len0rd/rockbox
1
0
Fork 1
mirror of https://github.com/Rockbox/rockbox.git synced 2025-11-19 18:12:54 -05:00
Code Issues Projects Releases Packages Wiki Activity Actions

Use Tremor IMDCT for WMA. Gives a ~20% speedup on ARM and coldfire. All my test samples gave equivilent or better accuracy (>>16 bit), but there may still be problems. Code becomes somewhat less readable since the IMDCT has bits of Vorbis in it, but this is not a serious issue and the old ffmpeg transform remains in SVN if anyone wants it. Also, WMA now passes Vorbis as the fastest transform codec on ARM, although MPC remains the fastest lossy codec overall.

Browse source 
git-svn-id: svn://svn.rockbox.org/rockbox/trunk@18084 a1c6a512-1295-4272-9138-f99709370657
This commit is contained in:
Michael Giacomelli 2008-07-17 17:43:49 +00:00
parent 5356e6205b
commit 1334738071
15 changed files with 2584 additions and 91 deletions
Download patch file Download diff file Expand all files Collapse all files

522
apps/codecs/libwma/mdct2.c Normal file
View file

@ -0,0 +1,522 @@
/********************************************************************
* *
* THIS FILE IS PART OF THE OggVorbis 'TREMOR' CODEC SOURCE CODE. *
* *
* USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
* GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
* IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
* *
* THE OggVorbis 'TREMOR' SOURCE CODE IS (C) COPYRIGHT 1994-2002 *
* BY THE Xiph.Org FOUNDATION http://www.xiph.org/ *
* *
********************************************************************
function: normalized modified discrete cosine transform
power of two length transform only [64 <= n ]
Original algorithm adapted long ago from _The use of multirate filter
banks for coding of high quality digital audio_, by T. Sporer,
K. Brandenburg and B. Edler, collection of the European Signal
Processing Conference (EUSIPCO), Amsterdam, June 1992, Vol.1, pp
211-214
The below code implements an algorithm that no longer looks much like
that presented in the paper, but the basic structure remains if you
dig deep enough to see it.
This module DOES NOT INCLUDE code to generate/apply the window
function. Everybody has their own weird favorite including me... I
happen to like the properties of y=sin(.5PI*sin^2(x)), but others may
vehemently disagree.
********************************************************************/
/*Tremor IMDCT adapted for use with libwmai*/
#include "mdct2.h"
#include "mdct_lookup.h"
#include <codecs/lib/codeclib.h>
#if defined(CPU_ARM) && CONFIG_CPU != S3C2440
/* C code is faster on S3C2440 */
extern void mdct_butterfly_32(ogg_int32_t *x);
extern void mdct_butterfly_generic_loop(ogg_int32_t *x1, ogg_int32_t *x2,
