mirror/opencv
1
0
Fork 0
mirror of https://github.com/opencv/opencv.git synced 2024-11-29 13:47:32 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

fixed gpu::sum* on CC1.0, updated some tests

Browse Source
This commit is contained in:
Alexey Spizhevoy 2011-02-02 07:23:55 +00:00
parent f7e62d89f8
commit 557dd39f03
6 changed files with 27 additions and 11 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
modules/gpu/include/opencv2/gpu/gpu.hpp
View File

@ -66,8 +66,14 @@ namespace cv
enum GpuFeature
{
NATIVE_DOUBLE,
ATOMICS
COMPUTE_10 = 10,
COMPUTE_11 = 11,
COMPUTE_12 = 12,
COMPUTE_13 = 13,
COMPUTE_20 = 20,
COMPUTE_21 = 21,
ATOMICS = COMPUTE_11,
NATIVE_DOUBLE = COMPUTE_13
};
class CV_EXPORTS TargetArchs

6
modules/gpu/src/cuda/matrix_reductions.cu
View File

@ -1394,7 +1394,7 @@ namespace cv { namespace gpu { namespace mathfunc
smem[tid] = res.x;
smem[tid + nthreads] = res.y;
smem[tid + 2 * nthreads] = res.z;
smem[tid + 3 * nthreads] = res.z;
smem[tid + 3 * nthreads] = res.w;
__syncthreads();
sumInSmem<nthreads, R>(smem, tid);
@ -1432,21 +1432,25 @@ namespace cv { namespace gpu { namespace mathfunc
src, (typename TypeVec<R, 1>::vec_t*)buf.ptr(0));
sumPass2Kernel<T, R, threads_x * threads_y><<<1, threads_x * threads_y>>>(
(typename TypeVec<R, 1>::vec_t*)buf.ptr(0), grid.x * grid.y);
break;
case 2:
sumKernel_C2<T, R, IdentityOp<R>, threads_x * threads_y><<<grid, threads>>>(
src, (typename TypeVec<R, 2>::vec_t*)buf.ptr(0));
sumPass2Kernel_C2<T, R, threads_x * threads_y><<<1, threads_x * threads_y>>>(
(typename TypeVec<R, 2>::vec_t*)buf.ptr(0), grid.x * grid.y);
break;
case 3:
sumKernel_C3<T, R, IdentityOp<R>, threads_x * threads_y><<<grid, threads>>>(
src, (typename TypeVec<R, 3>::vec_t*)buf.ptr(0));
sumPass2Kernel_C3<T, R, threads_x * threads_y><<<1, threads_x * threads_y>>>(
(typename TypeVec<R, 3>::vec_t*)buf.ptr(0), grid.x * grid.y);
break;
case 4:
sumKernel_C4<T, R, IdentityOp<R>, threads_x * threads_y><<<grid, threads>>>(
src, (typename TypeVec<R, 4>::vec_t*)buf.ptr(0));
sumPass2Kernel_C4<T, R, threads_x * threads_y><<<1, threads_x * threads_y>>>(
(typename TypeVec<R, 4>::vec_t*)buf.ptr(0), grid.x * grid.y);
break;
}
cudaSafeCall(cudaThreadSynchronize());

6
modules/gpu/src/initialization.cpp
View File

@ -71,11 +71,7 @@ namespace
CV_EXPORTS bool cv::gpu::TargetArchs::builtWith(cv::gpu::GpuFeature feature)
{
if (feature == NATIVE_DOUBLE)
return ::compareToSet(CUDA_ARCH_FEATURES, 13, std::greater_equal<int>());
if (feature == ATOMICS)
return ::compareToSet(CUDA_ARCH_FEATURES, 11, std::greater_equal<int>());
return true;
return ::compareToSet(CUDA_ARCH_FEATURES, feature, std::greater_equal<int>());
}

10
tests/gpu/src/arithm.cpp
View File

@ -947,6 +947,16 @@ struct CV_GpuSumTest: CvTest
// sum
//
gen(1 + rand() % 500, 1 + rand() % 500, CV_MAKETYPE(type, 1), src);
a = sum(src);
b = sum(GpuMat(src));
if (abs(a[0] - b[0]) > src.size().area() * max_err)
{
ts->printf(CvTS::CONSOLE, "1 cols: %d, rows: %d, expected: %f, actual: %f\n", src.cols, src.rows, a[0], b[0]);
ts->set_failed_test_info(CvTS::FAIL_INVALID_OUTPUT);
return;
}
gen(1 + rand() % 500, 1 + rand() % 500, CV_MAKETYPE(type, 2), src);
a = sum(src);
b = sum(GpuMat(src));

4
tests/gpu/src/meanshift.cpp
View File

@ -56,7 +56,7 @@ struct CV_GpuMeanShiftTest : public CvTest
cv::Mat img = cv::imread(std::string(ts->get_data_path()) + "meanshift/cones.png");
cv::Mat img_template;
if (cv::gpu::TargetArchs::hasEqualOrGreater(2, 0) && cv::gpu::DeviceInfo().major() >= 2)
if (cv::gpu::TargetArchs::builtWith(cv::gpu::COMPUTE_20) && cv::gpu::DeviceInfo().major() >= 2)
img_template = cv::imread(std::string(ts->get_data_path()) + "meanshift/con_result.png");
else
img_template = cv::imread(std::string(ts->get_data_path()) + "meanshift/con_result_CC1X.png");
@ -199,7 +199,7 @@ struct CV_GpuMeanShiftProcTest : public CvTest
cv::Mat spmap_template;
cv::FileStorage fs;
if (cv::gpu::TargetArchs::hasEqualOrGreater(2, 0) && cv::gpu::DeviceInfo().major() >= 2)
if (cv::gpu::TargetArchs::builtWith(cv::gpu::COMPUTE_20) && cv::gpu::DeviceInfo().major() >= 2)
fs.open(std::string(ts->get_data_path()) + "meanshift/spmap.yaml", cv::FileStorage::READ);
else
fs.open(std::string(ts->get_data_path()) + "meanshift/spmap_CC1X.yaml", cv::FileStorage::READ);

2
tests/gpu/src/mssegmentation.cpp
View File

@ -69,7 +69,7 @@ struct CV_GpuMeanShiftSegmentationTest : public CvTest {
{
stringstream path;
path << ts->get_data_path() << "meanshift/cones_segmented_sp10_sr10_minsize" << minsize;
if (TargetArchs::hasEqualOrGreater(2, 0) && DeviceInfo().major() >= 2)
if (TargetArchs::builtWith(COMPUTE_20) && DeviceInfo().major() >= 2)
path << ".png";
else
path << "_CC1X.png";