Skip to content

Commit

Permalink
[MLIR] Fix VectorEmulateNarrowType constant op mask bug (llvm#116064)
Browse files Browse the repository at this point in the history 
This commit adds support for handling mask constants generated by the
`arith.constant` op in the `VectorEmulateNarrowType` pattern.
Previously, this pattern would not match due to the lack of mask
constant handling in `getCompressedMaskOp`.

The changes include:

1. Updating `getCompressedMaskOp` to recognize and handle
`arith.constant` ops as mask value sources.

2. Handling cases where the mask is not aligned with the emulated load
width. The compressed mask is adjusted to account for the offset.

Limitations:
- The arith.constant op can only have 1-dimensional constant values.

Resolves: llvm#115742

Signed-off-by: Alan Li <[email protected]>
  • Loading branch information
@lialan
lialan authored Nov 15, 2024
1 parent 0398cb4 commit ef92aba
Show file tree
Hide file tree
Showing 3 changed files with 198 additions and 60 deletions.
169 changes: 109 additions & 60 deletions mlir/lib/Dialect/Vector/Transforms/VectorEmulateNarrowType.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -75,83 +75,134 @@ static FailureOr<Operation *> getCompressedMaskOp(OpBuilder &rewriter,
int numSrcElemsPerDest,
int numFrontPadElems = 0) {

assert(numFrontPadElems < numSrcElemsPerDest && "intraDataOffset must be less than scale");
assert(numFrontPadElems < numSrcElemsPerDest &&
"numFrontPadElems must be less than numSrcElemsPerDest");

auto numElements = (numFrontPadElems + numSrcElems + numSrcElemsPerDest - 1) /
numSrcElemsPerDest;
auto numDestElems =
(numFrontPadElems + numSrcElems + numSrcElemsPerDest - 1) /
numSrcElemsPerDest;

Operation *maskOp = mask.getDefiningOp();
SmallVector<vector::ExtractOp, 2> extractOps;
// TODO: add support to `vector.splat`.
// Finding the mask creation operation.
while (maskOp && !isa<vector::CreateMaskOp, vector::ConstantMaskOp>(maskOp)) {
while (maskOp &&
!isa<arith::ConstantOp, vector::CreateMaskOp, vector::ConstantMaskOp>(
maskOp)) {
if (auto extractOp = dyn_cast<vector::ExtractOp>(maskOp)) {
maskOp = extractOp.getVector().getDefiningOp();
extractOps.push_back(extractOp);
}
}
auto createMaskOp = dyn_cast_or_null<vector::CreateMaskOp>(maskOp);
auto constantMaskOp = dyn_cast_or_null<vector::ConstantMaskOp>(maskOp);
if (!createMaskOp && !constantMaskOp)

if (!isa<arith::ConstantOp, vector::CreateMaskOp, vector::ConstantMaskOp>(
maskOp))
return failure();

// Computing the "compressed" mask. All the emulation logic (i.e. computing
// new mask index) only happens on the last dimension of the vectors.
Operation *newMask = nullptr;
SmallVector<int64_t> shape(
SmallVector<int64_t> maskShape(
cast<VectorType>(maskOp->getResultTypes()[0]).getShape());
shape.back() = numElements;
auto newMaskType = VectorType::get(shape, rewriter.getI1Type());
if (createMaskOp) {
OperandRange maskOperands = createMaskOp.getOperands();
size_t numMaskOperands = maskOperands.size();
AffineExpr s0;
bindSymbols(rewriter.getContext(), s0);
s0 = s0 + numSrcElemsPerDest - 1;
s0 = s0.floorDiv(numSrcElemsPerDest);
OpFoldResult origIndex =
getAsOpFoldResult(maskOperands[numMaskOperands - 1]);
OpFoldResult maskIndex =
affine::makeComposedFoldedAffineApply(rewriter, loc, s0, origIndex);
SmallVector<Value> newMaskOperands(maskOperands.drop_back());
newMaskOperands.push_back(
getValueOrCreateConstantIndexOp(rewriter, loc, maskIndex));
newMask = rewriter.create<vector::CreateMaskOp>(loc, newMaskType,
newMaskOperands);
} else if (constantMaskOp) {
ArrayRef<int64_t> maskDimSizes = constantMaskOp.getMaskDimSizes();
size_t numMaskOperands = maskDimSizes.size();
int64_t origIndex = maskDimSizes[numMaskOperands - 1];
int64_t startIndex = numFrontPadElems / numSrcElemsPerDest;
int64_t maskIndex =
llvm::divideCeil(numFrontPadElems + origIndex, numSrcElemsPerDest);