LOOKUP_T *T0, int step,
LOOKUP_T *Ttop);
static inline void mdct_butterfly_generic(ogg_int32_t *x,int points, int step){
mdct_butterfly_generic_loop(x + points, x + (points>>1),
sincos_lookup0, step, sincos_lookup0+1024);
}
#else
/* 8 point butterfly (in place) */
static inline void mdct_butterfly_8(ogg_int32_t *x){
register ogg_int32_t r0 = x[4] + x[0];
register ogg_int32_t r1 = x[4] - x[0];
register ogg_int32_t r2 = x[5] + x[1];
register ogg_int32_t r3 = x[5] - x[1];
register ogg_int32_t r4 = x[6] + x[2];
register ogg_int32_t r5 = x[6] - x[2];
register ogg_int32_t r6 = x[7] + x[3];
register ogg_int32_t r7 = x[7] - x[3];
x[0] = r5 + r3;
x[1] = r7 - r1;
x[2] = r5 - r3;
x[3] = r7 + r1;
x[4] = r4 - r0;
x[5] = r6 - r2;
x[6] = r4 + r0;
x[7] = r6 + r2;
MB();
}
/* 16 point butterfly (in place, 4 register) */
static inline void mdct_butterfly_16(ogg_int32_t *x){
register ogg_int32_t r0, r1;
r0 = x[ 0] - x[ 8]; x[ 8] += x[ 0];
r1 = x[ 1] - x[ 9]; x[ 9] += x[ 1];
x[ 0] = MULT31((r0 + r1) , cPI2_8);
x[ 1] = MULT31((r1 - r0) , cPI2_8);
MB();
r0 = x[10] - x[ 2]; x[10] += x[ 2];
r1 = x[ 3] - x[11]; x[11] += x[ 3];
x[ 2] = r1; x[ 3] = r0;
MB();
r0 = x[12] - x[ 4]; x[12] += x[ 4];
r1 = x[13] - x[ 5]; x[13] += x[ 5];
x[ 4] = MULT31((r0 - r1) , cPI2_8);
x[ 5] = MULT31((r0 + r1) , cPI2_8);
MB();
r0 = x[14] - x[ 6]; x[14] += x[ 6];
r1 = x[15] - x[ 7]; x[15] += x[ 7];
x[ 6] = r0; x[ 7] = r1;
MB();
mdct_butterfly_8(x);
mdct_butterfly_8(x+8);
}
/* 32 point butterfly (in place, 4 register) */
static inline void mdct_butterfly_32(ogg_int32_t *x){
register ogg_int32_t r0, r1;
r0 = x[30] - x[14]; x[30] += x[14];
r1 = x[31] - x[15]; x[31] += x[15];
x[14] = r0; x[15] = r1;
MB();
r0 = x[28] - x[12]; x[28] += x[12];
r1 = x[29] - x[13]; x[29] += x[13];
XNPROD31( r0, r1, cPI1_8, cPI3_8, &x[12], &x[13] );
MB();
r0 = x[26] - x[10]; x[26] += x[10];
r1 = x[27] - x[11]; x[27] += x[11];
x[10] = MULT31((r0 - r1) , cPI2_8);
x[11] = MULT31((r0 + r1) , cPI2_8);
MB();
r0 = x[24] - x[ 8]; x[24] += x[ 8];
r1 = x[25] - x[ 9]; x[25] += x[ 9];
XNPROD31( r0, r1, cPI3_8, cPI1_8, &x[ 8], &x[ 9] );
MB();
r0 = x[22] - x[ 6]; x[22] += x[ 6];
r1 = x[ 7] - x[23]; x[23] += x[ 7];
x[ 6] = r1; x[ 7] = r0;
MB();
r0 = x[ 4] - x[20]; x[20] += x[ 4];
r1 = x[ 5] - x[21]; x[21] += x[ 5];
XPROD31 ( r0, r1, cPI3_8, cPI1_8, &x[ 4], &x[ 5] );
MB();
r0 = x[ 2] - x[18]; x[18] += x[ 2];
r1 = x[ 3] - x[19]; x[19] += x[ 3];
x[ 2] = MULT31((r1 + r0) , cPI2_8);
x[ 3] = MULT31((r1 - r0) , cPI2_8);
MB();
r0 = x[ 0] - x[16]; x[16] += x[ 0];
r1 = x[ 1] - x[17]; x[17] += x[ 1];
XPROD31 ( r0, r1, cPI1_8, cPI3_8, &x[ 0], &x[ 1] );
MB();
mdct_butterfly_16(x);
mdct_butterfly_16(x+16);
}
