1
0
mirror of https://github.com/opencv/opencv.git synced 2026-07-29 07:13:02 +04:00

optimizations:

- new reduce implementation (with kepler optimizations)
- saturate_cast via asm command
- video SIMD instructions in element operations
- float arithmetics instead of double
- new deviceSupports function
This commit is contained in:
Vladislav Vinogradov
2013-01-23 14:43:36 +04:00
parent ae6266e101
commit 281d036fcf
45 changed files with 9379 additions and 8846 deletions
+1 -5
View File
@@ -86,8 +86,7 @@ namespace cv { namespace gpu { namespace device
void icvCalcOrientation_gpu(const float* featureX, const float* featureY, const float* featureSize, float* featureDir, int nFeatures);
void compute_descriptors_gpu(const PtrStepSzf& descriptors,
const float* featureX, const float* featureY, const float* featureSize, const float* featureDir, int nFeatures);
void compute_descriptors_gpu(PtrStepSz<float4> descriptors, const float* featureX, const float* featureY, const float* featureSize, const float* featureDir, int nFeatures);
}
}}}
@@ -122,9 +121,6 @@ namespace
CV_Assert(mask.empty() || (mask.size() == img.size() && mask.type() == CV_8UC1));
CV_Assert(surf_.nOctaves > 0 && surf_.nOctaveLayers > 0);
if (!TargetArchs::builtWith(GLOBAL_ATOMICS) || !DeviceInfo().supports(GLOBAL_ATOMICS))
CV_Error(CV_StsNotImplemented, "The device doesn't support global atomics");
const int min_size = calcSize(surf_.nOctaves - 1, 0);
CV_Assert(img_rows - min_size >= 0);
CV_Assert(img_cols - min_size >= 0);