FFmpeg: libavresample/x86/audio_convert

00001 /*
00002  * Copyright (c) 2012 Justin Ruggles <justin.ruggles@gmail.com>
00003  *
00004  * This file is part of Libav.
00005  *
00006  * Libav is free software; you can redistribute it and/or
00007  * modify it under the terms of the GNU Lesser General Public
00008  * License as published by the Free Software Foundation; either
00009  * version 2.1 of the License, or (at your option) any later version.
00010  *
00011  * Libav is distributed in the hope that it will be useful,
00012  * but WITHOUT ANY WARRANTY; without even the implied warranty of
00013  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
00014  * Lesser General Public License for more details.
00015  *
00016  * You should have received a copy of the GNU Lesser General Public
00017  * License along with Libav; if not, write to the Free Software
00018  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
00019  */
00020 
00021 #include "config.h"
00022 #include "libavutil/cpu.h"
00023 #include "libavutil/x86/cpu.h"
00024 #include "libavresample/audio_convert.h"
00025 
00026 /* flat conversions */
00027 
00028 extern void ff_conv_s16_to_s32_sse2(int16_t *dst, const int32_t *src, int len);
00029 
00030 extern void ff_conv_s16_to_flt_sse2(float *dst, const int16_t *src, int len);
00031 extern void ff_conv_s16_to_flt_sse4(float *dst, const int16_t *src, int len);
00032 
00033 extern void ff_conv_s32_to_s16_mmx (int16_t *dst, const int32_t *src, int len);
00034 extern void ff_conv_s32_to_s16_sse2(int16_t *dst, const int32_t *src, int len);
00035 
00036 extern void ff_conv_s32_to_flt_sse2(float *dst, const int32_t *src, int len);
00037 extern void ff_conv_s32_to_flt_avx (float *dst, const int32_t *src, int len);
00038 
00039 extern void ff_conv_flt_to_s16_sse2(int16_t *dst, const float *src, int len);
00040 
00041 extern void ff_conv_flt_to_s32_sse2(int32_t *dst, const float *src, int len);
00042 extern void ff_conv_flt_to_s32_avx (int32_t *dst, const float *src, int len);
00043 
00044 /* interleave conversions */
00045 
00046 extern void ff_conv_s16p_to_s16_2ch_sse2(int16_t *dst, int16_t *const *src,
00047                                          int len, int channels);
00048 extern void ff_conv_s16p_to_s16_2ch_avx (int16_t *dst, int16_t *const *src,
00049                                          int len, int channels);
00050 
00051 extern void ff_conv_s16p_to_s16_6ch_sse2(int16_t *dst, int16_t *const *src,
00052                                          int len, int channels);
00053 extern void ff_conv_s16p_to_s16_6ch_sse2slow(int16_t *dst, int16_t *const *src,
00054                                              int len, int channels);
00055 extern void ff_conv_s16p_to_s16_6ch_avx (int16_t *dst, int16_t *const *src,
00056                                          int len, int channels);
00057 
00058 extern void ff_conv_s16p_to_flt_2ch_sse2(float *dst, int16_t *const *src,
00059                                          int len, int channels);
00060 extern void ff_conv_s16p_to_flt_2ch_avx (float *dst, int16_t *const *src,
00061                                          int len, int channels);
00062 
00063 extern void ff_conv_s16p_to_flt_6ch_sse2 (float *dst, int16_t *const *src,
00064                                           int len, int channels);
00065 extern void ff_conv_s16p_to_flt_6ch_ssse3(float *dst, int16_t *const *src,
00066                                          int len, int channels);
00067 extern void ff_conv_s16p_to_flt_6ch_avx  (float *dst, int16_t *const *src,
00068                                           int len, int channels);
00069 
00070 extern void ff_conv_fltp_to_s16_2ch_sse2 (int16_t *dst, float *const *src,
00071                                           int len, int channels);
00072 extern void ff_conv_fltp_to_s16_2ch_ssse3(int16_t *dst, float *const *src,
00073                                           int len, int channels);
00074 
00075 extern void ff_conv_fltp_to_s16_6ch_sse (int16_t *dst, float *const *src,
00076                                          int len, int channels);
00077 extern void ff_conv_fltp_to_s16_6ch_sse2(int16_t *dst, float *const *src,
00078                                          int len, int channels);
00079 extern void ff_conv_fltp_to_s16_6ch_avx (int16_t *dst, float *const *src,
