78 #ifndef INCLUDED_volk_32fc_x2_square_dist_32f_a_H
79 #define INCLUDED_volk_32fc_x2_square_dist_32f_a_H
91 unsigned int num_points)
93 const unsigned int num_bytes = num_points*8;
95 __m128 xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7;
99 int bound = num_bytes >> 5;
100 int leftovers0 = (num_bytes >> 4) & 1;
101 int leftovers1 = (num_bytes >> 3) & 1;
104 xmm1 = _mm_setzero_ps();
105 xmm1 = _mm_loadl_pi(xmm1, (__m64*)src0);
106 xmm2 = _mm_load_ps((
float*)&points[0]);
107 xmm1 = _mm_movelh_ps(xmm1, xmm1);
108 xmm3 = _mm_load_ps((
float*)&points[2]);
110 for(; i < bound - 1; ++i) {
111 xmm4 = _mm_sub_ps(xmm1, xmm2);
112 xmm5 = _mm_sub_ps(xmm1, xmm3);
114 xmm6 = _mm_mul_ps(xmm4, xmm4);
115 xmm7 = _mm_mul_ps(xmm5, xmm5);
117 xmm2 = _mm_load_ps((
float*)&points[0]);
119 xmm4 = _mm_hadd_ps(xmm6, xmm7);
121 xmm3 = _mm_load_ps((
float*)&points[2]);
123 _mm_store_ps(target, xmm4);
128 xmm4 = _mm_sub_ps(xmm1, xmm2);
129 xmm5 = _mm_sub_ps(xmm1, xmm3);
132 xmm6 = _mm_mul_ps(xmm4, xmm4);
133 xmm7 = _mm_mul_ps(xmm5, xmm5);
135 xmm4 = _mm_hadd_ps(xmm6, xmm7);
137 _mm_store_ps(target, xmm4);
141 for(i = 0; i < leftovers0; ++i) {
143 xmm2 = _mm_load_ps((
float*)&points[0]);
145 xmm4 = _mm_sub_ps(xmm1, xmm2);
149 xmm6 = _mm_mul_ps(xmm4, xmm4);
151 xmm4 = _mm_hadd_ps(xmm6, xmm6);
153 _mm_storeh_pi((__m64*)target, xmm4);
158 for(i = 0; i < leftovers1; ++i) {
160 diff = src0[0] - points[0];
172 #include <arm_neon.h>
174 volk_32fc_x2_square_dist_32f_neon(
float* target,
lv_32fc_t* src0,
lv_32fc_t* points,
unsigned int num_points)
176 const unsigned int quarter_points = num_points / 4;
179 float32x4x2_t a_vec, b_vec;
180 float32x4x2_t diff_vec;
181 float32x4_t tmp, tmp1, dist_sq;
182 a_vec.val[0] = vdupq_n_f32(
lv_creal(src0[0]) );
183 a_vec.val[1] = vdupq_n_f32(
lv_cimag(src0[0]) );
184 for(number=0; number < quarter_points; ++number) {
185 b_vec = vld2q_f32((
float*)points);
186 diff_vec.val[0] = vsubq_f32(a_vec.val[0], b_vec.val[0]);
187 diff_vec.val[1] = vsubq_f32(a_vec.val[1], b_vec.val[1]);
188 tmp = vmulq_f32(diff_vec.val[0], diff_vec.val[0]);
189 tmp1 = vmulq_f32(diff_vec.val[1], diff_vec.val[1]);
191 dist_sq = vaddq_f32(tmp, tmp1);
192 vst1q_f32(target, dist_sq);
196 for(number=quarter_points*4; number < num_points; ++number) {
204 #ifdef LV_HAVE_GENERIC
206 volk_32fc_x2_square_dist_32f_generic(
float* target,
lv_32fc_t* src0,
lv_32fc_t* points,
207 unsigned int num_points)
209 const unsigned int num_bytes = num_points*8;
215 for(; i < num_bytes >> 3; ++i) {
216 diff = src0[0] - points[i];
float complex lv_32fc_t
Definition: volk_complex.h:56
#define lv_creal(x)
Definition: volk_complex.h:76
#define lv_cimag(x)
Definition: volk_complex.h:78