h261 decoder by (Maarten Daniels <maarten.daniels at student dot luc dot ac dot be>)
[libav.git] / libavcodec / dsputil.h
1 /*
2 * DSP utils
3 * Copyright (c) 2000, 2001, 2002 Fabrice Bellard.
4 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
5 *
6 * This library 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 of the License, or (at your option) any later version.
10 *
11 * This library 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 this library; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
19 */
20
21 /**
22 * @file dsputil.h
23 * DSP utils.
24 * note, many functions in here may use MMX which trashes the FPU state, it is
25 * absolutely necessary to call emms_c() between dsp & float/double code
26 */
27
28 #ifndef DSPUTIL_H
29 #define DSPUTIL_H
30
31 #include "common.h"
32 #include "avcodec.h"
33
34
35 //#define DEBUG
36 /* dct code */
37 typedef short DCTELEM;
38
39 void fdct_ifast (DCTELEM *data);
40 void fdct_ifast248 (DCTELEM *data);
41 void ff_jpeg_fdct_islow (DCTELEM *data);
42 void ff_fdct248_islow (DCTELEM *data);
43
44 void j_rev_dct (DCTELEM *data);
45
46 void ff_fdct_mmx(DCTELEM *block);
47 void ff_fdct_mmx2(DCTELEM *block);
48 void ff_fdct_sse2(DCTELEM *block);
49
50 /* encoding scans */
51 extern const uint8_t ff_alternate_horizontal_scan[64];
52 extern const uint8_t ff_alternate_vertical_scan[64];
53 extern const uint8_t ff_zigzag_direct[64];
54 extern const uint8_t ff_zigzag248_direct[64];
55
56 /* pixel operations */
57 #define MAX_NEG_CROP 384
58
59 /* temporary */
60 extern uint32_t squareTbl[512];
61 extern uint8_t cropTbl[256 + 2 * MAX_NEG_CROP];
62
63 /* VP3 DSP functions */
64 void vp3_dsp_init_c(void);
65 void vp3_idct_c(int16_t *input_data, int16_t *dequant_matrix,
66 int coeff_count, DCTELEM *output_data);
67
68 void vp3_dsp_init_mmx(void);
69 void vp3_idct_mmx(int16_t *input_data, int16_t *dequant_matrix,
70 int coeff_count, DCTELEM *output_data);
71
72 void vp3_dsp_init_sse2(void);
73 void vp3_idct_sse2(int16_t *input_data, int16_t *dequant_matrix,
74 int coeff_count, DCTELEM *output_data);
75
76 /* minimum alignment rules ;)
77 if u notice errors in the align stuff, need more alignment for some asm code for some cpu
78 or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ...
79
80 !warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible)
81 i (michael) didnt check them, these are just the alignents which i think could be reached easily ...
82
83 !future video codecs might need functions with less strict alignment
84 */
85
86 /*
87 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
88 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
89 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
90 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
91 void clear_blocks_c(DCTELEM *blocks);
92 */
93
94 /* add and put pixel (decoding) */
95 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
96 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
97 typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
98 typedef void (*tpel_mc_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int w, int h);
99 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
100 typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
101
102 #define DEF_OLD_QPEL(name)\
103 void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
104 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
105 void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
106
107 DEF_OLD_QPEL(qpel16_mc11_old_c)
108 DEF_OLD_QPEL(qpel16_mc31_old_c)
109 DEF_OLD_QPEL(qpel16_mc12_old_c)
110 DEF_OLD_QPEL(qpel16_mc32_old_c)
111 DEF_OLD_QPEL(qpel16_mc13_old_c)
112 DEF_OLD_QPEL(qpel16_mc33_old_c)
113 DEF_OLD_QPEL(qpel8_mc11_old_c)
114 DEF_OLD_QPEL(qpel8_mc31_old_c)
115 DEF_OLD_QPEL(qpel8_mc12_old_c)
116 DEF_OLD_QPEL(qpel8_mc32_old_c)
117 DEF_OLD_QPEL(qpel8_mc13_old_c)
118 DEF_OLD_QPEL(qpel8_mc33_old_c)
119
120 #define CALL_2X_PIXELS(a, b, n)\
121 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
122 b(block , pixels , line_size, h);\
123 b(block+n, pixels+n, line_size, h);\
124 }
125
126 /* motion estimation */
127 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
128 // allthough currently h<4 is not used as functions with width <8 are not used and neither implemented
129 typedef int (*me_cmp_func)(void /*MpegEncContext*/ *s, uint8_t *blk1/*align width (8 or 16)*/, uint8_t *blk2/*align 1*/, int line_size, int h)/* __attribute__ ((const))*/;
130
131
132 /**
133 * DSPContext.
