Lines Matching refs:xsum1
42 __m128 xsum1, xsum2; in xcorr_kernel() local
43 xsum1 = _mm_loadu_ps(sum); in xcorr_kernel()
52 xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_shuffle_ps(x0,x0,0x00),yj)); in xcorr_kernel()
55 xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_shuffle_ps(x0,x0,0xaa), in xcorr_kernel()
61 xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_load1_ps(x+j),_mm_loadu_ps(y+j))); in xcorr_kernel()
67 xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(_mm_load1_ps(x+j),_mm_loadu_ps(y+j))); in xcorr_kernel()
71 _mm_storeu_ps(sum,_mm_add_ps(xsum1,xsum2)); in xcorr_kernel()
79 __m128 xsum1, xsum2; in dual_inner_prod() local
80 xsum1 = _mm_setzero_ps(); in dual_inner_prod()
87 xsum1 = _mm_add_ps(xsum1,_mm_mul_ps(xi, y1i)); in dual_inner_prod()
91 xsum1 = _mm_add_ps(xsum1, _mm_movehl_ps(xsum1, xsum1)); in dual_inner_prod()
92 xsum1 = _mm_add_ss(xsum1, _mm_shuffle_ps(xsum1, xsum1, 0x55)); in dual_inner_prod()
93 _mm_store_ss(xy1, xsum1); in dual_inner_prod()