1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/media/libtremor/lib/tremor_mdct.c Wed Dec 31 06:09:35 2014 +0100
1.3 @@ -0,0 +1,510 @@
1.4 +/********************************************************************
1.5 + * *
1.6 + * THIS FILE IS PART OF THE OggVorbis 'TREMOR' CODEC SOURCE CODE. *
1.7 + * *
1.8 + * USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
1.9 + * GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
1.10 + * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
1.11 + * *
1.12 + * THE OggVorbis 'TREMOR' SOURCE CODE IS (C) COPYRIGHT 1994-2002 *
1.13 + * BY THE Xiph.Org FOUNDATION http://www.xiph.org/ *
1.14 + * *
1.15 + ********************************************************************
1.16 +
1.17 + function: normalized modified discrete cosine transform
1.18 + power of two length transform only [64 <= n ]
1.19 + last mod: $Id: mdct.c,v 1.9 2002/10/16 09:17:39 xiphmont Exp $
1.20 +
1.21 + Original algorithm adapted long ago from _The use of multirate filter
1.22 + banks for coding of high quality digital audio_, by T. Sporer,
1.23 + K. Brandenburg and B. Edler, collection of the European Signal
1.24 + Processing Conference (EUSIPCO), Amsterdam, June 1992, Vol.1, pp
1.25 + 211-214
1.26 +
1.27 + The below code implements an algorithm that no longer looks much like
1.28 + that presented in the paper, but the basic structure remains if you
1.29 + dig deep enough to see it.
1.30 +
1.31 + This module DOES NOT INCLUDE code to generate/apply the window
1.32 + function. Everybody has their own weird favorite including me... I
1.33 + happen to like the properties of y=sin(.5PI*sin^2(x)), but others may
1.34 + vehemently disagree.
1.35 +
1.36 + ********************************************************************/
1.37 +
1.38 +#include "ivorbiscodec.h"
1.39 +#include "codebook.h"
1.40 +#include "misc.h"
1.41 +#include "mdct.h"
1.42 +#include "mdct_lookup.h"
1.43 +
1.44 +
1.45 +/* 8 point butterfly (in place) */
1.46 +STIN void mdct_butterfly_8(DATA_TYPE *x){
1.47 +
1.48 + REG_TYPE r0 = x[4] + x[0];
1.49 + REG_TYPE r1 = x[4] - x[0];
1.50 + REG_TYPE r2 = x[5] + x[1];
1.51 + REG_TYPE r3 = x[5] - x[1];
1.52 + REG_TYPE r4 = x[6] + x[2];
1.53 + REG_TYPE r5 = x[6] - x[2];
1.54 + REG_TYPE r6 = x[7] + x[3];
1.55 + REG_TYPE r7 = x[7] - x[3];
1.56 +
1.57 + x[0] = r5 + r3;
1.58 + x[1] = r7 - r1;
1.59 + x[2] = r5 - r3;
1.60 + x[3] = r7 + r1;
1.61 + x[4] = r4 - r0;
1.62 + x[5] = r6 - r2;
1.63 + x[6] = r4 + r0;
1.64 + x[7] = r6 + r2;
1.65 + MB();
1.66 +}
1.67 +
1.68 +/* 16 point butterfly (in place, 4 register) */