134 */
135 typedef struct DSPContext {
136 /* pixel ops : interface with DCT */
137 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
138 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
139 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
140 void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
141 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
142 /**
143 * translational global motion compensation.
144 */
145 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
146 /**
147 * global motion compensation.
148 */
149 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
150 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
151 void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
152 int (*pix_sum)(uint8_t * pix, int line_size);
153 int (*pix_norm1)(uint8_t * pix, int line_size);
154 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
155
156 me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */
157 me_cmp_func sse[5];
158 me_cmp_func hadamard8_diff[5];
159 me_cmp_func dct_sad[5];
160 me_cmp_func quant_psnr[5];
161 me_cmp_func bit[5];
162 me_cmp_func rd[5];
163 me_cmp_func vsad[5];
164 me_cmp_func vsse[5];
165
166 me_cmp_func me_pre_cmp[5];
167 me_cmp_func me_cmp[5];
168 me_cmp_func me_sub_cmp[5];
169 me_cmp_func mb_cmp[5];
170 me_cmp_func ildct_cmp[5]; //only width 16 used
171
172 /**
173 * Halfpel motion compensation with rounding (a+b+1)>>1.
174 * this is an array[4][4] of motion compensation funcions for 4
175 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
176 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
177 * @param block destination where the result is stored
178 * @param pixels source
179 * @param line_size number of bytes in a horizontal line of block
180 * @param h height
181 */
182 op_pixels_func put_pixels_tab[4][4];
183
184 /**
185 * Halfpel motion compensation with rounding (a+b+1)>>1.
186 * This is an array[4][4] of motion compensation functions for 4
187 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
188 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
189 * @param block destination into which the result is averaged (a+b+1)>>1
190 * @param pixels source
191 * @param line_size number of bytes in a horizontal line of block
192 * @param h height
193 */
194 op_pixels_func avg_pixels_tab[4][4];
195
196 /**
197 * Halfpel motion compensation with no rounding (a+b)>>1.
198 * this is an array[2][4] of motion compensation funcions for 2
199 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
200 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
201 * @param block destination where the result is stored
202 * @param pixels source
203 * @param line_size number of bytes in a horizontal line of block
204 * @param h height
205 */
206 op_pixels_func put_no_rnd_pixels_tab[2][4];
207
208 /**
209 * Halfpel motion compensation with no rounding (a+b)>>1.
210 * this is an array[2][4] of motion compensation funcions for 2
211 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
212 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
213 * @param block destination into which the result is averaged (a+b)>>1
214 * @param pixels source
215 * @param line_size number of bytes in a horizontal line of block
216 * @param h height
217 */
218 op_pixels_func avg_no_rnd_pixels_tab[2][4];
219
220 void (*put_no_rnd_pixels_l2[2])(uint8_t *block/*align width (8 or 16)*/, const uint8_t *a/*align 1*/, const uint8_t *b/*align 1*/, int line_size, int h);
221
222 /**
223 * Thirdpel motion compensation with rounding (a+b+1)>>1.
224 * this is an array[12] of motion compensation funcions for the 9 thirdpel positions<br>
225 * *pixels_tab[ xthirdpel + 4*ythirdpel ]
226 * @param block destination where the result is stored
227 * @param pixels source
228 * @param line_size number of bytes in a horizontal line of block
229 * @param h height
230 */
231 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
232 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
233
234 qpel_mc_func put_qpel_pixels_tab[2][16];
235 qpel_mc_func avg_qpel_pixels_tab[2][16];
236 qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
237 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
238 qpel_mc_func put_mspel_pixels_tab[8];
239
240 /**
241 * h264 Chram MC
242 */
243 h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
244 h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
245
246 qpel_mc_func put_h264_qpel_pixels_tab[3][16];
247 qpel_mc_func avg_h264_qpel_pixels_tab[3][16];
248
249 me_cmp_func pix_abs[2][4];
250
251 /* huffyuv specific */
252 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
253 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
254 /**
255 * subtract huffyuv's variant of median prediction
256 * note, this might read from src1[-1], src2[-1]
257 */
258 void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
259 void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w);
260
261 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
262 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
263
264 void (*h261_v_loop_filter)(uint8_t *dest,uint8_t *src, int stride);
265 void (*h261_h_loop_filter)(uint8_t *dest,uint8_t *src, int stride);
266
267 /* (I)DCT */
268 void (*fdct)(DCTELEM *block/* align 16*/);
269 void (*fdct248)(DCTELEM *block/* align 16*/);
270
271 /* IDCT really*/
272 void (*idct)(DCTELEM *block/* align 16*/);
273
274 /**
275 * block -> idct -> clip to unsigned 8 bit -> dest.
