[all-commits] [llvm/llvm-project] d79bb9: [flang][cuda] Carry over the stream information to...

[Valentin Clement (バレンタイン クレメン) via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: d79bb932785d312faa8cb8e82a0afc103436c54c
      https://github.com/llvm/llvm-project/commit/d79bb932785d312faa8cb8e82a0afc103436c54c
  Author: Valentin Clement (バレンタイン クレメン) <clementval at gmail.com>
  Date:   2025-04-18 (Fri, 18 Apr 2025)

  Changed paths:
    M flang-rt/lib/cuda/kernel.cpp
    M flang/include/flang/Optimizer/Dialect/CUF/CUFOps.td
    M flang/include/flang/Optimizer/Transforms/CUFGPUToLLVMConversion.h
    M flang/include/flang/Runtime/CUDA/kernel.h
    M flang/lib/Optimizer/Dialect/CUF/CUFOps.cpp
    M flang/lib/Optimizer/Transforms/CUFGPUToLLVMConversion.cpp
    M flang/test/Fir/CUDA/cuda-gpu-launch-func.mlir
    M flang/test/Fir/CUDA/cuda-launch.fir
    M flang/test/Fir/CUDA/cuda-stream.mlir

  Log Message:
  -----------
  [flang][cuda] Carry over the stream information to kernel launch (#136217)

In CUDA Fortran the stream is encoded in an INTEGER(cuda_stream_kind)
variable.

This information is carried over the GPU dialect through the
`cuf.stream_cast` and the token in the GPU ops.

When converting the `gpu.launch_func` to runtime call, the
`cuf.stream_cast` becomes a no-op and the reference to the stream is
passed to the runtime.

The runtime is adapted to take integer references instead of value for
stream.



To unsubscribe from these emails, change your notification settings at https://github.com/llvm/llvm-project/settings/notifications