137
|
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
|
|
64 /* minimum alignment rules ;)
|
|
65 if u notice errors in the align stuff, need more alignment for some asm code for some cpu
|
|
66 or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ...
|
|
67
|
|
68 !warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible)
|
|
69 i (michael) didnt check them, these are just the alignents which i think could be reached easily ...
|
|
70
|
|
71 !future video codecs might need functions with less strict alignment
|
|
72 */
|
|
73
|
|
74 /*
|
|
75 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
|
|
76 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
|
|
77 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
|
|
78 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
|
|
79 void clear_blocks_c(DCTELEM *blocks);
|
|
80 */
|
|
81
|
|
82 /* add and put pixel (decoding) */
|
|
83 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
|
|
84 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
|
|
85 typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
|
|
86 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);
|
|
87 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
|
|
88 typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
|
|
89
|
|
90 #define DEF_OLD_QPEL(name)\
|
|
91 void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
|
|
92 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
|
|
93 void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
|
|
94
|
|
95 DEF_OLD_QPEL(qpel16_mc11_old_c)
|
|
96 DEF_OLD_QPEL(qpel16_mc31_old_c)
|
|
97 DEF_OLD_QPEL(qpel16_mc12_old_c)
|
|
98 DEF_OLD_QPEL(qpel16_mc32_old_c)
|
|
99 DEF_OLD_QPEL(qpel16_mc13_old_c)
|
|
100 DEF_OLD_QPEL(qpel16_mc33_old_c)
|
|
101 DEF_OLD_QPEL(qpel8_mc11_old_c)
|
|
102 DEF_OLD_QPEL(qpel8_mc31_old_c)
|
|
103 DEF_OLD_QPEL(qpel8_mc12_old_c)
|
|
104 DEF_OLD_QPEL(qpel8_mc32_old_c)
|
|
105 DEF_OLD_QPEL(qpel8_mc13_old_c)
|
|
106 DEF_OLD_QPEL(qpel8_mc33_old_c)
|
|
107
|
|
108 #define CALL_2X_PIXELS(a, b, n)\
|
|
109 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
|
|
110 b(block , pixels , line_size, h);\
|
|
111 b(block+n, pixels+n, line_size, h);\
|
|
112 }
|
|
113
|
|
114 /* motion estimation */
|
|
115 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
|
|
116 // allthough currently h<4 is not used as functions with width <8 are not used and neither implemented
|
|
117 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))*/;
|
|
118
|
|
119
|
|
120 /**
|
|
121 * DSPContext.
|
|
122 */
|
|
123 typedef struct DSPContext {
|
|
124 /* pixel ops : interface with DCT */
|
|
125 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
|
|
126 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
|
|
127 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
|
|
128 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
|
|
129 /**
|
|
130 * translational global motion compensation.
|
|
131 */
|
|
132 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
|
|
133 /**
|
|
134 * global motion compensation.
|
|
135 */
|
|
136 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
|
|
137 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
|
|
138 void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
|
|
139 int (*pix_sum)(uint8_t * pix, int line_size);
|
|
140 int (*pix_norm1)(uint8_t * pix, int line_size);
|
|
141 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
|
|
142
|
|
143 me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */
|
|
144 me_cmp_func sse[5];
|
|
145 me_cmp_func hadamard8_diff[5];
|
|
146 me_cmp_func dct_sad[5];
|
|
147 me_cmp_func quant_psnr[5];
|
|
148 me_cmp_func bit[5];
|
|
149 me_cmp_func rd[5];
|
|
150 me_cmp_func vsad[5];
|
|
151 me_cmp_func vsse[5];
|
|
152
|
|
153 me_cmp_func me_pre_cmp[5];
|
|
154 me_cmp_func me_cmp[5];
|
|
155 me_cmp_func me_sub_cmp[5];
|
|
156 me_cmp_func mb_cmp[5];
|
|
157 me_cmp_func ildct_cmp[5]; //only width 16 used
|
|
158
|
|
159 /**
|
|
160 * Halfpel motion compensation with rounding (a+b+1)>>1.
|
|
161 * this is an array[4][4] of motion compensation funcions for 4
|
|
162 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
|
|
163 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
|
|
164 * @param block destination where the result is stored
|
|
165 * @param pixels source
|
|
166 * @param line_size number of bytes in a horizontal line of block
|
|
167 * @param h height
|
|
168 */
|
|
169 op_pixels_func put_pixels_tab[4][4];
|
|
170
|
|
171 /**
|
|
172 * Halfpel motion compensation with rounding (a+b+1)>>1.
