Files
llama.cpp/ggml/src/ggml-hip/CMakeLists.txt
T
Johannes Gäßler d6f3030047 ggml: backend-agnostic tensor parallelism (experimental) (#19378)
* ggml: backend-agnostic tensor parallelism

* support for GPT-OSS, Qwen 3 MoE

* partial Vulkan fix

* add support for 4/8 GPUs

* unconditional peer access

* re-use buffers + ggml contexts

* fix output pattern

* NCCL support

* GGML: HIP: add RCCL support

* Remove shfl and AllReduce from backend interface

* move allocation workaround out of ggml-alloc.c

* 2d tensor set/get support

* Fix the seg fault without NCCL

* Apply suggestion from JohannesGaessler

* support for tensor dims % n_devs != 0

* fix view_offs scaling

* arbitrary num. of GPUs/tensor split

* fix compilation

* better granularity estimate

* Support device-specific host buffer types if all underlying backends expose the same type. This allows using pinned memory instead of pageable memory for CUDA.

Fix compilation errors.

* partial Qwen 3 Next support

* Fix qwen3 30b (#8)

* Fix crash with Qwen-30B-A3B Q4_0

Qwen-30B-A3B Q4_0 has an intermediate dimension of 768. Using a granularity of 256 forces an uneven split between GPUs, which is not supported by the current implementation.

* Decide block size based on tensor quantization type

* Fix crashes due to KV cache serialization (#9)

KV cache serialization requires non-zero offsets on the tensor. Add support in the meta backend to set/get a tensor with a non-zero offset.

* metal : fix build (#7)

* static memory allocations, fix usage count

* fix tensor granularity

* more even memory distribution

* use BF16 for allreduce

* rebase fixup

* better error message for unsupported architectures

* Fix device mismatch during scatter of allReduce. (#11)

There is a mismatch between the dst buffer device and the backend device, causing the use of sync copies

* Enable the previous allreduce implementation. It is better in both perf and stability (#12)

* delay AllReduce for Moe for less I/O

* build : clean-up compile warnings

* backend : move most of the meta backend API to ggml-backend-impl.h

* cont : hide unused public API in the implementation

* llama : use llama_device + remove ggml_backend_dev_is_meta()

* ggml-backend : remove unused alloc include

* minor : remove regex include

* ggml : introduce ggml-ext.h for staging new APIs

* rebase fixup

* fix tests

* llama : more robust logic for determining Meta devices (#16)

* llama : more robust logic for determining Meta devices

* cont : fix devs size check

Co-authored-by: Johannes Gäßler <johannesg@5d6.de>

* cont : fix log type

Co-authored-by: Johannes Gäßler <johannesg@5d6.de>

---------

Co-authored-by: Johannes Gäßler <johannesg@5d6.de>

* disable roundtrip for meta backend

* fix arch selection

* Qwen 3.5 support

* fix Gemma 4 MoE

* fix OpenVino, SYCL

* fix test-llama-archs for CPU-only builds

* Fix Qwen 3.5 MoE

* disable meta backend tests for WebGPU

* tests : filter CPU-based devices from the Meta backend tests (#17)

* meta : formatting, naming, indentation (#18)

* formatting : llama-model.cpp

* formatting : ggml-ext.h

* formatting : ggml-backend-meta.cpp

* meta : add TODO

* add documentation

* better error messages

* fix GPT-OSS

---------

Co-authored-by: Carl Philipp Klemm <carl@uvos.xyz>
Co-authored-by: Gaurav Garg <gaugarg@nvidia.com>
Co-authored-by: Georgi Gerganov <ggerganov@gmail.com>
2026-04-09 16:42:19 +02:00

