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

use perf test replace performance sample

Browse Source
This commit is contained in:
yao
2013-03-27 12:04:48 +08:00
parent 55c9a7c87d
commit 5539e85a11
26 changed files with 3791 additions and 14305 deletions
Download Patch File Download Diff File Expand all files Collapse all files

105
modules/ocl/perf/perf_gemm.cpp
View File

@@ -16,6 +16,7 @@
//
// @Authors
// Fangfang Bai, fangfang@multicorewareinc.com
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
@@ -41,73 +42,47 @@
// the use of this software, even if advised of the possibility of such damage.
//
//M*/
#include "precomp.hpp"
using namespace std;
#ifdef HAVE_CLAMDBLAS
////////////////////////////////////////////////////////////////////////////
// GEMM
PARAM_TEST_CASE(Gemm, int, cv::Size, int)
///////////// gemm ////////////////////////
TEST(gemm)
{
int type;
cv::Size mat_size;
int flags;
vector<cv::ocl::Info> info;
virtual void SetUp()
Mat src1, src2, src3, dst;
ocl::oclMat d_src1, d_src2, d_src3, d_dst;
for (int size = Min_Size; size <= Max_Size; size *= Multiple)
{
type = GET_PARAM(0);
mat_size = GET_PARAM(1);
flags = GET_PARAM(2);
SUBTEST << size << 'x' << size;
cv::ocl::getDevice(info);
gen(src1, size, size, CV_32FC1, Scalar::all(-10), Scalar::all(10));
gen(src2, size, size, CV_32FC1, Scalar::all(-10), Scalar::all(10));
gen(src3, size, size, CV_32FC1, Scalar::all(-10), Scalar::all(10));
gemm(src1, src2, 1.0, src3, 1.0, dst);
CPU_ON;
gemm(src1, src2, 1.0, src3, 1.0, dst);
CPU_OFF;
d_src1.upload(src1);
d_src2.upload(src2);
d_src3.upload(src3);
WARMUP_ON;
ocl::gemm(d_src1, d_src2, 1.0, d_src3, 1.0, d_dst);
WARMUP_OFF;
GPU_ON;
ocl::gemm(d_src1, d_src2, 1.0, d_src3, 1.0, d_dst);
;
GPU_OFF;
GPU_FULL_ON;
d_src1.upload(src1);
d_src2.upload(src2);
d_src3.upload(src3);
ocl::gemm(d_src1, d_src2, 1.0, d_src3, 1.0, d_dst);
d_dst.download(dst);
GPU_FULL_OFF;
}
};
TEST_P(Gemm, Performance)
{
cv::Mat a = randomMat(mat_size, type, 0.0, 10.0);
cv::Mat b = randomMat(mat_size, type, 0.0, 10.0);
cv::Mat c = randomMat(mat_size, type, 0.0, 10.0);
cv::ocl::oclMat ocl_dst;
double totalgputick = 0;
double totalgputick_kernel = 0;
double t1 = 0;
double t2 = 0;
for(int j = 0; j < LOOP_TIMES + 1; j ++)
{
t1 = (double)cvGetTickCount();//gpu start1
cv::ocl::oclMat ga = cv::ocl::oclMat(a);//upload
cv::ocl::oclMat gb = cv::ocl::oclMat(b);//upload
cv::ocl::oclMat gc = cv::ocl::oclMat(c);//upload
t2 = (double)cvGetTickCount(); //kernel
cv::ocl::gemm(ga, gb, 1.0, gc, 1.0, ocl_dst, flags);
t2 = (double)cvGetTickCount() - t2;//kernel
cv::Mat cpu_dst;
ocl_dst.download (cpu_dst);//download
t1 = (double)cvGetTickCount() - t1;//gpu end
if(j == 0)
continue;
totalgputick = t1 + totalgputick;
totalgputick_kernel = t2 + totalgputick_kernel;
}
cout << "average gpu runtime is " << totalgputick / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
cout << "average gpu runtime without data transfer is " << totalgputick_kernel / ((double)cvGetTickFrequency()* LOOP_TIMES * 1000.) << "ms" << endl;
}
INSTANTIATE_TEST_CASE_P(ocl_gemm, Gemm, testing::Combine(
testing::Values(CV_32FC1, CV_32FC2/* , CV_64FC1, CV_64FC2*/),
testing::Values(cv::Size(512, 512), cv::Size(1024, 1024)),
testing::Values(0, (int)cv::GEMM_1_T, (int)cv::GEMM_2_T, (int)(cv::GEMM_1_T + cv::GEMM_2_T))));
#endif
}