mpv/audio/reorder_ch.c

1468 lines
48 KiB
C

/*
* common functions for reordering audio channels
*
* Copyright (C) 2007 Ulion <ulion A gmail P com>
*
* This file is part of MPlayer.
*
* MPlayer is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* MPlayer is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License along
* with MPlayer; if not, write to the Free Software Foundation, Inc.,
* 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
*/
#include <stdio.h>
#include <stdlib.h>
#include <inttypes.h>
#include <string.h>
#include "audio/reorder_ch.h"
#ifdef TEST
#define mp_msg(mod,lev, fmt, args... ) printf( fmt, ## args )
#else
#include "core/mp_msg.h"
#endif
#define REORDER_COPY_5(DEST,SRC,SAMPLES,S0,S1,S2,S3,S4) \
for (i = 0; i < SAMPLES; i += 5) {\
DEST[i] = SRC[i+S0];\
DEST[i+1] = SRC[i+S1];\
DEST[i+2] = SRC[i+S2];\
DEST[i+3] = SRC[i+S3];\
DEST[i+4] = SRC[i+S4];\
}
static int reorder_copy_5ch(void *dest, const void *src,
unsigned int samples, unsigned int samplesize,
int s0, int s1, int s2, int s3, int s4)
{
int i;
switch (samplesize) {
case 1:
{
int8_t *dest_8 = dest;
const int8_t *src_8 = src;
REORDER_COPY_5(dest_8,src_8,samples,s0,s1,s2,s3,s4);
break;
}
case 2:
{
int16_t *dest_16 = dest;
const int16_t *src_16 = src;
REORDER_COPY_5(dest_16,src_16,samples,s0,s1,s2,s3,s4);
break;
}
case 3:
{
int8_t *dest_8 = dest;
const int8_t *src_8 = src;
for (i = 0; i < samples * 3; i += 15) {
dest_8[i] = src_8[i+s0*3];
dest_8[i+1] = src_8[i+s0*3+1];
dest_8[i+2] = src_8[i+s0*3+2];
dest_8[i+3] = src_8[i+s1*3];
dest_8[i+4] = src_8[i+s1*3+1];
dest_8[i+5] = src_8[i+s1*3+2];
dest_8[i+6] = src_8[i+s2*3];
dest_8[i+7] = src_8[i+s2*3+1];
dest_8[i+8] = src_8[i+s2*3+2];
dest_8[i+9] = src_8[i+s3*3];
dest_8[i+10] = src_8[i+s3*3+1];
dest_8[i+11] = src_8[i+s3*3+2];
dest_8[i+12] = src_8[i+s4*3];
dest_8[i+13] = src_8[i+s4*3+1];
dest_8[i+14] = src_8[i+s4*3+2];
}
break;
}
case 4:
{
int32_t *dest_32 = dest;
const int32_t *src_32 = src;
REORDER_COPY_5(dest_32,src_32,samples,s0,s1,s2,s3,s4);
break;
}
case 8:
{
int64_t *dest_64 = dest;
const int64_t *src_64 = src;
REORDER_COPY_5(dest_64,src_64,samples,s0,s1,s2,s3,s4);
break;
}
default:
mp_msg(MSGT_GLOBAL, MSGL_WARN,
"[reorder_ch] Unsupported sample size: %d, please "
"report this error on the MPlayer mailing list.\n",samplesize);
return 0;
}
return 1;
}
#define REORDER_COPY_6(DEST,SRC,SAMPLES,S0,S1,S2,S3,S4,S5) \
for (i = 0; i < SAMPLES; i += 6) {\
DEST[i] = SRC[i+S0];\
DEST[i+1] = SRC[i+S1];\
DEST[i+2] = SRC[i+S2];\
DEST[i+3] = SRC[i+S3];\
DEST[i+4] = SRC[i+S4];\
DEST[i+5] = SRC[i+S5];\
}
static int reorder_copy_6ch(void *dest, const void *src,
unsigned int samples, uint8_t samplesize,
int s0, int s1, int s2, int s3, int s4, int s5)
{
int i;
switch (samplesize) {
case 1:
{
int8_t *dest_8 = dest;
const int8_t *src_8 = src;
REORDER_COPY_6(dest_8,src_8,samples,s0,s1,s2,s3,s4,s5);
break;
}
case 2:
{
int16_t *dest_16 = dest;
const int16_t *src_16 = src;
REORDER_COPY_6(dest_16,src_16,samples,s0,s1,s2,s3,s4,s5);
break;
}
case 3:
{
int8_t *dest_8 = dest;
const int8_t *src_8 = src;
for (i = 0; i < samples * 3; i += 18) {
dest_8[i] = src_8[i+s0*3];
dest_8[i+1] = src_8[i+s0*3+1];
dest_8[i+2] = src_8[i+s0*3+2];
dest_8[i+3] = src_8[i+s1*3];
dest_8[i+4] = src_8[i+s1*3+1];
dest_8[i+5] = src_8[i+s1*3+2];
dest_8[i+6] = src_8[i+s2*3];
dest_8[i+7] = src_8[i+s2*3+1];
dest_8[i+8] = src_8[i+s2*3+2];
dest_8[i+9] = src_8[i+s3*3];
dest_8[i+10] = src_8[i+s3*3+1];
dest_8[i+11] = src_8[i+s3*3+2];
dest_8[i+12] = src_8[i+s4*3];
dest_8[i+13] = src_8[i+s4*3+1];
dest_8[i+14] = src_8[i+s4*3+2];
dest_8[i+15] = src_8[i+s5*3];
dest_8[i+16] = src_8[i+s5*3+1];
dest_8[i+17] = src_8[i+s5*3+2];
}
break;
}
case 4:
{
int32_t *dest_32 = dest;
const int32_t *src_32 = src;
REORDER_COPY_6(dest_32,src_32,samples,s0,s1,s2,s3,s4,s5);
break;
}
case 8:
{
int64_t *dest_64 = dest;
const int64_t *src_64 = src;
REORDER_COPY_6(dest_64,src_64,samples,s0,s1,s2,s3,s4,s5);
break;
}
default:
mp_msg(MSGT_GLOBAL, MSGL_WARN,
"[reorder_ch] Unsupported sample size: %d, please "
"report this error on the MPlayer mailing list.\n",samplesize);
return 0;
}
return 1;
}
#define REORDER_COPY_8(DEST,SRC,SAMPLES,S0,S1,S2,S3,S4,S5,S6,S7) \
for (i = 0; i < SAMPLES; i += 8) {\
DEST[i] = SRC[i+S0];\
DEST[i+1] = SRC[i+S1];\
DEST[i+2] = SRC[i+S2];\
DEST[i+3] = SRC[i+S3];\
DEST[i+4] = SRC[i+S4];\
DEST[i+5] = SRC[i+S5];\
DEST[i+6] = SRC[i+S6];\
DEST[i+7] = SRC[i+S7];\
}
static int reorder_copy_8ch(void *dest, const void *src,
unsigned int samples, uint8_t samplesize,
int s0, int s1, int s2, int s3,
int s4, int s5, int s6, int s7)
{
int i;
switch (samplesize) {
case 1:
{
int8_t *dest_8 = dest;
const int8_t *src_8 = src;
REORDER_COPY_8(dest_8,src_8,samples,s0,s1,s2,s3,s4,s5,s6,s7);
