swscale: Consistently order input YUV pixel formats
[libav.git] / libswscale / swscale.c
CommitLineData
fe8054c0 1/*
d026b45e
DB
2 * Copyright (C) 2001-2003 Michael Niedermayer <michaelni@gmx.at>
3 *
2912e87a 4 * This file is part of Libav.
d026b45e 5 *
2912e87a 6 * Libav is free software; you can redistribute it and/or
819ee683
DB
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.
d026b45e 10 *
2912e87a 11 * Libav is distributed in the hope that it will be useful,
d026b45e 12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
819ee683
DB
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
d026b45e 15 *
819ee683 16 * You should have received a copy of the GNU Lesser General Public
2912e87a 17 * License along with Libav; if not, write to the Free Software
b19bcbaa 18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
d026b45e 19 */
783e9cc9 20
ef0ee7f6 21#include <assert.h>
d3f41512 22#include <inttypes.h>
077ea8a7 23#include <math.h>
c1b0bfb4 24#include <stdio.h>
ef0ee7f6
DB
25#include <string.h>
26
7248797c 27#include "libavutil/avutil.h"
83da2c6f 28#include "libavutil/bswap.h"
ef0ee7f6
DB
29#include "libavutil/cpu.h"
30#include "libavutil/intreadwrite.h"
31#include "libavutil/mathematics.h"
a9af75ae 32#include "libavutil/pixdesc.h"
ef0ee7f6
DB
33#include "config.h"
34#include "rgb2rgb.h"
35#include "swscale_internal.h"
36#include "swscale.h"
0d9f3d85 37
aa2ba8c9 38DECLARE_ALIGNED(8, const uint8_t, ff_dither_8x8_128)[8][8] = {
ef0ee7f6
DB
39 { 36, 68, 60, 92, 34, 66, 58, 90, },
40 { 100, 4, 124, 28, 98, 2, 122, 26, },
41 { 52, 84, 44, 76, 50, 82, 42, 74, },
42 { 116, 20, 108, 12, 114, 18, 106, 10, },
43 { 32, 64, 56, 88, 38, 70, 62, 94, },
44 { 96, 0, 120, 24, 102, 6, 126, 30, },
45 { 48, 80, 40, 72, 54, 86, 46, 78, },
46 { 112, 16, 104, 8, 118, 22, 110, 14, },
47};
48
d2585315 49DECLARE_ALIGNED(8, static const uint8_t, sws_pb_64)[8] = {
ef0ee7f6 50 64, 64, 64, 64, 64, 64, 64, 64
4e3e333a 51};
5cebb24b 52
ef0ee7f6
DB
53static av_always_inline void fillPlane(uint8_t *plane, int stride, int width,
54 int height, int y, uint8_t val)
dd68318c 55{
d4da3e47 56 int i;
ef0ee7f6
DB
57 uint8_t *ptr = plane + stride * y;
58 for (i = 0; i < height; i++) {
d4da3e47
CS
59 memset(ptr, val, width);
60 ptr += stride;
61 }
62}
63
26b5ad25
LB
64static void fill_plane9or10(uint8_t *plane, int stride, int width,
65 int height, int y, uint8_t val,
66 const int dst_depth, const int big_endian)
67{
68 int i, j;
69 uint16_t *dst = (uint16_t *) (plane + stride * y);
70#define FILL8TO9_OR_10(wfunc) \
71 for (i = 0; i < height; i++) { \
72 for (j = 0; j < width; j++) { \
73 wfunc(&dst[j], (val << (dst_depth - 8)) | \
74 (val >> (16 - dst_depth))); \
75 } \
76 dst += stride / 2; \
77 }
78 if (big_endian) {
79 FILL8TO9_OR_10(AV_WB16);
80 } else {
81 FILL8TO9_OR_10(AV_WL16);
82 }
83}
84
85
ef0ee7f6
DB
86static void hScale16To19_c(SwsContext *c, int16_t *_dst, int dstW,
87 const uint8_t *_src, const int16_t *filter,
2254b559 88 const int32_t *filterPos, int filterSize)
ef1ee362 89{
0a7068fa 90 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->srcFormat);
ef1ee362 91 int i;
ef0ee7f6 92 int32_t *dst = (int32_t *) _dst;
ef1ee362 93 const uint16_t *src = (const uint16_t *) _src;
2268db2c 94 int bits = desc->comp[0].depth - 1;
ef0ee7f6 95 int sh = bits - 4;
ef1ee362
RB
96
97 for (i = 0; i < dstW; i++) {
98 int j;
99 int srcPos = filterPos[i];
ef0ee7f6 100 int val = 0;
ef1ee362
RB
101
102 for (j = 0; j < filterSize; j++) {
103 val += src[srcPos + j] * filter[filterSize * i + j];
104 }
105 // filter=14 bit, input=16 bit, output=30 bit, >> 11 makes 19 bit
8a8d0ce2 106 dst[i] = FFMIN(val >> sh, (1 << 19) - 1);
ef1ee362
RB
107 }
108}
109
ef0ee7f6
DB
110static void hScale16To15_c(SwsContext *c, int16_t *dst, int dstW,
111 const uint8_t *_src, const int16_t *filter,
2254b559 112 const int32_t *filterPos, int filterSize)
28c1115a 113{
0a7068fa 114 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->srcFormat);
28c1115a
RB
115 int i;
116 const uint16_t *src = (const uint16_t *) _src;
2268db2c 117 int sh = desc->comp[0].depth - 1;
28c1115a
RB
118
119 for (i = 0; i < dstW; i++) {
120 int j;
121 int srcPos = filterPos[i];
ef0ee7f6 122 int val = 0;
28c1115a
RB
123
124 for (j = 0; j < filterSize; j++) {
125 val += src[srcPos + j] * filter[filterSize * i + j];
126 }
127 // filter=14 bit, input=16 bit, output=30 bit, >> 15 makes 15 bit
128 dst[i] = FFMIN(val >> sh, (1 << 15) - 1);
129 }
130}
131
1674bd2a 132// bilinear / bicubic scaling
ef0ee7f6
DB
133static void hScale8To15_c(SwsContext *c, int16_t *dst, int dstW,
134 const uint8_t *src, const int16_t *filter,
135 const int32_t *filterPos, int filterSize)
1674bd2a
RB
136{
137 int i;
ef0ee7f6 138 for (i = 0; i < dstW; i++) {
1674bd2a 139 int j;
ef0ee7f6
DB
140 int srcPos = filterPos[i];
141 int val = 0;
142 for (j = 0; j < filterSize; j++) {
143 val += ((int)src[srcPos + j]) * filter[filterSize * i + j];
1674bd2a 144 }
ef0ee7f6 145 dst[i] = FFMIN(val >> 7, (1 << 15) - 1); // the cubic equation does overflow ...
