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

added GPU_PERF_TESTP_P but haven't switched test to it (only perf4au app)

Browse Source
This commit is contained in:
Anatoly Baksheev
2012-12-25 19:53:58 +04:00
parent 3882bbbf35
commit be22891f71
3 changed files with 50 additions and 36 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
modules/gpu/src/cuda/optflowbm.cu
View File

@@ -210,7 +210,7 @@ namespace optflowbm_fast
{
}
__device__ void initSums_BruteForce(int i, int j, int* dist_sums, PtrStepi& col_sums, PtrStepi& up_col_sums) const
__device__ __forceinline__ void initSums_BruteForce(int i, int j, int* dist_sums, PtrStepi& col_sums, PtrStepi& up_col_sums) const
{
for (int index = threadIdx.x; index < search_window * search_window; index += STRIDE)
{
@@ -246,7 +246,7 @@ namespace optflowbm_fast
}
}
__device__ void shiftRight_FirstRow(int i, int j, int first, int* dist_sums, PtrStepi& col_sums, PtrStepi& up_col_sums) const
__device__ __forceinline__ void shiftRight_FirstRow(int i, int j, int first, int* dist_sums, PtrStepi& col_sums, PtrStepi& up_col_sums) const
{
for (int index = threadIdx.x; index < search_window * search_window; index += STRIDE)
{
@@ -271,7 +271,7 @@ namespace optflowbm_fast
}
}
__device__ void shiftRight_UpSums(int i, int j, int first, int* dist_sums, PtrStepi& col_sums, PtrStepi& up_col_sums) const
__device__ __forceinline__ void shiftRight_UpSums(int i, int j, int first, int* dist_sums, PtrStepi& col_sums, PtrStepi& up_col_sums) const
{
int ay = i;
int ax = j + block_radius;
@@ -298,7 +298,7 @@ namespace optflowbm_fast
}
}
__device__ void convolve_window(int i, int j, const int* dist_sums, float& velx, float& vely) const
__device__ __forceinline__ void convolve_window(int i, int j, const int* dist_sums, float& velx, float& vely) const
{
int bestDist = numeric_limits<int>::max();
int bestInd = -1;
@@ -328,7 +328,7 @@ namespace optflowbm_fast
}
}
__device__ void operator()(PtrStepf velx, PtrStepf vely) const
__device__ __forceinline__ void operator()(PtrStepf velx, PtrStepf vely) const
{
int tbx = blockIdx.x * TILE_COLS;
int tby = blockIdx.y * TILE_ROWS;