x86: check for AV_CPU_FLAG_AVXSLOW where useful
[libav.git] / libavcodec / x86 / dcadsp_init.c
CommitLineData
5b59a9fc
CG
1/*
2 * Copyright (c) 2012-2014 Christophe Gisquet <christophe.gisquet@gmail.com>
3 *
4 * This file is part of Libav.
5 *
6 * Libav is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
10 *
11 * Libav is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
15 *
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with Libav; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 */
20
21#include "libavutil/attributes.h"
22#include "libavutil/cpu.h"
23#include "libavutil/x86/cpu.h"
24#include "libavcodec/dcadsp.h"
25
4cb69642
CG
26void ff_decode_hf_sse(float dst[DCA_SUBBANDS][8], const int vq_num[DCA_SUBBANDS],
27 const int8_t hf_vq[1024][32], intptr_t vq_offset,
28 int scale[DCA_SUBBANDS][2], intptr_t start, intptr_t end);
29void ff_decode_hf_sse2(float dst[DCA_SUBBANDS][8], const int vq_num[DCA_SUBBANDS],
30 const int8_t hf_vq[1024][32], intptr_t vq_offset,
31 int scale[DCA_SUBBANDS][2], intptr_t start, intptr_t end);
32void ff_decode_hf_sse4(float dst[DCA_SUBBANDS][8], const int vq_num[DCA_SUBBANDS],
33 const int8_t hf_vq[1024][32], intptr_t vq_offset,
34 int scale[DCA_SUBBANDS][2], intptr_t start, intptr_t end);
ad507d79
CG
35void ff_dca_lfe_fir0_sse(float *out, const float *in, const float *coefs);
36void ff_dca_lfe_fir1_sse(float *out, const float *in, const float *coefs);
5b59a9fc
CG
37
38av_cold void ff_dcadsp_init_x86(DCADSPContext *s)
39{
40 int cpu_flags = av_get_cpu_flags();
41
42 if (EXTERNAL_SSE(cpu_flags)) {
43#if ARCH_X86_32
4cb69642 44 s->decode_hf = ff_decode_hf_sse;
5b59a9fc 45#endif
ad507d79
CG
46 s->lfe_fir[0] = ff_dca_lfe_fir0_sse;
47 s->lfe_fir[1] = ff_dca_lfe_fir1_sse;
5b59a9fc
CG
48 }
49
50 if (EXTERNAL_SSE2(cpu_flags)) {
4cb69642 51 s->decode_hf = ff_decode_hf_sse2;
5b59a9fc
CG
52 }
53
54 if (EXTERNAL_SSE4(cpu_flags)) {
4cb69642 55 s->decode_hf = ff_decode_hf_sse4;
5b59a9fc
CG
56 }
57}
08e3ea60 58
08e3ea60 59
2025d802
JA
60#define SYNTH_FILTER_FUNC(opt) \
61void ff_synth_filter_inner_##opt(float *synth_buf_ptr, float synth_buf2[32], \
62 const float window[512], \
63 float out[32], intptr_t offset, float scale); \
64static void synth_filter_##opt(FFTContext *imdct, \
65 float *synth_buf_ptr, int *synth_buf_offset, \
66 float synth_buf2[32], const float window[512], \
67 float out[32], const float in[32], float scale) \
68{ \
69 float *synth_buf= synth_buf_ptr + *synth_buf_offset; \
70 \
71 imdct->imdct_half(imdct, synth_buf, in); \
72 \
73 ff_synth_filter_inner_##opt(synth_buf, synth_buf2, window, \
74 out, *synth_buf_offset, scale); \
75 \
76 *synth_buf_offset = (*synth_buf_offset - 32) & 511; \
77} \
08e3ea60 78
2025d802
JA
79#if HAVE_YASM
80#if ARCH_X86_32
81SYNTH_FILTER_FUNC(sse)
82#endif
83SYNTH_FILTER_FUNC(sse2)
81e02fae 84SYNTH_FILTER_FUNC(avx)
c74b8669 85SYNTH_FILTER_FUNC(fma3)
3bfdee00 86#endif /* HAVE_YASM */
08e3ea60
CG
87
88av_cold void ff_synth_filter_init_x86(SynthFilterContext *s)
89{
3bfdee00 90#if HAVE_YASM
08e3ea60
CG
91 int cpu_flags = av_get_cpu_flags();
92
2025d802
JA
93#if ARCH_X86_32
94 if (EXTERNAL_SSE(cpu_flags)) {
95 s->synth_filter_float = synth_filter_sse;
96 }
97#endif
08e3ea60
CG
98 if (EXTERNAL_SSE2(cpu_flags)) {
99 s->synth_filter_float = synth_filter_sse2;
100 }
d68c0538 101 if (EXTERNAL_AVX_FAST(cpu_flags)) {
81e02fae
JA
102 s->synth_filter_float = synth_filter_avx;
103 }
d68c0538 104 if (EXTERNAL_FMA3(cpu_flags) && !(cpu_flags & AV_CPU_FLAG_AVXSLOW)) {
c74b8669
JA
105 s->synth_filter_float = synth_filter_fma3;
106 }
3bfdee00 107#endif /* HAVE_YASM */
08e3ea60 108}