ffmpeg / libswscale / utils.c @ 9c158e49
History | View | Annotate | Download (56.7 KB)
| 1 |
/*
|
|---|---|
| 2 |
* Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
|
| 3 |
*
|
| 4 |
* This file is part of FFmpeg.
|
| 5 |
*
|
| 6 |
* FFmpeg is free software; you can redistribute it and/or
|
| 7 |
* modify it under the terms of the GNU Lesser General Public
|
| 8 |
* License as published by the Free Software Foundation; either
|
| 9 |
* version 2.1 of the License, or (at your option) any later version.
|
| 10 |
*
|
| 11 |
* FFmpeg is distributed in the hope that it will be useful,
|
| 12 |
* but WITHOUT ANY WARRANTY; without even the implied warranty of
|
| 13 |
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
| 14 |
* Lesser General Public License for more details.
|
| 15 |
*
|
| 16 |
* You should have received a copy of the GNU Lesser General Public
|
| 17 |
* License along with FFmpeg; if not, write to the Free Software
|
| 18 |
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
| 19 |
*/
|
| 20 |
|
| 21 |
#define _SVID_SOURCE //needed for MAP_ANONYMOUS |
| 22 |
#define _DARWIN_C_SOURCE // needed for MAP_ANON |
| 23 |
#include <inttypes.h> |
| 24 |
#include <string.h> |
| 25 |
#include <math.h> |
| 26 |
#include <stdio.h> |
| 27 |
#include "config.h" |
| 28 |
#include <assert.h> |
| 29 |
#if HAVE_SYS_MMAN_H
|
| 30 |
#include <sys/mman.h> |
| 31 |
#if defined(MAP_ANON) && !defined(MAP_ANONYMOUS)
|
| 32 |
#define MAP_ANONYMOUS MAP_ANON
|
| 33 |
#endif
|
| 34 |
#endif
|
| 35 |
#if HAVE_VIRTUALALLOC
|
| 36 |
#define WIN32_LEAN_AND_MEAN
|
| 37 |
#include <windows.h> |
| 38 |
#endif
|
| 39 |
#include "swscale.h" |
| 40 |
#include "swscale_internal.h" |
| 41 |
#include "rgb2rgb.h" |
| 42 |
#include "libavutil/intreadwrite.h" |
| 43 |
#include "libavutil/x86_cpu.h" |
| 44 |
#include "libavutil/avutil.h" |
| 45 |
#include "libavutil/bswap.h" |
| 46 |
#include "libavutil/opt.h" |
| 47 |
#include "libavutil/pixdesc.h" |
| 48 |
|
| 49 |
unsigned swscale_version(void) |
| 50 |
{
|
| 51 |
return LIBSWSCALE_VERSION_INT;
|
| 52 |
} |
| 53 |
|
| 54 |
const char *swscale_configuration(void) |
| 55 |
{
|
| 56 |
return FFMPEG_CONFIGURATION;
|
| 57 |
} |
| 58 |
|
| 59 |
const char *swscale_license(void) |
| 60 |
{
|
| 61 |
#define LICENSE_PREFIX "libswscale license: " |
| 62 |
return LICENSE_PREFIX FFMPEG_LICENSE + sizeof(LICENSE_PREFIX) - 1; |
| 63 |
} |
| 64 |
|
| 65 |
#define RET 0xC3 //near return opcode for x86 |
| 66 |
|
| 67 |
#define isSupportedIn(x) ( \
|
| 68 |
(x)==PIX_FMT_YUV420P \ |
| 69 |
|| (x)==PIX_FMT_YUVA420P \ |
| 70 |
|| (x)==PIX_FMT_YUYV422 \ |
| 71 |
|| (x)==PIX_FMT_UYVY422 \ |
| 72 |
|| (x)==PIX_FMT_RGB48BE \ |
| 73 |
|| (x)==PIX_FMT_RGB48LE \ |
| 74 |
|| (x)==PIX_FMT_RGB32 \ |
| 75 |
|| (x)==PIX_FMT_RGB32_1 \ |
| 76 |
|| (x)==PIX_FMT_BGR24 \ |
| 77 |
|| (x)==PIX_FMT_BGR565 \ |
| 78 |
|| (x)==PIX_FMT_BGR555 \ |
| 79 |
|| (x)==PIX_FMT_BGR32 \ |
| 80 |
|| (x)==PIX_FMT_BGR32_1 \ |
| 81 |
|| (x)==PIX_FMT_RGB24 \ |
| 82 |
|| (x)==PIX_FMT_RGB565 \ |
| 83 |
|| (x)==PIX_FMT_RGB555 \ |
| 84 |
|| (x)==PIX_FMT_GRAY8 \ |
| 85 |
|| (x)==PIX_FMT_Y400A \ |
| 86 |
|| (x)==PIX_FMT_YUV410P \ |
| 87 |
|| (x)==PIX_FMT_YUV440P \ |
| 88 |
|| (x)==PIX_FMT_NV12 \ |
| 89 |
|| (x)==PIX_FMT_NV21 \ |
| 90 |
|| (x)==PIX_FMT_GRAY16BE \ |
| 91 |
|| (x)==PIX_FMT_GRAY16LE \ |
| 92 |
|| (x)==PIX_FMT_YUV444P \ |
| 93 |
|| (x)==PIX_FMT_YUV422P \ |
| 94 |
|| (x)==PIX_FMT_YUV411P \ |
| 95 |
|| (x)==PIX_FMT_YUVJ420P \ |
| 96 |
|| (x)==PIX_FMT_YUVJ422P \ |
| 97 |
|| (x)==PIX_FMT_YUVJ440P \ |
| 98 |
|| (x)==PIX_FMT_YUVJ444P \ |
| 99 |
|| (x)==PIX_FMT_PAL8 \ |
| 100 |
|| (x)==PIX_FMT_BGR8 \ |
| 101 |
|| (x)==PIX_FMT_RGB8 \ |
| 102 |
|| (x)==PIX_FMT_BGR4_BYTE \ |
| 103 |
|| (x)==PIX_FMT_RGB4_BYTE \ |
| 104 |
|| (x)==PIX_FMT_YUV440P \ |
| 105 |
|| (x)==PIX_FMT_MONOWHITE \ |
| 106 |
|| (x)==PIX_FMT_MONOBLACK \ |
| 107 |
|| (x)==PIX_FMT_YUV420P16LE \ |
| 108 |
|| (x)==PIX_FMT_YUV422P16LE \ |
| 109 |
|| (x)==PIX_FMT_YUV444P16LE \ |
| 110 |
|| (x)==PIX_FMT_YUV420P16BE \ |
| 111 |
|| (x)==PIX_FMT_YUV422P16BE \ |
| 112 |
|| (x)==PIX_FMT_YUV444P16BE \ |
| 113 |
) |
| 114 |
|
| 115 |
int sws_isSupportedInput(enum PixelFormat pix_fmt) |
| 116 |
{
|
| 117 |
return isSupportedIn(pix_fmt);
|
| 118 |
} |
| 119 |
|
| 120 |
#define isSupportedOut(x) ( \
|
| 121 |
(x)==PIX_FMT_YUV420P \ |
| 122 |
|| (x)==PIX_FMT_YUVA420P \ |
| 123 |
|| (x)==PIX_FMT_YUYV422 \ |
| 124 |
|| (x)==PIX_FMT_UYVY422 \ |
| 125 |
|| (x)==PIX_FMT_YUV444P \ |
| 126 |
|| (x)==PIX_FMT_YUV422P \ |
| 127 |
|| (x)==PIX_FMT_YUV411P \ |
| 128 |
|| (x)==PIX_FMT_YUVJ420P \ |
| 129 |
|| (x)==PIX_FMT_YUVJ422P \ |
| 130 |
|| (x)==PIX_FMT_YUVJ440P \ |
| 131 |
|| (x)==PIX_FMT_YUVJ444P \ |
| 132 |
|| isAnyRGB(x) \ |
| 133 |
|| (x)==PIX_FMT_NV12 \ |
| 134 |
|| (x)==PIX_FMT_NV21 \ |
| 135 |
|| (x)==PIX_FMT_GRAY16BE \ |
| 136 |
|| (x)==PIX_FMT_GRAY16LE \ |
| 137 |
|| (x)==PIX_FMT_GRAY8 \ |
| 138 |
|| (x)==PIX_FMT_YUV410P \ |
| 139 |
|| (x)==PIX_FMT_YUV440P \ |
| 140 |
|| (x)==PIX_FMT_YUV420P16LE \ |
| 141 |
|| (x)==PIX_FMT_YUV422P16LE \ |
| 142 |
|| (x)==PIX_FMT_YUV444P16LE \ |
| 143 |
|| (x)==PIX_FMT_YUV420P16BE \ |
| 144 |
|| (x)==PIX_FMT_YUV422P16BE \ |
| 145 |
|| (x)==PIX_FMT_YUV444P16BE \ |
| 146 |
) |
| 147 |
|
| 148 |
int sws_isSupportedOutput(enum PixelFormat pix_fmt) |
| 149 |
{
|
| 150 |
return isSupportedOut(pix_fmt);
|
| 151 |
} |
| 152 |
|
| 153 |
extern const int32_t ff_yuv2rgb_coeffs[8][4]; |
| 154 |
|
| 155 |
const char *sws_format_name(enum PixelFormat format) |
| 156 |
{
|
| 157 |
if ((unsigned)format < PIX_FMT_NB && av_pix_fmt_descriptors[format].name) |
| 158 |
return av_pix_fmt_descriptors[format].name;
|
| 159 |
else
|
| 160 |
return "Unknown format"; |
| 161 |
} |
| 162 |
|
| 163 |
static double getSplineCoeff(double a, double b, double c, double d, double dist) |
| 164 |
{
|
| 165 |
// printf("%f %f %f %f %f\n", a,b,c,d,dist);
|
| 166 |
if (dist<=1.0) return ((d*dist + c)*dist + b)*dist +a; |
| 167 |
else return getSplineCoeff( 0.0, |
| 168 |
b+ 2.0*c + 3.0*d, |
| 169 |
c + 3.0*d, |
| 170 |
-b- 3.0*c - 6.0*d, |
| 171 |
dist-1.0); |
| 172 |
} |
| 173 |
|
| 174 |
static int initFilter(int16_t **outFilter, int16_t **filterPos, int *outFilterSize, int xInc, |
| 175 |
int srcW, int dstW, int filterAlign, int one, int flags, |
| 176 |
SwsVector *srcFilter, SwsVector *dstFilter, double param[2]) |
| 177 |
{
|
| 178 |
int i;
|
| 179 |
int filterSize;
|
| 180 |
int filter2Size;
|
| 181 |
int minFilterSize;
|
| 182 |
int64_t *filter=NULL;
|
| 183 |
int64_t *filter2=NULL;
|
| 184 |
const int64_t fone= 1LL<<54; |
| 185 |
int ret= -1; |
| 186 |
#if ARCH_X86
|
| 187 |
if (flags & SWS_CPU_CAPS_MMX)
|
| 188 |
__asm__ volatile("emms\n\t"::: "memory"); //FIXME this should not be required but it IS (even for non-MMX versions) |
| 189 |
#endif
|
| 190 |
|
| 191 |
// NOTE: the +1 is for the MMX scaler which reads over the end
|
| 192 |
FF_ALLOC_OR_GOTO(NULL, *filterPos, (dstW+1)*sizeof(int16_t), fail); |
| 193 |
|
| 194 |
if (FFABS(xInc - 0x10000) <10) { // unscaled |
| 195 |
int i;
|
| 196 |
filterSize= 1;
|
| 197 |
FF_ALLOCZ_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail); |
| 198 |
|
| 199 |
for (i=0; i<dstW; i++) { |
| 200 |
filter[i*filterSize]= fone; |
| 201 |
(*filterPos)[i]=i; |
| 202 |
} |
| 203 |
|
| 204 |
} else if (flags&SWS_POINT) { // lame looking point sampling mode |
| 205 |
int i;
|
| 206 |
int xDstInSrc;
|
| 207 |
filterSize= 1;
|
| 208 |
FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail); |
| 209 |
|
| 210 |
xDstInSrc= xInc/2 - 0x8000; |
| 211 |
for (i=0; i<dstW; i++) { |
| 212 |