1674bd2a
RB
146 }
147}
148
ef0ee7f6
DB
149static void hScale8To19_c(SwsContext *c, int16_t *_dst, int dstW,
150 const uint8_t *src, const int16_t *filter,
151 const int32_t *filterPos, int filterSize)
28c1115a
RB
152{
153 int i;
154 int32_t *dst = (int32_t *) _dst;
ef0ee7f6 155 for (i = 0; i < dstW; i++) {
28c1115a 156 int j;
ef0ee7f6
DB
157 int srcPos = filterPos[i];
158 int val = 0;
159 for (j = 0; j < filterSize; j++) {
160 val += ((int)src[srcPos + j]) * filter[filterSize * i + j];
28c1115a 161 }
ef0ee7f6 162 dst[i] = FFMIN(val >> 3, (1 << 19) - 1); // the cubic equation does overflow ...
28c1115a
RB
163 }
164}
165
511cf612 166// FIXME all pal and rgb srcFormats could do this conversion as well
ef0ee7f6 167// FIXME all scalers more complex than bilinear could do half of this transform
0af8a71d 168static void chrRangeToJpeg_c(int16_t *dstU, int16_t *dstV, int width)
1674bd2a
RB
169{
170 int i;
171 for (i = 0; i < width; i++) {
ef0ee7f6
DB
172 dstU[i] = (FFMIN(dstU[i], 30775) * 4663 - 9289992) >> 12; // -264
173 dstV[i] = (FFMIN(dstV[i], 30775) * 4663 - 9289992) >> 12; // -264
1674bd2a
RB
174 }
175}
ef0ee7f6 176
0af8a71d 177static void chrRangeFromJpeg_c(int16_t *dstU, int16_t *dstV, int width)
1674bd2a
RB
178{
179 int i;
180 for (i = 0; i < width; i++) {
ef0ee7f6
DB
181 dstU[i] = (dstU[i] * 1799 + 4081085) >> 11; // 1469
182 dstV[i] = (dstV[i] * 1799 + 4081085) >> 11; // 1469
1674bd2a
RB
183 }
184}
ef0ee7f6 185
0af8a71d 186static void lumRangeToJpeg_c(int16_t *dst, int width)
1674bd2a
RB
187{
188 int i;
189 for (i = 0; i < width; i++)
ef0ee7f6 190 dst[i] = (FFMIN(dst[i], 30189) * 19077 - 39057361) >> 14;
1674bd2a 191}
ef0ee7f6 192
0af8a71d 193static void lumRangeFromJpeg_c(int16_t *dst, int width)
1674bd2a
RB
194{
195 int i;
196 for (i = 0; i < width; i++)
ef0ee7f6 197 dst[i] = (dst[i] * 14071 + 33561947) >> 14;
1674bd2a
RB
198}
199
ef1ee362
RB
200static void chrRangeToJpeg16_c(int16_t *_dstU, int16_t *_dstV, int width)
201{
202 int i;
203 int32_t *dstU = (int32_t *) _dstU;
204 int32_t *dstV = (int32_t *) _dstV;
205 for (i = 0; i < width; i++) {
ef0ee7f6
DB
206 dstU[i] = (FFMIN(dstU[i], 30775 << 4) * 4663 - (9289992 << 4)) >> 12; // -264
207 dstV[i] = (FFMIN(dstV[i], 30775 << 4) * 4663 - (9289992 << 4)) >> 12; // -264
ef1ee362
RB
208 }
209}
ef0ee7f6 210
ef1ee362
RB
211static void chrRangeFromJpeg16_c(int16_t *_dstU, int16_t *_dstV, int width)
212{
213 int i;
214 int32_t *dstU = (int32_t *) _dstU;
215 int32_t *dstV = (int32_t *) _dstV;
216 for (i = 0; i < width; i++) {
ef0ee7f6
DB
217 dstU[i] = (dstU[i] * 1799 + (4081085 << 4)) >> 11; // 1469
218 dstV[i] = (dstV[i] * 1799 + (4081085 << 4)) >> 11; // 1469
ef1ee362
RB
219 }
220}
ef0ee7f6 221
ef1ee362
RB
222static void lumRangeToJpeg16_c(int16_t *_dst, int width)
223{
224 int i;
225 int32_t *dst = (int32_t *) _dst;
226 for (i = 0; i < width; i++)
ef0ee7f6 227 dst[i] = (FFMIN(dst[i], 30189 << 4) * 4769 - (39057361 << 2)) >> 12;
ef1ee362 228}
ef0ee7f6 229
ef1ee362
RB
230static void lumRangeFromJpeg16_c(int16_t *_dst, int width)
231{
232 int i;
233 int32_t *dst = (int32_t *) _dst;
234 for (i = 0; i < width; i++)
ef0ee7f6 235 dst[i] = (dst[i] * 14071 + (33561947 << 4)) >> 14;
ef1ee362
RB
236}
237
2762ee30
RB
238static void hyscale_fast_c(SwsContext *c, int16_t *dst, int dstWidth,
239 const uint8_t *src, int srcW, int xInc)
1674bd2a
RB
240{
241 int i;
ef0ee7f6
DB
242 unsigned int xpos = 0;
243 for (i = 0; i < dstWidth; i++) {
244 register unsigned int xx = xpos >> 16;
245 register unsigned int xalpha = (xpos & 0xFFFF) >> 9;
246 dst[i] = (src[xx] << 7) + (src[xx + 1] - src[xx]) * xalpha;
247 xpos += xInc;
1674bd2a
RB
248 }
249}
250
2762ee30 251// *** horizontal scale Y line to temp buffer
