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) && FLAC__HAS_X86INTRIN
42 #include "private/fixed.h"
43 #ifdef FLAC__SSE4_2_SUPPORTED
44
45 #include <nmmintrin.h> /* SSE4.2 */
46 #include <math.h>
47 #include "private/macros.h"
48 #include "share/compat.h"
49 #include "FLAC/assert.h"
50
51 #ifdef local_abs64
52 #undef local_abs64
53 #endif
54 #define local_abs64(x) ((uint64_t)((x)<0? -(x) : (x)))
55
56 #define CHECK_ORDER_IS_VALID(macro_order) \
57 if(shadow_error_##macro_order <= INT32_MAX) { \
58 if(total_error_##macro_order < smallest_error) { \
59 order = macro_order; \
60 smallest_error = total_error_##macro_order ; \
61 } \
62 residual_bits_per_sample[ macro_order ] = (float)((total_error_0 > 0) ? log(M_LN2 * (double)total_error_0 / (double)data_len) / M_LN2 : 0.0); \
63 } \
64 else \
65 residual_bits_per_sample[ macro_order ] = 34.0f;
66
67 FLAC__SSE_TARGET("sse4.2")
FLAC__fixed_compute_best_predictor_limit_residual_intrin_sse42(const FLAC__int32 data[],uint32_t data_len,float residual_bits_per_sample[FLAC__MAX_FIXED_ORDER+1])68 uint32_t FLAC__fixed_compute_best_predictor_limit_residual_intrin_sse42(const FLAC__int32 data[], uint32_t data_len, float residual_bits_per_sample[FLAC__MAX_FIXED_ORDER + 1])
69 {
70 FLAC__uint64 total_error_0 = 0, total_error_1 = 0, total_error_2 = 0, total_error_3 = 0, total_error_4 = 0, smallest_error = UINT64_MAX;
71 FLAC__uint64 shadow_error_0 = 0, shadow_error_1 = 0, shadow_error_2 = 0, shadow_error_3 = 0, shadow_error_4 = 0;
72 FLAC__uint64 error_0, error_1, error_2, error_3, error_4;
73 FLAC__int32 i, data_len_int;
74 uint32_t order = 0;
75 __m128i total_err0, total_err1, total_err2, total_err3, total_err4;
76 __m128i shadow_err0, shadow_err1, shadow_err2, shadow_err3, shadow_err4;
77 __m128i prev_err0, prev_err1, prev_err2, prev_err3;
78 __m128i tempA, tempB, bitmask;
79 FLAC__int64 data_scalar[2];
80 FLAC__int64 prev_err0_scalar[2];
81 FLAC__int64 prev_err1_scalar[2];
82 FLAC__int64 prev_err2_scalar[2];
83 FLAC__int64 prev_err3_scalar[2];
84 total_err0 = _mm_setzero_si128();
85 total_err1 = _mm_setzero_si128();
86 total_err2 = _mm_setzero_si128();
87 total_err3 = _mm_setzero_si128();
88 total_err4 = _mm_setzero_si128();
89 shadow_err0 = _mm_setzero_si128();
90 shadow_err1 = _mm_setzero_si128();
91 shadow_err2 = _mm_setzero_si128();
92 shadow_err3 = _mm_setzero_si128();
93 shadow_err4 = _mm_setzero_si128();
94 data_len_int = data_len;
95
96 /* First take care of preceding samples */
97 for(i = -4; i < 0; i++) {
98 error_0 = local_abs64((FLAC__int64)data[i]);
99 error_1 = (i > -4) ? local_abs64((FLAC__int64)data[i] - data[i-1]) : 0 ;
100 error_2 = (i > -3) ? local_abs64((FLAC__int64)data[i] - 2 * (FLAC__int64)data[i-1] + data[i-2]) : 0;
