gaspersic/triton
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

save

Browse Source
This commit is contained in:
Philippe Tillet
2015-02-01 15:58:05 -05:00
parent b0bf235cc2
commit 3f1fa822f8
6 changed files with 10 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
bench/CMakeLists.txt
View File

@@ -38,7 +38,7 @@ foreach(PROG blas overhead)
cuda_add_cublas_to_target(${PROG}-bench) cuda_add_cublas_to_target(${PROG}-bench)
else() else()
add_executable(${PROG}-bench ${PROG}.cpp) add_executable(${PROG}-bench ${PROG}.cpp)
set_target_properties(${PROG}-bench PROPERTIES COMPILE_FLAGS "-Wall -Wextra ${BLAS_DEF_STR}") set_target_properties(${PROG}-bench PROPERTIES COMPILE_FLAGS "-Wall -Wextra ${BLAS_DEF_STR} -std=c++11")
endif() endif()
target_link_libraries(${PROG}-bench ${BLAS_LIBS}) target_link_libraries(${PROG}-bench ${BLAS_LIBS})
endforeach(PROG) endforeach(PROG)

2
bench/blas.cpp
View File

@@ -32,7 +32,7 @@ void bench(ad::numeric_type dtype)
total_time = 0;\ total_time = 0;\
OP;\ OP;\
ad::cl_ext::synchronize(ad::cl_ext::default_context());\ ad::cl_ext::synchronize(ad::cl_ext::default_context());\
while(total_time < 1e-2){\ while(total_time < 1e-1){\
timer.start(); \ timer.start(); \
OP;\ OP;\
SYNC;\ SYNC;\

5
cmake_modules/FindCLAMDBLAS.cmake
View File

@@ -1,9 +1,10 @@
file(GLOB CLAMDBLAS_ROOT /opt/clAmdBlas*) file(GLOB CLAMDBLAS_ROOT /opt/clBlas*)
set(CLAMDBLAS_INCLUDE_HINTS "${CLAMDBLAS_ROOT}/include") set(CLAMDBLAS_INCLUDE_HINTS "${CLAMDBLAS_ROOT}/include")
set(CLAMDBLAS_LIBRARIES_HINTS "${CLAMDBLAS_ROOT}/lib64") set(CLAMDBLAS_LIBRARIES_HINTS "${CLAMDBLAS_ROOT}/lib64")
find_path(CLAMDBLAS_INCLUDE_DIR clAmdBlas.h HINTS ${CLAMDBLAS_INCLUDE_HINTS}) find_path(CLAMDBLAS_INCLUDE_DIR clAmdBlas.h HINTS ${CLAMDBLAS_INCLUDE_HINTS})
find_library(CLAMDBLAS_LIBRARIES NAMES clAmdBlas HINTS ${CLAMDBLAS_LIBRARIES_HINTS}) find_library(CLAMDBLAS_LIBRARIES NAMES clBLAS HINTS ${CLAMDBLAS_LIBRARIES_HINTS})
if(CLAMDBLAS_LIBRARIES) if(CLAMDBLAS_LIBRARIES)
set(CLAMDBLAS_LIBRARIES ${CLAMDBLAS_LIBRARIES}) set(CLAMDBLAS_LIBRARIES ${CLAMDBLAS_LIBRARIES})

2
lib/CMakeLists.txt
View File

@@ -6,7 +6,7 @@ include_directories(${CMAKE_CURRENT_SOURCE_DIR})
add_library(atidlas SHARED ${LIBATIDLAS_SRC}) add_library(atidlas SHARED ${LIBATIDLAS_SRC})
set_target_properties(atidlas PROPERTIES set_target_properties(atidlas PROPERTIES
COMPILE_FLAGS "-Wno-sign-compare -D__CL_ENABLE_EXCEPTIONS -Wall -Wextra -pedantic") COMPILE_FLAGS "-Wno-sign-compare -D__CL_ENABLE_EXCEPTIONS -Wall -Wextra -pedantic -std=c++11")
#install(TARGETS atidlas LIBRARY DESTINATION lib) #install(TARGETS atidlas LIBRARY DESTINATION lib)

4
lib/model/model.cpp
View File

@@ -256,8 +256,8 @@ model_map_t init_models(cl::CommandQueue & queue)
res[std::make_pair(MATRIX_PRODUCT_NT_TYPE, DTYPE)] = ptr_t(new model(mproduct_nt(1, 8, 8, 8, 4, 1, 4, FETCH_FROM_LOCAL, FETCH_FROM_LOCAL, 8, 8), queue)); res[std::make_pair(MATRIX_PRODUCT_NT_TYPE, DTYPE)] = ptr_t(new model(mproduct_nt(1, 8, 8, 8, 4, 1, 4, FETCH_FROM_LOCAL, FETCH_FROM_LOCAL, 8, 8), queue));
res[std::make_pair(MATRIX_PRODUCT_TT_TYPE, DTYPE)] = ptr_t(new model(mproduct_tt(1, 8, 8, 8, 4, 1, 4, FETCH_FROM_LOCAL, FETCH_FROM_LOCAL, 8, 8), queue)); res[std::make_pair(MATRIX_PRODUCT_TT_TYPE, DTYPE)] = ptr_t(new model(mproduct_tt(1, 8, 8, 8, 4, 1, 4, FETCH_FROM_LOCAL, FETCH_FROM_LOCAL, 8, 8), queue));
} }
if(const char * cmodel_file = std::getenv("ATIDLAS_MODEL_DEVICE_0")) if(const char * homepath = std::getenv("HOME"))
import(std::string(cmodel_file), queue, res); import(std::string(homepath) + "/.atidlas/devices/device0.json", queue, res);
return res; return res;
} }

4
python/postinstall.py
View File

@@ -1,4 +1,4 @@
import sys import os, sys
def add_input(help, default): def add_input(help, default):
sys.stdout.write(help + "[" + default + "] : ") sys.stdout.write(help + "[" + default + "] : ")
@@ -31,6 +31,6 @@ if retune=='y':
print '----------------' print '----------------'
opts += ['--blas3-size'] + [add_input('BLAS3 sizes (M,N,K)', '1024,1024,1024').split(',')] opts += ['--blas3-size'] + [add_input('BLAS3 sizes (M,N,K)', '1024,1024,1024').split(',')]
print '----------------' print '----------------'
subprocess.call(["${CMAKE_BINARY_DIR}/python/autotune/dist/autotune", "tune"] + opts +['--json', 'tmp.json']) subprocess.call(["${CMAKE_BINARY_DIR}/python/autotune/dist/autotune", "tune"] + opts +['--json', os.environ['HOME'] + '/.atidlas/devices/device' + opts[1] + '.json'])