ef1ee362 252static av_always_inline void hyscale(SwsContext *c, int16_t *dst, int dstWidth,
ef0ee7f6
DB
253 const uint8_t *src_in[4],
254 int srcW, int xInc,
6e5a8d3c 255 const int16_t *hLumFilter,
ef0ee7f6
DB
256 const int32_t *hLumFilterPos,
257 int hLumFilterSize,
6e5a8d3c
RB
258 uint8_t *formatConvBuffer,
259 uint32_t *pal, int isAlpha)
1674bd2a 260{
ef0ee7f6
DB
261 void (*toYV12)(uint8_t *, const uint8_t *, int, uint32_t *) =
262 isAlpha ? c->alpToYV12 : c->lumToYV12;
0af8a71d 263 void (*convertRange)(int16_t *, int) = isAlpha ? NULL : c->lumConvertRange;
185655c6 264 const uint8_t *src = src_in[isAlpha ? 3 : 0];
1674bd2a 265
1674bd2a
RB
266 if (toYV12) {
267 toYV12(formatConvBuffer, src, srcW, pal);
ef0ee7f6 268 src = formatConvBuffer;
185655c6
RB
269 } else if (c->readLumPlanar && !isAlpha) {
270 c->readLumPlanar(formatConvBuffer, src_in, srcW);
271 src = formatConvBuffer;
92806167
PM
272 } else if (c->readAlpPlanar && isAlpha) {
273 c->readAlpPlanar(formatConvBuffer, src_in, srcW);
274 src = formatConvBuffer;
1674bd2a
RB
275 }
276
277 if (!c->hyscale_fast) {
ef0ee7f6
DB
278 c->hyScale(c, dst, dstWidth, src, hLumFilter,
279 hLumFilterPos, hLumFilterSize);
1674bd2a
RB
280 } else { // fast bilinear upscale / crap downscale
281 c->hyscale_fast(c, dst, dstWidth, src, srcW, xInc);
282 }
283
284 if (convertRange)
285 convertRange(dst, dstWidth);
286}
287
2762ee30
RB
288static void hcscale_fast_c(SwsContext *c, int16_t *dst1, int16_t *dst2,
289 int dstWidth, const uint8_t *src1,
290 const uint8_t *src2, int srcW, int xInc)
1674bd2a
RB
291{
292 int i;
ef0ee7f6
DB
293 unsigned int xpos = 0;
294 for (i = 0; i < dstWidth; i++) {
295 register unsigned int xx = xpos >> 16;
296 register unsigned int xalpha = (xpos & 0xFFFF) >> 9;
297 dst1[i] = (src1[xx] * (xalpha ^ 127) + src1[xx + 1] * xalpha);
298 dst2[i] = (src2[xx] * (xalpha ^ 127) + src2[xx + 1] * xalpha);
299 xpos += xInc;
1674bd2a
RB
300 }
301}
302
ef0ee7f6
DB
303static av_always_inline void hcscale(SwsContext *c, int16_t *dst1,
304 int16_t *dst2, int dstWidth,
185655c6 305 const uint8_t *src_in[4],
ef0ee7f6
DB
306 int srcW, int xInc,
307 const int16_t *hChrFilter,
308 const int32_t *hChrFilterPos,
309 int hChrFilterSize,
6e5a8d3c 310 uint8_t *formatConvBuffer, uint32_t *pal)
1674bd2a 311{
185655c6 312 const uint8_t *src1 = src_in[1], *src2 = src_in[2];
1674bd2a 313 if (c->chrToYV12) {
ef0ee7f6
DB
314 uint8_t *buf2 = formatConvBuffer +
315 FFALIGN(srcW * FFALIGN(c->srcBpc, 8) >> 3, 16);
1674bd2a 316 c->chrToYV12(formatConvBuffer, buf2, src1, src2, srcW, pal);
ef0ee7f6
DB
317 src1 = formatConvBuffer;
318 src2 = buf2;
185655c6 319 } else if (c->readChrPlanar) {
ef0ee7f6
DB
320 uint8_t *buf2 = formatConvBuffer +
321 FFALIGN(srcW * FFALIGN(c->srcBpc, 8) >> 3, 16);
185655c6 322 c->readChrPlanar(formatConvBuffer, buf2, src_in, srcW);
ef0ee7f6
DB
323 src1 = formatConvBuffer;
324 src2 = buf2;
1674bd2a
RB
325 }
326
327 if (!c->hcscale_fast) {
3f04ab4f
RB
328 c->hcScale(c, dst1, dstWidth, src1, hChrFilter, hChrFilterPos, hChrFilterSize);
329 c->hcScale(c, dst2, dstWidth, src2, hChrFilter, hChrFilterPos, hChrFilterSize);
1674bd2a
RB
330 } else { // fast bilinear upscale / crap downscale
331 c->hcscale_fast(c, dst1, dst2, dstWidth, src1, src2, srcW, xInc);
332 }
333
334 if (c->chrConvertRange)
335 c->chrConvertRange(dst1, dst2, dstWidth);
336}
337
338#define DEBUG_SWSCALE_BUFFERS 0
ef0ee7f6
DB
339#define DEBUG_BUFFERS(...) \
340 if (DEBUG_SWSCALE_BUFFERS) \
341 av_log(c, AV_LOG_DEBUG, __VA_ARGS__)
1674bd2a 342
1909f6b1 343static int swscale(SwsContext *c, const uint8_t *src[],
2762ee30 344 int srcStride[], int srcSliceY,
ef0ee7f6 345 int srcSliceH, uint8_t *dst[], int dstStride[])
1674bd2a 346{
ef0ee7f6
DB
347 /* load a few things into local vars to make the code more readable?
