Lines Matching refs:xsum2
46 __m128 xsum1, xsum2; in xcorr_kernel_sse() local
48 xsum2 = _mm_setzero_ps(); in xcorr_kernel_sse()
57 xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(_mm_shuffle_ps(x0,x0,0x55), in xcorr_kernel_sse()
61 xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(_mm_shuffle_ps(x0,x0,0xff),y3)); in xcorr_kernel_sse()
68 xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(_mm_load1_ps(x+j),_mm_loadu_ps(y+j))); in xcorr_kernel_sse()
75 _mm_storeu_ps(sum,_mm_add_ps(xsum1,xsum2)); in xcorr_kernel_sse()
83 __m128 xsum1, xsum2; in dual_inner_prod_sse() local
85 xsum2 = _mm_setzero_ps(); in dual_inner_prod_sse()
92 xsum2 = _mm_add_ps(xsum2,_mm_mul_ps(xi, y2i)); in dual_inner_prod_sse()
98 xsum2 = _mm_add_ps(xsum2, _mm_movehl_ps(xsum2, xsum2)); in dual_inner_prod_sse()
99 xsum2 = _mm_add_ss(xsum2, _mm_shuffle_ps(xsum2, xsum2, 0x55)); in dual_inner_prod_sse()
100 _mm_store_ss(xy2, xsum2); in dual_inner_prod_sse()