break;
}
case 2:
{
int16_t *dest_16 = dest;
const int16_t *src_16 = src;
REORDER_COPY_8(dest_16,src_16,samples,s0,s1,s2,s3,s4,s5,s6,s7);
break;
}
case 3:
{
int8_t *dest_8 = dest;
const int8_t *src_8 = src;
for (i = 0; i < samples * 3; i += 24) {
dest_8[i] = src_8[i+s0*3];
dest_8[i+1] = src_8[i+s0*3+1];
dest_8[i+2] = src_8[i+s0*3+2];
dest_8[i+3] = src_8[i+s1*3];
dest_8[i+4] = src_8[i+s1*3+1];
dest_8[i+5] = src_8[i+s1*3+2];
dest_8[i+6] = src_8[i+s2*3];
dest_8[i+7] = src_8[i+s2*3+1];
dest_8[i+8] = src_8[i+s2*3+2];
dest_8[i+9] = src_8[i+s3*3];
dest_8[i+10] = src_8[i+s3*3+1];
dest_8[i+11] = src_8[i+s3*3+2];
dest_8[i+12] = src_8[i+s4*3];
dest_8[i+13] = src_8[i+s4*3+1];
dest_8[i+14] = src_8[i+s4*3+2];
dest_8[i+15] = src_8[i+s5*3];
dest_8[i+16] = src_8[i+s5*3+1];
dest_8[i+17] = src_8[i+s5*3+2];
dest_8[i+18] = src_8[i+s6*3];
dest_8[i+19] = src_8[i+s6*3+1];
dest_8[i+20] = src_8[i+s6*3+2];
dest_8[i+21] = src_8[i+s7*3];
dest_8[i+22] = src_8[i+s7*3+1];
dest_8[i+23] = src_8[i+s7*3+2];
}
break;
}
case 4:
{
int32_t *dest_32 = dest;
const int32_t *src_32 = src;
REORDER_COPY_8(dest_32,src_32,samples,s0,s1,s2,s3,s4,s5,s6,s7);
break;
}
case 8:
{
int64_t *dest_64 = dest;
const int64_t *src_64 = src;
REORDER_COPY_8(dest_64,src_64,samples,s0,s1,s2,s3,s4,s5,s6,s7);
break;
}
default:
mp_msg(MSGT_GLOBAL, MSGL_WARN,
"[reorder_ch] Unsupported sample size: %d, please "
"report this error on the MPlayer mailing list.\n",samplesize);
return 0;
}
return 1;
}
void reorder_channel_copy(void *src,
int src_layout,
void *dest,
int dest_layout,
int samples,
int samplesize)
{
if (dest_layout==src_layout) {
memcpy(dest, src, samples*samplesize);
return;
}
if (!AF_IS_SAME_CH_NUM(dest_layout,src_layout)) {
mp_msg(MSGT_GLOBAL, MSGL_WARN, "[reorder_ch] different channel count "
"between src and dest: %x, %x\n",
AF_GET_CH_NUM_WITH_LFE(src_layout),
AF_GET_CH_NUM_WITH_LFE(dest_layout));
return;
}
switch ((src_layout<<16)|dest_layout) {
// AF_CHANNEL_LAYOUT_5_0_A L R C Ls Rs
// AF_CHANNEL_LAYOUT_5_0_B L R Ls Rs C
// AF_CHANNEL_LAYOUT_5_0_C L C R Ls Rs
// AF_CHANNEL_LAYOUT_5_0_D C L R Ls Rs
case AF_CHANNEL_LAYOUT_5_0_A << 16 | AF_CHANNEL_LAYOUT_5_0_B:
reorder_copy_5ch(dest, src, samples, samplesize, 0, 1, 3, 4, 2);
break;
case AF_CHANNEL_LAYOUT_5_0_A << 16 | AF_CHANNEL_LAYOUT_5_0_C:
reorder_copy_5ch(dest, src, samples, samplesize, 0, 2, 1, 3, 4);
break;
case AF_CHANNEL_LAYOUT_5_0_A << 16 | AF_CHANNEL_LAYOUT_5_0_D:
reorder_copy_5ch(dest, src, samples, samplesize, 2, 0, 1, 3, 4);
break;
case AF_CHANNEL_LAYOUT_5_0_B << 16 | AF_CHANNEL_LAYOUT_5_0_A:
reorder_copy_5ch(dest, src, samples, samplesize, 0, 1, 4, 2, 3);
break;
case AF_CHANNEL_LAYOUT_5_0_B << 16 | AF_CHANNEL_LAYOUT_5_0_C:
reorder_copy_5ch(dest, src, samples, samplesize, 0, 4, 1, 2, 3);
break;
case AF_CHANNEL_LAYOUT_5_0_B << 16 | AF_CHANNEL_LAYOUT_5_0_D:
reorder_copy_5ch(dest, src, samples, samplesize, 4, 0, 1, 2, 3);
break;
case AF_CHANNEL_LAYOUT_5_0_C << 16 | AF_CHANNEL_LAYOUT_5_0_A:
reorder_copy_5ch(dest, src, samples, samplesize, 0, 2, 1, 3, 4);
break;
case AF_CHANNEL_LAYOUT_5_0_C << 16 | AF_CHANNEL_LAYOUT_5_0_B:
reorder_copy_5ch(dest, src, samples, samplesize, 0, 2, 3, 4, 1);
break;
case AF_CHANNEL_LAYOUT_5_0_C << 16 | AF_CHANNEL_LAYOUT_5_0_D:
reorder_copy_5ch(dest, src, samples, samplesize, 1, 0, 2, 3, 4);
break;
case AF_CHANNEL_LAYOUT_5_0_D << 16 | AF_CHANNEL_LAYOUT_5_0_A:
reorder_copy_5ch(dest, src, samples, samplesize, 1, 2, 0, 3, 4);
break;
case AF_CHANNEL_LAYOUT_5_0_D << 16 | AF_CHANNEL_LAYOUT_5_0_B:
reorder_copy_5ch(dest, src, samples, samplesize, 1, 2, 3, 4, 0);
break;
case AF_CHANNEL_LAYOUT_5_0_D << 16 | AF_CHANNEL_LAYOUT_5_0_C:
reorder_copy_5ch(dest, src, samples, samplesize, 1, 0, 2, 3, 4);
break;
// AF_CHANNEL_LAYOUT_5_1_A L R C LFE Ls Rs
// AF_CHANNEL_LAYOUT_5_1_B L R Ls Rs C LFE
// AF_CHANNEL_LAYOUT_5_1_C L C R Ls Rs LFE
// AF_CHANNEL_LAYOUT_5_1_D C L R Ls Rs LFE
// AF_CHANNEL_LAYOUT_5_1_E LFE L C R Ls Rs
case AF_CHANNEL_LAYOUT_5_1_A << 16 | AF_CHANNEL_LAYOUT_5_1_B:
reorder_copy_6ch(dest, src, samples, samplesize, 0, 1, 4, 5, 2, 3);
break;
case AF_CHANNEL_LAYOUT_5_1_A << 16 | AF_CHANNEL_LAYOUT_5_1_C:
reorder_copy_6ch(dest, src, samples, samplesize, 0, 2, 1, 4, 5, 3);
break;
case AF_CHANNEL_LAYOUT_5_1_A << 16 | AF_CHANNEL_LAYOUT_5_1_D:
reorder_copy_6ch(dest, src, samples, samplesize, 2, 0, 1, 4, 5, 3);
break;
case AF_CHANNEL_LAYOUT_5_1_B << 16 | AF_CHANNEL_LAYOUT_5_1_A:
reorder_copy_6ch(dest, src, samples, samplesize, 0, 1, 4, 5, 2, 3);
break;
case AF_CHANNEL_LAYOUT_5_1_B << 16 | AF_CHANNEL_LAYOUT_5_1_C:
reorder_copy_6ch(dest, src, samples, samplesize, 0, 4, 1, 2, 3, 5);
break;
case AF_CHANNEL_LAYOUT_5_1_B << 16 | AF_CHANNEL_LAYOUT_5_1_D:
reorder_copy_6ch(dest, src, samples, samplesize, 4, 0, 1, 2, 3, 5);