276 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
277 * @param line_size size in bytes of a horizotal line of dest
278 */
279 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
280
281 /**
282 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
283 * @param line_size size in bytes of a horizotal line of dest
284 */
285 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
286
287 /**
288 * idct input permutation.
289 * several optimized IDCTs need a permutated input (relative to the normal order of the reference
290 * IDCT)
291 * this permutation must be performed before the idct_put/add, note, normally this can be merged
292 * with the zigzag/alternate scan<br>
293 * an example to avoid confusion:
294 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
295 * - (x -> referece dct -> reference idct -> x)
296 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
297 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
298 */
299 uint8_t idct_permutation[64];
300 int idct_permutation_type;
301 #define FF_NO_IDCT_PERM 1
302 #define FF_LIBMPEG2_IDCT_PERM 2
303 #define FF_SIMPLE_IDCT_PERM 3
304 #define FF_TRANSPOSE_IDCT_PERM 4
305
306 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
307 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
308 #define BASIS_SHIFT 16
309 #define RECON_SHIFT 6
310
311 /**
312 * This function handles any initialization for the VP3 DSP functions.
313 */
314 void (*vp3_dsp_init)(void);
315
316 /**
317 * This function is responsible for taking a block of zigzag'd,
318 * quantized DCT coefficients and reconstructing the original block of
319 * samples.
320 * @param input_data 64 zigzag'd, quantized DCT coefficients
321 * @param dequant_matrix 64 zigzag'd quantizer coefficients
322 * @param coeff_count index of the last coefficient
323 * @param output_samples space for 64 DCTELEMs where the transformed
324 * samples will be stored
325 */
326 void (*vp3_idct)(int16_t *input_data, int16_t *dequant_matrix,
327 int coeff_count, DCTELEM *output_samples);
328
329 } DSPContext;
330
331 void dsputil_static_init(void);
332 void dsputil_init(DSPContext* p, AVCodecContext *avctx);
333
334 /**
335 * permute block according to permuatation.
336 * @param last last non zero element in scantable order
337 */
338 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
339
340 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
341
342 #define BYTE_VEC32(c) ((c)*0x01010101UL)
343
344 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
345 {
346 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
347 }
348
349 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
350 {
351 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
352 }
353
354 /**
355 * Empty mmx state.
356 * this must be called between any dsp function and float/double code.
357 * for example sin(); dsp->idct_put(); emms_c(); cos()
358 */
359 #define emms_c()
360
361 /* should be defined by architectures supporting
362 one or more MultiMedia extension */
363 int mm_support(void);
364
365 #define __align16 __attribute__ ((aligned (16)))
366
367 #if defined(HAVE_MMX)
368
369 #undef emms_c
370
371 #define MM_MMX 0x0001 /* standard MMX */
372 #define MM_3DNOW 0x0004 /* AMD 3DNOW */
373 #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
374 #define MM_SSE 0x0008 /* SSE functions */
375 #define MM_SSE2 0x0010 /* PIV SSE2 functions */
376
377 extern int mm_flags;
378
379 void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
380 void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
381 void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
382
383 static inline void emms(void)
384 {
385 __asm __volatile ("emms;":::"memory");
386 }
387
388
389 #define emms_c() \
390 {\
391 if (mm_flags & MM_MMX)\
392 emms();\
393 }
394
395 #define __align8 __attribute__ ((aligned (8)))
396
397 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
398 void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
399
400 #elif defined(ARCH_ARMV4L)
401
402 /* This is to use 4 bytes read to the IDCT pointers for some 'zero'
403 line optimizations */
404 #define __align8 __attribute__ ((aligned (4)))
405
406 void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
407
408 #elif defined(HAVE_MLIB)
409
410 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
411 #define __align8 __attribute__ ((aligned (8)))
412
413 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
414
415 #elif defined(ARCH_SPARC)
416
417 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
