generic-library/opencv
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

add GPU module specific command line arguments

Browse Source
This commit is contained in:
marina.kolpakova 2012-10-10 15:16:28 +04:00
parent f88be98ec4
commit b28acfc12e
5 changed files with 112 additions and 87 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

73
modules/gpu/perf/perf_main.cpp
View File

@ -1,3 +1,74 @@
#include "perf_precomp.hpp"
CV_PERF_TEST_MAIN(gpu)
namespace{
static void printOsInfo()
{
#if defined _WIN32
# if defined _WIN64
printf("[----------]\n[ GPU INFO ] \tRun on OS Windows x64.\n[----------]\n"), fflush(stdout);
# else
printf("[----------]\n[ GPU INFO ] \tRun on OS Windows x32.\n[----------]\n"), fflush(stdout);
# endif
#elif defined linux
# if defined _LP64
printf("[----------]\n[ GPU INFO ] \tRun on OS Linux x64.\n[----------]\n"), fflush(stdout);
# else
printf("[----------]\n[ GPU INFO ] \tRun on OS Linux x32.\n[----------]\n"), fflush(stdout);
# endif
#elif defined __APPLE__
# if defined _LP64
printf("[----------]\n[ GPU INFO ] \tRun on OS Apple x64.\n[----------]\n"), fflush(stdout);
# else
printf("[----------]\n[ GPU INFO ] \tRun on OS Apple x32.\n[----------]\n"), fflush(stdout);
# endif
#endif
}
static void printCudaInfo()
{
printOsInfo();
#ifndef HAVE_CUDA
printf("[----------]\n[ GPU INFO ] \tOpenCV was built without CUDA support.\n[----------]\n"), fflush(stdout);
#else
int driver;
cudaDriverGetVersion(&driver);
printf("[----------]\n"), fflush(stdout);
printf("[ GPU INFO ] \tCUDA Driver version: %d.\n", driver), fflush(stdout);
printf("[ GPU INFO ] \tCUDA Runtime version: %d.\n", CUDART_VERSION), fflush(stdout);
printf("[----------]\n"), fflush(stdout);
printf("[----------]\n"), fflush(stdout);
printf("[ GPU INFO ] \tGPU module was compiled for the following GPU archs.\n"), fflush(stdout);
printf("[ BIN ] \t%s.\n", CUDA_ARCH_BIN), fflush(stdout);
printf("[ PTX ] \t%s.\n", CUDA_ARCH_PTX), fflush(stdout);
printf("[----------]\n"), fflush(stdout);
printf("[----------]\n"), fflush(stdout);
int deviceCount = cv::gpu::getCudaEnabledDeviceCount();
printf("[ GPU INFO ] \tCUDA device count:: %d.\n", deviceCount), fflush(stdout);
printf("[----------]\n"), fflush(stdout);
for (int i = 0; i < deviceCount; ++i)
{
cv::gpu::DeviceInfo info(i);
printf("[----------]\n"), fflush(stdout);
printf("[ DEVICE ] \t# %d %s.\n", i, info.name().c_str()), fflush(stdout);
printf("[ ] \tCompute capability: %d.%d\n", (int)info.majorVersion(), (int)info.minorVersion()), fflush(stdout);
printf("[ ] \tMulti Processor Count: %d\n", info.multiProcessorCount()), fflush(stdout);
printf("[ ] \tTotal memory: %d Mb\n", static_cast<int>(static_cast<int>(info.totalMemory() / 1024.0) / 1024.0)), fflush(stdout);
printf("[ ] \tFree memory: %d Mb\n", static_cast<int>(static_cast<int>(info.freeMemory() / 1024.0) / 1024.0)), fflush(stdout);
if (!info.isCompatible())
printf("[ GPU INFO ] \tThis device is NOT compatible with current GPU module build\n");
printf("[----------]\n"), fflush(stdout);
}
#endif
}
}
CV_PERF_TEST_MAIN(gpu, printCudaInfo())