break;
case AF_CHANNEL_LAYOUT_5_1_B << 16 | AF_CHANNEL_LAYOUT_5_1_E:
reorder_copy_6ch(dest, src, samples, samplesize, 5, 0, 4, 1, 2, 3);
break;
case AF_CHANNEL_LAYOUT_5_1_C << 16 | AF_CHANNEL_LAYOUT_5_1_A:
reorder_copy_6ch(dest, src, samples, samplesize, 0, 2, 1, 5, 3, 4);
break;
case AF_CHANNEL_LAYOUT_5_1_C << 16 | AF_CHANNEL_LAYOUT_5_1_B:
reorder_copy_6ch(dest, src, samples, samplesize, 0, 2, 3, 4, 1, 5);
break;
case AF_CHANNEL_LAYOUT_5_1_C << 16 | AF_CHANNEL_LAYOUT_5_1_D:
reorder_copy_6ch(dest, src, samples, samplesize, 1, 0, 2, 3, 4, 5);
break;
case AF_CHANNEL_LAYOUT_5_1_D << 16 | AF_CHANNEL_LAYOUT_5_1_A:
reorder_copy_6ch(dest, src, samples, samplesize, 1, 2, 0, 5, 3, 4);
break;
case AF_CHANNEL_LAYOUT_5_1_D << 16 | AF_CHANNEL_LAYOUT_5_1_B:
reorder_copy_6ch(dest, src, samples, samplesize, 1, 2, 3, 4, 0, 5);
break;
case AF_CHANNEL_LAYOUT_5_1_D << 16 | AF_CHANNEL_LAYOUT_5_1_C:
reorder_copy_6ch(dest, src, samples, samplesize, 1, 0, 2, 3, 4, 5);
break;
case AF_CHANNEL_LAYOUT_5_1_E << 16 | AF_CHANNEL_LAYOUT_5_1_B:
reorder_copy_6ch(dest, src, samples, samplesize, 1, 3, 4, 5, 2, 0);
break;
case AF_CHANNEL_LAYOUT_5_1_F << 16 | AF_CHANNEL_LAYOUT_5_1_B:
reorder_copy_6ch(dest, src, samples, samplesize, 1, 2, 4, 5, 0, 3);
break;
// AF_CHANNEL_LAYOUT_7_1_A L R C LFE Ls Rs Rls Rrs
// AF_CHANNEL_LAYOUT_7_1_B L R Ls Rs C LFE Rls Rrs
// AF_CHANNEL_LAYOUT_7_1_D C L R Ls Rs Rls Rrs LFE
case AF_CHANNEL_LAYOUT_7_1_A << 16 | AF_CHANNEL_LAYOUT_7_1_B:
case AF_CHANNEL_LAYOUT_7_1_B << 16 | AF_CHANNEL_LAYOUT_7_1_A:
reorder_copy_8ch(dest, src, samples, samplesize, 0, 1, 4, 5, 2, 3, 6, 7);
break;
case AF_CHANNEL_LAYOUT_7_1_D << 16 | AF_CHANNEL_LAYOUT_7_1_B:
reorder_copy_8ch(dest, src, samples, samplesize, 1, 2, 3, 4, 0, 7, 5, 6);
break;
default:
mp_msg(MSGT_GLOBAL, MSGL_WARN, "[reorder_channel_copy] unsupport "
"from %x to %x, %d * %d\n", src_layout, dest_layout,
samples, samplesize);
memcpy(dest, src, samples*samplesize);
}
}
#define REORDER_SELF_SWAP_2(SRC,TMP,SAMPLES,CHNUM,S0,S1) \
for (i = 0; i < SAMPLES; i += CHNUM) {\
TMP = SRC[i+S0];\
SRC[i+S0] = SRC[i+S1];\
SRC[i+S1] = TMP;\
}
static int reorder_self_2(void *src, unsigned int samples,
unsigned int samplesize, unsigned int chnum,
int s0, int s1)
{
int i;
switch (samplesize) {
case 1:
{
int8_t *src_8 = src;
int8_t tmp;
if (chnum==6) {
REORDER_SELF_SWAP_2(src_8,tmp,samples,6,s0,s1);
}
else if (chnum==8) {
REORDER_SELF_SWAP_2(src_8,tmp,samples,8,s0,s1);
}
else {
REORDER_SELF_SWAP_2(src_8,tmp,samples,5,s0,s1);
}
break;
}
case 2:
{
int16_t *src_16 = src;
int16_t tmp;
if (chnum==6) {
REORDER_SELF_SWAP_2(src_16,tmp,samples,6,s0,s1);
}
else if (chnum==3) {
REORDER_SELF_SWAP_2(src_16,tmp,samples,3,s0,s1);
}
else if (chnum==4) {
REORDER_SELF_SWAP_2(src_16,tmp,samples,3,s0,s1);
}
else {
REORDER_SELF_SWAP_2(src_16,tmp,samples,5,s0,s1);
}
break;
}
case 3:
{
int8_t *src_8 = src;
int8_t tmp0, tmp1, tmp2;
for (i = 0; i < samples * 3; i += chnum * 3) {
tmp0 = src_8[i+s0*3];
tmp1 = src_8[i+s0*3+1];
tmp2 = src_8[i+s0*3+2];
src_8[i+s0*3] = src_8[i+s1*3];
src_8[i+s0*3+1] = src_8[i+s1*3+1];
src_8[i+s0*3+2] = src_8[i+s1*3+2];
src_8[i+s1*3] = tmp0;
src_8[i+s1*3+1] = tmp1;
src_8[i+s1*3+2] = tmp2;
}
break;
}
case 4:
{
int32_t *src_32 = src;
int32_t tmp;
if (chnum==6) {
REORDER_SELF_SWAP_2(src_32,tmp,samples,6,s0,s1);
}
else if (chnum==3) {
REORDER_SELF_SWAP_2(src_32,tmp,samples,3,s0,s1);
}
else if (chnum==4) {
REORDER_SELF_SWAP_2(src_32,tmp,samples,4,s0,s1);
}
else {
REORDER_SELF_SWAP_2(src_32,tmp,samples,5,s0,s1);
}
break;
}
case 8:
{
int64_t *src_64 = src;
int64_t tmp;
if (chnum==6) {
REORDER_SELF_SWAP_2(src_64,tmp,samples,6,s0,s1);
}
else if (chnum==3) {
REORDER_SELF_SWAP_2(src_64,tmp,samples,3,s0,s1);
}
else if (chnum==4) {
REORDER_SELF_SWAP_2(src_64,tmp,samples,4,s0,s1);
}
else {
REORDER_SELF_SWAP_2(src_64,tmp,samples,5,s0,s1);
}
break;
}
default:
mp_msg(MSGT_GLOBAL, MSGL_WARN,
"[reorder_ch] Unsupported sample size: %d, please "
"report this error on the MPlayer mailing list.\n",samplesize);
return 0;
}
return 1;
}
#define REORDER_SELF_SWAP_3(SRC,TMP,SAMPLES,CHNUM,S0,S1,S2) \
for (i = 0; i < SAMPLES; i += CHNUM) {\
TMP = SRC[i+S0];\
SRC[i+S0] = SRC[i+S1];\
SRC[i+S1] = SRC[i+S2];\
SRC[i+S2] = TMP;\
}
static int reorder_self_3(void *src, unsigned int samples,
unsigned int samplesize, unsigned int chnum,
int s0, int s1, int s2)
{
int i;
switch (samplesize) {
case 1:
{
int8_t *src_8 = src;
int8_t tmp;
if (chnum==6) {
REORDER_SELF_SWAP_3(src_8,tmp,samples,6,s0,s1,s2);
}
else {
REORDER_SELF_SWAP_3(src_8,tmp,samples,5,s0,s1,s2);
}
break;
}
case 2:
{
int16_t *src_16 = src;
int16_t tmp;
if (chnum==6) {
REORDER_SELF_SWAP_3(src_16,tmp,samples,6,s0,s1,s2);
}
else {
REORDER_SELF_SWAP_3(src_16,tmp,samples,5,s0,s1,s2);