// TODO: we only want the mask between [startIndex, maskIndex] to be true,
// the rest are false.
if (numFrontPadElems != 0 && maskDimSizes.size() > 1)
return failure();

SmallVector<int64_t> newMaskDimSizes(maskDimSizes.drop_back());
newMaskDimSizes.push_back(maskIndex);

if (numFrontPadElems == 0) {
newMask = rewriter.create<vector::ConstantMaskOp>(loc, newMaskType,
newMaskDimSizes);
} else {
SmallVector<bool> newMaskValues;
for (int64_t i = 0; i < numElements; ++i)
newMaskValues.push_back(i >= startIndex && i < maskIndex);
auto denseAttr = DenseElementsAttr::get(newMaskType, newMaskValues);
newMask = rewriter.create<arith::ConstantOp>(loc, newMaskType, denseAttr);
}
}
maskShape.back() = numDestElems;
auto newMaskType = VectorType::get(maskShape, rewriter.getI1Type());
std::optional<Operation *> newMask =
TypeSwitch<Operation *, std::optional<Operation *>>(maskOp)
.Case<vector::CreateMaskOp>(
[&](auto createMaskOp) -> std::optional<Operation *> {
OperandRange maskOperands = createMaskOp.getOperands();
size_t numMaskOperands = maskOperands.size();
AffineExpr s0;
bindSymbols(rewriter.getContext(), s0);
s0 = s0 + numSrcElemsPerDest - 1;
s0 = s0.floorDiv(numSrcElemsPerDest);
OpFoldResult origIndex =
getAsOpFoldResult(maskOperands[numMaskOperands - 1]);
OpFoldResult maskIndex = affine::makeComposedFoldedAffineApply(
rewriter, loc, s0, origIndex);
SmallVector<Value> newMaskOperands(maskOperands.drop_back());
newMaskOperands.push_back(
getValueOrCreateConstantIndexOp(rewriter, loc, maskIndex));
return rewriter.create<vector::CreateMaskOp>(loc, newMaskType,
newMaskOperands);
})
.Case<vector::ConstantMaskOp>([&](auto constantMaskOp)
-> std::optional<Operation *> {
ArrayRef<int64_t> maskDimSizes = constantMaskOp.getMaskDimSizes();
size_t numMaskOperands = maskDimSizes.size();
int64_t origIndex = maskDimSizes[numMaskOperands - 1];
int64_t startIndex = numFrontPadElems / numSrcElemsPerDest;
int64_t maskIndex = llvm::divideCeil(numFrontPadElems + origIndex,
numSrcElemsPerDest);

// TODO: we only want the mask between [startIndex, maskIndex]
// to be true, the rest are false.
if (numFrontPadElems != 0 && maskDimSizes.size() > 1)
return std::nullopt;

SmallVector<int64_t> newMaskDimSizes(maskDimSizes.drop_back());
newMaskDimSizes.push_back(maskIndex);

if (numFrontPadElems == 0)
return rewriter.create<vector::ConstantMaskOp>(loc, newMaskType,
newMaskDimSizes);

