Line | Branch | Exec | Source |
---|---|---|---|
1 | /* | ||
2 | * Copyright (C) 2001-2011 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 | #include <stdint.h> | ||
22 | #include <stdio.h> | ||
23 | #include <string.h> | ||
24 | |||
25 | #include "libavutil/avassert.h" | ||
26 | #include "libavutil/bswap.h" | ||
27 | #include "libavutil/common.h" | ||
28 | #include "libavutil/cpu.h" | ||
29 | #include "libavutil/emms.h" | ||
30 | #include "libavutil/intreadwrite.h" | ||
31 | #include "libavutil/mem.h" | ||
32 | #include "libavutil/mem_internal.h" | ||
33 | #include "libavutil/pixdesc.h" | ||
34 | #include "config.h" | ||
35 | #include "swscale_internal.h" | ||
36 | #include "swscale.h" | ||
37 | |||
38 | DECLARE_ALIGNED(8, const uint8_t, ff_dither_8x8_128)[9][8] = { | ||
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 | { 36, 68, 60, 92, 34, 66, 58, 90, }, | ||
48 | }; | ||
49 | |||
50 | DECLARE_ALIGNED(8, static const uint8_t, sws_pb_64)[8] = { | ||
51 | 64, 64, 64, 64, 64, 64, 64, 64 | ||
52 | }; | ||
53 | |||
54 | 5751 | static av_always_inline void fillPlane(uint8_t *plane, int stride, int width, | |
55 | int height, int y, uint8_t val) | ||
56 | { | ||
57 | int i; | ||
58 | 5751 | uint8_t *ptr = plane + stride * y; | |
59 |
2/2✓ Branch 0 taken 181272 times.
✓ Branch 1 taken 5751 times.
|
187023 | for (i = 0; i < height; i++) { |
60 | 181272 | memset(ptr, val, width); | |
61 | 181272 | ptr += stride; | |
62 | } | ||
63 | 5751 | } | |
64 | |||
65 | 38904 | static void hScale16To19_c(SwsContext *c, int16_t *_dst, int dstW, | |
66 | const uint8_t *_src, const int16_t *filter, | ||
67 | const int32_t *filterPos, int filterSize) | ||
68 | { | ||
69 | 38904 | const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->srcFormat); | |
70 | int i; | ||
71 | 38904 | int32_t *dst = (int32_t *) _dst; | |
72 | 38904 | const uint16_t *src = (const uint16_t *) _src; | |
73 | 38904 | int bits = desc->comp[0].depth - 1; | |
74 | 38904 | int sh = bits - 4; | |
75 | |||
76 |
5/6✓ Branch 1 taken 19008 times.
✓ Branch 2 taken 19896 times.
✗ Branch 3 not taken.
✓ Branch 4 taken 19008 times.
✓ Branch 5 taken 2040 times.
✓ Branch 6 taken 17856 times.
|
38904 | if ((isAnyRGB(c->srcFormat) || c->srcFormat==AV_PIX_FMT_PAL8) && desc->comp[0].depth<16) { |
77 | 2040 | sh = 9; | |
78 |
2/2✓ Branch 0 taken 4608 times.
✓ Branch 1 taken 32256 times.
|
36864 | } else if (desc->flags & AV_PIX_FMT_FLAG_FLOAT) { /* float input are process like uint 16bpc */ |
79 | 4608 | sh = 16 - 1 - 4; | |
80 | } | ||
81 | |||
82 |
2/2✓ Branch 0 taken 7001280 times.
✓ Branch 1 taken 38904 times.
|
7040184 | for (i = 0; i < dstW; i++) { |
83 | int j; | ||
84 | 7001280 | int srcPos = filterPos[i]; | |
85 | 7001280 | int val = 0; | |
86 | |||
87 |
2/2✓ Branch 0 taken 48127680 times.
✓ Branch 1 taken 7001280 times.
|
55128960 | for (j = 0; j < filterSize; j++) { |
88 | 48127680 | val += src[srcPos + j] * filter[filterSize * i + j]; | |
89 | } | ||
90 | // filter=14 bit, input=16 bit, output=30 bit, >> 11 makes 19 bit | ||
91 | 7001280 | dst[i] = FFMIN(val >> sh, (1 << 19) - 1); | |
92 | } | ||
93 | 38904 | } | |
94 | |||
95 | 6691145 | static void hScale16To15_c(SwsContext *c, int16_t *dst, int dstW, | |
96 | const uint8_t *_src, const int16_t *filter, | ||
97 | const int32_t *filterPos, int filterSize) | ||
98 | { | ||
99 | 6691145 | const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->srcFormat); | |
100 | int i; | ||
101 | 6691145 | const uint16_t *src = (const uint16_t *) _src; | |
102 | 6691145 | int sh = desc->comp[0].depth - 1; | |
103 | |||
104 |
2/2✓ Branch 0 taken 6104421 times.
✓ Branch 1 taken 586724 times.
|
6691145 | if (sh<15) { |
105 |
4/4✓ Branch 1 taken 2372644 times.
✓ Branch 2 taken 3731777 times.
✓ Branch 3 taken 1832660 times.
✓ Branch 4 taken 539984 times.
|
6104421 | sh = isAnyRGB(c->srcFormat) || c->srcFormat==AV_PIX_FMT_PAL8 ? 13 : (desc->comp[0].depth - 1); |
106 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 586724 times.
|
586724 | } else if (desc->flags & AV_PIX_FMT_FLAG_FLOAT) { /* float input are process like uint 16bpc */ |
107 | ✗ | sh = 16 - 1; | |
108 | } | ||
109 | |||
110 |
2/2✓ Branch 0 taken 1734221649 times.
✓ Branch 1 taken 6691145 times.
|
1740912794 | for (i = 0; i < dstW; i++) { |
111 | int j; | ||
112 | 1734221649 | int srcPos = filterPos[i]; | |
113 | 1734221649 | int val = 0; | |
114 | |||
115 |
2/2✓ Branch 0 taken 3369871281 times.
✓ Branch 1 taken 1734221649 times.
|
5104092930 | for (j = 0; j < filterSize; j++) { |
116 | 3369871281 | val += src[srcPos + j] * filter[filterSize * i + j]; | |
117 | } | ||
118 | // filter=14 bit, input=16 bit, output=30 bit, >> 15 makes 15 bit | ||
119 | 1734221649 | dst[i] = FFMIN(val >> sh, (1 << 15) - 1); | |
120 | } | ||
121 | 6691145 | } | |
122 | |||
123 | // bilinear / bicubic scaling | ||
124 | 13011581 | static void hScale8To15_c(SwsContext *c, int16_t *dst, int dstW, | |
125 | const uint8_t *src, const int16_t *filter, | ||
126 | const int32_t *filterPos, int filterSize) | ||
127 | { | ||
128 | int i; | ||
129 |
2/2✓ Branch 0 taken 3853699660 times.
✓ Branch 1 taken 13011581 times.
|
3866711241 | for (i = 0; i < dstW; i++) { |
130 | int j; | ||
131 | 3853699660 | int srcPos = filterPos[i]; | |
132 | 3853699660 | int val = 0; | |
133 |
2/2✓ Branch 0 taken 10274366018 times.
✓ Branch 1 taken 3853699660 times.
|
14128065678 | for (j = 0; j < filterSize; j++) { |
134 | 10274366018 | val += ((int)src[srcPos + j]) * filter[filterSize * i + j]; | |
135 | } | ||
136 | 3853699660 | dst[i] = FFMIN(val >> 7, (1 << 15) - 1); // the cubic equation does overflow ... | |
137 | } | ||
138 | 13011581 | } | |
139 | |||
140 | 891519 | static void hScale8To19_c(SwsContext *c, int16_t *_dst, int dstW, | |
141 | const uint8_t *src, const int16_t *filter, | ||
142 | const int32_t *filterPos, int filterSize) | ||
143 | { | ||
144 | int i; | ||
145 | 891519 | int32_t *dst = (int32_t *) _dst; | |
146 |
2/2✓ Branch 0 taken 281929948 times.
✓ Branch 1 taken 891519 times.
|
282821467 | for (i = 0; i < dstW; i++) { |
147 | int j; | ||
148 | 281929948 | int srcPos = filterPos[i]; | |
149 | 281929948 | int val = 0; | |
150 |
2/2✓ Branch 0 taken 502198060 times.
✓ Branch 1 taken 281929948 times.
|
784128008 | for (j = 0; j < filterSize; j++) { |
151 | 502198060 | val += ((int)src[srcPos + j]) * filter[filterSize * i + j]; | |
152 | } | ||
153 | 281929948 | dst[i] = FFMIN(val >> 3, (1 << 19) - 1); // the cubic equation does overflow ... | |
154 | } | ||
155 | 891519 | } | |
156 | |||
157 | // FIXME all pal and rgb srcFormats could do this conversion as well | ||
158 | // FIXME all scalers more complex than bilinear could do half of this transform | ||
159 | 209164 | static void chrRangeToJpeg_c(int16_t *dstU, int16_t *dstV, int width) | |
160 | { | ||
161 | int i; | ||
162 |
2/2✓ Branch 0 taken 43081168 times.
✓ Branch 1 taken 209164 times.
|
43290332 | for (i = 0; i < width; i++) { |
163 | 43081168 | dstU[i] = (FFMIN(dstU[i], 30775) * 4663 - 9289992) >> 12; // -264 | |
164 | 43081168 | dstV[i] = (FFMIN(dstV[i], 30775) * 4663 - 9289992) >> 12; // -264 | |
165 | } | ||
166 | 209164 | } | |
167 | |||
168 | 207909 | static void chrRangeFromJpeg_c(int16_t *dstU, int16_t *dstV, int width) | |
169 | { | ||
170 | int i; | ||
171 |
2/2✓ Branch 0 taken 35516898 times.
✓ Branch 1 taken 207909 times.
|
35724807 | for (i = 0; i < width; i++) { |
172 | 35516898 | dstU[i] = (dstU[i] * 1799 + 4081085) >> 11; // 1469 | |
173 | 35516898 | dstV[i] = (dstV[i] * 1799 + 4081085) >> 11; // 1469 | |
174 | } | ||
175 | 207909 | } | |
176 | |||
177 | 628619 | static void lumRangeToJpeg_c(int16_t *dst, int width) | |
178 | { | ||
179 | int i; | ||
180 |
2/2✓ Branch 0 taken 212505936 times.
✓ Branch 1 taken 628619 times.
|
213134555 | for (i = 0; i < width; i++) |
181 | 212505936 | dst[i] = (FFMIN(dst[i], 30189) * 19077 - 39057361) >> 14; | |
182 | 628619 | } | |
183 | |||
184 | 337183 | static void lumRangeFromJpeg_c(int16_t *dst, int width) | |
185 | { | ||
186 | int i; | ||
187 |
2/2✓ Branch 0 taken 115873848 times.