}
break;
}
case 3:
{
int8_t *src_8 = src;
int8_t tmp0, tmp1, tmp2;
for (i = 0; i < samples * 3; i += chnum * 3) {
tmp0 = src_8[i+s0*3];
tmp1 = src_8[i+s0*3+1];
tmp2 = src_8[i+s0*3+2];
src_8[i+s0*3] = src_8[i+s1*3];
src_8[i+s0*3+1] = src_8[i+s1*3+1];
src_8[i+s0*3+2] = src_8[i+s1*3+2];
src_8[i+s1*3] = src_8[i+s2*3];
src_8[i+s1*3+1] = src_8[i+s2*3+1];
src_8[i+s1*3+2] = src_8[i+s2*3+2];
src_8[i+s2*3] = tmp0;
src_8[i+s2*3+1] = tmp1;
src_8[i+s2*3+2] = tmp2;
}
break;
}
case 4:
{
int32_t *src_32 = src;
int32_t tmp;
if (chnum==6) {
REORDER_SELF_SWAP_3(src_32,tmp,samples,6,s0,s1,s2);
}
else {
REORDER_SELF_SWAP_3(src_32,tmp,samples,5,s0,s1,s2);
}
break;
}
case 8:
{
int64_t *src_64 = src;
int64_t tmp;
if (chnum==6) {
REORDER_SELF_SWAP_3(src_64,tmp,samples,6,s0,s1,s2);
}
else {
REORDER_SELF_SWAP_3(src_64,tmp,samples,5,s0,s1,s2);
}
break;
}
default:
mp_msg(MSGT_GLOBAL, MSGL_WARN,
"[reorder_ch] Unsupported sample size: %d, please "
"report this error on the MPlayer mailing list.\n",samplesize);
return 0;
}
return 1;
}
#define REORDER_SELF_SWAP_4_STEP_1(SRC,TMP,SAMPLES,CHNUM,S0,S1,S2,S3) \
for (i = 0; i < SAMPLES; i += CHNUM) {\
TMP = SRC[i+S0];\
SRC[i+S0] = SRC[i+S1];\
SRC[i+S1] = SRC[i+S2];\
SRC[i+S2] = SRC[i+S3];\
SRC[i+S3] = TMP;\
}
static int reorder_self_4_step_1(void *src, unsigned int samples,
unsigned int samplesize, unsigned int chnum,
int s0, int s1, int s2, int s3)
{
int i;
switch (samplesize) {
case 1:
{
int8_t *src_8 = src;
int8_t tmp;
if (chnum==6) {
REORDER_SELF_SWAP_4_STEP_1(src_8,tmp,samples,6,s0,s1,s2,s3);
}
else if (chnum==8) {
REORDER_SELF_SWAP_4_STEP_1(src_8,tmp,samples,8,s0,s1,s2,s3);
}
else {
REORDER_SELF_SWAP_4_STEP_1(src_8,tmp,samples,5,s0,s1,s2,s3);
}
break;
}
case 2:
{
int16_t *src_16 = src;
int16_t tmp;
if (chnum==6) {
REORDER_SELF_SWAP_4_STEP_1(src_16,tmp,samples,6,s0,s1,s2,s3);
}
else if (chnum==8) {
REORDER_SELF_SWAP_4_STEP_1(src_16,tmp,samples,8,s0,s1,s2,s3);
}
else {
REORDER_SELF_SWAP_4_STEP_1(src_16,tmp,samples,5,s0,s1,s2,s3);
}
break;
}
case 3:
{
int8_t *src_8 = src;
int8_t tmp0, tmp1, tmp2;
for (i = 0; i < samples * 3; i += chnum * 3) {
tmp0 = src_8[i+s0*3];
tmp1 = src_8[i+s0*3+1];
tmp2 = src_8[i+s0*3+2];
src_8[i+s0*3] = src_8[i+s1*3];
src_8[i+s0*3+1] = src_8[i+s1*3+1];
src_8[i+s0*3+2] = src_8[i+s1*3+2];
src_8[i+s1*3] = src_8[i+s2*3];
src_8[i+s1*3+1] = src_8[i+s2*3+1];
src_8[i+s1*3+2] = src_8[i+s2*3+2];
src_8[i+s2*3] = src_8[i+s3*3];
src_8[i+s2*3+1] = src_8[i+s3*3+1];
src_8[i+s2*3+2] = src_8[i+s3*3+2];
src_8[i+s3*3] = tmp0;
src_8[i+s3*3+1] = tmp1;
src_8[i+s3*3+2] = tmp2;
}
break;
}
case 4:
{
int32_t *src_32 = src;
int32_t tmp;
if (chnum==6) {
REORDER_SELF_SWAP_4_STEP_1(src_32,tmp,samples,6,s0,s1,s2,s3);
}
else if (chnum==8) {
REORDER_SELF_SWAP_4_STEP_1(src_32,tmp,samples,8,s0,s1,s2,s3);
}
else {
REORDER_SELF_SWAP_4_STEP_1(src_32,tmp,samples,5,s0,s1,s2,s3);
}
break;
}
case 8:
{
int64_t *src_64 = src;
int64_t tmp;
if (chnum==6) {
REORDER_SELF_SWAP_4_STEP_1(src_64,tmp,samples,6,s0,s1,s2,s3);
}
else if (chnum==8) {
REORDER_SELF_SWAP_4_STEP_1(src_64,tmp,samples,8,s0,s1,s2,s3);
}
else {
REORDER_SELF_SWAP_4_STEP_1(src_64,tmp,samples,5,s0,s1,s2,s3);
}
break;
}
default:
mp_msg(MSGT_GLOBAL, MSGL_WARN,
"[reorder_ch] Unsupported sample size: %d, please "
"report this error on the MPlayer mailing list.\n",samplesize);
return 0;
}
return 1;
}
#define REORDER_SELF_SWAP_4_STEP_2(SRC,TMP,SAMPLES,CHNUM,S0,S1,S2,S3) \
for (i = 0; i < SAMPLES; i += CHNUM) {\
TMP = SRC[i+S0];\
SRC[i+S0] = SRC[i+S2];\
SRC[i+S2] = TMP;\
TMP = SRC[i+S1];\
SRC[i+S1] = SRC[i+S3];\
SRC[i+S3] = TMP;\
}
static int reorder_self_4_step_2(void *src, unsigned int samples,
unsigned int samplesize, unsigned int chnum,
int s0, int s1, int s2, int s3)
{
int i;
switch (samplesize) {
case 3:
{
int8_t *src_8 = src;
int8_t tmp0, tmp1, tmp2;
for (i = 0; i < samples * 3; i += chnum * 3) {
tmp0 = src_8[i+s0*3];
tmp1 = src_8[i+s0*3+1];
tmp2 = src_8[i+s0*3+2];
src_8[i+s0*3] = src_8[i+s2*3];
src_8[i+s0*3+1] = src_8[i+s2*3+1];
src_8[i+s0*3+2] = src_8[i+s2*3+2];
src_8[i+s2*3] = tmp0;
src_8[i+s2*3+1] = tmp1;
src_8[i+s2*3+2] = tmp2;
tmp0 = src_8[i+s1*3];
tmp1 = src_8[i+s1*3+1];
tmp2 = src_8[i+s1*3+2];
src_8[i+s1*3] = src_8[i+s3*3];
src_8[i+s1*3+1] = src_8[i+s3*3+1];
src_8[i+s1*3+2] = src_8[i+s3*3+2];
src_8[i+s3*3] = tmp0;
src_8[i+s3*3+1] = tmp1;
src_8[i+s3*3+2] = tmp2;
}
break;
}
default:
mp_msg(MSGT_GLOBAL, MSGL_WARN,
"[reorder_ch] Unsupported sample size: %d, please "
"report this error on the MPlayer mailing list.\n",samplesize);
return 0;
}
return 1;
}
#define REORDER_SELF_SWAP_5_STEP_1(SRC,TMP,SAMPLES,CHNUM,S0,S1,S2,S3,S4) \
for (i = 0; i < SAMPLES; i += CHNUM) {\
TMP = SRC[i+S0];\
SRC[i+S0] = SRC[i+S1];\
SRC[i+S1] = SRC[i+S2];\
SRC[i+S2] = SRC[i+S3];\
