaptor adaptor,
SmallVectorImpl<OpFoldResult> &results) {
/// copy(memrefcast) -> copy
bool folded = false;
Operation *op = *this;
for (OpOperand &operand : op->getOpOperands()) {
auto castOp = operand.get().getDefiningOp<memref::CastOp>();
if (castOp && memref::CastOp::canFoldIntoConsumerOp(castOp)) {
operand.set(castOp.getOperand());
folded = true;
}
}
return success(folded);
}
//===----------------------------------------------------------------------===//
// DeallocOp
//===----------------------------------------------------------------------===//
LogicalResult DeallocOp::fold(FoldAdaptor adaptor,
SmallVectorImpl<OpFoldResult> &results) {
/// dealloc(memrefcast) -> dealloc
return foldMemRefCast(*this);
}
//===----------------------------------------------------------------------===//
// DimOp
//===----------------------------------------------------------------------===//
void DimOp::getAsmResultNames(function_ref<void(Value, StringRef)> setNameFn) {
setNameFn(getResult(), "dim");
}
void DimOp::build(OpBuilder &builder, OperationState &result, Value source,
int64_t index) {
auto loc = result.location;
Value indexValue = builder.create<arith::ConstantIndexOp>(loc, index);
build(builder, result, source, indexValue);
}
std::optional<int64_t> DimOp::getConstantIndex() {
return getConstantIntValue(getIndex());
}
Speculation::Speculatability DimOp::getSpeculatability() {
auto constantIndex = getConstantIndex();
if (!constantIndex)
return Speculation::NotSpeculatable;
auto rankedSourceType = dyn_cast<MemRefType>(getSource().getType());
if (!rankedSourceType)
return Speculation::NotSpeculatable;
if (rankedSourceType.getRank() <= constantIndex)
return Speculation::NotSpeculatable;
return Speculation::Speculatable;
}
/// Return a map with key being elements in `vals` and data being number of
/// occurences of it. Use std::map, since the `vals` here are strides and the
/// dynamic stride value is the same as the tombstone value for
/// `DenseMap<int64_t>`.
static std::map<int64_t, unsigned> getNumOccurences(ArrayRef<int64_t> vals) {
std::map<int64_t, unsigned> numOccurences;
for (auto val : vals)
numOccurences[val]++;
return numOccurences;
}
/// Given the `originalType` and a `candidateReducedType` whose shape is assumed
/// to be a subset of `originalType` with some `1` entries erased, return the
/// set of indices that specifies which of the entries of `originalShape` are
/// dropped to obtain `reducedShape`.
/// This accounts for cases where there are multiple unit-dims, but only a
/// subset of those are dropped. For MemRefTypes these can be disambiguated
/// using the strides. If a dimension is dropped the stride must be dropped too.
static FailureOr<llvm::SmallBitVector>
computeMemRefRankReductionMask(MemRefType originalType, MemRefType reducedType,
ArrayRef<OpFoldResult> sizes) {
llvm::SmallBitVector unusedDims(originalType.getRank());
if (originalType.getRank() == reducedType.getRank())
return unusedDims;
for (const auto &dim : llvm::enumerate(sizes))
if (auto attr = llvm::dyn_cast_if_present<Attribute>(dim.value()))
if (llvm::cast<IntegerAttr>(attr).getInt() == 1)
unusedDims.set(dim.index());
// Early exit for the case where the number of unused dims matches the number
// of ranks reduced.
if (static_cast<int64_t>(unusedDims.count()) + reducedType.getRank() ==
originalType.getRank())
return unusedDims;
SmallVector<int64_t> originalStrides, candidateStrides;
int64_t originalOffset, candidateOffset;
if (failed(
getStridesAndOffset(originalType, originalStrides, originalOffset)) ||
failed(
getStridesAndOffset(reducedType, candidateStrides, candidateOffset)))
return failure();
// For memrefs, a dimension is truly dropped if its corresponding stride is
// also dropped. This is particularly important when more than one of the dims
// is 1. Track the number of occurences of the strides in the original type
// and the candidate type. For each unused dim that stride should not be
// present in the candidate type. Note that there could be multiple dimensions
// that have the same size. We dont need to exactly figure out which dim
// corresponds to which stride, we just need to verify that the number of
// reptitions of a stride in the original + number of unused dims with that
// stride == number of repititions of a stride in the candidate.
std::map<int64_t, unsigned> currUnaccountedStrides =
getNumOccurences(originalStrides);
std::map<int64_t, unsigned> candidateStridesNumOccurences =
getNumOccurences(candidateStrides);
for (size_t dim = 0, e = unusedDims.size(); dim != e; ++dim) {
if (!unusedDims.test(dim))
continue;
int64_t originalStride = originalStrides[dim];
if (currUnaccountedStrides[originalStride] >
candidateStridesNumOccurences[originalStride]) {
// This dim can be treated as dropped.
currUnaccountedStrides[originalStride]--;
continue;
}
if (currUnaccountedStrides[originalStride] ==
candidateStridesNumOccurences[originalStride]) {
// The stride for this is not dropped. Keep as is.
unusedDims.reset(dim);
continue;
}
if (currUnaccountedStrides[originalStride] <
candidateStridesNumOccurences[originalStride]) {
// This should never happen. Cant have a stride in the reduced rank type
// that wasnt in the original one.
return failure();
}
}
if ((int64_t)unusedDims.count() + reducedType.getRank() !=
originalType.getRank())
return failure();
return unusedDims;
}
llvm::SmallBitVector SubViewOp::getDroppedDims() {
MemRefType sourceType = getSourceType();
MemRefType resultType = getType();
FailureOr<llvm::SmallBitVector> unusedDims =
computeMemRefRankReductionMask(sourceType, resultType, getMixedSizes());
assert(succeeded(unusedDims) && "unable to find unused dims of subview");
return *unusedDims;
}
OpFoldResult DimOp::fold(FoldAdaptor adaptor) {
// All forms of folding require a known index.
auto index = llvm::dyn_cast_if_present<IntegerAttr>(adaptor.getIndex());
if (!index)
return {};
// Folding for unranked types (UnrankedMemRefType) is not supported.
auto memrefType = llvm::dyn_cast<MemRefType>(getSource().getType());
if (!memrefType)
return {};
// Out of bound indices produce undefined behavior but are still valid IR.
// Don't choke on them.
int64_t indexVal = index.getInt();
if (indexVal < 0 || indexVal >= memrefType.getRank())
return {};
// Fold if the shape extent along the given index is known.
if (!memrefType.isDynamicDim(index.getInt())) {
Builder builder(getContext());
return builder.getIndexAttr(memrefType.getShape()[index.getInt()]);
}
// The size at the given index is now known to be a dynamic size.
unsigned unsignedIndex = index.getValue().getZExtValue();
// Fold dim to the size argument for an `AllocOp`, `ViewOp`, or `SubViewOp`.
Operation *definingOp = getSource().getDefiningOp();
if (auto alloc = dyn_cast_or_null<AllocOp>(definingOp))
return *(alloc.getDynamicSizes().begin() +
memrefType.getDynamicDimIndex(unsignedIndex));
if (auto alloca = dyn_cast_or_null<AllocaOp>(definingOp))
return *(alloca.getDynamicSizes().begin() +
memrefType.getDynamicDimIndex(unsignedIndex));
if (auto view = dyn_cast_or_null<ViewOp>(definingOp))
return *(view.getDynamicSizes().begin() +
memrefType.getDynamicDimIndex(unsignedIndex));
if (auto subview = dyn_cast_or_null<SubViewOp>(definingOp)) {
llvm::SmallBitVector unusedDims = subview.getDroppedDims();
unsigned resultIndex = 0;
unsigned sourceRank = subview.getSourceType().getRank();
unsigned sourceIndex = 0;
for (auto i : llvm::seq<unsigned>(0, sourceRank)) {
if (unusedDims.test(i))
continue;
if (resultIndex == unsignedIndex) {
sourceIndex = i;
break;
}
resultIndex++;
}
assert(subview.isDynamicSize(sourceIndex) &&
"expected dynamic subview size");
return subview.getDynamicSize(sourceIndex);
}
if (auto sizeInterface =
dyn_cast_or_null<OffsetSizeAndStrideOpInterface>(definingOp)) {
assert(sizeInterface.isDynamicSize(unsignedIndex) &&
"Expected dynamic subview size");
return sizeInterface.getDynamicSize(unsignedIndex);
}
// dim(memrefcast) -> dim
if (succeeded(foldMemRefCast(*this)))
return getResult();
return {};
}
namespace {
/// Fold dim of a memref reshape operation to a load into the reshape's shape
/// operand.
struct DimOfMemRefReshape : public OpRewritePattern<DimOp> {
using OpRewritePattern<DimOp>::OpRewritePattern;
LogicalResult matchAndRewrite(DimOp dim,
PatternRewriter &rewriter) const override {
auto reshape = dim.getSource().getDefiningOp<ReshapeOp>();
if (!reshape)
return rewriter.notifyMatchFailure(
dim, "Dim op is not defined by a reshape op.");
// dim of a memref reshape can be folded if dim.getIndex() dominates the
// reshape. Instead of using `DominanceInfo` (which is usually costly) we
// cheaply check that either of the following conditions hold:
// 1. dim.getIndex() is defined in the same block as reshape but before
// reshape.
// 2. dim.getIndex() is defined in a parent block of
// reshape.
// Check condition 1
if (dim.getIndex().getParentBlock() == reshape->getBlock()) {
if (auto *definingOp = dim.getIndex().getDefiningOp()) {
if (reshape->isBeforeInBlock(definingOp)) {
return rewriter.notifyMatchFailure(
dim,
"dim.getIndex is not defined before reshape in the same block.");
}
} // else dim.getIndex is a block argument to reshape->getBlock and
// dominates reshape
} // Check condition 2
else if (dim->getBlock() != reshape->getBlock() &&
!dim.getIndex().getParentRegion()->isProperAncestor(
reshape->getParentRegion())) {
// If dim and reshape are in the same block but dim.getIndex() isn't, we
// already know dim.getIndex() dominates reshape without calling
// `isProperAncestor`
return rewriter.notifyMatchFailure(
dim, "dim.getIndex does not dominate reshape.");
}
// Place the load directly after the reshape to ensure that the shape memref
// was not mutated.
rewriter.setInsertionPointAfter(reshape);
Location loc = dim.getLoc();
Value load =
rewriter.create<LoadOp>(loc, reshape.getShape(), dim.getIndex());
if (load.getType() != dim.getType())
load = rewriter.create<arith::IndexCastOp>(loc, dim.getType(), load);
rewriter.replaceOp(dim, load);
return success();
}
};
} // namespace
void DimOp::getCanonicalizationPatterns(RewritePatternSet &results,
MLIRContext *context) {
results.add<DimOfMemRefReshape>(context);
}
// ---------------------------------------------------------------------------
// DmaStartOp
// ---------------------------------------------------------------------------
void DmaStartOp::build(OpBuilder &builder, OperationState &result,
Value srcMemRef, ValueRange srcIndices, Value destMemRef,
ValueRange destIndices, Value numElements,
Value tagMemRef, ValueRange tagIndices, Value stride,
Value elementsPerStride) {
result.addOperands(srcMemRef);
result.addOperands(srcIndices);
result.addOperands(destMemRef);
result.addOperands(destIndices);
result.addOperands({numElements, tagMemRef});
result.addOperands(tagIndices);
if (stride)
result.addOperands({stride, elementsPerStride});
}
void DmaStartOp::print(OpAsmPrinter &p) {
p << " " << getSrcMemRef() << '[' << getSrcIndices() << "], "
<< getDstMemRef() << '[' << getDstIndices() << "], " << getNumElements()
<< ", " << getTagMemRef() << '[' << getTagIndices() << ']';
if (isStrided())
p << ", " << getStride() << ", " << getNumElementsPerStride();
p.printOptionalAttrDict((*this)->getAttrs());
p << " : " << getSrcMemRef().getType() << ", " << getDstMemRef().getType()
<< ", " << getTagMemRef().getType();
}
// Parse DmaStartOp.
// Ex:
// %dma_id = dma_start %src[%i, %j], %dst[%k, %l], %size,
// %tag[%index], %stride, %num_elt_per_stride :
// : memref<3076 x f32, 0>,
// memref<1024 x f32, 2>,
// memref<1 x i32>
//
ParseResult DmaStartOp::parse(OpAsmParser &parser, OperationState &result) {
OpAsmParser::UnresolvedOperand srcMemRefInfo;
SmallVector<OpAsmParser::UnresolvedOperand, 4> srcIndexInfos;
OpAsmParser::UnresolvedOperand dstMemRefInfo;
SmallVector<OpAsmParser::UnresolvedOperand, 4> dstIndexInfos;
OpAsmParser::UnresolvedOperand numElementsInfo;
OpAsmParser::UnresolvedOperand tagMemrefInfo;
SmallVector<OpAsmParser::UnresolvedOperand, 4> tagIndexInfos;
SmallVector<OpAsmParser::UnresolvedOperand, 2> strideInfo;
SmallVector<Type, 3> types;
auto indexType = parser.getBuilder().getIndexType();
// Parse and resolve the following list of operands:
// *) source memref followed by its indices (in square brackets).