int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16; |
| 213 |
|
| 214 |
(*filterPos)[i]= xx; |
| 215 |
filter[i]= fone; |
| 216 |
xDstInSrc+= xInc; |
| 217 |
} |
| 218 |
} else if ((xInc <= (1<<16) && (flags&SWS_AREA)) || (flags&SWS_FAST_BILINEAR)) { // bilinear upscale |
| 219 |
int i;
|
| 220 |
int xDstInSrc;
|
| 221 |
filterSize= 2;
|
| 222 |
FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail); |
| 223 |
|
| 224 |
xDstInSrc= xInc/2 - 0x8000; |
| 225 |
for (i=0; i<dstW; i++) { |
| 226 |
int xx= (xDstInSrc - ((filterSize-1)<<15) + (1<<15))>>16; |
| 227 |
int j;
|
| 228 |
|
| 229 |
(*filterPos)[i]= xx; |
| 230 |
//bilinear upscale / linear interpolate / area averaging
|
| 231 |
for (j=0; j<filterSize; j++) { |
| 232 |
int64_t coeff= fone - FFABS((xx<<16) - xDstInSrc)*(fone>>16); |
| 233 |
if (coeff<0) coeff=0; |
| 234 |
filter[i*filterSize + j]= coeff; |
| 235 |
xx++; |
| 236 |
} |
| 237 |
xDstInSrc+= xInc; |
| 238 |
} |
| 239 |
} else {
|
| 240 |
int xDstInSrc;
|
| 241 |
int sizeFactor;
|
| 242 |
|
| 243 |
if (flags&SWS_BICUBIC) sizeFactor= 4; |
| 244 |
else if (flags&SWS_X) sizeFactor= 8; |
| 245 |
else if (flags&SWS_AREA) sizeFactor= 1; //downscale only, for upscale it is bilinear |
| 246 |
else if (flags&SWS_GAUSS) sizeFactor= 8; // infinite ;) |
| 247 |
else if (flags&SWS_LANCZOS) sizeFactor= param[0] != SWS_PARAM_DEFAULT ? ceil(2*param[0]) : 6; |
| 248 |
else if (flags&SWS_SINC) sizeFactor= 20; // infinite ;) |
| 249 |
else if (flags&SWS_SPLINE) sizeFactor= 20; // infinite ;) |
| 250 |
else if (flags&SWS_BILINEAR) sizeFactor= 2; |
| 251 |
else {
|
| 252 |
sizeFactor= 0; //GCC warning killer |
| 253 |
assert(0);
|
| 254 |
} |
| 255 |
|
| 256 |
if (xInc <= 1<<16) filterSize= 1 + sizeFactor; // upscale |
| 257 |
else filterSize= 1 + (sizeFactor*srcW + dstW - 1)/ dstW; |
| 258 |
|
| 259 |
if (filterSize > srcW-2) filterSize=srcW-2; |
| 260 |
|
| 261 |
FF_ALLOC_OR_GOTO(NULL, filter, dstW*sizeof(*filter)*filterSize, fail); |
| 262 |
|
| 263 |
xDstInSrc= xInc - 0x10000;
|
| 264 |
for (i=0; i<dstW; i++) { |
| 265 |
int xx= (xDstInSrc - ((filterSize-2)<<16)) / (1<<17); |
| 266 |
int j;
|
| 267 |
(*filterPos)[i]= xx; |
| 268 |
for (j=0; j<filterSize; j++) { |
| 269 |
int64_t d= ((int64_t)FFABS((xx<<17) - xDstInSrc))<<13; |
| 270 |
double floatd;
|
| 271 |
int64_t coeff; |
| 272 |
|
| 273 |
if (xInc > 1<<16) |
| 274 |
d= d*dstW/srcW; |
| 275 |
floatd= d * (1.0/(1<<30)); |
| 276 |
|
| 277 |
if (flags & SWS_BICUBIC) {
|
| 278 |
int64_t B= (param[0] != SWS_PARAM_DEFAULT ? param[0] : 0) * (1<<24); |
| 279 |
int64_t C= (param[1] != SWS_PARAM_DEFAULT ? param[1] : 0.6) * (1<<24); |
| 280 |
int64_t dd = ( d*d)>>30;
|
| 281 |
int64_t ddd= (dd*d)>>30;
|
| 282 |
|
| 283 |
if (d < 1LL<<30) |
| 284 |
coeff = (12*(1<<24)-9*B-6*C)*ddd + (-18*(1<<24)+12*B+6*C)*dd + (6*(1<<24)-2*B)*(1<<30); |
| 285 |
else if (d < 1LL<<31) |
| 286 |
coeff = (-B-6*C)*ddd + (6*B+30*C)*dd + (-12*B-48*C)*d + (8*B+24*C)*(1<<30); |
| 287 |
else
|
| 288 |
coeff=0.0; |
| 289 |
coeff *= fone>>(30+24); |
| 290 |
} |
| 291 |
/* else if (flags & SWS_X) {
|
| 292 |
double p= param ? param*0.01 : 0.3;
|
| 293 |
coeff = d ? sin(d*M_PI)/(d*M_PI) : 1.0;
|
| 294 |
coeff*= pow(2.0, - p*d*d);
|
| 295 |
}*/
|
| 296 |
else if (flags & SWS_X) { |
| 297 |
double A= param[0] != SWS_PARAM_DEFAULT ? param[0] : 1.0; |
| 298 |
double c;
|
| 299 |
|
| 300 |
if (floatd<1.0) |
| 301 |
c = cos(floatd*M_PI); |
| 302 |
else
|
| 303 |
c=-1.0; |
| 304 |
if (c<0.0) c= -pow(-c, A); |
| 305 |
else c= pow( c, A);
|
| 306 |
coeff= (c*0.5 + 0.5)*fone; |
| 307 |
} else if (flags & SWS_AREA) { |
| 308 |
int64_t d2= d - (1<<29); |
| 309 |
if (d2*xInc < -(1LL<<(29+16))) coeff= 1.0 * (1LL<<(30+16)); |
| 310 |
else if (d2*xInc < (1LL<<(29+16))) coeff= -d2*xInc + (1LL<<(29+16)); |
| 311 |
else coeff=0.0; |
| 312 |
coeff *= fone>>(30+16); |
| 313 |
} else if (flags & SWS_GAUSS) { |
| 314 |
double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0; |
| 315 |
coeff = (pow(2.0, - p*floatd*floatd))*fone; |
| 316 |
} else if (flags & SWS_SINC) { |
| 317 |
coeff = (d ? sin(floatd*M_PI)/(floatd*M_PI) : 1.0)*fone; |
| 318 |
} else if (flags & SWS_LANCZOS) { |
| 319 |
double p= param[0] != SWS_PARAM_DEFAULT ? param[0] : 3.0; |
| 320 |
coeff = (d ? sin(floatd*M_PI)*sin(floatd*M_PI/p)/(floatd*floatd*M_PI*M_PI/p) : 1.0)*fone; |
| 321 |
if (floatd>p) coeff=0; |
| 322 |
} else if (flags & SWS_BILINEAR) { |
| 323 |
coeff= (1<<30) - d; |
| 324 |
if (coeff<0) coeff=0; |
| 325 |
coeff *= fone >> 30;
|
| 326 |
} else if (flags & SWS_SPLINE) { |
| 327 |
double p=-2.196152422706632; |
| 328 |
coeff = getSplineCoeff(1.0, 0.0, p, -p-1.0, floatd) * fone; |
| 329 |
} else {
|
| 330 |
coeff= 0.0; //GCC warning killer |
| 331 |
assert(0);
|
| 332 |
} |
| 333 |
|
| 334 |
filter[i*filterSize + j]= coeff; |
| 335 |
xx++; |
| 336 |
} |
| 337 |
xDstInSrc+= 2*xInc;
|
| 338 |
} |
| 339 |
} |
| 340 |
|
| 341 |
/* apply src & dst Filter to filter -> filter2
|
| 342 |
av_free(filter);
|
| 343 |
*/
|
| 344 |
assert(filterSize>0);
|
| 345 |
filter2Size= filterSize; |
| 346 |
if (srcFilter) filter2Size+= srcFilter->length - 1; |
| 347 |
if (dstFilter) filter2Size+= dstFilter->length - 1; |
| 348 |
assert(filter2Size>0);
|
| 349 |
FF_ALLOCZ_OR_GOTO(NULL, filter2, filter2Size*dstW*sizeof(*filter2), fail); |
| 350 |
|
| 351 |
for (i=0; i<dstW; i++) { |
| 352 |
int j, k;
|
| 353 |
|
| 354 |
if(srcFilter) {
|
| 355 |
for (k=0; k<srcFilter->length; k++) { |
| 356 |
for (j=0; j<filterSize; j++) |
| 357 |
filter2[i*filter2Size + k + j] += srcFilter->coeff[k]*filter[i*filterSize + j]; |
| 358 |
} |
| 359 |
} else {
|
| 360 |
for (j=0; j<filterSize; j++) |
| 361 |
filter2[i*filter2Size + j]= filter[i*filterSize + j]; |
| 362 |
} |
| 363 |
//FIXME dstFilter
|
| 364 |
|
| 365 |
(*filterPos)[i]+= (filterSize-1)/2 - (filter2Size-1)/2; |
| 366 |
} |
| 367 |
av_freep(&filter); |
| 368 |
|
| 369 |
/* try to reduce the filter-size (step1 find size and shift left) */
|
| 370 |
// Assume it is near normalized (*0.5 or *2.0 is OK but * 0.001 is not).
|
| 371 |
minFilterSize= 0;
|
| 372 |
for (i=dstW-1; i>=0; i--) { |
| 373 |
int min= filter2Size;
|
| 374 |
int j;
|
| 375 |
int64_t cutOff=0.0; |
| 376 |
|
| 377 |
/* get rid of near zero elements on the left by shifting left */
|
| 378 |
for (j=0; j<filter2Size; j++) { |
| 379 |
int k;
|
| 380 |
cutOff += FFABS(filter2[i*filter2Size]); |
| 381 |
|
| 382 |
if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break; |
| 383 |
|
| 384 |
/* preserve monotonicity because the core can't handle the filter otherwise */
|
| 385 |
if (i<dstW-1 && (*filterPos)[i] >= (*filterPos)[i+1]) break; |
| 386 |
|
| 387 |
// move filter coefficients left
|
| 388 |
for (k=1; k<filter2Size; k++) |
| 389 |
filter2[i*filter2Size + k - 1]= filter2[i*filter2Size + k];
|
| 390 |
filter2[i*filter2Size + k - 1]= 0; |
| 391 |
(*filterPos)[i]++; |
| 392 |
} |
| 393 |
|
| 394 |
cutOff=0;
|
| 395 |
/* count near zeros on the right */
|
| 396 |
for (j=filter2Size-1; j>0; j--) { |
| 397 |
cutOff += FFABS(filter2[i*filter2Size + j]); |
| 398 |
|
| 399 |
if (cutOff > SWS_MAX_REDUCE_CUTOFF*fone) break; |
| 400 |
min--; |
| 401 |
} |
| 402 |
|
| 403 |
if (min>minFilterSize) minFilterSize= min;
|
| 404 |
} |
| 405 |
|
| 406 |
if (flags & SWS_CPU_CAPS_ALTIVEC) {
|
| 407 |
// we can handle the special case 4,
|
| 408 |
// so we don't want to go to the full 8
|
| 409 |
if (minFilterSize < 5) |
| 410 |
filterAlign = 4;
|
| 411 |
|
| 412 |
// We really don't want to waste our time
|
| 413 |
// doing useless computation, so fall back on
|
| 414 |
// the scalar C code for very small filters.