SmallVector<bool> newMaskValues;
for (int64_t i = 0; i < numDestElems; ++i)
newMaskValues.push_back(i >= startIndex && i < maskIndex);
auto newMask = DenseElementsAttr::get(newMaskType, newMaskValues);
return rewriter.create<arith::ConstantOp>(loc, newMaskType,
newMask);
})
.Case<arith::ConstantOp>([&](auto constantOp)
-> std::optional<Operation *> {
// TODO: Support multiple dimensions.
if (maskShape.size() != 1)
return std::nullopt;
// Rearrange the original mask values to cover the whole potential
// loading region. For example, in the case of using byte-size for
// emulation, given the following mask:
//
// %mask = [0, 1, 0, 1, 0, 0]
//
// With front offset of 1, the mask will be padded 0s in the front
// and back so that:
// 1. It is aligned with the effective loading bits
// 2. Its length is multiple of `numSrcElemPerDest` (and the total
// coverage size is mulitiple of bytes). The new mask will be like
// this before compressing:
//
// %new_mask = [0, 0, 1, 0, 1, 0, 0, 0]
auto originalMask =
cast<DenseIntElementsAttr>(constantOp.getValue());
SmallVector<bool> paddedMaskValues(numFrontPadElems, false);
paddedMaskValues.append(originalMask.template value_begin<bool>(),
originalMask.template value_end<bool>());
paddedMaskValues.resize(numDestElems * numSrcElemsPerDest, false);

// Compressing by combining every `numSrcElemsPerDest` elements:
SmallVector<bool> compressedMaskValues;
for (size_t i = 0; i < paddedMaskValues.size();
i += numSrcElemsPerDest) {
bool combinedValue = false;
for (int j = 0; j < numSrcElemsPerDest; ++j) {
combinedValue |= paddedMaskValues[i + j];
}
compressedMaskValues.push_back(combinedValue);
}
return rewriter.create<arith::ConstantOp>(
loc, DenseElementsAttr::get(newMaskType, compressedMaskValues));
});

if (!newMask)
return failure();

while (!extractOps.empty()) {
newMask = rewriter.create<vector::ExtractOp>(
loc, newMask->getResults()[0], extractOps.back().getMixedPosition());
loc, (*newMask)->getResults()[0], extractOps.back().getMixedPosition());
extractOps.pop_back();
}

return newMask;
return *newMask;
}

/// Extracts 1-D subvector from a 1-D vector. It is a wrapper function for
Expand Down Expand Up @@ -185,12 +236,10 @@ static Value staticallyExtractSubvector(OpBuilder &rewriter, Location loc,
/// `vector.insert_strided_slice`.
static Value staticallyInsertSubvector(OpBuilder &rewriter, Location loc,
Value src, Value dest, int64_t offset) {
auto srcType = cast<VectorType>(src.getType());
auto destType = cast<VectorType>(dest.getType());
[[maybe_unused]] auto srcType = cast<VectorType>(src.getType());
[[maybe_unused]] auto destType = cast<VectorType>(dest.getType());
assert(srcType.getRank() == 1 && destType.getRank() == 1 &&
"expected source and dest to be vector type");
(void)srcType;
(void)destType;
auto offsets = rewriter.getI64ArrayAttr({offset});
auto strides = rewriter.getI64ArrayAttr({1});
return rewriter.create<vector::InsertStridedSliceOp>(loc, dest.getType(), src,
Expand Down
38 changes: 38 additions & 0 deletions mlir/test/Dialect/Vector/vector-emulate-narrow-type-unaligned.mlir
View file
Original file line number Diff line number Diff line change
Expand Up @@ -249,3 +249,41 @@ func.func @vector_maskedload_i2_dynamic_indexing_mixed(%passthru: vector<3xi2>,
// CHECK: %[[IN8:.+]] = vector.insert %[[EX8]], %[[IN7]] [1] : i2 into vector<3xi2>
// CHECK: %[[EX9:.+]] = vector.extract %[[SELECT]][%[[INCIDX2]]] : i2 from vector<8xi2>
// CHECK: %[[IN9:.+]] = vector.insert %[[EX9]], %[[IN8]] [2] : i2 into vector<3xi2>

// -----

func.func @vector_maskedload_i4_constant_mask_unaligned(%passthru: vector<5xi2>) -> vector<5xi2> {
%0 = memref.alloc() : memref<3x5xi2>
%mask = arith.constant dense<[false, true, true, true, false]> : vector<5xi1>
%c0 = arith.constant 0 : index
%c1 = arith.constant 1 : index
%1 = vector.maskedload %0[%c1, %c0], %mask, %passthru :
memref<3x5xi2>, vector<5xi1>, vector<5xi2> into vector<5xi2>
return %1 : vector<5xi2>
}