✓ Branch 1 taken 337183 times.
|
116211031 | for (i = 0; i < width; i++) |
188 | 115873848 | dst[i] = (dst[i] * 14071 + 33561947) >> 14; | |
189 | 337183 | } | |
190 | |||
191 | ✗ | static void chrRangeToJpeg16_c(int16_t *_dstU, int16_t *_dstV, int width) | |
192 | { | ||
193 | int i; | ||
194 | ✗ | int32_t *dstU = (int32_t *) _dstU; | |
195 | ✗ | int32_t *dstV = (int32_t *) _dstV; | |
196 | ✗ | for (i = 0; i < width; i++) { | |
197 | ✗ | dstU[i] = (FFMIN(dstU[i], 30775 << 4) * 4663 - (9289992 << 4)) >> 12; // -264 | |
198 | ✗ | dstV[i] = (FFMIN(dstV[i], 30775 << 4) * 4663 - (9289992 << 4)) >> 12; // -264 | |
199 | } | ||
200 | ✗ | } | |
201 | |||
202 | ✗ | static void chrRangeFromJpeg16_c(int16_t *_dstU, int16_t *_dstV, int width) | |
203 | { | ||
204 | int i; | ||
205 | ✗ | int32_t *dstU = (int32_t *) _dstU; | |
206 | ✗ | int32_t *dstV = (int32_t *) _dstV; | |
207 | ✗ | for (i = 0; i < width; i++) { | |
208 | ✗ | dstU[i] = (dstU[i] * 1799 + (4081085 << 4)) >> 11; // 1469 | |
209 | ✗ | dstV[i] = (dstV[i] * 1799 + (4081085 << 4)) >> 11; // 1469 | |
210 | } | ||
211 | ✗ | } | |
212 | |||
213 | 33375 | static void lumRangeToJpeg16_c(int16_t *_dst, int width) | |
214 | { | ||
215 | int i; | ||
216 | 33375 | int32_t *dst = (int32_t *) _dst; | |
217 |
2/2✓ Branch 0 taken 11676600 times.
✓ Branch 1 taken 33375 times.
|
11709975 | for (i = 0; i < width; i++) { |
218 | 11676600 | dst[i] = ((int)(FFMIN(dst[i], 30189 << 4) * 4769U - (39057361 << 2))) >> 12; | |
219 | } | ||
220 | 33375 | } | |
221 | |||
222 | ✗ | static void lumRangeFromJpeg16_c(int16_t *_dst, int width) | |
223 | { | ||
224 | int i; | ||
225 | ✗ | int32_t *dst = (int32_t *) _dst; | |
226 | ✗ | for (i = 0; i < width; i++) | |
227 | ✗ | dst[i] = (dst[i]*(14071/4) + (33561947<<4)/4)>>12; | |
228 | ✗ | } | |
229 | |||
230 | |||
231 | #define DEBUG_SWSCALE_BUFFERS 0 | ||
232 | #define DEBUG_BUFFERS(...) \ | ||
233 | if (DEBUG_SWSCALE_BUFFERS) \ | ||
234 | av_log(c, AV_LOG_DEBUG, __VA_ARGS__) | ||
235 | |||
236 | 68469 | static int swscale(SwsContext *c, const uint8_t *src[], | |
237 | int srcStride[], int srcSliceY, int srcSliceH, | ||
238 | uint8_t *dst[], int dstStride[], | ||
239 | int dstSliceY, int dstSliceH) | ||
240 | { | ||
241 |
4/4✓ Branch 0 taken 35528 times.
✓ Branch 1 taken 32941 times.
✓ Branch 2 taken 4129 times.
✓ Branch 3 taken 31399 times.
|
68469 | const int scale_dst = dstSliceY > 0 || dstSliceH < c->dstH; |
242 | |||
243 | /* load a few things into local vars to make the code more readable? | ||
244 | * and faster */ | ||
245 | 68469 | const int dstW = c->dstW; | |
246 | 68469 | int dstH = c->dstH; | |
247 | |||
248 | 68469 | const enum AVPixelFormat dstFormat = c->dstFormat; | |
249 | 68469 | const int flags = c->flags; | |
250 | 68469 | int32_t *vLumFilterPos = c->vLumFilterPos; | |
251 | 68469 | int32_t *vChrFilterPos = c->vChrFilterPos; | |
252 | |||
253 | 68469 | const int vLumFilterSize = c->vLumFilterSize; | |
254 | 68469 | const int vChrFilterSize = c->vChrFilterSize; | |
255 | |||
256 | 68469 | yuv2planar1_fn yuv2plane1 = c->yuv2plane1; | |
257 | 68469 | yuv2planarX_fn yuv2planeX = c->yuv2planeX; | |
258 | 68469 | yuv2interleavedX_fn yuv2nv12cX = c->yuv2nv12cX; | |
259 | 68469 | yuv2packed1_fn yuv2packed1 = c->yuv2packed1; | |
260 | 68469 | yuv2packed2_fn yuv2packed2 = c->yuv2packed2; | |
261 | 68469 | yuv2packedX_fn yuv2packedX = c->yuv2packedX; | |
262 | 68469 | yuv2anyX_fn yuv2anyX = c->yuv2anyX; | |
263 | 68469 | const int chrSrcSliceY = srcSliceY >> c->chrSrcVSubSample; | |
264 | 68469 | const int chrSrcSliceH = AV_CEIL_RSHIFT(srcSliceH, c->chrSrcVSubSample); | |
265 |
4/4✓ Branch 1 taken 59195 times.
✓ Branch 2 taken 9274 times.
✓ Branch 3 taken 1117 times.
✓ Branch 4 taken 58078 times.
|
127664 | int should_dither = isNBPS(c->srcFormat) || |
266 | 59195 | is16BPS(c->srcFormat); | |
267 | int lastDstY; | ||
268 | |||
269 | /* vars which will change and which we need to store back in the context */ | ||
270 | 68469 | int dstY = c->dstY; | |
271 | 68469 | int lastInLumBuf = c->lastInLumBuf; | |
272 | 68469 | int lastInChrBuf = c->lastInChrBuf; | |
273 | |||
274 | 68469 | int lumStart = 0; | |
275 | 68469 | int lumEnd = c->descIndex[0]; | |
276 | 68469 | int chrStart = lumEnd; | |
277 | 68469 | int chrEnd = c->descIndex[1]; | |
278 | 68469 | int vStart = chrEnd; | |
279 | 68469 | int vEnd = c->numDesc; | |
280 | 68469 | SwsSlice *src_slice = &c->slice[lumStart]; | |
281 | 68469 | SwsSlice *hout_slice = &c->slice[c->numSlice-2]; | |
282 | 68469 | SwsSlice *vout_slice = &c->slice[c->numSlice-1]; | |
283 | 68469 | SwsFilterDescriptor *desc = c->desc; | |
284 | |||
285 | 68469 | int needAlpha = c->needAlpha; | |
286 | |||
287 | 68469 | int hasLumHoles = 1; | |
288 | 68469 | int hasChrHoles = 1; | |
289 | |||
290 |
2/2✓ Branch 1 taken 11052 times.
✓ Branch 2 taken 57417 times.
|
68469 | if (isPacked(c->srcFormat)) { |
291 | 11052 | src[1] = | |
292 | 11052 | src[2] = | |
293 | 11052 | src[3] = src[0]; | |
294 | 11052 | srcStride[1] = | |
295 | 11052 | srcStride[2] = | |
296 | 11052 | srcStride[3] = srcStride[0]; | |
297 | } | ||
298 | 68469 | srcStride[1] *= 1 << c->vChrDrop; | |
299 | 68469 | srcStride[2] *= 1 << c->vChrDrop; | |
300 | |||
301 | DEBUG_BUFFERS("swscale() %p[%d] %p[%d] %p[%d] %p[%d] -> %p[%d] %p[%d] %p[%d] %p[%d]\n", | ||
302 | src[0], srcStride[0], src[1], srcStride[1], | ||
303 | src[2], srcStride[2], src[3], srcStride[3], | ||
304 | dst[0], dstStride[0], dst[1], dstStride[1], | ||
305 | dst[2], dstStride[2], dst[3], dstStride[3]); | ||
306 | DEBUG_BUFFERS("srcSliceY: %d srcSliceH: %d dstY: %d dstH: %d\n", | ||
307 | srcSliceY, srcSliceH, dstY, dstH); | ||
308 | DEBUG_BUFFERS("vLumFilterSize: %d vChrFilterSize: %d\n", | ||
309 | vLumFilterSize, vChrFilterSize); | ||
310 | |||
311 |
4/4✓ Branch 0 taken 63518 times.
✓ Branch 1 taken 4951 times.
✓ Branch 2 taken 57971 times.
✓ Branch 3 taken 5547 times.
|
68469 | if (dstStride[0]&15 || dstStride[1]&15 || |
312 |
2/4✓ Branch 0 taken 57971 times.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✓ Branch 3 taken 57971 times.
|
57971 | dstStride[2]&15 || dstStride[3]&15) { |
313 |
2/2✓ Branch 0 taken 4302 times.
✓ Branch 1 taken 6196 times.
|
10498 | SwsContext *const ctx = c->parent ? c->parent : c; |
314 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 10498 times.
|
10498 | if (flags & SWS_PRINT_INFO && |
315 | ✗ | !atomic_exchange_explicit(&ctx->stride_unaligned_warned, 1, memory_order_relaxed)) { | |
316 | ✗ | av_log(c, AV_LOG_WARNING, | |
317 | "Warning: dstStride is not aligned!\n" | ||
318 | " ->cannot do aligned memory accesses anymore\n"); | ||
319 | } | ||
320 | } | ||
321 | |||
322 | #if ARCH_X86 | ||
323 |
5/6✓ Branch 0 taken 66062 times.
✓ Branch 1 taken 2407 times.
✓ Branch 2 taken 64379 times.
✓ Branch 3 taken 1683 times.
✓ Branch 4 taken 64379 times.
✗ Branch 5 not taken.
|
68469 | if ( (uintptr_t)dst[0]&15 || (uintptr_t)dst[1]&15 || (uintptr_t)dst[2]&15 |
324 |
5/6✓ Branch 0 taken 64364 times.
✓ Branch 1 taken 15 times.
✓ Branch 2 taken 61704 times.
✓ Branch 3 taken 2660 times.
✓ Branch 4 taken 61704 times.
✗ Branch 5 not taken.
|
64379 | || (uintptr_t)src[0]&15 || (uintptr_t)src[1]&15 || (uintptr_t)src[2]&15 |
325 |
6/8✓ Branch 0 taken 58305 times.
✓ Branch 1 taken 3399 times.
✓ Branch 2 taken 54765 times.
✓ Branch 3 taken 3540 times.
✓ Branch 4 taken 54765 times.
✗ Branch 5 not taken.
✓ Branch 6 taken 54765 times.
✗ Branch 7 not taken.
|
61704 | || dstStride[0]&15 || dstStride[1]&15 || dstStride[2]&15 || dstStride[3]&15 |
326 |
6/8✓ Branch 0 taken 54760 times.
✓ Branch 1 taken 5 times.
✓ Branch 2 taken 54758 times.
✓ Branch 3 taken 2 times.
✓ Branch 4 taken 54758 times.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
✓ Branch 7 taken 54758 times.
|
54765 | || srcStride[0]&15 || srcStride[1]&15 || srcStride[2]&15 || srcStride[3]&15 |
327 | ) { | ||
328 |
2/2✓ Branch 0 taken 6633 times.
✓ Branch 1 taken 7078 times.
|
13711 | SwsContext *const ctx = c->parent ? c->parent : c; |
329 | 13711 | int cpu_flags = av_get_cpu_flags(); | |
330 |
1/4✗ Branch 0 not taken.
✓ Branch 1 taken 13711 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
|
13711 | if (flags & SWS_PRINT_INFO && HAVE_MMXEXT && (cpu_flags & AV_CPU_FLAG_SSE2) && |
331 | ✗ | !atomic_exchange_explicit(&ctx->stride_unaligned_warned,1, memory_order_relaxed)) { | |
332 | ✗ | av_log(c, AV_LOG_WARNING, "Warning: data is not aligned! This can lead to a speed loss\n"); | |
333 | } | ||
334 | } | ||
335 | #endif | ||
336 | |||
337 |
2/2✓ Branch 0 taken 37070 times.