|
| 415 |
// Vectorizing is worth it only if you have a
|
| 416 |
// decent-sized vector.
|
| 417 |
if (minFilterSize < 3) |
| 418 |
filterAlign = 1;
|
| 419 |
} |
| 420 |
|
| 421 |
if (flags & SWS_CPU_CAPS_MMX) {
|
| 422 |
// special case for unscaled vertical filtering
|
| 423 |
if (minFilterSize == 1 && filterAlign == 2) |
| 424 |
filterAlign= 1;
|
| 425 |
} |
| 426 |
|
| 427 |
assert(minFilterSize > 0);
|
| 428 |
filterSize= (minFilterSize +(filterAlign-1)) & (~(filterAlign-1)); |
| 429 |
assert(filterSize > 0);
|
| 430 |
filter= av_malloc(filterSize*dstW*sizeof(*filter));
|
| 431 |
if (filterSize >= MAX_FILTER_SIZE*16/((flags&SWS_ACCURATE_RND) ? APCK_SIZE : 16) || !filter) |
| 432 |
goto fail;
|
| 433 |
*outFilterSize= filterSize; |
| 434 |
|
| 435 |
if (flags&SWS_PRINT_INFO)
|
| 436 |
av_log(NULL, AV_LOG_VERBOSE, "SwScaler: reducing / aligning filtersize %d -> %d\n", filter2Size, filterSize); |
| 437 |
/* try to reduce the filter-size (step2 reduce it) */
|
| 438 |
for (i=0; i<dstW; i++) { |
| 439 |
int j;
|
| 440 |
|
| 441 |
for (j=0; j<filterSize; j++) { |
| 442 |
if (j>=filter2Size) filter[i*filterSize + j]= 0; |
| 443 |
else filter[i*filterSize + j]= filter2[i*filter2Size + j];
|
| 444 |
if((flags & SWS_BITEXACT) && j>=minFilterSize)
|
| 445 |
filter[i*filterSize + j]= 0;
|
| 446 |
} |
| 447 |
} |
| 448 |
|
| 449 |
//FIXME try to align filterPos if possible
|
| 450 |
|
| 451 |
//fix borders
|
| 452 |
for (i=0; i<dstW; i++) { |
| 453 |
int j;
|
| 454 |
if ((*filterPos)[i] < 0) { |
| 455 |
// move filter coefficients left to compensate for filterPos
|
| 456 |
for (j=1; j<filterSize; j++) { |
| 457 |
int left= FFMAX(j + (*filterPos)[i], 0); |
| 458 |
filter[i*filterSize + left] += filter[i*filterSize + j]; |
| 459 |
filter[i*filterSize + j]=0;
|
| 460 |
} |
| 461 |
(*filterPos)[i]= 0;
|
| 462 |
} |
| 463 |
|
| 464 |
if ((*filterPos)[i] + filterSize > srcW) {
|
| 465 |
int shift= (*filterPos)[i] + filterSize - srcW;
|
| 466 |
// move filter coefficients right to compensate for filterPos
|
| 467 |
for (j=filterSize-2; j>=0; j--) { |
| 468 |
int right= FFMIN(j + shift, filterSize-1); |
| 469 |
filter[i*filterSize +right] += filter[i*filterSize +j]; |
| 470 |
filter[i*filterSize +j]=0;
|
| 471 |
} |
| 472 |
(*filterPos)[i]= srcW - filterSize; |
| 473 |
} |
| 474 |
} |
| 475 |
|
| 476 |
// Note the +1 is for the MMX scaler which reads over the end
|
| 477 |
/* align at 16 for AltiVec (needed by hScale_altivec_real) */
|
| 478 |
FF_ALLOCZ_OR_GOTO(NULL, *outFilter, *outFilterSize*(dstW+1)*sizeof(int16_t), fail); |
| 479 |
|
| 480 |
/* normalize & store in outFilter */
|
| 481 |
for (i=0; i<dstW; i++) { |
| 482 |
int j;
|
| 483 |
int64_t error=0;
|
| 484 |
int64_t sum=0;
|
| 485 |
|
| 486 |
for (j=0; j<filterSize; j++) { |
| 487 |
sum+= filter[i*filterSize + j]; |
| 488 |
} |
| 489 |
sum= (sum + one/2)/ one;
|
| 490 |
for (j=0; j<*outFilterSize; j++) { |
| 491 |
int64_t v= filter[i*filterSize + j] + error; |
| 492 |
int intV= ROUNDED_DIV(v, sum);
|
| 493 |
(*outFilter)[i*(*outFilterSize) + j]= intV; |
| 494 |
error= v - intV*sum; |
| 495 |
} |
| 496 |
} |
| 497 |
|
| 498 |
(*filterPos)[dstW]= (*filterPos)[dstW-1]; // the MMX scaler will read over the end |
| 499 |
for (i=0; i<*outFilterSize; i++) { |
| 500 |
int j= dstW*(*outFilterSize);
|
| 501 |
(*outFilter)[j + i]= (*outFilter)[j + i - (*outFilterSize)]; |
| 502 |
} |
| 503 |
|
| 504 |
ret=0;
|
| 505 |
fail:
|
| 506 |
av_free(filter); |
| 507 |
av_free(filter2); |
| 508 |
return ret;
|
| 509 |
} |
| 510 |
|
| 511 |
#if ARCH_X86 && (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT)
|
| 512 |
static int initMMX2HScaler(int dstW, int xInc, uint8_t *filterCode, int16_t *filter, int32_t *filterPos, int numSplits) |
| 513 |
{
|
| 514 |
uint8_t *fragmentA; |
| 515 |
x86_reg imm8OfPShufW1A; |
| 516 |
x86_reg imm8OfPShufW2A; |
| 517 |
x86_reg fragmentLengthA; |
| 518 |
uint8_t *fragmentB; |
| 519 |
x86_reg imm8OfPShufW1B; |
| 520 |
x86_reg imm8OfPShufW2B; |
| 521 |
x86_reg fragmentLengthB; |
| 522 |
int fragmentPos;
|
| 523 |
|
| 524 |
int xpos, i;
|
| 525 |
|
| 526 |
// create an optimized horizontal scaling routine
|
| 527 |
/* This scaler is made of runtime-generated MMX2 code using specially
|
| 528 |
* tuned pshufw instructions. For every four output pixels, if four
|
| 529 |
* input pixels are enough for the fast bilinear scaling, then a chunk
|
| 530 |
* of fragmentB is used. If five input pixels are needed, then a chunk
|
| 531 |
* of fragmentA is used.
|
| 532 |
*/
|
| 533 |
|
| 534 |
//code fragment
|
| 535 |
|
| 536 |
__asm__ volatile(
|
| 537 |
"jmp 9f \n\t"
|
| 538 |
// Begin
|
| 539 |
"0: \n\t"
|
| 540 |
"movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t" |
| 541 |
"movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t" |
| 542 |
"movd 1(%%"REG_c", %%"REG_S"), %%mm1 \n\t" |
| 543 |
"punpcklbw %%mm7, %%mm1 \n\t"
|
| 544 |
"punpcklbw %%mm7, %%mm0 \n\t"
|
| 545 |
"pshufw $0xFF, %%mm1, %%mm1 \n\t"
|
| 546 |
"1: \n\t"
|
| 547 |
"pshufw $0xFF, %%mm0, %%mm0 \n\t"
|
| 548 |
"2: \n\t"
|
| 549 |
"psubw %%mm1, %%mm0 \n\t"
|
| 550 |
"movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t" |
| 551 |
"pmullw %%mm3, %%mm0 \n\t"
|
| 552 |
"psllw $7, %%mm1 \n\t"
|
| 553 |
"paddw %%mm1, %%mm0 \n\t"
|
| 554 |
|
| 555 |
"movq %%mm0, (%%"REG_D", %%"REG_a") \n\t" |
| 556 |
|
| 557 |
"add $8, %%"REG_a" \n\t" |
| 558 |
// End
|
| 559 |
"9: \n\t"
|
| 560 |
// "int $3 \n\t"
|
| 561 |
"lea " LOCAL_MANGLE(0b) ", %0 \n\t" |
| 562 |
"lea " LOCAL_MANGLE(1b) ", %1 \n\t" |
| 563 |
"lea " LOCAL_MANGLE(2b) ", %2 \n\t" |
| 564 |
"dec %1 \n\t"
|
| 565 |
"dec %2 \n\t"
|
| 566 |
"sub %0, %1 \n\t"
|
| 567 |
"sub %0, %2 \n\t"
|
| 568 |
"lea " LOCAL_MANGLE(9b) ", %3 \n\t" |
| 569 |
"sub %0, %3 \n\t"
|
| 570 |
|
| 571 |
|
| 572 |
:"=r" (fragmentA), "=r" (imm8OfPShufW1A), "=r" (imm8OfPShufW2A), |
| 573 |
"=r" (fragmentLengthA)
|
| 574 |
); |
| 575 |
|
| 576 |
__asm__ volatile(
|
| 577 |
"jmp 9f \n\t"
|
| 578 |
// Begin
|
| 579 |
"0: \n\t"
|
| 580 |
"movq (%%"REG_d", %%"REG_a"), %%mm3 \n\t" |
| 581 |
"movd (%%"REG_c", %%"REG_S"), %%mm0 \n\t" |
| 582 |
"punpcklbw %%mm7, %%mm0 \n\t"
|
| 583 |
"pshufw $0xFF, %%mm0, %%mm1 \n\t"
|
| 584 |
"1: \n\t"
|
| 585 |
"pshufw $0xFF, %%mm0, %%mm0 \n\t"
|
| 586 |
"2: \n\t"
|
| 587 |
"psubw %%mm1, %%mm0 \n\t"
|
| 588 |
"movl 8(%%"REG_b", %%"REG_a"), %%esi \n\t" |
| 589 |
"pmullw %%mm3, %%mm0 \n\t"
|
| 590 |
"psllw $7, %%mm1 \n\t"
|
| 591 |
"paddw %%mm1, %%mm0 \n\t"
|
| 592 |
|
| 593 |
"movq %%mm0, (%%"REG_D", %%"REG_a") \n\t" |
| 594 |
|
| 595 |
"add $8, %%"REG_a" \n\t" |
| 596 |
// End
|
| 597 |
"9: \n\t"
|
| 598 |
// "int $3 \n\t"
|
| 599 |
"lea " LOCAL_MANGLE(0b) ", %0 \n\t" |
| 600 |
"lea " LOCAL_MANGLE(1b) ", %1 \n\t" |
| 601 |
"lea " LOCAL_MANGLE(2b) ", %2 \n\t" |
| 602 |
"dec %1 \n\t"
|
| 603 |
"dec %2 \n\t"
|
| 604 |
"sub %0, %1 \n\t"
|
| 605 |
"sub %0, %2 \n\t"
|
| 606 |
"lea " LOCAL_MANGLE(9b) ", %3 \n\t" |
| 607 |
"sub %0, %3 \n\t"
|
| 608 |
|
| 609 |
|
| 610 |
:"=r" (fragmentB), "=r" (imm8OfPShufW1B), "=r" (imm8OfPShufW2B), |
| 611 |
"=r" (fragmentLengthB)
|
| 612 |
); |
| 613 |
|
| 614 |
xpos= 0; //lumXInc/2 - 0x8000; // difference between pixel centers |
| 615 |
fragmentPos=0;
|
| 616 |
|
| 617 |
for (i=0; i<dstW/numSplits; i++) { |
| 618 |
int xx=xpos>>16; |
| 619 |
|
| 620 |
if ((i&3) == 0) { |
| 621 |
int a=0; |
| 622 |
int b=((xpos+xInc)>>16) - xx; |
| 623 |
int c=((xpos+xInc*2)>>16) - xx; |
| 624 |
int d=((xpos+xInc*3)>>16) - xx; |
| 625 |
int inc = (d+1<4); |
| 626 |
uint8_t *fragment = (d+1<4) ? fragmentB : fragmentA; |
| 627 |
x86_reg imm8OfPShufW1 = (d+1<4) ? imm8OfPShufW1B : imm8OfPShufW1A; |
| 628 |
x86_reg imm8OfPShufW2 = (d+1<4) ? imm8OfPShufW2B : imm8OfPShufW2A; |
| 629 |
x86_reg fragmentLength = (d+1<4) ? fragmentLengthB : fragmentLengthA; |
| 630 |
int maxShift= 3-(d+inc); |
| 631 |
int shift=0; |
| 632 |
|
| 633 |
if (filterCode) {
|
| 634 |
filter[i ] = (( xpos & 0xFFFF) ^ 0xFFFF)>>9; |
| 635 |
filter[i+1] = (((xpos+xInc ) & 0xFFFF) ^ 0xFFFF)>>9; |
| 636 |
filter[i+2] = (((xpos+xInc*2) & 0xFFFF) ^ 0xFFFF)>>9; |
| 637 |
filter[i+3] = (((xpos+xInc*3) & 0xFFFF) ^ 0xFFFF)>>9; |
| 638 |
filterPos[i/2]= xx;
|
| 639 |
|
| 640 |
memcpy(filterCode + fragmentPos, fragment, fragmentLength); |
| 641 |
|
| 642 |
filterCode[fragmentPos + imm8OfPShufW1]= |
| 643 |
(a+inc) | ((b+inc)<<2) | ((c+inc)<<4) | ((d+inc)<<6); |
| 644 |
filterCode[fragmentPos + imm8OfPShufW2]= |
| 645 |