00080                                          int len, int channels);
00081 
00082 extern void ff_conv_fltp_to_flt_2ch_sse(float *dst, float *const *src, int len,
00083                                         int channels);
00084 extern void ff_conv_fltp_to_flt_2ch_avx(float *dst, float *const *src, int len,
00085                                         int channels);
00086 
00087 extern void ff_conv_fltp_to_flt_6ch_mmx (float *dst, float *const *src, int len,
00088                                          int channels);
00089 extern void ff_conv_fltp_to_flt_6ch_sse4(float *dst, float *const *src, int len,
00090                                          int channels);
00091 extern void ff_conv_fltp_to_flt_6ch_avx (float *dst, float *const *src, int len,
00092                                          int channels);
00093 
00094 /* deinterleave conversions */
00095 
00096 extern void ff_conv_s16_to_s16p_2ch_sse2(int16_t *const *dst, int16_t *src,
00097                                          int len, int channels);
00098 extern void ff_conv_s16_to_s16p_2ch_ssse3(int16_t *const *dst, int16_t *src,
00099                                           int len, int channels);
00100 extern void ff_conv_s16_to_s16p_2ch_avx (int16_t *const *dst, int16_t *src,
00101                                          int len, int channels);
00102 
00103 extern void ff_conv_s16_to_s16p_6ch_sse2 (int16_t *const *dst, int16_t *src,
00104                                           int len, int channels);
00105 extern void ff_conv_s16_to_s16p_6ch_ssse3(int16_t *const *dst, int16_t *src,
00106                                           int len, int channels);
00107 extern void ff_conv_s16_to_s16p_6ch_avx  (int16_t *const *dst, int16_t *src,
00108                                           int len, int channels);
00109 
00110 extern void ff_conv_s16_to_fltp_2ch_sse2(float *const *dst, int16_t *src,
00111                                          int len, int channels);
00112 extern void ff_conv_s16_to_fltp_2ch_avx (float *const *dst, int16_t *src,
00113                                          int len, int channels);
00114 
00115 extern void ff_conv_s16_to_fltp_6ch_sse2 (float *const *dst, int16_t *src,
00116                                           int len, int channels);
00117 extern void ff_conv_s16_to_fltp_6ch_ssse3(float *const *dst, int16_t *src,
00118                                           int len, int channels);
00119 extern void ff_conv_s16_to_fltp_6ch_sse4 (float *const *dst, int16_t *src,
00120                                           int len, int channels);
00121 extern void ff_conv_s16_to_fltp_6ch_avx  (float *const *dst, int16_t *src,
00122                                           int len, int channels);
00123 
00124 extern void ff_conv_flt_to_s16p_2ch_sse2(int16_t *const *dst, float *src,
00125                                          int len, int channels);
00126 extern void ff_conv_flt_to_s16p_2ch_avx (int16_t *const *dst, float *src,
00127                                          int len, int channels);
00128 
00129 extern void ff_conv_flt_to_s16p_6ch_sse2 (int16_t *const *dst, float *src,
00130                                           int len, int channels);
00131 extern void ff_conv_flt_to_s16p_6ch_ssse3(int16_t *const *dst, float *src,
00132                                           int len, int channels);
00133 extern void ff_conv_flt_to_s16p_6ch_avx  (int16_t *const *dst, float *src,
00134                                           int len, int channels);
00135 
00136 extern void ff_conv_flt_to_fltp_2ch_sse(float *const *dst, float *src, int len,
00137                                         int channels);
00138 extern void ff_conv_flt_to_fltp_2ch_avx(float *const *dst, float *src, int len,
00139                                         int channels);
00140 
00141 extern void ff_conv_flt_to_fltp_6ch_sse2(float *const *dst, float *src, int len,
00142                                          int channels);
00143 extern void ff_conv_flt_to_fltp_6ch_avx (float *const *dst, float *src, int len,
00144                                          int channels);
00145 
00146 av_cold void ff_audio_convert_init_x86(AudioConvert *ac)
00147 {
00148     int mm_flags = av_get_cpu_flags();
00149 
00150     if (EXTERNAL_MMX(mm_flags)) {
