remove mpegvideo.c img resample dependancy
[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
5509bffa 18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
ff4ec49e 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;
059715a4 38typedef int DWTELEM;
de6d9b64 39
03c94ede 40void fdct_ifast (DCTELEM *data);
48b1f800 41void fdct_ifast248 (DCTELEM *data);
28db7fce 42void ff_jpeg_fdct_islow (DCTELEM *data);
10acc479 43void ff_fdct248_islow (DCTELEM *data);
de6d9b64
FB
44
45void j_rev_dct (DCTELEM *data);
178fcca8 46void j_rev_dct4 (DCTELEM *data);
9ca358b9 47void j_rev_dct2 (DCTELEM *data);
1aa8c57b 48void j_rev_dct1 (DCTELEM *data);
de6d9b64 49
3f09f52a 50void ff_fdct_mmx(DCTELEM *block);
cf3bf5bb 51void ff_fdct_mmx2(DCTELEM *block);
8fd19ab2 52void ff_fdct_sse2(DCTELEM *block);
de6d9b64 53
43efd19a 54void ff_h264_idct8_add_c(uint8_t *dst, DCTELEM *block, int stride);
0fa8158d 55void ff_h264_idct_add_c(uint8_t *dst, DCTELEM *block, int stride);
ef9d1d15
LM
56void ff_h264_idct8_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
57void ff_h264_idct_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
0fa8158d
MN
58void ff_h264_lowres_idct_add_c(uint8_t *dst, int stride, DCTELEM *block);
59void ff_h264_lowres_idct_put_c(uint8_t *dst, int stride, DCTELEM *block);
60
e0eac44e 61/* encoding scans */
0c1a9eda
ZK
62extern const uint8_t ff_alternate_horizontal_scan[64];
63extern const uint8_t ff_alternate_vertical_scan[64];
64extern const uint8_t ff_zigzag_direct[64];
10acc479 65extern const uint8_t ff_zigzag248_direct[64];
5a240838 66
de6d9b64 67/* pixel operations */
f2e92ef2 68#define MAX_NEG_CROP 1024
de6d9b64
FB
69
70/* temporary */
0c1a9eda
ZK
71extern uint32_t squareTbl[512];
72extern uint8_t cropTbl[256 + 2 * MAX_NEG_CROP];
de6d9b64 73
44cb64ee 74/* VP3 DSP functions */
8b6103da
MN
75void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
76void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
77void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
de6d9b64 78
54009d42
MN
79/* 1/2^n downscaling functions from imgconvert.c */
80void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
81void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
82void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
83void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
84
b7c27ee6 85/* minimum alignment rules ;)
eb4b3dd3 86if u notice errors in the align stuff, need more alignment for some asm code for some cpu
b7c27ee6
MN
87or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ...
88
89!warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible)
90i (michael) didnt check them, these are just the alignents which i think could be reached easily ...
de6d9b64 91
b7c27ee6
MN
92!future video codecs might need functions with less strict alignment
93*/
94
eb4b3dd3 95/*
0c1a9eda
ZK
96void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
97void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
98void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
99void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
649c00c9 100void clear_blocks_c(DCTELEM *blocks);
eb4b3dd3 101*/
de6d9b64
FB
102
103/* add and put pixel (decoding) */
b7c27ee6 104// blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
7d67aa9b 105//h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
0c1a9eda 106typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
669ac79c 107typedef 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 108typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
0da71265 109typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
9f2d1b4f 110typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset);
e8b56208 111typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset);
b3184779 112
db794953 113#define DEF_OLD_QPEL(name)\
0c1a9eda
ZK
114void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
115void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
116void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
db794953
MN
117
118DEF_OLD_QPEL(qpel16_mc11_old_c)
119DEF_OLD_QPEL(qpel16_mc31_old_c)
120DEF_OLD_QPEL(qpel16_mc12_old_c)
121DEF_OLD_QPEL(qpel16_mc32_old_c)
122DEF_OLD_QPEL(qpel16_mc13_old_c)
123DEF_OLD_QPEL(qpel16_mc33_old_c)
124DEF_OLD_QPEL(qpel8_mc11_old_c)
125DEF_OLD_QPEL(qpel8_mc31_old_c)
126DEF_OLD_QPEL(qpel8_mc12_old_c)
127DEF_OLD_QPEL(qpel8_mc32_old_c)
128DEF_OLD_QPEL(qpel8_mc13_old_c)
129DEF_OLD_QPEL(qpel8_mc33_old_c)
b3184779
MN
130
131#define CALL_2X_PIXELS(a, b, n)\
132static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
133 b(block , pixels , line_size, h);\
134 b(block+n, pixels+n, line_size, h);\
135}
44eb4951 136
de6d9b64 137/* motion estimation */
7d67aa9b
MN
138// h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
139// allthough currently h<4 is not used as functions with width <8 are not used and neither implemented
bb198e19 140typedef 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 141
0da71265 142
059715a4
RE
143// for snow slices
144typedef struct slice_buffer_s slice_buffer;
145
24641185
MN
146/**
147 * DSPContext.