a | (b<<2) | (c<<4) | (d<<6); |
| 646 |
|
| 647 |
if (i+4-inc>=dstW) shift=maxShift; //avoid overread |
| 648 |
else if ((filterPos[i/2]&3) <= maxShift) shift=filterPos[i/2]&3; //Align |
| 649 |
|
| 650 |
if (shift && i>=shift) {
|
| 651 |
filterCode[fragmentPos + imm8OfPShufW1]+= 0x55*shift;
|
| 652 |
filterCode[fragmentPos + imm8OfPShufW2]+= 0x55*shift;
|
| 653 |
filterPos[i/2]-=shift;
|
| 654 |
} |
| 655 |
} |
| 656 |
|
| 657 |
fragmentPos+= fragmentLength; |
| 658 |
|
| 659 |
if (filterCode)
|
| 660 |
filterCode[fragmentPos]= RET; |
| 661 |
} |
| 662 |
xpos+=xInc; |
| 663 |
} |
| 664 |
if (filterCode)
|
| 665 |
filterPos[((i/2)+1)&(~1)]= xpos>>16; // needed to jump to the next part |
| 666 |
|
| 667 |
return fragmentPos + 1; |
| 668 |
} |
| 669 |
#endif /* ARCH_X86 && (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT) */ |
| 670 |
|
| 671 |
static void getSubSampleFactors(int *h, int *v, enum PixelFormat format) |
| 672 |
{
|
| 673 |
*h = av_pix_fmt_descriptors[format].log2_chroma_w; |
| 674 |
*v = av_pix_fmt_descriptors[format].log2_chroma_h; |
| 675 |
} |
| 676 |
|
| 677 |
static int update_flags_cpu(int flags); |
| 678 |
|
| 679 |
int sws_setColorspaceDetails(SwsContext *c, const int inv_table[4], int srcRange, const int table[4], int dstRange, int brightness, int contrast, int saturation) |
| 680 |
{
|
| 681 |
memcpy(c->srcColorspaceTable, inv_table, sizeof(int)*4); |
| 682 |
memcpy(c->dstColorspaceTable, table, sizeof(int)*4); |
| 683 |
|
| 684 |
c->brightness= brightness; |
| 685 |
c->contrast = contrast; |
| 686 |
c->saturation= saturation; |
| 687 |
c->srcRange = srcRange; |
| 688 |
c->dstRange = dstRange; |
| 689 |
if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1; |
| 690 |
|
| 691 |
c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->dstFormat]); |
| 692 |
c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[c->srcFormat]); |
| 693 |
c->flags = update_flags_cpu(c->flags); |
| 694 |
|
| 695 |
ff_yuv2rgb_c_init_tables(c, inv_table, srcRange, brightness, contrast, saturation); |
| 696 |
//FIXME factorize
|
| 697 |
|
| 698 |
#if HAVE_ALTIVEC
|
| 699 |
if (c->flags & SWS_CPU_CAPS_ALTIVEC)
|
| 700 |
ff_yuv2rgb_init_tables_altivec(c, inv_table, brightness, contrast, saturation); |
| 701 |
#endif
|
| 702 |
return 0; |
| 703 |
} |
| 704 |
|
| 705 |
int sws_getColorspaceDetails(SwsContext *c, int **inv_table, int *srcRange, int **table, int *dstRange, int *brightness, int *contrast, int *saturation) |
| 706 |
{
|
| 707 |
if (isYUV(c->dstFormat) || isGray(c->dstFormat)) return -1; |
| 708 |
|
| 709 |
*inv_table = c->srcColorspaceTable; |
| 710 |
*table = c->dstColorspaceTable; |
| 711 |
*srcRange = c->srcRange; |
| 712 |
*dstRange = c->dstRange; |
| 713 |
*brightness= c->brightness; |
| 714 |
*contrast = c->contrast; |
| 715 |
*saturation= c->saturation; |
| 716 |
|
| 717 |
return 0; |
| 718 |
} |
| 719 |
|
| 720 |
static int handle_jpeg(enum PixelFormat *format) |
| 721 |
{
|
| 722 |
switch (*format) {
|
| 723 |
case PIX_FMT_YUVJ420P: *format = PIX_FMT_YUV420P; return 1; |
| 724 |
case PIX_FMT_YUVJ422P: *format = PIX_FMT_YUV422P; return 1; |
| 725 |
case PIX_FMT_YUVJ444P: *format = PIX_FMT_YUV444P; return 1; |
| 726 |
case PIX_FMT_YUVJ440P: *format = PIX_FMT_YUV440P; return 1; |
| 727 |
default: return 0; |
| 728 |
} |
| 729 |
} |
| 730 |
|
| 731 |
static int update_flags_cpu(int flags) |
| 732 |
{
|
| 733 |
#if !CONFIG_RUNTIME_CPUDETECT //ensure that the flags match the compiled variant if cpudetect is off |
| 734 |
flags &= ~( SWS_CPU_CAPS_MMX |
| 735 |
|SWS_CPU_CAPS_MMX2 |
| 736 |
|SWS_CPU_CAPS_3DNOW |
| 737 |
|SWS_CPU_CAPS_SSE2 |
| 738 |
|SWS_CPU_CAPS_ALTIVEC |
| 739 |
|SWS_CPU_CAPS_BFIN); |
| 740 |
flags |= ff_hardcodedcpuflags(); |
| 741 |
#endif /* CONFIG_RUNTIME_CPUDETECT */ |
| 742 |
return flags;
|
| 743 |
} |
| 744 |
|
| 745 |
SwsContext *sws_alloc_context(void)
|
| 746 |
{
|
| 747 |
SwsContext *c= av_mallocz(sizeof(SwsContext));
|
| 748 |
|
| 749 |
c->av_class = &sws_context_class; |
| 750 |
av_opt_set_defaults(c); |
| 751 |
|
| 752 |
return c;
|
| 753 |
} |
| 754 |
|
| 755 |
int sws_init_context(SwsContext *c, SwsFilter *srcFilter, SwsFilter *dstFilter)
|
| 756 |
{
|
| 757 |
int i;
|
| 758 |
int usesVFilter, usesHFilter;
|
| 759 |
int unscaled;
|
| 760 |
SwsFilter dummyFilter= {NULL, NULL, NULL, NULL};
|
| 761 |
int srcW= c->srcW;
|
| 762 |
int srcH= c->srcH;
|
| 763 |
int dstW= c->dstW;
|
| 764 |
int dstH= c->dstH;
|
| 765 |
int flags;
|
| 766 |
enum PixelFormat srcFormat= c->srcFormat;
|
| 767 |
enum PixelFormat dstFormat= c->dstFormat;
|
| 768 |
|
| 769 |
flags= c->flags = update_flags_cpu(c->flags); |
| 770 |
#if ARCH_X86
|
| 771 |
if (flags & SWS_CPU_CAPS_MMX)
|
| 772 |
__asm__ volatile("emms\n\t"::: "memory"); |
| 773 |
#endif
|
| 774 |
if (!rgb15to16) sws_rgb2rgb_init(flags);
|
| 775 |
|
| 776 |
unscaled = (srcW == dstW && srcH == dstH); |
| 777 |
|
| 778 |
if (!isSupportedIn(srcFormat)) {
|
| 779 |
av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as input pixel format\n", sws_format_name(srcFormat)); |
| 780 |
return AVERROR(EINVAL);
|
| 781 |
} |
| 782 |
if (!isSupportedOut(dstFormat)) {
|
| 783 |
av_log(NULL, AV_LOG_ERROR, "swScaler: %s is not supported as output pixel format\n", sws_format_name(dstFormat)); |
| 784 |
return AVERROR(EINVAL);
|
| 785 |
} |
| 786 |
|
| 787 |
i= flags & ( SWS_POINT |
| 788 |
|SWS_AREA |
| 789 |
|SWS_BILINEAR |
| 790 |
|SWS_FAST_BILINEAR |
| 791 |
|SWS_BICUBIC |
| 792 |
|SWS_X |
| 793 |
|SWS_GAUSS |
| 794 |
|SWS_LANCZOS |
| 795 |
|SWS_SINC |
| 796 |
|SWS_SPLINE |
| 797 |
|SWS_BICUBLIN); |
| 798 |
if(!i || (i & (i-1))) { |
| 799 |
av_log(NULL, AV_LOG_ERROR, "swScaler: Exactly one scaler algorithm must be chosen\n"); |
| 800 |
return AVERROR(EINVAL);
|
| 801 |
} |
| 802 |
/* sanity check */
|
| 803 |
if (srcW<4 || srcH<1 || dstW<8 || dstH<1) { //FIXME check if these are enough and try to lowwer them after fixing the relevant parts of the code |
| 804 |
av_log(NULL, AV_LOG_ERROR, "swScaler: %dx%d -> %dx%d is invalid scaling dimension\n", |
| 805 |
srcW, srcH, dstW, dstH); |
| 806 |
return AVERROR(EINVAL);
|
| 807 |
} |
| 808 |
if(srcW > VOFW || dstW > VOFW) {
|
| 809 |
av_log(NULL, AV_LOG_ERROR, "swScaler: Compile-time maximum width is "AV_STRINGIFY(VOFW)" change VOF/VOFW and recompile\n"); |
| 810 |
return AVERROR(EINVAL);
|
| 811 |
} |
| 812 |
|
| 813 |
if (!dstFilter) dstFilter= &dummyFilter;
|
| 814 |
if (!srcFilter) srcFilter= &dummyFilter;
|
| 815 |
|
| 816 |
c->lumXInc= ((srcW<<16) + (dstW>>1))/dstW; |
| 817 |
c->lumYInc= ((srcH<<16) + (dstH>>1))/dstH; |
| 818 |
c->dstFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[dstFormat]); |
| 819 |
c->srcFormatBpp = av_get_bits_per_pixel(&av_pix_fmt_descriptors[srcFormat]); |
| 820 |
c->vRounder= 4* 0x0001000100010001ULL; |
| 821 |
|
| 822 |
usesVFilter = (srcFilter->lumV && srcFilter->lumV->length>1) ||
|
| 823 |
(srcFilter->chrV && srcFilter->chrV->length>1) ||
|
| 824 |
(dstFilter->lumV && dstFilter->lumV->length>1) ||
|
| 825 |
(dstFilter->chrV && dstFilter->chrV->length>1);
|
| 826 |
usesHFilter = (srcFilter->lumH && srcFilter->lumH->length>1) ||
|
| 827 |
(srcFilter->chrH && srcFilter->chrH->length>1) ||
|
| 828 |
(dstFilter->lumH && dstFilter->lumH->length>1) ||
|
| 829 |
(dstFilter->chrH && dstFilter->chrH->length>1);
|
| 830 |
|
| 831 |
getSubSampleFactors(&c->chrSrcHSubSample, &c->chrSrcVSubSample, srcFormat); |
| 832 |
getSubSampleFactors(&c->chrDstHSubSample, &c->chrDstVSubSample, dstFormat); |
| 833 |
|
| 834 |
// reuse chroma for 2 pixels RGB/BGR unless user wants full chroma interpolation
|
| 835 |
if (isAnyRGB(dstFormat) && !(flags&SWS_FULL_CHR_H_INT)) c->chrDstHSubSample=1; |
| 836 |
|
| 837 |
// drop some chroma lines if the user wants it
|
| 838 |
c->vChrDrop= (flags&SWS_SRC_V_CHR_DROP_MASK)>>SWS_SRC_V_CHR_DROP_SHIFT; |
| 839 |
c->chrSrcVSubSample+= c->vChrDrop; |
| 840 |
|
| 841 |
// drop every other pixel for chroma calculation unless user wants full chroma
|
| 842 |
if (isAnyRGB(srcFormat) && !(flags&SWS_FULL_CHR_H_INP)
|
| 843 |
&& srcFormat!=PIX_FMT_RGB8 && srcFormat!=PIX_FMT_BGR8 |
| 844 |
&& srcFormat!=PIX_FMT_RGB4 && srcFormat!=PIX_FMT_BGR4 |
| 845 |
&& srcFormat!=PIX_FMT_RGB4_BYTE && srcFormat!=PIX_FMT_BGR4_BYTE |
| 846 |
&& ((dstW>>c->chrDstHSubSample) <= (srcW>>1) || (flags&SWS_FAST_BILINEAR)))
|
| 847 |
c->chrSrcHSubSample=1;
|
| 848 |
|
| 849 |
// Note the -((-x)>>y) is so that we always round toward +inf.