348 * and faster */
349 const int srcW = c->srcW;
350 const int dstW = c->dstW;
351 const int dstH = c->dstH;
352 const int chrDstW = c->chrDstW;
353 const int chrSrcW = c->chrSrcW;
354 const int lumXInc = c->lumXInc;
355 const int chrXInc = c->chrXInc;
716d413c 356 const enum AVPixelFormat dstFormat = c->dstFormat;
ef0ee7f6
DB
357 const int flags = c->flags;
358 int32_t *vLumFilterPos = c->vLumFilterPos;
359 int32_t *vChrFilterPos = c->vChrFilterPos;
360 int32_t *hLumFilterPos = c->hLumFilterPos;
361 int32_t *hChrFilterPos = c->hChrFilterPos;
362 int16_t *vLumFilter = c->vLumFilter;
363 int16_t *vChrFilter = c->vChrFilter;
364 int16_t *hLumFilter = c->hLumFilter;
365 int16_t *hChrFilter = c->hChrFilter;
366 int32_t *lumMmxFilter = c->lumMmxFilter;
367 int32_t *chrMmxFilter = c->chrMmxFilter;
368 const int vLumFilterSize = c->vLumFilterSize;
369 const int vChrFilterSize = c->vChrFilterSize;
370 const int hLumFilterSize = c->hLumFilterSize;
371 const int hChrFilterSize = c->hChrFilterSize;
372 int16_t **lumPixBuf = c->lumPixBuf;
373 int16_t **chrUPixBuf = c->chrUPixBuf;
374 int16_t **chrVPixBuf = c->chrVPixBuf;
375 int16_t **alpPixBuf = c->alpPixBuf;
376 const int vLumBufSize = c->vLumBufSize;
377 const int vChrBufSize = c->vChrBufSize;
378 uint8_t *formatConvBuffer = c->formatConvBuffer;
379 uint32_t *pal = c->pal_yuv;
380 yuv2planar1_fn yuv2plane1 = c->yuv2plane1;
381 yuv2planarX_fn yuv2planeX = c->yuv2planeX;
382 yuv2interleavedX_fn yuv2nv12cX = c->yuv2nv12cX;
383 yuv2packed1_fn yuv2packed1 = c->yuv2packed1;
384 yuv2packed2_fn yuv2packed2 = c->yuv2packed2;
385 yuv2packedX_fn yuv2packedX = c->yuv2packedX;
4eb93bed 386 yuv2anyX_fn yuv2anyX = c->yuv2anyX;
e8bc6422
CB
387 const int chrSrcSliceY = srcSliceY >> c->chrSrcVSubSample;
388 const int chrSrcSliceH = AV_CEIL_RSHIFT(srcSliceH, c->chrSrcVSubSample);
ef0ee7f6
DB
389 int should_dither = is9_OR_10BPS(c->srcFormat) ||
390 is16BPS(c->srcFormat);
1674bd2a 391 int lastDstY;
1674bd2a
RB
392
393 /* vars which will change and which we need to store back in the context */
ef0ee7f6
DB
394 int dstY = c->dstY;
395 int lumBufIndex = c->lumBufIndex;
396 int chrBufIndex = c->chrBufIndex;
397 int lastInLumBuf = c->lastInLumBuf;
398 int lastInChrBuf = c->lastInChrBuf;
1674bd2a
RB
399
400 if (isPacked(c->srcFormat)) {
ef0ee7f6
DB
401 src[0] =
402 src[1] =
403 src[2] =
404 src[3] = src[0];
405 srcStride[0] =
406 srcStride[1] =
407 srcStride[2] =
408 srcStride[3] = srcStride[0];
1674bd2a 409 }
ef0ee7f6
DB
410 srcStride[1] <<= c->vChrDrop;
411 srcStride[2] <<= c->vChrDrop;
1674bd2a 412
1909f6b1 413 DEBUG_BUFFERS("swscale() %p[%d] %p[%d] %p[%d] %p[%d] -> %p[%d] %p[%d] %p[%d] %p[%d]\n",
ef0ee7f6
DB
414 src[0], srcStride[0], src[1], srcStride[1],
415 src[2], srcStride[2], src[3], srcStride[3],
416 dst[0], dstStride[0], dst[1], dstStride[1],
417 dst[2], dstStride[2], dst[3], dstStride[3]);
1674bd2a 418 DEBUG_BUFFERS("srcSliceY: %d srcSliceH: %d dstY: %d dstH: %d\n",
ef0ee7f6 419 srcSliceY, srcSliceH, dstY, dstH);
1674bd2a 420 DEBUG_BUFFERS("vLumFilterSize: %d vLumBufSize: %d vChrFilterSize: %d vChrBufSize: %d\n",
ef0ee7f6 421 vLumFilterSize, vLumBufSize, vChrFilterSize, vChrBufSize);
1674bd2a 422
ef0ee7f6
DB
423 if (dstStride[0] % 8 != 0 || dstStride[1] % 8 != 0 ||
424 dstStride[2] % 8 != 0 || dstStride[3] % 8 != 0) {