// CHECK: func @vector_maskedload_i4_constant_mask_unaligned(
// CHECK-SAME: %[[PTH:.+]]: vector<5xi2>) -> vector<5xi2>
// CHECK: %[[ALLOC:.+]] = memref.alloc() : memref<4xi8>
// CHECK: %[[MASK:.+]] = arith.constant dense<[false, true, true, true, false]> : vector<5xi1>

// CHECK: %[[COMPRESSED_MASK:.+]] = arith.constant dense<true> : vector<2xi1>
// CHECK: %[[EMPTY:.+]] = arith.constant dense<0> : vector<8xi2>
// CHECK: %[[PTH_PADDED:.+]] = vector.insert_strided_slice %[[PTH]], %[[EMPTY]]
// CHECK-SAME: {offsets = [1], strides = [1]} : vector<5xi2> into vector<8xi2>

// Emulated masked load from alloc:
// CHECK: %[[PTH_PADDED_UPCAST:.+]] = vector.bitcast %[[PTH_PADDED]] : vector<8xi2> to vector<2xi8>
// CHECK: %[[C1:.+]] = arith.constant 1 : index
// CHECK: %[[MASKLOAD:.+]] = vector.maskedload %[[ALLOC]][%[[C1]]], %[[COMPRESSED_MASK]], %[[PTH_PADDED_UPCAST]]
// CHECK: %[[MASKLOAD_DOWNCAST:.+]] = vector.bitcast %[[MASKLOAD]] : vector<2xi8> to vector<8xi2>

// Select from emulated loaded vector and passthru vector:
// TODO: fold this part if possible.
// CHECK: %[[EMPTY_MASK:.+]] = arith.constant dense<false> : vector<8xi1>
// CHECK: %[[MASK_PADDED:.+]] = vector.insert_strided_slice %[[MASK]], %[[EMPTY_MASK]]
// CHECK-SAME: {offsets = [1], strides = [1]} : vector<5xi1> into vector<8xi1>
// CHECK: %[[SELECT:.+]] = arith.select %[[MASK_PADDED]], %[[MASKLOAD_DOWNCAST]], %[[PTH_PADDED]] : vector<8xi1>, vector<8xi2>
// CHECK: %[[RESULT:.+]] = vector.extract_strided_slice %[[SELECT]]
// CHECK-SAME: {offsets = [1], sizes = [5], strides = [1]} : vector<8xi2> to vector<5xi2>
// CHECK: return %[[RESULT]] : vector<5xi2>
51 changes: 51 additions & 0 deletions mlir/test/Dialect/Vector/vector-emulate-narrow-type.mlir
View file
Original file line number Diff line number Diff line change
Expand Up @@ -275,6 +275,30 @@ func.func @vector_maskedload_i4_constant_mask(%arg1: index, %arg2: index, %passt

// -----

func.func @vector_maskedload_i4_arith_constant(%passthru: vector<8xi4>) -> vector<8xi4> {
%0 = memref.alloc() : memref<3x8xi4>
%cst = arith.constant dense<0> : vector<8xi4>
%mask = arith.constant dense<[false, true, true, true, true, false, false, false]> : vector<8xi1>
%c0 = arith.constant 0 : index
%1 = vector.maskedload %0[%c0, %c0], %mask, %passthru :
memref<3x8xi4>, vector<8xi1>, vector<8xi4> into vector<8xi4>
return %1 : vector<8xi4>
}

// CHECK: func @vector_maskedload_i4_arith_constant(
// CHECK-SAME: %[[PASSTHRU:[a-zA-Z0-9]+]]
// CHECK: %[[ALLOC:.+]] = memref.alloc() : memref<12xi8>
// CHECK: %[[MASK:.+]] = arith.constant dense<[false, true, true, true, true, false, false, false]> : vector<8xi1>

