65 #ifndef INCLUDED_volk_32f_invsqrt_32f_a_H
66 #define INCLUDED_volk_32f_invsqrt_32f_a_H
76 const float threehalfs = 1.5F;
84 u.i = 0x5f3759df - (u.i >> 1);
85 u.f = u.f * (threehalfs - (x2 * u.f * u.f));
93 #include <immintrin.h>
98 unsigned int number = 0;
99 const unsigned int eighthPoints = num_points / 8;
101 float* cPtr = cVector;
102 const float* aPtr = aVector;
104 for (; number < eighthPoints; number++) {
105 aVal = _mm256_load_ps(aPtr);
106 cVal = _mm256_rsqrt_ps(aVal);
107 _mm256_store_ps(cPtr, cVal);
112 number = eighthPoints * 8;
113 for (; number < num_points; number++)
120 #include <xmmintrin.h>
125 unsigned int number = 0;
126 const unsigned int quarterPoints = num_points / 4;
128 float* cPtr = cVector;
129 const float* aPtr = aVector;
132 for (; number < quarterPoints; number++) {
134 aVal = _mm_load_ps(aPtr);
136 cVal = _mm_rsqrt_ps(aVal);
138 _mm_store_ps(cPtr, cVal);
144 number = quarterPoints * 4;
145 for (; number < num_points; number++) {
153 #include <arm_neon.h>
159 const unsigned int quarter_points = num_points / 4;
161 float* cPtr = cVector;
162 const float* aPtr = aVector;
163 float32x4_t a_val, c_val;
164 for (number = 0; number < quarter_points; ++number) {
165 a_val = vld1q_f32(aPtr);
166 c_val = vrsqrteq_f32(a_val);
167 vst1q_f32(cPtr, c_val);
172 for (number = quarter_points * 4; number < num_points; number++)
178 #ifdef LV_HAVE_GENERIC
181 const float* aVector,
182 unsigned int num_points)
184 float* cPtr = cVector;
185 const float* aPtr = aVector;
186 unsigned int number = 0;
187 for (number = 0; number < num_points; number++) {
194 #include <immintrin.h>
199 unsigned int number = 0;
200 const unsigned int eighthPoints = num_points / 8;
202 float* cPtr = cVector;
203 const float* aPtr = aVector;
205 for (; number < eighthPoints; number++) {
206 aVal = _mm256_loadu_ps(aPtr);
207 cVal = _mm256_rsqrt_ps(aVal);
208 _mm256_storeu_ps(cPtr, cVal);
213 number = eighthPoints * 8;
214 for (; number < num_points; number++)