425 static int warnedAlready = 0; // FIXME maybe move this into the context
1674bd2a 426 if (flags & SWS_PRINT_INFO && !warnedAlready) {
ef0ee7f6
DB
427 av_log(c, AV_LOG_WARNING,
428 "Warning: dstStride is not aligned!\n"
1674bd2a 429 " ->cannot do aligned memory accesses anymore\n");
ef0ee7f6 430 warnedAlready = 1;
1674bd2a
RB
431 }
432 }
433
434 /* Note the user might start scaling the picture in the middle so this
ef0ee7f6
DB
435 * will not get executed. This is not really intended but works
436 * currently, so people might do it. */
437 if (srcSliceY == 0) {
438 lumBufIndex = -1;
439 chrBufIndex = -1;
440 dstY = 0;
441 lastInLumBuf = -1;
442 lastInChrBuf = -1;
1674bd2a
RB
443 }
444
4e3e333a 445 if (!should_dither) {
d2585315 446 c->chrDither8 = c->lumDither8 = sws_pb_64;
4e3e333a 447 }
ef0ee7f6 448 lastDstY = dstY;
1674bd2a 449
ef0ee7f6
DB
450 for (; dstY < dstH; dstY++) {
451 const int chrDstY = dstY >> c->chrDstVSubSample;
452 uint8_t *dest[4] = {
13a09979
RB
453 dst[0] + dstStride[0] * dstY,
454 dst[1] + dstStride[1] * chrDstY,
455 dst[2] + dstStride[2] * chrDstY,
456 (CONFIG_SWSCALE_ALPHA && alpPixBuf) ? dst[3] + dstStride[3] * dstY : NULL,
457 };
1674bd2a 458
ef0ee7f6
DB
459 // First line needed as input
460 const int firstLumSrcY = FFMAX(1 - vLumFilterSize, vLumFilterPos[dstY]);
461 const int firstLumSrcY2 = FFMAX(1 - vLumFilterSize, vLumFilterPos[FFMIN(dstY | ((1 << c->chrDstVSubSample) - 1), dstH - 1)]);
462 // First line needed as input
463 const int firstChrSrcY = FFMAX(1 - vChrFilterSize, vChrFilterPos[chrDstY]);
d49352c7
RB
464
465 // Last line needed as input
466 int lastLumSrcY = FFMIN(c->srcH, firstLumSrcY + vLumFilterSize) - 1;
467 int lastLumSrcY2 = FFMIN(c->srcH, firstLumSrcY2 + vLumFilterSize) - 1;
468 int lastChrSrcY = FFMIN(c->chrSrcH, firstChrSrcY + vChrFilterSize) - 1;
1674bd2a
RB
469 int enough_lines;
470
ef0ee7f6
DB
471 // handle holes (FAST_BILINEAR & weird filters)
472 if (firstLumSrcY > lastInLumBuf)
473 lastInLumBuf = firstLumSrcY - 1;
474 if (firstChrSrcY > lastInChrBuf)
475 lastInChrBuf = firstChrSrcY - 1;
1674bd2a
RB
476 assert(firstLumSrcY >= lastInLumBuf - vLumBufSize + 1);
477 assert(firstChrSrcY >= lastInChrBuf - vChrBufSize + 1);
478
479 DEBUG_BUFFERS("dstY: %d\n", dstY);
480 DEBUG_BUFFERS("\tfirstLumSrcY: %d lastLumSrcY: %d lastInLumBuf: %d\n",
ef0ee7f6 481 firstLumSrcY, lastLumSrcY, lastInLumBuf);
1674bd2a 482 DEBUG_BUFFERS("\tfirstChrSrcY: %d lastChrSrcY: %d lastInChrBuf: %d\n",
ef0ee7f6 483 firstChrSrcY, lastChrSrcY, lastInChrBuf);
1674bd2a
RB
484
485 // Do we have enough lines in this slice to output the dstY line
ef0ee7f6 486 enough_lines = lastLumSrcY2 < srcSliceY + srcSliceH &&
e8bc6422 487 lastChrSrcY < AV_CEIL_RSHIFT(srcSliceY + srcSliceH, c->chrSrcVSubSample);
1674bd2a
RB
488
489 if (!enough_lines) {
490 lastLumSrcY = srcSliceY + srcSliceH - 1;
491 lastChrSrcY = chrSrcSliceY + chrSrcSliceH - 1;
492 DEBUG_BUFFERS("buffering slice: lastLumSrcY %d lastChrSrcY %d\n",
ef0ee7f6 493 lastLumSrcY, lastChrSrcY);
1674bd2a
RB
494 }
495
ef0ee7f6
DB
496 // Do horizontal scaling
497 while (lastInLumBuf < lastLumSrcY) {
185655c6
RB
498 const uint8_t *src1[4] = {
499 src[0] + (lastInLumBuf + 1 - srcSliceY) * srcStride[0],
500 src[1] + (lastInLumBuf + 1 - srcSliceY) * srcStride[1],