|
| 850 |
c->chrSrcW= -((-srcW) >> c->chrSrcHSubSample); |
| 851 |
c->chrSrcH= -((-srcH) >> c->chrSrcVSubSample); |
| 852 |
c->chrDstW= -((-dstW) >> c->chrDstHSubSample); |
| 853 |
c->chrDstH= -((-dstH) >> c->chrDstVSubSample); |
| 854 |
|
| 855 |
/* unscaled special cases */
|
| 856 |
if (unscaled && !usesHFilter && !usesVFilter && (c->srcRange == c->dstRange || isAnyRGB(dstFormat))) {
|
| 857 |
ff_get_unscaled_swscale(c); |
| 858 |
|
| 859 |
if (c->swScale) {
|
| 860 |
if (flags&SWS_PRINT_INFO)
|
| 861 |
av_log(c, AV_LOG_INFO, "using unscaled %s -> %s special converter\n",
|
| 862 |
sws_format_name(srcFormat), sws_format_name(dstFormat)); |
| 863 |
return 0; |
| 864 |
} |
| 865 |
} |
| 866 |
|
| 867 |
if (flags & SWS_CPU_CAPS_MMX2) {
|
| 868 |
c->canMMX2BeUsed= (dstW >=srcW && (dstW&31)==0 && (srcW&15)==0) ? 1 : 0; |
| 869 |
if (!c->canMMX2BeUsed && dstW >=srcW && (srcW&15)==0 && (flags&SWS_FAST_BILINEAR)) { |
| 870 |
if (flags&SWS_PRINT_INFO)
|
| 871 |
av_log(c, AV_LOG_INFO, "output width is not a multiple of 32 -> no MMX2 scaler\n");
|
| 872 |
} |
| 873 |
if (usesHFilter) c->canMMX2BeUsed=0; |
| 874 |
} |
| 875 |
else
|
| 876 |
c->canMMX2BeUsed=0;
|
| 877 |
|
| 878 |
c->chrXInc= ((c->chrSrcW<<16) + (c->chrDstW>>1))/c->chrDstW; |
| 879 |
c->chrYInc= ((c->chrSrcH<<16) + (c->chrDstH>>1))/c->chrDstH; |
| 880 |
|
| 881 |
// match pixel 0 of the src to pixel 0 of dst and match pixel n-2 of src to pixel n-2 of dst
|
| 882 |
// but only for the FAST_BILINEAR mode otherwise do correct scaling
|
| 883 |
// n-2 is the last chrominance sample available
|
| 884 |
// this is not perfect, but no one should notice the difference, the more correct variant
|
| 885 |
// would be like the vertical one, but that would require some special code for the
|
| 886 |
// first and last pixel
|
| 887 |
if (flags&SWS_FAST_BILINEAR) {
|
| 888 |
if (c->canMMX2BeUsed) {
|
| 889 |
c->lumXInc+= 20;
|
| 890 |
c->chrXInc+= 20;
|
| 891 |
} |
| 892 |
//we don't use the x86 asm scaler if MMX is available
|
| 893 |
else if (flags & SWS_CPU_CAPS_MMX) { |
| 894 |
c->lumXInc = ((srcW-2)<<16)/(dstW-2) - 20; |
| 895 |
c->chrXInc = ((c->chrSrcW-2)<<16)/(c->chrDstW-2) - 20; |
| 896 |
} |
| 897 |
} |
| 898 |
|
| 899 |
/* precalculate horizontal scaler filter coefficients */
|
| 900 |
{
|
| 901 |
#if ARCH_X86 && (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT)
|
| 902 |
// can't downscale !!!
|
| 903 |
if (c->canMMX2BeUsed && (flags & SWS_FAST_BILINEAR)) {
|
| 904 |
c->lumMmx2FilterCodeSize = initMMX2HScaler( dstW, c->lumXInc, NULL, NULL, NULL, 8); |
| 905 |
c->chrMmx2FilterCodeSize = initMMX2HScaler(c->chrDstW, c->chrXInc, NULL, NULL, NULL, 4); |
| 906 |
|
| 907 |
#ifdef MAP_ANONYMOUS
|
| 908 |
c->lumMmx2FilterCode = mmap(NULL, c->lumMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); |
| 909 |
c->chrMmx2FilterCode = mmap(NULL, c->chrMmx2FilterCodeSize, PROT_READ | PROT_WRITE, MAP_PRIVATE | MAP_ANONYMOUS, -1, 0); |
| 910 |
#elif HAVE_VIRTUALALLOC
|
| 911 |
c->lumMmx2FilterCode = VirtualAlloc(NULL, c->lumMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
|
| 912 |
c->chrMmx2FilterCode = VirtualAlloc(NULL, c->chrMmx2FilterCodeSize, MEM_COMMIT, PAGE_EXECUTE_READWRITE);
|
| 913 |
#else
|
| 914 |
c->lumMmx2FilterCode = av_malloc(c->lumMmx2FilterCodeSize); |
| 915 |
c->chrMmx2FilterCode = av_malloc(c->chrMmx2FilterCodeSize); |
| 916 |
#endif
|
| 917 |
|
| 918 |
if (!c->lumMmx2FilterCode || !c->chrMmx2FilterCode)
|
| 919 |
return AVERROR(ENOMEM);
|
| 920 |
FF_ALLOCZ_OR_GOTO(c, c->hLumFilter , (dstW /8+8)*sizeof(int16_t), fail); |
| 921 |
FF_ALLOCZ_OR_GOTO(c, c->hChrFilter , (c->chrDstW /4+8)*sizeof(int16_t), fail); |
| 922 |
FF_ALLOCZ_OR_GOTO(c, c->hLumFilterPos, (dstW /2/8+8)*sizeof(int32_t), fail); |
| 923 |
FF_ALLOCZ_OR_GOTO(c, c->hChrFilterPos, (c->chrDstW/2/4+8)*sizeof(int32_t), fail); |
| 924 |
|
| 925 |
initMMX2HScaler( dstW, c->lumXInc, c->lumMmx2FilterCode, c->hLumFilter, c->hLumFilterPos, 8);
|
| 926 |
initMMX2HScaler(c->chrDstW, c->chrXInc, c->chrMmx2FilterCode, c->hChrFilter, c->hChrFilterPos, 4);
|
| 927 |
|
| 928 |
#ifdef MAP_ANONYMOUS
|
| 929 |
mprotect(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize, PROT_EXEC | PROT_READ); |
| 930 |
mprotect(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize, PROT_EXEC | PROT_READ); |
| 931 |
#endif
|
| 932 |
} else
|
| 933 |
#endif /* ARCH_X86 && (HAVE_MMX2 || CONFIG_RUNTIME_CPUDETECT) */ |
| 934 |
{
|
| 935 |
const int filterAlign= |
| 936 |
(flags & SWS_CPU_CAPS_MMX) ? 4 :
|
| 937 |
(flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
|
| 938 |
1;
|
| 939 |
|
| 940 |
if (initFilter(&c->hLumFilter, &c->hLumFilterPos, &c->hLumFilterSize, c->lumXInc,
|
| 941 |
srcW , dstW, filterAlign, 1<<14, |
| 942 |
(flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags, |
| 943 |
srcFilter->lumH, dstFilter->lumH, c->param) < 0)
|
| 944 |
goto fail;
|
| 945 |
if (initFilter(&c->hChrFilter, &c->hChrFilterPos, &c->hChrFilterSize, c->chrXInc,
|
| 946 |
c->chrSrcW, c->chrDstW, filterAlign, 1<<14, |
| 947 |
(flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags, |
| 948 |
srcFilter->chrH, dstFilter->chrH, c->param) < 0)
|
| 949 |
goto fail;
|
| 950 |
} |
| 951 |
} // initialize horizontal stuff
|
| 952 |
|
| 953 |
/* precalculate vertical scaler filter coefficients */
|
| 954 |
{
|
| 955 |
const int filterAlign= |
| 956 |
(flags & SWS_CPU_CAPS_MMX) && (flags & SWS_ACCURATE_RND) ? 2 :
|
| 957 |
(flags & SWS_CPU_CAPS_ALTIVEC) ? 8 :
|
| 958 |
1;
|
| 959 |
|
| 960 |
if (initFilter(&c->vLumFilter, &c->vLumFilterPos, &c->vLumFilterSize, c->lumYInc,
|
| 961 |
srcH , dstH, filterAlign, (1<<12), |
| 962 |
(flags&SWS_BICUBLIN) ? (flags|SWS_BICUBIC) : flags, |
| 963 |
srcFilter->lumV, dstFilter->lumV, c->param) < 0)
|
| 964 |
goto fail;
|
| 965 |
if (initFilter(&c->vChrFilter, &c->vChrFilterPos, &c->vChrFilterSize, c->chrYInc,
|
| 966 |
c->chrSrcH, c->chrDstH, filterAlign, (1<<12), |
| 967 |
(flags&SWS_BICUBLIN) ? (flags|SWS_BILINEAR) : flags, |
| 968 |
srcFilter->chrV, dstFilter->chrV, c->param) < 0)
|
| 969 |
goto fail;
|
| 970 |
|
| 971 |
#if HAVE_ALTIVEC
|
| 972 |
FF_ALLOC_OR_GOTO(c, c->vYCoeffsBank, sizeof (vector signed short)*c->vLumFilterSize*c->dstH, fail); |
| 973 |
FF_ALLOC_OR_GOTO(c, c->vCCoeffsBank, sizeof (vector signed short)*c->vChrFilterSize*c->chrDstH, fail); |
| 974 |
|
| 975 |
for (i=0;i<c->vLumFilterSize*c->dstH;i++) { |
| 976 |
int j;
|
| 977 |
short *p = (short *)&c->vYCoeffsBank[i]; |
| 978 |
for (j=0;j<8;j++) |
| 979 |
p[j] = c->vLumFilter[i]; |
| 980 |
} |
| 981 |
|
| 982 |
for (i=0;i<c->vChrFilterSize*c->chrDstH;i++) { |
| 983 |
int j;
|
| 984 |
short *p = (short *)&c->vCCoeffsBank[i]; |
| 985 |
for (j=0;j<8;j++) |
| 986 |
p[j] = c->vChrFilter[i]; |
| 987 |
} |
| 988 |
#endif
|
| 989 |
} |
| 990 |
|
| 991 |
// calculate buffer sizes so that they won't run out while handling these damn slices
|
| 992 |
c->vLumBufSize= c->vLumFilterSize; |
| 993 |
c->vChrBufSize= c->vChrFilterSize; |
| 994 |
for (i=0; i<dstH; i++) { |
| 995 |
int chrI= i*c->chrDstH / dstH;
|
| 996 |
int nextSlice= FFMAX(c->vLumFilterPos[i ] + c->vLumFilterSize - 1, |
| 997 |
((c->vChrFilterPos[chrI] + c->vChrFilterSize - 1)<<c->chrSrcVSubSample));
|
| 998 |
|
| 999 |
nextSlice>>= c->chrSrcVSubSample; |
| 1000 |
nextSlice<<= c->chrSrcVSubSample; |
| 1001 |
if (c->vLumFilterPos[i ] + c->vLumBufSize < nextSlice)
|
| 1002 |
c->vLumBufSize= nextSlice - c->vLumFilterPos[i]; |