// *) destination memref followed by its indices (in square brackets).
// *) dma size in KiB.
if (parser.parseOperand(srcMemRefInfo) ||
parser.parseOperandList(srcIndexInfos, OpAsmParser::Delimiter::Square) ||
parser.parseComma() || parser.parseOperand(dstMemRefInfo) ||
parser.parseOperandList(dstIndexInfos, OpAsmParser::Delimiter::Square) ||
parser.parseComma() || parser.parseOperand(numElementsInfo) ||
parser.parseComma() || parser.parseOperand(tagMemrefInfo) ||
parser.parseOperandList(tagIndexInfos, OpAsmParser::Delimiter::Square))
return failure();
// Parse optional stride and elements per stride.
if (parser.parseTrailingOperandList(strideInfo))
return failure();
bool isStrided = strideInfo.size() == 2;
if (!strideInfo.empty() && !isStrided) {
return parser.emitError(parser.getNameLoc(),
"expected two stride related operands");
}
if (parser.parseColonTypeList(types))
return failure();
if (types.size() != 3)
return parser.emitError(parser.getNameLoc(), "fewer/more types expected");
if (parser.resolveOperand(srcMemRefInfo, types[0], result.operands) ||
parser.resolveOperands(srcIndexInfos, indexType, result.operands) ||
parser.resolveOperand(dstMemRefInfo, types[1], result.operands) ||
parser.resolveOperands(dstIndexInfos, indexType, result.operands) ||
// size should be an index.
parser.resolveOperand(numElementsInfo, indexType, result.operands) ||
parser.resolveOperand(tagMemrefInfo, types[2], result.operands) ||
// tag indices should be index.
parser.resolveOperands(tagIndexInfos, indexType, result.operands))
return failure();
if (isStrided) {
if (parser.resolveOperands(strideInfo, indexType, result.operands))
return failure();
}
return success();
}
LogicalResult DmaStartOp::verify() {
unsigned numOperands = getNumOperands();
// Mandatory non-variadic operands are: src memref, dst memref, tag memref and
// the number of elements.
if (numOperands < 4)
return emitOpError("expected at least 4 operands");
// Check types of operands. The order of these calls is important: the later
// calls rely on some type properties to compute the operand position.
// 1. Source memref.
if (!llvm::isa<MemRefType>(getSrcMemRef().getType()))
return emitOpError("expected source to be of memref type");
if (numOperands < getSrcMemRefRank() + 4)
return emitOpError() << "expected at least " << getSrcMemRefRank() + 4
<< " operands";
if (!getSrcIndices().empty() &&
!llvm::all_of(getSrcIndices().getTypes(),
[](Type t) { return t.isIndex(); }))
return emitOpError("expected source indices to be of index type");
// 2. Destination memref.
if (!llvm::isa<MemRefType>(getDstMemRef().getType()))
return emitOpError("expected destination to be of memref type");
unsigned numExpectedOperands = getSrcMemRefRank() + getDstMemRefRank() + 4;
if (numOperands < numExpectedOperands)
return emitOpError() << "expected at least " << numExpectedOperands
<< " operands";
if (!getDstIndices().empty() &&
!llvm::all_of(getDstIndices().getTypes(),
[](Type t) { return t.isIndex(); }))
return emitOpError("expected destination indices to be of index type");
// 3. Number of elements.
if (!getNumElements().getType().isIndex())
return emitOpError("expected num elements to be of index type");
// 4. Tag memref.
if (!llvm::isa<MemRefType>(getTagMemRef().getType()))
return emitOpError("expected tag to be of memref type");
numExpectedOperands += getTagMemRefRank();
if (numOperands < numExpectedOperands)
return emitOpError() << "expected at least " << numExpectedOperands
<< " operands";
if (!getTagIndices().empty() &&
!llvm::all_of(getTagIndices().getTypes(),
[](Type t) { return t.isIndex(); }))
return emitOpError("expected tag indices to be of index type");
// Optional stride-related operands must be either both present or both
// absent.
if (numOperands != numExpectedOperands &&
numOperands != numExpectedOperands + 2)
return emitOpError("incorrect number of operands");
// 5. Strides.
if (isStrided()) {
if (!getStride().getType().isIndex() ||
!getNumElementsPerStride().getType().isIndex())
return emitOpError(
"expected stride and num elements per stride to be of type index");
}
return success();
}
LogicalResult DmaStartOp::fold(FoldAdaptor adaptor,
SmallVectorImpl<OpFoldResult> &results) {
/// dma_start(memrefcast) -> dma_start
return foldMemRefCast(*this);
}
// ---------------------------------------------------------------------------
// DmaWaitOp
// ---------------------------------------------------------------------------
LogicalResult DmaWaitOp::fold(FoldAdaptor adaptor,
SmallVectorImpl<OpFoldResult> &results) {
/// dma_wait(memrefcast) -> dma_wait
return foldMemRefCast(*this);
}
LogicalResult DmaWaitOp::verify() {
// Check that the number of tag indices matches the tagMemRef rank.
unsigned numTagIndices = getTagIndices().size();
unsigned tagMemRefRank = getTagMemRefRank();
if (numTagIndices != tagMemRefRank)
return emitOpError() << "expected tagIndices to have the same number of "
"elements as the tagMemRef rank, expected "
<< tagMemRefRank << ", but got " << numTagIndices;
return success();
}
//===----------------------------------------------------------------------===//
// ExtractAlignedPointerAsIndexOp
//===----------------------------------------------------------------------===//
void ExtractAlignedPointerAsIndexOp::getAsmResultNames(
function_ref<void(Value, StringRef)> setNameFn) {
setNameFn(getResult(), "intptr");
}
//===----------------------------------------------------------------------===//
// ExtractStridedMetadataOp
//===----------------------------------------------------------------------===//
/// The number and type of the results are inferred from the
/// shape of the source.
LogicalResult ExtractStridedMetadataOp::inferReturnTypes(
MLIRContext *context, std::optional<Location> location,
ExtractStridedMetadataOp::Adaptor adaptor,
SmallVectorImpl<Type> &inferredReturnTypes) {
auto sourceType = llvm::dyn_cast<MemRefType>(adaptor.getSource().getType());
if (!sourceType)
return failure();
unsigned sourceRank = sourceType.getRank();
IndexType indexType = IndexType::get(context);
auto memrefType =
MemRefType::get({}, sourceType.getElementType(),
MemRefLayoutAttrInterface{}, sourceType.getMemorySpace());
// Base.
inferredReturnTypes.push_back(memrefType);
// Offset.
inferredReturnTypes.push_back(indexType);
// Sizes and strides.
for (unsigned i = 0; i < sourceRank * 2; ++i)
inferredReturnTypes.push_back(indexType);
return success();
}
void ExtractStridedMetadataOp::getAsmResultNames(
function_ref<void(Value, StringRef)> setNameFn) {
setNameFn(getBaseBuffer(), "base_buffer");
setNameFn(getOffset(), "offset");
// For multi-result to work properly with pretty names and packed syntax `x:3`
// we can only give a pretty name to the first value in the pack.
if (!getSizes().empty()) {
setNameFn(getSizes().front(), "sizes");
setNameFn(getStrides().front(), "strides");
}
}
/// Helper function to perform the replacement of all constant uses of `values`
/// by a materialized constant extracted from `maybeConstants`.
/// `values` and `maybeConstants` are expected to have the same size.
template <typename Container>
static bool replaceConstantUsesOf(OpBuilder &rewriter, Location loc,
Container values,
ArrayRef<OpFoldResult> maybeConstants) {
assert(values.size() == maybeConstants.size() &&
" expected values and maybeConstants of the same size");
bool atLeastOneReplacement = false;
for (auto [maybeConstant, result] : llvm::zip(maybeConstants, values)) {
// Don't materialize a constant if there are no uses: this would indice
// infinite loops in the driver.
if (result.use_empty() || maybeConstant == getAsOpFoldResult(result))
continue;
assert(maybeConstant.template is<Attribute>() &&
"The constified value should be either unchanged (i.e., == result) "
"or a constant");
Value constantVal = rewriter.create<arith::ConstantIndexOp>(
loc, llvm::cast<IntegerAttr>(maybeConstant.template get<Attribute>())
.getInt());
for (Operation *op : llvm::make_early_inc_range(result.getUsers())) {
// modifyOpInPlace: lambda cannot capture structured bindings in C++17
// yet.
op->replaceUsesOfWith(result, constantVal);
atLeastOneReplacement = true;
}
}
return atLeastOneReplacement;
}
LogicalResult
ExtractStridedMetadataOp::fold(FoldAdaptor adaptor,
SmallVectorImpl<OpFoldResult> &results) {
OpBuilder builder(*this);
bool atLeastOneReplacement = replaceConstantUsesOf(
builder, getLoc(), ArrayRef<TypedValue<IndexType>>(getOffset()),
getConstifiedMixedOffset());
atLeastOneReplacement |= replaceConstantUsesOf(builder, getLoc(), getSizes(),
getConstifiedMixedSizes());
atLeastOneReplacement |= replaceConstantUsesOf(
builder, getLoc(), getStrides(), getConstifiedMixedStrides());
return success(atLeastOneReplacement);
}
SmallVector<OpFoldResult> ExtractStridedMetadataOp::getConstifiedMixedSizes() {
SmallVector<OpFoldResult> values = getAsOpFoldResult(getSizes());
constifyIndexValues(values, getSource().getType(), getContext(),
getConstantSizes, ShapedType::isDynamic);
return values;
}
SmallVector<OpFoldResult>
ExtractStridedMetadataOp::getConstifiedMixedStrides() {
SmallVector<OpFoldResult> values = getAsOpFoldResult(getStrides());
constifyIndexValues(values, getSource().getType(), getContext(),
getConstantStrides, ShapedType::isDynamic);
return values;
}
OpFoldResult ExtractStridedMetadataOp::getConstifiedMixedOffset() {
OpFoldResult offsetOfr = getAsOpFoldResult(getOffset());
SmallVector<OpFoldResult> values(1, offsetOfr);
constifyIndexValues(values, getSource().getType(), getContext(),
getConstantOffset, ShapedType::isDynamic);
return values[0];
}
//===----------------------------------------------------------------------===//
// GenericAtomicRMWOp
//===----------------------------------------------------------------------===//
void GenericAtomicRMWOp::build(OpBuilder &builder, OperationState &result,
Value memref, ValueRange ivs) {
OpBuilder::InsertionGuard g(builder);
result.addOperands(memref);
result.addOperands(ivs);
if (auto memrefType = llvm::dyn_cast<MemRefType>(memref.getType())) {
Type elementType = memrefType.getElementType();
result.addTypes(elementType);
Region *bodyRegion = result.addRegion();
builder.createBlock(bodyRegion);
bodyRegion->addArgument(elementType, memref.getLoc());
}
}
LogicalResult GenericAtomicRMWOp::verify() {
auto &body = getRegion();
if (body.getNumArguments() != 1)
return emitOpError("expected single number of entry block arguments");
if (getResult().getType() != body.getArgument(0).getType())
return emitOpError("expected block argument of the same type result type");
bool hasSideEffects =
body.walk([&](Operation *nestedOp) {
if (isMemoryEffectFree(nestedOp))
return WalkResult::advance();
nestedOp->emitError(
"body of 'memref.generic_atomic_rmw' should contain "
"only operations with no side effects");
return WalkResult::interrupt();
})
.wasInterrupted();
return hasSideEffects ? failure() : success();
}
ParseResult GenericAtomicRMWOp::parse(OpAsmParser &parser,
OperationState &result) {
OpAsmParser::UnresolvedOperand memref;
Type memrefType;
SmallVector<OpAsmParser::UnresolvedOperand, 4> ivs;
Type indexType = parser.getBuilder().getIndexType();
if (parser.parseOperand(memref) ||
parser.parseOperandList(ivs, OpAsmParser::Delimiter::Square) ||
parser.parseColonType(memrefType) ||
parser.resolveOperand(memref, memrefType, result.operands) ||
parser.resolveOperands(ivs, indexType, result.operands))
return failure();
Region *body = result.addRegion();
if (parser.parseRegion(*body, {}) ||
parser.parseOptionalAttrDict(result.attributes))
return failure();
result.types.push_back(llvm::cast<MemRefType>(memrefType).getElementType());
return success();
}
void GenericAtomicRMWOp::print(OpAsmPrinter &p) {
p << ' ' << getMemref() << "[" << getIndices()
<< "] : " << getMemref().getType() << ' ';
p.printRegion(getRegion());
p.printOptionalAttrDict((*this)->getAttrs());
}
//===----------------------------------------------------------------------===//
// AtomicYieldOp
//===----------------------------------------------------------------------===//
LogicalResult AtomicYieldOp::verify() {
Type parentType = (*this)->getParentOp()->getResultTypes().front();
Type resultType = getResult().getType();
if (parentType != resultType)
return emitOpError() << "types mismatch between yield op: " << resultType
<< " and its parent: " << parentType;
return success();
}
//===----------------------------------------------------------------------===//
// GlobalOp
//===----------------------------------------------------------------------===//
static void printGlobalMemrefOpTypeAndInitialValue(OpAsmPrinter &p, GlobalOp op,
TypeAttr type,
Attribute initialValue) {
p << type;
if (!op.isExternal()) {
p << " = ";
if (op.isUninitialized())
p << "uninitialized";
else
p.printAttributeWithoutType(initialValue);
}
}
static ParseResult
parseGlobalMemrefOpTypeAndInitialValue(OpAsmParser &parser, TypeAttr &typeAttr,
Attribute &initialValue) {
Type type;
if (parser.parseType(type))
return failure();
auto memrefType = llvm::dyn_cast<MemRefType>(type);
if (!memrefType || !memrefType.hasStaticShape())
return parser.emitError(parser.getNameLoc())
<< "type should be static shaped memref, but got " << type;
typeAttr = TypeAttr::get(type);
if (parser.parseOptionalEqual())
return success();
if (succeeded(parser.parseOptionalKeyword("uninitialized"))) {
initialValue = UnitAttr::get(parser.getContext());
return success();
}
Type tensorType = getTensorTypeFromMemRefType(memrefType);
if (parser.parseAttribute(initialValue, tensorType))
return failure();
if (!llvm::isa<ElementsAttr>(initialValue))
return parser.emitError(parser.getNameLoc())
<< "initial value should be a unit or elements attribute";
return success();
}
LogicalResult GlobalOp::verify() {
auto memrefType = llvm::dyn_cast<MemRefType>(getType());
if (!memrefType || !memrefType.hasStaticShape())
return emitOpError("type should be static shaped memref, but got ")
<< getType();
// Verify that the initial value, if present, is either a unit attribute or
// an elements attribute.
