Skip to content

nix: Added missing packages and options for ROCm build #15747

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

Open
wants to merge 4 commits into
base: master
Choose a base branch
from
+28 −8
Open

nix: Added missing packages and options for ROCm build #15747

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
789ba53
nix: Added missing packages and options for ROCm build
SteelPh0enix Sep 2, 2025
5d3169f
nix: Renamed CMAKE_HIP_ARCHITECTURES to AMDGPU_TARGETS
SteelPh0enix Sep 2, 2025
50801be
nix: Added and enabled rocmWmma support along with ROWMMA_PATH option…
SteelPh0enix Sep 2, 2025
d99b83f
nix: Added enableUma option
SteelPh0enix Sep 2, 2025
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
31 changes: 24 additions & 7 deletions .devops/nix/package.nix
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32,8 +32,11 @@
useMpi ? false,
useRocm ? config.rocmSupport,
rocmGpuTargets ? builtins.concatStringsSep ";" rocmPackages.clr.gpuTargets,
rocmUseWmma ? true,
enableCurl ? true,
useVulkan ? false,
buildAllCudaFaQuants ? false,
enableUma ? false,
llamaVersion ? "0.0.0", # Arbitrary version, substituted by the flake

# It's necessary to consistently use backendStdenv when building with CUDA support,
Expand Down Expand Up @@ -91,11 +94,16 @@ let
libcublas
];

rocmBuildInputs = with rocmPackages; [
clr
hipblas
rocblas
];
rocmBuildInputs =
with rocmPackages;
[
clr
hipblas
rocblas
llvm.lld
llvm.bintools
]
++ optionals rocmUseWmma [ rocmPackages.rocwmma ];

vulkanBuildInputs = [
vulkan-headers
Expand Down Expand Up @@ -160,7 +168,8 @@ effectiveStdenv.mkDerivation (finalAttrs: {
buildInputs =
optionals effectiveStdenv.isDarwin darwinBuildInputs
++ optionals useCuda cudaBuildInputs
++ optionals useMpi [ mpi ]
++ optionals (useMpi && !useRocm) [ mpi ]
++ optionals (useMpi && useRocm) [ rocmPackages.mpi ]
++ optionals useRocm rocmBuildInputs
++ optionals useBlas [ blas ]
++ optionals useVulkan vulkanBuildInputs
Expand All @@ -187,10 +196,18 @@ effectiveStdenv.mkDerivation (finalAttrs: {
builtins.concatStringsSep ";" (map dropDot cudaCapabilities)
)
)
(cmakeBool "GGML_CUDA_FA_ALL_QUANTS" buildAllCudaFaQuants)
(cmakeBool "GGML_CUDA_ENABLE_UNIFIED_MEMORY" enableUma)
]
++ optionals useRocm [
(cmakeFeature "CMAKE_HIP_COMPILER" "${rocmPackages.llvm.clang}/bin/clang")
(cmakeFeature "CMAKE_HIP_ARCHITECTURES" rocmGpuTargets)
(cmakeFeature "AMDGPU_TARGETS" rocmGpuTargets)
(cmakeBool "GGML_CUDA_FA_ALL_QUANTS" buildAllCudaFaQuants)
(cmakeBool "GGML_CUDA_ENABLE_UNIFIED_MEMORY" enableUma)
]
++ optionals rocmUseWmma [
(cmakeBool "GGML_HIP_ROCWMMA_FATTN" rocmUseWmma)
(cmakeFeature "GGML_HIP_ROCWMMA_PATH" "${rocmPackages.rocwmma}")
]
++ optionals useMetalKit [
(lib.cmakeFeature "CMAKE_C_FLAGS" "-D__ARM_FEATURE_DOTPROD=1")
Expand Down
5 changes: 4 additions & 1 deletion ggml/src/ggml-hip/CMakeLists.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -39,7 +39,7 @@ endif()
find_package(hip REQUIRED)
find_package(hipblas REQUIRED)
find_package(rocblas REQUIRED)
if (GGML_HIP_ROCWMMA_FATTN)
if (GGML_HIP_ROCWMMA_FATTN AND NOT GGML_HIP_ROCWMMA_PATH)
CHECK_INCLUDE_FILE_CXX("rocwmma/rocwmma.hpp" FOUND_ROCWMMA)
if (NOT ${FOUND_ROCWMMA})
message(FATAL_ERROR "rocwmma has not been found")
Expand Down Expand Up @@ -111,6 +111,9 @@ endif()

if (GGML_HIP_ROCWMMA_FATTN)
add_compile_definitions(GGML_HIP_ROCWMMA_FATTN)
if (GGML_HIP_ROCWMMA_PATH)
target_include_directories(ggml-hip PRIVATE ${GGML_HIP_ROCWMMA_PATH}/include)
endif()
endif()

if (NOT GGML_HIP_MMQ_MFMA)
Expand Down