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

implemented gpu::gemm via CUBLAS

Browse Source
This commit is contained in:
Vladislav Vinogradov
2011-10-19 13:29:54 +00:00
parent 90ff3dd990
commit e7502e7641
6 changed files with 259 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

31
modules/gpu/perf/perf_arithm.cpp
View File

@@ -747,3 +747,34 @@ PERF_TEST_P(DevInfo_Size_MatType_FlipCode, reduce, testing::Combine(testing::Val
SANITY_CHECK(dst_host);
}
PERF_TEST_P(DevInfo_Size, gemm, testing::Combine(testing::ValuesIn(devices()),
testing::Values(cv::Size(512, 512), cv::Size(1024, 1024), cv::Size(2048, 2048), cv::Size(4096, 4096))))
{
DeviceInfo devInfo = std::tr1::get<0>(GetParam());
Size size = std::tr1::get<1>(GetParam());
setDevice(devInfo.deviceID());
Mat src1_host(size, CV_32FC1);
Mat src2_host(size, CV_32FC1);
Mat src3_host(size, CV_32FC1);
declare.in(src1_host, src2_host, src3_host, WARMUP_RNG);
GpuMat src1(src1_host);
GpuMat src2(src2_host);
GpuMat src3(src3_host);
GpuMat dst(size, CV_32FC1);
declare.time(5.0);
SIMPLE_TEST_CYCLE()
{
gemm(src1, src2, 1.0, src3, 1.0, dst);
}
Mat dst_host = dst;
SANITY_CHECK(dst_host);
}