lmin/lmax
[libav.git] / libavcodec / dsputil.h
CommitLineData
ff4ec49e
FB
1/*
2 * DSP utils
3 * Copyright (c) 2000, 2001, 2002 Fabrice Bellard.
8f2ab833 4 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
ff4ec49e
FB
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 */
24641185
MN
20
21/**
22 * @file dsputil.h
983e3246 23 * DSP utils.
5755c27f
MN
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
24641185
MN
26 */
27
de6d9b64
FB
28#ifndef DSPUTIL_H
29#define DSPUTIL_H
30
31#include "common.h"
43f1708f 32#include "avcodec.h"
de6d9b64 33
24641185 34
44eb4951 35//#define DEBUG
de6d9b64
FB
36/* dct code */
37typedef short DCTELEM;
38
03c94ede 39void fdct_ifast (DCTELEM *data);
48b1f800 40void fdct_ifast248 (DCTELEM *data);
28db7fce 41void ff_jpeg_fdct_islow (DCTELEM *data);
10acc479 42void ff_fdct248_islow (DCTELEM *data);
de6d9b64
FB
43
44void j_rev_dct (DCTELEM *data);
45
3f09f52a 46void ff_fdct_mmx(DCTELEM *block);
cf3bf5bb 47void ff_fdct_mmx2(DCTELEM *block);
8fd19ab2 48void ff_fdct_sse2(DCTELEM *block);
de6d9b64 49
e0eac44e 50/* encoding scans */
0c1a9eda
ZK
51extern const uint8_t ff_alternate_horizontal_scan[64];
52extern const uint8_t ff_alternate_vertical_scan[64];
53extern const uint8_t ff_zigzag_direct[64];
10acc479 54extern const uint8_t ff_zigzag248_direct[64];
5a240838 55
de6d9b64
FB
56/* pixel operations */
57#define MAX_NEG_CROP 384
58
59/* temporary */
0c1a9eda
ZK
60extern uint32_t squareTbl[512];
61extern uint8_t cropTbl[256 + 2 * MAX_NEG_CROP];
de6d9b64 62
44cb64ee
MM
63/* VP3 DSP functions */
64void vp3_dsp_init_c(void);
65void vp3_idct_put_c(int16_t *input_data, int16_t *dequant_matrix,
66 int coeff_count, uint8_t *dest, int stride);
67void vp3_idct_add_c(int16_t *input_data, int16_t *dequant_matrix,
68 int coeff_count, uint8_t *dest, int stride);
69
70void vp3_dsp_init_mmx(void);
71void vp3_idct_put_mmx(int16_t *input_data, int16_t *dequant_matrix,
72 int coeff_count, uint8_t *dest, int stride);
73void vp3_idct_add_mmx(int16_t *input_data, int16_t *dequant_matrix,
74 int coeff_count, uint8_t *dest, int stride);
75
38acbc3c
MM
76void vp3_dsp_init_sse2(void);
77void vp3_idct_put_sse2(int16_t *input_data, int16_t *dequant_matrix,
78 int coeff_count, uint8_t *dest, int stride);
79void vp3_idct_add_sse2(int16_t *input_data, int16_t *dequant_matrix,
80 int coeff_count, uint8_t *dest, int stride);
81
de6d9b64 82
b7c27ee6 83/* minimum alignment rules ;)
eb4b3dd3 84if u notice errors in the align stuff, need more alignment for some asm code for some cpu
b7c27ee6
MN
85or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ...
86
87!warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible)
88i (michael) didnt check them, these are just the alignents which i think could be reached easily ...
