Revert "[MLIR][XeGPU] XeVM lowering support for load_matrix/store_matrix" #163684
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
@llvm/pr-subscribers-mlir @llvm/pr-subscribers-mlir-gpu Author: Vitaly Buka (vitalybuka) ChangesReverts llvm/llvm-project#162780 Breaks build bots, see #162780. Patch is 57.84 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/163684.diff 13 Files Affected:
diff --git a/mlir/include/mlir/Dialect/XeGPU/IR/XeGPUAttrs.td b/mlir/include/mlir/Dialect/XeGPU/IR/XeGPUAttrs.td
index 19a52317956d2..5695d5d515d7f 100644
--- a/mlir/include/mlir/Dialect/XeGPU/IR/XeGPUAttrs.td
+++ b/mlir/include/mlir/Dialect/XeGPU/IR/XeGPUAttrs.td
@@ -712,14 +712,10 @@ def XeGPU_MemLayoutAttr : XeGPUAttr<"MemLayout", "mem_layout"> {
return getAttrs().contains(name);
}
- ArrayAttr getStrideAttr() {
+ ArrayAttr getStrides() {
return getAttrs().getAs<ArrayAttr>("stride");
}
- ArrayAttr getBlockAttr() {
- return getAttrs().getAs<ArrayAttr>("block");
- }
-
}];
}
diff --git a/mlir/include/mlir/Dialect/XeGPU/IR/XeGPUOps.td b/mlir/include/mlir/Dialect/XeGPU/IR/XeGPUOps.td
index 426377fcf598f..73f9061f5debe 100644
--- a/mlir/include/mlir/Dialect/XeGPU/IR/XeGPUOps.td
+++ b/mlir/include/mlir/Dialect/XeGPU/IR/XeGPUOps.td
@@ -1298,14 +1298,14 @@ def XeGPU_CreateMemDescOp: XeGPU_Op<"create_mem_desc", [Pure,
}
def XeGPU_LoadMatrixOp: XeGPU_Op<"load_matrix", [MemoryEffects<[MemRead]>,
- AllElementTypesMatch<["mem_desc", "res"]>]> {
+ AllElementTypesMatch<["mem_desc", "res"]>,
+ AllRanksMatch<["mem_desc", "res"]>]> {
let arguments = (ins XeGPU_MemDesc:$mem_desc,
Variadic<Index>: $offsets,
DenseI64ArrayAttr: $const_offsets,
- OptionalAttr<UnitAttr>:$subgroup_block_io,
OptionalAttr<DistributeLayoutAttr>:$layout
);
- let results = (outs AnyTypeOf<[XeGPU_ValueType, XeGPU_ScalarType]>:$res);
+ let results = (outs XeGPU_ValueType:$res);
let assemblyFormat = [{
$mem_desc `` custom<DynamicIndexList>($offsets, $const_offsets)
prop-dict attr-dict `` `:` type(operands) `->` type(results)
@@ -1319,9 +1319,6 @@ def XeGPU_LoadMatrixOp: XeGPU_Op<"load_matrix", [MemoryEffects<[MemRead]>,
Arguments:
- `mem_desc`: the memory descriptor identifying the SLM region.
- `offsets`: the coordinates within the matrix to read from.
- - `subgroup_block_io`: [optional] An attribute indicating that the operation can be
- lowered to a subgroup block load. When this attribute is present,
- the offsets are subgroup-uniform across all lanes.
- `layout`: [optional] An attribute for guiding distributions among
subgroups and/or work-items. It currently can accept either
LayoutAttr or SliceAttr.