if (getInitialValue().has_value()) {
Attribute initValue = getInitialValue().value();
if (!llvm::isa<UnitAttr>(initValue) && !llvm::isa<ElementsAttr>(initValue))
return emitOpError("initial value should be a unit or elements "
"attribute, but got ")
<< initValue;
// Check that the type of the initial value is compatible with the type of
// the global variable.
if (auto elementsAttr = llvm::dyn_cast<ElementsAttr>(initValue)) {
Type initType = elementsAttr.getType();
Type tensorType = getTensorTypeFromMemRefType(memrefType);
if (initType != tensorType)
return emitOpError("initial value expected to be of type ")
<< tensorType << ", but was of type " << initType;
}
}
if (std::optional<uint64_t> alignAttr = getAlignment()) {
uint64_t alignment = *alignAttr;
if (!llvm::isPowerOf2_64(alignment))
return emitError() << "alignment attribute value " << alignment
<< " is not a power of 2";
}
// TODO: verify visibility for declarations.
return success();
}
ElementsAttr GlobalOp::getConstantInitValue() {
auto initVal = getInitialValue();
if (getConstant() && initVal.has_value())
return llvm::cast<ElementsAttr>(initVal.value());
return {};
}
//===----------------------------------------------------------------------===//
// GetGlobalOp
//===----------------------------------------------------------------------===//
LogicalResult
GetGlobalOp::verifySymbolUses(SymbolTableCollection &symbolTable) {
// Verify that the result type is same as the type of the referenced
// memref.global op.
auto global =
symbolTable.lookupNearestSymbolFrom<GlobalOp>(*this, getNameAttr());
if (!global)
return emitOpError("'")
<< getName() << "' does not reference a valid global memref";
Type resultType = getResult().getType();
if (global.getType() != resultType)
return emitOpError("result type ")
<< resultType << " does not match type " << global.getType()
<< " of the global memref @" << getName();
return success();
}
//===----------------------------------------------------------------------===//
// LoadOp
//===----------------------------------------------------------------------===//
LogicalResult LoadOp::verify() {
if (static_cast<int64_t>(getIndices().size()) != getMemRefType().getRank()) {
return emitOpError("incorrect number of indices for load, expected ")
<< getMemRefType().getRank() << " but got " << getIndices().size();
}
return success();
}
OpFoldResult LoadOp::fold(FoldAdaptor adaptor) {
/// load(memrefcast) -> load
if (succeeded(foldMemRefCast(*this)))
return getResult();
return OpFoldResult();
}
//===----------------------------------------------------------------------===//
// MemorySpaceCastOp
//===----------------------------------------------------------------------===//
void MemorySpaceCastOp::getAsmResultNames(
function_ref<void(Value, StringRef)> setNameFn) {
setNameFn(getResult(), "memspacecast");
}
bool MemorySpaceCastOp::areCastCompatible(TypeRange inputs, TypeRange outputs) {
if (inputs.size() != 1 || outputs.size() != 1)
return false;
Type a = inputs.front(), b = outputs.front();
auto aT = llvm::dyn_cast<MemRefType>(a);
auto bT = llvm::dyn_cast<MemRefType>(b);
auto uaT = llvm::dyn_cast<UnrankedMemRefType>(a);
auto ubT = llvm::dyn_cast<UnrankedMemRefType>(b);
if (aT && bT) {
if (aT.getElementType() != bT.getElementType())
return false;
if (aT.getLayout() != bT.getLayout())
return false;
if (aT.getShape() != bT.getShape())
return false;
return true;
}
if (uaT && ubT) {
return uaT.getElementType() == ubT.getElementType();
}
return false;
}
OpFoldResult MemorySpaceCastOp::fold(FoldAdaptor adaptor) {
// memory_space_cast(memory_space_cast(v, t1), t2) -> memory_space_cast(v,
// t2)
if (auto parentCast = getSource().getDefiningOp<MemorySpaceCastOp>()) {
getSourceMutable().assign(parentCast.getSource());
return getResult();
}
return Value{};
}
//===----------------------------------------------------------------------===//
// PrefetchOp
//===----------------------------------------------------------------------===//
void PrefetchOp::print(OpAsmPrinter &p) {
p << " " << getMemref() << '[';
p.printOperands(getIndices());
p << ']' << ", " << (getIsWrite() ? "write" : "read");
p << ", locality<" << getLocalityHint();
p << ">, " << (getIsDataCache() ? "data" : "instr");
p.printOptionalAttrDict(
(*this)->getAttrs(),
/*elidedAttrs=*/{"localityHint", "isWrite", "isDataCache"});
p << " : " << getMemRefType();
}
ParseResult PrefetchOp::parse(OpAsmParser &parser, OperationState &result) {
OpAsmParser::UnresolvedOperand memrefInfo;
SmallVector<OpAsmParser::UnresolvedOperand, 4> indexInfo;
IntegerAttr localityHint;
MemRefType type;
StringRef readOrWrite, cacheType;
auto indexTy = parser.getBuilder().getIndexType();
auto i32Type = parser.getBuilder().getIntegerType(32);
if (parser.parseOperand(memrefInfo) ||
parser.parseOperandList(indexInfo, OpAsmParser::Delimiter::Square) ||
parser.parseComma() || parser.parseKeyword(&readOrWrite) ||
parser.parseComma() || parser.parseKeyword("locality") ||
parser.parseLess() ||
parser.parseAttribute(localityHint, i32Type, "localityHint",
result.attributes) ||
parser.parseGreater() || parser.parseComma() ||
parser.parseKeyword(&cacheType) || parser.parseColonType(type) ||
parser.resolveOperand(memrefInfo, type, result.operands) ||
parser.resolveOperands(indexInfo, indexTy, result.operands))
return failure();
if (!readOrWrite.equals("read") && !readOrWrite.equals("write"))
return parser.emitError(parser.getNameLoc(),
"rw specifier has to be 'read' or 'write'");
result.addAttribute(
PrefetchOp::getIsWriteAttrStrName(),
parser.getBuilder().getBoolAttr(readOrWrite.equals("write")));
if (!cacheType.equals("data") && !cacheType.equals("instr"))
return parser.emitError(parser.getNameLoc(),
"cache type has to be 'data' or 'instr'");
result.addAttribute(
PrefetchOp::getIsDataCacheAttrStrName(),
parser.getBuilder().getBoolAttr(cacheType.equals("data")));
return success();
}
LogicalResult PrefetchOp::verify() {
if (getNumOperands() != 1 + getMemRefType().getRank())
return emitOpError("too few indices");
return success();
}
LogicalResult PrefetchOp::fold(FoldAdaptor adaptor,
SmallVectorImpl<OpFoldResult> &results) {
// prefetch(memrefcast) -> prefetch
return foldMemRefCast(*this);
}
//===----------------------------------------------------------------------===//
// RankOp
//===----------------------------------------------------------------------===//
OpFoldResult RankOp::fold(FoldAdaptor adaptor) {
// Constant fold rank when the rank of the operand is known.
auto type = getOperand().getType();
auto shapedType = llvm::dyn_cast<ShapedType>(type);
if (shapedType && shapedType.hasRank())
return IntegerAttr::get(IndexType::get(getContext()), shapedType.getRank());
return IntegerAttr();
}
//===----------------------------------------------------------------------===//
// ReinterpretCastOp
//===----------------------------------------------------------------------===//
void ReinterpretCastOp::getAsmResultNames(
function_ref<void(Value, StringRef)> setNameFn) {
setNameFn(getResult(), "reinterpret_cast");
}
/// Build a ReinterpretCastOp with all dynamic entries: `staticOffsets`,
/// `staticSizes` and `staticStrides` are automatically filled with
/// source-memref-rank sentinel values that encode dynamic entries.
void ReinterpretCastOp::build(OpBuilder &b, OperationState &result,
MemRefType resultType, Value source,
OpFoldResult offset, ArrayRef<OpFoldResult> sizes,
ArrayRef<OpFoldResult> strides,
ArrayRef<NamedAttribute> attrs) {
SmallVector<int64_t> staticOffsets, staticSizes, staticStrides;
SmallVector<Value> dynamicOffsets, dynamicSizes, dynamicStrides;
dispatchIndexOpFoldResults(offset, dynamicOffsets, staticOffsets);
dispatchIndexOpFoldResults(sizes, dynamicSizes, staticSizes);
dispatchIndexOpFoldResults(strides, dynamicStrides, staticStrides);
build(b, result, resultType, source, dynamicOffsets, dynamicSizes,
dynamicStrides, b.getDenseI64ArrayAttr(staticOffsets),
b.getDenseI64ArrayAttr(staticSizes),
b.getDenseI64ArrayAttr(staticStrides));
result.addAttributes(attrs);
}
void ReinterpretCastOp::build(OpBuilder &b, OperationState &result,
MemRefType resultType, Value source,
int64_t offset, ArrayRef<int64_t> sizes,
ArrayRef<int64_t> strides,
ArrayRef<NamedAttribute> attrs) {
SmallVector<OpFoldResult> sizeValues =
llvm::to_vector<4>(llvm::map_range(sizes, [&](int64_t v) -> OpFoldResult {
return b.getI64IntegerAttr(v);
}));
SmallVector<OpFoldResult> strideValues = llvm::to_vector<4>(
llvm::map_range(strides, [&](int64_t v) -> OpFoldResult {
return b.getI64IntegerAttr(v);
}));
build(b, result, resultType, source, b.getI64IntegerAttr(offset), sizeValues,
strideValues, attrs);
}
void ReinterpretCastOp::build(OpBuilder &b, OperationState &result,
MemRefType resultType, Value source, Value offset,
ValueRange sizes, ValueRange strides,
ArrayRef<NamedAttribute> attrs) {
SmallVector<OpFoldResult> sizeValues = llvm::to_vector<4>(
llvm::map_range(sizes, [](Value v) -> OpFoldResult { return v; }));
SmallVector<OpFoldResult> strideValues = llvm::to_vector<4>(
llvm::map_range(strides, [](Value v) -> OpFoldResult { return v; }));
build(b, result, resultType, source, offset, sizeValues, strideValues, attrs);
}
// TODO: ponder whether we want to allow missing trailing sizes/strides that are
// completed automatically, like we have for subview and extract_slice.
