yading@11: /*
yading@11:  * audio conversion
yading@11:  * Copyright (c) 2006 Michael Niedermayer <michaelni@gmx.at>
yading@11:  *
yading@11:  * This file is part of FFmpeg.
yading@11:  *
yading@11:  * FFmpeg is free software; you can redistribute it and/or
yading@11:  * modify it under the terms of the GNU Lesser General Public
yading@11:  * License as published by the Free Software Foundation; either
yading@11:  * version 2.1 of the License, or (at your option) any later version.
yading@11:  *
yading@11:  * FFmpeg is distributed in the hope that it will be useful,
yading@11:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
yading@11:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
yading@11:  * Lesser General Public License for more details.
yading@11:  *
yading@11:  * You should have received a copy of the GNU Lesser General Public
yading@11:  * License along with FFmpeg; if not, write to the Free Software
yading@11:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
yading@11:  */
yading@11: 
yading@11: /**
yading@11:  * @file
yading@11:  * audio conversion
yading@11:  * @author Michael Niedermayer <michaelni@gmx.at>
yading@11:  */
yading@11: 
yading@11: #include "libavutil/avstring.h"
yading@11: #include "libavutil/avassert.h"
yading@11: #include "libavutil/libm.h"
yading@11: #include "libavutil/samplefmt.h"
yading@11: #include "audioconvert.h"
yading@11: 
yading@11: 
yading@11: #define CONV_FUNC_NAME(dst_fmt, src_fmt) conv_ ## src_fmt ## _to_ ## dst_fmt
yading@11: 
yading@11: //FIXME rounding ?
yading@11: #define CONV_FUNC(ofmt, otype, ifmt, expr)\
yading@11: static void CONV_FUNC_NAME(ofmt, ifmt)(uint8_t *po, const uint8_t *pi, int is, int os, uint8_t *end)\
yading@11: {\
yading@11:     uint8_t *end2 = end - 3*os;\
yading@11:     while(po < end2){\
yading@11:         *(otype*)po = expr; pi += is; po += os;\
yading@11:         *(otype*)po = expr; pi += is; po += os;\
yading@11:         *(otype*)po = expr; pi += is; po += os;\
yading@11:         *(otype*)po = expr; pi += is; po += os;\
yading@11:     }\
yading@11:     while(po < end){\
yading@11:         *(otype*)po = expr; pi += is; po += os;\
yading@11:     }\
yading@11: }
yading@11: 
yading@11: //FIXME put things below under ifdefs so we do not waste space for cases no codec will need
yading@11: CONV_FUNC(AV_SAMPLE_FMT_U8 , uint8_t, AV_SAMPLE_FMT_U8 ,  *(const uint8_t*)pi)
yading@11: CONV_FUNC(AV_SAMPLE_FMT_S16, int16_t, AV_SAMPLE_FMT_U8 , (*(const uint8_t*)pi - 0x80)<<8)
yading@11: CONV_FUNC(AV_SAMPLE_FMT_S32, int32_t, AV_SAMPLE_FMT_U8 , (*(const uint8_t*)pi - 0x80)<<24)
yading@11: CONV_FUNC(AV_SAMPLE_FMT_FLT, float  , AV_SAMPLE_FMT_U8 , (*(const uint8_t*)pi - 0x80)*(1.0f/ (1<<7)))
yading@11: CONV_FUNC(AV_SAMPLE_FMT_DBL, double , AV_SAMPLE_FMT_U8 , (*(const uint8_t*)pi - 0x80)*(1.0 / (1<<7)))
yading@11: CONV_FUNC(AV_SAMPLE_FMT_U8 , uint8_t, AV_SAMPLE_FMT_S16, (*(const int16_t*)pi>>8) + 0x80)
yading@11: CONV_FUNC(AV_SAMPLE_FMT_S16, int16_t, AV_SAMPLE_FMT_S16,  *(const int16_t*)pi)