67
modules/gpu/perf/utility.cpp
View File

@ -188,69 +188,4 @@ void PrintTo(const CvtColorInfo& info, ostream* os)
};
*os << str[info.code];
}
void ts::printOsInfo()
{
#if defined _WIN32
# if defined _WIN64
cout << "OS: Windows x64 \n" << endl;
# else
cout << "OS: Windows x32 \n" << endl;
# endif
#elif defined linux
# if defined _LP64
cout << "OS: Linux x64 \n" << endl;
# else
cout << "OS: Linux x32 \n" << endl;
# endif
#elif defined __APPLE__
# if defined _LP64
cout << "OS: Apple x64 \n" << endl;
# else
cout << "OS: Apple x32 \n" << endl;
# endif
#endif
}
void ts::printCudaInfo()
{
#if !defined HAVE_CUDA || defined(CUDA_DISABLER)
cout << "OpenCV was built without CUDA support \n" << endl;
#else
int driver;
cudaDriverGetVersion(&driver);
cout << "CUDA Driver version: " << driver << '\n';
cout << "CUDA Runtime version: " << CUDART_VERSION << '\n';
cout << endl;
cout << "GPU module was compiled for the following GPU archs:" << endl;
cout << " BIN: " << CUDA_ARCH_BIN << '\n';
cout << " PTX: " << CUDA_ARCH_PTX << '\n';
cout << endl;
int deviceCount = getCudaEnabledDeviceCount();
cout << "CUDA device count: " << deviceCount << '\n';
cout << endl;
for (int i = 0; i < deviceCount; ++i)
{
DeviceInfo info(i);
cout << "Device [" << i << "] \n";
cout << "\t Name: " << info.name() << '\n';
cout << "\t Compute capability: " << info.majorVersion() << '.' << info.minorVersion()<< '\n';
cout << "\t Multi Processor Count: " << info.multiProcessorCount() << '\n';
cout << "\t Total memory: " << static_cast<int>(static_cast<int>(info.totalMemory() / 1024.0) / 1024.0) << " Mb \n";
cout << "\t Free memory: " << static_cast<int>(static_cast<int>(info.freeMemory() / 1024.0) / 1024.0) << " Mb \n";
if (!info.isCompatible())
cout << "\t !!! This device is NOT compatible with current GPU module build \n";
cout << endl;
}
#endif
}
}

13
modules/gpu/perf/utility.hpp
View File

@ -46,21 +46,16 @@ DEF_PARAM_TEST(Sz_Depth_Cn, cv::Size, MatDepth, MatCn);
#define GPU_TYPICAL_MAT_SIZES testing::Values(perf::sz720p, perf::szSXGA, perf::sz1080p)
namespace ts {
void printOsInfo();
void printCudaInfo();
}
#define GPU_SANITY_CHECK(dmat, ...) \
do{ \
cv::Mat d##dmat(dmat); \
SANITY_CHECK(d##dmat, ## __VA_ARGS__); \
} while(0);
} while(0)
#define CPU_SANITY_CHECK(cmat, ...) \
do{ \
SANITY_CHECK(cmat, ## __VA_ARGS__); \
} while(0);
} while(0)
#define GPU_SANITY_CHECK_KEYPOINTS(alg, dmat, ...) \
do{ \
@ -75,13 +70,13 @@ namespace ts {
::perf::Regression::add(this, std::string(#dmat) + "-angle-row", __angle, ## __VA_ARGS__); \
::perf::Regression::add(this, std::string(#dmat) + "octave-row", __octave, ## __VA_ARGS__); \
::perf::Regression::add(this, std::string(#dmat) + "-pt-size-row", __size, ## __VA_ARGS__); \
} while(0);
} while(0)
#define GPU_SANITY_CHECK_RESPONSE(alg, dmat, ...) \
do{ \
cv::Mat d##dmat(dmat); \
cv::Mat __response = d##dmat.row(cv::gpu::alg##_GPU::RESPONSE_ROW); \
::perf::Regression::add(this, std::string(#dmat) + "-response-row", __response, ## __VA_ARGS__); \
} while(0);
} while(0)
#endif // __OPENCV_PERF_GPU_UTILITY_HPP__

