27 #ifndef INCLUDED_volk_32f_cos_32f_a_H
28 #define INCLUDED_volk_32f_cos_32f_a_H
31 #include <smmintrin.h>
38 static inline void volk_32f_cos_32f_a_sse4_1(
float* bVector,
const float* aVector,
unsigned int num_points){
40 float* bPtr = bVector;
41 const float* aPtr = aVector;
43 unsigned int number = 0;
44 unsigned int quarterPoints = num_points / 4;
47 __m128 aVal, s, m4pi, pio4A, pio4B, cp1, cp2, cp3, cp4, cp5, ffours, ftwos, fones, fzeroes;
48 __m128 sine, cosine, condition1, condition3;
49 __m128i q, r, ones, twos, fours;
51 m4pi = _mm_set1_ps(1.273239545);
52 pio4A = _mm_set1_ps(0.78515625);
53 pio4B = _mm_set1_ps(0.241876e-3);
54 ffours = _mm_set1_ps(4.0);
55 ftwos = _mm_set1_ps(2.0);
56 fones = _mm_set1_ps(1.0);
57 fzeroes = _mm_setzero_ps();
58 ones = _mm_set1_epi32(1);
59 twos = _mm_set1_epi32(2);
60 fours = _mm_set1_epi32(4);
62 cp1 = _mm_set1_ps(1.0);
63 cp2 = _mm_set1_ps(0.83333333e-1);
64 cp3 = _mm_set1_ps(0.2777778e-2);
65 cp4 = _mm_set1_ps(0.49603e-4);
66 cp5 = _mm_set1_ps(0.551e-6);
68 for(;number < quarterPoints; number++){
70 aVal = _mm_load_ps(aPtr);
71 s = _mm_sub_ps(aVal, _mm_and_ps(_mm_mul_ps(aVal, ftwos), _mm_cmplt_ps(aVal, fzeroes)));
72 q = _mm_cvtps_epi32(_mm_mul_ps(s, m4pi));
73 r = _mm_add_epi32(q, _mm_and_si128(q, ones));
75 s = _mm_sub_ps(s, _mm_mul_ps(_mm_cvtepi32_ps(r), pio4A));
76 s = _mm_sub_ps(s, _mm_mul_ps(_mm_cvtepi32_ps(r), pio4B));
78 s = _mm_div_ps(s, _mm_set1_ps(8.0));
81 s = _mm_mul_ps(_mm_add_ps(_mm_mul_ps(_mm_sub_ps(_mm_mul_ps(_mm_add_ps(_mm_mul_ps(_mm_sub_ps(_mm_mul_ps(s, cp5), cp4), s), cp3), s), cp2), s), cp1), s);
83 for(i = 0; i < 3; i++) s = _mm_mul_ps(s, _mm_sub_ps(ffours, s));
84 s = _mm_div_ps(s, ftwos);
86 sine = _mm_sqrt_ps(_mm_mul_ps(_mm_sub_ps(ftwos, s), s));
87 cosine = _mm_sub_ps(fones, s);
89 condition1 = _mm_cmpneq_ps(_mm_cvtepi32_ps(_mm_and_si128(_mm_add_epi32(q, ones), twos)), fzeroes);
91 condition3 = _mm_cmpneq_ps(_mm_cvtepi32_ps(_mm_and_si128(_mm_add_epi32(q, twos), fours)), fzeroes);
93 cosine = _mm_add_ps(cosine, _mm_and_ps(_mm_sub_ps(sine, cosine), condition1));
94 cosine = _mm_sub_ps(cosine, _mm_and_ps(_mm_mul_ps(cosine, _mm_set1_ps(2.0f)), condition3));
95 _mm_store_ps(bPtr, cosine);
100 number = quarterPoints * 4;
101 for(;number < num_points; number++){
102 *bPtr++ = cos(*aPtr++);
110 #ifndef INCLUDED_volk_32f_cos_32f_u_H
111 #define INCLUDED_volk_32f_cos_32f_u_H
113 #ifdef LV_HAVE_SSE4_1
114 #include <smmintrin.h>
121 static inline void volk_32f_cos_32f_u_sse4_1(
float* bVector,