LogicalResult ReinterpretCastOp::verify() {
// The source and result memrefs should be in the same memory space.
auto srcType = llvm::cast<BaseMemRefType>(getSource().getType());
auto resultType = llvm::cast<MemRefType>(getType());
if (srcType.getMemorySpace() != resultType.getMemorySpace())
return emitError("different memory spaces specified for source type ")
<< srcType << " and result memref type " << resultType;
if (srcType.getElementType() != resultType.getElementType())
return emitError("different element types specified for source type ")
<< srcType << " and result memref type " << resultType;
// Match sizes in result memref type and in static_sizes attribute.
for (auto [idx, resultSize, expectedSize] :
llvm::enumerate(resultType.getShape(), getStaticSizes())) {
if (!ShapedType::isDynamic(resultSize) &&
!ShapedType::isDynamic(expectedSize) && resultSize != expectedSize)
return emitError("expected result type with size = ")
<< expectedSize << " instead of " << resultSize
<< " in dim = " << idx;
}
// Match offset and strides in static_offset and static_strides attributes. If
// result memref type has no affine map specified, this will assume an
// identity layout.
int64_t resultOffset;
SmallVector<int64_t, 4> resultStrides;
if (failed(getStridesAndOffset(resultType, resultStrides, resultOffset)))
return emitError("expected result type to have strided layout but found ")
<< resultType;
// Match offset in result memref type and in static_offsets attribute.
int64_t expectedOffset = getStaticOffsets().front();
if (!ShapedType::isDynamic(resultOffset) &&
!ShapedType::isDynamic(expectedOffset) && resultOffset != expectedOffset)
return emitError("expected result type with offset = ")
<< expectedOffset << " instead of " << resultOffset;
// Match strides in result memref type and in static_strides attribute.
for (auto [idx, resultStride, expectedStride] :
llvm::enumerate(resultStrides, getStaticStrides())) {
if (!ShapedType::isDynamic(resultStride) &&
!ShapedType::isDynamic(expectedStride) &&
resultStride != expectedStride)
return emitError("expected result type with stride = ")
<< expectedStride << " instead of " << resultStride
<< " in dim = " << idx;
}
return success();
}
OpFoldResult ReinterpretCastOp::fold(FoldAdaptor /*operands*/) {
Value src = getSource();
auto getPrevSrc = [&]() -> Value {
// reinterpret_cast(reinterpret_cast(x)) -> reinterpret_cast(x).
if (auto prev = src.getDefiningOp<ReinterpretCastOp>())
return prev.getSource();
// reinterpret_cast(cast(x)) -> reinterpret_cast(x).
if (auto prev = src.getDefiningOp<CastOp>())
return prev.getSource();
// reinterpret_cast(subview(x)) -> reinterpret_cast(x) if subview offsets
// are 0.
if (auto prev = src.getDefiningOp<SubViewOp>())
if (llvm::all_of(prev.getMixedOffsets(), [](OpFoldResult val) {
return isConstantIntValue(val, 0);
}))
return prev.getSource();
return nullptr;
};
if (auto prevSrc = getPrevSrc()) {
getSourceMutable().assign(prevSrc);
return getResult();
}
// reinterpret_cast(x) w/o offset/shape/stride changes -> x
if (!ShapedType::isDynamicShape(getType().getShape()) &&
src.getType() == getType() && getStaticOffsets().front() == 0) {
return src;
}
return nullptr;
}
SmallVector<OpFoldResult> ReinterpretCastOp::getConstifiedMixedSizes() {
SmallVector<OpFoldResult> values = getMixedSizes();
constifyIndexValues(values, getType(), getContext(), getConstantSizes,
ShapedType::isDynamic);
return values;
}
SmallVector<OpFoldResult> ReinterpretCastOp::getConstifiedMixedStrides() {
SmallVector<OpFoldResult> values = getMixedStrides();
constifyIndexValues(values, getType(), getContext(), getConstantStrides,
ShapedType::isDynamic);
return values;
}
OpFoldResult ReinterpretCastOp::getConstifiedMixedOffset() {
SmallVector<OpFoldResult> values = getMixedOffsets();
assert(values.size() == 1 &&
"reinterpret_cast must have one and only one offset");
constifyIndexValues(values, getType(), getContext(), getConstantOffset,
ShapedType::isDynamic);
return values[0];
}
namespace {
/// Replace the sequence:
/// ```
/// base, offset, sizes, strides = extract_strided_metadata src
/// dst = reinterpret_cast base to offset, sizes, strides
/// ```
/// With
///
/// ```
/// dst = memref.cast src
/// ```
///
/// Note: The cast operation is only inserted when the type of dst and src
/// are not the same. E.g., when going from <4xf32> to <?xf32>.
///
/// This pattern also matches when the offset, sizes, and strides don't come
/// directly from the `extract_strided_metadata`'s results but it can be
/// statically proven that they would hold the same values.
///
/// For instance, the following sequence would be replaced:
/// ```
/// base, offset, sizes, strides =
/// extract_strided_metadata memref : memref<3x4xty>
/// dst = reinterpret_cast base to 0, [3, 4], strides
/// ```
/// Because we know (thanks to the type of the input memref) that variable
/// `offset` and `sizes` will respectively hold 0 and [3, 4].
///
/// Similarly, the following sequence would be replaced:
/// ```
/// c0 = arith.constant 0
/// c4 = arith.constant 4
/// base, offset, sizes, strides =
/// extract_strided_metadata memref : memref<3x4xty>
/// dst = reinterpret_cast base to c0, [3, c4], strides
/// ```
/// Because we know that `offset`and `c0` will hold 0
/// and `c4` will hold 4.
struct ReinterpretCastOpExtractStridedMetadataFolder
: public OpRewritePattern<ReinterpretCastOp> {
public:
using OpRewritePattern<ReinterpretCastOp>::OpRewritePattern;
LogicalResult matchAndRewrite(ReinterpretCastOp op,
PatternRewriter &rewriter) const override {
auto extractStridedMetadata =
op.getSource().getDefiningOp<ExtractStridedMetadataOp>();
if (!extractStridedMetadata)
return failure();
// Check if the reinterpret cast reconstructs a memref with the exact same
// properties as the extract strided metadata.
// First, check that the strides are the same.
SmallVector<OpFoldResult> extractStridesOfr =
extractStridedMetadata.getConstifiedMixedStrides();
SmallVector<OpFoldResult> reinterpretStridesOfr =
op.getConstifiedMixedStrides();
if (extractStridesOfr.size() != reinterpretStridesOfr.size())
return failure();
unsigned rank = op.getType().getRank();
for (unsigned i = 0; i < rank; ++i) {
if (extractStridesOfr[i] != reinterpretStridesOfr[i])
return failure();
}
// Second, check the sizes.
assert(extractStridedMetadata.getSizes().size() ==
op.getMixedSizes().size() &&
"Strides and sizes rank must match");
SmallVector<OpFoldResult> extractSizesOfr =
extractStridedMetadata.getConstifiedMixedSizes();
SmallVector<OpFoldResult> reinterpretSizesOfr =
op.getConstifiedMixedSizes();
for (unsigned i = 0; i < rank; ++i) {
if (extractSizesOfr[i] != reinterpretSizesOfr[i])
return failure();
}
// Finally, check the offset.
assert(op.getMixedOffsets().size() == 1 &&
"reinterpret_cast with more than one offset should have been "
"rejected by the verifier");
OpFoldResult extractOffsetOfr =
extractStridedMetadata.getConstifiedMixedOffset();
OpFoldResult reinterpretOffsetOfr = op.getConstifiedMixedOffset();
if (extractOffsetOfr != reinterpretOffsetOfr)
return failure();
// At this point, we know that the back and forth between extract strided
// metadata and reinterpret cast is a noop. However, the final type of the
// reinterpret cast may not be exactly the same as the original memref.
// E.g., it could be changing a dimension from static to dynamic. Check that
// here and add a cast if necessary.
Type srcTy = extractStridedMetadata.getSource().getType();
if (srcTy == op.getResult().getType())
rewriter.replaceOp(op, extractStridedMetadata.getSource());
else
rewriter.replaceOpWithNewOp<CastOp>(op, op.getType(),
extractStridedMetadata.getSource());
return success();
}
};
} // namespace
void ReinterpretCastOp::getCanonicalizationPatterns(RewritePatternSet &results,
MLIRContext *context) {
results.add<ReinterpretCastOpExtractStridedMetadataFolder>(context);
}
//===----------------------------------------------------------------------===//
// Reassociative reshape ops
//===----------------------------------------------------------------------===//
void CollapseShapeOp::getAsmResultNames(
function_ref<void(Value, StringRef)> setNameFn) {
setNameFn(getResult(), "collapse_shape");
}
void ExpandShapeOp::getAsmResultNames(
function_ref<void(Value, StringRef)> setNameFn) {
setNameFn(getResult(), "expand_shape");
}
/// Helper function for verifying the shape of ExpandShapeOp and ResultShapeOp
/// result and operand. Layout maps are verified separately.
///
/// If `allowMultipleDynamicDimsPerGroup`, multiple dynamic dimensions are
/// allowed in a reassocation group.
static LogicalResult
verifyCollapsedShape(Operation *op, ArrayRef<int64_t> collapsedShape,
ArrayRef<int64_t> expandedShape,
ArrayRef<ReassociationIndices> reassociation,
bool allowMultipleDynamicDimsPerGroup) {
// There must be one reassociation group per collapsed dimension.
if (collapsedShape.size() != reassociation.size())
return op->emitOpError("invalid number of reassociation groups: found ")
<< reassociation.size() << ", expected " << collapsedShape.size();
// The next expected expanded dimension index (while iterating over
// reassociation indices).
int64_t nextDim = 0;
for (const auto &it : llvm::enumerate(reassociation)) {
ReassociationIndices group = it.value();
int64_t collapsedDim = it.index();
bool foundDynamic = false;
for (int64_t expandedDim : group) {
if (expandedDim != nextDim++)
return op->emitOpError("reassociation indices must be contiguous");
if (expandedDim >= static_cast<int64_t>(expandedShape.size()))
return op->emitOpError("reassociation index ")
<< expandedDim << " is out of bounds";
// Check if there are multiple dynamic dims in a reassociation group.
if (ShapedType::isDynamic(expandedShape[expandedDim])) {
if (foundDynamic && !allowMultipleDynamicDimsPerGroup)
return op->emitOpError(
"at most one dimension in a reassociation group may be dynamic");
foundDynamic = true;
}
}
// ExpandShapeOp/CollapseShapeOp may not be used to cast dynamicity.
if (ShapedType::isDynamic(collapsedShape[collapsedDim]) != foundDynamic)
return op->emitOpError("collapsed dim (")
<< collapsedDim
<< ") must be dynamic if and only if reassociation group is "
"dynamic";
// If all dims in the reassociation group are static, the size of the
// collapsed dim can be verified.
if (!foundDynamic) {
int64_t groupSize = 1;
for (int64_t expandedDim : group)
groupSize *= expandedShape[expandedDim];
if (groupSize != collapsedShape[collapsedDim])
return op->emitOpError("collapsed dim size (")
<< collapsedShape[collapsedDim]
<< ") must equal reassociation group size (" << groupSize << ")";
}
}
if (collapsedShape.empty()) {
// Rank 0: All expanded dimensions must be 1.
for (int64_t d : expandedShape)
if (d != 1)
return op->emitOpError(
"rank 0 memrefs can only be extended/collapsed with/from ones");
} else if (nextDim != static_cast<int64_t>(expandedShape.size())) {
// Rank >= 1: Number of dimensions among all reassociation groups must match
// the result memref rank.
return op->emitOpError("expanded rank (")
<< expandedShape.size()
<< ") inconsistent with number of reassociation indices (" << nextDim
<< ")";
}
return success();
}
SmallVector<AffineMap, 4> CollapseShapeOp::getReassociationMaps() {
return getSymbolLessAffineMaps(getReassociationExprs());
}
SmallVector<ReassociationExprs, 4> CollapseShapeOp::getReassociationExprs() {
return convertReassociationIndicesToExprs(getContext(),
getReassociationIndices());
}
SmallVector<AffineMap, 4> ExpandShapeOp::getReassociationMaps() {
return getSymbolLessAffineMaps(getReassociationExprs());
}
SmallVector<ReassociationExprs, 4> ExpandShapeOp::getReassociationExprs() {
return convertReassociationIndicesToExprs(getContext(),
getReassociationIndices());
}
/// Compute the layout map after expanding a given source MemRef type with the
/// specified reassociation indices.
static FailureOr<StridedLayoutAttr>
computeExpandedLayoutMap(MemRefType srcType, ArrayRef<int64_t> resultShape,
ArrayRef<ReassociationIndices> reassociation) {
int64_t srcOffset;
SmallVector<int64_t> srcStrides;
if (failed(getStridesAndOffset(srcType, srcStrides, srcOffset)))
return failure();
assert(srcStrides.size() == reassociation.size() && "invalid reassociation");
// 1-1 mapping between srcStrides and reassociation packs.
// Each srcStride starts with the given value and gets expanded according to
// the proper entries in resultShape.
// Example:
// srcStrides = [10000, 1 , 100 ],
// reassociations = [ [0], [1], [2, 3, 4]],
// resultSizes = [2, 5, 4, 3, 2] = [ [2], [5], [4, 3, 2]]
// -> For the purpose of stride calculation, the useful sizes are:
// [x, x, x, 3, 2] = [ [x], [x], [x, 3, 2]].
// resultStrides = [10000, 1, 600, 200, 100]
// Note that a stride does not get expanded along the first entry of each
// shape pack.
SmallVector<int64_t> reverseResultStrides;
reverseResultStrides.reserve(resultShape.size());
unsigned shapeIndex = resultShape.size() - 1;
for (auto it : llvm::reverse(llvm::zip(reassociation, srcStrides))) {
ReassociationIndices reassoc = std::get<0>(it);
int64_t currentStrideToExpand = std::get<1>(it);
for (unsigned idx = 0, e = reassoc.size(); idx < e; ++idx) {
reverseResultStrides.push_back(currentStrideToExpand);
currentStrideToExpand =
(SaturatedInteger::wrap(currentStrideToExpand) *
SaturatedInteger::wrap(resultShape[shapeIndex--]))
.asInteger();
}
}
auto resultStrides = llvm::to_vector<8>(llvm::reverse(reverseResultStrides));
resultStrides.resize(resultShape.size(), 1);
return StridedLayoutAttr::get(srcType.getContext(), srcOffset, resultStrides);
}
FailureOr<MemRefType> ExpandShapeOp::computeExpandedType(
MemRefType srcType, ArrayRef<int64_t> resultShape,
ArrayRef<ReassociationIndices> reassociation) {
if (srcType.getLayout().isIdentity()) {
// If the source is contiguous (i.e., no layout map specified), so is the
// result.
