Commit | Line | Data |
---|---|---|
5c018ee1 VG |
1 | /* |
2 | * DirectDraw Surface image decoder | |
3 | * Copyright (C) 2015 Vittorio Giovara <vittorio.giovara@gmail.com> | |
4 | * | |
5 | * This file is part of Libav. | |
6 | * | |
7 | * Libav is free software; you can redistribute it and/or | |
8 | * modify it under the terms of the GNU Lesser General Public | |
9 | * License as published by the Free Software Foundation; either | |
10 | * version 2.1 of the License, or (at your option) any later version. | |
11 | * | |
12 | * Libav is distributed in the hope that it will be useful, | |
13 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
14 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
15 | * Lesser General Public License for more details. | |
16 | * | |
17 | * You should have received a copy of the GNU Lesser General Public | |
18 | * License along with Libav; if not, write to the Free Software | |
19 | * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | |
20 | */ | |
21 | ||
22 | /** | |
23 | * @file | |
24 | * DDS decoder | |
25 | * | |
26 | * https://msdn.microsoft.com/en-us/library/bb943982%28v=vs.85%29.aspx | |
27 | */ | |
28 | ||
29 | #include <stdint.h> | |
30 | ||
31 | #include "libavutil/imgutils.h" | |
32 | ||
33 | #include "avcodec.h" | |
34 | #include "bytestream.h" | |
35 | #include "internal.h" | |
36 | #include "texturedsp.h" | |
37 | #include "thread.h" | |
38 | ||
39 | #define DDPF_FOURCC (1 << 2) | |
40 | #define DDPF_PALETTE (1 << 5) | |
41 | #define DDPF_NORMALMAP (1 << 31) | |
42 | ||
43 | enum DDSPostProc { | |
44 | DDS_NONE = 0, | |
45 | DDS_ALPHA_EXP, | |
46 | DDS_NORMAL_MAP, | |
47 | DDS_RAW_YCOCG, | |
48 | DDS_SWAP_ALPHA, | |
49 | DDS_SWIZZLE_A2XY, | |
50 | DDS_SWIZZLE_RBXG, | |
51 | DDS_SWIZZLE_RGXB, | |
52 | DDS_SWIZZLE_RXBG, | |
53 | DDS_SWIZZLE_RXGB, | |
54 | DDS_SWIZZLE_XGBR, | |
55 | DDS_SWIZZLE_XRBG, | |
56 | DDS_SWIZZLE_XGXR, | |
ea4d46e7 | 57 | }; |
5c018ee1 VG |
58 | |
59 | enum DDSDXGIFormat { | |
60 | DXGI_FORMAT_R16G16B16A16_TYPELESS = 9, | |
61 | DXGI_FORMAT_R16G16B16A16_FLOAT = 10, | |
62 | DXGI_FORMAT_R16G16B16A16_UNORM = 11, | |
63 | DXGI_FORMAT_R16G16B16A16_UINT = 12, | |
64 | DXGI_FORMAT_R16G16B16A16_SNORM = 13, | |
65 | DXGI_FORMAT_R16G16B16A16_SINT = 14, | |
66 | ||
67 | DXGI_FORMAT_R8G8B8A8_TYPELESS = 27, | |
68 | DXGI_FORMAT_R8G8B8A8_UNORM = 28, | |
69 | DXGI_FORMAT_R8G8B8A8_UNORM_SRGB = 29, | |
70 | DXGI_FORMAT_R8G8B8A8_UINT = 30, | |
71 | DXGI_FORMAT_R8G8B8A8_SNORM = 31, | |
72 | DXGI_FORMAT_R8G8B8A8_SINT = 32, | |
73 | ||
74 | DXGI_FORMAT_BC1_TYPELESS = 70, | |
75 | DXGI_FORMAT_BC1_UNORM = 71, | |
76 | DXGI_FORMAT_BC1_UNORM_SRGB = 72, | |
77 | DXGI_FORMAT_BC2_TYPELESS = 73, | |
78 | DXGI_FORMAT_BC2_UNORM = 74, | |
79 | DXGI_FORMAT_BC2_UNORM_SRGB = 75, | |
80 | DXGI_FORMAT_BC3_TYPELESS = 76, | |
81 | DXGI_FORMAT_BC3_UNORM = 77, | |
82 | DXGI_FORMAT_BC3_UNORM_SRGB = 78, | |
83 | DXGI_FORMAT_BC4_TYPELESS = 79, | |
84 | DXGI_FORMAT_BC4_UNORM = 80, | |
85 | DXGI_FORMAT_BC4_SNORM = 81, | |
86 | DXGI_FORMAT_BC5_TYPELESS = 82, | |
87 | DXGI_FORMAT_BC5_UNORM = 83, | |
88 | DXGI_FORMAT_BC5_SNORM = 84, | |
89 | DXGI_FORMAT_B5G6R5_UNORM = 85, | |
90 | DXGI_FORMAT_B8G8R8A8_UNORM = 87, | |
91 | DXGI_FORMAT_B8G8R8X8_UNORM = 88, | |
92 | DXGI_FORMAT_B8G8R8A8_TYPELESS = 90, | |
93 | DXGI_FORMAT_B8G8R8A8_UNORM_SRGB = 91, | |
94 | DXGI_FORMAT_B8G8R8X8_TYPELESS = 92, | |
95 | DXGI_FORMAT_B8G8R8X8_UNORM_SRGB = 93, | |
ea4d46e7 | 96 | }; |
5c018ee1 VG |
97 | |