148 */
eb4b3dd3
ZK
149typedef struct DSPContext {
150 /* pixel ops : interface with DCT */
0c1a9eda
ZK
151 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
152 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
153 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
f9ed9d85 154 void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
0c1a9eda 155 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
36940eca
LM
156 void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
157 void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
d518aebd
MN
158 /**
159 * translational global motion compensation.
160 */
0c1a9eda 161 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
d518aebd
MN
162 /**
163 * global motion compensation.
164 */
0c1a9eda 165 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
bb270c08 166 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
eb4b3dd3 167 void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
0c1a9eda
ZK
168 int (*pix_sum)(uint8_t * pix, int line_size);
169 int (*pix_norm1)(uint8_t * pix, int line_size);
bb198e19 170// 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
115329f1 171
622348f9
MN
172 me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */
173 me_cmp_func sse[5];
174 me_cmp_func hadamard8_diff[5];
175 me_cmp_func dct_sad[5];
176 me_cmp_func quant_psnr[5];
177 me_cmp_func bit[5];
178 me_cmp_func rd[5];
179 me_cmp_func vsad[5];
180 me_cmp_func vsse[5];
e6a2ac34 181 me_cmp_func nsse[5];
26efc54e
MN
182 me_cmp_func w53[5];
183 me_cmp_func w97[5];
0fd6aea1 184 me_cmp_func dct_max[5];
27c61ac5 185 me_cmp_func dct264_sad[5];
1457ab52 186
bb198e19
MN
187 me_cmp_func me_pre_cmp[5];
188 me_cmp_func me_cmp[5];
189 me_cmp_func me_sub_cmp[5];
190 me_cmp_func mb_cmp[5];
622348f9 191 me_cmp_func ildct_cmp[5]; //only width 16 used
0fd6aea1 192 me_cmp_func frame_skip_cmp[5]; //only width 8 used
eb4b3dd3 193
d518aebd
MN
194 /**
195 * Halfpel motion compensation with rounding (a+b+1)>>1.
115329f1 196 * this is an array[4][4] of motion compensation funcions for 4
e5771f4f 197 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
5755c27f 198 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
d518aebd
MN
199 * @param block destination where the result is stored
200 * @param pixels source
201 * @param line_size number of bytes in a horizontal line of block
202 * @param h height
203 */
669ac79c 204 op_pixels_func put_pixels_tab[4][4];
d518aebd
MN
205
206 /**
207 * Halfpel motion compensation with rounding (a+b+1)>>1.
115329f1 208 * This is an array[4][4] of motion compensation functions for 4
e5771f4f 209 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
5755c27f 210 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
d518aebd
MN
211 * @param block destination into which the result is averaged (a+b+1)>>1
212 * @param pixels source
213 * @param line_size number of bytes in a horizontal line of block
214 * @param h height
215 */
da3b9756 216 op_pixels_func avg_pixels_tab[4][4];
d518aebd
MN
217
218 /**
219 * Halfpel motion compensation with no rounding (a+b)>>1.
115329f1 220 * this is an array[2][4] of motion compensation funcions for 2
eb14c713 221 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
5755c27f 222 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
d518aebd
MN
223 * @param block destination where the result is stored
224 * @param pixels source
225 * @param line_size number of bytes in a horizontal line of block
226 * @param h height
227 */
dbc56b39 228 op_pixels_func put_no_rnd_pixels_tab[4][4];
d518aebd
MN
229
230 /**
231 * Halfpel motion compensation with no rounding (a+b)>>1.