SRC[i+S3] = SRC[i+S4];\
SRC[i+S4] = TMP;\
}
static int reorder_self_5_step_1(void *src, unsigned int samples,
unsigned int samplesize, unsigned int chnum,
int s0, int s1, int s2, int s3, int s4)
{
int i;
switch (samplesize) {
case 1:
{
int8_t *src_8 = src;
int8_t tmp;
if (chnum==6) {
REORDER_SELF_SWAP_5_STEP_1(src_8,tmp,samples,6,s0,s1,s2,s3,s4);
}
else if (chnum==8) {
REORDER_SELF_SWAP_5_STEP_1(src_8,tmp,samples,8,s0,s1,s2,s3,s4);
}
else {
REORDER_SELF_SWAP_5_STEP_1(src_8,tmp,samples,5,s0,s1,s2,s3,s4);
}
break;
}
case 2:
{
int16_t *src_16 = src;
int16_t tmp;
if (chnum==6) {
REORDER_SELF_SWAP_5_STEP_1(src_16,tmp,samples,6,s0,s1,s2,s3,s4);
}
else if (chnum==8) {
REORDER_SELF_SWAP_5_STEP_1(src_16,tmp,samples,8,s0,s1,s2,s3,s4);
}
else {
REORDER_SELF_SWAP_5_STEP_1(src_16,tmp,samples,5,s0,s1,s2,s3,s4);
}
break;
}
case 3:
{
int8_t *src_8 = src;
int8_t tmp0, tmp1, tmp2;
for (i = 0; i < samples * 3; i += chnum * 3) {
tmp0 = src_8[i+s0*3];
tmp1 = src_8[i+s0*3+1];
tmp2 = src_8[i+s0*3+2];
src_8[i+s0*3] = src_8[i+s1*3];
src_8[i+s0*3+1] = src_8[i+s1*3+1];
src_8[i+s0*3+2] = src_8[i+s1*3+2];
src_8[i+s1*3] = src_8[i+s2*3];
src_8[i+s1*3+1] = src_8[i+s2*3+1];
src_8[i+s1*3+2] = src_8[i+s2*3+2];
src_8[i+s2*3] = src_8[i+s3*3];
src_8[i+s2*3+1] = src_8[i+s3*3+1];
src_8[i+s2*3+2] = src_8[i+s3*3+2];
src_8[i+s3*3] = src_8[i+s4*3];
src_8[i+s3*3+1] = src_8[i+s4*3+1];
src_8[i+s3*3+2] = src_8[i+s4*3+2];
src_8[i+s4*3] = tmp0;
src_8[i+s4*3+1] = tmp1;
src_8[i+s4*3+2] = tmp2;
}
break;
}
case 4:
{
int32_t *src_32 = src;
int32_t tmp;
if (chnum==6) {
REORDER_SELF_SWAP_5_STEP_1(src_32,tmp,samples,6,s0,s1,s2,s3,s4);
}
else if (chnum==8) {
REORDER_SELF_SWAP_5_STEP_1(src_32,tmp,samples,8,s0,s1,s2,s3,s4);
}
else {
REORDER_SELF_SWAP_5_STEP_1(src_32,tmp,samples,5,s0,s1,s2,s3,s4);
}
break;
}
case 8:
{
int64_t *src_64 = src;
int64_t tmp;
if (chnum==6) {
REORDER_SELF_SWAP_5_STEP_1(src_64,tmp,samples,6,s0,s1,s2,s3,s4);
}
else if (chnum==8) {
REORDER_SELF_SWAP_5_STEP_1(src_64,tmp,samples,8,s0,s1,s2,s3,s4);
}
else {
REORDER_SELF_SWAP_5_STEP_1(src_64,tmp,samples,5,s0,s1,s2,s3,s4);
}
break;
}
default:
mp_msg(MSGT_GLOBAL, MSGL_WARN,
"[reorder_ch] Unsupported sample size: %d, please "
"report this error on the MPlayer mailing list.\n",samplesize);
return 0;
}
return 1;
}
#define REORDER_SELF_SWAP_2_3(SRC,TMP,SAMPLES,CHNUM,S0,S1,S2,S3,S4) \
for (i = 0; i < SAMPLES; i += CHNUM) {\
TMP = SRC[i+S0];\
SRC[i+S0] = SRC[i+S1];\
SRC[i+S1] = TMP;\
TMP = SRC[i+S2];\
SRC[i+S2] = SRC[i+S3];\
SRC[i+S3] = SRC[i+S4];\
SRC[i+S4] = TMP;\
}
static int reorder_self_2_3(void *src, unsigned int samples,
unsigned int samplesize,
int s0, int s1, int s2, int s3, int s4)
{
int i;
switch (samplesize) {
case 1:
{
int8_t *src_8 = src;
int8_t tmp;
REORDER_SELF_SWAP_2_3(src_8,tmp,samples,6,s0,s1,s2,s3,s4);
break;
}
case 2:
{
int16_t *src_16 = src;
int16_t tmp;
REORDER_SELF_SWAP_2_3(src_16,tmp,samples,6,s0,s1,s2,s3,s4);
break;
}
case 3:
{
int8_t *src_8 = src;
int8_t tmp0, tmp1, tmp2;
for (i = 0; i < samples * 3; i += 18) {
tmp0 = src_8[i+s0*3];
tmp1 = src_8[i+s0*3+1];
tmp2 = src_8[i+s0*3+2];
src_8[i+s0*3] = src_8[i+s1*3];
src_8[i+s0*3+1] = src_8[i+s1*3+1];
src_8[i+s0*3+2] = src_8[i+s1*3+2];
src_8[i+s1*3] = tmp0;
src_8[i+s1*3+1] = tmp1;
src_8[i+s1*3+2] = tmp2;
tmp0 = src_8[i+s2*3];
tmp1 = src_8[i+s2*3+1];
tmp2 = src_8[i+s2*3+2];
src_8[i+s2*3] = src_8[i+s3*3];
src_8[i+s2*3+1] = src_8[i+s3*3+1];
src_8[i+s2*3+2] = src_8[i+s3*3+2];
src_8[i+s3*3] = src_8[i+s4*3];
src_8[i+s3*3+1] = src_8[i+s4*3+1];
src_8[i+s3*3+2] = src_8[i+s4*3+2];
src_8[i+s4*3] = tmp0;
src_8[i+s4*3+1] = tmp1;
src_8[i+s4*3+2] = tmp2;
}
break;
}
case 4:
{
int32_t *src_32 = src;
int32_t tmp;
REORDER_SELF_SWAP_2_3(src_32,tmp,samples,6,s0,s1,s2,s3,s4);
break;
}
case 8:
{
int64_t *src_64 = src;
int64_t tmp;
REORDER_SELF_SWAP_2_3(src_64,tmp,samples,6,s0,s1,s2,s3,s4);
break;
}
default:
mp_msg(MSGT_GLOBAL, MSGL_WARN,
"[reorder_ch] Unsupported sample size: %d, please "
"report this error on the MPlayer mailing list.\n",samplesize);
return 0;
}
return 1;
}
#define REORDER_SELF_SWAP_3_3(SRC,TMP,SAMPLES,CHNUM,S0,S1,S2,S3,S4,S5) \
for (i = 0; i < SAMPLES; i += CHNUM) {\
TMP = SRC[i+S0];\
SRC[i+S0] = SRC[i+S1];\
SRC[i+S1] = SRC[i+S2];\
SRC[i+S2] = TMP;\
TMP = SRC[i+S3];\
SRC[i+S3] = SRC[i+S4];\
SRC[i+S4] = SRC[i+S5];\
SRC[i+S5] = TMP;\
}
static int reorder_self_3_3(void *src, unsigned int samples,
unsigned int samplesize,
int s0, int s1, int s2, int s3, int s4, int s5)
{
int i;
switch (samplesize) {
case 1:
{
int8_t *src_8 = src;
int8_t tmp;
REORDER_SELF_SWAP_3_3(src_8,tmp,samples,6,s0,s1,s2,s3,s4,s5);
break;
}
case 2:
{
int16_t *src_16 = src;
int16_t tmp;
REORDER_SELF_SWAP_3_3(src_16,tmp,samples,6,s0,s1,s2,s3,s4,s5);
break;
}
case 3:
{
int8_t *src_8 = src;