const float* aVector,
unsigned int num_points){
123 float* bPtr = bVector;
124 const float* aPtr = aVector;
126 unsigned int number = 0;
127 unsigned int quarterPoints = num_points / 4;
130 __m128 aVal, s, m4pi, pio4A, pio4B, cp1, cp2, cp3, cp4, cp5, ffours, ftwos, fones, fzeroes;
131 __m128 sine, cosine, condition1, condition3;
132 __m128i q, r, ones, twos, fours;
134 m4pi = _mm_set1_ps(1.273239545);
135 pio4A = _mm_set1_ps(0.78515625);
136 pio4B = _mm_set1_ps(0.241876e-3);
137 ffours = _mm_set1_ps(4.0);
138 ftwos = _mm_set1_ps(2.0);
139 fones = _mm_set1_ps(1.0);
140 fzeroes = _mm_setzero_ps();
141 ones = _mm_set1_epi32(1);
142 twos = _mm_set1_epi32(2);
143 fours = _mm_set1_epi32(4);
145 cp1 = _mm_set1_ps(1.0);
146 cp2 = _mm_set1_ps(0.83333333e-1);
147 cp3 = _mm_set1_ps(0.2777778e-2);
148 cp4 = _mm_set1_ps(0.49603e-4);
149 cp5 = _mm_set1_ps(0.551e-6);
151 for(;number < quarterPoints; number++){
152 aVal = _mm_loadu_ps(aPtr);
153 s = _mm_sub_ps(aVal, _mm_and_ps(_mm_mul_ps(aVal, ftwos), _mm_cmplt_ps(aVal, fzeroes)));
154 q = _mm_cvtps_epi32(_mm_mul_ps(s, m4pi));
155 r = _mm_add_epi32(q, _mm_and_si128(q, ones));
157 s = _mm_sub_ps(s, _mm_mul_ps(_mm_cvtepi32_ps(r), pio4A));
158 s = _mm_sub_ps(s, _mm_mul_ps(_mm_cvtepi32_ps(r), pio4B));
160 s = _mm_div_ps(s, _mm_set1_ps(8.0));
161 s = _mm_mul_ps(s, s);
163 s = _mm_mul_ps(_mm_add_ps(_mm_mul_ps(_mm_sub_ps(_mm_mul_ps(_mm_add_ps(_mm_mul_ps(_mm_sub_ps(_mm_mul_ps(s, cp5), cp4), s), cp3), s), cp2), s), cp1), s);
165 for(i = 0; i < 3; i++){
166 s = _mm_mul_ps(s, _mm_sub_ps(ffours, s));
168 s = _mm_div_ps(s, ftwos);
170 sine = _mm_sqrt_ps(_mm_mul_ps(_mm_sub_ps(ftwos, s), s));
171 cosine = _mm_sub_ps(fones, s);
173 condition1 = _mm_cmpneq_ps(_mm_cvtepi32_ps(_mm_and_si128(_mm_add_epi32(q, ones), twos)), fzeroes);
175 condition3 = _mm_cmpneq_ps(_mm_cvtepi32_ps(_mm_and_si128(_mm_add_epi32(q, twos), fours)), fzeroes);
177 cosine = _mm_add_ps(cosine, _mm_and_ps(_mm_sub_ps(sine, cosine), condition1));
178 cosine = _mm_sub_ps(cosine, _mm_and_ps(_mm_mul_ps(cosine, _mm_set1_ps(2.0f)), condition3));
179 _mm_storeu_ps(bPtr, cosine);
184 number = quarterPoints * 4;
185 for(;number < num_points; number++){
186 *bPtr++ = cos(*aPtr++);
192 #ifdef LV_HAVE_GENERIC
199 static inline void volk_32f_cos_32f_generic(
float* bVector,
const float* aVector,
unsigned int num_points){
200 float* bPtr = bVector;
201 const float* aPtr = aVector;
202 unsigned int number = 0;
204 for(; number < num_points; number++){
205 *bPtr++ = cos(*aPtr++);