/* N/stage point generic N stage butterfly (in place, 4 register) */
void mdct_butterfly_generic(ogg_int32_t *x,int points, int step)
ICODE_ATTR_TREMOR_MDCT;
void mdct_butterfly_generic(ogg_int32_t *x,int points, int step){
LOOKUP_T *T = sincos_lookup0;
ogg_int32_t *x1 = x + points - 8;
ogg_int32_t *x2 = x + (points>>1) - 8;
register ogg_int32_t r0;
register ogg_int32_t r1;
register ogg_int32_t r2;
register ogg_int32_t r3;
do{
r0 = x1[6] - x2[6]; x1[6] += x2[6];
r1 = x2[7] - x1[7]; x1[7] += x2[7];
r2 = x1[4] - x2[4]; x1[4] += x2[4];
r3 = x2[5] - x1[5]; x1[5] += x2[5];
XPROD31( r1, r0, T[0], T[1], &x2[6], &x2[7] ); T+=step;
XPROD31( r3, r2, T[0], T[1], &x2[4], &x2[5] ); T+=step;
r0 = x1[2] - x2[2]; x1[2] += x2[2];
r1 = x2[3] - x1[3]; x1[3] += x2[3];
r2 = x1[0] - x2[0]; x1[0] += x2[0];
r3 = x2[1] - x1[1]; x1[1] += x2[1];
XPROD31( r1, r0, T[0], T[1], &x2[2], &x2[3] ); T+=step;
XPROD31( r3, r2, T[0], T[1], &x2[0], &x2[1] ); T+=step;
x1-=8; x2-=8;
}while(T<sincos_lookup0+1024);
do{
r0 = x1[6] - x2[6]; x1[6] += x2[6];
r1 = x1[7] - x2[7]; x1[7] += x2[7];
r2 = x1[4] - x2[4]; x1[4] += x2[4];
r3 = x1[5] - x2[5]; x1[5] += x2[5];
XNPROD31( r0, r1, T[0], T[1], &x2[6], &x2[7] ); T-=step;
XNPROD31( r2, r3, T[0], T[1], &x2[4], &x2[5] ); T-=step;
r0 = x1[2] - x2[2]; x1[2] += x2[2];
r1 = x1[3] - x2[3]; x1[3] += x2[3];
r2 = x1[0] - x2[0]; x1[0] += x2[0];
r3 = x1[1] - x2[1]; x1[1] += x2[1];
XNPROD31( r0, r1, T[0], T[1], &x2[2], &x2[3] ); T-=step;
XNPROD31( r2, r3, T[0], T[1], &x2[0], &x2[1] ); T-=step;
x1-=8; x2-=8;
}while(T>sincos_lookup0);
do{
r0 = x2[6] - x1[6]; x1[6] += x2[6];
r1 = x2[7] - x1[7]; x1[7] += x2[7];
r2 = x2[4] - x1[4]; x1[4] += x2[4];
r3 = x2[5] - x1[5]; x1[5] += x2[5];
XPROD31( r0, r1, T[0], T[1], &x2[6], &x2[7] ); T+=step;
XPROD31( r2, r3, T[0], T[1], &x2[4], &x2[5] ); T+=step;
r0 = x2[2] - x1[2]; x1[2] += x2[2];
r1 = x2[3] - x1[3]; x1[3] += x2[3];
r2 = x2[0] - x1[0]; x1[0] += x2[0];
r3 = x2[1] - x1[1]; x1[1] += x2[1];
XPROD31( r0, r1, T[0], T[1], &x2[2], &x2[3] ); T+=step;
XPROD31( r2, r3, T[0], T[1], &x2[0], &x2[1] ); T+=step;
x1-=8; x2-=8;
}while(T<sincos_lookup0+1024);
do{
r0 = x1[6] - x2[6]; x1[6] += x2[6];
r1 = x2[7] - x1[7]; x1[7] += x2[7];
r2 = x1[4] - x2[4]; x1[4] += x2[4];
r3 = x2[5] - x1[5]; x1[5] += x2[5];
XNPROD31( r1, r0, T[0], T[1], &x2[6], &x2[7] ); T-=step;
XNPROD31( r3, r2, T[0], T[1], &x2[4], &x2[5] ); T-=step;
r0 = x1[2] - x2[2]; x1[2] += x2[2];
r1 = x2[3] - x1[3]; x1[3] += x2[3];
r2 = x1[0] - x2[0]; x1[0] += x2[0];
r3 = x2[1] - x1[1]; x1[1] += x2[1];
XNPROD31( r1, r0, T[0], T[1], &x2[2], &x2[3] ); T-=step;
XNPROD31( r3, r2, T[0], T[1], &x2[0], &x2[1] ); T-=step;
x1-=8; x2-=8;
}while(T>sincos_lookup0);
}
#endif /* CPU_ARM */
