Skip to content

Fix bug in gpu.memcpy lowering for dynamically shaped operands. #128820

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Mar 3, 2025
Merged

Fix bug in gpu.memcpy lowering for dynamically shaped operands. #128820

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
18 changes: 10 additions & 8 deletions mlir/lib/Conversion/GPUCommon/GPUToLLVMConversion.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -76,14 +76,16 @@ class ConvertOpToGpuRuntimeCallPattern : public ConvertOpToLLVMPattern<OpTy> {
Value getNumElements(ConversionPatternRewriter &rewriter, Location loc,
MemRefType type, MemRefDescriptor desc) const {
Type indexType = ConvertToLLVMPattern::getIndexType();
return type.hasStaticShape()
? ConvertToLLVMPattern::createIndexAttrConstant(
rewriter, loc, indexType, type.getNumElements())
// For identity maps (verified by caller), the number of
// elements is stride[0] * size[0].
: rewriter.create<LLVM::MulOp>(loc,
desc.stride(rewriter, loc, 0),
desc.size(rewriter, loc, 0));
if (type.hasStaticShape())
return ConvertToLLVMPattern::createIndexAttrConstant(
rewriter, loc, indexType, type.getNumElements());
// Compute the number of elements by multiplying all the dim sizes.
uint64_t rank = type.getRank();
Value numElements = desc.size(rewriter, loc, /*pos=*/0);
for (unsigned i = 1; i < rank; i++)
numElements = rewriter.create<LLVM::MulOp>(
loc, numElements, desc.size(rewriter, loc, /*pos=*/i));
return numElements;
}

MLIRContext *context = &this->getTypeConverter()->getContext();
Expand Down
20 changes: 20 additions & 0 deletions mlir/test/Conversion/GPUCommon/lower-memcpy-to-gpu-runtime-calls.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,3 +17,23 @@ module attributes {gpu.container_module} {
return
}
}

// -----

module attributes {gpu.container_module} {

// CHECK: func @dynamic
func.func @dynamic(%dst : memref<?x?xf32, 1>, %src : memref<?x?xf32>) {
// CHECK: %[[T0:.*]] = llvm.call @mgpuStreamCreate
%t0 = gpu.wait async
%t1 = gpu.memcpy async [%t0] %dst, %src : memref<?x?xf32, 1>, memref<?x?xf32>
// CHECK: %[[DIM_SIZE_0:.*]] = llvm.extractvalue %{{.*}}[3, 0] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)>
// CHECK-NEXT: %[[DIM_SIZE_1:.*]] = llvm.extractvalue %{{.*}}[3, 1] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)>
// CHECK: %[[NUM_ELEMENTS:.*]] = llvm.mul %[[DIM_SIZE_0]], %[[DIM_SIZE_1]] : i64
// CHECK: %[[SIZE_PTR:.*]] = llvm.getelementptr %{{.*}}[%[[NUM_ELEMENTS]]] : (!llvm.ptr, i64) -> !llvm.ptr, f32
// CHECK-NEXT: %[[SIZE_INT:.*]] = llvm.ptrtoint %[[SIZE_PTR]] : !llvm.ptr to i64
// CHECK: %[[ADDR_CAST:.*]] = llvm.addrspacecast
// CHECK: llvm.call @mgpuMemcpy(%[[ADDR_CAST]], %{{.*}}, %[[SIZE_INT]], %[[T0]])
return
}
}