0307dbd6ab7a14270e46366d42040f0049678733
[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 (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
141 /**
142 * translational global motion compensation.
143 */
144 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
145 /**
146 * global motion compensation.
147 */
148 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
149 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
150 void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
151 int (*pix_sum)(uint8_t * pix, int line_size);
152 int (*pix_norm1)(uint8_t * pix, int line_size);
153 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
154
155 me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */
156 me_cmp_func sse[5];
157 me_cmp_func hadamard8_diff[5];
158 me_cmp_func dct_sad[5];
159 me_cmp_func quant_psnr[5];
160 me_cmp_func bit[5];
161 me_cmp_func rd[5];
162 me_cmp_func vsad[5];
163 me_cmp_func vsse[5];
164
165 me_cmp_func me_pre_cmp[5];
166 me_cmp_func me_cmp[5];
167 me_cmp_func me_sub_cmp[5];
168 me_cmp_func mb_cmp[5];
169 me_cmp_func ildct_cmp[5]; //only width 16 used
170
171 /**
172 * Halfpel motion compensation with rounding (a+b+1)>>1.
173 * this is an array[4][4] of motion compensation funcions for 4
174 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
175 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
176 * @param block destination where the result is stored
177 * @param pixels source
178 * @param line_size number of bytes in a horizontal line of block
179 * @param h height
180 */
181 op_pixels_func put_pixels_tab[4][4];
182
183 /**
184 * Halfpel motion compensation with rounding (a+b+1)>>1.
185 * This is an array[4][4] of motion compensation functions for 4
186 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
187 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
188 * @param block destination into which the result is averaged (a+b+1)>>1
189 * @param pixels source
190 * @param line_size number of bytes in a horizontal line of block
191 * @param h height
192 */
193 op_pixels_func avg_pixels_tab[4][4];
194
195 /**
196 * Halfpel motion compensation with no rounding (a+b)>>1.
197 * this is an array[2][4] of motion compensation funcions for 2
198 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
199 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
200 * @param block destination where the result is stored
201 * @param pixels source
202 * @param line_size number of bytes in a horizontal line of block
203 * @param h height
204 */
205 op_pixels_func put_no_rnd_pixels_tab[2][4];
206
207 /**
208 * Halfpel motion compensation with no rounding (a+b)>>1.
209 * this is an array[2][4] of motion compensation funcions for 2
210 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
211 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
212 * @param block destination into which the result is averaged (a+b)>>1
213 * @param pixels source
214 * @param line_size number of bytes in a horizontal line of block
215 * @param h height
216 */
217 op_pixels_func avg_no_rnd_pixels_tab[2][4];
218
219 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);
220
221 /**
222 * Thirdpel motion compensation with rounding (a+b+1)>>1.
223 * this is an array[12] of motion compensation funcions for the 9 thirdpel positions<br>
224 * *pixels_tab[ xthirdpel + 4*ythirdpel ]
225 * @param block destination where the result is stored
226 * @param pixels source
227 * @param line_size number of bytes in a horizontal line of block
228 * @param h height
229 */
230 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
231 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
232
233 qpel_mc_func put_qpel_pixels_tab[2][16];
234 qpel_mc_func avg_qpel_pixels_tab[2][16];
235 qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
236 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
237 qpel_mc_func put_mspel_pixels_tab[8];
238
239 /**
240 * h264 Chram MC
241 */
242 h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
243 h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
244
245 qpel_mc_func put_h264_qpel_pixels_tab[3][16];
246 qpel_mc_func avg_h264_qpel_pixels_tab[3][16];
247
248 me_cmp_func pix_abs[2][4];
249
250 /* huffyuv specific */
251 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
252 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
253 /**
254 * subtract huffyuv's variant of median prediction
255 * note, this might read from src1[-1], src2[-1]
256 */
257 void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
258 void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w);
259
260 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
261 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
262
263 /* (I)DCT */
264 void (*fdct)(DCTELEM *block/* align 16*/);
265 void (*fdct248)(DCTELEM *block/* align 16*/);
266
267 /* IDCT really*/
268 void (*idct)(DCTELEM *block/* align 16*/);
269
270 /**
271 * block -> idct -> clip to unsigned 8 bit -> dest.
