Commit | Line | Data |
---|---|---|
ff4ec49e FB |
1 | /* |
2 | * DSP utils | |
3 | * Copyright (c) 2000, 2001, 2002 Fabrice Bellard. | |
4 | * | |
5 | * This library is free software; you can redistribute it and/or | |
6 | * modify it under the terms of the GNU Lesser General Public | |
7 | * License as published by the Free Software Foundation; either | |
8 | * version 2 of the License, or (at your option) any later version. | |
9 | * | |
10 | * This library is distributed in the hope that it will be useful, | |
11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
13 | * Lesser General Public License for more details. | |
14 | * | |
15 | * You should have received a copy of the GNU Lesser General Public | |
16 | * License along with this library; if not, write to the Free Software | |
17 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
18 | */ | |
24641185 MN |
19 | |
20 | /** | |
21 | * @file dsputil.h | |
22 | * @brief DSP utils | |
23 | * | |
24 | */ | |
25 | ||
de6d9b64 FB |
26 | #ifndef DSPUTIL_H |
27 | #define DSPUTIL_H | |
28 | ||
29 | #include "common.h" | |
43f1708f | 30 | #include "avcodec.h" |
de6d9b64 | 31 | |
24641185 | 32 | |
44eb4951 | 33 | //#define DEBUG |
de6d9b64 FB |
34 | /* dct code */ |
35 | typedef short DCTELEM; | |
0e15384d | 36 | //typedef int DCTELEM; |
de6d9b64 | 37 | |
03c94ede | 38 | void fdct_ifast (DCTELEM *data); |
28db7fce | 39 | void ff_jpeg_fdct_islow (DCTELEM *data); |
de6d9b64 FB |
40 | |
41 | void j_rev_dct (DCTELEM *data); | |
42 | ||
3f09f52a | 43 | void ff_fdct_mmx(DCTELEM *block); |
de6d9b64 | 44 | |
e0eac44e | 45 | /* encoding scans */ |
0c1a9eda ZK |
46 | extern const uint8_t ff_alternate_horizontal_scan[64]; |
47 | extern const uint8_t ff_alternate_vertical_scan[64]; | |
48 | extern const uint8_t ff_zigzag_direct[64]; | |
5a240838 | 49 | |
de6d9b64 FB |
50 | /* pixel operations */ |
51 | #define MAX_NEG_CROP 384 | |
52 | ||
53 | /* temporary */ | |
0c1a9eda ZK |
54 | extern uint32_t squareTbl[512]; |
55 | extern uint8_t cropTbl[256 + 2 * MAX_NEG_CROP]; | |
de6d9b64 | 56 | |
de6d9b64 | 57 | |
b7c27ee6 | 58 | /* minimum alignment rules ;) |
eb4b3dd3 | 59 | if u notice errors in the align stuff, need more alignment for some asm code for some cpu |
b7c27ee6 MN |
60 | or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ... |
61 | ||
62 | !warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible) | |
63 | i (michael) didnt check them, these are just the alignents which i think could be reached easily ... | |
de6d9b64 | 64 | |
b7c27ee6 MN |
65 | !future video codecs might need functions with less strict alignment |
66 | */ | |
67 | ||
eb4b3dd3 | 68 | /* |
0c1a9eda ZK |
69 | void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size); |
70 | void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride); | |
71 | void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size); | |
72 | void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size); | |
649c00c9 | 73 | void clear_blocks_c(DCTELEM *blocks); |
eb4b3dd3 | 74 | */ |
de6d9b64 FB |
75 | |
76 | /* add and put pixel (decoding) */ | |
b7c27ee6 | 77 | // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16 |
0c1a9eda ZK |
78 | typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h); |
