parent
cbffab0a26
commit
68e778bed0
|
@ -53,6 +53,6 @@ public class LoadBackendTests {
|
|||
@Test
|
||||
public void loadCudaDLL() {
|
||||
System.load(
|
||||
"C:\\Users\\brian\\_projects\\deeplearning4j\\cavis-native\\cavis-native-lib\\build\\generated\\sources\\javacpp\\cuda\\windows-x86_64-avx2\\jni_nd4jcuda.dll");
|
||||
"C:\\Users\\brian\\_projects\\deeplearning4j\\cavis-native\\cavis-native-lib\\build\\generated\\sources\\javacpp\\cuda\\windows-x86_64-avx2\\jnind4jcuda.dll");
|
||||
}
|
||||
}
|
||||
|
|
|
@ -24,7 +24,9 @@ option(FLATBUFFERS_BUILD_FLATC "Enable the build of the flatbuffers compiler" OF
|
|||
set(FLATBUFFERS_BUILD_FLATC "OFF" CACHE STRING "Hack to disable flatc build" FORCE)
|
||||
|
||||
set(CMAKE_CXX_STANDARD 14)
|
||||
set(THREADS_PREFER_PTHREAD_FLAG ON)
|
||||
|
||||
set(CMAKE_THREAD_PREFER_PTHREAD TRUE)
|
||||
set(THREADS_PREFER_PTHREAD_FLAG TRUE)
|
||||
find_package(Threads REQUIRED)
|
||||
|
||||
#///////////////////////////////////////////////////////////////////////////////
|
||||
|
@ -364,7 +366,7 @@ if(SD_BUILD_TESTS)
|
|||
# tests are always compiled with all ops included
|
||||
set(SD_ALL_OPS true)
|
||||
set(SD_BUILD_MINIFIER true)
|
||||
add_subdirectory(tests_cpu)
|
||||
add_subdirectory(src/test/cpp/tests_cpu)
|
||||
endif()
|
||||
|
||||
|
||||
|
|
|
@ -251,7 +251,8 @@ chipList.each { String thisChip ->
|
|||
'--chip-extension', avxExtension,
|
||||
'-j', "${host_cores}",
|
||||
// '--helper', 'mkldnn',
|
||||
'--helper', 'cudnn']
|
||||
'--helper', 'cudnn',
|
||||
'--tests']
|
||||
} else if (thisChip.equals('cuda') && osdetector.os.startsWith("linux")) { //cuDNN requires CUDA
|
||||
it.buildCommand = ['bash', 'buildnativeoperations.sh',
|
||||
'-V',
|
||||
|
|
|
@ -216,32 +216,15 @@ if(SD_CUDA)
|
|||
endif()
|
||||
endif()
|
||||
|
||||
#
|
||||
#string( TOLOWER "${COMPUTE}" COMPUTE_CMP )
|
||||
# if ("${COMPUTE_CMP}" STREQUAL "all")
|
||||
# CUDA_SELECT_NVCC_ARCH_FLAGS(CUDA_ARCH_FLAGS "Common")
|
||||
# elseif("${COMPUTE_CMP}" STREQUAL "auto")
|
||||
# CUDA_SELECT_NVCC_ARCH_FLAGS(CUDA_ARCH_FLAGS "Auto")
|
||||
# elseif(COMPUTE_CMP MATCHES "^[0-9]+$")
|
||||
# #matches USER COMPUTE old way
|
||||
#set(CUDA_ARCH_FLAGS "-gencode arch=compute_${COMPUTE},code=sm_${COMPUTE} ")
|
||||
# else()
|
||||
# #matches numbers NAME | NUM.NUM | NUM.NUM(NUM.NUM) | NUM.NUM+PTX
|
||||
# #NAME: Fermi Kepler Maxwell Kepler+Tegra Kepler+Tesla Maxwell+Tegra Pascal
|
||||
# #NUM: 2.0 2.1 3.0 3.2 3.5 3.7 5.0 5.2 5.3 6.0 6.2 et cetera
|
||||
# CUDA_SELECT_NVCC_ARCH_FLAGS(CUDA_ARCH_FLAGS "${COMPUTE}")
|
||||
# endif()
|
||||
# list to spaces
|
||||
#string (REPLACE ";" " " CUDA_ARCH_FLAGS "${CUDA_ARCH_FLAGS}")
|
||||
|
||||
#set(CMAKE_CUDA_FLAGS " ${CMAKE_CUDA_FLAGS} -DCUDA_VERSION_MAJOR=${CUDA_VERSION_MAJOR} ${EXPM} -w --cudart=static --expt-extended-lambda -Xfatbin -compress-all ")
|
||||
set(CMAKE_CUDA_ARCHITECTURES OFF)
|
||||
set(CMAKE_CUDA_ARCHITECTURES "all-major")
|
||||
set(CMAKE_CUDA_RUNTIME_LIBRARY "shared")
|
||||