MemRefLayoutAttrInterface layout;
return MemRefType::get(resultShape, srcType.getElementType(), layout,
srcType.getMemorySpace());
}
// Source may not be contiguous. Compute the layout map.
FailureOr<StridedLayoutAttr> computedLayout =
computeExpandedLayoutMap(srcType, resultShape, reassociation);
if (failed(computedLayout))
return failure();
return MemRefType::get(resultShape, srcType.getElementType(), *computedLayout,
srcType.getMemorySpace());
}
void ExpandShapeOp::build(OpBuilder &builder, OperationState &result,
ArrayRef<int64_t> resultShape, Value src,
ArrayRef<ReassociationIndices> reassociation) {
// Only ranked memref source values are supported.
auto srcType = llvm::cast<MemRefType>(src.getType());
FailureOr<MemRefType> resultType =
ExpandShapeOp::computeExpandedType(srcType, resultShape, reassociation);
// Failure of this assertion usually indicates a problem with the source
// type, e.g., could not get strides/offset.
assert(succeeded(resultType) && "could not compute layout");
build(builder, result, *resultType, src, reassociation);
}
LogicalResult ExpandShapeOp::verify() {
MemRefType srcType = getSrcType();
MemRefType resultType = getResultType();
if (srcType.getRank() > resultType.getRank()) {
auto r0 = srcType.getRank();
auto r1 = resultType.getRank();
return emitOpError("has source rank ")
<< r0 << " and result rank " << r1 << ". This is not an expansion ("
<< r0 << " > " << r1 << ").";
}
// Verify result shape.
if (failed(verifyCollapsedShape(getOperation(), srcType.getShape(),
resultType.getShape(),
getReassociationIndices(),
/*allowMultipleDynamicDimsPerGroup=*/false)))
return failure();
// Compute expected result type (including layout map).
FailureOr<MemRefType> expectedResultType = ExpandShapeOp::computeExpandedType(
srcType, resultType.getShape(), getReassociationIndices());
if (failed(expectedResultType))
return emitOpError("invalid source layout map");
// Check actual result type.
if (*expectedResultType != resultType)
return emitOpError("expected expanded type to be ")
<< *expectedResultType << " but found " << resultType;
return success();
}
void ExpandShapeOp::getCanonicalizationPatterns(RewritePatternSet &results,
MLIRContext *context) {
results.add<ComposeReassociativeReshapeOps<ExpandShapeOp>,
ComposeExpandOfCollapseOp<ExpandShapeOp, CollapseShapeOp>>(
context);
}
/// Compute the layout map after collapsing a given source MemRef type with the
/// specified reassociation indices.
///
/// Note: All collapsed dims in a reassociation group must be contiguous. It is
/// not possible to check this by inspecting a MemRefType in the general case.
/// If non-contiguity cannot be checked statically, the collapse is assumed to
/// be valid (and thus accepted by this function) unless `strict = true`.
static FailureOr<StridedLayoutAttr>
computeCollapsedLayoutMap(MemRefType srcType,
ArrayRef<ReassociationIndices> reassociation,
bool strict = false) {
int64_t srcOffset;
SmallVector<int64_t> srcStrides;
auto srcShape = srcType.getShape();
if (failed(getStridesAndOffset(srcType, srcStrides, srcOffset)))
return failure();
// The result stride of a reassociation group is the stride of the last entry
// of the reassociation. (TODO: Should be the minimum stride in the
// reassociation because strides are not necessarily sorted. E.g., when using
// memref.transpose.) Dimensions of size 1 should be skipped, because their
// strides are meaningless and could have any arbitrary value.
SmallVector<int64_t> resultStrides;
resultStrides.reserve(reassociation.size());
for (const ReassociationIndices &reassoc : reassociation) {
ArrayRef<int64_t> ref = llvm::ArrayRef(reassoc);
while (srcShape[ref.back()] == 1 && ref.size() > 1)
ref = ref.drop_back();
if (!ShapedType::isDynamic(srcShape[ref.back()]) || ref.size() == 1) {
resultStrides.push_back(srcStrides[ref.back()]);
} else {
// Dynamically-sized dims may turn out to be dims of size 1 at runtime, so
// the corresponding stride may have to be skipped. (See above comment.)
// Therefore, the result stride cannot be statically determined and must
// be dynamic.
resultStrides.push_back(ShapedType::kDynamic);
}
}
// Validate that each reassociation group is contiguous.
unsigned resultStrideIndex = resultStrides.size() - 1;
for (const ReassociationIndices &reassoc : llvm::reverse(reassociation)) {
auto trailingReassocs = ArrayRef<int64_t>(reassoc).drop_front();
auto stride = SaturatedInteger::wrap(resultStrides[resultStrideIndex--]);
for (int64_t idx : llvm::reverse(trailingReassocs)) {
stride = stride * SaturatedInteger::wrap(srcShape[idx]);
// Both source and result stride must have the same static value. In that
// case, we can be sure, that the dimensions are collapsible (because they
// are contiguous).
// If `strict = false` (default during op verification), we accept cases
// where one or both strides are dynamic. This is best effort: We reject
// ops where obviously non-contiguous dims are collapsed, but accept ops
// where we cannot be sure statically. Such ops may fail at runtime. See
// the op documentation for details.
auto srcStride = SaturatedInteger::wrap(srcStrides[idx - 1]);
if (strict && (stride.saturated || srcStride.saturated))
return failure();
if (!stride.saturated && !srcStride.saturated && stride != srcStride)
return failure();
}
}
return StridedLayoutAttr::get(srcType.getContext(), srcOffset, resultStrides);
}
bool CollapseShapeOp::isGuaranteedCollapsible(
MemRefType srcType, ArrayRef<ReassociationIndices> reassociation) {
// MemRefs with identity layout are always collapsible.
if (srcType.getLayout().isIdentity())
return true;
return succeeded(computeCollapsedLayoutMap(srcType, reassociation,
/*strict=*/true));
}
MemRefType CollapseShapeOp::computeCollapsedType(
MemRefType srcType, ArrayRef<ReassociationIndices> reassociation) {
SmallVector<int64_t> resultShape;
resultShape.reserve(reassociation.size());
for (const ReassociationIndices &group : reassociation) {
auto groupSize = SaturatedInteger::wrap(1);
for (int64_t srcDim : group)
groupSize =
groupSize * SaturatedInteger::wrap(srcType.getDimSize(srcDim));
resultShape.push_back(groupSize.asInteger());
}
if (srcType.getLayout().isIdentity()) {
// If the source is contiguous (i.e., no layout map specified), so is the
// result.
MemRefLayoutAttrInterface layout;
return MemRefType::get(resultShape, srcType.getElementType(), layout,
srcType.getMemorySpace());
}
// Source may not be fully contiguous. Compute the layout map.
// Note: Dimensions that are collapsed into a single dim are assumed to be
// contiguous.
FailureOr<StridedLayoutAttr> computedLayout =
computeCollapsedLayoutMap(srcType, reassociation);
assert(succeeded(computedLayout) &&
"invalid source layout map or collapsing non-contiguous dims");
return MemRefType::get(resultShape, srcType.getElementType(), *computedLayout,
srcType.getMemorySpace());
}
void CollapseShapeOp::build(OpBuilder &b, OperationState &result, Value src,
ArrayRef<ReassociationIndices> reassociation,
ArrayRef<NamedAttribute> attrs) {
auto srcType = llvm::cast<MemRefType>(src.getType());
MemRefType resultType =
CollapseShapeOp::computeCollapsedType(srcType, reassociation);
build(b, result, resultType, src, attrs);
result.addAttribute(::mlir::getReassociationAttrName(),
getReassociationIndicesAttribute(b, reassociation));
}
LogicalResult CollapseShapeOp::verify() {
MemRefType srcType = getSrcType();
MemRefType resultType = getResultType();
if (srcType.getRank() < resultType.getRank()) {
auto r0 = srcType.getRank();
auto r1 = resultType.getRank();
return emitOpError("has source rank ")
<< r0 << " and result rank " << r1 << ". This is not a collapse ("
<< r0 << " < " << r1 << ").";
}
// Verify result shape.
if (failed(verifyCollapsedShape(getOperation(), resultType.getShape(),
srcType.getShape(), getReassociationIndices(),
/*allowMultipleDynamicDimsPerGroup=*/true)))
return failure();
// Compute expected result type (including layout map).
MemRefType expectedResultType;
if (srcType.getLayout().isIdentity()) {
// If the source is contiguous (i.e., no layout map specified), so is the
// result.
MemRefLayoutAttrInterface layout;
expectedResultType =
MemRefType::get(resultType.getShape(), srcType.getElementType(), layout,
srcType.getMemorySpace());
} else {
// Source may not be fully contiguous. Compute the layout map.
// Note: Dimensions that are collapsed into a single dim are assumed to be
// contiguous.
FailureOr<StridedLayoutAttr> computedLayout =
computeCollapsedLayoutMap(srcType, getReassociationIndices());
if (failed(computedLayout))
return emitOpError(
"invalid source layout map or collapsing non-contiguous dims");
expectedResultType =
MemRefType::get(resultType.getShape(), srcType.getElementType(),
*computedLayout, srcType.getMemorySpace());
}
if (expectedResultType != resultType)
return emitOpError("expected collapsed type to be ")
<< expectedResultType << " but found " << resultType;
return success();
}
struct CollapseShapeOpMemRefCastFolder
: public OpRewritePattern<CollapseShapeOp> {
public:
using OpRewritePattern<CollapseShapeOp>::OpRewritePattern;
LogicalResult matchAndRewrite(CollapseShapeOp op,
PatternRewriter &rewriter) const override {
auto cast = op.getOperand().getDefiningOp<CastOp>();
if (!cast)
return failure();
if (!CastOp::canFoldIntoConsumerOp(cast))
return failure();
Type newResultType = CollapseShapeOp::computeCollapsedType(
llvm::cast<MemRefType>(cast.getOperand().getType()),
op.getReassociationIndices());
if (newResultType == op.getResultType()) {
rewriter.modifyOpInPlace(
op, [&]() { op.getSrcMutable().assign(cast.getSource()); });
} else {
Value newOp = rewriter.create<CollapseShapeOp>(
op->getLoc(), cast.getSource(), op.getReassociationIndices());
rewriter.replaceOpWithNewOp<CastOp>(op, op.getType(), newOp);
}
return success();
}
};
void CollapseShapeOp::getCanonicalizationPatterns(RewritePatternSet &results,
MLIRContext *context) {
results.add<ComposeReassociativeReshapeOps<CollapseShapeOp>,
ComposeCollapseOfExpandOp<CollapseShapeOp, ExpandShapeOp, CastOp>,
CollapseShapeOpMemRefCastFolder>(context);
}
OpFoldResult ExpandShapeOp::fold(FoldAdaptor adaptor) {
return foldReshapeOp<ExpandShapeOp, CollapseShapeOp>(*this,
adaptor.getOperands());
}
OpFoldResult CollapseShapeOp::fold(FoldAdaptor adaptor) {
return foldReshapeOp<CollapseShapeOp, ExpandShapeOp>(*this,
adaptor.getOperands());
}
//===----------------------------------------------------------------------===//
// ReshapeOp
//===----------------------------------------------------------------------===//
void ReshapeOp::getAsmResultNames(
function_ref<void(Value, StringRef)> setNameFn) {
setNameFn(getResult(), "reshape");
}
LogicalResult ReshapeOp::verify() {
Type operandType = getSource().getType();
Type resultType = getResult().getType();
Type operandElementType =
llvm::cast<ShapedType>(operandType).getElementType();
Type resultElementType = llvm::cast<ShapedType>(resultType).getElementType();
if (operandElementType != resultElementType)
return emitOpError("element types of source and destination memref "
"types should be the same");
if (auto operandMemRefType = llvm::dyn_cast<MemRefType>(operandType))
if (!operandMemRefType.getLayout().isIdentity())
return emitOpError("source memref type should have identity affine map");
int64_t shapeSize =
llvm::cast<MemRefType>(getShape().getType()).getDimSize(0);
auto resultMemRefType = llvm::dyn_cast<MemRefType>(resultType);
if (resultMemRefType) {
if (!resultMemRefType.getLayout().isIdentity())
return emitOpError("result memref type should have identity affine map");
if (shapeSize == ShapedType::kDynamic)
return emitOpError("cannot use shape operand with dynamic length to "
"reshape to statically-ranked memref type");
if (shapeSize != resultMemRefType.getRank())
return emitOpError(
"length of shape operand differs from the result's memref rank");
}
return success();
}
//===----------------------------------------------------------------------===//
// StoreOp
//===----------------------------------------------------------------------===//
LogicalResult StoreOp::verify() {
if (getNumOperands() != 2 + getMemRefType().getRank())
return emitOpError("store index operand count not equal to memref rank");
return success();
}
LogicalResult StoreOp::fold(FoldAdaptor adaptor,
SmallVectorImpl<OpFoldResult> &results) {
/// store(memrefcast) -> store
return foldMemRefCast(*this, getValueToStore());
}
//===----------------------------------------------------------------------===//
// SubViewOp
//===----------------------------------------------------------------------===//
void SubViewOp::getAsmResultNames(
function_ref<void(Value, StringRef)> setNameFn) {
setNameFn(getResult(), "subview");
}
/// A subview result type can be fully inferred from the source type and the
/// static representation of offsets, sizes and strides. Special sentinels
/// encode the dynamic case.
