325     double* vsum = _vsum + (m+1)*5;  in FarnebackUpdateFlow_Blur()  local
330         vsum[x] = srow0[x]*(m+2);  in FarnebackUpdateFlow_Blur()
336             vsum[x] += srow0[x];  in FarnebackUpdateFlow_Blur()
350             vsum[x] += srow1[x] - srow0[x];  in FarnebackUpdateFlow_Blur()
355             vsum[-1-x] = vsum[4-x];  in FarnebackUpdateFlow_Blur()
356             vsum[width*5+x] = vsum[width*5+x-5];  in FarnebackUpdateFlow_Blur()
360         g11 = vsum[0]*(m+2);  in FarnebackUpdateFlow_Blur()
361         g12 = vsum[1]*(m+2);  in FarnebackUpdateFlow_Blur()
362         g22 = vsum[2]*(m+2);  in FarnebackUpdateFlow_Blur()
363         h1 = vsum[3]*(m+2);  in FarnebackUpdateFlow_Blur()
364         h2 = vsum[4]*(m+2);  in FarnebackUpdateFlow_Blur()
368             g11 += vsum[x*5];  in FarnebackUpdateFlow_Blur()
369             g12 += vsum[x*5+1];  in FarnebackUpdateFlow_Blur()
370             g22 += vsum[x*5+2];  in FarnebackUpdateFlow_Blur()
371             h1 += vsum[x*5+3];  in FarnebackUpdateFlow_Blur()
372             h2 += vsum[x*5+4];  in FarnebackUpdateFlow_Blur()
378             g11 += vsum[(x+m)*5] - vsum[(x-m)*5 - 5];  in FarnebackUpdateFlow_Blur()
379             g12 += vsum[(x+m)*5 + 1] - vsum[(x-m)*5 - 4];  in FarnebackUpdateFlow_Blur()
380             g22 += vsum[(x+m)*5 + 2] - vsum[(x-m)*5 - 3];  in FarnebackUpdateFlow_Blur()
381             h1 += vsum[(x+m)*5 + 3] - vsum[(x-m)*5 - 2];  in FarnebackUpdateFlow_Blur()
382             h2 += vsum[(x+m)*5 + 4] - vsum[(x-m)*5 - 1];  in FarnebackUpdateFlow_Blur()
420     float *vsum = alignPtr((float*)_vsum + (m+1)*5, 16), *hsum = alignPtr((float*)_hsum, 16);  in FarnebackUpdateFlow_GaussianBlur()  local
488                 _mm_store_ps(vsum + x, s0);  in FarnebackUpdateFlow_GaussianBlur()
489                 _mm_store_ps(vsum + x + 4, s1);  in FarnebackUpdateFlow_GaussianBlur()
490                 _mm_store_ps(vsum + x + 8, s2);  in FarnebackUpdateFlow_GaussianBlur()
491                 _mm_store_ps(vsum + x + 12, s3);  in FarnebackUpdateFlow_GaussianBlur()
507                 _mm_store_ps(vsum + x, s0);  in FarnebackUpdateFlow_GaussianBlur()
516             vsum[x] = s0;  in FarnebackUpdateFlow_GaussianBlur()
522             vsum[-1-x] = vsum[4-x];  in FarnebackUpdateFlow_GaussianBlur()
523             vsum[width*5+x] = vsum[width*5+x-5];  in FarnebackUpdateFlow_GaussianBlur()
534                 __m128 s0 = _mm_mul_ps(_mm_loadu_ps(vsum + x), g4);  in FarnebackUpdateFlow_GaussianBlur()
535                 __m128 s1 = _mm_mul_ps(_mm_loadu_ps(vsum + x + 4), g4);  in FarnebackUpdateFlow_GaussianBlur()
540                     __m128 x0 = _mm_add_ps(_mm_loadu_ps(vsum + x - i*5),  in FarnebackUpdateFlow_GaussianBlur()
541                                            _mm_loadu_ps(vsum + x + i*5));  in FarnebackUpdateFlow_GaussianBlur()
542                     __m128 x1 = _mm_add_ps(_mm_loadu_ps(vsum + x - i*5 + 4),  in FarnebackUpdateFlow_GaussianBlur()
543                                            _mm_loadu_ps(vsum + x + i*5 + 4));  in FarnebackUpdateFlow_GaussianBlur()
555             float sum = vsum[x]*kernel[0];  in FarnebackUpdateFlow_GaussianBlur()
557                 sum += kernel[i]*(vsum[x - i*5] + vsum[x + i*5]);  in FarnebackUpdateFlow_GaussianBlur()