98 | typedef struct DDSContext { | |
99 | TextureDSPContext texdsp; | |
100 | GetByteContext gbc; | |
101 | ||
102 | int compressed; | |
103 | int paletted; | |
104 | enum DDSPostProc postproc; | |
105 | ||
106 | const uint8_t *tex_data; // Compressed texture | |
107 | int tex_ratio; // Compression ratio | |
ebe8b5d9 | 108 | int slice_count; // Number of slices for threaded operations |
5c018ee1 VG |
109 | |
110 | /* Pointer to the selected compress or decompress function. */ | |
111 | int (*tex_funct)(uint8_t *dst, ptrdiff_t stride, const uint8_t *block); | |
112 | } DDSContext; | |
113 | ||
114 | static int parse_pixel_format(AVCodecContext *avctx) | |
115 | { | |
116 | DDSContext *ctx = avctx->priv_data; | |
117 | GetByteContext *gbc = &ctx->gbc; | |
118 | char buf[32]; | |
119 | uint32_t flags, fourcc, gimp_tag; | |
120 | enum DDSDXGIFormat dxgi; | |
121 | int size, bpp, r, g, b, a; | |
122 | int alpha_exponent, ycocg_classic, ycocg_scaled, normal_map, array; | |
123 | ||
124 | /* Alternative DDS implementations use reserved1 as custom header. */ | |
125 | bytestream2_skip(gbc, 4 * 3); | |
126 | gimp_tag = bytestream2_get_le32(gbc); | |
127 | alpha_exponent = gimp_tag == MKTAG('A', 'E', 'X', 'P'); | |
128 | ycocg_classic = gimp_tag == MKTAG('Y', 'C', 'G', '1'); | |
129 | ycocg_scaled = gimp_tag == MKTAG('Y', 'C', 'G', '2'); | |
130 | bytestream2_skip(gbc, 4 * 7); | |
131 | ||
132 | /* Now the real DDPF starts. */ | |
133 | size = bytestream2_get_le32(gbc); | |
134 | if (size != 32) { | |
135 | av_log(avctx, AV_LOG_ERROR, "Invalid pixel format header %d.\n", size); | |
136 | return AVERROR_INVALIDDATA; | |
137 | } | |
138 | flags = bytestream2_get_le32(gbc); | |
139 | ctx->compressed = flags & DDPF_FOURCC; | |
140 | ctx->paletted = flags & DDPF_PALETTE; | |
141 | normal_map = flags & DDPF_NORMALMAP; | |
142 | fourcc = bytestream2_get_le32(gbc); | |
143 | ||
144 | bpp = bytestream2_get_le32(gbc); // rgbbitcount | |
145 | r = bytestream2_get_le32(gbc); // rbitmask | |
146 | g = bytestream2_get_le32(gbc); // gbitmask | |
147 | b = bytestream2_get_le32(gbc); // bbitmask | |
148 | a = bytestream2_get_le32(gbc); // abitmask | |
149 | ||
150 | bytestream2_skip(gbc, 4); // caps | |
151 | bytestream2_skip(gbc, 4); // caps2 | |
152 | bytestream2_skip(gbc, 4); // caps3 | |
153 | bytestream2_skip(gbc, 4); // caps4 | |
154 | bytestream2_skip(gbc, 4); // reserved2 | |
155 | ||
156 | av_get_codec_tag_string(buf, sizeof(buf), fourcc); | |
157 | av_log(avctx, AV_LOG_VERBOSE, "fourcc %s bpp %d " | |
158 | "r 0x%x g 0x%x b 0x%x a 0x%x\n", buf, bpp, r, g, b, a); | |
159 | if (gimp_tag) { | |
160 | av_get_codec_tag_string(buf, sizeof(buf), gimp_tag); | |
161 | av_log(avctx, AV_LOG_VERBOSE, "and GIMP-DDS tag %s\n", buf); | |
162 | } | |
163 | ||
164 | if (ctx->compressed) | |
165 | avctx->pix_fmt = AV_PIX_FMT_RGBA; | |
166 | ||
167 | if (ctx->compressed) { | |
168 | switch (fourcc) { | |
169 | case MKTAG('D', 'X', 'T', '1'): | |
170 | ctx->tex_ratio = 8; | |
171 | ctx->tex_funct = ctx->texdsp.dxt1a_block; | |
172 | break; | |
173 | case MKTAG('D', 'X', 'T', '2'): | |
174 | ctx->tex_ratio = 16; | |
175 | ctx->tex_funct = ctx->texdsp.dxt2_block; | |
176 | break; | |
177 | case MKTAG('D', 'X', 'T', '3'): | |
178 | ctx->tex_ratio = 16; | |
179 | ctx->tex_funct = ctx->texdsp.dxt3_block; | |
180 | break; | |
181 | case MKTAG('D', 'X', 'T', '4'): | |
182 | ctx->tex_ratio = 16; | |
183 | ctx->tex_funct = ctx->texdsp.dxt4_block; | |
184 | break; | |