115329f1 232 * this is an array[2][4] of motion compensation funcions for 2
eb14c713 233 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
5755c27f 234 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
d518aebd
MN
235 * @param block destination into which the result is averaged (a+b)>>1
236 * @param pixels source
237 * @param line_size number of bytes in a horizontal line of block
238 * @param h height
239 */
dbc56b39 240 op_pixels_func avg_no_rnd_pixels_tab[4][4];
115329f1 241
c0a0170c 242 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);
115329f1 243
669ac79c
MN
244 /**
245 * Thirdpel motion compensation with rounding (a+b+1)>>1.
246 * this is an array[12] of motion compensation funcions for the 9 thirdpel positions<br>
247 * *pixels_tab[ xthirdpel + 4*ythirdpel ]
248 * @param block destination where the result is stored
249 * @param pixels source
250 * @param line_size number of bytes in a horizontal line of block
251 * @param h height
252 */
253 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
da3b9756
MM
254 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
255
eb4b3dd3
ZK
256 qpel_mc_func put_qpel_pixels_tab[2][16];
257 qpel_mc_func avg_qpel_pixels_tab[2][16];
258 qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
259 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
1457ab52 260 qpel_mc_func put_mspel_pixels_tab[8];
115329f1 261
0da71265
MN
262 /**
263 * h264 Chram MC
264 */
265 h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
266 h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
eb4b3dd3 267
80e44bc3
MN
268 qpel_mc_func put_h264_qpel_pixels_tab[4][16];
269 qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
115329f1 270
9f2d1b4f
LM
271 h264_weight_func weight_h264_pixels_tab[10];
272 h264_biweight_func biweight_h264_pixels_tab[10];
115329f1 273
bb198e19 274 me_cmp_func pix_abs[2][4];
115329f1 275
11f18faf 276 /* huffyuv specific */
11f18faf 277 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
1457ab52 278 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
84705403
MN
279 /**
280 * subtract huffyuv's variant of median prediction
281 * note, this might read from src1[-1], src2[-1]
282 */
283 void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
3d2e8cce 284 void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w);
42251a2a 285
5cf08f23
LM
286 void (*h264_v_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
287 void (*h264_h_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
288 void (*h264_v_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
289 void (*h264_h_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
290 void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
291 void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
115329f1 292
332f9ac4
MN
293 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
294 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
295
fdbbf2e0 296 void (*h261_loop_filter)(uint8_t *src, int stride);
c6148de2 297
b0368839
MN
298 /* (I)DCT */
299 void (*fdct)(DCTELEM *block/* align 16*/);
10acc479 300 void (*fdct248)(DCTELEM *block/* align 16*/);
115329f1 301
4fb518c3
MN
302 /* IDCT really*/
303 void (*idct)(DCTELEM *block/* align 16*/);
115329f1 304
24641185 305 /**
77c92c2d 306 * block -> idct -> clip to unsigned 8 bit -> dest.
24641185 307 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
a1adf436 308 * @param line_size size in bytes of a horizotal line of dest
24641185 309 */
b0368839 310 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
115329f1 311
24641185
MN
312 /**
313 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
a1adf436 314 * @param line_size size in bytes of a horizotal line of dest
24641185 315 */
b0368839 316 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
115329f1 317
24641185 318 /**
77c92c2d 319 * idct input permutation.
05493021
MN
320 * several optimized IDCTs need a permutated input (relative to the normal order of the reference
321 * IDCT)
322 * this permutation must be performed before the idct_put/add, note, normally this can be merged
323 * with the zigzag/alternate scan<br>
24641185
MN
324 * an example to avoid confusion:
325 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
326 * - (x -> referece dct -> reference idct -> x)
327 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
328 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
329 */
b0368839
MN
330 uint8_t idct_permutation[64];
331 int idct_permutation_type;
332#define FF_NO_IDCT_PERM 1
333#define FF_LIBMPEG2_IDCT_PERM 2
334#define FF_SIMPLE_IDCT_PERM 3
335#define FF_TRANSPOSE_IDCT_PERM 4
5773a746 336#define FF_PARTTRANS_IDCT_PERM 5
b0368839 337
364a1797
MN
338 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
339 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
340#define BASIS_SHIFT 16
341#define RECON_SHIFT 6
115329f1 342
0fa8158d 343 void (*h264_idct_add)(uint8_t *dst, DCTELEM *block, int stride);
43efd19a 344 void (*h264_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
ef9d1d15
LM
345 void (*h264_idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
346 void (*h264_idct8_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
059715a4
RE
347
348 /* snow wavelet */
349 void (*vertical_compose97i)(DWTELEM *b0, DWTELEM *b1, DWTELEM *b2, DWTELEM *b3, DWTELEM *b4, DWTELEM *b5, int width);
350 void (*horizontal_compose97i)(DWTELEM *b, int width);
351 void (*inner_add_yblock)(uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h, int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8);
513fbd8e
LM
352
353 void (*prefetch)(void *mem, int stride, int h);
54009d42
MN
354
355 void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
eb4b3dd3
ZK
356} DSPContext;
357
59cf08ce 358void dsputil_static_init(void);
b0368839 359void dsputil_init(DSPContext* p, AVCodecContext *avctx);
de6d9b64 360
7801d21d
MN
361/**
362 * permute block according to permuatation.
