[all-commits] [llvm/llvm-project] 5db136: [PGO][AMDGPU] Add basic HIP offload PGO support (#...

[Yaxun (Sam) Liu via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 5db13643f4b7038db0ca304d9f8900122502935c
      https://github.com/llvm/llvm-project/commit/5db13643f4b7038db0ca304d9f8900122502935c
  Author: Yaxun (Sam) Liu <yaxun.liu at amd.com>
  Date:   2026-05-27 (Wed, 27 May 2026)

  Changed paths:
    M clang/lib/CodeGen/CGCUDANV.cpp
    A clang/test/CodeGenHIP/offload-pgo-sections.hip
    M compiler-rt/lib/profile/CMakeLists.txt
    M compiler-rt/lib/profile/InstrProfilingFile.c
    A compiler-rt/lib/profile/InstrProfilingPlatformROCm.cpp
    M llvm/include/llvm/IR/RuntimeLibcalls.td
    M llvm/include/llvm/ProfileData/InstrProf.h
    M llvm/lib/Transforms/Instrumentation/InstrProfiling.cpp
    M llvm/lib/Transforms/Instrumentation/PGOInstrumentation.cpp
    A llvm/test/Instrumentation/InstrProfiling/amdgpu-instrumentation.ll
    A llvm/test/Instrumentation/InstrProfiling/amdgpu-profc-arrays.ll
    A llvm/test/Instrumentation/InstrProfiling/gpu-weak.ll
    A llvm/test/Transforms/PGOProfile/amdgpu-disable-value-profiling.ll

  Log Message:
  -----------
  [PGO][AMDGPU] Add basic HIP offload PGO support (#177665)

Provide the minimum HIP/offload path for device profile collection and
merge on HIP before layering profile-format and uniformity-specific
changes separately.

This adds the ROCm collection runtime, hooks device collection into the
host write-file path, lowers AMDGPU instrumentation to
__llvm_profile_instrument_gpu with regular counters, and disables GPU
indirect-call value profiling.



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