src/libFLAC/stream_decoder.c : Fix NULL de-reference.
[flac.git] / src / libFLAC / stream_encoder_intrin_avx2.c
1 /* libFLAC - Free Lossless Audio Codec library
2  * Copyright (C) 2000-2009  Josh Coalson
3  * Copyright (C) 2011-2014  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 #ifndef FLAC__NO_ASM
38 #if (defined FLAC__CPU_IA32 || defined FLAC__CPU_X86_64) && defined FLAC__HAS_X86INTRIN
39 #include "private/stream_encoder.h"
40 #include "private/bitmath.h"
41 #ifdef FLAC__AVX2_SUPPORTED
42
43 #include <stdlib.h>    /* for abs() */
44 #include <immintrin.h> /* AVX2 */
45 #include "FLAC/assert.h"
46
47 FLAC__SSE_TARGET("avx2")
48 void FLAC__precompute_partition_info_sums_intrin_avx2(const FLAC__int32 residual[], FLAC__uint64 abs_residual_partition_sums[],
49                 unsigned residual_samples, unsigned predictor_order, unsigned min_partition_order, unsigned max_partition_order, unsigned bps)
50 {
51         const unsigned default_partition_samples = (residual_samples + predictor_order) >> max_partition_order;
52         unsigned partitions = 1u << max_partition_order;
53
54         FLAC__ASSERT(default_partition_samples > predictor_order);
55
56         /* first do max_partition_order */
57         {
58                 unsigned partition, residual_sample, end = (unsigned)(-(int)predictor_order);
59                 __m256i res256, sum256;
60                 __m128i res128, sum128;
61
62                 if(FLAC__bitmath_ilog2(default_partition_samples) + bps + FLAC__MAX_EXTRA_RESIDUAL_BPS < 32) {
63                         for(partition = residual_sample = 0; partition < partitions; partition++) {
64                                 end += default_partition_samples;
65                                 sum256 = _mm256_setzero_si256();
66
67                                 for( ; (int)residual_sample < (int)end-7; residual_sample+=8) {
68                                         res256 = _mm256_abs_epi32(_mm256_loadu_si256((const __m256i*)(residual+residual_sample)));
69                                         sum256 = _mm256_add_epi32(sum256, res256);
70                                 }
71
72                                 sum128 = _mm_add_epi32(_mm256_extracti128_si256(sum256, 1), _mm256_castsi256_si128(sum256));
73
74                                 for( ; (int)residual_sample < (int)end-3; residual_sample+=4) {
75                                         res128 = _mm_abs_epi32(_mm_loadu_si128((const __m128i*)(residual+residual_sample)));
76                                         sum128 = _mm_add_epi32(sum128, res128);
77                                 }
78
79                                 for( ; residual_sample < end; residual_sample++) {
80                                         res128 = _mm_cvtsi32_si128(residual[residual_sample]);
81                                         res128 = _mm_abs_epi32(res128);
82                                         sum128 = _mm_add_epi32(sum128, res128);
83                                 }
84
85                                 sum128 = _mm_hadd_epi32(sum128, sum128);
86                                 sum128 = _mm_hadd_epi32(sum128, sum128);
87                                 abs_residual_partition_sums[partition] = (FLAC__uint32)_mm_cvtsi128_si32(sum128);
88                         }
89                 }
90                 else { /* have to pessimistically use 64 bits for accumulator */
91                         for(partition = residual_sample = 0; partition < partitions; partition++) {
92                                 end += default_partition_samples;
93                                 sum256 = _mm256_setzero_si256();
94
95                                 for( ; (int)residual_sample < (int)end-3; residual_sample+=4) {
96                                         res128 = _mm_abs_epi32(_mm_loadu_si128((const __m128i*)(residual+residual_sample)));
97                                         res256 = _mm256_cvtepu32_epi64(res128);
98                                         sum256 = _mm256_add_epi64(sum256, res256);
99                                 }
100
101                                 sum128 = _mm_add_epi64(_mm256_extracti128_si256(sum256, 1), _mm256_castsi256_si128(sum256));
102
103                                 for( ; (int)residual_sample < (int)end-1; residual_sample+=2) {
104                                         res128 = _mm_loadl_epi64((const __m128i*)(residual+residual_sample));
105                                         res128 = _mm_abs_epi32(res128);
106                                         res128 = _mm_cvtepu32_epi64(res128);
107                                         sum128 = _mm_add_epi64(sum128, res128);
108                                 }
109
110                                 for( ; residual_sample < end; residual_sample++) {
111                                         res128 = _mm_cvtsi32_si128(residual[residual_sample]);
112                                         res128 = _mm_abs_epi32(res128);
113                                         sum128 = _mm_add_epi64(sum128, res128);
114                                 }
115
116                                 sum128 = _mm_add_epi64(sum128, _mm_srli_si128(sum128, 8));
117                                 _mm_storel_epi64((__m128i*)(abs_residual_partition_sums+partition), sum128);
118                         }
119                 }
120         }
121
122         /* now merge partitions for lower orders */
123         {
124                 unsigned from_partition = 0, to_partition = partitions;
125                 int partition_order;
126                 for(partition_order = (int)max_partition_order - 1; partition_order >= (int)min_partition_order; partition_order--) {
127                         unsigned i;
128                         partitions >>= 1;
129                         for(i = 0; i < partitions; i++) {
130                                 abs_residual_partition_sums[to_partition++] =
131                                         abs_residual_partition_sums[from_partition  ] +
132                                         abs_residual_partition_sums[from_partition+1];
133                                 from_partition += 2;
134                         }
135                 }
136         }
137         _mm256_zeroupper();
138 }
139
140 #endif /* FLAC__AVX2_SUPPORTED */
141 #endif /* (FLAC__CPU_IA32 || FLAC__CPU_X86_64) && FLAC__HAS_X86INTRIN */
142 #endif /* FLAC__NO_ASM */