// Emit a new, compressed mask for emulated maskedload:
// CHECK: %[[COMPRESSED_MASK:.+]] = arith.constant dense<[true, true, true, false]> : vector<4xi1>
// CHECK: %[[PTHU_UPCAST:.+]] = vector.bitcast %[[PASSTHRU]] : vector<8xi4> to vector<4xi8>
// CHECK: %[[C0:.+]] = arith.constant 0 : index
// CHECK: %[[LOAD:.+]] = vector.maskedload %[[ALLOC]][%[[C0]]], %[[COMPRESSED_MASK]], %[[PTHU_UPCAST]]
// CHECK: %[[LOAD_DOWNCAST:.+]] = vector.bitcast %[[LOAD]] : vector<4xi8> to vector<8xi4>
// CHECK: %[[SELECT:.+]] = arith.select %[[MASK]], %[[LOAD_DOWNCAST]], %[[PASSTHRU]] : vector<8xi1>, vector<8xi4>
// CHECK: return %[[SELECT]] : vector<8xi4>

///----------------------------------------------------------------------------------------
/// vector.extract -> vector.masked_load
///----------------------------------------------------------------------------------------
Expand Down Expand Up @@ -624,3 +648,30 @@ func.func @vector_maskedstore_i4_constant_mask(
// CHECK32: %[[SELECT:.+]] = arith.select %[[ORIG_MASK]], %[[VAL_TO_STORE]], %[[BITCAST]] : vector<8xi1>, vector<8xi4>
// CHECK32: %[[NEW_VAL:.+]] = vector.bitcast %[[SELECT]] : vector<8xi4> to vector<1xi32>
// CHECK32: vector.maskedstore %[[ALLOC]]{{\[}}%[[LIDX]]], %[[NEW_MASK]], %[[NEW_VAL]] : memref<3xi32>, vector<1xi1>, vector<1xi32>

// -----

func.func @vector_maskedstore_i4_arith_constant(%val_to_store: vector<8xi4>) {
%0 = memref.alloc() : memref<5x8xi4>
%cst = arith.constant dense<0> : vector<8xi4>
%mask = arith.constant dense<[false, true, true, true, true, true, false, false]> : vector<8xi1>
%c0 = arith.constant 0 : index
%c3 = arith.constant 3 : index
vector.maskedstore %0[%c3, %c0], %mask, %val_to_store :
memref<5x8xi4>, vector<8xi1>, vector<8xi4>
return
}

// CHECK-LABEL: func @vector_maskedstore_i4_arith_constant
// CHECK-SAME: %[[VAL_TO_STORE:[a-zA-Z0-9]+]]:
// CHECK: %[[ALLOC:.+]] = memref.alloc() : memref<20xi8>
// CHECK: %[[MASK:.+]] = arith.constant dense<[false, true, true, true, true, true, false, false]> : vector<8xi1>
// %c3 * 4 bits = 12
// CHECK: %[[IDX_FLATTENED:.+]] = arith.constant 12 : index
// CHECK: %[[COMPRESSED_MASK:.+]] = arith.constant dense<[true, true, true, false]> : vector<4xi1>
// CHECK: %[[EMPTY:.+]] = arith.constant dense<0> : vector<4xi8>
// CHECK: %[[MASKEDLOAD:.+]] = vector.maskedload %[[ALLOC]][%[[IDX_FLATTENED]]], %[[COMPRESSED_MASK]], %[[EMPTY]]
// CHECK: %[[LOAD_UPCAST:.+]] = vector.bitcast %[[MASKEDLOAD]]
// CHECK: %[[SELECT:.+]] = arith.select %[[MASK]], %[[VAL_TO_STORE]], %[[LOAD_UPCAST]]
// CHECK: %[[SELECT_DOWNCAST:.+]] = vector.bitcast %[[SELECT]]
// CHECK: vector.maskedstore %[[ALLOC]][%[[IDX_FLATTENED]]], %[[COMPRESSED_MASK]], %[[SELECT_DOWNCAST]]

0 comments on commit ef92aba

Please sign in to comment.