| 1003 |
if (c->vChrFilterPos[chrI] + c->vChrBufSize < (nextSlice>>c->chrSrcVSubSample))
|
| 1004 |
c->vChrBufSize= (nextSlice>>c->chrSrcVSubSample) - c->vChrFilterPos[chrI]; |
| 1005 |
} |
| 1006 |
|
| 1007 |
// allocate pixbufs (we use dynamic allocation because otherwise we would need to
|
| 1008 |
// allocate several megabytes to handle all possible cases)
|
| 1009 |
FF_ALLOC_OR_GOTO(c, c->lumPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail); |
| 1010 |
FF_ALLOC_OR_GOTO(c, c->chrPixBuf, c->vChrBufSize*2*sizeof(int16_t*), fail); |
| 1011 |
if (CONFIG_SWSCALE_ALPHA && isALPHA(c->srcFormat) && isALPHA(c->dstFormat))
|
| 1012 |
FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf, c->vLumBufSize*2*sizeof(int16_t*), fail); |
| 1013 |
//Note we need at least one pixel more at the end because of the MMX code (just in case someone wanna replace the 4000/8000)
|
| 1014 |
/* align at 16 bytes for AltiVec */
|
| 1015 |
for (i=0; i<c->vLumBufSize; i++) { |
| 1016 |
FF_ALLOCZ_OR_GOTO(c, c->lumPixBuf[i+c->vLumBufSize], VOF+1, fail);
|
| 1017 |
c->lumPixBuf[i] = c->lumPixBuf[i+c->vLumBufSize]; |
| 1018 |
} |
| 1019 |
for (i=0; i<c->vChrBufSize; i++) { |
| 1020 |
FF_ALLOC_OR_GOTO(c, c->chrPixBuf[i+c->vChrBufSize], (VOF+1)*2, fail); |
| 1021 |
c->chrPixBuf[i] = c->chrPixBuf[i+c->vChrBufSize]; |
| 1022 |
} |
| 1023 |
if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf)
|
| 1024 |
for (i=0; i<c->vLumBufSize; i++) { |
| 1025 |
FF_ALLOCZ_OR_GOTO(c, c->alpPixBuf[i+c->vLumBufSize], VOF+1, fail);
|
| 1026 |
c->alpPixBuf[i] = c->alpPixBuf[i+c->vLumBufSize]; |
| 1027 |
} |
| 1028 |
|
| 1029 |
//try to avoid drawing green stuff between the right end and the stride end
|
| 1030 |
for (i=0; i<c->vChrBufSize; i++) memset(c->chrPixBuf[i], 64, (VOF+1)*2); |
| 1031 |
|
| 1032 |
assert(2*VOFW == VOF);
|
| 1033 |
|
| 1034 |
assert(c->chrDstH <= dstH); |
| 1035 |
|
| 1036 |
if (flags&SWS_PRINT_INFO) {
|
| 1037 |
if (flags&SWS_FAST_BILINEAR) av_log(c, AV_LOG_INFO, "FAST_BILINEAR scaler, "); |
| 1038 |
else if (flags&SWS_BILINEAR) av_log(c, AV_LOG_INFO, "BILINEAR scaler, "); |
| 1039 |
else if (flags&SWS_BICUBIC) av_log(c, AV_LOG_INFO, "BICUBIC scaler, "); |
| 1040 |
else if (flags&SWS_X) av_log(c, AV_LOG_INFO, "Experimental scaler, "); |
| 1041 |
else if (flags&SWS_POINT) av_log(c, AV_LOG_INFO, "Nearest Neighbor / POINT scaler, "); |
| 1042 |
else if (flags&SWS_AREA) av_log(c, AV_LOG_INFO, "Area Averaging scaler, "); |
| 1043 |
else if (flags&SWS_BICUBLIN) av_log(c, AV_LOG_INFO, "luma BICUBIC / chroma BILINEAR scaler, "); |
| 1044 |
else if (flags&SWS_GAUSS) av_log(c, AV_LOG_INFO, "Gaussian scaler, "); |
| 1045 |
else if (flags&SWS_SINC) av_log(c, AV_LOG_INFO, "Sinc scaler, "); |
| 1046 |
else if (flags&SWS_LANCZOS) av_log(c, AV_LOG_INFO, "Lanczos scaler, "); |
| 1047 |
else if (flags&SWS_SPLINE) av_log(c, AV_LOG_INFO, "Bicubic spline scaler, "); |
| 1048 |
else av_log(c, AV_LOG_INFO, "ehh flags invalid?! "); |
| 1049 |
|
| 1050 |
av_log(c, AV_LOG_INFO, "from %s to %s%s ",
|
| 1051 |
sws_format_name(srcFormat), |
| 1052 |
#ifdef DITHER1XBPP
|
| 1053 |
dstFormat == PIX_FMT_BGR555 || dstFormat == PIX_FMT_BGR565 || |
| 1054 |
dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE || |
| 1055 |
dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE ? "dithered " : "", |
| 1056 |
#else
|
| 1057 |
"",
|
| 1058 |
#endif
|
| 1059 |
sws_format_name(dstFormat)); |
| 1060 |
|
| 1061 |
if (flags & SWS_CPU_CAPS_MMX2) av_log(c, AV_LOG_INFO, "using MMX2\n"); |
| 1062 |
else if (flags & SWS_CPU_CAPS_3DNOW) av_log(c, AV_LOG_INFO, "using 3DNOW\n"); |
| 1063 |
else if (flags & SWS_CPU_CAPS_MMX) av_log(c, AV_LOG_INFO, "using MMX\n"); |
| 1064 |
else if (flags & SWS_CPU_CAPS_ALTIVEC) av_log(c, AV_LOG_INFO, "using AltiVec\n"); |
| 1065 |
else av_log(c, AV_LOG_INFO, "using C\n"); |
| 1066 |
|
| 1067 |
if (flags & SWS_CPU_CAPS_MMX) {
|
| 1068 |
if (c->canMMX2BeUsed && (flags&SWS_FAST_BILINEAR))
|
| 1069 |
av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR MMX2 scaler for horizontal scaling\n");
|
| 1070 |
else {
|
| 1071 |
if (c->hLumFilterSize==4) |
| 1072 |
av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal luminance scaling\n");
|
| 1073 |
else if (c->hLumFilterSize==8) |
| 1074 |
av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal luminance scaling\n");
|
| 1075 |
else
|
| 1076 |
av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal luminance scaling\n");
|
| 1077 |
|
| 1078 |
if (c->hChrFilterSize==4) |
| 1079 |
av_log(c, AV_LOG_VERBOSE, "using 4-tap MMX scaler for horizontal chrominance scaling\n");
|
| 1080 |
else if (c->hChrFilterSize==8) |
| 1081 |
av_log(c, AV_LOG_VERBOSE, "using 8-tap MMX scaler for horizontal chrominance scaling\n");
|
| 1082 |
else
|
| 1083 |
av_log(c, AV_LOG_VERBOSE, "using n-tap MMX scaler for horizontal chrominance scaling\n");
|
| 1084 |
} |
| 1085 |
} else {
|
| 1086 |
#if ARCH_X86
|
| 1087 |
av_log(c, AV_LOG_VERBOSE, "using x86 asm scaler for horizontal scaling\n");
|
| 1088 |
#else
|
| 1089 |
if (flags & SWS_FAST_BILINEAR)
|
| 1090 |
av_log(c, AV_LOG_VERBOSE, "using FAST_BILINEAR C scaler for horizontal scaling\n");
|
| 1091 |
else
|
| 1092 |
av_log(c, AV_LOG_VERBOSE, "using C scaler for horizontal scaling\n");
|
| 1093 |
#endif
|
| 1094 |
} |
| 1095 |
if (isPlanarYUV(dstFormat)) {
|
| 1096 |
if (c->vLumFilterSize==1) |
| 1097 |
av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"); |
| 1098 |
else
|
| 1099 |
av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (YV12 like)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"); |
| 1100 |
} else {
|
| 1101 |
if (c->vLumFilterSize==1 && c->vChrFilterSize==2) |
| 1102 |
av_log(c, AV_LOG_VERBOSE, "using 1-tap %s \"scaler\" for vertical luminance scaling (BGR)\n"
|
| 1103 |
" 2-tap scaler for vertical chrominance scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"); |
| 1104 |
else if (c->vLumFilterSize==2 && c->vChrFilterSize==2) |
| 1105 |
av_log(c, AV_LOG_VERBOSE, "using 2-tap linear %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"); |
| 1106 |
else
|
| 1107 |
av_log(c, AV_LOG_VERBOSE, "using n-tap %s scaler for vertical scaling (BGR)\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"); |
| 1108 |
} |
| 1109 |
|
| 1110 |
if (dstFormat==PIX_FMT_BGR24)
|
| 1111 |
av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR24 converter\n",
|
| 1112 |
(flags & SWS_CPU_CAPS_MMX2) ? "MMX2" : ((flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C")); |
| 1113 |
else if (dstFormat==PIX_FMT_RGB32) |
| 1114 |
av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR32 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"); |
| 1115 |
else if (dstFormat==PIX_FMT_BGR565) |
| 1116 |
av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR16 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"); |
| 1117 |
else if (dstFormat==PIX_FMT_BGR555) |
| 1118 |
av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR15 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"); |
| 1119 |
else if (dstFormat == PIX_FMT_RGB444BE || dstFormat == PIX_FMT_RGB444LE || |
| 1120 |
dstFormat == PIX_FMT_BGR444BE || dstFormat == PIX_FMT_BGR444LE) |
| 1121 |
av_log(c, AV_LOG_VERBOSE, "using %s YV12->BGR12 converter\n", (flags & SWS_CPU_CAPS_MMX) ? "MMX" : "C"); |
| 1122 |
|
| 1123 |
av_log(c, AV_LOG_VERBOSE, "%dx%d -> %dx%d\n", srcW, srcH, dstW, dstH);
|
| 1124 |
