1 /* Spa
2 *
3 * Copyright © 2019 Wim Taymans
4 *
5 * Permission is hereby granted, free of charge, to any person obtaining a
6 * copy of this software and associated documentation files (the "Software"),
7 * to deal in the Software without restriction, including without limitation
8 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
9 * and/or sell copies of the Software, and to permit persons to whom the
10 * Software is furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice (including the next
13 * paragraph) shall be included in all copies or substantial portions of the
14 * Software.
15 *
16 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
19 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
20 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
21 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
22 * DEALINGS IN THE SOFTWARE.
23 */
24
25 #include <string.h>
26 #include <stdio.h>
27 #include <math.h>
28
29 #include <spa/utils/defs.h>
30
31 #include "mix-ops.h"
32
33 #include <immintrin.h>
34
mix_4(float * dst,const float * SPA_RESTRICT src0,const float * SPA_RESTRICT src1,const float * SPA_RESTRICT src2,uint32_t n_samples)35 static inline void mix_4(float * dst,
36 const float * SPA_RESTRICT src0,
37 const float * SPA_RESTRICT src1,
38 const float * SPA_RESTRICT src2,
39 uint32_t n_samples)
40 {
41 uint32_t n, unrolled;
42
43 if (SPA_IS_ALIGNED(src0, 32) &&
44 SPA_IS_ALIGNED(src1, 32) &&
45 SPA_IS_ALIGNED(src2, 32) &&
46 SPA_IS_ALIGNED(dst, 32))
47 unrolled = n_samples & ~15;
48 else
49 unrolled = 0;
50
51 for (n = 0; n < unrolled; n += 16) {
52 __m256 in1[4], in2[4];
53
54 in1[0] = _mm256_load_ps(&dst[n + 0]);
55 in2[0] = _mm256_load_ps(&dst[n + 8]);
56 in1[1] = _mm256_load_ps(&src0[n + 0]);
57 in2[1] = _mm256_load_ps(&src0[n + 8]);
58 in1[2] = _mm256_load_ps(&src1[n + 0]);
59 in2[2] = _mm256_load_ps(&src1[n + 8]);
60 in1[3] = _mm256_load_ps(&src2[n + 0]);
61 in2[3] = _mm256_load_ps(&src2[n + 8]);
62
63 in1[0] = _mm256_add_ps(in1[0], in1[1]);
64 in2[0] = _mm256_add_ps(in2[0], in2[1]);
65 in1[2] = _mm256_add_ps(in1[2], in1[3]);
66 in2[2] = _mm256_add_ps(in2[2], in2[3]);
67 in1[0] = _mm256_add_ps(in1[0], in1[2]);
68 in2[0] = _mm256_add_ps(in2[0], in2[2]);
69
70 _mm256_store_ps(&dst[n + 0], in1[0]);
71 _mm256_store_ps(&dst[n + 8], in2[0]);
72 }
73 for (; n < n_samples; n++) {
74 __m128 in[4];
75 in[0] = _mm_load_ss(&dst[n]),
76 in[1] = _mm_load_ss(&src0[n]),
77 in[2] = _mm_load_ss(&src1[n]),
78 in[3] = _mm_load_ss(&src2[n]),
79 in[0] = _mm_add_ss(in[0], in[1]);
80 in[2] = _mm_add_ss(in[2], in[3]);
81 in[0] = _mm_add_ss(in[0], in[2]);
82 _mm_store_ss(&dst[n], in[0]);
83 }
84 }
85
86
mix_2(float * dst,const float * SPA_RESTRICT src,uint32_t n_samples)87 static inline void mix_2(float * dst, const float * SPA_RESTRICT src, uint32_t n_samples)
88 {
89 uint32_t n, unrolled;
90
91 if (SPA_IS_ALIGNED(src, 32) &&
92 SPA_IS_ALIGNED(dst, 32))
93 unrolled = n_samples & ~15;
94 else
95 unrolled = 0;
96
97 for (n = 0; n < unrolled; n += 16) {
98 __m256 in1[2], in2[2];
99
100 in1[0] = _mm256_load_ps(&dst[n + 0]);
101 in1[1] = _mm256_load_ps(&dst[n + 8]);
102 in2[0] = _mm256_load_ps(&src[n + 0]);
103 in2[1] = _mm256_load_ps(&src[n + 8]);
104
105 in1[0] = _mm256_add_ps(in1[0], in2[0]);
106 in1[1] = _mm256_add_ps(in1[1], in2[1]);
107
108 _mm256_store_ps(&dst[n + 0], in1[0]);
109 _mm256_store_ps(&dst[n + 8], in1[1]);
110 }
111 for (; n < n_samples; n++) {
112 __m128 in1[1], in2[1];
113 in1[0] = _mm_load_ss(&dst[n]),
114 in2[0] = _mm_load_ss(&src[n]),
115 in1[0] = _mm_add_ss(in1[0], in2[0]);
116 _mm_store_ss(&dst[n], in1[0]);
117 }
118 }
119
120 void
mix_f32_avx(struct mix_ops * ops,void * SPA_RESTRICT dst,const void * SPA_RESTRICT src[],uint32_t n_src,uint32_t n_samples)121 mix_f32_avx(struct mix_ops *ops, void * SPA_RESTRICT dst, const void * SPA_RESTRICT src[],
122 uint32_t n_src, uint32_t n_samples)
123 {
124 uint32_t i;
125
126 if (n_src == 0)
127 memset(dst, 0, n_samples * ops->n_channels * sizeof(float));
128 else if (dst != src[0])
129 memcpy(dst, src[0], n_samples * ops->n_channels * sizeof(float));
130
131 for (i = 1; i + 2 < n_src; i += 3)
132 mix_4(dst, src[i], src[i + 1], src[i + 2], n_samples);
133 for (; i < n_src; i++)
134 mix_2(dst, src[i], n_samples * ops->n_channels);
135 }
136