Vector Optimized Library of Kernels  2.4
Architecture-tuned implementations of math kernels
volk_16i_permute_and_scalar_add.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 GNU Radio
6  *
7  * GNU Radio is free software; you can redistribute it and/or modify
8  * it under the terms of the GNU General Public License as published by
9  * the Free Software Foundation; either version 3, or (at your option)
10  * any later version.
11  *
12  * GNU Radio is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
15  * GNU General Public License for more details.
16  *
17  * You should have received a copy of the GNU General Public License
18  * along with GNU Radio; see the file COPYING. If not, write to
19  * the Free Software Foundation, Inc., 51 Franklin Street,
20  * Boston, MA 02110-1301, USA.
21  */
22 
59 #ifndef INCLUDED_volk_16i_permute_and_scalar_add_a_H
60 #define INCLUDED_volk_16i_permute_and_scalar_add_a_H
61 
62 #include <inttypes.h>
63 #include <stdio.h>
64 
65 #ifdef LV_HAVE_SSE2
66 
67 #include <emmintrin.h>
68 #include <xmmintrin.h>
69 
70 static inline void volk_16i_permute_and_scalar_add_a_sse2(short* target,
71  short* src0,
72  short* permute_indexes,
73  short* cntl0,
74  short* cntl1,
75  short* cntl2,
76  short* cntl3,
77  short* scalars,
78  unsigned int num_points)
79 {
80 
81  const unsigned int num_bytes = num_points * 2;
82 
83  __m128i xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7;
84 
85  __m128i *p_target, *p_cntl0, *p_cntl1, *p_cntl2, *p_cntl3, *p_scalars;
86 
87  short* p_permute_indexes = permute_indexes;
88 
89  p_target = (__m128i*)target;
90  p_cntl0 = (__m128i*)cntl0;
91  p_cntl1 = (__m128i*)cntl1;
92  p_cntl2 = (__m128i*)cntl2;
93  p_cntl3 = (__m128i*)cntl3;
94  p_scalars = (__m128i*)scalars;
95 
96  int i = 0;
97 
98  int bound = (num_bytes >> 4);
99  int leftovers = (num_bytes >> 1) & 7;
100 
101  xmm0 = _mm_load_si128(p_scalars);
102 
103  xmm1 = _mm_shufflelo_epi16(xmm0, 0);
104  xmm2 = _mm_shufflelo_epi16(xmm0, 0x55);
105  xmm3 = _mm_shufflelo_epi16(xmm0, 0xaa);
106  xmm4 = _mm_shufflelo_epi16(xmm0, 0xff);
107 
108  xmm1 = _mm_shuffle_epi32(xmm1, 0x00);
109  xmm2 = _mm_shuffle_epi32(xmm2, 0x00);
110  xmm3 = _mm_shuffle_epi32(xmm3, 0x00);
111  xmm4 = _mm_shuffle_epi32(xmm4, 0x00);
112 
113 
114  for (; i < bound; ++i) {
115  xmm0 = _mm_setzero_si128();
116  xmm5 = _mm_setzero_si128();
117  xmm6 = _mm_setzero_si128();
118  xmm7 = _mm_setzero_si128();
119 
120  xmm0 = _mm_insert_epi16(xmm0, src0[p_permute_indexes[0]], 0);
121  xmm5 = _mm_insert_epi16(xmm5, src0[p_permute_indexes[1]], 1);
122  xmm6 = _mm_insert_epi16(xmm6, src0[p_permute_indexes[2]], 2);
123  xmm7 = _mm_insert_epi16(xmm7, src0[p_permute_indexes[3]], 3);
124  xmm0 = _mm_insert_epi16(xmm0, src0[p_permute_indexes[4]], 4);
125  xmm5 = _mm_insert_epi16(xmm5, src0[p_permute_indexes[5]], 5);
126  xmm6 = _mm_insert_epi16(xmm6, src0[p_permute_indexes[6]], 6);
127  xmm7 = _mm_insert_epi16(xmm7, src0[p_permute_indexes[7]], 7);
128 
129  xmm0 = _mm_add_epi16(xmm0, xmm5);
130  xmm6 = _mm_add_epi16(xmm6, xmm7);
131 
132  p_permute_indexes += 8;
133 
134  xmm0 = _mm_add_epi16(xmm0, xmm6);
135 
136  xmm5 = _mm_load_si128(p_cntl0);
137  xmm6 = _mm_load_si128(p_cntl1);
138  xmm7 = _mm_load_si128(p_cntl2);
139 
140  xmm5 = _mm_and_si128(xmm5, xmm1);
141  xmm6 = _mm_and_si128(xmm6, xmm2);
142  xmm7 = _mm_and_si128(xmm7, xmm3);
143 
144  xmm0 = _mm_add_epi16(xmm0, xmm5);
145 
146  xmm5 = _mm_load_si128(p_cntl3);
147 
148  xmm6 = _mm_add_epi16(xmm6, xmm7);
149 
150  p_cntl0 += 1;
151 
152  xmm5 = _mm_and_si128(xmm5, xmm4);
153 
154  xmm0 = _mm_add_epi16(xmm0, xmm6);
155 
156  p_cntl1 += 1;
157  p_cntl2 += 1;
158 
159  xmm0 = _mm_add_epi16(xmm0, xmm5);
160 
161  p_cntl3 += 1;
162 
163  _mm_store_si128(p_target, xmm0);
164 
165  p_target += 1;
166  }
167 
168  for (i = bound * 8; i < (bound * 8) + leftovers; ++i) {
169  target[i] = src0[permute_indexes[i]] + (cntl0[i] & scalars[0]) +
170  (cntl1[i] & scalars[1]) + (cntl2[i] & scalars[2]) +
171  (cntl3[i] & scalars[3]);
172  }
173 }
174 #endif /*LV_HAVE_SSE*/
175 
176 
177 #ifdef LV_HAVE_GENERIC
178 static inline void volk_16i_permute_and_scalar_add_generic(short* target,
179  short* src0,
180  short* permute_indexes,
181  short* cntl0,
182  short* cntl1,
183  short* cntl2,
184  short* cntl3,
185  short* scalars,
186  unsigned int num_points)
187 {
188  const unsigned int num_bytes = num_points * 2;
189 
190  int i = 0;
191 
192  int bound = num_bytes >> 1;
193 
194  for (i = 0; i < bound; ++i) {
195  target[i] = src0[permute_indexes[i]] + (cntl0[i] & scalars[0]) +
196  (cntl1[i] & scalars[1]) + (cntl2[i] & scalars[2]) +
197  (cntl3[i] & scalars[3]);
198  }
199 }
200 
201 #endif /*LV_HAVE_GENERIC*/
202 
203 #endif /*INCLUDED_volk_16i_permute_and_scalar_add_a_H*/
static void volk_16i_permute_and_scalar_add_a_sse2(short *target, short *src0, short *permute_indexes, short *cntl0, short *cntl1, short *cntl2, short *cntl3, short *scalars, unsigned int num_points)
Definition: volk_16i_permute_and_scalar_add.h:70
static void volk_16i_permute_and_scalar_add_generic(short *target, short *src0, short *permute_indexes, short *cntl0, short *cntl1, short *cntl2, short *cntl3, short *scalars, unsigned int num_points)
Definition: volk_16i_permute_and_scalar_add.h:178
for i
Definition: volk_config_fixed.tmpl.h:25