static inline void mdct_butterflies(ogg_int32_t *x,int points,int shift) {
int stages=8-shift;
int i,j;
for(i=0;--stages>0;i++){
for(j=0;j<(1<<i);j++)
mdct_butterfly_generic(x+(points>>i)*j,points>>i,4<<(i+shift));
}
for(j=0;j<points;j+=32)
mdct_butterfly_32(x+j);
}
static const unsigned char bitrev[16] ICONST_ATTR =
{0,8,4,12,2,10,6,14,1,9,5,13,3,11,7,15};
static inline int bitrev12(int x){
return bitrev[x>>8]|(bitrev[(x&0x0f0)>>4]<<4)|(((int)bitrev[x&0x00f])<<8);
}
static inline void mdct_bitreverse(ogg_int32_t *x,int n,int step,int shift) {
int bit = 0;
ogg_int32_t *w0 = x;
ogg_int32_t *w1 = x = w0+(n>>1);
LOOKUP_T *T = (step>=4)?(sincos_lookup0+(step>>1)):sincos_lookup1;
LOOKUP_T *Ttop = T+1024;
register ogg_int32_t r2;
do{
register ogg_int32_t r3 = bitrev12(bit++);
ogg_int32_t *x0 = x + ((r3 ^ 0xfff)>>shift) -1;
ogg_int32_t *x1 = x + (r3>>shift);
register ogg_int32_t r0 = x0[0] + x1[0];
register ogg_int32_t r1 = x1[1] - x0[1];
XPROD32( r0, r1, T[1], T[0], r2, r3 ); T+=step;
w1 -= 4;
r0 = (x0[1] + x1[1])>>1;
r1 = (x0[0] - x1[0])>>1;
w0[0] = r0 + r2;
w0[1] = r1 + r3;
w1[2] = r0 - r2;
w1[3] = r3 - r1;
r3 = bitrev12(bit++);
x0 = x + ((r3 ^ 0xfff)>>shift) -1;
x1 = x + (r3>>shift);
r0 = x0[0] + x1[0];
r1 = x1[1] - x0[1];
XPROD32( r0, r1, T[1], T[0], r2, r3 ); T+=step;
r0 = (x0[1] + x1[1])>>1;
r1 = (x0[0] - x1[0])>>1;
w0[2] = r0 + r2;
w0[3] = r1 + r3;
w1[0] = r0 - r2;
w1[1] = r3 - r1;
w0 += 4;
}while(T<Ttop);
do{
register ogg_int32_t r3 = bitrev12(bit++);
ogg_int32_t *x0 = x + ((r3 ^ 0xfff)>>shift) -1;
ogg_int32_t *x1 = x + (r3>>shift);
register ogg_int32_t r0 = x0[0] + x1[0];
register ogg_int32_t r1 = x1[1] - x0[1];
T-=step; XPROD32( r0, r1, T[0], T[1], r2, r3 );
w1 -= 4;
r0 = (x0[1] + x1[1])>>1;
r1 = (x0[0] - x1[0])>>1;
w0[0] = r0 + r2;
w0[1] = r1 + r3;
w1[2] = r0 - r2;
w1[3] = r3 - r1;
r3 = bitrev12(bit++);
x0 = x + ((r3 ^ 0xfff)>>shift) -1;
x1 = x + (r3>>shift);
r0 = x0[0] + x1[0];
r1 = x1[1] - x0[1];
T-=step; XPROD32( r0, r1, T[0], T[1], r2, r3 );
r0 = (x0[1] + x1[1])>>1;
r1 = (x0[0] - x1[0])>>1;
w0[2] = r0 + r2;
w0[3] = r1 + r3;
w1[0] = r0 - r2;
w1[1] = r3 - r1;
w0 += 4;
}while(w0<w1);
}
void mdct_backward(int n, ogg_int32_t *in, ogg_int32_t *out)
ICODE_ATTR_TREMOR_MDCT;
void mdct_backward(int n, ogg_int32_t *in, ogg_int32_t *out) {
int n2=n>>1;
int n4=n>>2;
ogg_int32_t *iX;
ogg_int32_t *oX;
LOOKUP_T *T;
LOOKUP_T *V;
int shift;
int step;
for (shift=6;!(n&(1<<shift));shift++);
shift=13-shift;
step=2<<shift;
/* rotate */
iX = in+n2-7;
oX = out+n2+n4;
T = sincos_lookup0;
do{
oX-=4;
XPROD31( iX[4], iX[6], T[0], T[1], &oX[2], &oX[3] ); T+=step;
XPROD31( iX[0], iX[2], T[0], T[1], &oX[0], &oX[1] ); T+=step;
iX-=8;
}while(iX>=in+n4);
do{
oX-=4;