101 error_3 = (i > -2) ? local_abs64((FLAC__int64)data[i] - 3 * (FLAC__int64)data[i-1] + 3 * (FLAC__int64)data[i-2] - data[i-3]) : 0;
102
103 total_error_0 += error_0;
104 total_error_1 += error_1;
105 total_error_2 += error_2;
106 total_error_3 += error_3;
107
108 shadow_error_0 |= error_0;
109 shadow_error_1 |= error_1;
110 shadow_error_2 |= error_2;
111 shadow_error_3 |= error_3;
112 }
113
114 for(i = 0; i < 2; i++){
115 prev_err0_scalar[i] = data[-1+i*(data_len_int/2)];
116 prev_err1_scalar[i] = (FLAC__int64)(data[-1+i*(data_len_int/2)]) - data[-2+i*(data_len_int/2)];
117 prev_err2_scalar[i] = prev_err1_scalar[i] - ((FLAC__int64)(data[-2+i*(data_len_int/2)]) - data[-3+i*(data_len_int/2)]);
118 prev_err3_scalar[i] = prev_err2_scalar[i] - ((FLAC__int64)(data[-2+i*(data_len_int/2)]) - 2*(FLAC__int64)(data[-3+i*(data_len_int/2)]) + data[-4+i*(data_len_int/2)]);
119 }
120 prev_err0 = _mm_loadu_si128((const __m128i*)prev_err0_scalar);
121 prev_err1 = _mm_loadu_si128((const __m128i*)prev_err1_scalar);
122 prev_err2 = _mm_loadu_si128((const __m128i*)prev_err2_scalar);
123 prev_err3 = _mm_loadu_si128((const __m128i*)prev_err3_scalar);
124 for(i = 0; i < data_len_int / 2; i++){
125 data_scalar[0] = data[i];
126 data_scalar[1] = data[i+data_len/2];
127 tempA = _mm_loadu_si128((const __m128i*)data_scalar);
128 /* Next three intrinsics calculate tempB as abs of tempA */
129 bitmask = _mm_cmpgt_epi64(_mm_set1_epi64x(0), tempA);
130 tempB = _mm_xor_si128(tempA, bitmask);
131 tempB = _mm_sub_epi64(tempB, bitmask);
132 total_err0 = _mm_add_epi64(total_err0,tempB);
133 shadow_err0 = _mm_or_si128(shadow_err0,tempB);
134 tempB = _mm_sub_epi64(tempA,prev_err0);
135 prev_err0 = tempA;
136 /* Next three intrinsics calculate tempA as abs of tempB */
137 bitmask = _mm_cmpgt_epi64(_mm_set1_epi64x(0), tempB);
138 tempA = _mm_xor_si128(tempB, bitmask);
139 tempA = _mm_sub_epi64(tempA, bitmask);
140 total_err1 = _mm_add_epi64(total_err1,tempA);
141 shadow_err1 = _mm_or_si128(shadow_err1,tempA);
142 tempA = _mm_sub_epi64(tempB,prev_err1);
143 prev_err1 = tempB;
144 /* Next three intrinsics calculate tempB as abs of tempA */
145 bitmask = _mm_cmpgt_epi64(_mm_set1_epi64x(0), tempA);
146 tempB = _mm_xor_si128(tempA, bitmask);
147 tempB = _mm_sub_epi64(tempB, bitmask);
148 total_err2 = _mm_add_epi64(total_err2,tempB);
149 shadow_err2 = _mm_or_si128(shadow_err2,tempB);
150 tempB = _mm_sub_epi64(tempA,prev_err2);
151 prev_err2 = tempA;
152 /* Next three intrinsics calculate tempA as abs of tempB */
153 bitmask = _mm_cmpgt_epi64(_mm_set1_epi64x(0), tempB);
154 tempA = _mm_xor_si128(tempB, bitmask);
155 tempA = _mm_sub_epi64(tempA, bitmask);
156 total_err3 = _mm_add_epi64(total_err3,tempA);
157 shadow_err3 = _mm_or_si128(shadow_err3,tempA);
