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")

# 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 AMDGPU_TARGETS to CMAKE_HIP_ARCHITECTURES.
    if (AMDGPU_TARGETS AND NOT CMAKE_HIP_ARCHITECTURES)
        set(CMAKE_HIP_ARCHITECTURES ${AMDGPU_TARGETS})
    endif()
    cmake_minimum_required(VERSION 3.21)
    enable_language(HIP)
endif()

find_package(hip     REQUIRED)
find_package(hipblas REQUIRED)
find_package(rocblas REQUIRED)

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-wmma*.cu")
list(APPEND GGML_SOURCES_ROCM ${SRCS})
file(GLOB   SRCS "../ggml-cuda/template-instances/mmq*.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()
    file(GLOB   SRCS "../ggml-cuda/template-instances/fattn-vec*q4_0-q4_0.cu")
    list(APPEND GGML_SOURCES_ROCM ${SRCS})
    file(GLOB   SRCS "../ggml-cuda/template-instances/fattn-vec*q8_0-q8_0.cu")
    list(APPEND GGML_SOURCES_ROCM ${SRCS})
    file(GLOB   SRCS "../ggml-cuda/template-instances/fattn-vec*f16-f16.cu")
    list(APPEND GGML_SOURCES_ROCM ${SRCS})
endif()

ggml_add_backend_library(ggml-hip
                         ${GGML_HEADERS_ROCM}
                         ${GGML_SOURCES_ROCM}
                        )

# TODO: do not use CUDA definitions for HIP
target_compile_definitions(ggml PUBLIC GGML_USE_CUDA)

add_compile_definitions(GGML_USE_HIP)

if (GGML_HIP_UMA)
    add_compile_definitions(GGML_HIP_UMA)
endif()

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 (CXX_IS_HIPCC)
    set_source_files_properties(${GGML_SOURCES_ROCM} PROPERTIES LANGUAGE CXX)
    target_link_libraries(ggml-hip PRIVATE hip::device)
else()
    set_source_files_properties(${GGML_SOURCES_ROCM} PROPERTIES LANGUAGE HIP)
endif()

if (GGML_STATIC)
    message(FATAL_ERROR "Static linking not supported for HIP/ROCm")
endif()

target_link_libraries(ggml-hip PRIVATE ggml-base hip::host roc::rocblas roc::hipblas)