79 | typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride); | |
b3184779 | 80 | |
db794953 | 81 | #define DEF_OLD_QPEL(name)\ |
0c1a9eda ZK |
82 | void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ |
83 | void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ | |
84 | void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride); | |
db794953 MN |
85 | |
86 | DEF_OLD_QPEL(qpel16_mc11_old_c) | |
87 | DEF_OLD_QPEL(qpel16_mc31_old_c) | |
88 | DEF_OLD_QPEL(qpel16_mc12_old_c) | |
89 | DEF_OLD_QPEL(qpel16_mc32_old_c) | |
90 | DEF_OLD_QPEL(qpel16_mc13_old_c) | |
91 | DEF_OLD_QPEL(qpel16_mc33_old_c) | |
92 | DEF_OLD_QPEL(qpel8_mc11_old_c) | |
93 | DEF_OLD_QPEL(qpel8_mc31_old_c) | |
94 | DEF_OLD_QPEL(qpel8_mc12_old_c) | |
95 | DEF_OLD_QPEL(qpel8_mc32_old_c) | |
96 | DEF_OLD_QPEL(qpel8_mc13_old_c) | |
97 | DEF_OLD_QPEL(qpel8_mc33_old_c) | |
b3184779 MN |
98 | |
99 | #define CALL_2X_PIXELS(a, b, n)\ | |
100 | static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ | |
101 | b(block , pixels , line_size, h);\ | |
102 | b(block+n, pixels+n, line_size, h);\ | |
103 | } | |
44eb4951 | 104 | |
de6d9b64 FB |
105 | /* motion estimation */ |
106 | ||
0c1a9eda | 107 | typedef int (*op_pixels_abs_func)(uint8_t *blk1/*align width (8 or 16)*/, uint8_t *blk2/*align 1*/, int line_size)/* __attribute__ ((const))*/; |
1457ab52 | 108 | |
0c1a9eda | 109 | typedef int (*me_cmp_func)(void /*MpegEncContext*/ *s, uint8_t *blk1/*align width (8 or 16)*/, uint8_t *blk2/*align 1*/, int line_size)/* __attribute__ ((const))*/; |
1457ab52 | 110 | |
24641185 MN |
111 | /** |
112 | * DSPContext. | |
113 | */ | |
eb4b3dd3 ZK |
114 | typedef struct DSPContext { |
115 | /* pixel ops : interface with DCT */ | |
0c1a9eda ZK |
116 | void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size); |
117 | void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride); | |
118 | void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); | |
119 | void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); | |
120 | void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder); | |
121 | void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy, | |
eb4b3dd3 ZK |
122 | int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height); |
123 | void (*clear_blocks)(DCTELEM *blocks/*align 16*/); | |
0c1a9eda ZK |
124 | int (*pix_sum)(uint8_t * pix, int line_size); |
125 | int (*pix_norm1)(uint8_t * pix, int line_size); | |
1457ab52 MN |
126 | me_cmp_func sad[2]; /* identical to pix_absAxA except additional void * */ |
127 | me_cmp_func sse[2]; | |
128 | me_cmp_func hadamard8_diff[2]; | |
129 | me_cmp_func dct_sad[2]; | |
130 | me_cmp_func quant_psnr[2]; | |
3a87ac94 MN |
131 | me_cmp_func bit[2]; |
132 | me_cmp_func rd[2]; | |
1457ab52 MN |
133 | int (*hadamard8_abs )(uint8_t *src, int stride, int mean); |
134 | ||
826f429a | 135 | me_cmp_func me_pre_cmp[11]; |
1457ab52 MN |
136 | me_cmp_func me_cmp[11]; |
137 | me_cmp_func me_sub_cmp[11]; | |
138 | me_cmp_func mb_cmp[11]; | |
eb4b3dd3 ZK |
139 | |
140 | /* maybe create an array for 16/8 functions */ | |
141 | op_pixels_func put_pixels_tab[2][4]; | |
142 | op_pixels_func avg_pixels_tab[2][4]; | |
143 | op_pixels_func put_no_rnd_pixels_tab[2][4]; | |