de6d9b64 89
b7c27ee6
MN
90!future video codecs might need functions with less strict alignment
91*/
92
eb4b3dd3 93/*
0c1a9eda
ZK
94void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
95void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
96void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
97void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
649c00c9 98void clear_blocks_c(DCTELEM *blocks);
eb4b3dd3 99*/
de6d9b64
FB
100
101/* add and put pixel (decoding) */
b7c27ee6 102// blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
7d67aa9b 103//h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
0c1a9eda 104typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
669ac79c 105typedef 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);
0c1a9eda 106typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
0da71265 107typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
b3184779 108
db794953 109#define DEF_OLD_QPEL(name)\
0c1a9eda
ZK
110void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
111void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
112void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
db794953
MN
113
114DEF_OLD_QPEL(qpel16_mc11_old_c)
115DEF_OLD_QPEL(qpel16_mc31_old_c)
116DEF_OLD_QPEL(qpel16_mc12_old_c)
117DEF_OLD_QPEL(qpel16_mc32_old_c)
118DEF_OLD_QPEL(qpel16_mc13_old_c)
119DEF_OLD_QPEL(qpel16_mc33_old_c)
120DEF_OLD_QPEL(qpel8_mc11_old_c)
121DEF_OLD_QPEL(qpel8_mc31_old_c)
122DEF_OLD_QPEL(qpel8_mc12_old_c)
123DEF_OLD_QPEL(qpel8_mc32_old_c)
124DEF_OLD_QPEL(qpel8_mc13_old_c)
125DEF_OLD_QPEL(qpel8_mc33_old_c)
b3184779
MN
126
127#define CALL_2X_PIXELS(a, b, n)\
128static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
129 b(block , pixels , line_size, h);\
130 b(block+n, pixels+n, line_size, h);\
131}
44eb4951 132
de6d9b64 133/* motion estimation */
7d67aa9b
MN
134// h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
135// allthough currently h<4 is not used as functions with width <8 are not used and neither implemented
bb198e19 136typedef 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))*/;
1457ab52 137
0da71265 138
24641185
MN
139/**
140 * DSPContext.
141 */
eb4b3dd3
ZK
142typedef struct DSPContext {
143 /* pixel ops : interface with DCT */
0c1a9eda
ZK
144 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
145 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
146 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
147 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
d518aebd
MN
148 /**
149 * translational global motion compensation.
150 */
0c1a9eda 151 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
d518aebd
MN
152 /**
153 * global motion compensation.
154 */
0c1a9eda 155 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
eb4b3dd3
ZK
156 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
157 void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
0c1a9eda
ZK
158 int (*pix_sum)(uint8_t * pix, int line_size);
159 int (*pix_norm1)(uint8_t * pix, int line_size);
bb198e19
MN
160// 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
161
622348f9
MN
162 me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */
163 me_cmp_func sse[5];
164 me_cmp_func hadamard8_diff[5];
165 me_cmp_func dct_sad[5];
166 me_cmp_func quant_psnr[5];
167 me_cmp_func bit[5];
168 me_cmp_func rd[5];
169 me_cmp_func vsad[5];
170 me_cmp_func vsse[5];
1457ab52 171
bb198e19
MN
172 me_cmp_func me_pre_cmp[5];
173 me_cmp_func me_cmp[5];
174 me_cmp_func me_sub_cmp[5];
175 me_cmp_func mb_cmp[5];
622348f9 176 me_cmp_func ildct_cmp[5]; //only width 16 used
eb4b3dd3 177
d518aebd
MN
178 /**
179 * Halfpel motion compensation with rounding (a+b+1)>>1.
669ac79c 180 * this is an array[4][4] of motion compensation funcions for 4
e5771f4f 181 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
5755c27f 182 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
d518aebd
MN
183 * @param block destination where the result is stored
184 * @param pixels source
185 * @param line_size number of bytes in a horizontal line of block
186 * @param h height
187 */
669ac79c 188 op_pixels_func put_pixels_tab[4][4];
d518aebd
MN
189
190 /**
191 * Halfpel motion compensation with rounding (a+b+1)>>1.
4e8eed2f 192 * This is an array[4][4] of motion compensation functions for 4
e5771f4f 193 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
5755c27f 194 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
d518aebd
MN
195 * @param block destination into which the result is averaged (a+b+1)>>1
196 * @param pixels source
197 * @param line_size number of bytes in a horizontal line of block
198 * @param h height
199 */
da3b9756 200 op_pixels_func avg_pixels_tab[4][4];
d518aebd
MN
201
202 /**
203 * Halfpel motion compensation with no rounding (a+b)>>1.
eb14c713
MN
204 * this is an array[2][4] of motion compensation funcions for 2
205 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
5755c27f 206 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
d518aebd
MN
207 * @param block destination where the result is stored
208 * @param pixels source
209 * @param line_size number of bytes in a horizontal line of block
210 * @param h height
211 */
eb4b3dd3 212 op_pixels_func put_no_rnd_pixels_tab[2][4];
d518aebd
MN
213
214 /**
215 * Halfpel motion compensation with no rounding (a+b)>>1.
eb14c713
MN
216 * this is an array[2][4] of motion compensation funcions for 2
217 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
5755c27f 218 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
d518aebd
MN
219 * @param block destination into which the result is averaged (a+b)>>1
220 * @param pixels source
221 * @param line_size number of bytes in a horizontal line of block
222 * @param h height
223 */
eb4b3dd3 224 op_pixels_func avg_no_rnd_pixels_tab[2][4];
669ac79c 225
c0a0170c
MN
226 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);
227
669ac79c
MN
228 /**
229 * Thirdpel motion compensation with rounding (a+b+1)>>1.
