Vector Optimized Library of Kernels  3.1.2
Architecture-tuned implementations of math kernels
volk_16i_x5_add_quad_16i_x4.h
Go to the documentation of this file.
1 /* -*- c++ -*- */
2 /*
3  * Copyright 2012, 2014 Free Software Foundation, Inc.
4  *
5  * This file is part of VOLK
6  *
7  * SPDX-License-Identifier: LGPL-3.0-or-later
8  */
9 
47 #ifndef INCLUDED_volk_16i_x5_add_quad_16i_x4_a_H
48 #define INCLUDED_volk_16i_x5_add_quad_16i_x4_a_H
49 
50 #include <inttypes.h>
51 #include <stdio.h>
52 
53 #ifdef LV_HAVE_SSE2
54 #include <emmintrin.h>
55 #include <xmmintrin.h>
56 
57 static inline void volk_16i_x5_add_quad_16i_x4_a_sse2(short* target0,
58  short* target1,
59  short* target2,
60  short* target3,
61  short* src0,
62  short* src1,
63  short* src2,
64  short* src3,
65  short* src4,
66  unsigned int num_points)
67 {
68  const unsigned int num_bytes = num_points * 2;
69 
70  __m128i xmm0, xmm1, xmm2, xmm3, xmm4;
71  __m128i *p_target0, *p_target1, *p_target2, *p_target3, *p_src0, *p_src1, *p_src2,
72  *p_src3, *p_src4;
73  p_target0 = (__m128i*)target0;
74  p_target1 = (__m128i*)target1;
75  p_target2 = (__m128i*)target2;
76  p_target3 = (__m128i*)target3;
77 
78  p_src0 = (__m128i*)src0;
79  p_src1 = (__m128i*)src1;
80  p_src2 = (__m128i*)src2;
81  p_src3 = (__m128i*)src3;
82  p_src4 = (__m128i*)src4;
83 
84  int i = 0;
85 
86  int bound = (num_bytes >> 4);
87  int leftovers = (num_bytes >> 1) & 7;
88 
89  for (; i < bound; ++i) {
90  xmm0 = _mm_load_si128(p_src0);
91  xmm1 = _mm_load_si128(p_src1);
92  xmm2 = _mm_load_si128(p_src2);
93  xmm3 = _mm_load_si128(p_src3);
94  xmm4 = _mm_load_si128(p_src4);
95 
96  p_src0 += 1;
97  p_src1 += 1;
98 
99  xmm1 = _mm_add_epi16(xmm0, xmm1);
100  xmm2 = _mm_add_epi16(xmm0, xmm2);
101  xmm3 = _mm_add_epi16(xmm0, xmm3);
102  xmm4 = _mm_add_epi16(xmm0, xmm4);
103 
104 
105  p_src2 += 1;
106  p_src3 += 1;
107  p_src4 += 1;
108 
109  _mm_store_si128(p_target0, xmm1);
110  _mm_store_si128(p_target1, xmm2);
111  _mm_store_si128(p_target2, xmm3);
112  _mm_store_si128(p_target3, xmm4);
113 
114  p_target0 += 1;
115  p_target1 += 1;
116  p_target2 += 1;
117  p_target3 += 1;
118  }
119 
120  for (i = bound * 8; i < (bound * 8) + leftovers; ++i) {
121  target0[i] = src0[i] + src1[i];
122  target1[i] = src0[i] + src2[i];
123  target2[i] = src0[i] + src3[i];
124  target3[i] = src0[i] + src4[i];
125  }
126 }
127 #endif /*LV_HAVE_SSE2*/
128 
129 #ifdef LV_HAVE_NEON
130 #include <arm_neon.h>
131 
132 static inline void volk_16i_x5_add_quad_16i_x4_neon(short* target0,
133  short* target1,
134  short* target2,
135  short* target3,
136  short* src0,
137  short* src1,
138  short* src2,
139  short* src3,
140  short* src4,
141  unsigned int num_points)
142 {
143  const unsigned int eighth_points = num_points / 8;
144  unsigned int number = 0;
145 
146  int16x8_t src0_vec, src1_vec, src2_vec, src3_vec, src4_vec;
147  int16x8_t target0_vec, target1_vec, target2_vec, target3_vec;
148  for (number = 0; number < eighth_points; ++number) {
149  src0_vec = vld1q_s16(src0);
150  src1_vec = vld1q_s16(src1);
151  src2_vec = vld1q_s16(src2);
152  src3_vec = vld1q_s16(src3);
153  src4_vec = vld1q_s16(src4);
154 
155  target0_vec = vaddq_s16(src0_vec, src1_vec);
156  target1_vec = vaddq_s16(src0_vec, src2_vec);
157  target2_vec = vaddq_s16(src0_vec, src3_vec);
158  target3_vec = vaddq_s16(src0_vec, src4_vec);
159 
160  vst1q_s16(target0, target0_vec);
161  vst1q_s16(target1, target1_vec);
162  vst1q_s16(target2, target2_vec);
163  vst1q_s16(target3, target3_vec);
164  src0 += 8;
165  src1 += 8;
166  src2 += 8;
167  src3 += 8;
168  src4 += 8;
169  target0 += 8;
170  target1 += 8;
171  target2 += 8;
172  target3 += 8;
173  }
174 
175  for (number = eighth_points * 8; number < num_points; ++number) {
176  *target0++ = *src0 + *src1++;
177  *target1++ = *src0 + *src2++;
178  *target2++ = *src0 + *src3++;
179  *target3++ = *src0++ + *src4++;
180  }
181 }
182 
183 #endif /* LV_HAVE_NEON */
184 
185 #ifdef LV_HAVE_GENERIC
186 
187 static inline void volk_16i_x5_add_quad_16i_x4_generic(short* target0,
188  short* target1,
189  short* target2,
190  short* target3,
191  short* src0,
192  short* src1,
193  short* src2,
194  short* src3,
195  short* src4,
196  unsigned int num_points)
197 {
198  const unsigned int num_bytes = num_points * 2;
199 
200  int i = 0;
201 
202  int bound = num_bytes >> 1;
203 
204  for (i = 0; i < bound; ++i) {
205  target0[i] = src0[i] + src1[i];
206  target1[i] = src0[i] + src2[i];
207  target2[i] = src0[i] + src3[i];
208  target3[i] = src0[i] + src4[i];
209  }
210 }
211 
212 #endif /* LV_HAVE_GENERIC */
213 
214 #endif /*INCLUDED_volk_16i_x5_add_quad_16i_x4_a_H*/
for i
Definition: volk_config_fixed.tmpl.h:13
static void volk_16i_x5_add_quad_16i_x4_neon(short *target0, short *target1, short *target2, short *target3, short *src0, short *src1, short *src2, short *src3, short *src4, unsigned int num_points)
Definition: volk_16i_x5_add_quad_16i_x4.h:132
static void volk_16i_x5_add_quad_16i_x4_generic(short *target0, short *target1, short *target2, short *target3, short *src0, short *src1, short *src2, short *src3, short *src4, unsigned int num_points)
Definition: volk_16i_x5_add_quad_16i_x4.h:187
static void volk_16i_x5_add_quad_16i_x4_a_sse2(short *target0, short *target1, short *target2, short *target3, short *src0, short *src1, short *src2, short *src3, short *src4, unsigned int num_points)
Definition: volk_16i_x5_add_quad_16i_x4.h:57