1.69 +STIN void mdct_butterfly_16(DATA_TYPE *x){
1.70 +
1.71 + REG_TYPE r0, r1;
1.72 +
1.73 + r0 = x[ 0] - x[ 8]; x[ 8] += x[ 0];
1.74 + r1 = x[ 1] - x[ 9]; x[ 9] += x[ 1];
1.75 + x[ 0] = MULT31((r0 + r1) , cPI2_8);
1.76 + x[ 1] = MULT31((r1 - r0) , cPI2_8);
1.77 + MB();
1.78 +
1.79 + r0 = x[10] - x[ 2]; x[10] += x[ 2];
1.80 + r1 = x[ 3] - x[11]; x[11] += x[ 3];
1.81 + x[ 2] = r1; x[ 3] = r0;
1.82 + MB();
1.83 +
1.84 + r0 = x[12] - x[ 4]; x[12] += x[ 4];
1.85 + r1 = x[13] - x[ 5]; x[13] += x[ 5];
1.86 + x[ 4] = MULT31((r0 - r1) , cPI2_8);
1.87 + x[ 5] = MULT31((r0 + r1) , cPI2_8);
1.88 + MB();
1.89 +
1.90 + r0 = x[14] - x[ 6]; x[14] += x[ 6];
1.91 + r1 = x[15] - x[ 7]; x[15] += x[ 7];
1.92 + x[ 6] = r0; x[ 7] = r1;
1.93 + MB();
1.94 +
1.95 + mdct_butterfly_8(x);
1.96 + mdct_butterfly_8(x+8);
1.97 +}
1.98 +
1.99 +/* 32 point butterfly (in place, 4 register) */
1.100 +STIN void mdct_butterfly_32(DATA_TYPE *x){
1.101 +
1.102 + REG_TYPE r0, r1;
1.103 +
1.104 + r0 = x[30] - x[14]; x[30] += x[14];
1.105 + r1 = x[31] - x[15]; x[31] += x[15];
1.106 + x[14] = r0; x[15] = r1;
1.107 + MB();
1.108 +
1.109 + r0 = x[28] - x[12]; x[28] += x[12];
1.110 + r1 = x[29] - x[13]; x[29] += x[13];
1.111 + XNPROD31( r0, r1, cPI1_8, cPI3_8, &x[12], &x[13] );
1.112 + MB();
1.113 +
1.114 + r0 = x[26] - x[10]; x[26] += x[10];
1.115 + r1 = x[27] - x[11]; x[27] += x[11];
1.116 + x[10] = MULT31((r0 - r1) , cPI2_8);
1.117 + x[11] = MULT31((r0 + r1) , cPI2_8);
1.118 + MB();
1.119 +
1.120 + r0 = x[24] - x[ 8]; x[24] += x[ 8];
1.121 + r1 = x[25] - x[ 9]; x[25] += x[ 9];
1.122 + XNPROD31( r0, r1, cPI3_8, cPI1_8, &x[ 8], &x[ 9] );
1.123 + MB();
1.124 +
1.125 + r0 = x[22] - x[ 6]; x[22] += x[ 6];
1.126 + r1 = x[ 7] - x[23]; x[23] += x[ 7];
1.127 + x[ 6] = r1; x[ 7] = r0;
1.128 + MB();
1.129 +
1.130 + r0 = x[ 4] - x[20]; x[20] += x[ 4];
1.131 + r1 = x[ 5] - x[21]; x[21] += x[ 5];
1.132 + XPROD31 ( r0, r1, cPI3_8, cPI1_8, &x[ 4], &x[ 5] );
1.133 + MB();
1.134 +
1.135 + r0 = x[ 2] - x[18]; x[18] += x[ 2];
1.136 + r1 = x[ 3] - x[19]; x[19] += x[ 3];
1.137 + x[ 2] = MULT31((r1 + r0) , cPI2_8);
1.138 + x[ 3] = MULT31((r1 - r0) , cPI2_8);
1.139 + MB();
1.140 +
1.141 + r0 = x[ 0] - x[16]; x[16] += x[ 0];
1.142 + r1 = x[ 1] - x[17]; x[17] += x[ 1];
1.143 + XPROD31 ( r0, r1, cPI1_8, cPI3_8, &x[ 0], &x[ 1] );
1.144 + MB();
1.145 +
1.146 + mdct_butterfly_16(x);
1.147 + mdct_butterfly_16(x+16);
1.148 +}
1.149 +
1.150 +/* N/stage point generic N stage butterfly (in place, 2 register) */
1.151 +STIN void mdct_butterfly_generic(DATA_TYPE *x,int points,int step){
1.152 +
1.153 + LOOKUP_T *T = sincos_lookup0;