Type SubViewOp::inferResultType(MemRefType sourceMemRefType,
ArrayRef<int64_t> staticOffsets,
ArrayRef<int64_t> staticSizes,
ArrayRef<int64_t> staticStrides) {
unsigned rank = sourceMemRefType.getRank();
(void)rank;
assert(staticOffsets.size() == rank && "staticOffsets length mismatch");
assert(staticSizes.size() == rank && "staticSizes length mismatch");
assert(staticStrides.size() == rank && "staticStrides length mismatch");
// Extract source offset and strides.
auto [sourceStrides, sourceOffset] = getStridesAndOffset(sourceMemRefType);
// Compute target offset whose value is:
// `sourceOffset + sum_i(staticOffset_i * sourceStrides_i)`.
int64_t targetOffset = sourceOffset;
for (auto it : llvm::zip(staticOffsets, sourceStrides)) {
auto staticOffset = std::get<0>(it), targetStride = std::get<1>(it);
targetOffset = (SaturatedInteger::wrap(targetOffset) +
SaturatedInteger::wrap(staticOffset) *
SaturatedInteger::wrap(targetStride))
.asInteger();
}
// Compute target stride whose value is:
// `sourceStrides_i * staticStrides_i`.
SmallVector<int64_t, 4> targetStrides;
targetStrides.reserve(staticOffsets.size());
for (auto it : llvm::zip(sourceStrides, staticStrides)) {
auto sourceStride = std::get<0>(it), staticStride = std::get<1>(it);
targetStrides.push_back((SaturatedInteger::wrap(sourceStride) *
SaturatedInteger::wrap(staticStride))
.asInteger());
}
// The type is now known.
return MemRefType::get(staticSizes, sourceMemRefType.getElementType(),
StridedLayoutAttr::get(sourceMemRefType.getContext(),
targetOffset, targetStrides),
sourceMemRefType.getMemorySpace());
}
Type SubViewOp::inferResultType(MemRefType sourceMemRefType,
ArrayRef<OpFoldResult> offsets,
ArrayRef<OpFoldResult> sizes,
ArrayRef<OpFoldResult> strides) {
SmallVector<int64_t> staticOffsets, staticSizes, staticStrides;
SmallVector<Value> dynamicOffsets, dynamicSizes, dynamicStrides;
dispatchIndexOpFoldResults(offsets, dynamicOffsets, staticOffsets);
dispatchIndexOpFoldResults(sizes, dynamicSizes, staticSizes);
dispatchIndexOpFoldResults(strides, dynamicStrides, staticStrides);
if (!hasValidSizesOffsets(staticOffsets))
return {};
if (!hasValidSizesOffsets(staticSizes))
return {};
if (!hasValidStrides(staticStrides))
return {};
return SubViewOp::inferResultType(sourceMemRefType, staticOffsets,
staticSizes, staticStrides);
}
Type SubViewOp::inferRankReducedResultType(ArrayRef<int64_t> resultShape,
MemRefType sourceRankedTensorType,
ArrayRef<int64_t> offsets,
ArrayRef<int64_t> sizes,
ArrayRef<int64_t> strides) {
auto inferredType = llvm::cast<MemRefType>(
inferResultType(sourceRankedTensorType, offsets, sizes, strides));
assert(inferredType.getRank() >= static_cast<int64_t>(resultShape.size()) &&
"expected ");
if (inferredType.getRank() == static_cast<int64_t>(resultShape.size()))
return inferredType;
// Compute which dimensions are dropped.
std::optional<llvm::SmallDenseSet<unsigned>> dimsToProject =
computeRankReductionMask(inferredType.getShape(), resultShape);
assert(dimsToProject.has_value() && "invalid rank reduction");
// Compute the layout and result type.
auto inferredLayout = llvm::cast<StridedLayoutAttr>(inferredType.getLayout());
SmallVector<int64_t> rankReducedStrides;
rankReducedStrides.reserve(resultShape.size());
for (auto [idx, value] : llvm::enumerate(inferredLayout.getStrides())) {
if (!dimsToProject->contains(idx))
rankReducedStrides.push_back(value);
}
return MemRefType::get(resultShape, inferredType.getElementType(),
StridedLayoutAttr::get(inferredLayout.getContext(),
inferredLayout.getOffset(),
rankReducedStrides),
inferredType.getMemorySpace());
}
Type SubViewOp::inferRankReducedResultType(ArrayRef<int64_t> resultShape,
MemRefType sourceRankedTensorType,
ArrayRef<OpFoldResult> offsets,
ArrayRef<OpFoldResult> sizes,
ArrayRef<OpFoldResult> strides) {
SmallVector<int64_t> staticOffsets, staticSizes, staticStrides;
SmallVector<Value> dynamicOffsets, dynamicSizes, dynamicStrides;
dispatchIndexOpFoldResults(offsets, dynamicOffsets, staticOffsets);
dispatchIndexOpFoldResults(sizes, dynamicSizes, staticSizes);
dispatchIndexOpFoldResults(strides, dynamicStrides, staticStrides);
return SubViewOp::inferRankReducedResultType(
resultShape, sourceRankedTensorType, staticOffsets, staticSizes,
staticStrides);
}
// Build a SubViewOp with mixed static and dynamic entries and custom result
// type. If the type passed is nullptr, it is inferred.
void SubViewOp::build(OpBuilder &b, OperationState &result,
MemRefType resultType, Value source,
ArrayRef<OpFoldResult> offsets,
ArrayRef<OpFoldResult> sizes,
ArrayRef<OpFoldResult> strides,
ArrayRef<NamedAttribute> attrs) {
SmallVector<int64_t> staticOffsets, staticSizes, staticStrides;
SmallVector<Value> dynamicOffsets, dynamicSizes, dynamicStrides;
dispatchIndexOpFoldResults(offsets, dynamicOffsets, staticOffsets);
dispatchIndexOpFoldResults(sizes, dynamicSizes, staticSizes);
dispatchIndexOpFoldResults(strides, dynamicStrides, staticStrides);
auto sourceMemRefType = llvm::cast<MemRefType>(source.getType());
// Structuring implementation this way avoids duplication between builders.
if (!resultType) {
resultType = llvm::cast<MemRefType>(SubViewOp::inferResultType(
sourceMemRefType, staticOffsets, staticSizes, staticStrides));
}
build(b, result, resultType, source, dynamicOffsets, dynamicSizes,
dynamicStrides, b.getDenseI64ArrayAttr(staticOffsets),
b.getDenseI64ArrayAttr(staticSizes),
b.getDenseI64ArrayAttr(staticStrides));
result.addAttributes(attrs);
}
// Build a SubViewOp with mixed static and dynamic entries and inferred result
// type.
void SubViewOp::build(OpBuilder &b, OperationState &result, Value source,
ArrayRef<OpFoldResult> offsets,
ArrayRef<OpFoldResult> sizes,
ArrayRef<OpFoldResult> strides,
ArrayRef<NamedAttribute> attrs) {
build(b, result, MemRefType(), source, offsets, sizes, strides, attrs);
}
// Build a SubViewOp with static entries and inferred result type.
void SubViewOp::build(OpBuilder &b, OperationState &result, Value source,
ArrayRef<int64_t> offsets, ArrayRef<int64_t> sizes,
ArrayRef<int64_t> strides,
ArrayRef<NamedAttribute> attrs) {
SmallVector<OpFoldResult> offsetValues = llvm::to_vector<4>(
llvm::map_range(offsets, [&](int64_t v) -> OpFoldResult {
return b.getI64IntegerAttr(v);
}));
SmallVector<OpFoldResult> sizeValues =
llvm::to_vector<4>(llvm::map_range(sizes, [&](int64_t v) -> OpFoldResult {
return b.getI64IntegerAttr(v);
}));
SmallVector<OpFoldResult> strideValues = llvm::to_vector<4>(
llvm::map_range(strides, [&](int64_t v) -> OpFoldResult {
return b.getI64IntegerAttr(v);
}));
build(b, result, source, offsetValues, sizeValues, strideValues, attrs);
}
// Build a SubViewOp with dynamic entries and custom result type. If the
// type passed is nullptr, it is inferred.
void SubViewOp::build(OpBuilder &b, OperationState &result,
MemRefType resultType, Value source,
ArrayRef<int64_t> offsets, ArrayRef<int64_t> sizes,
ArrayRef<int64_t> strides,
ArrayRef<NamedAttribute> attrs) {
SmallVector<OpFoldResult> offsetValues = llvm::to_vector<4>(
llvm::map_range(offsets, [&](int64_t v) -> OpFoldResult {
return b.getI64IntegerAttr(v);
}));
SmallVector<OpFoldResult> sizeValues =
llvm::to_vector<4>(llvm::map_range(sizes, [&](int64_t v) -> OpFoldResult {
return b.getI64IntegerAttr(v);
}));
SmallVector<OpFoldResult> strideValues = llvm::to_vector<4>(
llvm::map_range(strides, [&](int64_t v) -> OpFoldResult {
return b.getI64IntegerAttr(v);
}));
build(b, result, resultType, source, offsetValues, sizeValues, strideValues,
attrs);
}
// Build a SubViewOp with dynamic entries and custom result type. If the type
// passed is nullptr, it is inferred.
void SubViewOp::build(OpBuilder &b, OperationState &result,
MemRefType resultType, Value source, ValueRange offsets,
ValueRange sizes, ValueRange strides,
ArrayRef<NamedAttribute> attrs) {
SmallVector<OpFoldResult> offsetValues = llvm::to_vector<4>(
llvm::map_range(offsets, [](Value v) -> OpFoldResult { return v; }));
SmallVector<OpFoldResult> sizeValues = llvm::to_vector<4>(
llvm::map_range(sizes, [](Value v) -> OpFoldResult { return v; }));
SmallVector<OpFoldResult> strideValues = llvm::to_vector<4>(
llvm::map_range(strides, [](Value v) -> OpFoldResult { return v; }));
build(b, result, resultType, source, offsetValues, sizeValues, strideValues);
}
// Build a SubViewOp with dynamic entries and inferred result type.
void SubViewOp::build(OpBuilder &b, OperationState &result, Value source,
ValueRange offsets, ValueRange sizes, ValueRange strides,
ArrayRef<NamedAttribute> attrs) {
build(b, result, MemRefType(), source, offsets, sizes, strides, attrs);
}
/// For ViewLikeOpInterface.
Value SubViewOp::getViewSource() { return getSource(); }
/// Return true if `t1` and `t2` have equal offsets (both dynamic or of same
/// static value).
static bool haveCompatibleOffsets(MemRefType t1, MemRefType t2) {
int64_t t1Offset, t2Offset;
SmallVector<int64_t> t1Strides, t2Strides;
auto res1 = getStridesAndOffset(t1, t1Strides, t1Offset);
auto res2 = getStridesAndOffset(t2, t2Strides, t2Offset);
return succeeded(res1) && succeeded(res2) && t1Offset == t2Offset;
}
/// Return true if `t1` and `t2` have equal strides (both dynamic or of same
/// static value). Dimensions of `t1` may be dropped in `t2`; these must be
/// marked as dropped in `droppedDims`.
static bool haveCompatibleStrides(MemRefType t1, MemRefType t2,
const llvm::SmallBitVector &droppedDims) {
assert(size_t(t1.getRank()) == droppedDims.size() && "incorrect number of bits");
assert(size_t(t1.getRank() - t2.getRank()) == droppedDims.count() &&
"incorrect number of dropped dims");
int64_t t1Offset, t2Offset;
SmallVector<int64_t> t1Strides, t2Strides;
auto res1 = getStridesAndOffset(t1, t1Strides, t1Offset);
auto res2 = getStridesAndOffset(t2, t2Strides, t2Offset);
if (failed(res1) || failed(res2))
return false;
for (int64_t i = 0, j = 0, e = t1.getRank(); i < e; ++i) {
if (droppedDims[i])
continue;
if (t1Strides[i] != t2Strides[j])
return false;
++j;
}
return true;
}
static LogicalResult produceSubViewErrorMsg(SliceVerificationResult result,
Operation *op, Type expectedType) {
auto memrefType = llvm::cast<ShapedType>(expectedType);
switch (result) {
case SliceVerificationResult::Success:
return success();
case SliceVerificationResult::RankTooLarge:
return op->emitError("expected result rank to be smaller or equal to ")
<< "the source rank. ";
case SliceVerificationResult::SizeMismatch:
return op->emitError("expected result type to be ")
<< expectedType
<< " or a rank-reduced version. (mismatch of result sizes) ";
case SliceVerificationResult::ElemTypeMismatch:
return op->emitError("expected result element type to be ")
<< memrefType.getElementType();
case SliceVerificationResult::MemSpaceMismatch:
return op->emitError("expected result and source memory spaces to match.");
case SliceVerificationResult::LayoutMismatch:
return op->emitError("expected result type to be ")
<< expectedType
<< " or a rank-reduced version. (mismatch of result layout) ";
}
llvm_unreachable("unexpected subview verification result");
}
/// Verifier for SubViewOp.
