update CMake

1 files changed, 158 insertions, 3 deletions

diff --git a/CMakeLists.txt b/CMakeLists.txt
index ac01604..f88a2a9 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -38,6 +38,128 @@ set_property(CACHE CMAKE_BUILD_TYPE PROPERTY STRINGS "${BUILD_TYPE}")
 option(CMAKE_LINK_STATIC "link as much as possible libraries static" OFF)
 
 ################################################################################
+# Find CUDA
+################################################################################
+#option(CUDA_USE_STATIC_CUDA_RUNTIME "Use the static version of the CUDA runtime library if available" OFF)
+#set(CUDA_USE_STATIC_CUDA_RUNTIME OFF CACHE BOOL "Use the static version of the CUDA runtime library if available" FORCE)
+
+find_package(CUDA 7.5 QUIET)
+
+if(CUDA_FOUND)
+
+    option(XMR-STAK_LARGEGRID "Support large CUDA block count > 128" ON)
+    if(XMR-STAK_LARGEGRID)
+        add_definitions("-DXMR_STAK_LARGEGRID=${XMR-STAK_LARGEGRID}")
+    endif()
+
+    set(DEVICE_COMPILER "nvcc")
+    set(CUDA_COMPILER "${DEVICE_COMPILER}" CACHE STRING "Select the device compiler")
+
+    if(${CMAKE_CXX_COMPILER_ID} STREQUAL "Clang")
+        list(APPEND DEVICE_COMPILER "clang")
+    endif()
+
+    set_property(CACHE CUDA_COMPILER PROPERTY STRINGS "${DEVICE_COMPILER}")
+
+    set(XMR-STAK_THREADS 0 CACHE STRING "Set maximum number of threads (for compile time optimization)")
+    if(NOT XMR-STAK_THREADS EQUAL 0)
+        message(STATUS "xmr-stak-nvidia: set max threads per block to ${XMR-STAK_THREADS}")
+        add_definitions("-DXMR_STAK_THREADS=${XMR-STAK_THREADS}")
+    endif()
+
+    set(CUDA_ARCH "20;30;35;37;50;52;60;61;62" CACHE STRING "Set GPU architecture (semicolon separated list, e.g. '-DCUDA_ARCH=20;35;60')")
+
+    # validate architectures (only numbers are allowed)
+    foreach(CUDA_ARCH_ELEM ${CUDA_ARCH})
+        string(REGEX MATCH "^[0-9]+$" IS_NUMBER ${CUDA_ARCH})
+        if(NOT IS_NUMBER)
+            message(FATAL_ERROR "Defined compute architecture '${CUDA_ARCH_ELEM}' in "
+                                "'${CUDA_ARCH}' is not an integral number, use e.g. '30' (for compute architecture 3.0).")
+        endif()
+        unset(IS_NUMBER)
+
+        if(${CUDA_ARCH_ELEM} LESS 20)
+            message(FATAL_ERROR "Unsupported CUDA architecture '${CUDA_ARCH_ELEM}' specified. "
+                                "Use '20' (for compute architecture 2.0) or higher.")
+        endif()
+    endforeach()
+
+    option(CUDA_SHOW_REGISTER "Show registers used for each kernel and compute architecture" OFF)
+    option(CUDA_KEEP_FILES "Keep all intermediate files that are generated during internal compilation steps" OFF)
+
+    if("${CUDA_COMPILER}" STREQUAL "clang")
+        set(CLANG_BUILD_FLAGS "-O3 -x cuda --cuda-path=${CUDA_TOOLKIT_ROOT_DIR}")
+        # activation usage of FMA
+        set(CLANG_BUILD_FLAGS "${CLANG_BUILD_FLAGS} -ffp-contract=fast")
+
+        if(CUDA_SHOW_REGISTER)
+            set(CLANG_BUILD_FLAGS "${CLANG_BUILD_FLAGS} -Xcuda-ptxas -v")
+        endif(CUDA_SHOW_REGISTER)
+
+        if(CUDA_KEEP_FILES)
+            set(CLANG_BUILD_FLAGS "${CLANG_BUILD_FLAGS} -save-temps=${PROJECT_BINARY_DIR}")
+        endif(CUDA_KEEP_FILES)
+
+        foreach(CUDA_ARCH_ELEM ${CUDA_ARCH})
+            # set flags to create device code for the given architectures
+            set(CLANG_BUILD_FLAGS "${CLANG_BUILD_FLAGS} --cuda-gpu-arch=sm_${CUDA_ARCH_ELEM}")
+        endforeach()
+
+    elseif("${CUDA_COMPILER}" STREQUAL "nvcc")
+        # avoid that nvcc in CUDA < 8 tries to use libc `memcpy` within the kernel
+        if(CUDA_VERSION VERSION_LESS 8.0)
+            add_definitions(-D_FORCE_INLINES)
+        endif()
+        foreach(CUDA_ARCH_ELEM ${CUDA_ARCH})
+            # set flags to create device code for the given architecture
+            set(CUDA_NVCC_FLAGS ${CUDA_NVCC_FLAGS}
+                "--generate-code arch=compute_${CUDA_ARCH_ELEM},code=sm_${CUDA_ARCH_ELEM} --generate-code arch=compute_${CUDA_ARCH_ELEM},code=compute_${CUDA_ARCH_ELEM}")
+        endforeach()
+
+        # give each thread an independent default stream
+        set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS} --default-stream per-thread")
+
+        option(CUDA_SHOW_CODELINES "Show kernel lines in cuda-gdb and cuda-memcheck" OFF)
+
+        if(CUDA_SHOW_CODELINES)
+            set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS}" --source-in-ptx -lineinfo)
+            set(CUDA_KEEP_FILES ON CACHE BOOL "activate keep files" FORCE)
+        endif(CUDA_SHOW_CODELINES)
+
+        if(CUDA_SHOW_REGISTER)
+            set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS}" -Xptxas=-v)
+        endif(CUDA_SHOW_REGISTER)
+
+        if(CUDA_KEEP_FILES)
+            set(CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS}" --keep --keep-dir "${PROJECT_BINARY_DIR}")
+        endif(CUDA_KEEP_FILES)
+
+    else()
+        message(FATAL_ERROR "selected CUDA compiler '${CUDA_COMPILER}' is not supported")
+    endif()
+else()
+    add_definitions("-DCONF_NO_CUDA")
+endif()
+
+# help to find AMD app SDK on systems with a software module system
+list(APPEND CMAKE_PREFIX_PATH "$ENV{AMDAPPSDKROOT}")
+# allow user to extent CMAKE_PREFIX_PATH via environment variable
+list(APPEND CMAKE_PREFIX_PATH "$ENV{CMAKE_PREFIX_PATH}")
+
+###############################################################################
+# Find OpenCL
+###############################################################################
+
+find_package(OpenCL)
+include_directories(SYSTEM ${OpenCL_INCLUDE_DIRS})
+#set(LIBS ${LIBS} ${OpenCL_LIBRARY})
+link_directories(${OpenCL_LIBRARY})
+
+if(NOT OpenCL_FOUND)
+    add_definitions("-DCONF_NO_OPENCL")
+endif()
+
+################################################################################
 # Find PThreads
 ################################################################################
 