501 src[2] + (lastInLumBuf + 1 - srcSliceY) * srcStride[2],
502 src[3] + (lastInLumBuf + 1 - srcSliceY) * srcStride[3],
503 };
1674bd2a 504 lumBufIndex++;
ef0ee7f6 505 assert(lumBufIndex < 2 * vLumBufSize);
1674bd2a
RB
506 assert(lastInLumBuf + 1 - srcSliceY < srcSliceH);
507 assert(lastInLumBuf + 1 - srcSliceY >= 0);
ef0ee7f6 508 hyscale(c, lumPixBuf[lumBufIndex], dstW, src1, srcW, lumXInc,
2762ee30 509 hLumFilter, hLumFilterPos, hLumFilterSize,
ef0ee7f6 510 formatConvBuffer, pal, 0);
1674bd2a 511 if (CONFIG_SWSCALE_ALPHA && alpPixBuf)
ef0ee7f6 512 hyscale(c, alpPixBuf[lumBufIndex], dstW, src1, srcW,
2762ee30 513 lumXInc, hLumFilter, hLumFilterPos, hLumFilterSize,
ef0ee7f6 514 formatConvBuffer, pal, 1);
1674bd2a
RB
515 lastInLumBuf++;
516 DEBUG_BUFFERS("\t\tlumBufIndex %d: lastInLumBuf: %d\n",
ef0ee7f6 517 lumBufIndex, lastInLumBuf);
1674bd2a 518 }
ef0ee7f6 519 while (lastInChrBuf < lastChrSrcY) {
185655c6
RB
520 const uint8_t *src1[4] = {
521 src[0] + (lastInChrBuf + 1 - chrSrcSliceY) * srcStride[0],
522 src[1] + (lastInChrBuf + 1 - chrSrcSliceY) * srcStride[1],
523 src[2] + (lastInChrBuf + 1 - chrSrcSliceY) * srcStride[2],
524 src[3] + (lastInChrBuf + 1 - chrSrcSliceY) * srcStride[3],
525 };
1674bd2a 526 chrBufIndex++;
ef0ee7f6 527 assert(chrBufIndex < 2 * vChrBufSize);
1674bd2a
RB
528 assert(lastInChrBuf + 1 - chrSrcSliceY < (chrSrcSliceH));
529 assert(lastInChrBuf + 1 - chrSrcSliceY >= 0);
ef0ee7f6 530 // FIXME replace parameters through context struct (some at least)
1674bd2a
RB
531
532 if (c->needs_hcscale)
2762ee30 533 hcscale(c, chrUPixBuf[chrBufIndex], chrVPixBuf[chrBufIndex],
ef0ee7f6
DB
534 chrDstW, src1, chrSrcW, chrXInc,
535 hChrFilter, hChrFilterPos, hChrFilterSize,
536 formatConvBuffer, pal);
1674bd2a
RB
537 lastInChrBuf++;
538 DEBUG_BUFFERS("\t\tchrBufIndex %d: lastInChrBuf: %d\n",
ef0ee7f6 539 chrBufIndex, lastInChrBuf);
1674bd2a 540 }
ef0ee7f6
DB
541 // wrap buf index around to stay inside the ring buffer
542 if (lumBufIndex >= vLumBufSize)
543 lumBufIndex -= vLumBufSize;
544 if (chrBufIndex >= vChrBufSize)
545 chrBufIndex -= vChrBufSize;
1674bd2a 546 if (!enough_lines)
ef0ee7f6 547 break; // we can't output a dstY line so let's try with the next slice
1674bd2a 548
17337f54 549#if HAVE_MMX_INLINE
ef0ee7f6
DB
550 updateMMXDitherTables(c, dstY, lumBufIndex, chrBufIndex,
551 lastInLumBuf, lastInChrBuf);
1674bd2a 552#endif
4e3e333a 553 if (should_dither) {
aa2ba8c9
DB
554 c->chrDither8 = ff_dither_8x8_128[chrDstY & 7];
555 c->lumDither8 = ff_dither_8x8_128[dstY & 7];
4e3e333a 556 }
ef0ee7f6
DB
557 if (dstY >= dstH - 2) {
558 /* hmm looks like we can't use MMX here without overwriting
559 * this array's tail */
560 ff_sws_init_output_funcs(c, &yuv2plane1, &yuv2planeX, &yuv2nv12cX,
4eb93bed 561 &yuv2packed1, &yuv2packed2, &yuv2packedX, &yuv2anyX);
edeb56fa
RB
562 }
563
564 {
ef0ee7f6
DB
565 const int16_t **lumSrcPtr = (const int16_t **)lumPixBuf + lumBufIndex + firstLumSrcY - lastInLumBuf + vLumBufSize;
566 const int16_t **chrUSrcPtr = (const int16_t **)chrUPixBuf + chrBufIndex + firstChrSrcY - lastInChrBuf + vChrBufSize;
567 const int16_t **chrVSrcPtr = (const int16_t **)chrVPixBuf + chrBufIndex + firstChrSrcY - lastInChrBuf + vChrBufSize;
568 const int16_t **alpSrcPtr = (CONFIG_SWSCALE_ALPHA && alpPixBuf) ?