1.154 + DATA_TYPE *x1 = x + points - 8;
1.155 + DATA_TYPE *x2 = x + (points>>1) - 8;
1.156 + REG_TYPE r0;
1.157 + REG_TYPE r1;
1.158 +
1.159 + do{
1.160 + r0 = x1[6] - x2[6]; x1[6] += x2[6];
1.161 + r1 = x2[7] - x1[7]; x1[7] += x2[7];
1.162 + XPROD31( r1, r0, T[0], T[1], &x2[6], &x2[7] ); T+=step;
1.163 +
1.164 + r0 = x1[4] - x2[4]; x1[4] += x2[4];
1.165 + r1 = x2[5] - x1[5]; x1[5] += x2[5];
1.166 + XPROD31( r1, r0, T[0], T[1], &x2[4], &x2[5] ); T+=step;
1.167 +
1.168 + r0 = x1[2] - x2[2]; x1[2] += x2[2];
1.169 + r1 = x2[3] - x1[3]; x1[3] += x2[3];
1.170 + XPROD31( r1, r0, T[0], T[1], &x2[2], &x2[3] ); T+=step;
1.171 +
1.172 + r0 = x1[0] - x2[0]; x1[0] += x2[0];
1.173 + r1 = x2[1] - x1[1]; x1[1] += x2[1];
1.174 + XPROD31( r1, r0, T[0], T[1], &x2[0], &x2[1] ); T+=step;
1.175 +
1.176 + x1-=8; x2-=8;
1.177 + }while(T<sincos_lookup0+1024);
1.178 + do{
1.179 + r0 = x1[6] - x2[6]; x1[6] += x2[6];
1.180 + r1 = x1[7] - x2[7]; x1[7] += x2[7];
1.181 + XNPROD31( r0, r1, T[0], T[1], &x2[6], &x2[7] ); T-=step;
1.182 +
1.183 + r0 = x1[4] - x2[4]; x1[4] += x2[4];
1.184 + r1 = x1[5] - x2[5]; x1[5] += x2[5];
1.185 + XNPROD31( r0, r1, T[0], T[1], &x2[4], &x2[5] ); T-=step;
1.186 +
1.187 + r0 = x1[2] - x2[2]; x1[2] += x2[2];
1.188 + r1 = x1[3] - x2[3]; x1[3] += x2[3];
1.189 + XNPROD31( r0, r1, T[0], T[1], &x2[2], &x2[3] ); T-=step;
1.190 +
1.191 + r0 = x1[0] - x2[0]; x1[0] += x2[0];
1.192 + r1 = x1[1] - x2[1]; x1[1] += x2[1];
1.193 + XNPROD31( r0, r1, T[0], T[1], &x2[0], &x2[1] ); T-=step;
1.194 +
1.195 + x1-=8; x2-=8;
1.196 + }while(T>sincos_lookup0);
1.197 + do{
1.198 + r0 = x2[6] - x1[6]; x1[6] += x2[6];
1.199 + r1 = x2[7] - x1[7]; x1[7] += x2[7];
1.200 + XPROD31( r0, r1, T[0], T[1], &x2[6], &x2[7] ); T+=step;
1.201 +
1.202 + r0 = x2[4] - x1[4]; x1[4] += x2[4];
1.203 + r1 = x2[5] - x1[5]; x1[5] += x2[5];
1.204 + XPROD31( r0, r1, T[0], T[1], &x2[4], &x2[5] ); T+=step;
1.205 +
1.206 + r0 = x2[2] - x1[2]; x1[2] += x2[2];
1.207 + r1 = x2[3] - x1[3]; x1[3] += x2[3];
1.208 + XPROD31( r0, r1, T[0], T[1], &x2[2], &x2[3] ); T+=step;
1.209 +
1.210 + r0 = x2[0] - x1[0]; x1[0] += x2[0];
1.211 + r1 = x2[1] - x1[1]; x1[1] += x2[1];
1.212 + XPROD31( r0, r1, T[0], T[1], &x2[0], &x2[1] ); T+=step;
1.213 +
1.214 + x1-=8; x2-=8;
1.215 + }while(T<sincos_lookup0+1024);
1.216 + do{
1.217 + r0 = x1[6] - x2[6]; x1[6] += x2[6];
1.218 + r1 = x2[7] - x1[7]; x1[7] += x2[7];
1.219 + XNPROD31( r1, r0, T[0], T[1], &x2[6], &x2[7] ); T-=step;
1.220 +
1.221 + r0 = x1[4] - x2[4]; x1[4] += x2[4];
1.222 + r1 = x2[5] - x1[5]; x1[5] += x2[5];
1.223 + XNPROD31( r1, r0, T[0], T[1], &x2[4], &x2[5] ); T-=step;