272 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
273 * @param line_size size in bytes of a horizotal line of dest
274 */
275 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
276
277 /**
278 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
279 * @param line_size size in bytes of a horizotal line of dest
280 */
281 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
282
283 /**
284 * idct input permutation.
285 * several optimized IDCTs need a permutated input (relative to the normal order of the reference
286 * IDCT)
287 * this permutation must be performed before the idct_put/add, note, normally this can be merged
288 * with the zigzag/alternate scan<br>
289 * an example to avoid confusion:
290 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
291 * - (x -> referece dct -> reference idct -> x)
292 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
293 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
294 */
295 uint8_t idct_permutation[64];
296 int idct_permutation_type;
297 #define FF_NO_IDCT_PERM 1
298 #define FF_LIBMPEG2_IDCT_PERM 2
299 #define FF_SIMPLE_IDCT_PERM 3
300 #define FF_TRANSPOSE_IDCT_PERM 4
301
302 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
303 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
304 #define BASIS_SHIFT 16
305 #define RECON_SHIFT 6
306
307 /**
308 * This function handles any initialization for the VP3 DSP functions.
309 */
310 void (*vp3_dsp_init)(void);
311
312 /**
313 * This function is responsible for taking a block of zigzag'd,
314 * quantized DCT coefficients and reconstructing the original block of
315 * samples.
316 * @param input_data 64 zigzag'd, quantized DCT coefficients
317 * @param dequant_matrix 64 zigzag'd quantizer coefficients
318 * @param coeff_count index of the last coefficient
319 * @param output_samples space for 64 DCTELEMs where the transformed
320 * samples will be stored
321 */
322 void (*vp3_idct)(int16_t *input_data, int16_t *dequant_matrix,
323 int coeff_count, DCTELEM *output_samples);
324
325 } DSPContext;
326
327 void dsputil_static_init(void);
328 void dsputil_init(DSPContext* p, AVCodecContext *avctx);
329
330 /**
331 * permute block according to permuatation.
332 * @param last last non zero element in scantable order
333 */
334 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
335
336 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
337
338 #define BYTE_VEC32(c) ((c)*0x01010101UL)
339
340 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
341 {
342 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
343 }
344
345 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
346 {
347 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
348 }
349
350 /**
351 * Empty mmx state.
352 * this must be called between any dsp function and float/double code.
353 * for example sin(); dsp->idct_put(); emms_c(); cos()
354 */
355 #define emms_c()
356
357 /* should be defined by architectures supporting
358 one or more MultiMedia extension */
359 int mm_support(void);
360
361 #define __align16 __attribute__ ((aligned (16)))
362
363 #if defined(HAVE_MMX)
364
365 #undef emms_c
366
367 #define MM_MMX 0x0001 /* standard MMX */
368 #define MM_3DNOW 0x0004 /* AMD 3DNOW */
369 #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
370 #define MM_SSE 0x0008 /* SSE functions */
371 #define MM_SSE2 0x0010 /* PIV SSE2 functions */
372
373 extern int mm_flags;
374
375 void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
376 void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
377
378 static inline void emms(void)
379 {
380 __asm __volatile ("emms;":::"memory");
381 }
382
383
384 #define emms_c() \
385 {\
386 if (mm_flags & MM_MMX)\
387 emms();\
388 }
389
390 #define __align8 __attribute__ ((aligned (8)))
391
392 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
393 void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
394
395 #elif defined(ARCH_ARMV4L)
396
397 /* This is to use 4 bytes read to the IDCT pointers for some 'zero'
398 line optimizations */
399 #define __align8 __attribute__ ((aligned (4)))
400
401 void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
402
403 #elif defined(HAVE_MLIB)
404
405 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
406 #define __align8 __attribute__ ((aligned (8)))
407
408 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
409
410 #elif defined(ARCH_SPARC)
411
412 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