185 | case MKTAG('D', 'X', 'T', '5'): | |
186 | ctx->tex_ratio = 16; | |
187 | if (ycocg_scaled) | |
188 | ctx->tex_funct = ctx->texdsp.dxt5ys_block; | |
189 | else if (ycocg_classic) | |
190 | ctx->tex_funct = ctx->texdsp.dxt5y_block; | |
191 | else | |
192 | ctx->tex_funct = ctx->texdsp.dxt5_block; | |
193 | break; | |
194 | case MKTAG('R', 'X', 'G', 'B'): | |
195 | ctx->tex_ratio = 16; | |
196 | ctx->tex_funct = ctx->texdsp.dxt5_block; | |
197 | /* This format may be considered as a normal map, | |
198 | * but it is handled differently in a separate postproc. */ | |
199 | ctx->postproc = DDS_SWIZZLE_RXGB; | |
200 | normal_map = 0; | |
201 | break; | |
202 | case MKTAG('A', 'T', 'I', '1'): | |
203 | case MKTAG('B', 'C', '4', 'U'): | |
204 | ctx->tex_ratio = 8; | |
205 | ctx->tex_funct = ctx->texdsp.rgtc1u_block; | |
206 | break; | |
207 | case MKTAG('B', 'C', '4', 'S'): | |
208 | ctx->tex_ratio = 8; | |
209 | ctx->tex_funct = ctx->texdsp.rgtc1s_block; | |
210 | break; | |
211 | case MKTAG('A', 'T', 'I', '2'): | |
212 | /* RGT2 variant with swapped R and G (3Dc)*/ | |
213 | ctx->tex_ratio = 16; | |
214 | ctx->tex_funct = ctx->texdsp.dxn3dc_block; | |
215 | break; | |
216 | case MKTAG('B', 'C', '5', 'U'): | |
217 | ctx->tex_ratio = 16; | |
218 | ctx->tex_funct = ctx->texdsp.rgtc2u_block; | |
219 | break; | |
220 | case MKTAG('B', 'C', '5', 'S'): | |
221 | ctx->tex_ratio = 16; | |
222 | ctx->tex_funct = ctx->texdsp.rgtc2s_block; | |
223 | break; | |
224 | case MKTAG('U', 'Y', 'V', 'Y'): | |
225 | ctx->compressed = 0; | |
226 | avctx->pix_fmt = AV_PIX_FMT_UYVY422; | |
227 | break; | |
228 | case MKTAG('Y', 'U', 'Y', '2'): | |
229 | ctx->compressed = 0; | |
230 | avctx->pix_fmt = AV_PIX_FMT_YUYV422; | |
231 | break; | |
232 | case MKTAG('P', '8', ' ', ' '): | |
233 | /* ATI Palette8, same as normal palette */ | |
234 | ctx->compressed = 0; | |
235 | ctx->paletted = 1; | |
236 | avctx->pix_fmt = AV_PIX_FMT_PAL8; | |
237 | break; | |
238 | case MKTAG('D', 'X', '1', '0'): | |
239 | /* DirectX 10 extra header */ | |
240 | dxgi = bytestream2_get_le32(gbc); | |
241 | bytestream2_skip(gbc, 4); // resourceDimension | |
242 | bytestream2_skip(gbc, 4); // miscFlag | |
243 | array = bytestream2_get_le32(gbc); | |
244 | bytestream2_skip(gbc, 4); // miscFlag2 | |
245 | ||
246 | if (array != 0) | |
247 | av_log(avctx, AV_LOG_VERBOSE, | |
248 | "Found array of size %d (ignored).\n", array); | |
249 | ||
250 | /* Only BC[1-5] are actually compressed. */ | |
251 | ctx->compressed = (dxgi >= 70) && (dxgi <= 84); | |
252 | ||
253 | av_log(avctx, AV_LOG_VERBOSE, "DXGI format %d.\n", dxgi); | |
254 | switch (dxgi) { | |
255 | /* RGB types. */ | |
256 | case DXGI_FORMAT_R16G16B16A16_TYPELESS: | |
257 | case DXGI_FORMAT_R16G16B16A16_FLOAT: | |
258 | case DXGI_FORMAT_R16G16B16A16_UNORM: | |
259 | case DXGI_FORMAT_R16G16B16A16_UINT: | |
260 | case DXGI_FORMAT_R16G16B16A16_SNORM: | |
261 | case DXGI_FORMAT_R16G16B16A16_SINT: | |
262 | avctx->pix_fmt = AV_PIX_FMT_BGRA64; | |
263 | break; | |
264 | case DXGI_FORMAT_R8G8B8A8_UNORM_SRGB: | |
265 | avctx->colorspace = AVCOL_SPC_RGB; | |
266 | case DXGI_FORMAT_R8G8B8A8_TYPELESS: | |
267 | case DXGI_FORMAT_R8G8B8A8_UNORM: | |
268 | case DXGI_FORMAT_R8G8B8A8_UINT: | |
269 | case DXGI_FORMAT_R8G8B8A8_SNORM: | |
270 | case DXGI_FORMAT_R8G8B8A8_SINT: | |
271 | avctx->pix_fmt = AV_PIX_FMT_BGRA; | |
272 | break; | |
273 | case DXGI_FORMAT_B8G8R8A8_UNORM_SRGB: | |
274 | avctx->colorspace = AVCOL_SPC_RGB; | |
