Chiebot-Mirror
/
opencv
8
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Merge pull request #13115 from catree:fix_dis_flow_intrin

Browse Source
pull/13134/head
Alexander Alekhin 7 years ago committed by GitHub
parent 5459c11e99 c96850ba59
commit 326be0d7ba
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
2 changed files with 13 additions and 12 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Unified View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 17
      modules/video/src/dis_flow.cpp
  2. 8
      modules/video/src/variational_refinement.cpp

17
modules/video/src/dis_flow.cpp
Unescape View File

@ -41,6 +41,7 @@
//M*/ //M*/
#include "precomp.hpp" #include "precomp.hpp"
#include "opencv2/core/hal/intrin.hpp"
#include "opencl_kernels_video.hpp" #include "opencl_kernels_video.hpp"
using namespace std; using namespace std;
@ -517,7 +518,7 @@ inline float processPatch(float &dst_dUx, float &dst_dUy, uchar *I0_ptr, uchar *
int I0_stride, int I1_stride, float w00, float w01, float w10, float w11, int patch_sz) int I0_stride, int I1_stride, float w00, float w01, float w10, float w11, int patch_sz)
{ {
float SSD = 0.0f; float SSD = 0.0f;
#ifdef CV_SIMD128 #if CV_SIMD128
if (patch_sz == 8) if (patch_sz == 8)
{ {
/* Variables to accumulate the sums */ /* Variables to accumulate the sums */
@ -569,7 +570,7 @@ inline float processPatch(float &dst_dUx, float &dst_dUy, uchar *I0_ptr, uchar *
dst_dUx += diff * I0x_ptr[i * I0_stride + j]; dst_dUx += diff * I0x_ptr[i * I0_stride + j];
dst_dUy += diff * I0y_ptr[i * I0_stride + j]; dst_dUy += diff * I0y_ptr[i * I0_stride + j];
} }
#ifdef CV_SIMD128 #if CV_SIMD128
} }
#endif #endif
return SSD; return SSD;
@ -586,7 +587,7 @@ inline float processPatchMeanNorm(float &dst_dUx, float &dst_dUy, uchar *I0_ptr,
float sum_I0x_mul = 0.0, sum_I0y_mul = 0.0; float sum_I0x_mul = 0.0, sum_I0y_mul = 0.0;
float n = (float)patch_sz * patch_sz; float n = (float)patch_sz * patch_sz;
#ifdef CV_SIMD128 #if CV_SIMD128
if (patch_sz == 8) if (patch_sz == 8)
{ {
/* Variables to accumulate the sums */ /* Variables to accumulate the sums */
@ -641,7 +642,7 @@ inline float processPatchMeanNorm(float &dst_dUx, float &dst_dUy, uchar *I0_ptr,
sum_I0x_mul += diff * I0x_ptr[i * I0_stride + j]; sum_I0x_mul += diff * I0x_ptr[i * I0_stride + j];
sum_I0y_mul += diff * I0y_ptr[i * I0_stride + j]; sum_I0y_mul += diff * I0y_ptr[i * I0_stride + j];
} }
#ifdef CV_SIMD128 #if CV_SIMD128
} }
#endif #endif
dst_dUx = sum_I0x_mul - sum_diff * x_grad_sum / n; dst_dUx = sum_I0x_mul - sum_diff * x_grad_sum / n;
@ -654,7 +655,7 @@ inline float computeSSD(uchar *I0_ptr, uchar *I1_ptr, int I0_stride, int I1_stri
float w11, int patch_sz) float w11, int patch_sz)
{ {
float SSD = 0.0f; float SSD = 0.0f;
#ifdef CV_SIMD128 #if CV_SIMD128
if (patch_sz == 8) if (patch_sz == 8)
{ {
v_float32x4 SSD_vec = v_setall_f32(0); v_float32x4 SSD_vec = v_setall_f32(0);
@ -679,7 +680,7 @@ inline float computeSSD(uchar *I0_ptr, uchar *I1_ptr, int I0_stride, int I1_stri
I0_ptr[i * I0_stride + j]; I0_ptr[i * I0_stride + j];
SSD += diff * diff; SSD += diff * diff;
} }
#ifdef CV_SIMD128 #if CV_SIMD128
} }
#endif #endif
return SSD; return SSD;
@ -691,7 +692,7 @@ inline float computeSSDMeanNorm(uchar *I0_ptr, uchar *I1_ptr, int I0_stride, int
{ {
float sum_diff = 0.0f, sum_diff_sq = 0.0f; float sum_diff = 0.0f, sum_diff_sq = 0.0f;
float n = (float)patch_sz * patch_sz; float n = (float)patch_sz * patch_sz;
#ifdef CV_SIMD128 #if CV_SIMD128
if (patch_sz == 8) if (patch_sz == 8)
{ {
v_float32x4 sum_diff_vec = v_setall_f32(0); v_float32x4 sum_diff_vec = v_setall_f32(0);
@ -721,7 +722,7 @@ inline float computeSSDMeanNorm(uchar *I0_ptr, uchar *I1_ptr, int I0_stride, int
sum_diff += diff; sum_diff += diff;
sum_diff_sq += diff * diff; sum_diff_sq += diff * diff;
} }
#ifdef CV_SIMD128 #if CV_SIMD128
} }
#endif #endif
return sum_diff_sq - sum_diff * sum_diff / n; return sum_diff_sq - sum_diff * sum_diff / n;