413 #define __align8 __attribute__ ((aligned (8)))
414 void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
415
416 #elif defined(ARCH_ALPHA)
417
418 #define __align8 __attribute__ ((aligned (8)))
419
420 void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
421
422 #elif defined(ARCH_POWERPC)
423
424 #define MM_ALTIVEC 0x0001 /* standard AltiVec */
425
426 extern int mm_flags;
427
428 #if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN)
429 #define pixel altivec_pixel
430 #include <altivec.h>
431 #undef pixel
432 #endif
433
434 #define __align8 __attribute__ ((aligned (16)))
435
436 void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
437
438 #elif defined(HAVE_MMI)
439
440 #define __align8 __attribute__ ((aligned (16)))
441
442 void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
443
444 #elif defined(ARCH_SH4)
445
446 #define __align8 __attribute__ ((aligned (8)))
447
448 void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
449
450 #else
451
452 #define __align8
453
454 #endif
455
456 #ifdef __GNUC__
457
458 struct unaligned_64 { uint64_t l; } __attribute__((packed));
459 struct unaligned_32 { uint32_t l; } __attribute__((packed));
460 struct unaligned_16 { uint16_t l; } __attribute__((packed));
461
462 #define LD16(a) (((const struct unaligned_16 *) (a))->l)
463 #define LD32(a) (((const struct unaligned_32 *) (a))->l)
464 #define LD64(a) (((const struct unaligned_64 *) (a))->l)
465
466 #define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)
467
468 #else /* __GNUC__ */
469
470 #define LD16(a) (*((uint16_t*)(a)))
471 #define LD32(a) (*((uint32_t*)(a)))
472 #define LD64(a) (*((uint64_t*)(a)))
473
474 #define ST32(a, b) *((uint32_t*)(a)) = (b)
475
476 #endif /* !__GNUC__ */
477
478 /* PSNR */
479 void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
480 int orig_linesize[3], int coded_linesize,
481 AVCodecContext *avctx);
482
483 /* FFT computation */
484
485 /* NOTE: soon integer code will be added, so you must use the
486 FFTSample type */
487 typedef float FFTSample;
488
489 typedef struct FFTComplex {
490 FFTSample re, im;
491 } FFTComplex;
492
493 typedef struct FFTContext {
494 int nbits;
495 int inverse;
496 uint16_t *revtab;
497 FFTComplex *exptab;
498 FFTComplex *exptab1; /* only used by SSE code */
499 void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
500 } FFTContext;
501
502 int ff_fft_init(FFTContext *s, int nbits, int inverse);
503 void ff_fft_permute(FFTContext *s, FFTComplex *z);
504 void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
505 void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
506 void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
507
508 static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
509 {
510 s->fft_calc(s, z);
511 }
512 void ff_fft_end(FFTContext *s);
513
514 /* MDCT computation */
515
516 typedef struct MDCTContext {
517 int n; /* size of MDCT (i.e. number of input data * 2) */
518 int nbits; /* n = 2^nbits */
519 /* pre/post rotation tables */
520 FFTSample *tcos;
521 FFTSample *tsin;
522 FFTContext fft;
523 } MDCTContext;
524
525 int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
526 void ff_imdct_calc(MDCTContext *s, FFTSample *output,
527 const FFTSample *input, FFTSample *tmp);
528 void ff_mdct_calc(MDCTContext *s, FFTSample *out,
529 const FFTSample *input, FFTSample *tmp);
530 void ff_mdct_end(MDCTContext *s);
531
532 #define WARPER8_16(name8, name16)\
533 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
534 return name8(s, dst , src , stride, h)\
535 +name8(s, dst+8 , src+8 , stride, h);\
536 }
537
538 #define WARPER8_16_SQ(name8, name16)\
539 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
540 int score=0;\
541 score +=name8(s, dst , src , stride, 8);\
542 score +=name8(s, dst+8 , src+8 , stride, 8);\
543 if(h==16){\
544 dst += 8*stride;\
545 src += 8*stride;\
546 score +=name8(s, dst , src , stride, 8);\
547 score +=name8(s, dst+8 , src+8 , stride, 8);\
548 }\
549 return score;\
550 }
551
552 #ifndef HAVE_LRINTF
553 /* XXX: add ISOC specific test to avoid specific BSD testing. */
554 /* better than nothing implementation. */
555 /* btw, rintf() is existing on fbsd too -- alex */
556 static inline long int lrintf(float x)
557 {
558 #ifdef CONFIG_WIN32
559 /* XXX: incorrect, but make it compile */
560 return (int)(x);
561 #else
562 return (int)(rint(x));
563 #endif
564 }
565 #endif
566
567 #endif