|
|
173 * This is an array[4][4] of motion compensation functions 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 into which the result is averaged (a+b+1)>>1
|
|
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 avg_pixels_tab[4][4];
|
|
182
|
|
183 /**
|
|
184 * Halfpel motion compensation with no rounding (a+b)>>1.
|
|
185 * this is an array[2][4] of motion compensation funcions for 2
|
|
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 where the result is stored
|
|
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 put_no_rnd_pixels_tab[2][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 into which the result is averaged (a+b)>>1
|
|
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 avg_no_rnd_pixels_tab[2][4];
|
|
206
|
|
207 /**
|
|
208 * Thirdpel motion compensation with rounding (a+b+1)>>1.
|
|
209 * this is an array[12] of motion compensation funcions for the 9 thirdpel positions<br>
|
|
210 * *pixels_tab[ xthirdpel + 4*ythirdpel ]
|
|
211 * @param block destination where the result is stored
|
|
212 * @param pixels source
|
|
213 * @param line_size number of bytes in a horizontal line of block
|
|
214 * @param h height
|
|
215 */
|
|
216 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
|
|
217 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
|
|
218
|
|
219 qpel_mc_func put_qpel_pixels_tab[2][16];
|
|
220 qpel_mc_func avg_qpel_pixels_tab[2][16];
|
|
221 qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
|
|
222 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
|
|
223 qpel_mc_func put_mspel_pixels_tab[8];
|
|
224
|
|
225 /**
|
|
226 * h264 Chram MC
|
|
227 */
|
|
228 h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
|
|
229 h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
|
|
230
|
|
231 qpel_mc_func put_h264_qpel_pixels_tab[3][16];
|
|
232 qpel_mc_func avg_h264_qpel_pixels_tab[3][16];
|
|
233
|
|
234 me_cmp_func pix_abs[2][4];
|
|
235
|
|
236 /* huffyuv specific */
|
|
237 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
|
|
238 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
|
|
239 /**
|
|
240 * subtract huffyuv's variant of median prediction
|
|
241 * note, this might read from src1[-1], src2[-1]
|
|
242 */
|
|
243 void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
|
|
244 void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w);
|
|
245
|
|
246 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
|
|
247 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
|
|
248
|
|
249 /* (I)DCT */
|
|
250 void (*fdct)(DCTELEM *block/* align 16*/);
|
|
251 void (*fdct248)(DCTELEM *block/* align 16*/);
|
|
252
|
|
253 /* IDCT really*/
|
|
254 void (*idct)(DCTELEM *block/* align 16*/);
|
|
255
|
|
256 /**
|
|
257 * block -> idct -> clip to unsigned 8 bit -> dest.
|
|
258 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
|
|
259 * @param line_size size in bytes of a horizotal line of dest
|
|
260 */
|
|
261 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
|
|
262
|
|
263 /**
|
|
264 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
|
|
265 * @param line_size size in bytes of a horizotal line of dest
|
|
266 */
|
|
267 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
|
|
268
|
|
269 /**
|
|
270 * idct input permutation.
|
|
271 * several optimized IDCTs need a permutated input (relative to the normal order of the reference
|
|
272 * IDCT)
|
|
273 * this permutation must be performed before the idct_put/add, note, normally this can be merged
|
|
274 * with the zigzag/alternate scan<br>
|
|
275 * an example to avoid confusion:
|
|
276 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
|
|
277 * - (x -> referece dct -> reference idct -> x)
|
|
278 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
|
|
279 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
|
|
280 */
|
|
281 uint8_t idct_permutation[64];
|
|
282 int idct_permutation_type;
|
|
283 #define FF_NO_IDCT_PERM 1
|
|
284 #define FF_LIBMPEG2_IDCT_PERM 2
|
|
285 #define FF_SIMPLE_IDCT_PERM 3
|
|
286 #define FF_TRANSPOSE_IDCT_PERM 4
|
|
287
|
|
288 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
|
|
289 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
|
|
290 #define BASIS_SHIFT 16
|
|
291 #define RECON_SHIFT 6
|
|
292
|
|
293 } DSPContext;
|
|
294
|
|
295 void dsputil_static_init(void);
|
|
296 void dsputil_init(DSPContext* p, AVCodecContext *avctx);
|
|
297
|
|
298 /**
|
|
299 * permute block according to permuatation.