8
modules/video/src/variational_refinement.cpp
Unescape View File

@ -595,7 +595,7 @@ void VariationalRefinementImpl::ComputeDataTerm_ParBody::operator()(const Range
#undef INIT_ROW_POINTERS #undef INIT_ROW_POINTERS
int j = 0; int j = 0;
#ifdef CV_SIMD128 #if CV_SIMD128
v_float32x4 zeta_vec = v_setall_f32(zeta_squared); v_float32x4 zeta_vec = v_setall_f32(zeta_squared);
v_float32x4 eps_vec = v_setall_f32(epsilon_squared); v_float32x4 eps_vec = v_setall_f32(epsilon_squared);
v_float32x4 delta_vec = v_setall_f32(delta2); v_float32x4 delta_vec = v_setall_f32(delta2);
@ -801,7 +801,7 @@ void VariationalRefinementImpl::ComputeSmoothnessTermHorPass_ParBody::operator()
pA_v_next[j] += pWeight[j]; pA_v_next[j] += pWeight[j];
int j = 0; int j = 0;
#ifdef CV_SIMD128 #if CV_SIMD128
v_float32x4 alpha2_vec = v_setall_f32(alpha2); v_float32x4 alpha2_vec = v_setall_f32(alpha2);
v_float32x4 eps_vec = v_setall_f32(epsilon_squared); v_float32x4 eps_vec = v_setall_f32(epsilon_squared);
v_float32x4 cW_u_vec, cW_v_vec; v_float32x4 cW_u_vec, cW_v_vec;
@ -911,7 +911,7 @@ void VariationalRefinementImpl::ComputeSmoothnessTermVertPass_ParBody::operator(
#undef INIT_ROW_POINTERS #undef INIT_ROW_POINTERS
int j = 0; int j = 0;
#ifdef CV_SIMD128 #if CV_SIMD128
v_float32x4 pWeight_vec, uy_vec, vy_vec; v_float32x4 pWeight_vec, uy_vec, vy_vec;
for (; j < len - 3; j += 4) for (; j < len - 3; j += 4)
{ {
@ -1013,7 +1013,7 @@ void VariationalRefinementImpl::RedBlackSOR_ParBody::operator()(const Range &ran
#undef INIT_ROW_POINTERS #undef INIT_ROW_POINTERS
j = 0; j = 0;
#ifdef CV_SIMD128 #if CV_SIMD128
v_float32x4 pW_prev_vec = v_setall_f32(pW_next[-1]); v_float32x4 pW_prev_vec = v_setall_f32(pW_next[-1]);
v_float32x4 pdu_prev_vec = v_setall_f32(pdu_next[-1]); v_float32x4 pdu_prev_vec = v_setall_f32(pdu_next[-1]);
v_float32x4 pdv_prev_vec = v_setall_f32(pdv_next[-1]); v_float32x4 pdv_prev_vec = v_setall_f32(pdv_next[-1]);

Write Preview
Loading…
Cancel
Save