@@ -1339,10 +1336,7 @@ def XeGPU_LoadMatrixOp: XeGPU_Op<"load_matrix", [MemoryEffects<[MemRead]>,
}
ArrayRef<int64_t> getDataShape() {
- auto resTy = getRes().getType();
- if (auto vecTy = llvm::dyn_cast<VectorType>(resTy))
- return vecTy.getShape();
- return {};
+ return getRes().getType().getShape();
}
}];
@@ -1350,13 +1344,13 @@ def XeGPU_LoadMatrixOp: XeGPU_Op<"load_matrix", [MemoryEffects<[MemRead]>,
}
def XeGPU_StoreMatrixOp: XeGPU_Op<"store_matrix", [MemoryEffects<[MemWrite]>,
- AllElementTypesMatch<["mem_desc", "data"]>]> {
+ AllElementTypesMatch<["mem_desc", "data"]>,
+ AllRanksMatch<["mem_desc", "data"]>]> {
let arguments = (ins
- AnyTypeOf<[XeGPU_ValueType, XeGPU_ScalarType]>:$data,
+ XeGPU_ValueType:$data,
XeGPU_MemDesc:$mem_desc,
Variadic<Index>: $offsets,
DenseI64ArrayAttr: $const_offsets,
- OptionalAttr<UnitAttr>:$subgroup_block_io,
OptionalAttr<DistributeLayoutAttr>:$layout
);
let assemblyFormat = [{ $data `,` $mem_desc `` custom<DynamicIndexList>($offsets, $const_offsets)
@@ -1370,9 +1364,6 @@ def XeGPU_StoreMatrixOp: XeGPU_Op<"store_matrix", [MemoryEffects<[MemWrite]>,
- `mem_desc`: the memory descriptor specifying the SLM region.
- `offsets`: the coordinates within the matrix where the data will be written.
- `data`: the values to be stored in the matrix.
- - `subgroup_block_io`: [optional] An attribute indicating that the operation can be
- lowered to a subgroup block store. When this attribute is present,
- the offsets are subgroup-uniform across all lanes.
- `layout`: [optional] An attribute for guiding distributions among
subgroups and/or work-items. It currently can accept either
LayoutAttr or SliceAttr.
@@ -1387,10 +1378,7 @@ def XeGPU_StoreMatrixOp: XeGPU_Op<"store_matrix", [MemoryEffects<[MemWrite]>,
}
ArrayRef<int64_t> getDataShape() {
- auto DataTy = getData().getType();
- if (auto vecTy = llvm::dyn_cast<VectorType>(DataTy))
- return vecTy.getShape();
- return {};
+ return getData().getType().getShape();
}
}];
@@ -1398,4 +1386,41 @@ def XeGPU_StoreMatrixOp: XeGPU_Op<"store_matrix", [MemoryEffects<[MemWrite]>,
let hasVerifier = 1;
}
+def XeGPU_MemDescSubviewOp: XeGPU_Op<"mem_desc_subview",
+ [Pure, ViewLikeOpInterface, AllElementTypesMatch<["src", "res"]>]> {
+ let description = [{
+ Creates a subview of a memory descriptor. The resulting memory descriptor can have
+ a lower rank than the source; in this case, the result dimensions correspond to the
+ higher-order dimensions of the source memory descriptor.
+
+ Arguments:
+ - `src` : a memory descriptor.
+ - `offsets` : the coordinates within the matrix the subview will be created from.
+
+ Results:
+ - `res` : a memory descriptor with smaller size.