230 * this is an array[12] of motion compensation funcions for the 9 thirdpel positions<br>
231 * *pixels_tab[ xthirdpel + 4*ythirdpel ]
232 * @param block destination where the result is stored
233 * @param pixels source
234 * @param line_size number of bytes in a horizontal line of block
235 * @param h height
236 */
237 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
da3b9756
MM
238 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
239
eb4b3dd3
ZK
240 qpel_mc_func put_qpel_pixels_tab[2][16];
241 qpel_mc_func avg_qpel_pixels_tab[2][16];
242 qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
243 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
1457ab52 244 qpel_mc_func put_mspel_pixels_tab[8];
0da71265
MN
245
246 /**
247 * h264 Chram MC
248 */
249 h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
250 h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
eb4b3dd3 251
0da71265
MN
252 qpel_mc_func put_h264_qpel_pixels_tab[3][16];
253 qpel_mc_func avg_h264_qpel_pixels_tab[3][16];
254
bb198e19 255 me_cmp_func pix_abs[2][4];
11f18faf
MN
256
257 /* huffyuv specific */
11f18faf 258 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
1457ab52 259 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
84705403
MN
260 /**
261 * subtract huffyuv's variant of median prediction
262 * note, this might read from src1[-1], src2[-1]
263 */
264 void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
3d2e8cce 265 void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w);
b0368839 266
332f9ac4
MN
267 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
268 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
269
b0368839
MN
270 /* (I)DCT */
271 void (*fdct)(DCTELEM *block/* align 16*/);
10acc479 272 void (*fdct248)(DCTELEM *block/* align 16*/);
4fb518c3
MN
273
274 /* IDCT really*/
275 void (*idct)(DCTELEM *block/* align 16*/);
24641185
MN
276
277 /**
77c92c2d 278 * block -> idct -> clip to unsigned 8 bit -> dest.
24641185 279 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
a1adf436 280 * @param line_size size in bytes of a horizotal line of dest
24641185 281 */
b0368839 282 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
24641185
MN
283
284 /**
285 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
a1adf436 286 * @param line_size size in bytes of a horizotal line of dest
24641185 287 */
b0368839 288 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
24641185
MN
289
290 /**
77c92c2d 291 * idct input permutation.
05493021
MN
292 * several optimized IDCTs need a permutated input (relative to the normal order of the reference
293 * IDCT)
294 * this permutation must be performed before the idct_put/add, note, normally this can be merged
295 * with the zigzag/alternate scan<br>
24641185
MN
296 * an example to avoid confusion:
297 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
298 * - (x -> referece dct -> reference idct -> x)
299 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
300 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
301 */
b0368839
MN
302 uint8_t idct_permutation[64];
303 int idct_permutation_type;
304#define FF_NO_IDCT_PERM 1
305#define FF_LIBMPEG2_IDCT_PERM 2
306#define FF_SIMPLE_IDCT_PERM 3
307#define FF_TRANSPOSE_IDCT_PERM 4
308
364a1797
MN
309 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
310 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
311#define BASIS_SHIFT 16
312#define RECON_SHIFT 6
313
44cb64ee
MM
314 /**
315 * This function handles any initialization for the VP3 DSP functions.
316 */
317 void (*vp3_dsp_init)(void);
318
319 /**
320 * This function is responsible for taking a block of zigzag'd,
321 * quantized DCT coefficients, reconstructing the original block of
322 * samples, and placing it into the output.
323 * @param input_data 64 zigzag'd, quantized DCT coefficients
324 * @param dequant_matrix 64 zigzag'd quantizer coefficients
325 * @param coeff_count index of the last coefficient
326 * @param dest the final output location where the transformed samples
327 * are to be placed
328 * @param stride the width in 8-bit samples of a line on this plane
329 */
330 void (*vp3_idct_put)(int16_t *input_data, int16_t *dequant_matrix,
331 int coeff_count, uint8_t *dest, int stride);
332
333 /**
334 * This function is responsible for taking a block of zigzag'd,
335 * quantized DCT coefficients, reconstructing the original block of
336 * samples, and adding the transformed samples to an existing block of
337 * samples in the output.
