1 /* libFLAC - Free Lossless Audio Codec library
2 * Copyright (C) 2000-2009 Josh Coalson
3 * Copyright (C) 2011-2023 Xiph.Org Foundation
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 *
9 * - Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 *
12 * - Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 *
16 * - Neither the name of the Xiph.org Foundation nor the names of its
17 * contributors may be used to endorse or promote products derived from
18 * this software without specific prior written permission.
19 *
20 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
22 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
23 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR
24 * CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
25 * EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
26 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
27 * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
28 * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
29 * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
30 * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
31 */
32
33 #ifdef HAVE_CONFIG_H
34 # include <config.h>
35 #endif
36
37 #include "private/cpu.h"
38
39 #ifndef FLAC__INTEGER_ONLY_LIBRARY
40 #ifndef FLAC__NO_ASM
41 #if (defined FLAC__CPU_IA32 || defined FLAC__CPU_X86_64) && defined FLAC__HAS_X86INTRIN
42 #include "private/fixed.h"
43 #ifdef FLAC__SSE2_SUPPORTED
44
45 #include <emmintrin.h> /* SSE2 */
46 #include <math.h>
47 #include "private/macros.h"
48 #include "share/compat.h"
49 #include "FLAC/assert.h"
50
51 #ifdef FLAC__CPU_IA32
52 #define m128i_to_i64(dest, src) _mm_storel_epi64((__m128i*)&dest, src)
53 #else
54 #define m128i_to_i64(dest, src) dest = _mm_cvtsi128_si64(src)
55 #endif
56
57 #ifdef local_abs
58 #undef local_abs
59 #endif
60 #define local_abs(x) ((uint32_t)((x)<0? -(x) : (x)))
61
62 FLAC__SSE_TARGET("sse2")
FLAC__fixed_compute_best_predictor_intrin_sse2(const FLAC__int32 data[],uint32_t data_len,float residual_bits_per_sample[FLAC__MAX_FIXED_ORDER+1])63 uint32_t FLAC__fixed_compute_best_predictor_intrin_sse2(const FLAC__int32 data[], uint32_t data_len, float residual_bits_per_sample[FLAC__MAX_FIXED_ORDER + 1])
64 {
65 FLAC__uint32 total_error_0, total_error_1, total_error_2, total_error_3, total_error_4;
66 FLAC__int32 i, data_len_int;
67 uint32_t order;
68 __m128i total_err0, total_err1, total_err2, total_err3, total_err4;
69 __m128i prev_err0, prev_err1, prev_err2, prev_err3;
70 __m128i tempA, tempB, bitmask;
71 FLAC__int32 data_scalar[4];
72 FLAC__int32 prev_err0_scalar[4];
73 FLAC__int32 prev_err1_scalar[4];
74 FLAC__int32 prev_err2_scalar[4];
75 FLAC__int32 prev_err3_scalar[4];
76 total_err0 = _mm_setzero_si128();
77 total_err1 = _mm_setzero_si128();
78 total_err2 = _mm_setzero_si128();
79 total_err3 = _mm_setzero_si128();
80 total_err4 = _mm_setzero_si128();
81 data_len_int = data_len;
82
83 for(i = 0; i < 4; i++){
84 prev_err0_scalar[i] = data[-1+i*(data_len_int/4)];
85 prev_err1_scalar[i] = data[-1+i*(data_len_int/4)] - data[-2+i*(data_len_int/4)];
86 prev_err2_scalar[i] = prev_err1_scalar[i] - (data[-2+i*(data_len_int/4)] - data[-3+i*(data_len_int/4)]);
87 prev_err3_scalar[i] = prev_err2_scalar[i] - (data[-2+i*(data_len_int/4)] - 2*data[-3+i*(data_len_int/4)] + data[-4+i*(data_len_int/4)]);
88 }
89 prev_err0 = _mm_loadu_si128((const __m128i*)prev_err0_scalar);
90 prev_err1 = _mm_loadu_si128((const __m128i*)prev_err1_scalar);
91 prev_err2 = _mm_loadu_si128((const __m128i*)prev_err2_scalar);
92 prev_err3 = _mm_loadu_si128((const __m128i*)prev_err3_scalar);
93 for(i = 0; i < data_len_int / 4; i++){
94 data_scalar[0] = data[i];
95 data_scalar[1] = data[i+data_len/4];
96 data_scalar[2] = data[i+2*(data_len/4)];
97 data_scalar[3] = data[i+3*(data_len/4)];
98 tempA = _mm_loadu_si128((const __m128i*)data_scalar);
99 /* Next three intrinsics calculate tempB as abs of tempA */
100 bitmask = _mm_srai_epi32(tempA, 31);
101 tempB = _mm_xor_si128(tempA, bitmask);
102 tempB = _mm_sub_epi32(tempB, bitmask);
103 total_err0 = _mm_add_epi32(total_err0,tempB);
104 tempB = _mm_sub_epi32(tempA,prev_err0);
105 prev_err0 = tempA;
106 /* Next three intrinsics calculate tempA as abs of tempB */
107 bitmask = _mm_srai_epi32(tempB, 31);
108 tempA = _mm_xor_si128(tempB, bitmask);
109 tempA = _mm_sub_epi32(tempA, bitmask);