#set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --generate-code \"arch=compute_53,code=[compute_53,sm_53]\" " )
|
||||
set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --generate-code \"arch=compute_61,code=[compute_61,sm_61]\" " )
|
||||
#set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --generate-code \"arch=compute_61,code=[compute_61,sm_61]\" " )
|
||||
#set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --generate-code \"arch=compute_75,code=[compute_75,sm_75]\" " )
|
||||
set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --extended-lambda ")
|
||||
set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} --expt-relaxed-constexpr ")
|
||||
set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -DCUDA_VERSION_MAJOR=12 -w --cudart=static -Xfatbin -compress-all")
|
||||
set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -DCUDA_VERSION_MAJOR=12 -w -Xfatbin -compress-all")
|
||||
set(CUDAHOSTCXX "${CMAKE_CXX_COMPILER}")
|
||||
if(WIN32)
|
||||
set(CMAKE_CUDA_FLAGS "${CMAKE_CUDA_FLAGS} -Xcompiler=/EHsc")
|
||||
|
|
|
@ -1,5 +1,5 @@
|
|||
include_directories(${gtest_SOURCE_DIR}/include ${gtest_SOURCE_DIR})
|
||||
include_directories(${CMAKE_CURRENT_SOURCE_DIR}/include)
|
||||
include_directories(../../../../../src/main/cpp/blas)
|
||||
if(LINUX)
|
||||
link_directories(/usr/local/lib)
|
||||
link_directories(/usr/lib)
|
||||
|
@ -22,9 +22,13 @@ if(WIN32)
|
|||
endif()
|
||||
|
||||
if (SD_CUDA)
|
||||
find_package(CUDA)
|
||||
message("Tests CUDA include directory: ${CUDA_INCLUDE_DIRS}")
|
||||
include_directories(${CUDA_INCLUDE_DIRS})
|
||||
find_package(CUDAToolkit 12.2 REQUIRED)
|
||||
enable_language(CUDA)
|
||||
|
||||
set(CMAKE_CUDA_STANDARD 17)
|
||||
set(CMAKE_CXX_STANDARD 14)
|
||||
message("Tests CUDA include directory: ${CUDAToolkit_INCLUDE_DIRS}")
|
||||
include_directories(${CUDAToolkit_INCLUDE_DIRS})
|
||||
add_definitions(-D__CUDABLAS__=true)
|
||||
|
||||
if(WIN32)
|
||||
|
@ -34,17 +38,14 @@ if (SD_CUDA)
|
|||
|
||||
string( TOLOWER "${COMPUTE}" COMPUTE_CMP )
|
||||
if ("${COMPUTE_CMP}" STREQUAL "all")
|
||||
CUDA_SELECT_NVCC_ARCH_FLAGS(CUDA_ARCH_FLAGS "Common")
|
||||
set(CMAKE_CUDA_ARCHITECTURES "all")
|
||||
elseif("${COMPUTE_CMP}" STREQUAL "auto")
|
||||
CUDA_SELECT_NVCC_ARCH_FLAGS(CUDA_ARCH_FLAGS "Auto")
|
||||
elseif(COMPUTE_CMP MATCHES "^[0-9]+$")
|
||||
#matches USER COMPUTE old way
|
||||
set(CUDA_ARCH_FLAGS "-gencode arch=compute_${COMPUTE},code=sm_${COMPUTE} ")
|
||||
set(CMAKE_CUDA_ARCHITECTURES "all-major")
|
||||
else()
|
||||
#matches numbers NAME | NUM.NUM | NUM.NUM(NUM.NUM) | NUM.NUM+PTX
|
||||
#NAME: Fermi Kepler Maxwell Kepler+Tegra Kepler+Tesla Maxwell+Tegra Pascal
|
||||
#NUM: 2.0 2.1 3.0 3.2 3.5 3.7 5.0 5.2 5.3 6.0 6.2 et cetera
|
||||
CUDA_SELECT_NVCC_ARCH_FLAGS(CUDA_ARCH_FLAGS "${COMPUTE}")
|
||||
set(CMAKE_CUDA_ARCHITECTURES "all")
|
||||
endif()
|
||||
# list to spaces
|
||||
string (REPLACE ";" " " CUDA_ARCH_FLAGS "${CUDA_ARCH_FLAGS}")
|
Some files were not shown because too many files have changed in this diff Show More
Loading…
Reference in New Issue