av_log(c, AV_LOG_DEBUG, "lum srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
|
| 1125 |
c->srcW, c->srcH, c->dstW, c->dstH, c->lumXInc, c->lumYInc); |
| 1126 |
av_log(c, AV_LOG_DEBUG, "chr srcW=%d srcH=%d dstW=%d dstH=%d xInc=%d yInc=%d\n",
|
| 1127 |
c->chrSrcW, c->chrSrcH, c->chrDstW, c->chrDstH, c->chrXInc, c->chrYInc); |
| 1128 |
} |
| 1129 |
|
| 1130 |
c->swScale= ff_getSwsFunc(c); |
| 1131 |
return 0; |
| 1132 |
fail: //FIXME replace things by appropriate error codes |
| 1133 |
return -1; |
| 1134 |
} |
| 1135 |
|
| 1136 |
#if FF_API_SWS_GETCONTEXT
|
| 1137 |
SwsContext *sws_getContext(int srcW, int srcH, enum PixelFormat srcFormat, |
| 1138 |
int dstW, int dstH, enum PixelFormat dstFormat, int flags, |
| 1139 |
SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param) |
| 1140 |
{
|
| 1141 |
SwsContext *c; |
| 1142 |
|
| 1143 |
if(!(c=sws_alloc_context()))
|
| 1144 |
return NULL; |
| 1145 |
|
| 1146 |
c->flags= flags; |
| 1147 |
c->srcW= srcW; |
| 1148 |
c->srcH= srcH; |
| 1149 |
c->dstW= dstW; |
| 1150 |
c->dstH= dstH; |
| 1151 |
c->srcRange = handle_jpeg(&srcFormat); |
| 1152 |
c->dstRange = handle_jpeg(&dstFormat); |
| 1153 |
c->srcFormat= srcFormat; |
| 1154 |
c->dstFormat= dstFormat; |
| 1155 |
|
| 1156 |
if (param) {
|
| 1157 |
c->param[0] = param[0]; |
| 1158 |
c->param[1] = param[1]; |
| 1159 |
} |
| 1160 |
sws_setColorspaceDetails(c, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], c->srcRange, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/, c->dstRange, 0, 1<<16, 1<<16); |
| 1161 |
|
| 1162 |
if(sws_init_context(c, srcFilter, dstFilter) < 0){ |
| 1163 |
sws_freeContext(c); |
| 1164 |
return NULL; |
| 1165 |
} |
| 1166 |
|
| 1167 |
return c;
|
| 1168 |
} |
| 1169 |
#endif
|
| 1170 |
|
| 1171 |
SwsFilter *sws_getDefaultFilter(float lumaGBlur, float chromaGBlur, |
| 1172 |
float lumaSharpen, float chromaSharpen, |
| 1173 |
float chromaHShift, float chromaVShift, |
| 1174 |
int verbose)
|
| 1175 |
{
|
| 1176 |
SwsFilter *filter= av_malloc(sizeof(SwsFilter));
|
| 1177 |
if (!filter)
|
| 1178 |
return NULL; |
| 1179 |
|
| 1180 |
if (lumaGBlur!=0.0) { |
| 1181 |
filter->lumH= sws_getGaussianVec(lumaGBlur, 3.0); |
| 1182 |
filter->lumV= sws_getGaussianVec(lumaGBlur, 3.0); |
| 1183 |
} else {
|
| 1184 |
filter->lumH= sws_getIdentityVec(); |
| 1185 |
filter->lumV= sws_getIdentityVec(); |
| 1186 |
} |
| 1187 |
|
| 1188 |
if (chromaGBlur!=0.0) { |
| 1189 |
filter->chrH= sws_getGaussianVec(chromaGBlur, 3.0); |
| 1190 |
filter->chrV= sws_getGaussianVec(chromaGBlur, 3.0); |
| 1191 |
} else {
|
| 1192 |
filter->chrH= sws_getIdentityVec(); |
| 1193 |
filter->chrV= sws_getIdentityVec(); |
| 1194 |
} |
| 1195 |
|
| 1196 |
if (chromaSharpen!=0.0) { |
| 1197 |
SwsVector *id= sws_getIdentityVec(); |
| 1198 |
sws_scaleVec(filter->chrH, -chromaSharpen); |
| 1199 |
sws_scaleVec(filter->chrV, -chromaSharpen); |
| 1200 |
sws_addVec(filter->chrH, id); |
| 1201 |
sws_addVec(filter->chrV, id); |
| 1202 |
sws_freeVec(id); |
| 1203 |
} |
| 1204 |
|
| 1205 |
if (lumaSharpen!=0.0) { |
| 1206 |
SwsVector *id= sws_getIdentityVec(); |
| 1207 |
sws_scaleVec(filter->lumH, -lumaSharpen); |
| 1208 |
sws_scaleVec(filter->lumV, -lumaSharpen); |
| 1209 |
sws_addVec(filter->lumH, id); |
| 1210 |
sws_addVec(filter->lumV, id); |
| 1211 |
sws_freeVec(id); |
| 1212 |
} |
| 1213 |
|
| 1214 |
if (chromaHShift != 0.0) |
| 1215 |
sws_shiftVec(filter->chrH, (int)(chromaHShift+0.5)); |
| 1216 |
|
| 1217 |
if (chromaVShift != 0.0) |
| 1218 |
sws_shiftVec(filter->chrV, (int)(chromaVShift+0.5)); |
| 1219 |
|
| 1220 |
sws_normalizeVec(filter->chrH, 1.0); |
| 1221 |
sws_normalizeVec(filter->chrV, 1.0); |
| 1222 |
sws_normalizeVec(filter->lumH, 1.0); |
| 1223 |
sws_normalizeVec(filter->lumV, 1.0); |
| 1224 |
|
| 1225 |
if (verbose) sws_printVec2(filter->chrH, NULL, AV_LOG_DEBUG); |
| 1226 |
if (verbose) sws_printVec2(filter->lumH, NULL, AV_LOG_DEBUG); |
| 1227 |
|
| 1228 |
return filter;
|
| 1229 |
} |
| 1230 |
|
| 1231 |
SwsVector *sws_allocVec(int length)
|
| 1232 |
{
|
| 1233 |
SwsVector *vec = av_malloc(sizeof(SwsVector));
|
| 1234 |
if (!vec)
|
| 1235 |
return NULL; |
| 1236 |
vec->length = length; |
| 1237 |
vec->coeff = av_malloc(sizeof(double) * length); |
| 1238 |
if (!vec->coeff)
|
| 1239 |
av_freep(&vec); |
| 1240 |
return vec;
|
| 1241 |
} |
| 1242 |
|
| 1243 |
SwsVector *sws_getGaussianVec(double variance, double quality) |
| 1244 |
{
|
| 1245 |
const int length= (int)(variance*quality + 0.5) | 1; |
| 1246 |
int i;
|
| 1247 |
double middle= (length-1)*0.5; |
| 1248 |
SwsVector *vec= sws_allocVec(length); |
| 1249 |
|
| 1250 |
if (!vec)
|
| 1251 |
return NULL; |
| 1252 |
|
| 1253 |
for (i=0; i<length; i++) { |
| 1254 |
double dist= i-middle;
|
| 1255 |
vec->coeff[i]= exp(-dist*dist/(2*variance*variance)) / sqrt(2*variance*M_PI); |
| 1256 |
} |
| 1257 |
|
| 1258 |
sws_normalizeVec(vec, 1.0); |
| 1259 |
|
| 1260 |
return vec;
|
| 1261 |
} |
| 1262 |
|
| 1263 |
SwsVector *sws_getConstVec(double c, int length) |
| 1264 |
{
|
| 1265 |
int i;
|
| 1266 |
SwsVector *vec= sws_allocVec(length); |
| 1267 |
|
| 1268 |
if (!vec)
|
| 1269 |
return NULL; |
| 1270 |
|
| 1271 |
for (i=0; i<length; i++) |
| 1272 |
vec->coeff[i]= c; |
| 1273 |
|
| 1274 |
return vec;
|
| 1275 |
} |
| 1276 |
|
| 1277 |
SwsVector *sws_getIdentityVec(void)
|
| 1278 |
{
|
| 1279 |
return sws_getConstVec(1.0, 1); |
| 1280 |
} |
| 1281 |
|
| 1282 |
static double sws_dcVec(SwsVector *a) |
| 1283 |
{
|
| 1284 |
int i;
|
| 1285 |
double sum=0; |
| 1286 |
|
| 1287 |
for (i=0; i<a->length; i++) |
| 1288 |
sum+= a->coeff[i]; |
| 1289 |
|
| 1290 |
return sum;
|
| 1291 |
} |
| 1292 |
|
| 1293 |
void sws_scaleVec(SwsVector *a, double scalar) |
| 1294 |
{
|
| 1295 |
int i;
|
| 1296 |
|
| 1297 |
for (i=0; i<a->length; i++) |
| 1298 |
a->coeff[i]*= scalar; |
| 1299 |
} |
| 1300 |
|
| 1301 |
void sws_normalizeVec(SwsVector *a, double height) |
| 1302 |
{
|
| 1303 |
sws_scaleVec(a, height/sws_dcVec(a)); |
| 1304 |
} |
| 1305 |
|
| 1306 |
static SwsVector *sws_getConvVec(SwsVector *a, SwsVector *b)
|
| 1307 |
{
|
| 1308 |
int length= a->length + b->length - 1; |
| 1309 |
int i, j;
|
| 1310 |
SwsVector *vec= sws_getConstVec(0.0, length); |
| 1311 |
|
| 1312 |
if (!vec)
|
| 1313 |
return NULL; |
| 1314 |
|
| 1315 |
for (i=0; i<a->length; i++) { |
| 1316 |
for (j=0; j<b->length; j++) { |
| 1317 |
vec->coeff[i+j]+= a->coeff[i]*b->coeff[j]; |
| 1318 |
} |
| 1319 |
} |
| 1320 |
|
| 1321 |
return vec;
|
| 1322 |
} |
| 1323 |
|
| 1324 |
static SwsVector *sws_sumVec(SwsVector *a, SwsVector *b)
|
| 1325 |
{
|
| 1326 |
int length= FFMAX(a->length, b->length);
|
| 1327 |
int i;
|
| 1328 |
SwsVector *vec= sws_getConstVec(0.0, length); |
| 1329 |
|
| 1330 |
if (!vec)
|
| 1331 |
return NULL; |
| 1332 |
|
| 1333 |
for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i]; |
| 1334 |
for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]+= b->coeff[i]; |
| 1335 |
|
| 1336 |
return vec;
|
| 1337 |
} |
| 1338 |
|
| 1339 |
static SwsVector *sws_diffVec(SwsVector *a, SwsVector *b)
|
| 1340 |
{
|
| 1341 |
int length= FFMAX(a->length, b->length);
|
| 1342 |
int i;
|
| 1343 |
SwsVector *vec= sws_getConstVec(0.0, length); |
| 1344 |
|
| 1345 |
if (!vec)
|
| 1346 |
return NULL; |
| 1347 |
|
| 1348 |
for (i=0; i<a->length; i++) vec->coeff[i + (length-1)/2 - (a->length-1)/2]+= a->coeff[i]; |
| 1349 |
for (i=0; i<b->length; i++) vec->coeff[i + (length-1)/2 - (b->length-1)/2]-= b->coeff[i]; |