✓ Branch 1 taken 31399 times.
|
68469 | if (scale_dst) { |
338 | 37070 | dstY = dstSliceY; | |
339 | 37070 | dstH = dstY + dstSliceH; | |
340 | 37070 | lastInLumBuf = -1; | |
341 | 37070 | lastInChrBuf = -1; | |
342 |
2/2✓ Branch 0 taken 31343 times.
✓ Branch 1 taken 56 times.
|
31399 | } else if (srcSliceY == 0) { |
343 | /* Note the user might start scaling the picture in the middle so this | ||
344 | * will not get executed. This is not really intended but works | ||
345 | * currently, so people might do it. */ | ||
346 | 31343 | dstY = 0; | |
347 | 31343 | lastInLumBuf = -1; | |
348 | 31343 | lastInChrBuf = -1; | |
349 | } | ||
350 | |||
351 |
2/2✓ Branch 0 taken 58078 times.
✓ Branch 1 taken 10391 times.
|
68469 | if (!should_dither) { |
352 | 58078 | c->chrDither8 = c->lumDither8 = sws_pb_64; | |
353 | } | ||
354 | 68469 | lastDstY = dstY; | |
355 | |||
356 | 68469 | ff_init_vscale_pfn(c, yuv2plane1, yuv2planeX, yuv2nv12cX, | |
357 | yuv2packed1, yuv2packed2, yuv2packedX, yuv2anyX, c->use_mmx_vfilter); | ||
358 | |||
359 | 68469 | ff_init_slice_from_src(src_slice, (uint8_t**)src, srcStride, c->srcW, | |
360 | srcSliceY, srcSliceH, chrSrcSliceY, chrSrcSliceH, 1); | ||
361 | |||
362 | 68469 | ff_init_slice_from_src(vout_slice, (uint8_t**)dst, dstStride, c->dstW, | |
363 | 68469 | dstY, dstSliceH, dstY >> c->chrDstVSubSample, | |
364 | 68469 | AV_CEIL_RSHIFT(dstSliceH, c->chrDstVSubSample), scale_dst); | |
365 |
2/2✓ Branch 0 taken 68413 times.
✓ Branch 1 taken 56 times.
|
68469 | if (srcSliceY == 0) { |
366 | 68413 | hout_slice->plane[0].sliceY = lastInLumBuf + 1; | |
367 | 68413 | hout_slice->plane[1].sliceY = lastInChrBuf + 1; | |
368 | 68413 | hout_slice->plane[2].sliceY = lastInChrBuf + 1; | |
369 | 68413 | hout_slice->plane[3].sliceY = lastInLumBuf + 1; | |
370 | |||
371 | 68413 | hout_slice->plane[0].sliceH = | |
372 | 68413 | hout_slice->plane[1].sliceH = | |
373 | 68413 | hout_slice->plane[2].sliceH = | |
374 | 68413 | hout_slice->plane[3].sliceH = 0; | |
375 | 68413 | hout_slice->width = dstW; | |
376 | } | ||
377 | |||
378 |
2/2✓ Branch 0 taken 11355519 times.
✓ Branch 1 taken 68413 times.
|
11423932 | for (; dstY < dstH; dstY++) { |
379 | 11355519 | const int chrDstY = dstY >> c->chrDstVSubSample; | |
380 | 11355519 | int use_mmx_vfilter= c->use_mmx_vfilter; | |
381 | |||
382 | // First line needed as input | ||
383 | 11355519 | const int firstLumSrcY = FFMAX(1 - vLumFilterSize, vLumFilterPos[dstY]); | |
384 |
2/2✓ Branch 0 taken 100 times.
✓ Branch 1 taken 11355419 times.
|
11355519 | const int firstLumSrcY2 = FFMAX(1 - vLumFilterSize, vLumFilterPos[FFMIN(dstY | ((1 << c->chrDstVSubSample) - 1), c->dstH - 1)]); |
385 | // First line needed as input | ||
386 | 11355519 | const int firstChrSrcY = FFMAX(1 - vChrFilterSize, vChrFilterPos[chrDstY]); | |
387 | |||
388 | // Last line needed as input | ||
389 | 11355519 | int lastLumSrcY = FFMIN(c->srcH, firstLumSrcY + vLumFilterSize) - 1; | |
390 | 11355519 | int lastLumSrcY2 = FFMIN(c->srcH, firstLumSrcY2 + vLumFilterSize) - 1; | |
391 | 11355519 | int lastChrSrcY = FFMIN(c->chrSrcH, firstChrSrcY + vChrFilterSize) - 1; | |
392 | int enough_lines; | ||
393 | |||
394 | int i; | ||
395 | int posY, cPosY, firstPosY, lastPosY, firstCPosY, lastCPosY; | ||
396 | |||
397 | // handle holes (FAST_BILINEAR & weird filters) | ||
398 |
2/2✓ Branch 0 taken 7560740 times.
✓ Branch 1 taken 3794779 times.
|
11355519 | if (firstLumSrcY > lastInLumBuf) { |
399 | |||
400 | 7560740 | hasLumHoles = lastInLumBuf != firstLumSrcY - 1; | |
401 |
2/2✓ Branch 0 taken 217041 times.
✓ Branch 1 taken 7343699 times.
|
7560740 | if (hasLumHoles) { |
402 | 217041 | hout_slice->plane[0].sliceY = firstLumSrcY; | |
403 | 217041 | hout_slice->plane[3].sliceY = firstLumSrcY; | |
404 | 217041 | hout_slice->plane[0].sliceH = | |
405 | 217041 | hout_slice->plane[3].sliceH = 0; | |
406 | } | ||
407 | |||
408 | 7560740 | lastInLumBuf = firstLumSrcY - 1; | |
409 | } | ||
410 |
2/2✓ Branch 0 taken 1400353 times.
✓ Branch 1 taken 9955166 times.
|
11355519 | if (firstChrSrcY > lastInChrBuf) { |
411 | |||
412 | 1400353 | hasChrHoles = lastInChrBuf != firstChrSrcY - 1; | |
413 |
2/2✓ Branch 0 taken 308691 times.
✓ Branch 1 taken 1091662 times.
|
1400353 | if (hasChrHoles) { |
414 | 308691 | hout_slice->plane[1].sliceY = firstChrSrcY; | |
415 | 308691 | hout_slice->plane[2].sliceY = firstChrSrcY; | |
416 | 308691 | hout_slice->plane[1].sliceH = | |
417 | 308691 | hout_slice->plane[2].sliceH = 0; | |
418 | } | ||
419 | |||
420 | 1400353 | lastInChrBuf = firstChrSrcY - 1; | |
421 | } | ||
422 | |||
423 | DEBUG_BUFFERS("dstY: %d\n", dstY); | ||
424 | DEBUG_BUFFERS("\tfirstLumSrcY: %d lastLumSrcY: %d lastInLumBuf: %d\n", | ||
425 | firstLumSrcY, lastLumSrcY, lastInLumBuf); | ||
426 | DEBUG_BUFFERS("\tfirstChrSrcY: %d lastChrSrcY: %d lastInChrBuf: %d\n", | ||
427 | firstChrSrcY, lastChrSrcY, lastInChrBuf); | ||
428 | |||
429 | // Do we have enough lines in this slice to output the dstY line | ||
430 |
2/2✓ Branch 0 taken 11355468 times.
✓ Branch 1 taken 51 times.
|
22710987 | enough_lines = lastLumSrcY2 < srcSliceY + srcSliceH && |
431 |
2/2✓ Branch 0 taken 11355463 times.
✓ Branch 1 taken 5 times.
|
11355468 | lastChrSrcY < AV_CEIL_RSHIFT(srcSliceY + srcSliceH, c->chrSrcVSubSample); |
432 | |||
433 |
2/2✓ Branch 0 taken 56 times.
✓ Branch 1 taken 11355463 times.
|
11355519 | if (!enough_lines) { |
434 | 56 | lastLumSrcY = srcSliceY + srcSliceH - 1; | |
435 | 56 | lastChrSrcY = chrSrcSliceY + chrSrcSliceH - 1; | |
436 | DEBUG_BUFFERS("buffering slice: lastLumSrcY %d lastChrSrcY %d\n", | ||
437 | lastLumSrcY, lastChrSrcY); | ||
438 | } | ||
439 | |||
440 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 11355519 times.
|
11355519 | av_assert0((lastLumSrcY - firstLumSrcY + 1) <= hout_slice->plane[0].available_lines); |
441 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 11355519 times.
|
11355519 | av_assert0((lastChrSrcY - firstChrSrcY + 1) <= hout_slice->plane[1].available_lines); |
442 | |||
443 | |||
444 | 11355519 | posY = hout_slice->plane[0].sliceY + hout_slice->plane[0].sliceH; | |
445 |
4/4✓ Branch 0 taken 1862066 times.
✓ Branch 1 taken 9493453 times.
✓ Branch 2 taken 1393876 times.
✓ Branch 3 taken 468190 times.
|
11355519 | if (posY <= lastLumSrcY && !hasLumHoles) { |
446 | 1393876 | firstPosY = FFMAX(firstLumSrcY, posY); | |
447 |
2/2✓ Branch 0 taken 22023 times.
✓ Branch 1 taken 1371853 times.
|
1393876 | lastPosY = FFMIN(firstLumSrcY + hout_slice->plane[0].available_lines - 1, srcSliceY + srcSliceH - 1); |
448 | } else { | ||
449 | 9961643 | firstPosY = posY; | |
450 | 9961643 | lastPosY = lastLumSrcY; | |
451 | } | ||
452 | |||
453 | 11355519 | cPosY = hout_slice->plane[1].sliceY + hout_slice->plane[1].sliceH; | |
454 |
4/4✓ Branch 0 taken 1603655 times.
✓ Branch 1 taken 9751864 times.
✓ Branch 2 taken 905507 times.
✓ Branch 3 taken 698148 times.
|
11355519 | if (cPosY <= lastChrSrcY && !hasChrHoles) { |
455 | 905507 | firstCPosY = FFMAX(firstChrSrcY, cPosY); | |
456 | 905507 | lastCPosY = FFMIN(firstChrSrcY + hout_slice->plane[1].available_lines - 1, AV_CEIL_RSHIFT(srcSliceY + srcSliceH, c->chrSrcVSubSample) - 1); | |
457 | } else { | ||
458 | 10450012 | firstCPosY = cPosY; | |
459 | 10450012 | lastCPosY = lastChrSrcY; | |
460 | } | ||
461 | |||
462 | 11355519 | ff_rotate_slice(hout_slice, lastPosY, lastCPosY); | |
463 | |||
464 |
2/2✓ Branch 0 taken 1862066 times.
✓ Branch 1 taken 9493453 times.
|
11355519 | if (posY < lastLumSrcY + 1) { |
465 |
2/2✓ Branch 0 taken 2137887 times.
✓ Branch 1 taken 1862066 times.
|
3999953 | for (i = lumStart; i < lumEnd; ++i) |
466 | 2137887 | desc[i].process(c, &desc[i], firstPosY, lastPosY - firstPosY + 1); | |
467 | } | ||
468 | |||
469 | 11355519 | lastInLumBuf = lastLumSrcY; | |
470 | |||
471 |
2/2✓ Branch 0 taken 1603655 times.
✓ Branch 1 taken 9751864 times.
|
11355519 | if (cPosY < lastChrSrcY + 1) { |
472 |
2/2✓ Branch 0 taken 1972259 times.
✓ Branch 1 taken 1603655 times.
|
3575914 | for (i = chrStart; i < chrEnd; ++i) |
473 | 1972259 | desc[i].process(c, &desc[i], firstCPosY, lastCPosY - firstCPosY + 1); | |
474 | } | ||
475 | |||
476 | 11355519 | lastInChrBuf = lastChrSrcY; | |
477 | |||
478 |
2/2✓ Branch 0 taken 56 times.