XPROD31( iX[4], iX[6], T[1], T[0], &oX[2], &oX[3] ); T-=step;
XPROD31( iX[0], iX[2], T[1], T[0], &oX[0], &oX[1] ); T-=step;
iX-=8;
}while(iX>=in);
iX = in+n2-8;
oX = out+n2+n4;
T = sincos_lookup0;
do{
T+=step; XNPROD31( iX[6], iX[4], T[0], T[1], &oX[0], &oX[1] );
T+=step; XNPROD31( iX[2], iX[0], T[0], T[1], &oX[2], &oX[3] );
iX-=8;
oX+=4;
}while(iX>=in+n4);
do{
T-=step; XNPROD31( iX[6], iX[4], T[1], T[0], &oX[0], &oX[1] );
T-=step; XNPROD31( iX[2], iX[0], T[1], T[0], &oX[2], &oX[3] );
iX-=8;
oX+=4;
}while(iX>=in);
mdct_butterflies(out+n2,n2,shift);
mdct_bitreverse(out,n,step,shift);
/* rotate + window */
step>>=2;
{
ogg_int32_t *oX1=out+n2+n4;
ogg_int32_t *oX2=out+n2+n4;
ogg_int32_t *iX =out;
switch(step) {
default: {
T=(step>=4)?(sincos_lookup0+(step>>1)):sincos_lookup1;
do{
oX1-=4;
XPROD31( iX[0], -iX[1], T[0], T[1], &oX1[3], &oX2[0] ); T+=step;
XPROD31( iX[2], -iX[3], T[0], T[1], &oX1[2], &oX2[1] ); T+=step;
XPROD31( iX[4], -iX[5], T[0], T[1], &oX1[1], &oX2[2] ); T+=step;
XPROD31( iX[6], -iX[7], T[0], T[1], &oX1[0], &oX2[3] ); T+=step;
oX2+=4;
iX+=8;
}while(iX<oX1);
break;
}
case 1: {
/* linear interpolation between table values: offset=0.5, step=1 */
register ogg_int32_t t0,t1,v0,v1;
T = sincos_lookup0;
V = sincos_lookup1;
t0 = (*T++)>>1;
t1 = (*T++)>>1;
do{
oX1-=4;
t0 += (v0 = (*V++)>>1);
t1 += (v1 = (*V++)>>1);
XPROD31( iX[0], -iX[1], t0, t1, &oX1[3], &oX2[0] );
v0 += (t0 = (*T++)>>1);
v1 += (t1 = (*T++)>>1);
XPROD31( iX[2], -iX[3], v0, v1, &oX1[2], &oX2[1] );
t0 += (v0 = (*V++)>>1);
t1 += (v1 = (*V++)>>1);
XPROD31( iX[4], -iX[5], t0, t1, &oX1[1], &oX2[2] );
v0 += (t0 = (*T++)>>1);
v1 += (t1 = (*T++)>>1);
XPROD31( iX[6], -iX[7], v0, v1, &oX1[0], &oX2[3] );
oX2+=4;
iX+=8;
}while(iX<oX1);
break;
}
case 0: {
/* linear interpolation between table values: offset=0.25, step=0.5 */
register ogg_int32_t t0,t1,v0,v1,q0,q1;
T = sincos_lookup0;
V = sincos_lookup1;
t0 = *T++;
t1 = *T++;
do{
oX1-=4;
v0 = *V++;
v1 = *V++;
t0 += (q0 = (v0-t0)>>2);
t1 += (q1 = (v1-t1)>>2);
XPROD31( iX[0], -iX[1], t0, t1, &oX1[3], &oX2[0] );
t0 = v0-q0;
t1 = v1-q1;
XPROD31( iX[2], -iX[3], t0, t1, &oX1[2], &oX2[1] );
t0 = *T++;
t1 = *T++;
v0 += (q0 = (t0-v0)>>2);
v1 += (q1 = (t1-v1)>>2);
XPROD31( iX[4], -iX[5], v0, v1, &oX1[1], &oX2[2] );
v0 = t0-q0;
v1 = t1-q1;
XPROD31( iX[6], -iX[7], v0, v1, &oX1[0], &oX2[3] );
oX2+=4;
iX+=8;
}while(iX<oX1);
break;
}
}
iX=out+n2+n4;
oX1=out+n4;
oX2=oX1;
do{
oX1-=4;
iX-=4;
oX2[0] = -(oX1[3] = iX[3]);
oX2[1] = -(oX1[2] = iX[2]);
oX2[2] = -(oX1[1] = iX[1]);
oX2[3] = -(oX1[0] = iX[0]);
oX2+=4;
}while(oX2<iX);
iX=out+n2+n4;
oX1=out+n2+n4;
oX2=out+n2;
do{
oX1-=4;
oX1[0]= iX[3];
oX1[1]= iX[2];
oX1[2]= iX[1];
oX1[3]= iX[0];
iX+=4;
}while(oX1>oX2);
}
}