@@ -111,7 +233,7 @@ if(HWLOC_ENABLE)
             /usr/local
             /usr
             ENV "PROGRAMFILES(X86)"
-            ENV "HWLOC_ROOT"
+            ENV "MICROHTTPD_ROOT"
         PATH_SUFFIXES
             include)
 
@@ -160,7 +282,28 @@ if(CMAKE_LINK_STATIC)
     set(LIBS "-static-libgcc -static-libstdc++ ${LIBS}")
 endif()
 
-file(GLOB SRCFILES_CPP "*.cpp" "crypto/*.cpp")
+if(CUDA_FOUND)
+    file(GLOB CUDASRCFILES "backend/nvidia/nvcc_code/*.cu" "backend/nvidia/*.cpp")
+
+    #  build device code with nvcc
+    cuda_add_library(xmrstak_cuda_backend
+        SHARED
+        ${CUDASRCFILES}
+    )
+    target_link_libraries(xmrstak_cuda_backend  ${CUDA_LIBRARIES})
+endif()
+
+if(OpenCL_FOUND)
+    file(GLOB OPENCLSRCFILES_C  "backend/amd/amd_gpu/*.cpp" "console.cpp" "backend/amd/*.cpp")
+    add_library(xmrstak_opencl_backend
+        SHARED
+        ${OPENCLSRCFILES_C}
+    )
+    #set_property(TARGET xmrstak_opencl_backend PROPERTY C_STANDARD 99)
+    target_link_libraries(xmrstak_opencl_backend PUBLIC ${OpenCL_LIBRARY})
+endif()
+
+file(GLOB SRCFILES_CPP "*.cpp" "crypto/*.cpp"  "backend/cpu/*.cpp" "backend/*.cpp")
 file(GLOB SRCFILES_C "crypto/*.c")
 
 add_library(xmr-stak-c
@@ -175,7 +318,9 @@ add_executable(xmr-stak-cpu
 )
 
 set(EXECUTABLE_OUTPUT_PATH "bin")
-target_link_libraries(xmr-stak-cpu ${LIBS} xmr-stak-c)
+set(LIBRARY_OUTPUT_PATH "bin")
+
+target_link_libraries(xmr-stak-cpu ${LIBS} xmr-stak-c ${CMAKE_DL_LIBS})
 
 ################################################################################
 # Install
@@ -185,8 +330,18 @@ target_link_libraries(xmr-stak-cpu ${LIBS} xmr-stak-c)
 if( NOT "${CMAKE_INSTALL_PREFIX}" STREQUAL "${PROJECT_BINARY_DIR}" )
     install(TARGETS xmr-stak-cpu
             RUNTIME DESTINATION "${CMAKE_INSTALL_PREFIX}/bin")
+    if(CUDA_FOUND)
+        install(TARGETS xmrstak_cuda_backend
+            LIBRARY DESTINATION "${CMAKE_INSTALL_PREFIX}/bin")
+    endif()
+    if(OpenCL_FOUND)
+        install(TARGETS xmrstak_opencl_backend
+            LIBRARY DESTINATION "${CMAKE_INSTALL_PREFIX}/bin")
+    endif()
 endif()
 
+install(DIRECTORY "${CMAKE_CURRENT_SOURCE_DIR}/opencl"
+        DESTINATION "${CMAKE_INSTALL_PREFIX}/bin")
 
 # avoid overwrite of user defined settings
 # install `config.txt`if file not exists in `${CMAKE_INSTALL_PREFIX}/bin`