|
|
300 * @param last last non zero element in scantable order
|
|
301 */
|
|
302 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
|
|
303
|
|
304 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
|
|
305
|
|
306 #define BYTE_VEC32(c) ((c)*0x01010101UL)
|
|
307
|
|
308 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
|
|
309 {
|
|
310 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
|
|
311 }
|
|
312
|
|
313 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
|
|
314 {
|
|
315 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
|
|
316 }
|
|
317
|
|
318 /**
|
|
319 * Empty mmx state.
|
|
320 * this must be called between any dsp function and float/double code.
|
|
321 * for example sin(); dsp->idct_put(); emms_c(); cos()
|
|
322 */
|
|
323 #define emms_c()
|
|
324
|
|
325 /* should be defined by architectures supporting
|
|
326 one or more MultiMedia extension */
|
|
327 int mm_support(void);
|
|
328
|
|
329 #if defined(HAVE_MMX)
|
|
330
|
|
331 #undef emms_c
|
|
332
|
|
333 #define MM_MMX 0x0001 /* standard MMX */
|
|
334 #define MM_3DNOW 0x0004 /* AMD 3DNOW */
|
|
335 #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
|
|
336 #define MM_SSE 0x0008 /* SSE functions */
|
|
337 #define MM_SSE2 0x0010 /* PIV SSE2 functions */
|
|
338
|
|
339 extern int mm_flags;
|
|
340
|
|
341 void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
|
|
342 void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
|
|
343
|
|
344 static inline void emms(void)
|
|
345 {
|
|
346 __asm __volatile ("emms;":::"memory");
|
|
347 }
|
|
348
|
|
349
|
|
350 #define emms_c() \
|
|
351 {\
|
|
352 if (mm_flags & MM_MMX)\
|
|
353 emms();\
|
|
354 }
|
|
355
|
|
356 #define __align8 __attribute__ ((aligned (8)))
|
|
357
|
|
358 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
|
|
359 void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
|
|
360
|
|
361 #elif defined(ARCH_ARMV4L)
|
|
362
|
|
363 /* This is to use 4 bytes read to the IDCT pointers for some 'zero'
|
|
364 line ptimizations */
|
|
365 #define __align8 __attribute__ ((aligned (4)))
|
|
366
|
|
367 void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
|
|
368
|
|
369 #elif defined(HAVE_MLIB)
|
|
370
|
|
371 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
|
|
372 #define __align8 __attribute__ ((aligned (8)))
|
|
373
|
|
374 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
|
|
375
|
|
376 #elif defined(ARCH_ALPHA)
|
|
377
|
|
378 #define __align8 __attribute__ ((aligned (8)))
|
|
379
|
|
380 void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
|
|
381
|
|
382 #elif defined(ARCH_POWERPC)
|
|
383
|
|
384 #define MM_ALTIVEC 0x0001 /* standard AltiVec */
|
|
385
|
|
386 extern int mm_flags;
|
|
387
|
|
388 #if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN)
|
|
389 #define pixel altivec_pixel
|
|
390 #include <altivec.h>
|
|
391 #undef pixel
|
|
392 #endif
|
|
393
|
|
394 #define __align8 __attribute__ ((aligned (16)))
|
|
395
|
|
396 void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
|
|
397
|
|
398 #elif defined(HAVE_MMI)
|
|
399
|
|
400 #define __align8 __attribute__ ((aligned (16)))
|
|
401
|
|
402 void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
|
|
403
|
|
404 #elif defined(ARCH_SH4)
|
|
405
|
|
406 #define __align8 __attribute__ ((aligned (8)))
|
|
407
|
|
408 void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
|
|
409
|
|
410 #else
|
|
411
|
|
412 #define __align8
|
|
413
|
|
414 #endif
|
|
415
|
|
416 #ifdef __GNUC__
|
|
417
|
|
418 struct unaligned_64 { uint64_t l; } __attribute__((packed));