569 (const int16_t **)alpPixBuf + lumBufIndex + firstLumSrcY - lastInLumBuf + vLumBufSize : NULL;
d49352c7
RB
570
571 if (firstLumSrcY < 0 || firstLumSrcY + vLumFilterSize > c->srcH) {
ef0ee7f6
DB
572 const int16_t **tmpY = (const int16_t **)lumPixBuf +
573 2 * vLumBufSize;
574 int neg = -firstLumSrcY, i;
575 int end = FFMIN(c->srcH - firstLumSrcY, vLumFilterSize);
576 for (i = 0; i < neg; i++)
d49352c7 577 tmpY[i] = lumSrcPtr[neg];
ef0ee7f6 578 for (; i < end; i++)
d49352c7 579 tmpY[i] = lumSrcPtr[i];
ef0ee7f6
DB
580 for (; i < vLumFilterSize; i++)
581 tmpY[i] = tmpY[i - 1];
d49352c7
RB
582 lumSrcPtr = tmpY;
583
584 if (alpSrcPtr) {
ef0ee7f6
DB
585 const int16_t **tmpA = (const int16_t **)alpPixBuf +
586 2 * vLumBufSize;
587 for (i = 0; i < neg; i++)
d49352c7 588 tmpA[i] = alpSrcPtr[neg];
ef0ee7f6 589 for (; i < end; i++)
d49352c7 590 tmpA[i] = alpSrcPtr[i];
ef0ee7f6 591 for (; i < vLumFilterSize; i++)
d49352c7
RB
592 tmpA[i] = tmpA[i - 1];
593 alpSrcPtr = tmpA;
594 }
595 }
ef0ee7f6
DB
596 if (firstChrSrcY < 0 ||
597 firstChrSrcY + vChrFilterSize > c->chrSrcH) {
598 const int16_t **tmpU = (const int16_t **)chrUPixBuf + 2 * vChrBufSize,
599 **tmpV = (const int16_t **)chrVPixBuf + 2 * vChrBufSize;
600 int neg = -firstChrSrcY, i;
601 int end = FFMIN(c->chrSrcH - firstChrSrcY, vChrFilterSize);
602 for (i = 0; i < neg; i++) {
d49352c7
RB
603 tmpU[i] = chrUSrcPtr[neg];
604 tmpV[i] = chrVSrcPtr[neg];
605 }
ef0ee7f6 606 for (; i < end; i++) {
d49352c7
RB
607 tmpU[i] = chrUSrcPtr[i];
608 tmpV[i] = chrVSrcPtr[i];
609 }
ef0ee7f6 610 for (; i < vChrFilterSize; i++) {
d49352c7
RB
611 tmpU[i] = tmpU[i - 1];
612 tmpV[i] = tmpV[i - 1];
613 }
614 chrUSrcPtr = tmpU;
615 chrVSrcPtr = tmpV;
616 }
617
ef0ee7f6
DB
618 if (isPlanarYUV(dstFormat) ||
619 (isGray(dstFormat) && !isALPHA(dstFormat))) { // YV12 like
620 const int chrSkipMask = (1 << c->chrDstVSubSample) - 1;
34e8d147 621
109f62e8
RB
622 if (vLumFilterSize == 1) {
623 yuv2plane1(lumSrcPtr[0], dest[0], dstW, c->lumDither8, 0);
624 } else {
ef0ee7f6
DB
625 yuv2planeX(vLumFilter + dstY * vLumFilterSize,
626 vLumFilterSize, lumSrcPtr, dest[0],
627 dstW, c->lumDither8, 0);
109f62e8 628 }
34e8d147 629
ef0ee7f6 630 if (!((dstY & chrSkipMask) || isGray(dstFormat))) {
109f62e8 631 if (yuv2nv12cX) {
ef0ee7f6
DB
632 yuv2nv12cX(c, vChrFilter + chrDstY * vChrFilterSize,
633 vChrFilterSize, chrUSrcPtr, chrVSrcPtr,
634 dest[1], chrDstW);
109f62e8
RB
635 } else if (vChrFilterSize == 1) {
636 yuv2plane1(chrUSrcPtr[0], dest[1], chrDstW, c->chrDither8, 0);
637 yuv2plane1(chrVSrcPtr[0], dest[2], chrDstW, c->chrDither8, 3);
638 } else {
ef0ee7f6
DB
639 yuv2planeX(vChrFilter + chrDstY * vChrFilterSize,
640 vChrFilterSize, chrUSrcPtr, dest[1],
641 chrDstW, c->chrDither8, 0);
642 yuv2planeX(vChrFilter + chrDstY * vChrFilterSize,
643 vChrFilterSize, chrVSrcPtr, dest[2],
644 chrDstW, c->chrDither8, 3);
34e8d147 645 }
109f62e8 646 }
34e8d147 647
ef0ee7f6 648 if (CONFIG_SWSCALE_ALPHA && alpPixBuf) {
109f62e8 649 if (vLumFilterSize == 1) {
ef0ee7f6
DB
650 yuv2plane1(alpSrcPtr[0], dest[3], dstW,
651 c->lumDither8, 0);
109f62e8 652 } else {
ef0ee7f6
DB
653 yuv2planeX(vLumFilter + dstY * vLumFilterSize,
654 vLumFilterSize, alpSrcPtr, dest[3],
655 dstW, c->lumDither8, 0);
ff7913ae 656 }
1674bd2a 657 }
4eb93bed 658 } else if (yuv2packedX) {
ef0ee7f6
DB
659 if (c->yuv2packed1 && vLumFilterSize == 1 &&
660 vChrFilterSize <= 2) { // unscaled RGB
9d9c8464 661 int chrAlpha = vChrFilterSize == 1 ? 0 : vChrFilter[2 * dstY + 1];
13a09979
RB
662 yuv2packed1(c, *lumSrcPtr, chrUSrcPtr, chrVSrcPtr,
663 alpPixBuf ? *alpSrcPtr : NULL,
664 dest[0], dstW, chrAlpha, dstY);
ef0ee7f6
DB
665 } else if (c->yuv2packed2 && vLumFilterSize == 2 &&
666 vChrFilterSize == 2) { // bilinear upscale RGB
13a09979
RB
667 int lumAlpha = vLumFilter[2 * dstY + 1];
668 int chrAlpha = vChrFilter[2 * dstY + 1];