1.224 +
1.225 + r0 = x1[2] - x2[2]; x1[2] += x2[2];
1.226 + r1 = x2[3] - x1[3]; x1[3] += x2[3];
1.227 + XNPROD31( r1, r0, T[0], T[1], &x2[2], &x2[3] ); T-=step;
1.228 +
1.229 + r0 = x1[0] - x2[0]; x1[0] += x2[0];
1.230 + r1 = x2[1] - x1[1]; x1[1] += x2[1];
1.231 + XNPROD31( r1, r0, T[0], T[1], &x2[0], &x2[1] ); T-=step;
1.232 +
1.233 + x1-=8; x2-=8;
1.234 + }while(T>sincos_lookup0);
1.235 +}
1.236 +
1.237 +STIN void mdct_butterflies(DATA_TYPE *x,int points,int shift){
1.238 +
1.239 + int stages=8-shift;
1.240 + int i,j;
1.241 +
1.242 + for(i=0;--stages>0;i++){
1.243 + for(j=0;j<(1<<i);j++)
1.244 + mdct_butterfly_generic(x+(points>>i)*j,points>>i,4<<(i+shift));
1.245 + }
1.246 +
1.247 + for(j=0;j<points;j+=32)
1.248 + mdct_butterfly_32(x+j);
1.249 +
1.250 +}
1.251 +
1.252 +static unsigned char bitrev[16]={0,8,4,12,2,10,6,14,1,9,5,13,3,11,7,15};
1.253 +
1.254 +STIN int bitrev12(int x){
1.255 + return bitrev[x>>8]|(bitrev[(x&0x0f0)>>4]<<4)|(((int)bitrev[x&0x00f])<<8);
1.256 +}
1.257 +
1.258 +STIN void mdct_bitreverse(DATA_TYPE *x,int n,int step,int shift){
1.259 +
1.260 + int bit = 0;
1.261 + DATA_TYPE *w0 = x;
1.262 + DATA_TYPE *w1 = x = w0+(n>>1);
1.263 + LOOKUP_T *T = (step>=4)?(sincos_lookup0+(step>>1)):sincos_lookup1;
1.264 + LOOKUP_T *Ttop = T+1024;
1.265 + DATA_TYPE r2;
1.266 +
1.267 + do{
1.268 + DATA_TYPE r3 = bitrev12(bit++);
1.269 + DATA_TYPE *x0 = x + ((r3 ^ 0xfff)>>shift) -1;
1.270 + DATA_TYPE *x1 = x + (r3>>shift);
1.271 +
1.272 + REG_TYPE r0 = x0[0] + x1[0];
1.273 + REG_TYPE r1 = x1[1] - x0[1];
1.274 +
1.275 + XPROD32( r0, r1, T[1], T[0], &r2, &r3 ); T+=step;
1.276 +
1.277 + w1 -= 4;
1.278 +
1.279 + r0 = (x0[1] + x1[1])>>1;
1.280 + r1 = (x0[0] - x1[0])>>1;
1.281 + w0[0] = r0 + r2;
1.282 + w0[1] = r1 + r3;
1.283 + w1[2] = r0 - r2;
1.284 + w1[3] = r3 - r1;
1.285 +
1.286 + r3 = bitrev12(bit++);
1.287 + x0 = x + ((r3 ^ 0xfff)>>shift) -1;
1.288 + x1 = x + (r3>>shift);
1.289 +
1.290 + r0 = x0[0] + x1[0];
1.291 + r1 = x1[1] - x0[1];
1.292 +
1.293 + XPROD32( r0, r1, T[1], T[0], &r2, &r3 ); T+=step;
1.294 +
1.295 + r0 = (x0[1] + x1[1])>>1;
1.296 + r1 = (x0[0] - x1[0])>>1;
1.297 + w0[2] = r0 + r2;
1.298 + w0[3] = r1 + r3;
1.299 + w1[0] = r0 - r2;
1.300 + w1[1] = r3 - r1;
1.301 +
1.302 + w0 += 4;
1.303 + }while(T<Ttop);
1.304 + do{
1.305 + DATA_TYPE r3 = bitrev12(bit++);
1.306 + DATA_TYPE *x0 = x + ((r3 ^ 0xfff)>>shift) -1;
1.307 + DATA_TYPE *x1 = x + (r3>>shift);
1.308 +
1.309 + REG_TYPE r0 = x0[0] + x1[0];
1.310 + REG_TYPE r1 = x1[1] - x0[1];
1.311 +
1.312 + T-=step; XPROD32( r0, r1, T[0], T[1], &r2, &r3 );
1.313 +