275 | case DXGI_FORMAT_B8G8R8A8_TYPELESS: | |
276 | case DXGI_FORMAT_B8G8R8A8_UNORM: | |
277 | avctx->pix_fmt = AV_PIX_FMT_RGBA; | |
278 | break; | |
279 | case DXGI_FORMAT_B8G8R8X8_UNORM_SRGB: | |
280 | avctx->colorspace = AVCOL_SPC_RGB; | |
281 | case DXGI_FORMAT_B8G8R8X8_TYPELESS: | |
282 | case DXGI_FORMAT_B8G8R8X8_UNORM: | |
283 | avctx->pix_fmt = AV_PIX_FMT_RGBA; // opaque | |
284 | break; | |
285 | case DXGI_FORMAT_B5G6R5_UNORM: | |
286 | avctx->pix_fmt = AV_PIX_FMT_RGB565LE; | |
287 | break; | |
288 | /* Texture types. */ | |
289 | case DXGI_FORMAT_BC1_UNORM_SRGB: | |
290 | avctx->colorspace = AVCOL_SPC_RGB; | |
291 | case DXGI_FORMAT_BC1_TYPELESS: | |
292 | case DXGI_FORMAT_BC1_UNORM: | |
293 | ctx->tex_ratio = 8; | |
294 | ctx->tex_funct = ctx->texdsp.dxt1a_block; | |
295 | break; | |
296 | case DXGI_FORMAT_BC2_UNORM_SRGB: | |
297 | avctx->colorspace = AVCOL_SPC_RGB; | |
298 | case DXGI_FORMAT_BC2_TYPELESS: | |
299 | case DXGI_FORMAT_BC2_UNORM: | |
300 | ctx->tex_ratio = 16; | |
301 | ctx->tex_funct = ctx->texdsp.dxt3_block; | |
302 | break; | |
303 | case DXGI_FORMAT_BC3_UNORM_SRGB: | |
304 | avctx->colorspace = AVCOL_SPC_RGB; | |
305 | case DXGI_FORMAT_BC3_TYPELESS: | |
306 | case DXGI_FORMAT_BC3_UNORM: | |
307 | ctx->tex_ratio = 16; | |
308 | ctx->tex_funct = ctx->texdsp.dxt5_block; | |
309 | break; | |
310 | case DXGI_FORMAT_BC4_TYPELESS: | |
311 | case DXGI_FORMAT_BC4_UNORM: | |
312 | ctx->tex_ratio = 8; | |
313 | ctx->tex_funct = ctx->texdsp.rgtc1u_block; | |
314 | break; | |
315 | case DXGI_FORMAT_BC4_SNORM: | |
316 | ctx->tex_ratio = 8; | |
317 | ctx->tex_funct = ctx->texdsp.rgtc1s_block; | |
318 | break; | |
319 | case DXGI_FORMAT_BC5_TYPELESS: | |
320 | case DXGI_FORMAT_BC5_UNORM: | |
321 | ctx->tex_ratio = 16; | |
322 | ctx->tex_funct = ctx->texdsp.rgtc2u_block; | |
323 | break; | |
324 | case DXGI_FORMAT_BC5_SNORM: | |
325 | ctx->tex_ratio = 16; | |
326 | ctx->tex_funct = ctx->texdsp.rgtc2s_block; | |
327 | break; | |
328 | default: | |
329 | av_log(avctx, AV_LOG_ERROR, | |
330 | "Unsupported DXGI format %d.\n", dxgi); | |
331 | return AVERROR_INVALIDDATA; | |
332 | } | |
333 | break; | |
334 | default: | |
335 | av_log(avctx, AV_LOG_ERROR, "Unsupported %s fourcc.\n", buf); | |
336 | return AVERROR_INVALIDDATA; | |
337 | } | |
338 | } else if (ctx->paletted) { | |
339 | if (bpp == 8) { | |
340 | avctx->pix_fmt = AV_PIX_FMT_PAL8; | |
341 | } else { | |
342 | av_log(avctx, AV_LOG_ERROR, "Unsupported palette bpp %d.\n", bpp); | |
343 | return AVERROR_INVALIDDATA; | |
344 | } | |
345 | } else { | |
346 | /* 8 bpp */ | |
347 | if (bpp == 8 && r == 0xff && g == 0 && b == 0 && a == 0) | |
348 | avctx->pix_fmt = AV_PIX_FMT_GRAY8; | |
349 | /* 16 bpp */ | |
350 | else if (bpp == 16 && r == 0xff && g == 0 && b == 0 && a == 0xff00) | |
351 | avctx->pix_fmt = AV_PIX_FMT_YA8; | |
352 | else if (bpp == 16 && r == 0xffff && g == 0 && b == 0 && a == 0) | |
353 | avctx->pix_fmt = AV_PIX_FMT_GRAY16LE; | |
354 | else if (bpp == 16 && r == 0xf800 && g == 0x7e0 && b == 0x1f && a == 0) | |
355 | avctx->pix_fmt = AV_PIX_FMT_RGB565LE; | |
356 | /* 24 bpp */ | |
357 | else if (bpp == 24 && r == 0xff0000 && g == 0xff00 && b == 0xff && a == 0) | |
358 | avctx->pix_fmt = AV_PIX_FMT_BGR24; | |
359 | /* 32 bpp */ | |
360 | else if (bpp == 32 && r == 0xff0000 && g == 0xff00 && b == 0xff && a == 0) | |
5c018ee1 | 361 | avctx->pix_fmt = AV_PIX_FMT_BGRA; // opaque |
d08d8b61 MN |
362 | else if (bpp == 32 && r == 0xff && g == 0xff00 && b == 0xff0000 && a == 0) |