158 lines
4.7 KiB
CMake

if (NOT EXISTS $ENV{ROCM_PATH})
if (NOT EXISTS /opt/rocm)
set(ROCM_PATH /usr)
else()
set(ROCM_PATH /opt/rocm)
endif()
else()
set(ROCM_PATH $ENV{ROCM_PATH})
endif()
list(APPEND CMAKE_PREFIX_PATH ${ROCM_PATH})
list(APPEND CMAKE_PREFIX_PATH "${ROCM_PATH}/lib64/cmake")
if (NOT DEFINED CMAKE_HIP_FLAGS_DEBUG)
set(CMAKE_HIP_FLAGS_DEBUG "-g -O2")
endif()
# CMake on Windows doesn't support the HIP language yet
if (WIN32)
set(CXX_IS_HIPCC TRUE)
else()
string(REGEX MATCH "hipcc(\.bat)?$" CXX_IS_HIPCC "${CMAKE_CXX_COMPILER}")
endif()
if (CXX_IS_HIPCC)
if (LINUX)
if (NOT ${CMAKE_CXX_COMPILER_ID} MATCHES "Clang")
message(WARNING "Only LLVM is supported for HIP, hint: CXX=/opt/rocm/llvm/bin/clang++")
endif()
message(WARNING "Setting hipcc as the C++ compiler is legacy behavior."
" Prefer setting the HIP compiler directly. See README for details.")
endif()
else()
# Forward (AMD)GPU_TARGETS to CMAKE_HIP_ARCHITECTURES.
if(AMDGPU_TARGETS AND NOT GPU_TARGETS)
set(GPU_TARGETS ${AMDGPU_TARGETS})
endif()
if(GPU_TARGETS AND NOT CMAKE_HIP_ARCHITECTURES)
set(CMAKE_HIP_ARCHITECTURES ${GPU_TARGETS})
endif()
cmake_minimum_required(VERSION 3.21)
enable_language(HIP)
endif()
find_package(hip REQUIRED)
find_package(hipblas REQUIRED)
find_package(rocblas REQUIRED)
if (GGML_HIP_RCCL)
find_package(rccl REQUIRED)
endif()
if (${hip_VERSION} VERSION_LESS 6.1)
message(FATAL_ERROR "At least ROCM/HIP V6.1 is required")
endif()
message(STATUS "HIP and hipBLAS found")
file(GLOB GGML_HEADERS_ROCM "../ggml-cuda/*.cuh")
list(APPEND GGML_HEADERS_ROCM "../../include/ggml-cuda.h")
file(GLOB GGML_SOURCES_ROCM "../ggml-cuda/*.cu")
file(GLOB SRCS "../ggml-cuda/template-instances/fattn-tile*.cu")
list(APPEND GGML_SOURCES_ROCM ${SRCS})
file(GLOB SRCS "../ggml-cuda/template-instances/fattn-mma*.cu")
list(APPEND GGML_SOURCES_ROCM ${SRCS})
file(GLOB SRCS "../ggml-cuda/template-instances/mmq*.cu")
list(APPEND GGML_SOURCES_ROCM ${SRCS})
file(GLOB SRCS "../ggml-cuda/template-instances/mmf*.cu")
list(APPEND GGML_SOURCES_ROCM ${SRCS})
if (GGML_CUDA_FA_ALL_QUANTS)
file(GLOB SRCS "../ggml-cuda/template-instances/fattn-vec*.cu")
list(APPEND GGML_SOURCES_ROCM ${SRCS})
add_compile_definitions(GGML_CUDA_FA_ALL_QUANTS)
else()
list(APPEND GGML_SOURCES_ROCM
../ggml-cuda/template-instances/fattn-vec-instance-f16-f16.cu
../ggml-cuda/template-instances/fattn-vec-instance-q4_0-q4_0.cu
../ggml-cuda/template-instances/fattn-vec-instance-q8_0-q8_0.cu
../ggml-cuda/template-instances/fattn-vec-instance-bf16-bf16.cu)
endif()
ggml_add_backend_library(ggml-hip
${GGML_HEADERS_ROCM}
${GGML_SOURCES_ROCM}
)
# TODO: do not use CUDA definitions for HIP
if (NOT GGML_BACKEND_DL)
target_compile_definitions(ggml PUBLIC GGML_USE_CUDA)
endif()
add_compile_definitions(GGML_USE_HIP)
if (GGML_CUDA_FORCE_MMQ)
add_compile_definitions(GGML_CUDA_FORCE_MMQ)
endif()
if (GGML_CUDA_FORCE_CUBLAS)
add_compile_definitions(GGML_CUDA_FORCE_CUBLAS)
endif()
if (GGML_CUDA_NO_PEER_COPY)
add_compile_definitions(GGML_CUDA_NO_PEER_COPY)
endif()
if (GGML_HIP_GRAPHS)
add_compile_definitions(GGML_HIP_GRAPHS)
endif()
if (GGML_HIP_NO_VMM)
add_compile_definitions(GGML_HIP_NO_VMM)
endif()
if (GGML_HIP_ROCWMMA_FATTN)
add_compile_definitions(GGML_HIP_ROCWMMA_FATTN)
endif()
if (NOT GGML_HIP_MMQ_MFMA)
add_compile_definitions(GGML_HIP_NO_MMQ_MFMA)
endif()
if (GGML_HIP_RCCL)
add_compile_definitions(GGML_USE_NCCL) # RCCL has the same interface as NCCL.
endif()
if (GGML_HIP_EXPORT_METRICS)
set(CMAKE_HIP_FLAGS "${CMAKE_HIP_FLAGS} -Rpass-analysis=kernel-resource-usage --save-temps")
endif()
if (NOT GGML_CUDA_FA)
add_compile_definitions(GGML_CUDA_NO_FA)
endif()
if (CXX_IS_HIPCC)
set_source_files_properties(${GGML_SOURCES_ROCM} PROPERTIES LANGUAGE CXX)
if (WIN32 AND CMAKE_BUILD_TYPE STREQUAL "Debug")
# CMake on Windows doesn't support the HIP language yet.
# Therefore we workaround debug build's failure on HIP backend this way.
set_source_files_properties(${GGML_SOURCES_ROCM} PROPERTIES COMPILE_FLAGS "-O2 -g")
endif()
target_link_libraries(ggml-hip PRIVATE hip::device)
else()
set_source_files_properties(${GGML_SOURCES_ROCM} PROPERTIES LANGUAGE HIP)
endif()
if (GGML_STATIC)
message(FATAL_ERROR "Static linking not supported for HIP/ROCm")
endif()
if (GGML_HIP_RCCL)
target_link_libraries(ggml-hip PRIVATE ggml-base roc::rccl)
endif()
target_link_libraries(ggml-hip PRIVATE ggml-base hip::host roc::rocblas roc::hipblas)