00151         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_S32,
00152                                   0, 1, 8, "MMX", ff_conv_s32_to_s16_mmx);
00153         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_FLTP,
00154                                   6, 1, 4, "MMX", ff_conv_fltp_to_flt_6ch_mmx);
00155     }
00156     if (EXTERNAL_SSE(mm_flags)) {
00157         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_FLTP,
00158                                   6, 1, 2, "SSE", ff_conv_fltp_to_s16_6ch_sse);
00159         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_FLTP,
00160                                   2, 16, 8, "SSE", ff_conv_fltp_to_flt_2ch_sse);
00161         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_FLT,
00162                                   2, 16, 4, "SSE", ff_conv_flt_to_fltp_2ch_sse);
00163     }
00164     if (EXTERNAL_SSE2(mm_flags)) {
00165         if (!(mm_flags & AV_CPU_FLAG_SSE2SLOW)) {
00166             ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_S32,
00167                                       0, 16, 16, "SSE2", ff_conv_s32_to_s16_sse2);
00168             ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_S16P,
00169                                       6, 16, 8, "SSE2", ff_conv_s16p_to_s16_6ch_sse2);
00170             ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_FLTP,
00171                                       6, 16, 4, "SSE2", ff_conv_fltp_to_s16_6ch_sse2);
00172         } else {
00173             ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_S16P,
00174                                       6, 1, 4, "SSE2SLOW", ff_conv_s16p_to_s16_6ch_sse2slow);
00175         }
00176         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S32, AV_SAMPLE_FMT_S16,
00177                                   0, 16, 8, "SSE2", ff_conv_s16_to_s32_sse2);
00178         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16,
00179                                   0, 16, 8, "SSE2", ff_conv_s16_to_flt_sse2);
00180         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S32,
00181                                   0, 16, 8, "SSE2", ff_conv_s32_to_flt_sse2);
00182         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_FLT,
00183                                   0, 16, 16, "SSE2", ff_conv_flt_to_s16_sse2);
00184         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S32, AV_SAMPLE_FMT_FLT,
00185                                   0, 16, 16, "SSE2", ff_conv_flt_to_s32_sse2);
00186         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_S16P,
00187                                   2, 16, 16, "SSE2", ff_conv_s16p_to_s16_2ch_sse2);
00188         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16P,
00189                                   2, 16, 8, "SSE2", ff_conv_s16p_to_flt_2ch_sse2);
00190         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16P,
00191                                   6, 16, 4, "SSE2", ff_conv_s16p_to_flt_6ch_sse2);
00192         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_FLTP,
00193                                   2, 16, 4, "SSE2", ff_conv_fltp_to_s16_2ch_sse2);
00194         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16P, AV_SAMPLE_FMT_S16,
00195                                   2, 16, 8, "SSE2", ff_conv_s16_to_s16p_2ch_sse2);
00196         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16P, AV_SAMPLE_FMT_S16,
00197                                   6, 16, 4, "SSE2", ff_conv_s16_to_s16p_6ch_sse2);
00198         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_S16,
00199                                   2, 16, 8, "SSE2", ff_conv_s16_to_fltp_2ch_sse2);
00200         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_S16,
00201                                   6, 16, 4, "SSE2", ff_conv_s16_to_fltp_6ch_sse2);
00202         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16P, AV_SAMPLE_FMT_FLT,
00203                                   2, 16, 8, "SSE2", ff_conv_flt_to_s16p_2ch_sse2);
00204         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16P, AV_SAMPLE_FMT_FLT,
00205                                   6, 16, 4, "SSE2", ff_conv_flt_to_s16p_6ch_sse2);
00206         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_FLT,
