Commits

Breadcrumbs

History for

ondev

User selector

All users

All time

Commit History

Commits on Jan 17, 2025

• GPU: Rename OpenCL2 to OpenCL, now that we have only 1 OpenCL implementation

  davidrohr

  committed
  843479b

  Copy full SHA for 843479b

• GPU:Remove OpenCL 1.2, now that OpenCL 2 supports all its functionality

  davidrohr

  committed
  90ca34c

  Copy full SHA for 90ca34c

Commits on Sep 11, 2024

• GPU: GPUCA_BUILD_DEBUG makes sure all possible source files are compiled for all kernels

  davidrohr

  committed
  cc64785

  Copy full SHA for cc64785

• GPU CMake: Add check for minimum ROCm version

  davidrohr

  committed
  158e6f8

  Copy full SHA for 158e6f8

Commits on Sep 7, 2024

• GPU: More WORKAROUNDs for CUDA bug exposing device code via host symbols

  davidrohr

  committed
  b1438a9

  Copy full SHA for b1438a9

• GPU CMake: Raise required CUDA version to 12.6

  davidrohr

  committed
  1a27e5d

  Copy full SHA for 1a27e5d

Commits on Aug 23, 2024

• GPU Compilation: print CUDA resource usage only if requested by GPUCA_KERNEL_RESOURCE_USAGE_VERBOSE flag, also print HIP resource usage

  davidrohr

  committed
  3e82457

  Copy full SHA for 3e82457

• GPU CMake: With new ROCm/Cmake, GCC usage seems to work, no need any more to enforce gcc-toolchain

  davidrohr

  committed
  9b6e47a

  Copy full SHA for 9b6e47a

Commits on Aug 20, 2024

• GPU CMake: Remove unsed O2_HIP_CMAKE_LINK_FLAGS and create HIP --offload-arch arguments only when needed

  davidrohr

  committed
  c481ad4

  Copy full SHA for c481ad4

Commits on Aug 9, 2024

• GPU CMake: Set amdgpu-function-calls=true, in case not set by the ROCm scripts / cmake files

  davidrohr

  committed
  d40ccfa

  Copy full SHA for d40ccfa

• GPU Cmake: allow unsupported GCC only if we don't autodetect CUDA

  davidrohr

  committed
  ded3e55

  Copy full SHA for ded3e55

Commits on Aug 8, 2024

• GPU CMake: Make compatible to ROCm >= 6.2

  davidrohr

  committed
  6c44a3f

  Copy full SHA for 6c44a3f

Commits on Aug 5, 2024

• GPU CMake: Workaround for CMake adding old cmake library paths implicitly if we request an old GCC for CUDA

  davidrohr

  committed
  bf35985

  Copy full SHA for bf35985

• GPU CMake: Bugfix: fix hardcoded old llvm version in case of manually specified clang binary

  davidrohr

  committed
  0457082

  Copy full SHA for 0457082

Commits on Apr 24, 2024

• GPU: Fix GPU build flags once more, HIP flags got their -O3 overridden by CXX flags O2

  davidrohr

  committed
  18126db

  Copy full SHA for 18126db

Commits on Apr 22, 2024

• GPU CMake: Fix -W option must be passed only to compiler, not to device compilation

  davidrohr

  committed
  5cb50ca

  Copy full SHA for 5cb50ca

Commits on Apr 6, 2024

• GPU: Silence some compiler warnings

  davidrohr

  committed
  36f0b9c

  Copy full SHA for 36f0b9c

Commits on Apr 1, 2024

• GPU CXX Standard: should not force it to 20 but use CMAKE_CXX_STANDARD

  davidrohr

  committed
  872ff06

  Copy full SHA for 872ff06

Commits on Mar 31, 2024

• CMake: Make sure CMAKE_BUILD_TYPE_UPPER is defined if GPU Cmake is used outside of O2

  davidrohr

  committed
  df31977

  Copy full SHA for df31977

Commits on Mar 30, 2024

• GPU: Bump GPU compilation to C++ 20

  davidrohr

  committed
  0280ce7

  Copy full SHA for 0280ce7

Commits on Mar 22, 2024

• GPU CMake: Fix FindO2GPU.cmake

  davidrohr

  committed
  5df07e6

  Copy full SHA for 5df07e6

• GPU HIP CMake: More fixes for ROCm detection, wasn't working on my laptop

  davidrohr

  committed
  c280308

  Copy full SHA for c280308

Commits on Mar 21, 2024

• GPU ROCM CMake: With HIP as language, it seems we may no longer set CMAKE_HIP_COMPILER and HIP_CLANG_PATH

  davidrohr

  committed
  03a5feb

  Copy full SHA for 03a5feb

• GPU HIP CMake: Automatically set -amdgpu-legacy-sgpr-spill-lowering=true workaround for ROCm >= 6.0

  davidrohr

  committed
  444a65a

  Copy full SHA for 444a65a

• GPU CMake: Use global hip_HIPIFY_PERL_EXECUTABLE variable

  davidrohr

  committed
  eb18b0a

  Copy full SHA for eb18b0a

• GPU Cmake: remove obsolete code

  davidrohr

  committed
  db32a53

  Copy full SHA for db32a53

• GPU HIP: -ffast-math seems to work now, and is needed for full performance

  davidrohr

  committed
  f0e9661

  Copy full SHA for f0e9661

Commits on Mar 20, 2024

• Follow conventions for CMAKE_BUILD_TYPE

  Show description for 2b56f82
  

  ktf

  authored and

  davidrohr

  committed
  2b56f82

  Copy full SHA for 2b56f82

Commits on Mar 14, 2024

• GPU: Extend CUDA compute target detection

  davidrohr

  committed
  facd33b

  Copy full SHA for facd33b

Commits on Jan 31, 2024

• GPU: Automatically disable math optimizations for TPC track model encoding/decoding code

  davidrohr

  committed
  3860e7f

  Copy full SHA for 3860e7f

• GPU: Need to disable some more stuff with GPUCA_NO_FAST_MATH

  davidrohr

  committed
  f758ea1

  Copy full SHA for f758ea1

Commits on Jan 14, 2024

• GPU: Suppress bogus CUDA compiler warning

  davidrohr

  committed
  e8309a8

  Copy full SHA for e8309a8

• GPU: Get rid of GPUReconstructionKernels file, and auto-generate based on kernels.cmake

  davidrohr

  committed
  fb33365

  Copy full SHA for fb33365

Commits on Jan 2, 2024

• GPU: Deprecated paths /opt/rocm/hip removed in ROCm6, just use /opt/rocm/bin

  davidrohr

  committed
  b0dc1b0

  Copy full SHA for b0dc1b0

• GPU: If we specify a clang version for OpenCL, use that also for llvm-spirv

  davidrohr

  committed
  7ac484f

  Copy full SHA for 7ac484f