+
+ }];
+ let arguments = (ins XeGPU_MemDesc:$src,
+ Variadic<Index>:$offsets,
+ DenseI64ArrayAttr:$const_offsets);
+ let results = (outs XeGPU_MemDesc:$res);
+ let assemblyFormat = [{$src `` custom<DynamicIndexList>($offsets, $const_offsets) prop-dict
+ attr-dict `` `:` qualified(type($src)) `->` qualified(type($res))}];
+ let builders = [
+ OpBuilder<(ins "Type": $res, "Value":$src, "llvm::ArrayRef<OpFoldResult>": $offsets)>
+ ];
+
+ let extraClassDeclaration = [{
+ mlir::Value getViewSource() { return getSrc(); }
+
+ SmallVector<OpFoldResult> getMixedOffsets() {
+ return getMixedValues(getConstOffsets(), getOffsets(), getContext());
+ }
+ }];
+
+ let hasVerifier = 1;
+}
+
+
#endif // MLIR_DIALECT_XEGPU_IR_XEGPUOPS_TD
diff --git a/mlir/include/mlir/Dialect/XeGPU/IR/XeGPUTypes.td b/mlir/include/mlir/Dialect/XeGPU/IR/XeGPUTypes.td
index b1196fbe9c66a..84902b2039643 100644
--- a/mlir/include/mlir/Dialect/XeGPU/IR/XeGPUTypes.td
+++ b/mlir/include/mlir/Dialect/XeGPU/IR/XeGPUTypes.td
@@ -237,11 +237,12 @@ def XeGPU_MemDesc: XeGPUTypeDef<"MemDesc", "mem_desc", [ShapedTypeInterface], "m
return MemDescType::get(getContext(), shape.value_or(getShape()), elementType, getMemLayout());
}
- ArrayAttr getStrideAttr() {
+ ArrayAttr getStrides() {
auto layout = getMemLayout();
if (layout && layout.hasAttr("stride")) {
- return layout.getStrideAttr();
+ return layout.getStrides();
}
+
// derive and return default strides
SmallVector<int64_t> defaultStrides;
llvm::append_range(defaultStrides, getShape().drop_front());
@@ -249,63 +250,6 @@ def XeGPU_MemDesc: XeGPUTypeDef<"MemDesc", "mem_desc", [ShapedTypeInterface], "m
Builder builder(getContext());
return builder.getI64ArrayAttr(defaultStrides);
}
-
- ArrayAttr getBlockAttr() {
- auto layout = getMemLayout();
- if (layout && layout.hasAttr("block")) {
- return layout.getBlockAttr();
- }
- Builder builder(getContext());
- return builder.getI64ArrayAttr({});
- }
-
- /// Heuristic to determine if the MemDesc uses column-major layout,
- /// based on the rank and the value of the first stride dimension.
- bool isColMajor() {
- auto dim0 = dyn_cast<IntegerAttr>(getStrideAttr()[0]);
- return getRank() == 2 && dim0.getInt() == 1;
- }
-
- // Get the Blocking shape for a MemDescType, Which is represented
- // as an attribute in MemDescType. By default it is the shape
- // of the mdescTy
- SmallVector<int64_t> getBlockShape() {
- SmallVector<int64_t> size(getShape());
- ArrayAttr blockAttr = getBlockAttr();
- if (!blockAttr.empty()) {
- size.clear();
- for (auto attr : blockAttr.getValue()) {
- size.push_back(cast<IntegerAttr>(attr).getInt());
- }
- }
- return size;
- }
-
- // Get strides as vector of integer.
- // If it contains block attribute, the strides are blocked strides.
- //
- // The blocking is applied to the base matrix shape derived from the
- // memory descriptor's stride information. If the matrix described by
- // the memory descriptor is not contiguous, it is assumed that the base
- // matrix is contiguous and follows the same memory layout.
- //
- // It first computes the original matrix shape using the stride info,
- // then computes the number of blocks in each dimension of original shape,
- // then compute the outer block shape and stride,
- // then combines the inner and outer block shape and stride
- // e.g. for `mem_desc<32x256xf16, @block=[16, 8], @strides=[1, 32]>`
- // its memory layout tuple is ([2,32,16,8],[128,256,1,16])
- // for `mem_desc<256x32xf16, @block=[8, 16]>` with default @stride[32, 1]
- // its memory layout tuple is ([32,2,8,16],[256,128,16,1])
