diff options
| -rw-r--r-- | volk/include/volk/volk_32f_x2_dot_prod_32f_u.h | 51 | ||||
| -rw-r--r-- | volk/lib/testqa.cc | 2 |
2 files changed, 25 insertions, 28 deletions
diff --git a/volk/include/volk/volk_32f_x2_dot_prod_32f_u.h b/volk/include/volk/volk_32f_x2_dot_prod_32f_u.h index ab33a25876..b24e8b1f79 100644 --- a/volk/include/volk/volk_32f_x2_dot_prod_32f_u.h +++ b/volk/include/volk/volk_32f_x2_dot_prod_32f_u.h @@ -48,14 +48,14 @@ static inline void volk_32f_x2_dot_prod_32f_u_sse( float* result, const float* for(;number < sixteenthPoints; number++){ - a0Val = _mm_load_ps(aPtr); - a1Val = _mm_load_ps(aPtr+4); - a2Val = _mm_load_ps(aPtr+8); - a3Val = _mm_load_ps(aPtr+12); - b0Val = _mm_load_ps(bPtr); - b1Val = _mm_load_ps(bPtr+4); - b2Val = _mm_load_ps(bPtr+8); - b3Val = _mm_load_ps(bPtr+12); + a0Val = _mm_loadu_ps(aPtr); + a1Val = _mm_loadu_ps(aPtr+4); + a2Val = _mm_loadu_ps(aPtr+8); + a3Val = _mm_loadu_ps(aPtr+12); + b0Val = _mm_loadu_ps(bPtr); + b1Val = _mm_loadu_ps(bPtr+4); + b2Val = _mm_loadu_ps(bPtr+8); + b3Val = _mm_loadu_ps(bPtr+12); c0Val = _mm_mul_ps(a0Val, b0Val); c1Val = _mm_mul_ps(a1Val, b1Val); @@ -87,9 +87,6 @@ static inline void volk_32f_x2_dot_prod_32f_u_sse( float* result, const float* number = sixteenthPoints*16; for(;number < num_points; number++){ dotProduct += ((*aPtr++) * (*bPtr++)); - dotProduct += ((*aPtr++) * (*bPtr++)); - dotProduct += ((*aPtr++) * (*bPtr++)); - dotProduct += ((*aPtr++) * (*bPtr++)); } *result = dotProduct; @@ -121,14 +118,14 @@ static inline void volk_32f_x2_dot_prod_32f_u_sse3(float * result, const float * for(;number < sixteenthPoints; number++){ - a0Val = _mm_load_ps(aPtr); - a1Val = _mm_load_ps(aPtr+4); - a2Val = _mm_load_ps(aPtr+8); - a3Val = _mm_load_ps(aPtr+12); - b0Val = _mm_load_ps(bPtr); - b1Val = _mm_load_ps(bPtr+4); - b2Val = _mm_load_ps(bPtr+8); - b3Val = _mm_load_ps(bPtr+12); + a0Val = _mm_loadu_ps(aPtr); + a1Val = _mm_loadu_ps(aPtr+4); + a2Val = _mm_loadu_ps(aPtr+8); + a3Val = _mm_loadu_ps(aPtr+12); + b0Val = _mm_loadu_ps(bPtr); + b1Val = _mm_loadu_ps(bPtr+4); + b2Val = _mm_loadu_ps(bPtr+8); + b3Val = _mm_loadu_ps(bPtr+12); c0Val = _mm_mul_ps(a0Val, b0Val); c1Val = _mm_mul_ps(a1Val, b1Val); @@ -187,15 +184,15 @@ static inline void volk_32f_x2_dot_prod_32f_u_sse4_1(float * result, const float for(;number < sixteenthPoints; number++){ - aVal1 = _mm_load_ps(aPtr); aPtr += 4; - aVal2 = _mm_load_ps(aPtr); aPtr += 4; - aVal3 = _mm_load_ps(aPtr); aPtr += 4; - aVal4 = _mm_load_ps(aPtr); aPtr += 4; + aVal1 = _mm_loadu_ps(aPtr); aPtr += 4; + aVal2 = _mm_loadu_ps(aPtr); aPtr += 4; + aVal3 = _mm_loadu_ps(aPtr); aPtr += 4; + aVal4 = _mm_loadu_ps(aPtr); aPtr += 4; - bVal1 = _mm_load_ps(bPtr); bPtr += 4; - bVal2 = _mm_load_ps(bPtr); bPtr += 4; - bVal3 = _mm_load_ps(bPtr); bPtr += 4; - bVal4 = _mm_load_ps(bPtr); bPtr += 4; + bVal1 = _mm_loadu_ps(bPtr); bPtr += 4; + bVal2 = _mm_loadu_ps(bPtr); bPtr += 4; + bVal3 = _mm_loadu_ps(bPtr); bPtr += 4; + bVal4 = _mm_loadu_ps(bPtr); bPtr += 4; cVal1 = _mm_dp_ps(aVal1, bVal1, 0xF1); cVal2 = _mm_dp_ps(aVal2, bVal2, 0xF2); diff --git a/volk/lib/testqa.cc b/volk/lib/testqa.cc index d1eb1cacbb..507c787772 100644 --- a/volk/lib/testqa.cc +++ b/volk/lib/testqa.cc @@ -54,7 +54,7 @@ VOLK_RUN_TESTS(volk_32fc_x2_square_dist_32f_a, 1e-4, 0, 20460, 1); VOLK_RUN_TESTS(volk_32fc_x2_s32f_square_dist_scalar_mult_32f_a, 1e-4, 10, 20460, 1); VOLK_RUN_TESTS(volk_32f_x2_divide_32f_a, 1e-4, 0, 20460, 1); VOLK_RUN_TESTS(volk_32f_x2_dot_prod_32f_a, 1e-4, 0, 204600, 1); -//VOLK_RUN_TESTS(volk_32f_x2_dot_prod_32f_u, 1e-4, 0, 204600, 1); +VOLK_RUN_TESTS(volk_32f_x2_dot_prod_32f_u, 1e-4, 0, 204600, 1); VOLK_RUN_TESTS(volk_32f_x2_dot_prod_16i_a, 1e-4, 0, 204600, 1); //VOLK_RUN_TESTS(volk_32f_s32f_32f_fm_detect_32f_a, 1e-4, 2046, 10000); VOLK_RUN_TESTS(volk_32f_index_max_16u_a, 3, 0, 20460, 1); |