338 * @param input_data 64 zigzag'd, quantized DCT coefficients
339 * @param dequant_matrix 64 zigzag'd quantizer coefficients
340 * @param coeff_count index of the last coefficient
341 * @param dest the final output location where the transformed samples
342 * are to be placed
343 * @param stride the width in 8-bit samples of a line on this plane
344 */
345 void (*vp3_idct_add)(int16_t *input_data, int16_t *dequant_matrix,
346 int coeff_count, uint8_t *dest, int stride);
347
eb4b3dd3
ZK
348} DSPContext;
349
59cf08ce 350void dsputil_static_init(void);
b0368839 351void dsputil_init(DSPContext* p, AVCodecContext *avctx);
de6d9b64 352
7801d21d
MN
353/**
354 * permute block according to permuatation.
355 * @param last last non zero element in scantable order
356 */
0c1a9eda 357void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
e0eac44e 358
622348f9
MN
359void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
360
d8085ea7
MN
361#define BYTE_VEC32(c) ((c)*0x01010101UL)
362
363static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
364{
365 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
366}
367
368static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
369{
370 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
371}
372
24641185 373/**
77c92c2d 374 * Empty mmx state.
24641185
MN
375 * this must be called between any dsp function and float/double code.
376 * for example sin(); dsp->idct_put(); emms_c(); cos()
377 */
eb4b3dd3
ZK
378#define emms_c()
379
e629ab68
RD
380/* should be defined by architectures supporting
381 one or more MultiMedia extension */
382int mm_support(void);
383
92a69cf8
MM
384#define __align16 __attribute__ ((aligned (16)))
385
3d03c0a2 386#if defined(HAVE_MMX)
de6d9b64 387
18f77016 388#undef emms_c
eb4b3dd3 389
de6d9b64
FB
390#define MM_MMX 0x0001 /* standard MMX */
391#define MM_3DNOW 0x0004 /* AMD 3DNOW */
392#define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
393#define MM_SSE 0x0008 /* SSE functions */
394#define MM_SSE2 0x0010 /* PIV SSE2 functions */
395
396extern int mm_flags;
397
0c1a9eda
ZK
398void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
399void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
de6d9b64
FB
400
401static inline void emms(void)
402{
fb16b7e7
FB
403 __asm __volatile ("emms;":::"memory");
404}
405
1457ab52 406
fb16b7e7
FB
407#define emms_c() \
408{\
409 if (mm_flags & MM_MMX)\
410 emms();\
de6d9b64
FB
411}
412
413#define __align8 __attribute__ ((aligned (8)))
414
b0368839
MN
415void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
416void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
2720569a 417
3d03c0a2
FB
418#elif defined(ARCH_ARMV4L)
419
3d03c0a2 420/* This is to use 4 bytes read to the IDCT pointers for some 'zero'
92a69cf8 421 line optimizations */
3d03c0a2
FB
422#define __align8 __attribute__ ((aligned (4)))
423
b0368839 424void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
3d03c0a2 425
c34270f5 426#elif defined(HAVE_MLIB)
c34270f5
FB
427
428/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
429#define __align8 __attribute__ ((aligned (8)))
430
b0368839 431void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
c34270f5 432
44f54ceb
MN
433#elif defined(ARCH_SPARC)
434
435/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
436#define __align8 __attribute__ ((aligned (8)))
437void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
438
1e98dffb
NK
439#elif defined(ARCH_ALPHA)
440
1e98dffb
NK
441#define __align8 __attribute__ ((aligned (8)))
442
b0368839 443void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
1e98dffb 444
59925ef2
BF
445#elif defined(ARCH_POWERPC)
446
404d2241
BF
447#define MM_ALTIVEC 0x0001 /* standard AltiVec */
448
449extern int mm_flags;
450
3b991c54 451#if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN)
9a197a24 452#define pixel altivec_pixel
3b991c54 453#include <altivec.h>
9a197a24 454#undef pixel
3b991c54
RD
455#endif
456
59925ef2
BF
457#define __align8 __attribute__ ((aligned (16)))
458
b0368839 459void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
59925ef2 460
d46aba26