| 1350 |
|
| 1351 |
return vec;
|
| 1352 |
} |
| 1353 |
|
| 1354 |
/* shift left / or right if "shift" is negative */
|
| 1355 |
static SwsVector *sws_getShiftedVec(SwsVector *a, int shift) |
| 1356 |
{
|
| 1357 |
int length= a->length + FFABS(shift)*2; |
| 1358 |
int i;
|
| 1359 |
SwsVector *vec= sws_getConstVec(0.0, length); |
| 1360 |
|
| 1361 |
if (!vec)
|
| 1362 |
return NULL; |
| 1363 |
|
| 1364 |
for (i=0; i<a->length; i++) { |
| 1365 |
vec->coeff[i + (length-1)/2 - (a->length-1)/2 - shift]= a->coeff[i]; |
| 1366 |
} |
| 1367 |
|
| 1368 |
return vec;
|
| 1369 |
} |
| 1370 |
|
| 1371 |
void sws_shiftVec(SwsVector *a, int shift) |
| 1372 |
{
|
| 1373 |
SwsVector *shifted= sws_getShiftedVec(a, shift); |
| 1374 |
av_free(a->coeff); |
| 1375 |
a->coeff= shifted->coeff; |
| 1376 |
a->length= shifted->length; |
| 1377 |
av_free(shifted); |
| 1378 |
} |
| 1379 |
|
| 1380 |
void sws_addVec(SwsVector *a, SwsVector *b)
|
| 1381 |
{
|
| 1382 |
SwsVector *sum= sws_sumVec(a, b); |
| 1383 |
av_free(a->coeff); |
| 1384 |
a->coeff= sum->coeff; |
| 1385 |
a->length= sum->length; |
| 1386 |
av_free(sum); |
| 1387 |
} |
| 1388 |
|
| 1389 |
void sws_subVec(SwsVector *a, SwsVector *b)
|
| 1390 |
{
|
| 1391 |
SwsVector *diff= sws_diffVec(a, b); |
| 1392 |
av_free(a->coeff); |
| 1393 |
a->coeff= diff->coeff; |
| 1394 |
a->length= diff->length; |
| 1395 |
av_free(diff); |
| 1396 |
} |
| 1397 |
|
| 1398 |
void sws_convVec(SwsVector *a, SwsVector *b)
|
| 1399 |
{
|
| 1400 |
SwsVector *conv= sws_getConvVec(a, b); |
| 1401 |
av_free(a->coeff); |
| 1402 |
a->coeff= conv->coeff; |
| 1403 |
a->length= conv->length; |
| 1404 |
av_free(conv); |
| 1405 |
} |
| 1406 |
|
| 1407 |
SwsVector *sws_cloneVec(SwsVector *a) |
| 1408 |
{
|
| 1409 |
int i;
|
| 1410 |
SwsVector *vec= sws_allocVec(a->length); |
| 1411 |
|
| 1412 |
if (!vec)
|
| 1413 |
return NULL; |
| 1414 |
|
| 1415 |
for (i=0; i<a->length; i++) vec->coeff[i]= a->coeff[i]; |
| 1416 |
|
| 1417 |
return vec;
|
| 1418 |
} |
| 1419 |
|
| 1420 |
void sws_printVec2(SwsVector *a, AVClass *log_ctx, int log_level) |
| 1421 |
{
|
| 1422 |
int i;
|
| 1423 |
double max=0; |
| 1424 |
double min=0; |
| 1425 |
double range;
|
| 1426 |
|
| 1427 |
for (i=0; i<a->length; i++) |
| 1428 |
if (a->coeff[i]>max) max= a->coeff[i];
|
| 1429 |
|
| 1430 |
for (i=0; i<a->length; i++) |
| 1431 |
if (a->coeff[i]<min) min= a->coeff[i];
|
| 1432 |
|
| 1433 |
range= max - min; |
| 1434 |
|
| 1435 |
for (i=0; i<a->length; i++) { |
| 1436 |
int x= (int)((a->coeff[i]-min)*60.0/range +0.5); |
| 1437 |
av_log(log_ctx, log_level, "%1.3f ", a->coeff[i]);
|
| 1438 |
for (;x>0; x--) av_log(log_ctx, log_level, " "); |
| 1439 |
av_log(log_ctx, log_level, "|\n");
|
| 1440 |
} |
| 1441 |
} |
| 1442 |
|
| 1443 |
#if LIBSWSCALE_VERSION_MAJOR < 1 |
| 1444 |
void sws_printVec(SwsVector *a)
|
| 1445 |
{
|
| 1446 |
sws_printVec2(a, NULL, AV_LOG_DEBUG);
|
| 1447 |
} |
| 1448 |
#endif
|
| 1449 |
|
| 1450 |
void sws_freeVec(SwsVector *a)
|
| 1451 |
{
|
| 1452 |
if (!a) return; |
| 1453 |
av_freep(&a->coeff); |
| 1454 |
a->length=0;
|
| 1455 |
av_free(a); |
| 1456 |
} |
| 1457 |
|
| 1458 |
void sws_freeFilter(SwsFilter *filter)
|
| 1459 |
{
|
| 1460 |
if (!filter) return; |
| 1461 |
|
| 1462 |
if (filter->lumH) sws_freeVec(filter->lumH);
|
| 1463 |
if (filter->lumV) sws_freeVec(filter->lumV);
|
| 1464 |
if (filter->chrH) sws_freeVec(filter->chrH);
|
| 1465 |
if (filter->chrV) sws_freeVec(filter->chrV);
|
| 1466 |
av_free(filter); |
| 1467 |
} |
| 1468 |
|
| 1469 |
void sws_freeContext(SwsContext *c)
|
| 1470 |
{
|
| 1471 |
int i;
|
| 1472 |
if (!c) return; |
| 1473 |
|
| 1474 |
if (c->lumPixBuf) {
|
| 1475 |
for (i=0; i<c->vLumBufSize; i++) |
| 1476 |
av_freep(&c->lumPixBuf[i]); |
| 1477 |
av_freep(&c->lumPixBuf); |
| 1478 |
} |
| 1479 |
|
| 1480 |
if (c->chrPixBuf) {
|
| 1481 |
for (i=0; i<c->vChrBufSize; i++) |
| 1482 |
av_freep(&c->chrPixBuf[i]); |
| 1483 |
av_freep(&c->chrPixBuf); |
| 1484 |
} |
| 1485 |
|
| 1486 |
if (CONFIG_SWSCALE_ALPHA && c->alpPixBuf) {
|
| 1487 |
for (i=0; i<c->vLumBufSize; i++) |
| 1488 |
av_freep(&c->alpPixBuf[i]); |
| 1489 |
av_freep(&c->alpPixBuf); |
| 1490 |
} |
| 1491 |
|
| 1492 |
av_freep(&c->vLumFilter); |
| 1493 |
av_freep(&c->vChrFilter); |
| 1494 |
av_freep(&c->hLumFilter); |
| 1495 |
av_freep(&c->hChrFilter); |
| 1496 |
#if HAVE_ALTIVEC
|
| 1497 |
av_freep(&c->vYCoeffsBank); |
| 1498 |
av_freep(&c->vCCoeffsBank); |
| 1499 |
#endif
|
| 1500 |
|
| 1501 |
av_freep(&c->vLumFilterPos); |
| 1502 |
av_freep(&c->vChrFilterPos); |
| 1503 |
av_freep(&c->hLumFilterPos); |
| 1504 |
av_freep(&c->hChrFilterPos); |
| 1505 |
|
| 1506 |
#if ARCH_X86
|
| 1507 |
#ifdef MAP_ANONYMOUS
|
| 1508 |
if (c->lumMmx2FilterCode) munmap(c->lumMmx2FilterCode, c->lumMmx2FilterCodeSize);
|
| 1509 |
if (c->chrMmx2FilterCode) munmap(c->chrMmx2FilterCode, c->chrMmx2FilterCodeSize);
|
| 1510 |
#elif HAVE_VIRTUALALLOC
|
| 1511 |
if (c->lumMmx2FilterCode) VirtualFree(c->lumMmx2FilterCode, 0, MEM_RELEASE); |
| 1512 |
if (c->chrMmx2FilterCode) VirtualFree(c->chrMmx2FilterCode, 0, MEM_RELEASE); |
| 1513 |
#else
|
| 1514 |
av_free(c->lumMmx2FilterCode); |
| 1515 |
av_free(c->chrMmx2FilterCode); |
| 1516 |
#endif
|
| 1517 |
c->lumMmx2FilterCode=NULL;
|
| 1518 |
c->chrMmx2FilterCode=NULL;
|
| 1519 |
#endif /* ARCH_X86 */ |
| 1520 |
|
| 1521 |
av_freep(&c->yuvTable); |
| 1522 |
|
| 1523 |
av_free(c); |
| 1524 |
} |
| 1525 |
|
| 1526 |
struct SwsContext *sws_getCachedContext(struct SwsContext *context, |
| 1527 |
int srcW, int srcH, enum PixelFormat srcFormat, |
| 1528 |
int dstW, int dstH, enum PixelFormat dstFormat, int flags, |
| 1529 |
SwsFilter *srcFilter, SwsFilter *dstFilter, const double *param) |
| 1530 |
{
|
| 1531 |
static const double default_param[2] = {SWS_PARAM_DEFAULT, SWS_PARAM_DEFAULT}; |
| 1532 |
|
| 1533 |
if (!param)
|
| 1534 |
param = default_param; |
| 1535 |
|
| 1536 |
flags = update_flags_cpu(flags); |
| 1537 |
|
| 1538 |
if (context &&
|
| 1539 |
(context->srcW != srcW || |
| 1540 |
context->srcH != srcH || |
| 1541 |
context->srcFormat != srcFormat || |
| 1542 |
context->dstW != dstW || |
| 1543 |
context->dstH != dstH || |
| 1544 |
context->dstFormat != dstFormat || |
| 1545 |
context->flags != flags || |
| 1546 |
context->param[0] != param[0] || |
| 1547 |
context->param[1] != param[1])) { |
| 1548 |
sws_freeContext(context); |
| 1549 |
context = NULL;
|
| 1550 |
} |
| 1551 |
|
| 1552 |
if (!context) {
|
| 1553 |
if (!(context = sws_alloc_context()))
|
| 1554 |
return NULL; |
| 1555 |
context->srcW = srcW; |
| 1556 |
context->srcH = srcH; |
| 1557 |
context->srcRange = handle_jpeg(&srcFormat); |
| 1558 |
context->srcFormat = srcFormat; |
| 1559 |
context->dstW = dstW; |
| 1560 |
context->dstH = dstH; |
| 1561 |
context->dstRange = handle_jpeg(&dstFormat); |
| 1562 |
context->dstFormat = dstFormat; |
| 1563 |
context->flags = flags; |
| 1564 |
context->param[0] = param[0]; |
| 1565 |
context->param[1] = param[1]; |
| 1566 |
sws_setColorspaceDetails(context, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT], context->srcRange, ff_yuv2rgb_coeffs[SWS_CS_DEFAULT] /* FIXME*/, context->dstRange, 0, 1<<16, 1<<16); |
| 1567 |
if (sws_init_context(context, srcFilter, dstFilter) < 0) { |
| 1568 |
sws_freeContext(context); |
| 1569 |
return NULL; |
| 1570 |
} |
| 1571 |
} |
| 1572 |
return context;
|
| 1573 |
} |
| 1574 |
|