✓ Branch 1 taken 11355463 times.
|
11355519 | if (!enough_lines) |
479 | 56 | break; // we can't output a dstY line so let's try with the next slice | |
480 | |||
481 | #if HAVE_MMX_INLINE | ||
482 | 11355463 | ff_updateMMXDitherTables(c, dstY); | |
483 | #endif | ||
484 |
2/2✓ Branch 0 taken 3018178 times.
✓ Branch 1 taken 8337285 times.
|
11355463 | if (should_dither) { |
485 | 3018178 | c->chrDither8 = ff_dither_8x8_128[chrDstY & 7]; | |
486 | 3018178 | c->lumDither8 = ff_dither_8x8_128[dstY & 7]; | |
487 | } | ||
488 |
2/2✓ Branch 0 taken 70944 times.
✓ Branch 1 taken 11284519 times.
|
11355463 | if (dstY >= c->dstH - 2) { |
489 | /* hmm looks like we can't use MMX here without overwriting | ||
490 | * this array's tail */ | ||
491 | 70944 | ff_sws_init_output_funcs(c, &yuv2plane1, &yuv2planeX, &yuv2nv12cX, | |
492 | &yuv2packed1, &yuv2packed2, &yuv2packedX, &yuv2anyX); | ||
493 | 70944 | use_mmx_vfilter= 0; | |
494 | 70944 | ff_init_vscale_pfn(c, yuv2plane1, yuv2planeX, yuv2nv12cX, | |
495 | yuv2packed1, yuv2packed2, yuv2packedX, yuv2anyX, use_mmx_vfilter); | ||
496 | } | ||
497 | |||
498 |
2/2✓ Branch 0 taken 20173797 times.
✓ Branch 1 taken 11355463 times.
|
31529260 | for (i = vStart; i < vEnd; ++i) |
499 | 20173797 | desc[i].process(c, &desc[i], dstY, 1); | |
500 | } | ||
501 |
6/6✓ Branch 1 taken 50101 times.
✓ Branch 2 taken 18368 times.
✓ Branch 4 taken 8345 times.
✓ Branch 5 taken 41756 times.
✓ Branch 6 taken 6493 times.
✓ Branch 7 taken 1852 times.
|
68469 | if (isPlanar(dstFormat) && isALPHA(dstFormat) && !needAlpha) { |
502 | 6493 | int offset = lastDstY - dstSliceY; | |
503 | 6493 | int length = dstW; | |
504 | 6493 | int height = dstY - lastDstY; | |
505 | |||
506 |
4/4✓ Branch 1 taken 6167 times.
✓ Branch 2 taken 326 times.
✓ Branch 4 taken 334 times.
✓ Branch 5 taken 5833 times.
|
7153 | if (is16BPS(dstFormat) || isNBPS(dstFormat)) { |
507 | 660 | const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(dstFormat); | |
508 | 660 | fillPlane16(dst[3], dstStride[3], length, height, offset, | |
509 | 660 | 1, desc->comp[3].depth, | |
510 | isBE(dstFormat)); | ||
511 |
2/2✓ Branch 1 taken 82 times.
✓ Branch 2 taken 5751 times.
|
5833 | } else if (is32BPS(dstFormat)) { |
512 | 82 | const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(dstFormat); | |
513 | 164 | fillPlane32(dst[3], dstStride[3], length, height, offset, | |
514 | 82 | 1, desc->comp[3].depth, | |
515 | 82 | isBE(dstFormat), desc->flags & AV_PIX_FMT_FLAG_FLOAT); | |
516 | } else | ||
517 | 5751 | fillPlane(dst[3], dstStride[3], length, height, offset, 255); | |
518 | } | ||
519 | |||
520 | #if HAVE_MMXEXT_INLINE | ||
521 |
2/2✓ Branch 1 taken 183 times.
✓ Branch 2 taken 68286 times.
|
68469 | if (av_get_cpu_flags() & AV_CPU_FLAG_MMXEXT) |
522 | 183 | __asm__ volatile ("sfence" ::: "memory"); | |
523 | #endif | ||
524 | 68469 | emms_c(); | |
525 | |||
526 | /* store changed local vars back in the context */ | ||
527 | 68469 | c->dstY = dstY; | |
528 | 68469 | c->lastInLumBuf = lastInLumBuf; | |
529 | 68469 | c->lastInChrBuf = lastInChrBuf; | |
530 | |||
531 | 68469 | return dstY - lastDstY; | |
532 | } | ||
533 | |||
534 | 23689 | av_cold void ff_sws_init_range_convert(SwsContext *c) | |
535 | { | ||
536 | 23689 | c->lumConvertRange = NULL; | |
537 | 23689 | c->chrConvertRange = NULL; | |
538 |
4/4✓ Branch 0 taken 1577 times.
✓ Branch 1 taken 22112 times.
✓ Branch 3 taken 1206 times.
✓ Branch 4 taken 371 times.
|
23689 | if (c->srcRange != c->dstRange && !isAnyRGB(c->dstFormat)) { |
539 |
2/2✓ Branch 0 taken 1147 times.
✓ Branch 1 taken 59 times.
|
1206 | if (c->dstBpc <= 14) { |
540 |
2/2✓ Branch 0 taken 384 times.
✓ Branch 1 taken 763 times.
|
1147 | if (c->srcRange) { |
541 | 384 | c->lumConvertRange = lumRangeFromJpeg_c; | |
542 | 384 | c->chrConvertRange = chrRangeFromJpeg_c; | |
543 | } else { | ||
544 | 763 | c->lumConvertRange = lumRangeToJpeg_c; | |
545 | 763 | c->chrConvertRange = chrRangeToJpeg_c; | |
546 | } | ||
547 | } else { | ||
548 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 59 times.
|
59 | if (c->srcRange) { |
549 | ✗ | c->lumConvertRange = lumRangeFromJpeg16_c; | |
550 | ✗ | c->chrConvertRange = chrRangeFromJpeg16_c; | |
551 | } else { | ||
552 | 59 | c->lumConvertRange = lumRangeToJpeg16_c; | |
553 | 59 | c->chrConvertRange = chrRangeToJpeg16_c; | |
554 | } | ||
555 | } | ||
556 | } | ||
557 | 23689 | } | |
558 | |||
559 | 16731 | static av_cold void sws_init_swscale(SwsContext *c) | |
560 | { | ||
561 | 16731 | enum AVPixelFormat srcFormat = c->srcFormat; | |
562 | |||
563 | 16731 | ff_sws_init_output_funcs(c, &c->yuv2plane1, &c->yuv2planeX, | |
564 | &c->yuv2nv12cX, &c->yuv2packed1, | ||
565 | &c->yuv2packed2, &c->yuv2packedX, &c->yuv2anyX); | ||
566 | |||
567 | 16731 | ff_sws_init_input_funcs(c); | |
568 | |||
569 |
2/2✓ Branch 0 taken 15658 times.
✓ Branch 1 taken 1073 times.
|
16731 | if (c->srcBpc == 8) { |
570 |
2/2✓ Branch 0 taken 14686 times.
✓ Branch 1 taken 972 times.
|
15658 | if (c->dstBpc <= 14) { |
571 | 14686 | c->hyScale = c->hcScale = hScale8To15_c; | |
572 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 14686 times.
|
14686 | if (c->flags & SWS_FAST_BILINEAR) { |
573 | ✗ | c->hyscale_fast = ff_hyscale_fast_c; | |
574 | ✗ | c->hcscale_fast = ff_hcscale_fast_c; | |
575 | } | ||
576 | } else { | ||
577 | 972 | c->hyScale = c->hcScale = hScale8To19_c; | |
578 | } | ||
579 | } else { | ||
580 |
2/2✓ Branch 0 taken 107 times.
✓ Branch 1 taken 966 times.
|
1073 | c->hyScale = c->hcScale = c->dstBpc > 14 ? hScale16To19_c |
581 | : hScale16To15_c; | ||
582 | } | ||
583 | |||
584 | 16731 | ff_sws_init_range_convert(c); | |
585 | |||
586 |
8/8✓ Branch 1 taken 16543 times.
✓ Branch 2 taken 188 times.
✓ Branch 4 taken 16274 times.
✓ Branch 5 taken 269 times.
✓ Branch 6 taken 16272 times.
✓ Branch 7 taken 2 times.
✓ Branch 8 taken 16266 times.
✓ Branch 9 taken 6 times.
|
16731 | if (!(isGray(srcFormat) || isGray(c->dstFormat) || |
587 | srcFormat == AV_PIX_FMT_MONOBLACK || srcFormat == AV_PIX_FMT_MONOWHITE)) | ||
588 | 16266 | c->needs_hcscale = 1; | |
589 | 16731 | } | |
590 | |||
591 | 16731 | void ff_sws_init_scale(SwsContext *c) | |
592 | { | ||
593 | 16731 | sws_init_swscale(c); | |
594 | |||
595 | #if ARCH_PPC | ||
596 | ff_sws_init_swscale_ppc(c); | ||
597 | #elif ARCH_X86 | ||
598 | 16731 | ff_sws_init_swscale_x86(c); | |
599 | #elif ARCH_AARCH64 | ||
600 | ff_sws_init_swscale_aarch64(c); | ||
601 | #elif ARCH_ARM | ||
602 | ff_sws_init_swscale_arm(c); | ||
603 | #elif ARCH_LOONGARCH64 | ||
604 | ff_sws_init_swscale_loongarch(c); | ||
605 | #endif | ||
606 | 16731 | } | |
607 | |||
608 | 250374 | static void reset_ptr(const uint8_t *src[], enum AVPixelFormat format) | |
609 | { | ||
610 |
2/2✓ Branch 1 taken 181132 times.
✓ Branch 2 taken 69242 times.
|
250374 | if (!isALPHA(format)) |
611 | 181132 | src[3] = NULL; | |
612 |
2/2✓ Branch 1 taken 134102 times.
✓ Branch 2 taken 116272 times.
|
250374 | if (!isPlanar(format)) { |
613 | 134102 | src[3] = src[2] = NULL; | |
614 | |||
615 |
2/2✓ Branch 1 taken 96381 times.
✓ Branch 2 taken 37721 times.
|
134102 | if (!usePal(format)) |
616 | 96381 | src[1] = NULL; | |
617 | } | ||
618 | 250374 | } | |
619 | |||
620 | 250392 | static int check_image_pointers(const uint8_t * const data[4], enum AVPixelFormat pix_fmt, | |
621 | const int linesizes[4]) | ||
622 | { | ||
623 | 250392 | const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(pix_fmt); | |
624 | int i; | ||
625 | |||
626 | av_assert2(desc); | ||
627 | |||
628 |
2/2✓ Branch 0 taken 1001568 times.
✓ Branch 1 taken 250392 times.
|
1251960 | for (i = 0; i < 4; i++) { |
629 | 1001568 | int plane = desc->comp[i].plane; | |
630 |
2/4✓ Branch 0 taken 1001568 times.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✓ Branch 3 taken 1001568 times.
|
1001568 | if (!data[plane] || !linesizes[plane]) |
631 | ✗ | return 0; | |
632 | } | ||
633 | |||
634 | 250392 | return 1; | |
635 | } | ||
636 | |||
637 | 3 | static void xyz12Torgb48(struct SwsContext *c, uint16_t *dst, | |
638 | const uint16_t *src, int stride, int h) | ||
639 | { | ||
640 | int xp,yp; | ||
641 | 3 | const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->srcFormat); | |
642 | |||
643 |
2/2✓ Branch 0 taken 864 times.
