27 #ifndef INCLUDED_volk_32f_sin_32f_a_H
28 #define INCLUDED_volk_32f_sin_32f_a_H
31 #include <smmintrin.h>
38 static inline void volk_32f_sin_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, condition2;
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++){
69 aVal = _mm_load_ps(aPtr);
70 s = _mm_sub_ps(aVal, _mm_and_ps(_mm_mul_ps(aVal, ftwos), _mm_cmplt_ps(aVal, fzeroes)));
71 q = _mm_cvtps_epi32(_mm_mul_ps(s, m4pi));
72 r = _mm_add_epi32(q, _mm_and_si128(q, ones));
74 s = _mm_sub_ps(s, _mm_mul_ps(_mm_cvtepi32_ps(r), pio4A));
75 s = _mm_sub_ps(s, _mm_mul_ps(_mm_cvtepi32_ps(r), pio4B));
77 s = _mm_div_ps(s, _mm_set1_ps(8.0));
80 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);
82 for(i = 0; i < 3; i++) {
83 s = _mm_mul_ps(s, _mm_sub_ps(ffours, s));
85 s = _mm_div_ps(s, ftwos);
87 sine = _mm_sqrt_ps(_mm_mul_ps(_mm_sub_ps(ftwos, s), s));
88 cosine = _mm_sub_ps(fones, s);
90 condition1 = _mm_cmpneq_ps(_mm_cvtepi32_ps(_mm_and_si128(_mm_add_epi32(q, ones), twos)), fzeroes);
91 condition2 = _mm_cmpneq_ps(_mm_cmpneq_ps(_mm_cvtepi32_ps(_mm_and_si128(q, fours)), fzeroes), _mm_cmplt_ps(aVal, fzeroes));
95 sine = _mm_add_ps(sine, _mm_and_ps(_mm_sub_ps(cosine, sine), condition1));
96 sine = _mm_sub_ps(sine, _mm_and_ps(_mm_mul_ps(sine, _mm_set1_ps(2.0f)), condition2));
97 _mm_store_ps(bPtr, sine);
102 number = quarterPoints * 4;
103 for(;number < num_points; number++){
104 *bPtr++ = sin(*aPtr++);
111 #ifndef INCLUDED_volk_32f_sin_32f_u_H
112 #define INCLUDED_volk_32f_sin_32f_u_H
114 #ifdef LV_HAVE_SSE4_1
115 #include <smmintrin.h>
122 static inline void volk_32f_sin_32f_u_sse4_1(
float* bVector,
const float* aVector,
unsigned int num_points){
124 float* bPtr = bVector;
125 const float* aPtr = aVector;
127 unsigned int number = 0;
128 unsigned int quarterPoints = num_points / 4;
131 __m128 aVal, s, m4pi, pio4A, pio4B, cp1, cp2, cp3, cp4, cp5, ffours, ftwos, fones, fzeroes;
132 __m128 sine, cosine, condition1, condition2;
133 __m128i q, r, ones, twos, fours;
135 m4pi = _mm_set1_ps(1.273239545);
136 pio4A = _mm_set1_ps(0.78515625);
137 pio4B = _mm_set1_ps(0.241876e-3);
138 ffours = _mm_set1_ps(4.0);
139 ftwos = _mm_set1_ps(2.0);
140 fones = _mm_set1_ps(1.0);
141 fzeroes = _mm_setzero_ps();
142 ones = _mm_set1_epi32(1);
143 twos = _mm_set1_epi32(2);
144 fours = _mm_set1_epi32(4);
146 cp1 = _mm_set1_ps(1.0);
147 cp2 = _mm_set1_ps(0.83333333e-1);
148 cp3 = _mm_set1_ps(0.2777778e-2);
149 cp4 = _mm_set1_ps(0.49603e-4);
150 cp5 = _mm_set1_ps(0.551e-6);
152 for(;number < quarterPoints; number++){
153 aVal = _mm_loadu_ps(aPtr);
154 s = _mm_sub_ps(aVal, _mm_and_ps(_mm_mul_ps(aVal, ftwos), _mm_cmplt_ps(aVal, fzeroes)));
155 q = _mm_cvtps_epi32(_mm_mul_ps(s, m4pi));
156 r = _mm_add_epi32(q, _mm_and_si128(q, ones));
158 s = _mm_sub_ps(s, _mm_mul_ps(_mm_cvtepi32_ps(r), pio4A));
159 s = _mm_sub_ps(s, _mm_mul_ps(_mm_cvtepi32_ps(r), pio4B));
161 s = _mm_div_ps(s, _mm_set1_ps(8.0));
162 s = _mm_mul_ps(s, s);
164 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);
166 for(i = 0; i < 3; i++) {
167 s = _mm_mul_ps(s, _mm_sub_ps(ffours, s));
169 s = _mm_div_ps(s, ftwos);
171 sine = _mm_sqrt_ps(_mm_mul_ps(_mm_sub_ps(ftwos, s), s));
172 cosine = _mm_sub_ps(fones, s);
174 condition1 = _mm_cmpneq_ps(_mm_cvtepi32_ps(_mm_and_si128(_mm_add_epi32(q, ones), twos)), fzeroes);
175 condition2 = _mm_cmpneq_ps(_mm_cmpneq_ps(_mm_cvtepi32_ps(_mm_and_si128(q, fours)), fzeroes), _mm_cmplt_ps(aVal, fzeroes));
177 sine = _mm_add_ps(sine, _mm_and_ps(_mm_sub_ps(cosine, sine), condition1));
178 sine = _mm_sub_ps(sine, _mm_and_ps(_mm_mul_ps(sine, _mm_set1_ps(2.0f)), condition2));
179 _mm_storeu_ps(bPtr, sine);
184 number = quarterPoints * 4;
185 for(;number < num_points; number++){
186 *bPtr++ = sin(*aPtr++);
192 #ifdef LV_HAVE_GENERIC
199 static inline void volk_32f_sin_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 = 0; number < num_points; number++){
205 *bPtr++ = sin(*aPtr++);