144 | op_pixels_func avg_no_rnd_pixels_tab[2][4]; | |
145 | qpel_mc_func put_qpel_pixels_tab[2][16]; | |
146 | qpel_mc_func avg_qpel_pixels_tab[2][16]; | |
147 | qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16]; | |
148 | qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16]; | |
1457ab52 | 149 | qpel_mc_func put_mspel_pixels_tab[8]; |
eb4b3dd3 ZK |
150 | |
151 | op_pixels_abs_func pix_abs16x16; | |
152 | op_pixels_abs_func pix_abs16x16_x2; | |
153 | op_pixels_abs_func pix_abs16x16_y2; | |
154 | op_pixels_abs_func pix_abs16x16_xy2; | |
155 | op_pixels_abs_func pix_abs8x8; | |
156 | op_pixels_abs_func pix_abs8x8_x2; | |
157 | op_pixels_abs_func pix_abs8x8_y2; | |
158 | op_pixels_abs_func pix_abs8x8_xy2; | |
11f18faf MN |
159 | |
160 | /* huffyuv specific */ | |
11f18faf | 161 | void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w); |
1457ab52 | 162 | void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w); |
b0368839 MN |
163 | |
164 | /* (I)DCT */ | |
165 | void (*fdct)(DCTELEM *block/* align 16*/); | |
24641185 MN |
166 | |
167 | /** | |
168 | * block -> idct -> clip to unsigned 8 bit -> dest.<br> | |
169 | * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...) | |
170 | * @param line_size size in pixels of a horizotal line of dest | |
171 | */ | |
b0368839 | 172 | void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
24641185 MN |
173 | |
174 | /** | |
175 | * block -> idct -> add dest -> clip to unsigned 8 bit -> dest. | |
176 | * @param line_size size in pixels of a horizotal line of dest | |
177 | */ | |
b0368839 | 178 | void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
24641185 MN |
179 | |
180 | /** | |
181 | * idct input permutation.<br> | |
182 | * an example to avoid confusion: | |
183 | * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...) | |
184 | * - (x -> referece dct -> reference idct -> x) | |
185 | * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x) | |
186 | * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...) | |
187 | */ | |
b0368839 MN |
188 | uint8_t idct_permutation[64]; |
189 | int idct_permutation_type; | |
190 | #define FF_NO_IDCT_PERM 1 | |
191 | #define FF_LIBMPEG2_IDCT_PERM 2 | |
192 | #define FF_SIMPLE_IDCT_PERM 3 | |
193 | #define FF_TRANSPOSE_IDCT_PERM 4 | |
194 | ||
eb4b3dd3 ZK |
195 | } DSPContext; |
196 | ||
b0368839 | 197 | void dsputil_init(DSPContext* p, AVCodecContext *avctx); |
de6d9b64 | 198 | |
7801d21d MN |
199 | /** |
200 | * permute block according to permuatation. | |
201 | * @param last last non zero element in scantable order | |
202 | */ | |
0c1a9eda | 203 | void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last); |
e0eac44e | 204 | |
24641185 MN |
205 | /** |
206 | * Empty mmx state.<br> | |
207 | * this must be called between any dsp function and float/double code. | |
208 | * for example sin(); dsp->idct_put(); emms_c(); cos() | |
209 | */ | |
eb4b3dd3 ZK |
210 | #define emms_c() |
211 | ||
e629ab68 RD |
212 | /* should be defined by architectures supporting |
213 | one or more MultiMedia extension */ | |
214 | int mm_support(void); | |
215 | ||
3d03c0a2 | 216 | #if defined(HAVE_MMX) |
de6d9b64 | 217 | |
18f77016 | 218 | #undef emms_c |
eb4b3dd3 | 219 | |