- SmallVector<int64_t> getStrideShape();
-
- /// Generates instructions to compute the linearize offset
- // if the memory descriptor is blocked, it returns linearize offset based on the blocked layout
- // the strides of memory descriptor is always considered regardless of blocked or not
- Value getLinearOffsets(OpBuilder &builder,
- Location loc, ArrayRef<OpFoldResult> offsets);
-
-
}];
let hasCustomAssemblyFormat = true;
diff --git a/mlir/lib/Conversion/XeGPUToXeVM/CMakeLists.txt b/mlir/lib/Conversion/XeGPUToXeVM/CMakeLists.txt
index dd9edc43a1657..84b25809f1ed0 100644
--- a/mlir/lib/Conversion/XeGPUToXeVM/CMakeLists.txt
+++ b/mlir/lib/Conversion/XeGPUToXeVM/CMakeLists.txt
@@ -21,7 +21,6 @@ add_mlir_conversion_library(MLIRXeGPUToXeVM
MLIRIndexDialect
MLIRSCFDialect
MLIRXeGPUDialect
- MLIRXeGPUUtils
MLIRPass
MLIRTransforms
MLIRSCFTransforms
diff --git a/mlir/lib/Conversion/XeGPUToXeVM/XeGPUToXeVM.cpp b/mlir/lib/Conversion/XeGPUToXeVM/XeGPUToXeVM.cpp
index fcbf66dbe9e45..ddcbc44f2652a 100644
--- a/mlir/lib/Conversion/XeGPUToXeVM/XeGPUToXeVM.cpp
+++ b/mlir/lib/Conversion/XeGPUToXeVM/XeGPUToXeVM.cpp
@@ -22,7 +22,6 @@
#include "mlir/Dialect/SCF/Transforms/Patterns.h"
#include "mlir/Dialect/Vector/IR/VectorOps.h"
#include "mlir/Dialect/XeGPU/IR/XeGPU.h"
-#include "mlir/Dialect/XeGPU/Utils/XeGPUUtils.h"
#include "mlir/Pass/Pass.h"
#include "mlir/Support/LLVM.h"
#include "llvm/ADT/STLExtras.h"
@@ -64,7 +63,6 @@ static int32_t getNumericXeVMAddrSpace(xegpu::MemorySpace xeGpuMemspace) {
case xegpu::MemorySpace::SLM:
return static_cast<int>(xevm::AddrSpace::SHARED);
}
- llvm_unreachable("Unknown XeGPU memory space");
}
// Get same bitwidth flat vector type of new element type.
@@ -188,7 +186,6 @@ class CreateNdDescToXeVMPattern
int64_t rank = mixedSizes.size();
if (rank != 2)
return rewriter.notifyMatchFailure(op, "Expected 2D shape.");
-
auto sourceTy = source.getType();
auto sourceMemrefTy = dyn_cast<MemRefType>(sourceTy);
// If source is a memref, we need to extract the aligned pointer as index.
@@ -367,11 +364,10 @@ class LoadStorePrefetchNdToXeVMPattern : public OpConversionPattern<OpType> {
// Add a builder that creates
// offset * elemByteSize + baseAddr
-static Value addOffsetToBaseAddr(ConversionPatternRewriter &rewriter,
- Location loc, Value baseAddr, Value offset,
- int64_t elemByteSize) {
+static Value addOffset(ConversionPatternRewriter &rewriter, Location loc,
+ Value baseAddr, Value offset, int64_t elemByteSize) {
Value byteSize = arith::ConstantIntOp::create(
- rewriter, loc, baseAddr.getType(), elemByteSize);
+ rewriter, loc, rewriter.getI64Type(), elemByteSize);
Value byteOffset = arith::MulIOp::create(rewriter, loc, offset, byteSize);
Value newAddr = arith::AddIOp::create(rewriter, loc, baseAddr, byteOffset);
return newAddr;
@@ -447,8 +443,7 @@ class LoadStoreToXeVMPattern : public OpConversionPattern<OpType> {
// If offset is provided, we add them to the base pointer.
// Offset is in number of elements, we need to multiply by
// element byte size.
- basePtrI64 =
- addOffsetToBaseAddr(rewriter, loc, basePtrI64, offset, elemByteSize);
+ basePtrI64 = addOffset(rewriter, loc, basePtrI64, offset, elemByteSize);
}
// Convert base pointer (i64) to LLVM pointer type.
Value basePtrLLVM =
@@ -511,147 +506,6 @@ class LoadStoreToXeVMPattern : public OpConversionPattern<OpType> {
}
};
-// Lower xegpu::CreateMemDescOp to memref::ViewOp. Since SLM access instructions
-// on Xe2 and Xe3 operate on 32-bit or 64-bit units, all data types smaller than
-// 32 bits will be converted to 32 bits.