110 total_err1 = _mm_add_epi32(total_err1,tempA);
111 tempA = _mm_sub_epi32(tempB,prev_err1);
112 prev_err1 = tempB;
113 /* Next three intrinsics calculate tempB as abs of tempA */
114 bitmask = _mm_srai_epi32(tempA, 31);
115 tempB = _mm_xor_si128(tempA, bitmask);
116 tempB = _mm_sub_epi32(tempB, bitmask);
117 total_err2 = _mm_add_epi32(total_err2,tempB);
118 tempB = _mm_sub_epi32(tempA,prev_err2);
119 prev_err2 = tempA;
120 /* Next three intrinsics calculate tempA as abs of tempB */
121 bitmask = _mm_srai_epi32(tempB, 31);
122 tempA = _mm_xor_si128(tempB, bitmask);
123 tempA = _mm_sub_epi32(tempA, bitmask);
124 total_err3 = _mm_add_epi32(total_err3,tempA);
125 tempA = _mm_sub_epi32(tempB,prev_err3);
126 prev_err3 = tempB;
127 /* Next three intrinsics calculate tempB as abs of tempA */
128 bitmask = _mm_srai_epi32(tempA, 31);
129 tempB = _mm_xor_si128(tempA, bitmask);
130 tempB = _mm_sub_epi32(tempB, bitmask);
131 total_err4 = _mm_add_epi32(total_err4,tempB);
132 }
133 _mm_storeu_si128((__m128i*)data_scalar,total_err0);
134 total_error_0 = data_scalar[0] + data_scalar[1] + data_scalar[2] + data_scalar[3];
135 _mm_storeu_si128((__m128i*)data_scalar,total_err1);
136 total_error_1 = data_scalar[0] + data_scalar[1] + data_scalar[2] + data_scalar[3];
137 _mm_storeu_si128((__m128i*)data_scalar,total_err2);
138 total_error_2 = data_scalar[0] + data_scalar[1] + data_scalar[2] + data_scalar[3];
139 _mm_storeu_si128((__m128i*)data_scalar,total_err3);
140 total_error_3 = data_scalar[0] + data_scalar[1] + data_scalar[2] + data_scalar[3];
141 _mm_storeu_si128((__m128i*)data_scalar,total_err4);
142 total_error_4 = data_scalar[0] + data_scalar[1] + data_scalar[2] + data_scalar[3];
143
144 /* Now the remainder of samples needs to be processed */
145 i *= 4;
146 if(data_len % 4 > 0){
147 FLAC__int32 last_error_0 = data[i-1];
148 FLAC__int32 last_error_1 = data[i-1] - data[i-2];
149 FLAC__int32 last_error_2 = last_error_1 - (data[i-2] - data[i-3]);
150 FLAC__int32 last_error_3 = last_error_2 - (data[i-2] - 2*data[i-3] + data[i-4]);
151 FLAC__int32 error, save;
152 for(; i < data_len_int; i++) {
153 error = data[i] ; total_error_0 += local_abs(error); save = error;
154 error -= last_error_0; total_error_1 += local_abs(error); last_error_0 = save; save = error;
155 error -= last_error_1; total_error_2 += local_abs(error); last_error_1 = save; save = error;
156 error -= last_error_2; total_error_3 += local_abs(error); last_error_2 = save; save = error;
157 error -= last_error_3; total_error_4 += local_abs(error); last_error_3 = save;
158 }
159 }
160
161 /* prefer lower order */
162 if(total_error_0 <= flac_min(flac_min(flac_min(total_error_1, total_error_2), total_error_3), total_error_4))
163 order = 0;
164 else if(total_error_1 <= flac_min(flac_min(total_error_2, total_error_3), total_error_4))
165 order = 1;
166 else if(total_error_2 <= flac_min(total_error_3, total_error_4))
167 order = 2;
168 else if(total_error_3 <= total_error_4)
169 order = 3;
170 else
171 order = 4;
172
173 /* Estimate the expected number of bits per residual signal sample. */
174 /* 'total_error*' is linearly related to the variance of the residual */
175 /* signal, so we use it directly to compute E(|x|) */
176 FLAC__ASSERT(data_len > 0 || total_error_0 == 0);
177 FLAC__ASSERT(data_len > 0 || total_error_1 == 0);
178 FLAC__ASSERT(data_len > 0 || total_error_2 == 0);
179 FLAC__ASSERT(data_len > 0 || total_error_3 == 0);
180 FLAC__ASSERT(data_len > 0 || total_error_4 == 0);
181
182 residual_bits_per_sample[0] = (float)((total_error_0 > 0) ? log(M_LN2 * (double)total_error_0 / (double)data_len) / M_LN2 : 0.0);
183 residual_bits_per_sample[1] = (float)((total_error_1 > 0) ? log(M_LN2 * (double)total_error_1 / (double)data_len) / M_LN2 : 0.0);
184 residual_bits_per_sample[2] = (float)((total_error_2 > 0) ? log(M_LN2 * (double)total_error_2 / (double)data_len) / M_LN2 : 0.0);
185 residual_bits_per_sample[3] = (float)((total_error_3 > 0) ? log(M_LN2 * (double)total_error_3 / (double)data_len) / M_LN2 : 0.0);
186 residual_bits_per_sample[4] = (float)((total_error_4 > 0) ? log(M_LN2 * (double)total_error_4 / (double)data_len) / M_LN2 : 0.0);
187
188 return order;
189 }
190
191 #endif /* FLAC__SSE2_SUPPORTED */
192 #endif /* (FLAC__CPU_IA32 || FLAC__CPU_X86_64) && FLAC__HAS_X86INTRIN */
193 #endif /* FLAC__NO_ASM */
194 #endif /* FLAC__INTEGER_ONLY_LIBRARY */
195