LS
461#elif defined(HAVE_MMI)
462
d46aba26
LS
463#define __align8 __attribute__ ((aligned (16)))
464
b0368839 465void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
d46aba26 466
0c6bd2ea
B
467#elif defined(ARCH_SH4)
468
469#define __align8 __attribute__ ((aligned (8)))
470
471void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
472
de6d9b64
FB
473#else
474
475#define __align8
476
477#endif
478
6d4985bb
FB
479#ifdef __GNUC__
480
481struct unaligned_64 { uint64_t l; } __attribute__((packed));
482struct unaligned_32 { uint32_t l; } __attribute__((packed));
669ac79c 483struct unaligned_16 { uint16_t l; } __attribute__((packed));
6d4985bb 484
669ac79c 485#define LD16(a) (((const struct unaligned_16 *) (a))->l)
6d4985bb
FB
486#define LD32(a) (((const struct unaligned_32 *) (a))->l)
487#define LD64(a) (((const struct unaligned_64 *) (a))->l)
488
489#define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)
490
491#else /* __GNUC__ */
492
669ac79c 493#define LD16(a) (*((uint16_t*)(a)))
6d4985bb
FB
494#define LD32(a) (*((uint32_t*)(a)))
495#define LD64(a) (*((uint64_t*)(a)))
496
497#define ST32(a, b) *((uint32_t*)(a)) = (b)
498
499#endif /* !__GNUC__ */
500
43f1708f 501/* PSNR */
0c1a9eda 502void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
43f1708f
J
503 int orig_linesize[3], int coded_linesize,
504 AVCodecContext *avctx);
bb6f5690
FB
505
506/* FFT computation */
507
508/* NOTE: soon integer code will be added, so you must use the
509 FFTSample type */
510typedef float FFTSample;
511
512typedef struct FFTComplex {
513 FFTSample re, im;
514} FFTComplex;
515
516typedef struct FFTContext {
517 int nbits;
518 int inverse;
519 uint16_t *revtab;
520 FFTComplex *exptab;
521 FFTComplex *exptab1; /* only used by SSE code */
522 void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
523} FFTContext;
524
68951ecf
GB
525int ff_fft_init(FFTContext *s, int nbits, int inverse);
526void ff_fft_permute(FFTContext *s, FFTComplex *z);
527void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
528void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
529void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
8d268a7d 530
68951ecf 531static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
bb6f5690
FB
532{
533 s->fft_calc(s, z);
534}
68951ecf 535void ff_fft_end(FFTContext *s);
bb6f5690
FB
536
537/* MDCT computation */
538
539typedef struct MDCTContext {
540 int n; /* size of MDCT (i.e. number of input data * 2) */
541 int nbits; /* n = 2^nbits */
542 /* pre/post rotation tables */
543 FFTSample *tcos;
544 FFTSample *tsin;
545 FFTContext fft;
546} MDCTContext;
547
82696bee 548int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
eb4b3dd3 549void ff_imdct_calc(MDCTContext *s, FFTSample *output,
bb6f5690 550 const FFTSample *input, FFTSample *tmp);
eb4b3dd3 551void ff_mdct_calc(MDCTContext *s, FFTSample *out,
bb6f5690 552 const FFTSample *input, FFTSample *tmp);
82696bee 553void ff_mdct_end(MDCTContext *s);
bb6f5690 554
bb198e19
MN
555#define WARPER8_16(name8, name16)\
556static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
557 return name8(s, dst , src , stride, h)\
558 +name8(s, dst+8 , src+8 , stride, h);\
559}
560
561#define WARPER8_16_SQ(name8, name16)\
562static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
563 int score=0;\
564 score +=name8(s, dst , src , stride, 8);\
565 score +=name8(s, dst+8 , src+8 , stride, 8);\
566 if(h==16){\
567 dst += 8*stride;\
568 src += 8*stride;\
569 score +=name8(s, dst , src , stride, 8);\
570 score +=name8(s, dst+8 , src+8 , stride, 8);\
571 }\
572 return score;\
1457ab52
MN
573}
574
95e2ce4a 575#ifndef HAVE_LRINTF
9d85cbd9
FB
576/* XXX: add ISOC specific test to avoid specific BSD testing. */
577/* better than nothing implementation. */
6234d753 578/* btw, rintf() is existing on fbsd too -- alex */
9d85cbd9
FB
579static inline long int lrintf(float x)
580{
ea937d01
FB
581#ifdef CONFIG_WIN32
582 /* XXX: incorrect, but make it compile */
583 return (int)(x);
584#else
9d85cbd9 585 return (int)(rint(x));
ea937d01 586#endif
9d85cbd9
FB
587}
588#endif
589
de6d9b64 590#endif