yading@11: CONV_FUNC(AV_SAMPLE_FMT_S32, int32_t, AV_SAMPLE_FMT_S16,  *(const int16_t*)pi<<16)
yading@11: CONV_FUNC(AV_SAMPLE_FMT_FLT, float  , AV_SAMPLE_FMT_S16,  *(const int16_t*)pi*(1.0f/ (1<<15)))
yading@11: CONV_FUNC(AV_SAMPLE_FMT_DBL, double , AV_SAMPLE_FMT_S16,  *(const int16_t*)pi*(1.0 / (1<<15)))
yading@11: CONV_FUNC(AV_SAMPLE_FMT_U8 , uint8_t, AV_SAMPLE_FMT_S32, (*(const int32_t*)pi>>24) + 0x80)
yading@11: CONV_FUNC(AV_SAMPLE_FMT_S16, int16_t, AV_SAMPLE_FMT_S32,  *(const int32_t*)pi>>16)
yading@11: CONV_FUNC(AV_SAMPLE_FMT_S32, int32_t, AV_SAMPLE_FMT_S32,  *(const int32_t*)pi)
yading@11: CONV_FUNC(AV_SAMPLE_FMT_FLT, float  , AV_SAMPLE_FMT_S32,  *(const int32_t*)pi*(1.0f/ (1U<<31)))
yading@11: CONV_FUNC(AV_SAMPLE_FMT_DBL, double , AV_SAMPLE_FMT_S32,  *(const int32_t*)pi*(1.0 / (1U<<31)))
yading@11: CONV_FUNC(AV_SAMPLE_FMT_U8 , uint8_t, AV_SAMPLE_FMT_FLT, av_clip_uint8(  lrintf(*(const float*)pi * (1<<7)) + 0x80))
yading@11: CONV_FUNC(AV_SAMPLE_FMT_S16, int16_t, AV_SAMPLE_FMT_FLT, av_clip_int16(  lrintf(*(const float*)pi * (1<<15))))
yading@11: CONV_FUNC(AV_SAMPLE_FMT_S32, int32_t, AV_SAMPLE_FMT_FLT, av_clipl_int32(llrintf(*(const float*)pi * (1U<<31))))
yading@11: CONV_FUNC(AV_SAMPLE_FMT_FLT, float  , AV_SAMPLE_FMT_FLT, *(const float*)pi)
yading@11: CONV_FUNC(AV_SAMPLE_FMT_DBL, double , AV_SAMPLE_FMT_FLT, *(const float*)pi)
yading@11: CONV_FUNC(AV_SAMPLE_FMT_U8 , uint8_t, AV_SAMPLE_FMT_DBL, av_clip_uint8(  lrint(*(const double*)pi * (1<<7)) + 0x80))
yading@11: CONV_FUNC(AV_SAMPLE_FMT_S16, int16_t, AV_SAMPLE_FMT_DBL, av_clip_int16(  lrint(*(const double*)pi * (1<<15))))
yading@11: CONV_FUNC(AV_SAMPLE_FMT_S32, int32_t, AV_SAMPLE_FMT_DBL, av_clipl_int32(llrint(*(const double*)pi * (1U<<31))))
yading@11: CONV_FUNC(AV_SAMPLE_FMT_FLT, float  , AV_SAMPLE_FMT_DBL, *(const double*)pi)
yading@11: CONV_FUNC(AV_SAMPLE_FMT_DBL, double , AV_SAMPLE_FMT_DBL, *(const double*)pi)
yading@11: 
yading@11: #define FMT_PAIR_FUNC(out, in) [out + AV_SAMPLE_FMT_NB*in] = CONV_FUNC_NAME(out, in)
yading@11: 
yading@11: static conv_func_type * const fmt_pair_to_conv_functions[AV_SAMPLE_FMT_NB*AV_SAMPLE_FMT_NB] = {
yading@11:     FMT_PAIR_FUNC(AV_SAMPLE_FMT_U8 , AV_SAMPLE_FMT_U8 ),
yading@11:     FMT_PAIR_FUNC(AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_U8 ),
yading@11:     FMT_PAIR_FUNC(AV_SAMPLE_FMT_S32, AV_SAMPLE_FMT_U8 ),
yading@11:     FMT_PAIR_FUNC(AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_U8 ),
yading@11:     FMT_PAIR_FUNC(AV_SAMPLE_FMT_DBL, AV_SAMPLE_FMT_U8 ),
yading@11:     FMT_PAIR_FUNC(AV_SAMPLE_FMT_U8 , AV_SAMPLE_FMT_S16),
yading@11:     FMT_PAIR_FUNC(AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_S16),
yading@11:     FMT_PAIR_FUNC(AV_SAMPLE_FMT_S32, AV_SAMPLE_FMT_S16),
yading@11:     FMT_PAIR_FUNC(AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S16),
yading@11:     FMT_PAIR_FUNC(AV_SAMPLE_FMT_DBL, AV_SAMPLE_FMT_S16),
yading@11:     FMT_PAIR_FUNC(AV_SAMPLE_FMT_U8 , AV_SAMPLE_FMT_S32),
yading@11:     FMT_PAIR_FUNC(AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_S32),