-class CreateMemDescOpPattern final
- : public OpConversionPattern<xegpu::CreateMemDescOp> {
-public:
- using OpConversionPattern<xegpu::CreateMemDescOp>::OpConversionPattern;
- LogicalResult
- matchAndRewrite(xegpu::CreateMemDescOp op, OpAdaptor adaptor,
- ConversionPatternRewriter &rewriter) const override {
-
- auto resTy = op.getMemDesc();
-
- // Create the result MemRefType with the same shape, element type, and
- // memory space
- auto newResTy = getTypeConverter()->convertType<MemRefType>(resTy);
-
- Value zero = arith::ConstantIndexOp::create(rewriter, op.getLoc(), 0);
- auto viewOp = memref::ViewOp::create(rewriter, op.getLoc(), newResTy,
- op.getSource(), zero, ValueRange());
- rewriter.replaceOp(op, viewOp);
- return success();
- }
-};
-
-template <typename OpType,
- typename = std::enable_if_t<llvm::is_one_of<
- OpType, xegpu::LoadMatrixOp, xegpu::StoreMatrixOp>::value>>
-class LoadStoreMatrixToXeVMPattern : public OpConversionPattern<OpType> {
- using OpConversionPattern<OpType>::OpConversionPattern;
- LogicalResult
- matchAndRewrite(OpType op, typename OpType::Adaptor adaptor,
- ConversionPatternRewriter &rewriter) const override {
-
- SmallVector<OpFoldResult> offsets = op.getMixedOffsets();
- if (offsets.empty())
- return rewriter.notifyMatchFailure(op, "Expected offset to be provided.");
-
- auto loc = op.getLoc();
- auto ctxt = rewriter.getContext();
- Value basePtrStruct = adaptor.getMemDesc();
- Value mdescVal = op.getMemDesc();
- // Load result or Store value Type can be vector or scalar.
- Value data;
- if constexpr (std::is_same_v<OpType, xegpu::LoadMatrixOp>)
- data = op.getResult();
- else
- data = adaptor.getData();
- VectorType valOrResVecTy = dyn_cast<VectorType>(data.getType());
- if (!valOrResVecTy)
- valOrResVecTy = VectorType::get(1, data.getType());
-
- int64_t elemBitWidth =
- valOrResVecTy.getElementType().getIntOrFloatBitWidth();
- // Element type must be multiple of 8 bits.
- if (elemBitWidth % 8 != 0)
- return rewriter.notifyMatchFailure(
- op, "Expected element type bit width to be multiple of 8.");
- int64_t elemByteSize = elemBitWidth / 8;
-
- // Default memory space is SLM.
- LLVM::LLVMPointerType ptrTypeLLVM = LLVM::LLVMPointerType::get(
- ctxt, getNumericXeVMAddrSpace(xegpu::MemorySpace::SLM));
-
- auto mdescTy = cast<xegpu::MemDescType>(mdescVal.getType());
-
- Value basePtrLLVM = memref::ExtractAlignedPointerAsIndexOp::create(
- rewriter, loc, basePtrStruct);
-
- // Convert base pointer (ptr) to i32
- Value basePtrI32 = arith::IndexCastUIOp::create(
- rewriter, loc, rewriter.getI32Type(), basePtrLLVM);
-
- Value linearOffset = mdescTy.getLinearOffsets(rewriter, loc, offsets);
- linearOffset = arith::IndexCastUIOp::create(
- rewriter, loc, rewriter.getI32Type(), linearOffset);
- basePtrI32 = addOffsetToBaseAddr(rewriter, loc, basePtrI32, linearOffset,
- elemByteSize);
-
- // convert base pointer (i32) to LLVM pointer type
- basePtrLLVM =
- LLVM::IntToPtrOp::create(rewriter, loc, ptrTypeLLVM, basePtrI32);
-
- if (op.getSubgroupBlockIoAttr()) {
- // if the attribute 'subgroup_block_io' is set to true, it lowers to
- // xevm.blockload
-
- Type intElemTy = rewriter.getIntegerType(elemBitWidth);
- VectorType intVecTy =
- VectorType::get(valOrResVecTy.getShape(), intElemTy);