1.314 + w1 -= 4;
1.315 +
1.316 + r0 = (x0[1] + x1[1])>>1;
1.317 + r1 = (x0[0] - x1[0])>>1;
1.318 + w0[0] = r0 + r2;
1.319 + w0[1] = r1 + r3;
1.320 + w1[2] = r0 - r2;
1.321 + w1[3] = r3 - r1;
1.322 +
1.323 + r3 = bitrev12(bit++);
1.324 + x0 = x + ((r3 ^ 0xfff)>>shift) -1;
1.325 + x1 = x + (r3>>shift);
1.326 +
1.327 + r0 = x0[0] + x1[0];
1.328 + r1 = x1[1] - x0[1];
1.329 +
1.330 + T-=step; XPROD32( r0, r1, T[0], T[1], &r2, &r3 );
1.331 +
1.332 + r0 = (x0[1] + x1[1])>>1;
1.333 + r1 = (x0[0] - x1[0])>>1;
1.334 + w0[2] = r0 + r2;
1.335 + w0[3] = r1 + r3;
1.336 + w1[0] = r0 - r2;
1.337 + w1[1] = r3 - r1;
1.338 +
1.339 + w0 += 4;
1.340 + }while(w0<w1);
1.341 +}
1.342 +
1.343 +void mdct_backward(int n, DATA_TYPE *in, DATA_TYPE *out){
1.344 + int n2=n>>1;
1.345 + int n4=n>>2;
1.346 + DATA_TYPE *iX;
1.347 + DATA_TYPE *oX;
1.348 + LOOKUP_T *T;
1.349 + LOOKUP_T *V;
1.350 + int shift;
1.351 + int step;
1.352 +
1.353 + for (shift=6;!(n&(1<<shift));shift++);
1.354 + shift=13-shift;
1.355 + step=2<<shift;
1.356 +
1.357 + /* rotate */
1.358 +
1.359 + iX = in+n2-7;
1.360 + oX = out+n2+n4;
1.361 + T = sincos_lookup0;
1.362 +
1.363 + do{
1.364 + oX-=4;
1.365 + XPROD31( iX[4], iX[6], T[0], T[1], &oX[2], &oX[3] ); T+=step;
1.366 + XPROD31( iX[0], iX[2], T[0], T[1], &oX[0], &oX[1] ); T+=step;
1.367 + iX-=8;
1.368 + }while(iX>=in+n4);
1.369 + do{
1.370 + oX-=4;
1.371 + XPROD31( iX[4], iX[6], T[1], T[0], &oX[2], &oX[3] ); T-=step;
1.372 + XPROD31( iX[0], iX[2], T[1], T[0], &oX[0], &oX[1] ); T-=step;
1.373 + iX-=8;
1.374 + }while(iX>=in);
1.375 +
1.376 + iX = in+n2-8;
1.377 + oX = out+n2+n4;
1.378 + T = sincos_lookup0;
1.379 +
1.380 + do{
1.381 + T+=step; XNPROD31( iX[6], iX[4], T[0], T[1], &oX[0], &oX[1] );
1.382 + T+=step; XNPROD31( iX[2], iX[0], T[0], T[1], &oX[2], &oX[3] );
1.383 + iX-=8;
1.384 + oX+=4;
1.385 + }while(iX>=in+n4);
1.386 + do{
1.387 + T-=step; XNPROD31( iX[6], iX[4], T[1], T[0], &oX[0], &oX[1] );
1.388 + T-=step; XNPROD31( iX[2], iX[0], T[1], T[0], &oX[2], &oX[3] );
1.389 + iX-=8;
1.390 + oX+=4;
1.391 + }while(iX>=in);
1.392 +
1.393 + mdct_butterflies(out+n2,n2,shift);
1.394 + mdct_bitreverse(out,n,step,shift);
1.395 +
1.396 + /* rotate + window */
1.397 +
1.398 + step>>=2;
1.399 + {
1.400 + DATA_TYPE *oX1=out+n2+n4;
1.401 + DATA_TYPE *oX2=out+n2+n4;
1.402 + DATA_TYPE *iX =out;
1.403 +
1.404 + switch(step) {
1.405 + default: {
1.406 + T=(step>=4)?(sincos_lookup0+(step>>1)):sincos_lookup1;
1.407 + do{
1.408 + oX1-=4;
1.409 + XPROD31( iX[0], -iX[1], T[0], T[1], &oX1[3], &oX2[0] ); T+=step;
1.410 + XPROD31( iX[2], -iX[3], T[0], T[1], &oX1[2], &oX2[1] ); T+=step;