int8_t tmp0, tmp1, tmp2;
for (i = 0; i < samples * 3; i += 18) {
tmp0 = src_8[i+s0*3];
tmp1 = src_8[i+s0*3+1];
tmp2 = src_8[i+s0*3+2];
src_8[i+s0*3] = src_8[i+s1*3];
src_8[i+s0*3+1] = src_8[i+s1*3+1];
src_8[i+s0*3+2] = src_8[i+s1*3+2];
src_8[i+s1*3] = src_8[i+s2*3];
src_8[i+s1*3+1] = src_8[i+s2*3+1];
src_8[i+s1*3+2] = src_8[i+s2*3+2];
src_8[i+s2*3] = tmp0;
src_8[i+s2*3+1] = tmp1;
src_8[i+s2*3+2] = tmp2;
tmp0 = src_8[i+s3*3];
tmp1 = src_8[i+s3*3+1];
tmp2 = src_8[i+s3*3+2];
src_8[i+s3*3] = src_8[i+s4*3];
src_8[i+s3*3+1] = src_8[i+s4*3+1];
src_8[i+s3*3+2] = src_8[i+s4*3+2];
src_8[i+s4*3] = src_8[i+s5*3];
src_8[i+s4*3+1] = src_8[i+s5*3+1];
src_8[i+s4*3+2] = src_8[i+s5*3+2];
src_8[i+s5*3] = tmp0;
src_8[i+s5*3+1] = tmp1;
src_8[i+s5*3+2] = tmp2;
}
break;
}
case 4:
{
int32_t *src_32 = src;
int32_t tmp;
REORDER_SELF_SWAP_3_3(src_32,tmp,samples,6,s0,s1,s2,s3,s4,s5);
break;
}
case 8:
{
int64_t *src_64 = src;
int64_t tmp;
REORDER_SELF_SWAP_3_3(src_64,tmp,samples,6,s0,s1,s2,s3,s4,s5);
break;
}
default:
mp_msg(MSGT_GLOBAL, MSGL_WARN,
"[reorder_ch] Unsupported sample size: %d, please "
"report this error on the MPlayer mailing list.\n",samplesize);
return 0;
}
return 1;
}
#define REORDER_SELF_SWAP_2_4(SRC,TMP,SAMPLES,CHNUM,S0,S1,S2,S3,S4,S5) \
for (i = 0; i < SAMPLES; i += CHNUM) {\
TMP = SRC[i+S0];\
SRC[i+S0] = SRC[i+S1];\
SRC[i+S1] = TMP;\
TMP = SRC[i+S2];\
SRC[i+S2] = SRC[i+S3];\
SRC[i+S3] = SRC[i+S4];\
SRC[i+S4] = SRC[i+S5];\
SRC[i+S5] = TMP;\
}
static int reorder_self_2_4(void *src, unsigned int samples,
unsigned int samplesize, int chnum,
int s0, int s1, int s2, int s3, int s4, int s5)
{
int i;
switch (samplesize) {
case 1:
{
int8_t *src_8 = src;
int8_t tmp;
if (chnum==6) {
REORDER_SELF_SWAP_2_4(src_8,tmp,samples,6,s0,s1,s2,s3,s4,s5);
} else {
REORDER_SELF_SWAP_2_4(src_8,tmp,samples,8,s0,s1,s2,s3,s4,s5);
}
break;
}
case 2:
{
int16_t *src_16 = src;
int16_t tmp;
if (chnum==6) {
REORDER_SELF_SWAP_2_4(src_16,tmp,samples,6,s0,s1,s2,s3,s4,s5);
} else {
REORDER_SELF_SWAP_2_4(src_16,tmp,samples,8,s0,s1,s2,s3,s4,s5);
}
break;
}
case 3:
{
int8_t *src_8 = src;
int8_t tmp0, tmp1, tmp2;
for (i = 0; i < samples * 3; i += chnum * 3) {
tmp0 = src_8[i+s0*3];
tmp1 = src_8[i+s0*3+1];
tmp2 = src_8[i+s0*3+2];
src_8[i+s0*3] = src_8[i+s1*3];
src_8[i+s0*3+1] = src_8[i+s1*3+1];
src_8[i+s0*3+2] = src_8[i+s1*3+2];
src_8[i+s1*3] = tmp0;
src_8[i+s1*3+1] = tmp1;
src_8[i+s1*3+2] = tmp2;
tmp0 = src_8[i+s2*3];
tmp1 = src_8[i+s2*3+1];
tmp2 = src_8[i+s2*3+2];
src_8[i+s2*3] = src_8[i+s3*3];
src_8[i+s2*3+1] = src_8[i+s3*3+1];
src_8[i+s2*3+2] = src_8[i+s3*3+2];
src_8[i+s3*3] = src_8[i+s4*3];
src_8[i+s3*3+1] = src_8[i+s4*3+1];
src_8[i+s3*3+2] = src_8[i+s4*3+2];
src_8[i+s4*3] = src_8[i+s5*3];
src_8[i+s4*3+1] = src_8[i+s5*3+1];
src_8[i+s4*3+2] = src_8[i+s5*3+2];
src_8[i+s5*3] = tmp0;
src_8[i+s5*3+1] = tmp1;
src_8[i+s5*3+2] = tmp2;
}
break;
}
case 4:
{
int32_t *src_32 = src;
int32_t tmp;
if (chnum==6) {
REORDER_SELF_SWAP_2_4(src_32,tmp,samples,6,s0,s1,s2,s3,s4,s5);
} else {
REORDER_SELF_SWAP_2_4(src_32,tmp,samples,8,s0,s1,s2,s3,s4,s5);
}
break;
}
case 8:
{
int64_t *src_64 = src;
int64_t tmp;
if (chnum==6) {
REORDER_SELF_SWAP_2_4(src_64,tmp,samples,6,s0,s1,s2,s3,s4,s5);
} else {
REORDER_SELF_SWAP_2_4(src_64,tmp,samples,8,s0,s1,s2,s3,s4,s5);
}
break;
}
default:
mp_msg(MSGT_GLOBAL, MSGL_WARN,
"[reorder_ch] Unsupported sample size: %d, please "
"report this error on the MPlayer mailing list.\n",samplesize);
return 0;
}
return 1;
}
void reorder_channel(void *src,
int src_layout,
int dest_layout,
int samples,
int samplesize)
{
if (dest_layout==src_layout)
return;
if (!AF_IS_SAME_CH_NUM(dest_layout,src_layout)) {
mp_msg(MSGT_GLOBAL, MSGL_WARN,
"[reorder_channel] different channel count "
"between current and target: %x, %x\n",
AF_GET_CH_NUM_WITH_LFE(src_layout),
AF_GET_CH_NUM_WITH_LFE(dest_layout));
return;
}
switch ((src_layout<<16)|dest_layout) {
// AF_CHANNEL_LAYOUT_5_0_A L R C Ls Rs
// AF_CHANNEL_LAYOUT_5_0_B L R Ls Rs C
// AF_CHANNEL_LAYOUT_5_0_C L C R Ls Rs
// AF_CHANNEL_LAYOUT_5_0_D C L R Ls Rs
case AF_CHANNEL_LAYOUT_5_0_A << 16 | AF_CHANNEL_LAYOUT_5_0_B:
reorder_self_3(src, samples, samplesize, 5, 2, 3, 4);
break;
case AF_CHANNEL_LAYOUT_5_0_A << 16 | AF_CHANNEL_LAYOUT_5_0_C:
reorder_self_2(src, samples, samplesize, 5, 1, 2);
break;
case AF_CHANNEL_LAYOUT_5_0_A << 16 | AF_CHANNEL_LAYOUT_5_0_D:
reorder_self_3(src, samples, samplesize, 5, 2, 1, 0);
break;
case AF_CHANNEL_LAYOUT_5_0_B << 16 | AF_CHANNEL_LAYOUT_5_0_A:
reorder_self_3(src, samples, samplesize, 5, 4, 3, 2);
break;
case AF_CHANNEL_LAYOUT_5_0_B << 16 | AF_CHANNEL_LAYOUT_5_0_C:
reorder_self_4_step_1(src, samples, samplesize, 5, 4, 3, 2, 1);
break;
case AF_CHANNEL_LAYOUT_5_0_B << 16 | AF_CHANNEL_LAYOUT_5_0_D:
reorder_self_5_step_1(src, samples, samplesize, 5, 4, 3, 2, 1, 0);
break;
case AF_CHANNEL_LAYOUT_5_0_C << 16 | AF_CHANNEL_LAYOUT_5_0_A:
reorder_self_2(src, samples, samplesize, 5, 1, 2);
break;
case AF_CHANNEL_LAYOUT_5_0_C << 16 | AF_CHANNEL_LAYOUT_5_0_B:
reorder_self_4_step_1(src, samples, samplesize, 5, 1, 2, 3, 4);
break;
case AF_CHANNEL_LAYOUT_5_0_C << 16 | AF_CHANNEL_LAYOUT_5_0_D:
reorder_self_2(src, samples, samplesize, 5, 0, 1);
break;
case AF_CHANNEL_LAYOUT_5_0_D << 16 | AF_CHANNEL_LAYOUT_5_0_A:
reorder_self_3(src, samples, samplesize, 5, 0, 1, 2);
break;
case AF_CHANNEL_LAYOUT_5_0_D << 16 | AF_CHANNEL_LAYOUT_5_0_B:
reorder_self_5_step_1(src, samples, samplesize, 5, 0, 1, 2, 3, 4);
break;
case AF_CHANNEL_LAYOUT_5_0_D << 16 | AF_CHANNEL_LAYOUT_5_0_C:
reorder_self_2(src, samples, samplesize, 5, 0, 1);
break;
// AF_CHANNEL_LAYOUT_5_1_A L R C LFE Ls Rs
// AF_CHANNEL_LAYOUT_5_1_B L R Ls Rs C LFE
// AF_CHANNEL_LAYOUT_5_1_C L C R Ls Rs LFE
// AF_CHANNEL_LAYOUT_5_1_D C L R Ls Rs LFE
// AF_CHANNEL_LAYOUT_5_1_E LFE L C R Ls Rs
case AF_CHANNEL_LAYOUT_5_1_A << 16 | AF_CHANNEL_LAYOUT_5_1_B:
if (samplesize != 3)
reorder_self_2(src, samples/2, samplesize*2, 3, 1, 2);
else
reorder_self_4_step_2(src, samples, samplesize, 6, 2, 3, 4, 5);
break;
case AF_CHANNEL_LAYOUT_5_1_A << 16 | AF_CHANNEL_LAYOUT_5_1_C:
reorder_self_2_3(src, samples, samplesize, 1, 2, 3, 4, 5);
break;
case AF_CHANNEL_LAYOUT_5_1_A << 16 | AF_CHANNEL_LAYOUT_5_1_D:
reorder_self_3_3(src, samples, samplesize, 2, 1, 0, 3, 4, 5);
break;
case AF_CHANNEL_LAYOUT_5_1_B << 16 | AF_CHANNEL_LAYOUT_5_1_A:
if (samplesize != 3)
reorder_self_2(src, samples/2, samplesize*2, 3, 1, 2);
else
reorder_self_4_step_2(src, samples, samplesize, 6, 2, 3, 4, 5);
break;
case AF_CHANNEL_LAYOUT_5_1_B << 16 | AF_CHANNEL_LAYOUT_5_1_C:
reorder_self_4_step_1(src, samples, samplesize, 6, 4, 3, 2, 1);
break;
case AF_CHANNEL_LAYOUT_5_1_B << 16 | AF_CHANNEL_LAYOUT_5_1_D:
reorder_self_5_step_1(src, samples, samplesize, 6, 4, 3, 2, 1, 0);
break;
case AF_CHANNEL_LAYOUT_5_1_B << 16 | AF_CHANNEL_LAYOUT_5_1_E:
reorder_self_2_4(src, samples, samplesize, 6, 2, 4, 5, 3, 1, 0);
break;
case AF_CHANNEL_LAYOUT_5_1_C << 16 | AF_CHANNEL_LAYOUT_5_1_A:
reorder_self_2_3(src, samples, samplesize, 1, 2, 5, 4, 3);
break;
case AF_CHANNEL_LAYOUT_5_1_C << 16 | AF_CHANNEL_LAYOUT_5_1_B:
reorder_self_4_step_1(src, samples, samplesize, 6, 1, 2, 3, 4);
break;
case AF_CHANNEL_LAYOUT_5_1_C << 16 | AF_CHANNEL_LAYOUT_5_1_D:
reorder_self_2(src, samples, samplesize, 6, 0, 1);
break;
case AF_CHANNEL_LAYOUT_5_1_D << 16 | AF_CHANNEL_LAYOUT_5_1_A:
reorder_self_3_3(src, samples, samplesize, 0, 1, 2, 5, 4, 3);
break;
case AF_CHANNEL_LAYOUT_5_1_D << 16 | AF_CHANNEL_LAYOUT_5_1_B:
reorder_self_5_step_1(src, samples, samplesize, 6, 0, 1, 2, 3, 4);
break;
case AF_CHANNEL_LAYOUT_5_1_D << 16 | AF_CHANNEL_LAYOUT_5_1_C:
reorder_self_2(src, samples, samplesize, 6, 0, 1);
break;
case AF_CHANNEL_LAYOUT_5_1_E << 16 | AF_CHANNEL_LAYOUT_5_1_B:
reorder_self_2_4(src, samples, samplesize, 6, 2, 4, 0, 1, 3, 5);
break;
case AF_CHANNEL_LAYOUT_5_1_F << 16 | AF_CHANNEL_LAYOUT_5_1_B:
reorder_self_2_4(src, samples, samplesize, 6, 3, 5, 0, 1, 2, 4);
break;
// AF_CHANNEL_LAYOUT_7_1_A L R C LFE Ls Rs Rls Rrs
// AF_CHANNEL_LAYOUT_7_1_B L R Ls Rs C LFE Rls Rrs
// AF_CHANNEL_LAYOUT_7_1_C L C R Ls Rs LFE Rls Rrs
// AF_CHANNEL_LAYOUT_7_1_D C L R Ls Rs Rls Rrs LFE
// AF_CHANNEL_LAYOUT_7_1_F C L R LFE Ls Rs Rls Rrs
case AF_CHANNEL_LAYOUT_7_1_A << 16 | AF_CHANNEL_LAYOUT_7_1_B:
case AF_CHANNEL_LAYOUT_7_1_B << 16 | AF_CHANNEL_LAYOUT_7_1_A:
if (samplesize != 3)
reorder_self_2(src, samples/2, samplesize*2, 4, 1, 2);
else
reorder_self_4_step_2(src, samples, samplesize, 8, 2, 3, 4, 5);
break;
case AF_CHANNEL_LAYOUT_7_1_B << 16 | AF_CHANNEL_LAYOUT_7_1_D:
// First convert to AF_CHANNEL_LAYOUT_7_1_F
reorder_self_2_4(src, samples, samplesize, 8, 3, 5, 4, 2, 1, 0);
// then convert to AF_CHANNEL_LAYOUT_7_1_D
reorder_self_5_step_1(src, samples, samplesize, 8, 3, 4, 5, 6, 7);
break;
case AF_CHANNEL_LAYOUT_7_1_C << 16 | AF_CHANNEL_LAYOUT_7_1_B:
reorder_self_4_step_1(src, samples, samplesize, 8, 1, 2, 3, 4);
break;
case AF_CHANNEL_LAYOUT_7_1_F << 16 | AF_CHANNEL_LAYOUT_7_1_B:
reorder_self_2_4(src, samples, samplesize, 8, 3, 5, 0, 1, 2, 4);
break;
default:
mp_msg(MSGT_GLOBAL, MSGL_WARN,
"[reorder_channel] unsupported from %x to %x, %d * %d\n",
src_layout, dest_layout, samples, samplesize);
}
}
static int channel_layout_mapping_5ch[AF_CHANNEL_LAYOUT_SOURCE_NUM] = {
AF_CHANNEL_LAYOUT_ALSA_5CH_DEFAULT,
AF_CHANNEL_LAYOUT_AAC_5CH_DEFAULT,
AF_CHANNEL_LAYOUT_WAVEEX_5CH_DEFAULT,
AF_CHANNEL_LAYOUT_LAVC_5CH_DEFAULT,
AF_CHANNEL_LAYOUT_VORBIS_5CH_DEFAULT,
};
static int channel_layout_mapping_6ch[AF_CHANNEL_LAYOUT_SOURCE_NUM] = {