✓ Branch 1 taken 3 times.
|
867 | for (yp=0; yp<h; yp++) { |
644 |
2/2✓ Branch 0 taken 304128 times.
✓ Branch 1 taken 864 times.
|
304992 | for (xp=0; xp+2<stride; xp+=3) { |
645 | int x, y, z, r, g, b; | ||
646 | |||
647 |
2/2✓ Branch 0 taken 101376 times.
✓ Branch 1 taken 202752 times.
|
304128 | if (desc->flags & AV_PIX_FMT_FLAG_BE) { |
648 | 101376 | x = AV_RB16(src + xp + 0); | |
649 | 101376 | y = AV_RB16(src + xp + 1); | |
650 | 101376 | z = AV_RB16(src + xp + 2); | |
651 | } else { | ||
652 | 202752 | x = AV_RL16(src + xp + 0); | |
653 | 202752 | y = AV_RL16(src + xp + 1); | |
654 | 202752 | z = AV_RL16(src + xp + 2); | |
655 | } | ||
656 | |||
657 | 304128 | x = c->xyzgamma[x>>4]; | |
658 | 304128 | y = c->xyzgamma[y>>4]; | |
659 | 304128 | z = c->xyzgamma[z>>4]; | |
660 | |||
661 | // convert from XYZlinear to sRGBlinear | ||
662 | 304128 | r = c->xyz2rgb_matrix[0][0] * x + | |
663 | 304128 | c->xyz2rgb_matrix[0][1] * y + | |
664 | 304128 | c->xyz2rgb_matrix[0][2] * z >> 12; | |
665 | 304128 | g = c->xyz2rgb_matrix[1][0] * x + | |
666 | 304128 | c->xyz2rgb_matrix[1][1] * y + | |
667 | 304128 | c->xyz2rgb_matrix[1][2] * z >> 12; | |
668 | 304128 | b = c->xyz2rgb_matrix[2][0] * x + | |
669 | 304128 | c->xyz2rgb_matrix[2][1] * y + | |
670 | 304128 | c->xyz2rgb_matrix[2][2] * z >> 12; | |
671 | |||
672 | // limit values to 12-bit depth | ||
673 | 304128 | r = av_clip_uintp2(r, 12); | |
674 | 304128 | g = av_clip_uintp2(g, 12); | |
675 | 304128 | b = av_clip_uintp2(b, 12); | |
676 | |||
677 | // convert from sRGBlinear to RGB and scale from 12bit to 16bit | ||
678 |
2/2✓ Branch 0 taken 101376 times.
✓ Branch 1 taken 202752 times.
|
304128 | if (desc->flags & AV_PIX_FMT_FLAG_BE) { |
679 | 101376 | AV_WB16(dst + xp + 0, c->rgbgamma[r] << 4); | |
680 | 101376 | AV_WB16(dst + xp + 1, c->rgbgamma[g] << 4); | |
681 | 101376 | AV_WB16(dst + xp + 2, c->rgbgamma[b] << 4); | |
682 | } else { | ||
683 | 202752 | AV_WL16(dst + xp + 0, c->rgbgamma[r] << 4); | |
684 | 202752 | AV_WL16(dst + xp + 1, c->rgbgamma[g] << 4); | |
685 | 202752 | AV_WL16(dst + xp + 2, c->rgbgamma[b] << 4); | |
686 | } | ||
687 | } | ||
688 | 864 | src += stride; | |
689 | 864 | dst += stride; | |
690 | } | ||
691 | 3 | } | |
692 | |||
693 | 34 | static void rgb48Toxyz12(struct SwsContext *c, uint16_t *dst, | |
694 | const uint16_t *src, int stride, int h) | ||
695 | { | ||
696 | int xp,yp; | ||
697 | 34 | const AVPixFmtDescriptor *desc = av_pix_fmt_desc_get(c->dstFormat); | |
698 | |||
699 |
2/2✓ Branch 0 taken 9416 times.
✓ Branch 1 taken 34 times.
|
9450 | for (yp=0; yp<h; yp++) { |
700 |
2/2✓ Branch 0 taken 3081280 times.
✓ Branch 1 taken 9416 times.
|
3090696 | for (xp=0; xp+2<stride; xp+=3) { |
701 | int x, y, z, r, g, b; | ||
702 | |||
703 |
2/2✓ Branch 0 taken 1439264 times.
✓ Branch 1 taken 1642016 times.
|
3081280 | if (desc->flags & AV_PIX_FMT_FLAG_BE) { |
704 | 1439264 | r = AV_RB16(src + xp + 0); | |
705 | 1439264 | g = AV_RB16(src + xp + 1); | |
706 | 1439264 | b = AV_RB16(src + xp + 2); | |
707 | } else { | ||
708 | 1642016 | r = AV_RL16(src + xp + 0); | |
709 | 1642016 | g = AV_RL16(src + xp + 1); | |
710 | 1642016 | b = AV_RL16(src + xp + 2); | |
711 | } | ||
712 | |||
713 | 3081280 | r = c->rgbgammainv[r>>4]; | |
714 | 3081280 | g = c->rgbgammainv[g>>4]; | |
715 | 3081280 | b = c->rgbgammainv[b>>4]; | |
716 | |||
717 | // convert from sRGBlinear to XYZlinear | ||
718 | 3081280 | x = c->rgb2xyz_matrix[0][0] * r + | |
719 | 3081280 | c->rgb2xyz_matrix[0][1] * g + | |
720 | 3081280 | c->rgb2xyz_matrix[0][2] * b >> 12; | |
721 | 3081280 | y = c->rgb2xyz_matrix[1][0] * r + | |
722 | 3081280 | c->rgb2xyz_matrix[1][1] * g + | |
723 | 3081280 | c->rgb2xyz_matrix[1][2] * b >> 12; | |
724 | 3081280 | z = c->rgb2xyz_matrix[2][0] * r + | |
725 | 3081280 | c->rgb2xyz_matrix[2][1] * g + | |
726 | 3081280 | c->rgb2xyz_matrix[2][2] * b >> 12; | |
727 | |||
728 | // limit values to 12-bit depth | ||
729 | 3081280 | x = av_clip_uintp2(x, 12); | |
730 | 3081280 | y = av_clip_uintp2(y, 12); | |
731 | 3081280 | z = av_clip_uintp2(z, 12); | |
732 | |||
733 | // convert from XYZlinear to X'Y'Z' and scale from 12bit to 16bit | ||
734 |
2/2✓ Branch 0 taken 1439264 times.
✓ Branch 1 taken 1642016 times.
|
3081280 | if (desc->flags & AV_PIX_FMT_FLAG_BE) { |
735 | 1439264 | AV_WB16(dst + xp + 0, c->xyzgammainv[x] << 4); | |
736 | 1439264 | AV_WB16(dst + xp + 1, c->xyzgammainv[y] << 4); | |
737 | 1439264 | AV_WB16(dst + xp + 2, c->xyzgammainv[z] << 4); | |
738 | } else { | ||
739 | 1642016 | AV_WL16(dst + xp + 0, c->xyzgammainv[x] << 4); | |
740 | 1642016 | AV_WL16(dst + xp + 1, c->xyzgammainv[y] << 4); | |
741 | 1642016 | AV_WL16(dst + xp + 2, c->xyzgammainv[z] << 4); | |
742 | } | ||
743 | } | ||
744 | 9416 | src += stride; | |
745 | 9416 | dst += stride; | |
746 | } | ||
747 | 34 | } | |
748 | |||
749 | 33301 | static void update_palette(SwsContext *c, const uint32_t *pal) | |
750 | { | ||
751 |
2/2✓ Branch 0 taken 8525056 times.
✓ Branch 1 taken 33301 times.
|
8558357 | for (int i = 0; i < 256; i++) { |
752 | 8525056 | int r, g, b, y, u, v, a = 0xff; | |
753 |
2/2✓ Branch 0 taken 8518656 times.
✓ Branch 1 taken 6400 times.
|
8525056 | if (c->srcFormat == AV_PIX_FMT_PAL8) { |
754 | 8518656 | uint32_t p = pal[i]; | |
755 | 8518656 | a = (p >> 24) & 0xFF; | |
756 | 8518656 | r = (p >> 16) & 0xFF; | |
757 | 8518656 | g = (p >> 8) & 0xFF; | |
758 | 8518656 | b = p & 0xFF; | |
759 |
2/2✓ Branch 0 taken 256 times.
✓ Branch 1 taken 6144 times.
|
6400 | } else if (c->srcFormat == AV_PIX_FMT_RGB8) { |
760 | 256 | r = ( i >> 5 ) * 36; | |
761 | 256 | g = ((i >> 2) & 7) * 36; | |
762 | 256 | b = ( i & 3) * 85; | |
763 |
2/2✓ Branch 0 taken 256 times.
✓ Branch 1 taken 5888 times.
|
6144 | } else if (c->srcFormat == AV_PIX_FMT_BGR8) { |
764 | 256 | b = ( i >> 6 ) * 85; | |
765 | 256 | g = ((i >> 3) & 7) * 36; | |
766 | 256 | r = ( i & 7) * 36; | |
767 |
2/2✓ Branch 0 taken 256 times.
✓ Branch 1 taken 5632 times.
|
5888 | } else if (c->srcFormat == AV_PIX_FMT_RGB4_BYTE) { |
768 | 256 | r = ( i >> 3 ) * 255; | |
769 | 256 | g = ((i >> 1) & 3) * 85; | |
770 | 256 | b = ( i & 1) * 255; | |
771 |
3/4✓ Branch 0 taken 256 times.
✓ Branch 1 taken 5376 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 256 times.
|
5632 | } else if (c->srcFormat == AV_PIX_FMT_GRAY8 || c->srcFormat == AV_PIX_FMT_GRAY8A) { |
772 | 5376 | r = g = b = i; | |
773 | } else { | ||
774 | av_assert1(c->srcFormat == AV_PIX_FMT_BGR4_BYTE); | ||
775 | 256 | b = ( i >> 3 ) * 255; | |
776 | 256 | g = ((i >> 1) & 3) * 85; | |
777 | 256 | r = ( i & 1) * 255; | |
778 | } | ||
779 | #define RGB2YUV_SHIFT 15 | ||
780 | #define BY ( (int) (0.114 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5)) | ||
781 | #define BV (-(int) (0.081 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5)) | ||
782 | #define BU ( (int) (0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5)) | ||
783 | #define GY ( (int) (0.587 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5)) | ||
784 | #define GV (-(int) (0.419 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5)) | ||
785 | #define GU (-(int) (0.331 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5)) | ||
786 | #define RY ( (int) (0.299 * 219 / 255 * (1 << RGB2YUV_SHIFT) + 0.5)) | ||
787 | #define RV ( (int) (0.500 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5)) | ||
788 | #define RU (-(int) (0.169 * 224 / 255 * (1 << RGB2YUV_SHIFT) + 0.5)) | ||
789 | |||
790 | 8525056 | y = av_clip_uint8((RY * r + GY * g + BY * b + ( 33 << (RGB2YUV_SHIFT - 1))) >> RGB2YUV_SHIFT); | |
791 | 8525056 | u = av_clip_uint8((RU * r + GU * g + BU * b + (257 << (RGB2YUV_SHIFT - 1))) >> RGB2YUV_SHIFT); | |
792 | 8525056 | v = av_clip_uint8((RV * r + GV * g + BV * b + (257 << (RGB2YUV_SHIFT - 1))) >> RGB2YUV_SHIFT); | |
793 | 8525056 | c->pal_yuv[i]= y + (u<<8) + (v<<16) + ((unsigned)a<<24); | |
794 | |||
795 |
2/4✓ Branch 0 taken 7630592 times.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✓ Branch 3 taken 894464 times.