158 tempA = _mm_sub_epi64(tempB,prev_err3);
159 prev_err3 = tempB;
160 /* Next three intrinsics calculate tempB as abs of tempA */
161 bitmask = _mm_cmpgt_epi64(_mm_set1_epi64x(0), tempA);
162 tempB = _mm_xor_si128(tempA, bitmask);
163 tempB = _mm_sub_epi64(tempB, bitmask);
164 total_err4 = _mm_add_epi64(total_err4,tempB);
165 shadow_err4 = _mm_or_si128(shadow_err4,tempB);
166 }
167 _mm_storeu_si128((__m128i*)data_scalar,total_err0);
168 total_error_0 += data_scalar[0] + data_scalar[1];
169 _mm_storeu_si128((__m128i*)data_scalar,total_err1);
170 total_error_1 += data_scalar[0] + data_scalar[1];
171 _mm_storeu_si128((__m128i*)data_scalar,total_err2);
172 total_error_2 += data_scalar[0] + data_scalar[1];
173 _mm_storeu_si128((__m128i*)data_scalar,total_err3);
174 total_error_3 += data_scalar[0] + data_scalar[1];
175 _mm_storeu_si128((__m128i*)data_scalar,total_err4);
176 total_error_4 += data_scalar[0] + data_scalar[1];
177 _mm_storeu_si128((__m128i*)data_scalar,shadow_err0);
178 shadow_error_0 |= data_scalar[0] | data_scalar[1];
179 _mm_storeu_si128((__m128i*)data_scalar,shadow_err1);
180 shadow_error_1 |= data_scalar[0] | data_scalar[1];
181 _mm_storeu_si128((__m128i*)data_scalar,shadow_err2);
182 shadow_error_2 |= data_scalar[0] | data_scalar[1];
183 _mm_storeu_si128((__m128i*)data_scalar,shadow_err3);
184 shadow_error_3 |= data_scalar[0] | data_scalar[1];
185 _mm_storeu_si128((__m128i*)data_scalar,shadow_err4);
186 shadow_error_4 |= data_scalar[0] | data_scalar[1];
187
188 /* Take care of remaining sample */
189 if(data_len_int % 2 > 0) {
190 i += data_len/2;
191 error_0 = local_abs64((FLAC__int64)data[i]);
192 error_1 = local_abs64((FLAC__int64)data[i] - data[i-1]);
193 error_2 = local_abs64((FLAC__int64)data[i] - 2 * (FLAC__int64)data[i-1] + data[i-2]);
194 error_3 = local_abs64((FLAC__int64)data[i] - 3 * (FLAC__int64)data[i-1] + 3 * (FLAC__int64)data[i-2] - data[i-3]);
195 error_4 = local_abs64((FLAC__int64)data[i] - 4 * (FLAC__int64)data[i-1] + 6 * (FLAC__int64)data[i-2] - 4 * (FLAC__int64)data[i-3] + data[i-4]);
196
197 total_error_0 += error_0;
198 total_error_1 += error_1;
199 total_error_2 += error_2;
200 total_error_3 += error_3;
201 total_error_4 += error_4;
202
203 shadow_error_0 |= error_0;
204 shadow_error_1 |= error_1;
205 shadow_error_2 |= error_2;
206 shadow_error_3 |= error_3;
207 shadow_error_4 |= error_4;
208 }
209
210
211 CHECK_ORDER_IS_VALID(0);
212 CHECK_ORDER_IS_VALID(1);
213 CHECK_ORDER_IS_VALID(2);
214 CHECK_ORDER_IS_VALID(3);
215 CHECK_ORDER_IS_VALID(4);
216
217 return order;
218 }
219
220 #endif /* FLAC__SSE4_2_SUPPORTED */
221 #endif /* (FLAC__CPU_IA32 || FLAC__CPU_X86_64) && FLAC__HAS_X86INTRIN */
222 #endif /* FLAC__NO_ASM */
223 #endif /* FLAC__INTEGER_ONLY_LIBRARY */
224