LogicalResult SubViewOp::verify() {
MemRefType baseType = getSourceType();
MemRefType subViewType = getType();
// The base memref and the view memref should be in the same memory space.
if (baseType.getMemorySpace() != subViewType.getMemorySpace())
return emitError("different memory spaces specified for base memref "
"type ")
<< baseType << " and subview memref type " << subViewType;
// Verify that the base memref type has a strided layout map.
if (!isStrided(baseType))
return emitError("base type ") << baseType << " is not strided";
// Compute the expected result type, assuming that there are no rank
// reductions.
auto expectedType = cast<MemRefType>(SubViewOp::inferResultType(
baseType, getStaticOffsets(), getStaticSizes(), getStaticStrides()));
// Verify all properties of a shaped type: rank, element type and dimension
// sizes. This takes into account potential rank reductions.
auto shapedTypeVerification = isRankReducedType(
/*originalType=*/expectedType, /*candidateReducedType=*/subViewType);
if (shapedTypeVerification != SliceVerificationResult::Success)
return produceSubViewErrorMsg(shapedTypeVerification, *this, expectedType);
// Make sure that the memory space did not change.
if (expectedType.getMemorySpace() != subViewType.getMemorySpace())
return produceSubViewErrorMsg(SliceVerificationResult::MemSpaceMismatch,
*this, expectedType);
// Verify the offset of the layout map.
if (!haveCompatibleOffsets(expectedType, subViewType))
return produceSubViewErrorMsg(SliceVerificationResult::LayoutMismatch,
*this, expectedType);
// The only thing that's left to verify now are the strides. First, compute
// the unused dimensions due to rank reductions. We have to look at sizes and
// strides to decide which dimensions were dropped. This function also
// partially verifies strides in case of rank reductions.
auto unusedDims = computeMemRefRankReductionMask(expectedType, subViewType,
getMixedSizes());
if (failed(unusedDims))
return produceSubViewErrorMsg(SliceVerificationResult::LayoutMismatch,
*this, expectedType);
// Strides must match.
if (!haveCompatibleStrides(expectedType, subViewType, *unusedDims))
return produceSubViewErrorMsg(SliceVerificationResult::LayoutMismatch,
*this, expectedType);
return success();
}
raw_ostream &mlir::operator<<(raw_ostream &os, const Range &range) {
return os << "range " << range.offset << ":" << range.size << ":"
<< range.stride;
}
/// Return the list of Range (i.e. offset, size, stride). Each Range
/// entry contains either the dynamic value or a ConstantIndexOp constructed
/// with `b` at location `loc`.
SmallVector<Range, 8> mlir::getOrCreateRanges(OffsetSizeAndStrideOpInterface op,
OpBuilder &b, Location loc) {
std::array<unsigned, 3> ranks = op.getArrayAttrMaxRanks();
assert(ranks[0] == ranks[1] && "expected offset and sizes of equal ranks");
assert(ranks[1] == ranks[2] && "expected sizes and strides of equal ranks");
SmallVector<Range, 8> res;
unsigned rank = ranks[0];
res.reserve(rank);
for (unsigned idx = 0; idx < rank; ++idx) {
Value offset =
op.isDynamicOffset(idx)
? op.getDynamicOffset(idx)
: b.create<arith::ConstantIndexOp>(loc, op.getStaticOffset(idx));
Value size =
op.isDynamicSize(idx)
? op.getDynamicSize(idx)
: b.create<arith::ConstantIndexOp>(loc, op.getStaticSize(idx));
Value stride =
op.isDynamicStride(idx)
? op.getDynamicStride(idx)
: b.create<arith::ConstantIndexOp>(loc, op.getStaticStride(idx));
res.emplace_back(Range{offset, size, stride});
}
return res;
}
/// Compute the canonical result type of a SubViewOp. Call `inferResultType`
/// to deduce the result type for the given `sourceType`. Additionally, reduce
/// the rank of the inferred result type if `currentResultType` is lower rank
/// than `currentSourceType`. Use this signature if `sourceType` is updated
/// together with the result type. In this case, it is important to compute
/// the dropped dimensions using `currentSourceType` whose strides align with
/// `currentResultType`.
static MemRefType getCanonicalSubViewResultType(
MemRefType currentResultType, MemRefType currentSourceType,
MemRefType sourceType, ArrayRef<OpFoldResult> mixedOffsets,
ArrayRef<OpFoldResult> mixedSizes, ArrayRef<OpFoldResult> mixedStrides) {
auto nonRankReducedType = llvm::cast<MemRefType>(SubViewOp::inferResultType(
sourceType, mixedOffsets, mixedSizes, mixedStrides));
FailureOr<llvm::SmallBitVector> unusedDims = computeMemRefRankReductionMask(
currentSourceType, currentResultType, mixedSizes);
if (failed(unusedDims))
return nullptr;
auto layout = llvm::cast<StridedLayoutAttr>(nonRankReducedType.getLayout());
SmallVector<int64_t> shape, strides;
unsigned numDimsAfterReduction =
nonRankReducedType.getRank() - unusedDims->count();
shape.reserve(numDimsAfterReduction);
strides.reserve(numDimsAfterReduction);
for (const auto &[idx, size, stride] :
llvm::zip(llvm::seq<unsigned>(0, nonRankReducedType.getRank()),
nonRankReducedType.getShape(), layout.getStrides())) {
if (unusedDims->test(idx))
continue;
shape.push_back(size);
strides.push_back(stride);
}
return MemRefType::get(shape, nonRankReducedType.getElementType(),
StridedLayoutAttr::get(sourceType.getContext(),
layout.getOffset(), strides),
nonRankReducedType.getMemorySpace());
}
Value mlir::memref::createCanonicalRankReducingSubViewOp(
OpBuilder &b, Location loc, Value memref, ArrayRef<int64_t> targetShape) {
auto memrefType = llvm::cast<MemRefType>(memref.getType());
unsigned rank = memrefType.getRank();
SmallVector<OpFoldResult> offsets(rank, b.getIndexAttr(0));
SmallVector<OpFoldResult> sizes = getMixedSizes(b, loc, memref);
SmallVector<OpFoldResult> strides(rank, b.getIndexAttr(1));
auto targetType =
llvm::cast<MemRefType>(SubViewOp::inferRankReducedResultType(
targetShape, memrefType, offsets, sizes, strides));
return b.createOrFold<memref::SubViewOp>(loc, targetType, memref, offsets,
sizes, strides);
}
FailureOr<Value> SubViewOp::rankReduceIfNeeded(OpBuilder &b, Location loc,
Value value,
ArrayRef<int64_t> desiredShape) {
auto sourceMemrefType = llvm::dyn_cast<MemRefType>(value.getType());
assert(sourceMemrefType && "not a ranked memref type");
auto sourceShape = sourceMemrefType.getShape();
if (sourceShape.equals(desiredShape))
return value;
auto maybeRankReductionMask =
mlir::computeRankReductionMask(sourceShape, desiredShape);
if (!maybeRankReductionMask)
return failure();
return createCanonicalRankReducingSubViewOp(b, loc, value, desiredShape);
}
/// Helper method to check if a `subview` operation is trivially a no-op. This
/// is the case if the all offsets are zero, all strides are 1, and the source
/// shape is same as the size of the subview. In such cases, the subview can
/// be folded into its source.
static bool isTrivialSubViewOp(SubViewOp subViewOp) {
if (subViewOp.getSourceType().getRank() != subViewOp.getType().getRank())
return false;
auto mixedOffsets = subViewOp.getMixedOffsets();
auto mixedSizes = subViewOp.getMixedSizes();
auto mixedStrides = subViewOp.getMixedStrides();
// Check offsets are zero.
if (llvm::any_of(mixedOffsets, [](OpFoldResult ofr) {
std::optional<int64_t> intValue = getConstantIntValue(ofr);
return !intValue || intValue.value() != 0;
}))
return false;
// Check strides are one.
if (llvm::any_of(mixedStrides, [](OpFoldResult ofr) {
std::optional<int64_t> intValue = getConstantIntValue(ofr);
return !intValue || intValue.value() != 1;
}))
return false;
// Check all size values are static and matches the (static) source shape.
ArrayRef<int64_t> sourceShape = subViewOp.getSourceType().getShape();
for (const auto &size : llvm::enumerate(mixedSizes)) {
std::optional<int64_t> intValue = getConstantIntValue(size.value());
if (!intValue || *intValue != sourceShape[size.index()])
return false;
}
// All conditions met. The `SubViewOp` is foldable as a no-op.
return true;
}
namespace {
/// Pattern to rewrite a subview op with MemRefCast arguments.
/// This essentially pushes memref.cast past its consuming subview when
/// `canFoldIntoConsumerOp` is true.
///
/// Example:
/// ```
/// %0 = memref.cast %V : memref<16x16xf32> to memref<?x?xf32>
/// %1 = memref.subview %0[0, 0][3, 4][1, 1] :
/// memref<?x?xf32> to memref<3x4xf32, strided<[?, 1], offset: ?>>
/// ```
/// is rewritten into:
/// ```
/// %0 = memref.subview %V: memref<16x16xf32> to memref<3x4xf32, #[[map0]]>
/// %1 = memref.cast %0: memref<3x4xf32, strided<[16, 1], offset: 0>> to
/// memref<3x4xf32, strided<[?, 1], offset: ?>>
/// ```
class SubViewOpMemRefCastFolder final : public OpRewritePattern<SubViewOp> {
public:
using OpRewritePattern<SubViewOp>::OpRewritePattern;
LogicalResult matchAndRewrite(SubViewOp subViewOp,
PatternRewriter &rewriter) const override {
// Any constant operand, just return to let SubViewOpConstantFolder kick
// in.
if (llvm::any_of(subViewOp.getOperands(), [](Value operand) {
return matchPattern(operand, matchConstantIndex());
}))
return failure();
auto castOp = subViewOp.getSource().getDefiningOp<CastOp>();
if (!castOp)
return failure();
if (!CastOp::canFoldIntoConsumerOp(castOp))
return failure();
// Compute the SubViewOp result type after folding the MemRefCastOp. Use
// the MemRefCastOp source operand type to infer the result type and the
// current SubViewOp source operand type to compute the dropped dimensions
// if the operation is rank-reducing.
auto resultType = getCanonicalSubViewResultType(
subViewOp.getType(), subViewOp.getSourceType(),
llvm::cast<MemRefType>(castOp.getSource().getType()),
subViewOp.getMixedOffsets(), subViewOp.getMixedSizes(),
subViewOp.getMixedStrides());
if (!resultType)
return failure();
Value newSubView = rewriter.create<SubViewOp>(
subViewOp.getLoc(), resultType, castOp.getSource(),
subViewOp.getOffsets(), subViewOp.getSizes(), subViewOp.getStrides(),
subViewOp.getStaticOffsets(), subViewOp.getStaticSizes(),
subViewOp.getStaticStrides());
rewriter.replaceOpWithNewOp<CastOp>(subViewOp, subViewOp.getType(),
newSubView);
return success();
}
};
/// Canonicalize subview ops that are no-ops. When the source shape is not
/// same as a result shape due to use of `affine_map`.
class TrivialSubViewOpFolder final : public OpRewritePattern<SubViewOp> {
public:
using OpRewritePattern<SubViewOp>::OpRewritePattern;
LogicalResult matchAndRewrite(SubViewOp subViewOp,
PatternRewriter &rewriter) const override {
if (!isTrivialSubViewOp(subViewOp))
return failure();
if (subViewOp.getSourceType() == subViewOp.getType()) {
rewriter.replaceOp(subViewOp, subViewOp.getSource());
return success();
}
rewriter.replaceOpWithNewOp<CastOp>(subViewOp, subViewOp.getType(),
subViewOp.getSource());
return success();
}
};
} // namespace
/// Return the canonical type of the result of a subview.
struct SubViewReturnTypeCanonicalizer {
MemRefType operator()(SubViewOp op, ArrayRef<OpFoldResult> mixedOffsets,
ArrayRef<OpFoldResult> mixedSizes,
ArrayRef<OpFoldResult> mixedStrides) {
// Infer a memref type without taking into account any rank reductions.
auto resTy = SubViewOp::inferResultType(op.getSourceType(), mixedOffsets,
mixedSizes, mixedStrides);
if (!resTy)
return {};
MemRefType nonReducedType = cast<MemRefType>(resTy);
// Directly return the non-rank reduced type if there are no dropped dims.
llvm::SmallBitVector droppedDims = op.getDroppedDims();
if (droppedDims.none())
return nonReducedType;
// Take the strides and offset from the non-rank reduced type.
auto [nonReducedStrides, offset] = getStridesAndOffset(nonReducedType);
// Drop dims from shape and strides.