669 lumMmxFilter[2] =
ef0ee7f6 670 lumMmxFilter[3] = vLumFilter[2 * dstY] * 0x10001;
13a09979
RB
671 chrMmxFilter[2] =
672 chrMmxFilter[3] = vChrFilter[2 * chrDstY] * 0x10001;
673 yuv2packed2(c, lumSrcPtr, chrUSrcPtr, chrVSrcPtr,
674 alpPixBuf ? alpSrcPtr : NULL,
675 dest[0], dstW, lumAlpha, chrAlpha, dstY);
ef0ee7f6 676 } else { // general RGB
13a09979
RB
677 yuv2packedX(c, vLumFilter + dstY * vLumFilterSize,
678 lumSrcPtr, vLumFilterSize,
679 vChrFilter + dstY * vChrFilterSize,
680 chrUSrcPtr, chrVSrcPtr, vChrFilterSize,
681 alpSrcPtr, dest[0], dstW, dstY);
1674bd2a 682 }
4eb93bed
MN
683 } else {
684 yuv2anyX(c, vLumFilter + dstY * vLumFilterSize,
685 lumSrcPtr, vLumFilterSize,
686 vChrFilter + dstY * vChrFilterSize,
687 chrUSrcPtr, chrVSrcPtr, vChrFilterSize,
688 alpSrcPtr, dest, dstW, dstY);
1674bd2a 689 }
1674bd2a
RB
690 }
691 }
692
26b5ad25
LB
693 if (isPlanar(dstFormat) && isALPHA(dstFormat) && !alpPixBuf) {
694 int length = dstW;
695 int height = dstY - lastDstY;
696 if (is16BPS(c->dstFormat))
697 length *= 2;
698
699 if (is9_OR_10BPS(dstFormat)) {
700 const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(dstFormat);
701 fill_plane9or10(dst[3], dstStride[3], length, height, lastDstY,
2268db2c 702 255, desc->comp[3].depth, isBE(dstFormat));
26b5ad25
LB
703 } else
704 fillPlane(dst[3], dstStride[3], length, height, lastDstY, 255);
705 }
1674bd2a 706
1169f0d0 707#if HAVE_MMXEXT_INLINE
239fdf1b 708 if (av_get_cpu_flags() & AV_CPU_FLAG_MMXEXT)
ef0ee7f6 709 __asm__ volatile ("sfence" ::: "memory");
1674bd2a
RB
710#endif
711 emms_c();
712
713 /* store changed local vars back in the context */
ef0ee7f6
DB
714 c->dstY = dstY;
715 c->lumBufIndex = lumBufIndex;
716 c->chrBufIndex = chrBufIndex;
717 c->lastInLumBuf = lastInLumBuf;
718 c->lastInChrBuf = lastInChrBuf;
1674bd2a
RB
719
720 return dstY - lastDstY;
721}
722
1909f6b1 723static av_cold void sws_init_swscale(SwsContext *c)
1674bd2a 724{
716d413c 725 enum AVPixelFormat srcFormat = c->srcFormat;
1674bd2a 726
21449410
RB
727 ff_sws_init_output_funcs(c, &c->yuv2plane1, &c->yuv2planeX,
728 &c->yuv2nv12cX, &c->yuv2packed1,
4eb93bed 729 &c->yuv2packed2, &c->yuv2packedX, &c->yuv2anyX);
1674bd2a 730
2dd7a1c0 731 ff_sws_init_input_funcs(c);
1674bd2a 732
28c1115a
RB
733 if (c->srcBpc == 8) {
734 if (c->dstBpc <= 10) {
3f04ab4f 735 c->hyScale = c->hcScale = hScale8To15_c;
28c1115a
RB
736 if (c->flags & SWS_FAST_BILINEAR) {
737 c->hyscale_fast = hyscale_fast_c;
738 c->hcscale_fast = hcscale_fast_c;
739 }
62ee0e6a 740 } else {
3f04ab4f 741 c->hyScale = c->hcScale = hScale8To19_c;
1674bd2a 742 }
ef1ee362 743 } else {
ef0ee7f6
DB
744 c->hyScale = c->hcScale = c->dstBpc > 10 ? hScale16To19_c
745 : hScale16To15_c;
28c1115a 746 }
ef1ee362 747
28c1115a
RB
748 if (c->srcRange != c->dstRange && !isAnyRGB(c->dstFormat)) {
749 if (c->dstBpc <= 10) {
750 if (c->srcRange) {
751 c->lumConvertRange = lumRangeFromJpeg_c;
752 c->chrConvertRange = chrRangeFromJpeg_c;
753 } else {
754 c->lumConvertRange = lumRangeToJpeg_c;
755 c->chrConvertRange = chrRangeToJpeg_c;
756 }
757 } else {
ef1ee362
RB
758 if (c->srcRange) {
759 c->lumConvertRange = lumRangeFromJpeg16_c;
760 c->chrConvertRange = chrRangeFromJpeg16_c;
761 } else {
762 c->lumConvertRange = lumRangeToJpeg16_c;
763 c->chrConvertRange = chrRangeToJpeg16_c;
764 }
765 }
766 }
1674bd2a
RB
767
768 if (!(isGray(srcFormat) || isGray(c->dstFormat) ||
716d413c 769 srcFormat == AV_PIX_FMT_MONOBLACK || srcFormat == AV_PIX_FMT_MONOWHITE))
1674bd2a
RB
770 c->needs_hcscale = 1;
771}
397c035e 772
a4388ebd 773SwsFunc ff_getSwsFunc(SwsContext *c)
14d5c18c 774{
1909f6b1 775 sws_init_swscale(c);
86330b4c 776
c2503d9c
DB
777 if (ARCH_PPC)
778 ff_sws_init_swscale_ppc(c);
a5195839
DB
779 if (ARCH_X86)
780 ff_sws_init_swscale_x86(c);
33a0421b 781
1909f6b1 782 return swscale;
31190492 783}