de6d9b64 FB |
220 | #define MM_MMX 0x0001 /* standard MMX */ |
221 | #define MM_3DNOW 0x0004 /* AMD 3DNOW */ | |
222 | #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */ | |
223 | #define MM_SSE 0x0008 /* SSE functions */ | |
224 | #define MM_SSE2 0x0010 /* PIV SSE2 functions */ | |
225 | ||
226 | extern int mm_flags; | |
227 | ||
0c1a9eda ZK |
228 | void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size); |
229 | void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size); | |
de6d9b64 FB |
230 | |
231 | static inline void emms(void) | |
232 | { | |
fb16b7e7 FB |
233 | __asm __volatile ("emms;":::"memory"); |
234 | } | |
235 | ||
1457ab52 | 236 | |
fb16b7e7 FB |
237 | #define emms_c() \ |
238 | {\ | |
239 | if (mm_flags & MM_MMX)\ | |
240 | emms();\ | |
de6d9b64 FB |
241 | } |
242 | ||
243 | #define __align8 __attribute__ ((aligned (8))) | |
244 | ||
b0368839 MN |
245 | void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx); |
246 | void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx); | |
2720569a | 247 | |
3d03c0a2 FB |
248 | #elif defined(ARCH_ARMV4L) |
249 | ||
3d03c0a2 FB |
250 | /* This is to use 4 bytes read to the IDCT pointers for some 'zero' |
251 | line ptimizations */ | |
252 | #define __align8 __attribute__ ((aligned (4))) | |
253 | ||
b0368839 | 254 | void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx); |
3d03c0a2 | 255 | |
c34270f5 | 256 | #elif defined(HAVE_MLIB) |
c34270f5 FB |
257 | |
258 | /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */ | |
259 | #define __align8 __attribute__ ((aligned (8))) | |
260 | ||
b0368839 | 261 | void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx); |
c34270f5 | 262 | |
1e98dffb NK |
263 | #elif defined(ARCH_ALPHA) |
264 | ||
1e98dffb NK |
265 | #define __align8 __attribute__ ((aligned (8))) |
266 | ||
b0368839 | 267 | void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx); |
1e98dffb | 268 | |
59925ef2 BF |
269 | #elif defined(ARCH_POWERPC) |
270 | ||
404d2241 BF |
271 | #define MM_ALTIVEC 0x0001 /* standard AltiVec */ |
272 | ||
273 | extern int mm_flags; | |
274 | ||
3b991c54 RD |
275 | #if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN) |
276 | #include <altivec.h> | |
277 | #endif | |
278 | ||
59925ef2 BF |
279 | #define __align8 __attribute__ ((aligned (16))) |
280 | ||
b0368839 | 281 | void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx); |
59925ef2 | 282 | |
d46aba26 LS |
283 | #elif defined(HAVE_MMI) |
284 | ||
d46aba26 LS |
285 | #define __align8 __attribute__ ((aligned (16))) |
286 | ||
b0368839 | 287 | void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx); |
d46aba26 | 288 | |
de6d9b64 FB |
289 | #else |
290 | ||
291 | #define __align8 | |
292 | ||
293 | #endif | |
294 | ||
6d4985bb FB |
295 | #ifdef __GNUC__ |
296 | ||
297 | struct unaligned_64 { uint64_t l; } __attribute__((packed)); | |
298 | struct unaligned_32 { uint32_t l; } __attribute__((packed)); | |
299 | ||
300 | #define LD32(a) (((const struct unaligned_32 *) (a))->l) | |
301 | #define LD64(a) (((const struct unaligned_64 *) (a))->l) | |
302 | ||
303 | #define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b) | |
304 | ||
305 | #else /* __GNUC__ */ | |
306 | ||