SmallVector<int64_t> targetShape;
SmallVector<int64_t> targetStrides;
for (int64_t i = 0; i < static_cast<int64_t>(mixedSizes.size()); ++i) {
if (droppedDims.test(i))
continue;
targetStrides.push_back(nonReducedStrides[i]);
targetShape.push_back(nonReducedType.getDimSize(i));
}
return MemRefType::get(targetShape, nonReducedType.getElementType(),
StridedLayoutAttr::get(nonReducedType.getContext(),
offset, targetStrides),
nonReducedType.getMemorySpace());
}
};
/// A canonicalizer wrapper to replace SubViewOps.
struct SubViewCanonicalizer {
void operator()(PatternRewriter &rewriter, SubViewOp op, SubViewOp newOp) {
rewriter.replaceOpWithNewOp<CastOp>(op, op.getType(), newOp);
}
};
void SubViewOp::getCanonicalizationPatterns(RewritePatternSet &results,
MLIRContext *context) {
results
.add<OpWithOffsetSizesAndStridesConstantArgumentFolder<
SubViewOp, SubViewReturnTypeCanonicalizer, SubViewCanonicalizer>,
SubViewOpMemRefCastFolder, TrivialSubViewOpFolder>(context);
}
OpFoldResult SubViewOp::fold(FoldAdaptor adaptor) {
auto resultShapedType = llvm::cast<ShapedType>(getResult().getType());
auto sourceShapedType = llvm::cast<ShapedType>(getSource().getType());
if (resultShapedType.hasStaticShape() &&
resultShapedType == sourceShapedType) {
return getViewSource();
}
// Fold subview(subview(x)), where both subviews have the same size and the
// second subview's offsets are all zero. (I.e., the second subview is a
// no-op.)
if (auto srcSubview = getViewSource().getDefiningOp<SubViewOp>()) {
auto srcSizes = srcSubview.getMixedSizes();
auto sizes = getMixedSizes();
auto offsets = getMixedOffsets();
bool allOffsetsZero = llvm::all_of(
offsets, [](OpFoldResult ofr) { return isConstantIntValue(ofr, 0); });
auto strides = getMixedStrides();
bool allStridesOne = llvm::all_of(
strides, [](OpFoldResult ofr) { return isConstantIntValue(ofr, 1); });
bool allSizesSame = llvm::equal(sizes, srcSizes);
if (allOffsetsZero && allStridesOne && allSizesSame &&
resultShapedType == sourceShapedType)
return getViewSource();
}
return {};
}
//===----------------------------------------------------------------------===//
// TransposeOp
//===----------------------------------------------------------------------===//
void TransposeOp::getAsmResultNames(
function_ref<void(Value, StringRef)> setNameFn) {
setNameFn(getResult(), "transpose");
}
/// Build a strided memref type by applying `permutationMap` to `memRefType`.
static MemRefType inferTransposeResultType(MemRefType memRefType,
AffineMap permutationMap) {
auto originalSizes = memRefType.getShape();
auto [originalStrides, offset] = getStridesAndOffset(memRefType);
assert(originalStrides.size() == static_cast<unsigned>(memRefType.getRank()));
// Compute permuted sizes and strides.
auto sizes = applyPermutationMap<int64_t>(permutationMap, originalSizes);
auto strides = applyPermutationMap<int64_t>(permutationMap, originalStrides);
return MemRefType::Builder(memRefType)
.setShape(sizes)
.setLayout(
StridedLayoutAttr::get(memRefType.getContext(), offset, strides));
}
void TransposeOp::build(OpBuilder &b, OperationState &result, Value in,
AffineMapAttr permutation,
ArrayRef<NamedAttribute> attrs) {
auto permutationMap = permutation.getValue();
assert(permutationMap);
auto memRefType = llvm::cast<MemRefType>(in.getType());
// Compute result type.
MemRefType resultType = inferTransposeResultType(memRefType, permutationMap);
build(b, result, resultType, in, attrs);
result.addAttribute(TransposeOp::getPermutationAttrStrName(), permutation);
}
// transpose $in $permutation attr-dict : type($in) `to` type(results)
void TransposeOp::print(OpAsmPrinter &p) {
p << " " << getIn() << " " << getPermutation();
p.printOptionalAttrDict((*this)->getAttrs(), {getPermutationAttrStrName()});
p << " : " << getIn().getType() << " to " << getType();
}
ParseResult TransposeOp::parse(OpAsmParser &parser, OperationState &result) {
OpAsmParser::UnresolvedOperand in;
AffineMap permutation;
MemRefType srcType, dstType;
if (parser.parseOperand(in) || parser.parseAffineMap(permutation) ||
parser.parseOptionalAttrDict(result.attributes) ||
parser.parseColonType(srcType) ||
parser.resolveOperand(in, srcType, result.operands) ||
parser.parseKeywordType("to", dstType) ||
parser.addTypeToList(dstType, result.types))
return failure();
result.addAttribute(TransposeOp::getPermutationAttrStrName(),
AffineMapAttr::get(permutation));
return success();
}
LogicalResult TransposeOp::verify() {
if (!getPermutation().isPermutation())
return emitOpError("expected a permutation map");
if (getPermutation().getNumDims() != getIn().getType().getRank())
return emitOpError("expected a permutation map of same rank as the input");
auto srcType = llvm::cast<MemRefType>(getIn().getType());
auto resultType = llvm::cast<MemRefType>(getType());
auto canonicalResultType = canonicalizeStridedLayout(
inferTransposeResultType(srcType, getPermutation()));
if (canonicalizeStridedLayout(resultType) != canonicalResultType)
return emitOpError("result type ")
<< resultType
<< " is not equivalent to the canonical transposed input type "
<< canonicalResultType;
return success();
}
OpFoldResult TransposeOp::fold(FoldAdaptor) {
// First check for identity permutation, we can fold it away if input and
// result types are identical already.
if (getPermutation().isIdentity() && getType() == getIn().getType())
return getIn();
// Fold two consecutive memref.transpose Ops into one by composing their
// permutation maps.
if (auto otherTransposeOp = getIn().getDefiningOp<memref::TransposeOp>()) {
AffineMap composedPermutation =
getPermutation().compose(otherTransposeOp.getPermutation());
getInMutable().assign(otherTransposeOp.getIn());
setPermutation(composedPermutation);
return getResult();
}
return {};
}
//===----------------------------------------------------------------------===//
// ViewOp
//===----------------------------------------------------------------------===//
void ViewOp::getAsmResultNames(function_ref<void(Value, StringRef)> setNameFn) {
setNameFn(getResult(), "view");
}
LogicalResult ViewOp::verify() {
auto baseType = llvm::cast<MemRefType>(getOperand(0).getType());
auto viewType = getType();
// The base memref should have identity layout map (or none).
if (!baseType.getLayout().isIdentity())
return emitError("unsupported map for base memref type ") << baseType;
// The result memref should have identity layout map (or none).
if (!viewType.getLayout().isIdentity())
return emitError("unsupported map for result memref type ") << viewType;
// The base memref and the view memref should be in the same memory space.
if (baseType.getMemorySpace() != viewType.getMemorySpace())
return emitError("different memory spaces specified for base memref "
"type ")
<< baseType << " and view memref type " << viewType;
// Verify that we have the correct number of sizes for the result type.
unsigned numDynamicDims = viewType.getNumDynamicDims();
if (getSizes().size() != numDynamicDims)
return emitError("incorrect number of size operands for type ") << viewType;
return success();
}
Value ViewOp::getViewSource() { return getSource(); }
namespace {
struct ViewOpShapeFolder : public OpRewritePattern<ViewOp> {
using OpRewritePattern<ViewOp>::OpRewritePattern;
LogicalResult matchAndRewrite(ViewOp viewOp,
PatternRewriter &rewriter) const override {
// Return if none of the operands are constants.
if (llvm::none_of(viewOp.getOperands(), [](Value operand) {
return matchPattern(operand, matchConstantIndex());
}))
return failure();
// Get result memref type.
auto memrefType = viewOp.getType();
// Get offset from old memref view type 'memRefType'.
int64_t oldOffset;
SmallVector<int64_t, 4> oldStrides;
if (failed(getStridesAndOffset(memrefType, oldStrides, oldOffset)))
return failure();
assert(oldOffset == 0 && "Expected 0 offset");
SmallVector<Value, 4> newOperands;
// Offset cannot be folded into result type.
// Fold any dynamic dim operands which are produced by a constant.
SmallVector<int64_t, 4> newShapeConstants;
newShapeConstants.reserve(memrefType.getRank());
unsigned dynamicDimPos = 0;
unsigned rank = memrefType.getRank();
for (unsigned dim = 0, e = rank; dim < e; ++dim) {
int64_t dimSize = memrefType.getDimSize(dim);
// If this is already static dimension, keep it.
if (!ShapedType::isDynamic(dimSize)) {
newShapeConstants.push_back(dimSize);
continue;
}
auto *defOp = viewOp.getSizes()[dynamicDimPos].getDefiningOp();
if (auto constantIndexOp =
dyn_cast_or_null<arith::ConstantIndexOp>(defOp)) {
// Dynamic shape dimension will be folded.
newShapeConstants.push_back(constantIndexOp.value());
} else {
// Dynamic shape dimension not folded; copy operand from old memref.
newShapeConstants.push_back(dimSize);
newOperands.push_back(viewOp.getSizes()[dynamicDimPos]);
}
dynamicDimPos++;
}
// Create new memref type with constant folded dims.
MemRefType newMemRefType =
MemRefType::Builder(memrefType).setShape(newShapeConstants);
// Nothing new, don't fold.
if (newMemRefType == memrefType)
return failure();
// Create new ViewOp.
auto newViewOp = rewriter.create<ViewOp>(
viewOp.getLoc(), newMemRefType, viewOp.getOperand(0),
viewOp.getByteShift(), newOperands);
// Insert a cast so we have the same type as the old memref type.
rewriter.replaceOpWithNewOp<CastOp>(viewOp, viewOp.getType(), newViewOp);
return success();
}
};
struct ViewOpMemrefCastFolder : public OpRewritePattern<ViewOp> {
using OpRewritePattern<ViewOp>::OpRewritePattern;
LogicalResult matchAndRewrite(ViewOp viewOp,
PatternRewriter &rewriter) const override {
Value memrefOperand = viewOp.getOperand(0);
CastOp memrefCastOp = memrefOperand.getDefiningOp<CastOp>();
if (!memrefCastOp)
return failure();
Value allocOperand = memrefCastOp.getOperand();
AllocOp allocOp = allocOperand.getDefiningOp<AllocOp>();
if (!allocOp)
return failure();
rewriter.replaceOpWithNewOp<ViewOp>(viewOp, viewOp.getType(), allocOperand,
viewOp.getByteShift(),
viewOp.getSizes());
return success();
}
};
} // namespace
void ViewOp::getCanonicalizationPatterns(RewritePatternSet &results,
MLIRContext *context) {
results.add<ViewOpShapeFolder, ViewOpMemrefCastFolder>(context);
}
//===----------------------------------------------------------------------===//
// AtomicRMWOp
//===----------------------------------------------------------------------===//
LogicalResult AtomicRMWOp::verify() {
if (getMemRefType().getRank() != getNumOperands() - 2)
return emitOpError(
"expects the number of subscripts to be equal to memref rank");
switch (getKind()) {
case arith::AtomicRMWKind::addf:
case arith::AtomicRMWKind::maximumf:
case arith::AtomicRMWKind::minimumf:
case arith::AtomicRMWKind::mulf:
if (!llvm::isa<FloatType>(getValue().getType()))
return emitOpError() << "with kind '"
<< arith::stringifyAtomicRMWKind(getKind())
<< "' expects a floating-point type";
break;
case arith::AtomicRMWKind::addi:
case arith::AtomicRMWKind::maxs:
case arith::AtomicRMWKind::maxu:
case arith::AtomicRMWKind::mins:
case arith::AtomicRMWKind::minu:
case arith::AtomicRMWKind::muli:
case arith::AtomicRMWKind::ori:
case arith::AtomicRMWKind::andi:
if (!llvm::isa<IntegerType>(getValue().getType()))
return emitOpError() << "with kind '"
<< arith::stringifyAtomicRMWKind(getKind())
<< "' expects an integer type";
break;
default:
break;
}
return success();
}
OpFoldResult AtomicRMWOp::fold(FoldAdaptor adaptor) {
/// atomicrmw(memrefcast) -> atomicrmw
if (succeeded(foldMemRefCast(*this, getValue())))
return getResult();
return OpFoldResult();
}
//===----------------------------------------------------------------------===//
// TableGen'd op method definitions
//===----------------------------------------------------------------------===//
#define GET_OP_CLASSES
#include "mlir/Dialect/MemRef/IR/MemRefOps.cpp.inc"