-
- if constexpr (std::is_same_v<OpType, xegpu::LoadMatrixOp>) {
- Value loadOp =
- xevm::BlockLoadOp::create(rewriter, loc, intVecTy, basePtrLLVM);
- if (intVecTy != valOrResVecTy) {
- loadOp =
- vector::BitCastOp::create(rewriter, loc, valOrResVecTy, loadOp);
- }
- rewriter.replaceOp(op, loadOp);
- } else {
- Value dataToStore = adaptor.getData();
- if (valOrResVecTy != intVecTy) {
- dataToStore =
- vector::BitCastOp::create(rewriter, loc, intVecTy, dataToStore);
- }
- xevm::BlockStoreOp::create(rewriter, loc, basePtrLLVM, dataToStore,
- nullptr);
- rewriter.eraseOp(op);
- }
- return success();
- }
-
- if (valOrResVecTy.getNumElements() >= 1) {
- auto chipOpt = xegpu::getChipStr(op);
- if (!chipOpt || (*chipOpt != "pvc" && *chipOpt != "bmg")) {
- // the lowering for chunk load only works for pvc and bmg
- return rewriter.notifyMatchFailure(
- op, "The lowering is specific to pvc or bmg.");
- }
- }
-
- if constexpr (std::is_same_v<OpType, xegpu::LoadMatrixOp>) {
- // if the size of valOrResVecTy is 1, it lowers to a scalar load/store
- // operation. LLVM load/store does not support vector of size 1, so we
- // need to handle this case separately.
- auto scalarTy = valOrResVecTy.getElementType();
- LLVM::LoadOp loadOp;
- if (valOrResVecTy.getNumElements() == 1)
- loadOp = LLVM::LoadOp::create(rewriter, loc, scalarTy, basePtrLLVM);
- else
- loadOp =
- LLVM::LoadOp::create(rewriter, loc, valOrResVecTy, basePtrLLVM);
- rewriter.replaceOp(op, loadOp);
- } else {
- LLVM::StoreOp::create(rewriter, loc, adaptor.getData(), basePtrLLVM);
- rewriter.eraseOp(op);
- }
- return success();
- }
-};
-
class PrefetchToXeVMPattern : public OpConversionPattern<xegpu::PrefetchOp> {
using OpConversionPattern::OpConversionPattern;
LogicalResult
@@ -694,8 +548,8 @@ class PrefetchToXeVMPattern : public OpConversionPattern<xegpu::PrefetchOp> {
op, "Expected element type bit width to be multiple of 8.");
elemByteSize = elemBitWidth / 8;
}
- basePtrI64 = addOffsetToBaseAddr(rewriter, loc, basePtrI64, offsets,
- elemByteSize);
+ basePtrI64 =
+ addOffset(rewriter, loc, basePtrI64, offsets, elemByteSize);
}
}
// Default memory space is global.
@@ -932,13 +786,6 @@ struct ConvertXeGPUToXeVMPass
auto i32Type = IntegerType::get(&getContext(), 32);
return VectorType::get(8, i32Type);
});
- // Convert MemDescType into flattened MemRefType for SLM
- typeConverter.addConversion([&](xegpu::MemDescType type) -> Type {
- Type elemTy = type.getElementType();
- int numElems = type.getNumElements();
- return MemRefType::get(numElems, elemTy, AffineMap(), 3);
- });
-
typeConverter.addConversion([&](MemRefType type) -> Type {
// Convert MemRefType to i64 type.
return IntegerType::get(&getContext(), 64);
@@ -1093,9 +940,6 @@ void mlir::populateXeGPUToXeVMConversionPatterns(
LoadStoreToXeVMPattern<xegpu::LoadGatherOp>,
LoadStoreToXeVMPattern<xegpu::StoreScatterOp>>(
typeConverter, patterns.getContext());
- patterns.add<LoadStoreMatrixToXeVMPattern<xegpu::LoadMatrixOp>,
- LoadStoreMat...
[truncated]
|
vitalybuka
added
the
skip-precommit-approval
vitalybuka
deleted the
revert-162780-users/Jianhui-Li/XeGPU/load-matrix-WI-attributes
branch
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Reverts #162780
Breaks build bots, see #162780.