Skip to content

[flang][cuda] Carry over the dynamic shared memory size to gpu.launch_func #132837

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 25, 2025
Merged

[flang][cuda] Carry over the dynamic shared memory size to gpu.launch_func #132837

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
7 changes: 4 additions & 3 deletions flang/lib/Optimizer/Transforms/CUFOpConversion.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -810,7 +810,7 @@ struct CUFLaunchOpConversion
mlir::PatternRewriter &rewriter) const override {
mlir::Location loc = op.getLoc();
auto idxTy = mlir::IndexType::get(op.getContext());
auto zero = rewriter.create<mlir::arith::ConstantOp>(
mlir::Value zero = rewriter.create<mlir::arith::ConstantOp>(
loc, rewriter.getIntegerType(32), rewriter.getI32IntegerAttr(0));
auto gridSizeX =
rewriter.create<mlir::arith::IndexCastOp>(loc, idxTy, op.getGridX());
Expand Down Expand Up @@ -869,10 +869,11 @@ struct CUFLaunchOpConversion
}
args.push_back(arg);
}

mlir::Value dynamicShmemSize = op.getBytes() ? op.getBytes() : zero;
auto gpuLaunchOp = rewriter.create<mlir::gpu::LaunchFuncOp>(
loc, kernelName, mlir::gpu::KernelDim3{gridSizeX, gridSizeY, gridSizeZ},
mlir::gpu::KernelDim3{blockSizeX, blockSizeY, blockSizeZ}, zero, args);
mlir::gpu::KernelDim3{blockSizeX, blockSizeY, blockSizeZ},
dynamicShmemSize, args);
if (clusterDimX && clusterDimY && clusterDimZ) {
gpuLaunchOp.getClusterSizeXMutable().assign(clusterDimX);
gpuLaunchOp.getClusterSizeYMutable().assign(clusterDimY);
Expand Down
4 changes: 4 additions & 0 deletions flang/test/Fir/CUDA/cuda-launch.fir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -23,11 +23,15 @@ module attributes {gpu.container_module, dlti.dl_spec = #dlti.dl_spec<#dlti.dl_e
// CHECK: %[[ALLOCA:.*]] = fir.alloca f32
%c1 = arith.constant 1 : index
%c11_i32 = arith.constant 11 : i32
%c1024_i32 = arith.constant 1024 : i32
%c6_i32 = arith.constant 6 : i32
%c1_i32 = arith.constant 1 : i32
// CHECK: gpu.launch_func @cuda_device_mod::@_QPsub_device1 blocks in (%{{.*}}, %{{.*}}, %{{.*}}) threads in (%{{.*}}, %{{.*}}, %{{.*}}) dynamic_shared_memory_size %c0{{.*}}
cuf.kernel_launch @cuda_device_mod::@_QPsub_device1<<<%c1_i32, %c1_i32, %c1_i32, %c1_i32, %c1_i32, %c1_i32>>>()

// CHECK: gpu.launch_func @cuda_device_mod::@_QPsub_device1 blocks in (%{{.*}}, %{{.*}}, %{{.*}}) threads in (%{{.*}}, %{{.*}}, %{{.*}}) dynamic_shared_memory_size %c1024{{.*}}
cuf.kernel_launch @cuda_device_mod::@_QPsub_device1<<<%c1_i32, %c1_i32, %c1_i32, %c1_i32, %c1_i32, %c1_i32, %c1024_i32>>>()

// CHECK: gpu.launch_func @cuda_device_mod::@_QPsub_device2 blocks in (%{{.*}}, %{{.*}}, %{{.*}}) threads in (%{{.*}}, %{{.*}}, %{{.*}}) dynamic_shared_memory_size %c0{{.*}} args(%[[ALLOCA]] : !fir.ref<f32>)
cuf.kernel_launch @cuda_device_mod::@_QPsub_device2<<<%c1_i32, %c1_i32, %c1_i32, %c1_i32, %c1_i32, %c1_i32>>>(%0) : (!fir.ref<f32>)
return
Expand Down