363 | avctx->pix_fmt = AV_PIX_FMT_RGBA; // opaque | |
5c018ee1 | 364 | else if (bpp == 32 && r == 0xff0000 && g == 0xff00 && b == 0xff && a == 0xff000000) |
5c018ee1 | 365 | avctx->pix_fmt = AV_PIX_FMT_BGRA; |
d08d8b61 MN |
366 | else if (bpp == 32 && r == 0xff && g == 0xff00 && b == 0xff0000 && a == 0xff000000) |
367 | avctx->pix_fmt = AV_PIX_FMT_RGBA; | |
5c018ee1 VG |
368 | /* give up */ |
369 | else { | |
370 | av_log(avctx, AV_LOG_ERROR, "Unknown pixel format " | |
371 | "[bpp %d r 0x%x g 0x%x b 0x%x a 0x%x].\n", bpp, r, g, b, a); | |
372 | return AVERROR_INVALIDDATA; | |
373 | } | |
374 | } | |
375 | ||
376 | /* Set any remaining post-proc that should happen before frame is ready. */ | |
377 | if (alpha_exponent) | |
378 | ctx->postproc = DDS_ALPHA_EXP; | |
379 | else if (normal_map) | |
380 | ctx->postproc = DDS_NORMAL_MAP; | |
381 | else if (ycocg_classic && !ctx->compressed) | |
382 | ctx->postproc = DDS_RAW_YCOCG; | |
383 | else if (avctx->pix_fmt == AV_PIX_FMT_YA8) | |
384 | ctx->postproc = DDS_SWAP_ALPHA; | |
385 | ||
386 | /* ATI/NVidia variants sometimes add swizzling in bpp. */ | |
387 | switch (bpp) { | |
388 | case MKTAG('A', '2', 'X', 'Y'): | |
389 | ctx->postproc = DDS_SWIZZLE_A2XY; | |
390 | break; | |
391 | case MKTAG('x', 'G', 'B', 'R'): | |
392 | ctx->postproc = DDS_SWIZZLE_XGBR; | |
393 | break; | |
394 | case MKTAG('x', 'R', 'B', 'G'): | |
395 | ctx->postproc = DDS_SWIZZLE_XRBG; | |
396 | break; | |
397 | case MKTAG('R', 'B', 'x', 'G'): | |
398 | ctx->postproc = DDS_SWIZZLE_RBXG; | |
399 | break; | |
400 | case MKTAG('R', 'G', 'x', 'B'): | |
401 | ctx->postproc = DDS_SWIZZLE_RGXB; | |
402 | break; | |
403 | case MKTAG('R', 'x', 'B', 'G'): | |
404 | ctx->postproc = DDS_SWIZZLE_RXBG; | |
405 | break; | |
406 | case MKTAG('x', 'G', 'x', 'R'): | |
407 | ctx->postproc = DDS_SWIZZLE_XGXR; | |
408 | break; | |
409 | case MKTAG('A', '2', 'D', '5'): | |
410 | ctx->postproc = DDS_NORMAL_MAP; | |
411 | break; | |
412 | } | |
413 | ||
414 | return 0; | |
415 | } | |
416 | ||
417 | static int decompress_texture_thread(AVCodecContext *avctx, void *arg, | |
6b2b26e7 | 418 | int slice, int thread_nb) |
5c018ee1 VG |
419 | { |
420 | DDSContext *ctx = avctx->priv_data; | |
421 | AVFrame *frame = arg; | |
6b2b26e7 LB |
422 | const uint8_t *d = ctx->tex_data; |
423 | int w_block = avctx->coded_width / TEXTURE_BLOCK_W; | |
ebe8b5d9 | 424 | int h_block = avctx->coded_height / TEXTURE_BLOCK_H; |
6b2b26e7 LB |
425 | int x, y; |
426 | int start_slice, end_slice; | |
ebe8b5d9 TB |
427 | int base_blocks_per_slice = h_block / ctx->slice_count; |
428 | int remainder_blocks = h_block % ctx->slice_count; | |
6b2b26e7 | 429 | |
ebe8b5d9 TB |
430 | /* When the frame height (in blocks) doesn't divide evenly between the |
431 | * number of slices, spread the remaining blocks evenly between the first | |
432 | * operations */ | |
433 | start_slice = slice * base_blocks_per_slice; | |
434 | /* Add any extra blocks (one per slice) that have been added before this slice */ | |
435 | start_slice += FFMIN(slice, remainder_blocks); | |
6b2b26e7 | 436 | |
ebe8b5d9 TB |
437 | end_slice = start_slice + base_blocks_per_slice; |
438 | /* Add an extra block if there are still remainder blocks to be accounted for */ | |
439 | if (slice < remainder_blocks) | |
440 | end_slice++; | |
6b2b26e7 LB |
441 | |
442 | for (y = start_slice; y < end_slice; y++) { | |
443 | uint8_t *p = frame->data[0] + y * frame->linesize[0] * TEXTURE_BLOCK_H; | |