|
|
419 struct unaligned_32 { uint32_t l; } __attribute__((packed));
|
|
420 struct unaligned_16 { uint16_t l; } __attribute__((packed));
|
|
421
|
|
422 #define LD16(a) (((const struct unaligned_16 *) (a))->l)
|
|
423 #define LD32(a) (((const struct unaligned_32 *) (a))->l)
|
|
424 #define LD64(a) (((const struct unaligned_64 *) (a))->l)
|
|
425
|
|
426 #define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)
|
|
427
|
|
428 #else /* __GNUC__ */
|
|
429
|
|
430 #define LD16(a) (*((uint16_t*)(a)))
|
|
431 #define LD32(a) (*((uint32_t*)(a)))
|
|
432 #define LD64(a) (*((uint64_t*)(a)))
|
|
433
|
|
434 #define ST32(a, b) *((uint32_t*)(a)) = (b)
|
|
435
|
|
436 #endif /* !__GNUC__ */
|
|
437
|
|
438 /* PSNR */
|
|
439 void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
|
|
440 int orig_linesize[3], int coded_linesize,
|
|
441 AVCodecContext *avctx);
|
|
442
|
|
443 /* FFT computation */
|
|
444
|
|
445 /* NOTE: soon integer code will be added, so you must use the
|
|
446 FFTSample type */
|
|
447 typedef float FFTSample;
|
|
448
|
|
449 typedef struct FFTComplex {
|
|
450 FFTSample re, im;
|
|
451 } FFTComplex;
|
|
452
|
|
453 typedef struct FFTContext {
|
|
454 int nbits;
|
|
455 int inverse;
|
|
456 uint16_t *revtab;
|
|
457 FFTComplex *exptab;
|
|
458 FFTComplex *exptab1; /* only used by SSE code */
|
|
459 void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
|
|
460 } FFTContext;
|
|
461
|
|
462 int fft_inits(FFTContext *s, int nbits, int inverse);
|
|
463 void fft_permute(FFTContext *s, FFTComplex *z);
|
|
464 void fft_calc_c(FFTContext *s, FFTComplex *z);
|
|
465 void fft_calc_sse(FFTContext *s, FFTComplex *z);
|
|
466 void fft_calc_altivec(FFTContext *s, FFTComplex *z);
|
|
467
|
|
468 static inline void fft_calc(FFTContext *s, FFTComplex *z)
|
|
469 {
|
|
470 s->fft_calc(s, z);
|
|
471 }
|
|
472 void fft_end(FFTContext *s);
|
|
473
|
|
474 /* MDCT computation */
|
|
475
|
|
476 typedef struct MDCTContext {
|
|
477 int n; /* size of MDCT (i.e. number of input data * 2) */
|
|
478 int nbits; /* n = 2^nbits */
|
|
479 /* pre/post rotation tables */
|
|
480 FFTSample *tcos;
|
|
481 FFTSample *tsin;
|
|
482 FFTContext fft;
|
|
483 } MDCTContext;
|
|
484
|
|
485 int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
|
|
486 void ff_imdct_calc(MDCTContext *s, FFTSample *output,
|
|
487 const FFTSample *input, FFTSample *tmp);
|
|
488 void ff_mdct_calc(MDCTContext *s, FFTSample *out,
|
|
489 const FFTSample *input, FFTSample *tmp);
|
|
490 void ff_mdct_end(MDCTContext *s);
|
|
491
|
|
492 #define WARPER8_16(name8, name16)\
|
|
493 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
|
|
494 return name8(s, dst , src , stride, h)\
|
|
495 +name8(s, dst+8 , src+8 , stride, h);\
|
|
496 }
|
|
497
|
|
498 #define WARPER8_16_SQ(name8, name16)\
|
|
499 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
|
|
500 int score=0;\
|
|
501 score +=name8(s, dst , src , stride, 8);\
|
|
502 score +=name8(s, dst+8 , src+8 , stride, 8);\
|
|
503 if(h==16){\
|
|
504 dst += 8*stride;\
|
|
505 src += 8*stride;\
|
|
506 score +=name8(s, dst , src , stride, 8);\
|
|
507 score +=name8(s, dst+8 , src+8 , stride, 8);\
|
|
508 }\
|
|
509 return score;\
|
|
510 }
|
|
511
|
|
512 #ifndef HAVE_LRINTF
|
|
513 /* XXX: add ISOC specific test to avoid specific BSD testing. */
|
|
514 /* better than nothing implementation. */
|
|
515 /* btw, rintf() is existing on fbsd too -- alex */
|
|
516 static inline long int lrintf(float x)
|
|
517 {
|
|
518 return (int)(rint(x));
|
|
519 }
|
|
520 #endif
|
|
521
|
|
522 #endif
|