git.sesse.net Git - ffmpeg/blob - libavcodec/x86/dcadsp_init.c

   1 /*
   2  * Copyright (c) 2012-2014 Christophe Gisquet <christophe.gisquet@gmail.com>
   3  *
   4  * This file is part of FFmpeg.
   5  *
   6  * FFmpeg 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  * FFmpeg 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 FFmpeg; 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
  26 void 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);
  29 void 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);
  32 void 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);
  35 void ff_dca_lfe_fir0_sse(float *out, const float *in, const float *coefs);
  36 void ff_dca_lfe_fir1_sse(float *out, const float *in, const float *coefs);
  37 void ff_dca_lfe_fir0_fma3(float *out, const float *in, const float *coefs);
  38
  39 av_cold void ff_dcadsp_init_x86(DCADSPContext *s)
  40 {
  41     int cpu_flags = av_get_cpu_flags();
  42
  43     if (EXTERNAL_SSE(cpu_flags)) {
  44 #if ARCH_X86_32
  45         s->decode_hf = ff_decode_hf_sse;
  46 #endif
  47         s->lfe_fir[0]        = ff_dca_lfe_fir0_sse;
  48         s->lfe_fir[1]        = ff_dca_lfe_fir1_sse;
  49     }
  50
  51     if (EXTERNAL_SSE2(cpu_flags)) {
  52         s->decode_hf = ff_decode_hf_sse2;
  53     }
  54
  55     if (EXTERNAL_SSE4(cpu_flags)) {
  56         s->decode_hf = ff_decode_hf_sse4;
  57     }
  58
  59     if (EXTERNAL_FMA3(cpu_flags)) {
  60         s->lfe_fir[0]        = ff_dca_lfe_fir0_fma3;
  61     }
  62 }
  63
  64
  65 #define SYNTH_FILTER_FUNC(opt)                                                 \
  66 void ff_synth_filter_inner_##opt(float *synth_buf_ptr, float synth_buf2[32],   \
  67                                  const float window[512],                      \
  68                                  float out[32], intptr_t offset, float scale); \
  69 static void synth_filter_##opt(FFTContext *imdct,                              \
  70                                float *synth_buf_ptr, int *synth_buf_offset,    \
  71                                float synth_buf2[32], const float window[512],  \
  72                                float out[32], const float in[32], float scale) \
  73 {                                                                              \
  74     float *synth_buf= synth_buf_ptr + *synth_buf_offset;                       \
  75                                                                                \
  76     imdct->imdct_half(imdct, synth_buf, in);                                   \
  77                                                                                \
  78     ff_synth_filter_inner_##opt(synth_buf, synth_buf2, window,                 \
  79                                 out, *synth_buf_offset, scale);                \
  80                                                                                \
  81     *synth_buf_offset = (*synth_buf_offset - 32) & 511;                        \
  82 }                                                                              \
  83
  84 #if HAVE_YASM
  85 #if ARCH_X86_32
  86 SYNTH_FILTER_FUNC(sse)
  87 #endif
  88 SYNTH_FILTER_FUNC(sse2)
  89 SYNTH_FILTER_FUNC(avx)
  90 SYNTH_FILTER_FUNC(fma3)
  91 #endif /* HAVE_YASM */
  92
  93 av_cold void ff_synth_filter_init_x86(SynthFilterContext *s)
  94 {
  95 #if HAVE_YASM
  96     int cpu_flags = av_get_cpu_flags();
  97
  98 #if ARCH_X86_32
  99     if (EXTERNAL_SSE(cpu_flags)) {
 100         s->synth_filter_float = synth_filter_sse;
 101     }
 102 #endif
 103     if (EXTERNAL_SSE2(cpu_flags)) {
 104         s->synth_filter_float = synth_filter_sse2;
 105     }
 106     if (EXTERNAL_AVX_FAST(cpu_flags)) {
 107         s->synth_filter_float = synth_filter_avx;
 108     }
 109     if (EXTERNAL_FMA3(cpu_flags) && !(cpu_flags & AV_CPU_FLAG_AVXSLOW)) {
 110         s->synth_filter_float = synth_filter_fma3;
 111     }
 112 #endif /* HAVE_YASM */
 113 }