444 | int off = y * w_block; | |
445 | for (x = 0; x < w_block; x++) { | |
446 | ctx->tex_funct(p + x * 16, frame->linesize[0], | |
447 | d + (off + x) * ctx->tex_ratio); | |
448 | } | |
449 | } | |
5c018ee1 | 450 | |
5c018ee1 VG |
451 | return 0; |
452 | } | |
453 | ||
454 | static void do_swizzle(AVFrame *frame, int x, int y) | |
455 | { | |
456 | int i; | |
457 | for (i = 0; i < frame->linesize[0] * frame->height; i += 4) { | |
458 | uint8_t *src = frame->data[0] + i; | |
459 | FFSWAP(uint8_t, src[x], src[y]); | |
460 | } | |
461 | } | |
462 | ||
463 | static void run_postproc(AVCodecContext *avctx, AVFrame *frame) | |
464 | { | |
465 | DDSContext *ctx = avctx->priv_data; | |
466 | int i, x_off; | |
467 | ||
468 | switch (ctx->postproc) { | |
469 | case DDS_ALPHA_EXP: | |
470 | /* Alpha-exponential mode divides each channel by the maximum | |
471 | * R, G or B value, and stores the multiplying factor in the | |
472 | * alpha channel. */ | |
473 | av_log(avctx, AV_LOG_DEBUG, "Post-processing alpha exponent.\n"); | |
474 | ||
475 | for (i = 0; i < frame->linesize[0] * frame->height; i += 4) { | |
476 | uint8_t *src = frame->data[0] + i; | |
477 | int r = src[0]; | |
478 | int g = src[1]; | |
479 | int b = src[2]; | |
480 | int a = src[3]; | |
481 | ||
482 | src[0] = r * a / 255; | |
483 | src[1] = g * a / 255; | |
484 | src[2] = b * a / 255; | |
485 | src[3] = 255; | |
486 | } | |
487 | break; | |
488 | case DDS_NORMAL_MAP: | |
489 | /* Normal maps work in the XYZ color space and they encode | |
490 | * X in R or in A, depending on the texture type, Y in G and | |
491 | * derive Z with a square root of the distance. | |
492 | * | |
493 | * http://www.realtimecollisiondetection.net/blog/?p=28 */ | |
494 | av_log(avctx, AV_LOG_DEBUG, "Post-processing normal map.\n"); | |
495 | ||
496 | x_off = ctx->tex_ratio == 8 ? 0 : 3; | |
497 | for (i = 0; i < frame->linesize[0] * frame->height; i += 4) { | |
498 | uint8_t *src = frame->data[0] + i; | |
499 | int x = src[x_off]; | |
500 | int y = src[1]; | |
501 | int z = 127; | |
502 | ||
503 | int d = (255 * 255 - x * x - y * y) / 2; | |
504 | if (d > 0) | |
505 | z = rint(sqrtf(d)); | |
506 | ||
507 | src[0] = x; | |
508 | src[1] = y; | |
509 | src[2] = z; | |
510 | src[3] = 255; | |
511 | } | |
512 | break; | |
513 | case DDS_RAW_YCOCG: | |
514 | /* Data is Y-Co-Cg-A and not RGBA, but they are represented | |
515 | * with the same masks in the DDPF header. */ | |
516 | av_log(avctx, AV_LOG_DEBUG, "Post-processing raw YCoCg.\n"); | |
517 | ||
518 | for (i = 0; i < frame->linesize[0] * frame->height; i += 4) { | |
519 | uint8_t *src = frame->data[0] + i; | |
520 | int a = src[0]; | |
521 | int cg = src[1] - 128; | |
522 | int co = src[2] - 128; | |
523 | int y = src[3]; | |
524 | ||
525 | src[0] = av_clip_uint8(y + co - cg); | |
526 | src[1] = av_clip_uint8(y + cg); | |
527 | src[2] = av_clip_uint8(y - co - cg); | |
528 | src[3] = a; | |
529 | } | |
530 | break; | |
531 | case DDS_SWAP_ALPHA: | |
532 | /* Alpha and Luma are stored swapped. */ | |
533 | av_log(avctx, AV_LOG_DEBUG, "Post-processing swapped Luma/Alpha.\n"); | |
534 | ||
535 | for (i = 0; i < frame->linesize[0] * frame->height; i += 2) { | |
536 | uint8_t *src = frame->data[0] + i; | |
537 | FFSWAP(uint8_t, src[0], src[1]); | |
538 | } | |
539 | break; | |
540 | case DDS_SWIZZLE_A2XY: | |
541 | /* Swap R and G, often used to restore a standard RGTC2. */ | |
542 | av_log(avctx, AV_LOG_DEBUG, "Post-processing A2XY swizzle.\n"); | |