|
8525056 | switch (c->dstFormat) { |
796 | 7630592 | case AV_PIX_FMT_BGR32: | |
797 | #if !HAVE_BIGENDIAN | ||
798 | case AV_PIX_FMT_RGB24: | ||
799 | #endif | ||
800 | 7630592 | c->pal_rgb[i]= r + (g<<8) + (b<<16) + ((unsigned)a<<24); | |
801 | 7630592 | break; | |
802 | ✗ | case AV_PIX_FMT_BGR32_1: | |
803 | #if HAVE_BIGENDIAN | ||
804 | case AV_PIX_FMT_BGR24: | ||
805 | #endif | ||
806 | ✗ | c->pal_rgb[i]= a + (r<<8) + (g<<16) + ((unsigned)b<<24); | |
807 | ✗ | break; | |
808 | ✗ | case AV_PIX_FMT_RGB32_1: | |
809 | #if HAVE_BIGENDIAN | ||
810 | case AV_PIX_FMT_RGB24: | ||
811 | #endif | ||
812 | ✗ | c->pal_rgb[i]= a + (b<<8) + (g<<16) + ((unsigned)r<<24); | |
813 | ✗ | break; | |
814 | 894464 | case AV_PIX_FMT_RGB32: | |
815 | #if !HAVE_BIGENDIAN | ||
816 | case AV_PIX_FMT_BGR24: | ||
817 | #endif | ||
818 | default: | ||
819 | 894464 | c->pal_rgb[i]= b + (g<<8) + (r<<16) + ((unsigned)a<<24); | |
820 | } | ||
821 | } | ||
822 | 33301 | } | |
823 | |||
824 | static int scale_internal(SwsContext *c, | ||
825 | const uint8_t * const srcSlice[], const int srcStride[], | ||
826 | int srcSliceY, int srcSliceH, | ||
827 | uint8_t *const dstSlice[], const int dstStride[], | ||
828 | int dstSliceY, int dstSliceH); | ||
829 | |||
830 | ✗ | static int scale_gamma(SwsContext *c, | |
831 | const uint8_t * const srcSlice[], const int srcStride[], | ||
832 | int srcSliceY, int srcSliceH, | ||
833 | uint8_t * const dstSlice[], const int dstStride[], | ||
834 | int dstSliceY, int dstSliceH) | ||
835 | { | ||
836 | ✗ | int ret = scale_internal(c->cascaded_context[0], | |
837 | srcSlice, srcStride, srcSliceY, srcSliceH, | ||
838 | ✗ | c->cascaded_tmp, c->cascaded_tmpStride, 0, c->srcH); | |
839 | |||
840 | ✗ | if (ret < 0) | |
841 | ✗ | return ret; | |
842 | |||
843 | ✗ | if (c->cascaded_context[2]) | |
844 | ✗ | ret = scale_internal(c->cascaded_context[1], (const uint8_t * const *)c->cascaded_tmp, | |
845 | ✗ | c->cascaded_tmpStride, srcSliceY, srcSliceH, | |
846 | ✗ | c->cascaded1_tmp, c->cascaded1_tmpStride, 0, c->dstH); | |
847 | else | ||
848 | ✗ | ret = scale_internal(c->cascaded_context[1], (const uint8_t * const *)c->cascaded_tmp, | |
849 | ✗ | c->cascaded_tmpStride, srcSliceY, srcSliceH, | |
850 | dstSlice, dstStride, dstSliceY, dstSliceH); | ||
851 | |||
852 | ✗ | if (ret < 0) | |
853 | ✗ | return ret; | |
854 | |||
855 | ✗ | if (c->cascaded_context[2]) { | |
856 | ✗ | ret = scale_internal(c->cascaded_context[2], (const uint8_t * const *)c->cascaded1_tmp, | |
857 | ✗ | c->cascaded1_tmpStride, c->cascaded_context[1]->dstY - ret, | |
858 | ✗ | c->cascaded_context[1]->dstY, | |
859 | dstSlice, dstStride, dstSliceY, dstSliceH); | ||
860 | } | ||
861 | ✗ | return ret; | |
862 | } | ||
863 | |||
864 | 9 | static int scale_cascaded(SwsContext *c, | |
865 | const uint8_t * const srcSlice[], const int srcStride[], | ||
866 | int srcSliceY, int srcSliceH, | ||
867 | uint8_t * const dstSlice[], const int dstStride[], | ||
868 | int dstSliceY, int dstSliceH) | ||
869 | { | ||
870 | 9 | int ret = scale_internal(c->cascaded_context[0], | |
871 | srcSlice, srcStride, srcSliceY, srcSliceH, | ||
872 | 9 | c->cascaded_tmp, c->cascaded_tmpStride, | |
873 | 9 | 0, c->cascaded_context[0]->dstH); | |
874 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 9 times.
|
9 | if (ret < 0) |
875 | ✗ | return ret; | |
876 | 9 | ret = scale_internal(c->cascaded_context[1], | |
877 | 9 | (const uint8_t * const * )c->cascaded_tmp, c->cascaded_tmpStride, | |
878 | 9 | 0, c->cascaded_context[0]->dstH, | |
879 | dstSlice, dstStride, dstSliceY, dstSliceH); | ||
880 | 9 | return ret; | |
881 | } | ||
882 | |||
883 | 125196 | static int scale_internal(SwsContext *c, | |
884 | const uint8_t * const srcSlice[], const int srcStride[], | ||
885 | int srcSliceY, int srcSliceH, | ||
886 | uint8_t *const dstSlice[], const int dstStride[], | ||
887 | int dstSliceY, int dstSliceH) | ||
888 | { | ||
889 |
4/4✓ Branch 0 taken 42740 times.
✓ Branch 1 taken 82456 times.
✓ Branch 2 taken 10364 times.
✓ Branch 3 taken 32376 times.
|
125196 | const int scale_dst = dstSliceY > 0 || dstSliceH < c->dstH; |
890 |
4/4✓ Branch 0 taken 32376 times.
✓ Branch 1 taken 92820 times.
✓ Branch 2 taken 32320 times.
✓ Branch 3 taken 56 times.
|
125196 | const int frame_start = scale_dst || !c->sliceDir; |
891 | int i, ret; | ||
892 | const uint8_t *src2[4]; | ||
893 | uint8_t *dst2[4]; | ||
894 |
1/2✓ Branch 1 taken 125196 times.
✗ Branch 2 not taken.
|
125196 | int macro_height_src = isBayer(c->srcFormat) ? 2 : (1 << c->chrSrcVSubSample); |
895 |
1/2✓ Branch 1 taken 125196 times.
✗ Branch 2 not taken.
|
125196 | int macro_height_dst = isBayer(c->dstFormat) ? 2 : (1 << c->chrDstVSubSample); |
896 | // copy strides, so they can safely be modified | ||
897 | int srcStride2[4]; | ||
898 | int dstStride2[4]; | ||
899 | 125196 | int srcSliceY_internal = srcSliceY; | |
900 | |||
901 |
4/8✓ Branch 0 taken 125196 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 125196 times.
✗ Branch 3 not taken.
✓ Branch 4 taken 125196 times.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
✓ Branch 7 taken 125196 times.
|
125196 | if (!srcStride || !dstStride || !dstSlice || !srcSlice) { |
902 | ✗ | av_log(c, AV_LOG_ERROR, "One of the input parameters to sws_scale() is NULL, please check the calling code\n"); | |
903 | ✗ | return AVERROR(EINVAL); | |
904 | } | ||
905 | |||
906 |
1/2✓ Branch 0 taken 125196 times.
✗ Branch 1 not taken.
|
125196 | if ((srcSliceY & (macro_height_src - 1)) || |
907 |
3/4✓ Branch 0 taken 81 times.
✓ Branch 1 taken 125115 times.
✓ Branch 2 taken 81 times.
✗ Branch 3 not taken.
|
125196 | ((srcSliceH & (macro_height_src - 1)) && srcSliceY + srcSliceH != c->srcH) || |
908 |
2/4✓ Branch 0 taken 125196 times.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✓ Branch 3 taken 125196 times.
|
250392 | srcSliceY + srcSliceH > c->srcH || |
909 |
0/2✗ Branch 1 not taken.
✗ Branch 2 not taken.
|
125196 | (isBayer(c->srcFormat) && srcSliceH <= 1)) { |
910 | ✗ | av_log(c, AV_LOG_ERROR, "Slice parameters %d, %d are invalid\n", srcSliceY, srcSliceH); | |
911 | ✗ | return AVERROR(EINVAL); | |
912 | } | ||
913 | |||
914 |
1/2✓ Branch 0 taken 125196 times.
✗ Branch 1 not taken.
|
125196 | if ((dstSliceY & (macro_height_dst - 1)) || |
915 |
3/4✓ Branch 0 taken 50 times.
✓ Branch 1 taken 125146 times.
✓ Branch 2 taken 50 times.
✗ Branch 3 not taken.
|
125196 | ((dstSliceH & (macro_height_dst - 1)) && dstSliceY + dstSliceH != c->dstH) || |
916 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 125196 times.
|
125196 | dstSliceY + dstSliceH > c->dstH) { |
917 | ✗ | av_log(c, AV_LOG_ERROR, "Slice parameters %d, %d are invalid\n", dstSliceY, dstSliceH); | |
918 | ✗ | return AVERROR(EINVAL); | |
919 | } | ||
920 | |||
921 |
1/2✗ Branch 1 not taken.
✓ Branch 2 taken 125196 times.
|
125196 | if (!check_image_pointers(srcSlice, c->srcFormat, srcStride)) { |
922 | ✗ | av_log(c, AV_LOG_ERROR, "bad src image pointers\n"); | |
923 | ✗ | return AVERROR(EINVAL); | |
924 | } | ||
925 |
1/2✗ Branch 1 not taken.
✓ Branch 2 taken 125196 times.
|
125196 | if (!check_image_pointers((const uint8_t* const*)dstSlice, c->dstFormat, dstStride)) { |
926 | ✗ | av_log(c, AV_LOG_ERROR, "bad dst image pointers\n"); | |
927 | ✗ | return AVERROR(EINVAL); | |
928 | } | ||
929 | |||
930 | // do not mess up sliceDir if we have a "trailing" 0-size slice | ||
931 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 125196 times.
|
125196 | if (srcSliceH == 0) |
932 | ✗ | return 0; | |
933 | |||
934 |
1/4✗ Branch 0 not taken.
✓ Branch 1 taken 125196 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
|
125196 | if (c->gamma_flag && c->cascaded_context[0]) |
935 | ✗ | return scale_gamma(c, srcSlice, srcStride, srcSliceY, srcSliceH, | |
936 | dstSlice, dstStride, dstSliceY, dstSliceH); | ||
937 | |||
938 |
4/6✓ Branch 0 taken 9 times.
✓ Branch 1 taken 125187 times.
✓ Branch 2 taken 9 times.
✗ Branch 3 not taken.
✓ Branch 4 taken 9 times.
✗ Branch 5 not taken.
|
125196 | if (c->cascaded_context[0] && srcSliceY == 0 && srcSliceH == c->cascaded_context[0]->srcH) |
939 | 9 | return scale_cascaded(c, srcSlice, srcStride, srcSliceY, srcSliceH, | |
940 | dstSlice, dstStride, dstSliceY, dstSliceH); | ||
941 | |||
942 |
7/8✓ Branch 0 taken 125131 times.