yading@11:     FMT_PAIR_FUNC(AV_SAMPLE_FMT_S32, AV_SAMPLE_FMT_S32),
yading@11:     FMT_PAIR_FUNC(AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_S32),
yading@11:     FMT_PAIR_FUNC(AV_SAMPLE_FMT_DBL, AV_SAMPLE_FMT_S32),
yading@11:     FMT_PAIR_FUNC(AV_SAMPLE_FMT_U8 , AV_SAMPLE_FMT_FLT),
yading@11:     FMT_PAIR_FUNC(AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_FLT),
yading@11:     FMT_PAIR_FUNC(AV_SAMPLE_FMT_S32, AV_SAMPLE_FMT_FLT),
yading@11:     FMT_PAIR_FUNC(AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_FLT),
yading@11:     FMT_PAIR_FUNC(AV_SAMPLE_FMT_DBL, AV_SAMPLE_FMT_FLT),
yading@11:     FMT_PAIR_FUNC(AV_SAMPLE_FMT_U8 , AV_SAMPLE_FMT_DBL),
yading@11:     FMT_PAIR_FUNC(AV_SAMPLE_FMT_S16, AV_SAMPLE_FMT_DBL),
yading@11:     FMT_PAIR_FUNC(AV_SAMPLE_FMT_S32, AV_SAMPLE_FMT_DBL),
yading@11:     FMT_PAIR_FUNC(AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_DBL),
yading@11:     FMT_PAIR_FUNC(AV_SAMPLE_FMT_DBL, AV_SAMPLE_FMT_DBL),
yading@11: };
yading@11: 
yading@11: static void cpy1(uint8_t **dst, const uint8_t **src, int len){
yading@11:     memcpy(*dst, *src, len);
yading@11: }
yading@11: static void cpy2(uint8_t **dst, const uint8_t **src, int len){
yading@11:     memcpy(*dst, *src, 2*len);
yading@11: }
yading@11: static void cpy4(uint8_t **dst, const uint8_t **src, int len){
yading@11:     memcpy(*dst, *src, 4*len);
yading@11: }
yading@11: static void cpy8(uint8_t **dst, const uint8_t **src, int len){
yading@11:     memcpy(*dst, *src, 8*len);
yading@11: }
yading@11: 
yading@11: AudioConvert *swri_audio_convert_alloc(enum AVSampleFormat out_fmt,
yading@11:                                        enum AVSampleFormat in_fmt,
yading@11:                                        int channels, const int *ch_map,
yading@11:                                        int flags)
yading@11: {
yading@11:     AudioConvert *ctx;
yading@11:     conv_func_type *f = fmt_pair_to_conv_functions[av_get_packed_sample_fmt(out_fmt) + AV_SAMPLE_FMT_NB*av_get_packed_sample_fmt(in_fmt)];
yading@11: 
yading@11:     if (!f)
yading@11:         return NULL;
yading@11:     ctx = av_mallocz(sizeof(*ctx));
yading@11:     if (!ctx)
yading@11:         return NULL;
yading@11: 
yading@11:     if(channels == 1){
yading@11:          in_fmt = av_get_planar_sample_fmt( in_fmt);
yading@11:         out_fmt = av_get_planar_sample_fmt(out_fmt);
yading@11:     }
yading@11: 
yading@11:     ctx->channels = channels;
yading@11:     ctx->conv_f   = f;
yading@11:     ctx->ch_map   = ch_map;
yading@11:     if (in_fmt == AV_SAMPLE_FMT_U8 || in_fmt == AV_SAMPLE_FMT_U8P)
yading@11:         memset(ctx->silence, 0x80, sizeof(ctx->silence));
yading@11: 
yading@11:     if(out_fmt == in_fmt && !ch_map) {
yading@11:         switch(av_get_bytes_per_sample(in_fmt)){
yading@11:             case 1:ctx->simd_f = cpy1; break;
yading@11:             case 2:ctx->simd_f = cpy2; break;
yading@11:             case 4:ctx->simd_f = cpy4; break;