543 | do_swizzle(frame, 0, 1); | |
544 | break; | |
545 | case DDS_SWIZZLE_RBXG: | |
546 | /* Swap G and A, then B and new A (G). */ | |
547 | av_log(avctx, AV_LOG_DEBUG, "Post-processing RBXG swizzle.\n"); | |
548 | do_swizzle(frame, 1, 3); | |
549 | do_swizzle(frame, 2, 3); | |
550 | break; | |
551 | case DDS_SWIZZLE_RGXB: | |
552 | /* Swap B and A. */ | |
553 | av_log(avctx, AV_LOG_DEBUG, "Post-processing RGXB swizzle.\n"); | |
554 | do_swizzle(frame, 2, 3); | |
555 | break; | |
556 | case DDS_SWIZZLE_RXBG: | |
557 | /* Swap G and A. */ | |
558 | av_log(avctx, AV_LOG_DEBUG, "Post-processing RXBG swizzle.\n"); | |
559 | do_swizzle(frame, 1, 3); | |
560 | break; | |
561 | case DDS_SWIZZLE_RXGB: | |
562 | /* Swap R and A (misleading name). */ | |
563 | av_log(avctx, AV_LOG_DEBUG, "Post-processing RXGB swizzle.\n"); | |
564 | do_swizzle(frame, 0, 3); | |
565 | break; | |
566 | case DDS_SWIZZLE_XGBR: | |
567 | /* Swap B and A, then R and new A (B). */ | |
568 | av_log(avctx, AV_LOG_DEBUG, "Post-processing XGBR swizzle.\n"); | |
569 | do_swizzle(frame, 2, 3); | |
570 | do_swizzle(frame, 0, 3); | |
571 | break; | |
572 | case DDS_SWIZZLE_XGXR: | |
573 | /* Swap G and A, then R and new A (G), then new R (G) and new G (A). | |
574 | * This variant does not store any B component. */ | |
575 | av_log(avctx, AV_LOG_DEBUG, "Post-processing XGXR swizzle.\n"); | |
576 | do_swizzle(frame, 1, 3); | |
577 | do_swizzle(frame, 0, 3); | |
578 | do_swizzle(frame, 0, 1); | |
579 | break; | |
580 | case DDS_SWIZZLE_XRBG: | |
581 | /* Swap G and A, then R and new A (G). */ | |
582 | av_log(avctx, AV_LOG_DEBUG, "Post-processing XRBG swizzle.\n"); | |
583 | do_swizzle(frame, 1, 3); | |
584 | do_swizzle(frame, 0, 3); | |
585 | break; | |
586 | } | |
587 | } | |
588 | ||
589 | static int dds_decode(AVCodecContext *avctx, void *data, | |
590 | int *got_frame, AVPacket *avpkt) | |
591 | { | |
592 | DDSContext *ctx = avctx->priv_data; | |
593 | GetByteContext *gbc = &ctx->gbc; | |
594 | AVFrame *frame = data; | |
6b2b26e7 | 595 | int mipmap; |
5c018ee1 VG |
596 | int ret; |
597 | ||
598 | ff_texturedsp_init(&ctx->texdsp); | |
599 | bytestream2_init(gbc, avpkt->data, avpkt->size); | |
600 | ||
601 | if (bytestream2_get_bytes_left(gbc) < 128) { | |
7b2211bf | 602 | av_log(avctx, AV_LOG_ERROR, "Frame is too small (%d).\n", |
5c018ee1 VG |
603 | bytestream2_get_bytes_left(gbc)); |
604 | return AVERROR_INVALIDDATA; | |
605 | } | |
606 | ||
607 | if (bytestream2_get_le32(gbc) != MKTAG('D', 'D', 'S', ' ') || | |
608 | bytestream2_get_le32(gbc) != 124) { // header size | |
7b2211bf | 609 | av_log(avctx, AV_LOG_ERROR, "Invalid DDS header.\n"); |
5c018ee1 VG |
610 | return AVERROR_INVALIDDATA; |
611 | } | |
612 | ||
613 | bytestream2_skip(gbc, 4); // flags | |
614 | ||
615 | avctx->height = bytestream2_get_le32(gbc); | |
616 | avctx->width = bytestream2_get_le32(gbc); | |
617 | ret = av_image_check_size(avctx->width, avctx->height, 0, avctx); | |
618 | if (ret < 0) { | |
619 | av_log(avctx, AV_LOG_ERROR, "Invalid image size %dx%d.\n", | |
620 | avctx->width, avctx->height); | |
621 | return ret; | |
622 | } | |
623 | ||
624 | /* Since codec is based on 4x4 blocks, size is aligned to 4. */ | |
625 | avctx->coded_width = FFALIGN(avctx->width, TEXTURE_BLOCK_W); | |
626 | avctx->coded_height = FFALIGN(avctx->height, TEXTURE_BLOCK_H); | |
627 | ||
628 | bytestream2_skip(gbc, 4); // pitch | |
629 | bytestream2_skip(gbc, 4); // depth | |