✓ Branch 1 taken 56 times.
✓ Branch 2 taken 59250 times.
✓ Branch 3 taken 65881 times.
✓ Branch 4 taken 870 times.
✓ Branch 5 taken 58380 times.
✓ Branch 6 taken 870 times.
✗ Branch 7 not taken.
|
125187 | if (!srcSliceY && (c->flags & SWS_BITEXACT) && c->dither == SWS_DITHER_ED && c->dither_error[0]) |
943 |
2/2✓ Branch 0 taken 3480 times.
✓ Branch 1 taken 870 times.
|
4350 | for (i = 0; i < 4; i++) |
944 | 3480 | memset(c->dither_error[i], 0, sizeof(c->dither_error[0][0]) * (c->dstW+2)); | |
945 | |||
946 |
2/2✓ Branch 1 taken 33301 times.
✓ Branch 2 taken 91886 times.
|
125187 | if (usePal(c->srcFormat)) |
947 | 33301 | update_palette(c, (const uint32_t *)srcSlice[1]); | |
948 | |||
949 | 125187 | memcpy(src2, srcSlice, sizeof(src2)); | |
950 | 125187 | memcpy(dst2, dstSlice, sizeof(dst2)); | |
951 | 125187 | memcpy(srcStride2, srcStride, sizeof(srcStride2)); | |
952 | 125187 | memcpy(dstStride2, dstStride, sizeof(dstStride2)); | |
953 | |||
954 |
4/4✓ Branch 0 taken 125131 times.
✓ Branch 1 taken 56 times.
✓ Branch 2 taken 32320 times.
✓ Branch 3 taken 92811 times.
|
125187 | if (frame_start && !scale_dst) { |
955 |
1/4✗ Branch 0 not taken.
✓ Branch 1 taken 32320 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
|
32320 | if (srcSliceY != 0 && srcSliceY + srcSliceH != c->srcH) { |
956 | ✗ | av_log(c, AV_LOG_ERROR, "Slices start in the middle!\n"); | |
957 | ✗ | return AVERROR(EINVAL); | |
958 | } | ||
959 | |||
960 |
1/2✓ Branch 0 taken 32320 times.
✗ Branch 1 not taken.
|
32320 | c->sliceDir = (srcSliceY == 0) ? 1 : -1; |
961 |
2/2✓ Branch 0 taken 92811 times.
✓ Branch 1 taken 56 times.
|
92867 | } else if (scale_dst) |
962 | 92811 | c->sliceDir = 1; | |
963 | |||
964 |
6/6✓ Branch 0 taken 4697 times.
✓ Branch 1 taken 120490 times.
✓ Branch 2 taken 4693 times.
✓ Branch 3 taken 4 times.
✓ Branch 5 taken 54 times.
✓ Branch 6 taken 4639 times.
|
125187 | if (c->src0Alpha && !c->dst0Alpha && isALPHA(c->dstFormat)) { |
965 | uint8_t *base; | ||
966 | int x,y; | ||
967 | |||
968 | 54 | av_fast_malloc(&c->rgb0_scratch, &c->rgb0_scratch_allocated, | |
969 | 54 | FFABS(srcStride[0]) * srcSliceH + 32); | |
970 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 54 times.
|
54 | if (!c->rgb0_scratch) |
971 | ✗ | return AVERROR(ENOMEM); | |
972 | |||
973 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 54 times.
|
54 | base = srcStride[0] < 0 ? c->rgb0_scratch - srcStride[0] * (srcSliceH-1) : |
974 | c->rgb0_scratch; | ||
975 |
2/2✓ Branch 0 taken 41958 times.
✓ Branch 1 taken 54 times.
|
42012 | for (y=0; y<srcSliceH; y++){ |
976 | 41958 | memcpy(base + srcStride[0]*y, src2[0] + srcStride[0]*y, 4*c->srcW); | |
977 |
2/2✓ Branch 0 taken 25822818 times.
✓ Branch 1 taken 41958 times.
|
25864776 | for (x=c->src0Alpha-1; x<4*c->srcW; x+=4) { |
978 | 25822818 | base[ srcStride[0]*y + x] = 0xFF; | |
979 | } | ||
980 | } | ||
981 | 54 | src2[0] = base; | |
982 | } | ||
983 | |||
984 |
5/8✓ Branch 0 taken 3 times.
✓ Branch 1 taken 125184 times.
✓ Branch 2 taken 2 times.
✓ Branch 3 taken 1 times.
✗ Branch 4 not taken.
✓ Branch 5 taken 2 times.
✗ Branch 6 not taken.
✗ Branch 7 not taken.
|
125187 | if (c->srcXYZ && !(c->dstXYZ && c->srcW==c->dstW && c->srcH==c->dstH)) { |
985 | uint8_t *base; | ||
986 | |||
987 | 3 | av_fast_malloc(&c->xyz_scratch, &c->xyz_scratch_allocated, | |
988 | 3 | FFABS(srcStride[0]) * srcSliceH + 32); | |
989 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 3 times.
|
3 | if (!c->xyz_scratch) |
990 | ✗ | return AVERROR(ENOMEM); | |
991 | |||
992 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 3 times.
|
3 | base = srcStride[0] < 0 ? c->xyz_scratch - srcStride[0] * (srcSliceH-1) : |
993 | c->xyz_scratch; | ||
994 | |||
995 | 3 | xyz12Torgb48(c, (uint16_t*)base, (const uint16_t*)src2[0], srcStride[0]/2, srcSliceH); | |
996 | 3 | src2[0] = base; | |
997 | } | ||
998 | |||
999 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 125187 times.
|
125187 | if (c->sliceDir != 1) { |
1000 | // slices go from bottom to top => we flip the image internally | ||
1001 | ✗ | for (i=0; i<4; i++) { | |
1002 | ✗ | srcStride2[i] *= -1; | |
1003 | ✗ | dstStride2[i] *= -1; | |
1004 | } | ||
1005 | |||
1006 | ✗ | src2[0] += (srcSliceH - 1) * srcStride[0]; | |
1007 | ✗ | if (!usePal(c->srcFormat)) | |
1008 | ✗ | src2[1] += ((srcSliceH >> c->chrSrcVSubSample) - 1) * srcStride[1]; | |
1009 | ✗ | src2[2] += ((srcSliceH >> c->chrSrcVSubSample) - 1) * srcStride[2]; | |
1010 | ✗ | src2[3] += (srcSliceH - 1) * srcStride[3]; | |
1011 | ✗ | dst2[0] += ( c->dstH - 1) * dstStride[0]; | |
1012 | ✗ | dst2[1] += ((c->dstH >> c->chrDstVSubSample) - 1) * dstStride[1]; | |
1013 | ✗ | dst2[2] += ((c->dstH >> c->chrDstVSubSample) - 1) * dstStride[2]; | |
1014 | ✗ | dst2[3] += ( c->dstH - 1) * dstStride[3]; | |
1015 | |||
1016 | ✗ | srcSliceY_internal = c->srcH-srcSliceY-srcSliceH; | |
1017 | } | ||
1018 | 125187 | reset_ptr(src2, c->srcFormat); | |
1019 | 125187 | reset_ptr((void*)dst2, c->dstFormat); | |
1020 | |||
1021 |
2/2✓ Branch 0 taken 56718 times.
✓ Branch 1 taken 68469 times.
|
125187 | if (c->convert_unscaled) { |
1022 | 56718 | int offset = srcSliceY_internal; | |
1023 | 56718 | int slice_h = srcSliceH; | |
1024 | |||
1025 | // for dst slice scaling, offset the pointers to match the unscaled API | ||
1026 |
2/2✓ Branch 0 taken 55741 times.
✓ Branch 1 taken 977 times.
|
56718 | if (scale_dst) { |
1027 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 55741 times.
|
55741 | av_assert0(offset == 0); |
1028 |
4/4✓ Branch 0 taken 118208 times.
✓ Branch 1 taken 288 times.
✓ Branch 2 taken 95100 times.
✓ Branch 3 taken 23108 times.
|
118496 | for (i = 0; i < 4 && src2[i]; i++) { |
1029 |
5/6✓ Branch 0 taken 95100 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 39359 times.
✓ Branch 3 taken 55741 times.
✓ Branch 5 taken 7014 times.
✓ Branch 6 taken 32345 times.
|
95100 | if (!src2[i] || (i > 0 && usePal(c->srcFormat))) |
1030 | break; | ||
1031 |
4/4✓ Branch 0 taken 58942 times.
✓ Branch 1 taken 3813 times.
✓ Branch 2 taken 2913 times.
✓ Branch 3 taken 56029 times.
|
62755 | src2[i] += (dstSliceY >> ((i == 1 || i == 2) ? c->chrSrcVSubSample : 0)) * srcStride2[i]; |
1032 | } | ||
1033 | |||
1034 |
4/4✓ Branch 0 taken 118064 times.
✓ Branch 1 taken 501 times.
✓ Branch 2 taken 62824 times.
✓ Branch 3 taken 55240 times.
|
118565 | for (i = 0; i < 4 && dst2[i]; i++) { |
1035 |
4/6✓ Branch 0 taken 62824 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 7083 times.
✓ Branch 3 taken 55741 times.
✓ Branch 5 taken 7083 times.
✗ Branch 6 not taken.
|
62824 | if (!dst2[i] || (i > 0 && usePal(c->dstFormat))) |
1036 | break; | ||
1037 |
4/4✓ Branch 0 taken 59083 times.
✓ Branch 1 taken 3741 times.
✓ Branch 2 taken 2841 times.
✓ Branch 3 taken 56242 times.
|
62824 | dst2[i] -= (dstSliceY >> ((i == 1 || i == 2) ? c->chrDstVSubSample : 0)) * dstStride2[i]; |
1038 | } | ||
1039 | 55741 | offset = dstSliceY; | |
1040 | 55741 | slice_h = dstSliceH; | |
1041 | } | ||
1042 | |||
1043 | 56718 | ret = c->convert_unscaled(c, src2, srcStride2, offset, slice_h, | |
1044 | dst2, dstStride2); | ||
1045 |
2/2✓ Branch 0 taken 55741 times.
✓ Branch 1 taken 977 times.
|
56718 | if (scale_dst) |
1046 | 55741 | dst2[0] += dstSliceY * dstStride2[0]; | |
1047 | } else { | ||
1048 | 68469 | ret = swscale(c, src2, srcStride2, srcSliceY_internal, srcSliceH, | |
1049 | dst2, dstStride2, dstSliceY, dstSliceH); | ||
1050 | } | ||
1051 | |||
1052 |
5/8✓ Branch 0 taken 34 times.
✓ Branch 1 taken 125153 times.
✓ Branch 2 taken 2 times.
✓ Branch 3 taken 32 times.
✗ Branch 4 not taken.
✓ Branch 5 taken 2 times.
✗ Branch 6 not taken.
✗ Branch 7 not taken.
|
125187 | if (c->dstXYZ && !(c->srcXYZ && c->srcW==c->dstW && c->srcH==c->dstH)) { |
1053 | uint16_t *dst16; | ||
1054 | |||
1055 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 34 times.
|
34 | if (scale_dst) { |
1056 | ✗ | dst16 = (uint16_t *)dst2[0]; | |
1057 | } else { | ||
1058 |
1/2✓ Branch 0 taken 34 times.