1.411 + XPROD31( iX[4], -iX[5], T[0], T[1], &oX1[1], &oX2[2] ); T+=step;
1.412 + XPROD31( iX[6], -iX[7], T[0], T[1], &oX1[0], &oX2[3] ); T+=step;
1.413 + oX2+=4;
1.414 + iX+=8;
1.415 + }while(iX<oX1);
1.416 + break;
1.417 + }
1.418 +
1.419 + case 1: {
1.420 + /* linear interpolation between table values: offset=0.5, step=1 */
1.421 + REG_TYPE t0,t1,v0,v1;
1.422 + T = sincos_lookup0;
1.423 + V = sincos_lookup1;
1.424 + t0 = (*T++)>>1;
1.425 + t1 = (*T++)>>1;
1.426 + do{
1.427 + oX1-=4;
1.428 +
1.429 + t0 += (v0 = (*V++)>>1);
1.430 + t1 += (v1 = (*V++)>>1);
1.431 + XPROD31( iX[0], -iX[1], t0, t1, &oX1[3], &oX2[0] );
1.432 + v0 += (t0 = (*T++)>>1);
1.433 + v1 += (t1 = (*T++)>>1);
1.434 + XPROD31( iX[2], -iX[3], v0, v1, &oX1[2], &oX2[1] );
1.435 + t0 += (v0 = (*V++)>>1);
1.436 + t1 += (v1 = (*V++)>>1);
1.437 + XPROD31( iX[4], -iX[5], t0, t1, &oX1[1], &oX2[2] );
1.438 + v0 += (t0 = (*T++)>>1);
1.439 + v1 += (t1 = (*T++)>>1);
1.440 + XPROD31( iX[6], -iX[7], v0, v1, &oX1[0], &oX2[3] );
1.441 +
1.442 + oX2+=4;
1.443 + iX+=8;
1.444 + }while(iX<oX1);
1.445 + break;
1.446 + }
1.447 +
1.448 + case 0: {
1.449 + /* linear interpolation between table values: offset=0.25, step=0.5 */
1.450 + REG_TYPE t0,t1,v0,v1,q0,q1;
1.451 + T = sincos_lookup0;
1.452 + V = sincos_lookup1;
1.453 + t0 = *T++;
1.454 + t1 = *T++;
1.455 + do{
1.456 + oX1-=4;
1.457 +
1.458 + v0 = *V++;
1.459 + v1 = *V++;
1.460 + t0 += (q0 = (v0-t0)>>2);
1.461 + t1 += (q1 = (v1-t1)>>2);
1.462 + XPROD31( iX[0], -iX[1], t0, t1, &oX1[3], &oX2[0] );
1.463 + t0 = v0-q0;
1.464 + t1 = v1-q1;
1.465 + XPROD31( iX[2], -iX[3], t0, t1, &oX1[2], &oX2[1] );
1.466 +
1.467 + t0 = *T++;
1.468 + t1 = *T++;
1.469 + v0 += (q0 = (t0-v0)>>2);
1.470 + v1 += (q1 = (t1-v1)>>2);
1.471 + XPROD31( iX[4], -iX[5], v0, v1, &oX1[1], &oX2[2] );
1.472 + v0 = t0-q0;
1.473 + v1 = t1-q1;
1.474 + XPROD31( iX[6], -iX[7], v0, v1, &oX1[0], &oX2[3] );
1.475 +
1.476 + oX2+=4;
1.477 + iX+=8;
1.478 + }while(iX<oX1);
1.479 + break;
1.480 + }
1.481 + }
1.482 +
1.483 + iX=out+n2+n4;
1.484 + oX1=out+n4;
1.485 + oX2=oX1;
1.486 +
1.487 + do{
1.488 + oX1-=4;
1.489 + iX-=4;
1.490 +
1.491 + oX2[0] = -(oX1[3] = iX[3]);
1.492 + oX2[1] = -(oX1[2] = iX[2]);
1.493 + oX2[2] = -(oX1[1] = iX[1]);
1.494 + oX2[3] = -(oX1[0] = iX[0]);
1.495 +
1.496 + oX2+=4;
1.497 + }while(oX2<iX);
1.498 +
1.499 + iX=out+n2+n4;
1.500 + oX1=out+n2+n4;
1.501 + oX2=out+n2;
1.502 +
1.503 + do{
1.504 + oX1-=4;
1.505 + oX1[0]= iX[3];
1.506 + oX1[1]= iX[2];
1.507 + oX1[2]= iX[1];
1.508 + oX1[3]= iX[0];
1.509 + iX+=4;
1.510 + }while(oX1>oX2);
1.511 + }
1.512 +}
1.513 +