630 | mipmap = bytestream2_get_le32(gbc); | |
631 | if (mipmap != 0) | |
632 | av_log(avctx, AV_LOG_VERBOSE, "Found %d mipmaps (ignored).\n", mipmap); | |
633 | ||
634 | /* Extract pixel format information, considering additional elements | |
635 | * in reserved1 and reserved2. */ | |
636 | ret = parse_pixel_format(avctx); | |
637 | if (ret < 0) | |
638 | return ret; | |
639 | ||
640 | ret = ff_get_buffer(avctx, frame, 0); | |
641 | if (ret < 0) | |
642 | return ret; | |
643 | ||
644 | if (ctx->compressed) { | |
29b17528 AC |
645 | int size = (avctx->coded_height / TEXTURE_BLOCK_H) * |
646 | (avctx->coded_width / TEXTURE_BLOCK_W) * ctx->tex_ratio; | |
ebe8b5d9 TB |
647 | ctx->slice_count = av_clip(avctx->thread_count, 1, |
648 | avctx->coded_height / TEXTURE_BLOCK_H); | |
6b2b26e7 | 649 | |
29b17528 AC |
650 | if (bytestream2_get_bytes_left(gbc) < size) { |
651 | av_log(avctx, AV_LOG_ERROR, | |
652 | "Compressed Buffer is too small (%d < %d).\n", | |
653 | bytestream2_get_bytes_left(gbc), size); | |
654 | return AVERROR_INVALIDDATA; | |
655 | } | |
656 | ||
5c018ee1 VG |
657 | /* Use the decompress function on the texture, one block per thread. */ |
658 | ctx->tex_data = gbc->buffer; | |
ebe8b5d9 | 659 | avctx->execute2(avctx, decompress_texture_thread, frame, NULL, ctx->slice_count); |
5c018ee1 VG |
660 | } else { |
661 | int linesize = av_image_get_linesize(avctx->pix_fmt, frame->width, 0); | |
662 | ||
663 | if (ctx->paletted) { | |
4d554843 | 664 | int i; |
44f7df0c | 665 | uint32_t *p = (uint32_t*) frame->data[1]; |
4d554843 | 666 | |
5c018ee1 | 667 | /* Use the first 1024 bytes as palette, then copy the rest. */ |
57214b2f | 668 | for (i = 0; i < 256; i++) { |
44f7df0c MS |
669 | uint32_t rgba = 0; |
670 | rgba |= bytestream2_get_byte(gbc) << 16; | |
671 | rgba |= bytestream2_get_byte(gbc) << 8; | |
672 | rgba |= bytestream2_get_byte(gbc) << 0; | |
673 | rgba |= bytestream2_get_byte(gbc) << 24; | |
674 | p[i] = rgba; | |
57214b2f | 675 | } |
4d554843 | 676 | |
5c018ee1 VG |
677 | frame->palette_has_changed = 1; |
678 | } | |
679 | ||
e6459c65 AC |
680 | if (bytestream2_get_bytes_left(gbc) < frame->height * linesize) { |
681 | av_log(avctx, AV_LOG_ERROR, "Buffer is too small (%d < %d).\n", | |
682 | bytestream2_get_bytes_left(gbc), frame->height * linesize); | |
683 | return AVERROR_INVALIDDATA; | |
684 | } | |
685 | ||
5c018ee1 VG |
686 | av_image_copy_plane(frame->data[0], frame->linesize[0], |
687 | gbc->buffer, linesize, | |
688 | linesize, frame->height); | |
689 | } | |
690 | ||
691 | /* Run any post processing here if needed. */ | |
d08d8b61 MN |
692 | if (avctx->pix_fmt == AV_PIX_FMT_BGRA || |
693 | avctx->pix_fmt == AV_PIX_FMT_RGBA || | |
694 | avctx->pix_fmt == AV_PIX_FMT_YA8) | |
5c018ee1 VG |
695 | run_postproc(avctx, frame); |
696 | ||
697 | /* Frame is ready to be output. */ | |
698 | frame->pict_type = AV_PICTURE_TYPE_I; | |
699 | frame->key_frame = 1; | |
700 | *got_frame = 1; | |
701 | ||
702 | return avpkt->size; | |
703 | } | |
704 | ||
705 | AVCodec ff_dds_decoder = { | |
706 | .name = "dds", | |
707 | .long_name = NULL_IF_CONFIG_SMALL("DirectDraw Surface image decoder"), | |
708 | .type = AVMEDIA_TYPE_VIDEO, | |
709 | .id = AV_CODEC_ID_DDS, | |
710 | .decode = dds_decode, | |
711 | .priv_data_size = sizeof(DDSContext), | |
def97856 | 712 | .capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_SLICE_THREADS, |
5c018ee1 VG |
713 | .caps_internal = FF_CODEC_CAP_INIT_THREADSAFE |
714 | }; |