yading@11:             case 8:ctx->simd_f = cpy8; break;
yading@11:         }
yading@11:     }
yading@11: 
yading@11:     if(HAVE_YASM && HAVE_MMX) swri_audio_convert_init_x86(ctx, out_fmt, in_fmt, channels);
yading@11:     if(ARCH_ARM)              swri_audio_convert_init_arm(ctx, out_fmt, in_fmt, channels);
yading@11: 
yading@11:     return ctx;
yading@11: }
yading@11: 
yading@11: void swri_audio_convert_free(AudioConvert **ctx)
yading@11: {
yading@11:     av_freep(ctx);
yading@11: }
yading@11: 
yading@11: int swri_audio_convert(AudioConvert *ctx, AudioData *out, AudioData *in, int len)
yading@11: {
yading@11:     int ch;
yading@11:     int off=0;
yading@11:     const int os= (out->planar ? 1 :out->ch_count) *out->bps;
yading@11:     unsigned misaligned = 0;
yading@11: 
yading@11:     av_assert0(ctx->channels == out->ch_count);
yading@11: 
yading@11:     if (ctx->in_simd_align_mask) {
yading@11:         int planes = in->planar ? in->ch_count : 1;
yading@11:         unsigned m = 0;
yading@11:         for (ch = 0; ch < planes; ch++)
yading@11:             m |= (intptr_t)in->ch[ch];
yading@11:         misaligned |= m & ctx->in_simd_align_mask;
yading@11:     }
yading@11:     if (ctx->out_simd_align_mask) {
yading@11:         int planes = out->planar ? out->ch_count : 1;
yading@11:         unsigned m = 0;
yading@11:         for (ch = 0; ch < planes; ch++)
yading@11:             m |= (intptr_t)out->ch[ch];
yading@11:         misaligned |= m & ctx->out_simd_align_mask;
yading@11:     }
yading@11: 
yading@11:     //FIXME optimize common cases
yading@11: 
yading@11:     if(ctx->simd_f && !ctx->ch_map && !misaligned){
yading@11:         off = len&~15;
yading@11:         av_assert1(off>=0);
yading@11:         av_assert1(off<=len);
yading@11:         av_assert2(ctx->channels == SWR_CH_MAX || !in->ch[ctx->channels]);
yading@11:         if(off>0){
yading@11:             if(out->planar == in->planar){
yading@11:                 int planes = out->planar ? out->ch_count : 1;
yading@11:                 for(ch=0; ch<planes; ch++){
yading@11:                     ctx->simd_f(out->ch+ch, (const uint8_t **)in->ch+ch, off * (out->planar ? 1 :out->ch_count));
yading@11:                 }
yading@11:             }else{
yading@11:                 ctx->simd_f(out->ch, (const uint8_t **)in->ch, off);
yading@11:             }
yading@11:         }
yading@11:         if(off == len)
yading@11:             return 0;
yading@11:     }
yading@11: 
yading@11:     for(ch=0; ch<ctx->channels; ch++){
yading@11:         const int ich= ctx->ch_map ? ctx->ch_map[ch] : ch;
yading@11:         const int is= ich < 0 ? 0 : (in->planar ? 1 : in->ch_count) * in->bps;
yading@11:         const uint8_t *pi= ich < 0 ? ctx->silence : in->ch[ich];
yading@11:         uint8_t       *po= out->ch[ch];
yading@11:         uint8_t *end= po + os*len;
yading@11:         if(!po)
yading@11:             continue;
yading@11:         ctx->conv_f(po+off*os, pi+off*is, is, os, end);
yading@11:     }
yading@11:     return 0;
yading@11: }