6
modules/ts/include/opencv2/ts/ts_perf.hpp
View File

@ -206,7 +206,6 @@ private:
#define SANITY_CHECK_MATCHES(array, ...) ::perf::Regression::addMatches(this, #array, array , ## __VA_ARGS__)
#ifdef HAVE_CUDA
//#error "CUDA"
class CV_EXPORTS GpuPerf
{
public:
@ -215,7 +214,7 @@ public:
# define PERF_RUN_GPU() ::perf::GpuPerf::targetDevice()
#else
# define PERF_RUN_GPU()
# define PERF_RUN_GPU() false
#endif
@ -478,9 +477,10 @@ CV_EXPORTS void PrintTo(const Size& sz, ::std::ostream* os);
void fixture##_##name::PerfTestBody()
#define CV_PERF_TEST_MAIN(testsuitname) \
#define CV_PERF_TEST_MAIN(testsuitname, ...) \
int main(int argc, char **argv)\
{\
__VA_ARGS__;\
::perf::Regression::Init(#testsuitname);\
::perf::TestBase::Init(argc, argv);\
::testing::InitGoogleTest(&argc, argv);\

40
modules/ts/src/ts_perf.cpp
View File

@ -17,18 +17,20 @@ const std::string command_line_keys =
"{ perf_seed |809564 |seed for random numbers generator}"
"{ perf_tbb_nthreads |-1 |if TBB is enabled, the number of TBB threads}"
"{ perf_write_sanity | |allow to create new records for sanity checks}"
#ifdef ANDROID
#ifdef ANDROID
"{ perf_time_limit |6.0 |default time limit for a single test (in seconds)}"
"{ perf_affinity_mask |0 |set affinity mask for the main thread}"
"{ perf_log_power_checkpoints | |additional xml logging for power measurement}"
#else
#else
"{ perf_time_limit |3.0 |default time limit for a single test (in seconds)}"
#endif
#endif
"{ perf_max_deviation |1.0 |}"
"{ help h | |print help info}"
#ifdef HAVE_CUDA
"{ perf_run_cpu |false |run GPU performance tests for analogy CPU functions}"
#endif
#ifdef HAVE_CUDA
"{ perf_run_cpu |false |run GPU performance tests for analogical CPU functions}"
"{ perf_cuda_device |0 |run GPU test suite onto specific CUDA capable device}"
"{ perf_cuda_info_only |false |print an information about system and an available CUDA devices and then exit.}"
#endif
;
static double param_max_outliers;
@ -41,6 +43,7 @@ static int param_tbb_nthreads;
static bool param_write_sanity;
#ifdef HAVE_CUDA
static bool param_run_cpu;
static int param_cuda_device;
#endif
#ifdef ANDROID
static int param_affinity_mask;
@ -64,6 +67,10 @@ static void setCurrentThreadAffinityMask(int mask)
#endif
#ifdef HAVE_CUDA
# include <opencv2/core/gpumat.hpp>
#endif
static void randu(cv::Mat& m)
{
const int bigValue = 0x00000FFF;
@ -617,12 +624,30 @@ void TestBase::Init(int argc, const char* const argv[])
#endif
#ifdef HAVE_CUDA
bool printOnly = args.has("perf_cuda_info_only");
if (printOnly)
exit(0);
param_run_cpu = args.has("perf_run_cpu");
param_cuda_device = std::max(0, std::min(cv::gpu::getCudaEnabledDeviceCount(), args.get<int>("perf_cuda_device")));
if (param_run_cpu)
printf("[----------]\n[ GPU INFO ] \tRun test suite on CPU.\n[----------]\n"), fflush(stdout);
else
printf("[----------]\n[ GPU INFO ] \tRun test suite on GPU.\n[----------]\n"), fflush(stdout);
{
cv::gpu::DeviceInfo info(param_cuda_device);
if (!info.isCompatible())
{
printf("[----------]\n[ FAILURE ] \tDevice %s is NOT compatible with current GPU module build.\n[----------]\n", info.name().c_str()), fflush(stdout);
exit(-1);
}
cv::gpu::setDevice(param_cuda_device);
printf("[----------]\n[ GPU INFO ] \tRun test suite on %s GPU.\n[----------]\n", info.name().c_str()), fflush(stdout);
}
#endif
if (!args.check())
@ -1212,7 +1237,6 @@ bool perf::GpuPerf::targetDevice()
}
#endif
/*****************************************************************************************\
* ::perf::PrintTo
\*****************************************************************************************/