✗ Branch 1 not taken.
|
34 | int dstY = c->dstY ? c->dstY : srcSliceY + srcSliceH; |
1059 | |||
1060 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 34 times.
|
34 | av_assert0(dstY >= ret); |
1061 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 34 times.
|
34 | av_assert0(ret >= 0); |
1062 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 34 times.
|
34 | av_assert0(c->dstH >= dstY); |
1063 | 34 | dst16 = (uint16_t*)(dst2[0] + (dstY - ret) * dstStride2[0]); | |
1064 | } | ||
1065 | |||
1066 | /* replace on the same data */ | ||
1067 | 34 | rgb48Toxyz12(c, dst16, dst16, dstStride2[0]/2, ret); | |
1068 | } | ||
1069 | |||
1070 | /* reset slice direction at end of frame */ | ||
1071 |
3/4✓ Branch 0 taken 56 times.
✓ Branch 1 taken 125131 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 56 times.
|
125187 | if ((srcSliceY_internal + srcSliceH == c->srcH) || scale_dst) |
1072 | 125131 | c->sliceDir = 0; | |
1073 | |||
1074 | 125187 | return ret; | |
1075 | } | ||
1076 | |||
1077 | 42592 | void sws_frame_end(struct SwsContext *c) | |
1078 | { | ||
1079 | 42592 | av_frame_unref(c->frame_src); | |
1080 | 42592 | av_frame_unref(c->frame_dst); | |
1081 | 42592 | c->src_ranges.nb_ranges = 0; | |
1082 | 42592 | } | |
1083 | |||
1084 | 42592 | int sws_frame_start(struct SwsContext *c, AVFrame *dst, const AVFrame *src) | |
1085 | { | ||
1086 | 42592 | int ret, allocated = 0; | |
1087 | |||
1088 | 42592 | ret = av_frame_ref(c->frame_src, src); | |
1089 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 42592 times.
|
42592 | if (ret < 0) |
1090 | ✗ | return ret; | |
1091 | |||
1092 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 42592 times.
|
42592 | if (!dst->buf[0]) { |
1093 | ✗ | dst->width = c->dstW; | |
1094 | ✗ | dst->height = c->dstH; | |
1095 | ✗ | dst->format = c->dstFormat; | |
1096 | |||
1097 | ✗ | ret = av_frame_get_buffer(dst, 0); | |
1098 | ✗ | if (ret < 0) | |
1099 | ✗ | return ret; | |
1100 | ✗ | allocated = 1; | |
1101 | } | ||
1102 | |||
1103 | 42592 | ret = av_frame_ref(c->frame_dst, dst); | |
1104 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 42592 times.
|
42592 | if (ret < 0) { |
1105 | ✗ | if (allocated) | |
1106 | ✗ | av_frame_unref(dst); | |
1107 | |||
1108 | ✗ | return ret; | |
1109 | } | ||
1110 | |||
1111 | 42592 | return 0; | |
1112 | } | ||
1113 | |||
1114 | 42592 | int sws_send_slice(struct SwsContext *c, unsigned int slice_start, | |
1115 | unsigned int slice_height) | ||
1116 | { | ||
1117 | int ret; | ||
1118 | |||
1119 | 42592 | ret = ff_range_add(&c->src_ranges, slice_start, slice_height); | |
1120 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 42592 times.
|
42592 | if (ret < 0) |
1121 | ✗ | return ret; | |
1122 | |||
1123 | 42592 | return 0; | |
1124 | } | ||
1125 | |||
1126 | 42592 | unsigned int sws_receive_slice_alignment(const struct SwsContext *c) | |
1127 | { | ||
1128 |
2/2✓ Branch 0 taken 11228 times.
✓ Branch 1 taken 31364 times.
|
42592 | if (c->slice_ctx) |
1129 | 11228 | return c->slice_ctx[0]->dst_slice_align; | |
1130 | |||
1131 | 31364 | return c->dst_slice_align; | |
1132 | } | ||
1133 | |||
1134 | 42592 | int sws_receive_slice(struct SwsContext *c, unsigned int slice_start, | |
1135 | unsigned int slice_height) | ||
1136 | { | ||
1137 | 42592 | unsigned int align = sws_receive_slice_alignment(c); | |
1138 | uint8_t *dst[4]; | ||
1139 | |||
1140 | /* wait until complete input has been received */ | ||
1141 |
1/2✓ Branch 0 taken 42592 times.
✗ Branch 1 not taken.
|
42592 | if (!(c->src_ranges.nb_ranges == 1 && |
1142 |
1/2✓ Branch 0 taken 42592 times.
✗ Branch 1 not taken.
|
42592 | c->src_ranges.ranges[0].start == 0 && |
1143 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 42592 times.
|
42592 | c->src_ranges.ranges[0].len == c->srcH)) |
1144 | ✗ | return AVERROR(EAGAIN); | |
1145 | |||
1146 |
2/4✓ Branch 0 taken 42592 times.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✓ Branch 3 taken 42592 times.
|
42592 | if ((slice_start > 0 || slice_height < c->dstH) && |
1147 | ✗ | (slice_start % align || slice_height % align)) { | |
1148 | ✗ | av_log(c, AV_LOG_ERROR, | |
1149 | "Incorrectly aligned output: %u/%u not multiples of %u\n", | ||
1150 | slice_start, slice_height, align); | ||
1151 | ✗ | return AVERROR(EINVAL); | |
1152 | } | ||
1153 | |||
1154 |
2/2✓ Branch 0 taken 11228 times.
✓ Branch 1 taken 31364 times.
|
42592 | if (c->slicethread) { |
1155 |
2/2✓ Branch 0 taken 10363 times.
✓ Branch 1 taken 865 times.
|
11228 | int nb_jobs = c->slice_ctx[0]->dither == SWS_DITHER_ED ? 1 : c->nb_slice_ctx; |
1156 | 11228 | int ret = 0; | |
1157 | |||
1158 | 11228 | c->dst_slice_start = slice_start; | |
1159 | 11228 | c->dst_slice_height = slice_height; | |
1160 | |||
1161 | 11228 | avpriv_slicethread_execute(c->slicethread, nb_jobs, 0); | |
1162 | |||
1163 |
2/2✓ Branch 0 taken 94057 times.
✓ Branch 1 taken 11228 times.
|
105285 | for (int i = 0; i < c->nb_slice_ctx; i++) { |
1164 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 94057 times.
|
94057 | if (c->slice_err[i] < 0) { |
1165 | ✗ | ret = c->slice_err[i]; | |
1166 | ✗ | break; | |
1167 | } | ||
1168 | } | ||
1169 | |||
1170 | 11228 | memset(c->slice_err, 0, c->nb_slice_ctx * sizeof(*c->slice_err)); | |
1171 | |||
1172 | 11228 | return ret; | |
1173 | } | ||
1174 | |||
1175 |
2/2✓ Branch 0 taken 125456 times.
✓ Branch 1 taken 31364 times.
|
156820 | for (int i = 0; i < FF_ARRAY_ELEMS(dst); i++) { |
1176 | 125456 | ptrdiff_t offset = c->frame_dst->linesize[i] * (slice_start >> c->chrDstVSubSample); | |
1177 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 125456 times.
|
125456 | dst[i] = FF_PTR_ADD(c->frame_dst->data[i], offset); |
1178 | } | ||
1179 | |||
1180 | 31364 | return scale_internal(c, (const uint8_t * const *)c->frame_src->data, | |
1181 | 31364 | c->frame_src->linesize, 0, c->srcH, | |
1182 | 31364 | dst, c->frame_dst->linesize, slice_start, slice_height); | |
1183 | } | ||
1184 | |||
1185 | 42592 | int sws_scale_frame(struct SwsContext *c, AVFrame *dst, const AVFrame *src) | |
1186 | { | ||
1187 | int ret; | ||
1188 | |||
1189 | 42592 | ret = sws_frame_start(c, dst, src); | |
1190 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 42592 times.
|
42592 | if (ret < 0) |
1191 | ✗ | return ret; | |
1192 | |||
1193 | 42592 | ret = sws_send_slice(c, 0, src->height); | |
1194 |
1/2✓ Branch 0 taken 42592 times.
✗ Branch 1 not taken.
|
42592 | if (ret >= 0) |
1195 | 42592 | ret = sws_receive_slice(c, 0, dst->height); | |
1196 | |||
1197 | 42592 | sws_frame_end(c); | |
1198 | |||
1199 | 42592 | return ret; | |
1200 | } | ||
1201 | |||
1202 | /** | ||
1203 | * swscale wrapper, so we don't need to export the SwsContext. | ||
1204 | * Assumes planar YUV to be in YUV order instead of YVU. | ||
1205 | */ | ||
1206 | 138 | int attribute_align_arg sws_scale(struct SwsContext *c, | |
1207 | const uint8_t * const srcSlice[], | ||
1208 | const int srcStride[], int srcSliceY, | ||
1209 | int srcSliceH, uint8_t *const dst[], | ||
1210 | const int dstStride[]) | ||
1211 | { | ||
1212 |
1/2✗ Branch 0 not taken.
✓ Branch 1 taken 138 times.
|
138 | if (c->nb_slice_ctx) |
1213 | ✗ | c = c->slice_ctx[0]; | |
1214 | |||
1215 | 138 | return scale_internal(c, srcSlice, srcStride, srcSliceY, srcSliceH, | |
1216 | dst, dstStride, 0, c->dstH); | ||
1217 | } | ||
1218 | |||
1219 | 94057 | void ff_sws_slice_worker(void *priv, int jobnr, int threadnr, | |
1220 | int nb_jobs, int nb_threads) | ||
1221 | { | ||
1222 | 94057 | SwsContext *parent = priv; | |
1223 | 94057 | SwsContext *c = parent->slice_ctx[threadnr]; | |
1224 | |||
1225 | 94057 | const int slice_height = FFALIGN(FFMAX((parent->dst_slice_height + nb_jobs - 1) / nb_jobs, 1), | |
1226 | c->dst_slice_align); | ||
1227 | 94057 | const int slice_start = jobnr * slice_height; | |
1228 | 94057 | const int slice_end = FFMIN((jobnr + 1) * slice_height, parent->dst_slice_height); | |
1229 | 94057 | int err = 0; | |
1230 | |||
1231 |
2/2✓ Branch 0 taken 93676 times.
✓ Branch 1 taken 381 times.
|
94057 | if (slice_end > slice_start) { |
1232 | 93676 | uint8_t *dst[4] = { NULL }; | |
1233 | |||
1234 |
4/4✓ Branch 0 taken 247439 times.
✓ Branch 1 taken 7798 times.
✓ Branch 2 taken 161561 times.
✓ Branch 3 taken 85878 times.
|
255237 | for (int i = 0; i < FF_ARRAY_ELEMS(dst) && parent->frame_dst->data[i]; i++) { |
1235 |
4/4✓ Branch 0 taken 130981 times.
✓ Branch 1 taken 30580 times.
✓ Branch 2 taken 29507 times.
✓ Branch 3 taken 101474 times.
|
161561 | const int vshift = (i == 1 || i == 2) ? c->chrDstVSubSample : 0; |
1236 | 161561 | const ptrdiff_t offset = parent->frame_dst->linesize[i] * | |
1237 | 161561 | ((slice_start + parent->dst_slice_start) >> vshift); | |
1238 | |||
1239 | 161561 | dst[i] = parent->frame_dst->data[i] + offset; | |
1240 | } | ||
1241 | |||
1242 | 93676 | err = scale_internal(c, (const uint8_t * const *)parent->frame_src->data, | |
1243 | 93676 | parent->frame_src->linesize, 0, c->srcH, | |
1244 | 93676 | dst, parent->frame_dst->linesize, | |
1245 | 93676 | parent->dst_slice_start + slice_start, slice_end - slice_start); | |
1246 | } | ||
1247 | |||
1248 | 94057 | parent->slice_err[threadnr] = err; | |
1249 | 94057 | } | |
1250 |