AF_CHANNEL_LAYOUT_ALSA_6CH_DEFAULT,
AF_CHANNEL_LAYOUT_AAC_6CH_DEFAULT,
AF_CHANNEL_LAYOUT_WAVEEX_6CH_DEFAULT,
AF_CHANNEL_LAYOUT_LAVC_6CH_DEFAULT,
AF_CHANNEL_LAYOUT_VORBIS_6CH_DEFAULT,
};
static int channel_layout_mapping_8ch[AF_CHANNEL_LAYOUT_SOURCE_NUM] = {
AF_CHANNEL_LAYOUT_ALSA_8CH_DEFAULT,
AF_CHANNEL_LAYOUT_AAC_8CH_DEFAULT,
AF_CHANNEL_LAYOUT_WAVEEX_8CH_DEFAULT,
AF_CHANNEL_LAYOUT_LAVC_8CH_DEFAULT,
AF_CHANNEL_LAYOUT_VORBIS_8CH_DEFAULT,
};
void reorder_channel_copy_nch(void *src,
int src_layout,
void *dest,
int dest_layout,
int chnum,
int samples,
int samplesize)
{
if (chnum < 5 || chnum == 7 || chnum > 8 ||
src_layout < 0 || dest_layout < 0 ||
src_layout >= AF_CHANNEL_LAYOUT_SOURCE_NUM ||
dest_layout >= AF_CHANNEL_LAYOUT_SOURCE_NUM)
memcpy(dest, src, samples*samplesize);
else if (chnum == 6)
reorder_channel_copy(src, channel_layout_mapping_6ch[src_layout],
dest, channel_layout_mapping_6ch[dest_layout],
samples, samplesize);
else if (chnum == 8)
reorder_channel_copy(src, channel_layout_mapping_8ch[src_layout],
dest, channel_layout_mapping_8ch[dest_layout],
samples, samplesize);
else
reorder_channel_copy(src, channel_layout_mapping_5ch[src_layout],
dest, channel_layout_mapping_5ch[dest_layout],
samples, samplesize);
}
void reorder_channel_nch(void *buf,
int src_layout,
int dest_layout,
int chnum,
int samples,
int samplesize)
{
if (src_layout == dest_layout || chnum < 5 || chnum == 7 || chnum > 8 ||
src_layout < 0 || dest_layout < 0 ||
src_layout >= AF_CHANNEL_LAYOUT_SOURCE_NUM ||
dest_layout >= AF_CHANNEL_LAYOUT_SOURCE_NUM ||
src_layout == dest_layout)
return;
if (chnum == 6)
reorder_channel(buf, channel_layout_mapping_6ch[src_layout],
channel_layout_mapping_6ch[dest_layout],
samples, samplesize);
else if (chnum == 8)
reorder_channel(buf, channel_layout_mapping_8ch[src_layout],
channel_layout_mapping_8ch[dest_layout],
samples, samplesize);
else
reorder_channel(buf, channel_layout_mapping_5ch[src_layout],
channel_layout_mapping_5ch[dest_layout],
samples, samplesize);
}
#ifdef TEST
static void test_copy(int channels) {
int samples = 12*1024*1024;
int samplesize = 2;
int i;
unsigned char *bufin = malloc((samples+100)*samplesize);
unsigned char *bufout = malloc((samples+100)*samplesize);
memset(bufin, 0xFF, samples*samplesize);
for (i = 0;i < 100; ++i)
reorder_channel_copy(bufin, AF_CHANNEL_LAYOUT_5_1_A,
bufout, AF_CHANNEL_LAYOUT_5_1_B,
samples, samplesize);
// reorder_channel(bufin, AF_CHANNEL_LAYOUT_5_1_B,
// AF_CHANNEL_LAYOUT_5_1_D,
// samples, samplesize);
free(bufin);
free(bufout);
}
int main(int argc, char *argv[]) {
int channels = 6;
if (argc > 1)
channels = atoi(argv[1]);
test_copy(channels);
return 0;
}
#endif
static inline void reorder_to_planar_(void *restrict out, const void *restrict in,
size_t size, size_t nchan, size_t nmemb)
{
size_t i, c;
char *outptr = (char *) out;
size_t instep = nchan * size;
for (c = 0; c < nchan; ++c) {
const char *inptr = ((const char *) in) + c * size;
for (i = 0; i < nmemb; ++i, inptr += instep, outptr += size) {
memcpy(outptr, inptr, size);
}
}
}
void reorder_to_planar(void *restrict out, const void *restrict in,
size_t size, size_t nchan, size_t nmemb)
{
// special case for mono (nothing to do...)
if (nchan == 1)
memcpy(out, in, size * nchan * nmemb);
// these calls exist to insert an inline copy of to_planar_ here with known
// value of size to help the compiler replace the memcpy calls by mov
// instructions
else if (size == 1)
reorder_to_planar_(out, in, 1, nchan, nmemb);
else if (size == 2)
reorder_to_planar_(out, in, 2, nchan, nmemb);
else if (size == 4)
reorder_to_planar_(out, in, 4, nchan, nmemb);
// general case (calls memcpy a lot, should actually never happen, but
// stays here for correctness purposes)
else
reorder_to_planar_(out, in, size, nchan, nmemb);
}
static inline void reorder_to_packed_(uint8_t *out, uint8_t **in,
size_t size, size_t nchan, size_t nmemb)
{
size_t outstep = nchan * size;
for (size_t c = 0; c < nchan; ++c) {
char *outptr = out + c * size;
char *inptr = in[c];
for (size_t i = 0; i < nmemb; ++i, outptr += outstep, inptr += size) {
memcpy(outptr, inptr, size);
}
}
}
// out = destination array of packed samples of given size, nmemb frames
// in[channel] = source array of samples for the given channel
void reorder_to_packed(uint8_t *out, uint8_t **in,
size_t size, size_t nchan, size_t nmemb)
{
if (nchan == 1)
memcpy(out, in, size * nchan * nmemb);
// See reorder_to_planar() why this is done this way
else if (size == 1)
reorder_to_packed_(out, in, 1, nchan, nmemb);
else if (size == 2)
reorder_to_packed_(out, in, 2, nchan, nmemb);
else if (size == 4)
reorder_to_packed_(out, in, 4, nchan, nmemb);
else
reorder_to_packed_(out, in, size, nchan, nmemb);
}