418 #define __align8 __attribute__ ((aligned (8)))
419 void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
420
421 #elif defined(ARCH_ALPHA)
422
423 #define __align8 __attribute__ ((aligned (8)))
424
425 void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
426
427 #elif defined(ARCH_POWERPC)
428
429 #define MM_ALTIVEC 0x0001 /* standard AltiVec */
430
431 extern int mm_flags;
432
433 #if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN)
434 #define pixel altivec_pixel
435 #include <altivec.h>
436 #undef pixel
437 #endif
438
439 #define __align8 __attribute__ ((aligned (16)))
440
441 void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
442
443 #elif defined(HAVE_MMI)
444
445 #define __align8 __attribute__ ((aligned (16)))
446
447 void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
448
449 #elif defined(ARCH_SH4)
450
451 #define __align8 __attribute__ ((aligned (8)))
452
453 void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
454
455 #else
456
457 #define __align8
458
459 #endif
460
461 #ifdef __GNUC__
462
463 struct unaligned_64 { uint64_t l; } __attribute__((packed));
464 struct unaligned_32 { uint32_t l; } __attribute__((packed));
465 struct unaligned_16 { uint16_t l; } __attribute__((packed));
466
467 #define LD16(a) (((const struct unaligned_16 *) (a))->l)
468 #define LD32(a) (((const struct unaligned_32 *) (a))->l)
469 #define LD64(a) (((const struct unaligned_64 *) (a))->l)
470
471 #define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)
472
473 #else /* __GNUC__ */
474
475 #define LD16(a) (*((uint16_t*)(a)))
476 #define LD32(a) (*((uint32_t*)(a)))
477 #define LD64(a) (*((uint64_t*)(a)))
478
479 #define ST32(a, b) *((uint32_t*)(a)) = (b)
480
481 #endif /* !__GNUC__ */
482
483 /* PSNR */
484 void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
485 int orig_linesize[3], int coded_linesize,
486 AVCodecContext *avctx);
487
488 /* FFT computation */
489
490 /* NOTE: soon integer code will be added, so you must use the
491 FFTSample type */
492 typedef float FFTSample;
493
494 typedef struct FFTComplex {
495 FFTSample re, im;
496 } FFTComplex;
497
498 typedef struct FFTContext {
499 int nbits;
500 int inverse;
501 uint16_t *revtab;
502 FFTComplex *exptab;
503 FFTComplex *exptab1; /* only used by SSE code */
504 void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
505 } FFTContext;
506
507 int ff_fft_init(FFTContext *s, int nbits, int inverse);
508 void ff_fft_permute(FFTContext *s, FFTComplex *z);
509 void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
510 void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
511 void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
512
513 static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
514 {
515 s->fft_calc(s, z);
516 }
517 void ff_fft_end(FFTContext *s);
518
519 /* MDCT computation */
520
521 typedef struct MDCTContext {
522 int n; /* size of MDCT (i.e. number of input data * 2) */
523 int nbits; /* n = 2^nbits */
524 /* pre/post rotation tables */
525 FFTSample *tcos;
526 FFTSample *tsin;
527 FFTContext fft;
528 } MDCTContext;
529
530 int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
531 void ff_imdct_calc(MDCTContext *s, FFTSample *output,
532 const FFTSample *input, FFTSample *tmp);
533 void ff_mdct_calc(MDCTContext *s, FFTSample *out,
534 const FFTSample *input, FFTSample *tmp);
535 void ff_mdct_end(MDCTContext *s);
536
537 #define WARPER8_16(name8, name16)\
538 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
539 return name8(s, dst , src , stride, h)\
540 +name8(s, dst+8 , src+8 , stride, h);\
541 }
542
543 #define WARPER8_16_SQ(name8, name16)\
544 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
545 int score=0;\
546 score +=name8(s, dst , src , stride, 8);\
547 score +=name8(s, dst+8 , src+8 , stride, 8);\
548 if(h==16){\
549 dst += 8*stride;\
550 src += 8*stride;\
551 score +=name8(s, dst , src , stride, 8);\
552 score +=name8(s, dst+8 , src+8 , stride, 8);\
553 }\
554 return score;\
555 }
556
557 #ifndef HAVE_LRINTF
558 /* XXX: add ISOC specific test to avoid specific BSD testing. */
559 /* better than nothing implementation. */
560 /* btw, rintf() is existing on fbsd too -- alex */
561 static inline long int lrintf(float x)
562 {
563 #ifdef CONFIG_WIN32
564 /* XXX: incorrect, but make it compile */
565 return (int)(x);
566 #else
567 return (int)(rint(x));
568 #endif
569 }
570 #else
571 #ifndef _ISOC9X_SOURCE
572 #define _ISOC9X_SOURCE
573 #endif
574 #include <math.h>
575 #endif
576
577 #endif