00207                                   6, 16, 4, "SSE2", ff_conv_flt_to_fltp_6ch_sse2);
00208     }
00209     if (EXTERNAL_SSSE3(mm_flags)) {
00210         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16P,
00211                                   6, 16, 4, "SSSE3", ff_conv_s16p_to_flt_6ch_ssse3);
00212         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_FLTP,
00213                                   2, 16, 4, "SSSE3", ff_conv_fltp_to_s16_2ch_ssse3);
00214         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16P, AV_SAMPLE_FMT_S16,
00215                                   2, 16, 8, "SSSE3", ff_conv_s16_to_s16p_2ch_ssse3);
00216         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16P, AV_SAMPLE_FMT_S16,
00217                                   6, 16, 4, "SSSE3", ff_conv_s16_to_s16p_6ch_ssse3);
00218         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_S16,
00219                                   6, 16, 4, "SSSE3", ff_conv_s16_to_fltp_6ch_ssse3);
00220         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16P, AV_SAMPLE_FMT_FLT,
00221                                   6, 16, 4, "SSSE3", ff_conv_flt_to_s16p_6ch_ssse3);
00222     }
00223     if (EXTERNAL_SSE4(mm_flags)) {
00224         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16,
00225                                   0, 16, 8, "SSE4", ff_conv_s16_to_flt_sse4);
00226         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_FLTP,
00227                                   6, 16, 4, "SSE4", ff_conv_fltp_to_flt_6ch_sse4);
00228     }
00229     if (EXTERNAL_AVX(mm_flags)) {
00230         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S32,
00231                                   0, 32, 16, "AVX", ff_conv_s32_to_flt_avx);
00232         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S32, AV_SAMPLE_FMT_FLT,
00233                                   0, 32, 32, "AVX", ff_conv_flt_to_s32_avx);
00234         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_S16P,
00235                                   2, 16, 16, "AVX", ff_conv_s16p_to_s16_2ch_avx);
00236         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_S16P,
00237                                   6, 16, 8, "AVX", ff_conv_s16p_to_s16_6ch_avx);
00238         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16P,
00239                                   2, 16, 8, "AVX", ff_conv_s16p_to_flt_2ch_avx);
00240         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16P,
00241                                   6, 16, 4, "AVX", ff_conv_s16p_to_flt_6ch_avx);
00242         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_FLTP,
00243                                   6, 16, 4, "AVX", ff_conv_fltp_to_s16_6ch_avx);
00244         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_FLTP,
00245                                   6, 16, 4, "AVX", ff_conv_fltp_to_flt_6ch_avx);
00246         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16P, AV_SAMPLE_FMT_S16,
00247                                   2, 16, 8, "AVX", ff_conv_s16_to_s16p_2ch_avx);
00248         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16P, AV_SAMPLE_FMT_S16,
00249                                   6, 16, 4, "AVX", ff_conv_s16_to_s16p_6ch_avx);
00250         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_S16,
00251                                   2, 16, 8, "AVX", ff_conv_s16_to_fltp_2ch_avx);
00252         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_S16,
00253                                   6, 16, 4, "AVX", ff_conv_s16_to_fltp_6ch_avx);
00254         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16P, AV_SAMPLE_FMT_FLT,
00255                                   2, 16, 8, "AVX", ff_conv_flt_to_s16p_2ch_avx);
00256         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_S16P, AV_SAMPLE_FMT_FLT,
00257                                   6, 16, 4, "AVX", ff_conv_flt_to_s16p_6ch_avx);
00258         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_FLT,
00259                                   2, 16, 4, "AVX", ff_conv_flt_to_fltp_2ch_avx);
00260         ff_audio_convert_set_func(ac, AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_FLT,
00261                                   6, 16, 4, "AVX", ff_conv_flt_to_fltp_6ch_avx);
00262     }
00263 }
libavresample/x86/audio_convert_init.c