307 | #define LD32(a) (*((uint32_t*)(a))) | |
308 | #define LD64(a) (*((uint64_t*)(a))) | |
309 | ||
310 | #define ST32(a, b) *((uint32_t*)(a)) = (b) | |
311 | ||
312 | #endif /* !__GNUC__ */ | |
313 | ||
43f1708f | 314 | /* PSNR */ |
0c1a9eda | 315 | void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3], |
43f1708f J |
316 | int orig_linesize[3], int coded_linesize, |
317 | AVCodecContext *avctx); | |
bb6f5690 FB |
318 | |
319 | /* FFT computation */ | |
320 | ||
321 | /* NOTE: soon integer code will be added, so you must use the | |
322 | FFTSample type */ | |
323 | typedef float FFTSample; | |
324 | ||
325 | typedef struct FFTComplex { | |
326 | FFTSample re, im; | |
327 | } FFTComplex; | |
328 | ||
329 | typedef struct FFTContext { | |
330 | int nbits; | |
331 | int inverse; | |
332 | uint16_t *revtab; | |
333 | FFTComplex *exptab; | |
334 | FFTComplex *exptab1; /* only used by SSE code */ | |
335 | void (*fft_calc)(struct FFTContext *s, FFTComplex *z); | |
336 | } FFTContext; | |
337 | ||
338 | int fft_init(FFTContext *s, int nbits, int inverse); | |
339 | void fft_permute(FFTContext *s, FFTComplex *z); | |
340 | void fft_calc_c(FFTContext *s, FFTComplex *z); | |
341 | void fft_calc_sse(FFTContext *s, FFTComplex *z); | |
8d268a7d FB |
342 | void fft_calc_altivec(FFTContext *s, FFTComplex *z); |
343 | ||
bb6f5690 FB |
344 | static inline void fft_calc(FFTContext *s, FFTComplex *z) |
345 | { | |
346 | s->fft_calc(s, z); | |
347 | } | |
348 | void fft_end(FFTContext *s); | |
349 | ||
350 | /* MDCT computation */ | |
351 | ||
352 | typedef struct MDCTContext { | |
353 | int n; /* size of MDCT (i.e. number of input data * 2) */ | |
354 | int nbits; /* n = 2^nbits */ | |
355 | /* pre/post rotation tables */ | |
356 | FFTSample *tcos; | |
357 | FFTSample *tsin; | |
358 | FFTContext fft; | |
359 | } MDCTContext; | |
360 | ||
82696bee | 361 | int ff_mdct_init(MDCTContext *s, int nbits, int inverse); |
eb4b3dd3 | 362 | void ff_imdct_calc(MDCTContext *s, FFTSample *output, |
bb6f5690 | 363 | const FFTSample *input, FFTSample *tmp); |
eb4b3dd3 | 364 | void ff_mdct_calc(MDCTContext *s, FFTSample *out, |
bb6f5690 | 365 | const FFTSample *input, FFTSample *tmp); |
82696bee | 366 | void ff_mdct_end(MDCTContext *s); |
bb6f5690 | 367 | |
1457ab52 MN |
368 | #define WARPER88_1616(name8, name16)\ |
369 | static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride){\ | |
370 | return name8(s, dst , src , stride)\ | |
371 | +name8(s, dst+8 , src+8 , stride)\ | |
372 | +name8(s, dst +8*stride, src +8*stride, stride)\ | |
373 | +name8(s, dst+8+8*stride, src+8+8*stride, stride);\ | |
374 | } | |
375 | ||
95e2ce4a | 376 | #ifndef HAVE_LRINTF |
9d85cbd9 FB |
377 | /* XXX: add ISOC specific test to avoid specific BSD testing. */ |
378 | /* better than nothing implementation. */ | |
6234d753 | 379 | /* btw, rintf() is existing on fbsd too -- alex */ |
9d85cbd9 FB |
380 | static inline long int lrintf(float x) |
381 | { | |
ea937d01 FB |
382 | #ifdef CONFIG_WIN32 |
383 | /* XXX: incorrect, but make it compile */ | |
384 | return (int)(x); | |
385 | #else | |
9d85cbd9 | 386 | return (int)(rint(x)); |
ea937d01 | 387 | #endif |
9d85cbd9 FB |
388 | } |
389 | #endif | |
390 | ||
de6d9b64 | 391 | #endif |