363 * @param last last non zero element in scantable order
364 */
0c1a9eda 365void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
e0eac44e 366
622348f9
MN
367void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
368
bb270c08 369#define BYTE_VEC32(c) ((c)*0x01010101UL)
d8085ea7
MN
370
371static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
372{
373 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
374}
375
376static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
377{
378 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
379}
380
26efc54e
MN
381static inline int get_penalty_factor(int lambda, int lambda2, int type){
382 switch(type&0xFF){
383 default:
384 case FF_CMP_SAD:
385 return lambda>>FF_LAMBDA_SHIFT;
386 case FF_CMP_DCT:
387 return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
388 case FF_CMP_W53:
389 return (4*lambda)>>(FF_LAMBDA_SHIFT);
390 case FF_CMP_W97:
391 return (2*lambda)>>(FF_LAMBDA_SHIFT);
392 case FF_CMP_SATD:
27c61ac5 393 case FF_CMP_DCT264:
26efc54e
MN
394 return (2*lambda)>>FF_LAMBDA_SHIFT;
395 case FF_CMP_RD:
396 case FF_CMP_PSNR:
397 case FF_CMP_SSE:
398 case FF_CMP_NSSE:
399 return lambda2>>FF_LAMBDA_SHIFT;
400 case FF_CMP_BIT:
401 return 1;
402 }
403}
404
24641185 405/**
77c92c2d 406 * Empty mmx state.
24641185
MN
407 * this must be called between any dsp function and float/double code.
408 * for example sin(); dsp->idct_put(); emms_c(); cos()
409 */
eb4b3dd3
ZK
410#define emms_c()
411
e629ab68
RD
412/* should be defined by architectures supporting
413 one or more MultiMedia extension */
414int mm_support(void);
415
68b51e58
SH
416#ifdef __GNUC__
417 #define DECLARE_ALIGNED_16(t,v) t v __attribute__ ((aligned (16)))
418#else
419 #define DECLARE_ALIGNED_16(t,v) __declspec(align(16)) t v
420#endif
92a69cf8 421
3d03c0a2 422#if defined(HAVE_MMX)
de6d9b64 423
18f77016 424#undef emms_c
eb4b3dd3 425
de6d9b64
FB
426#define MM_MMX 0x0001 /* standard MMX */
427#define MM_3DNOW 0x0004 /* AMD 3DNOW */
428#define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
429#define MM_SSE 0x0008 /* SSE functions */
430#define MM_SSE2 0x0010 /* PIV SSE2 functions */
e42a152b 431#define MM_3DNOWEXT 0x0020 /* AMD 3DNowExt */
de6d9b64
FB
432
433extern int mm_flags;
434
0c1a9eda
ZK
435void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
436void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
f9ed9d85 437void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
de6d9b64
FB
438
439static inline void emms(void)
440{
fb16b7e7
FB
441 __asm __volatile ("emms;":::"memory");
442}
443
1457ab52 444
fb16b7e7
FB
445#define emms_c() \
446{\
447 if (mm_flags & MM_MMX)\
448 emms();\
de6d9b64
FB
449}
450
68b51e58
SH
451#ifdef __GNUC__
452 #define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8)))
453#else
454 #define DECLARE_ALIGNED_8(t,v) __declspec(align(8)) t v
455#endif
456
3237f731 457#define STRIDE_ALIGN 8
de6d9b64 458
b0368839
MN
459void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
460void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
2720569a 461
3d03c0a2
FB
462#elif defined(ARCH_ARMV4L)
463
3d03c0a2 464/* This is to use 4 bytes read to the IDCT pointers for some 'zero'
92a69cf8 465 line optimizations */
68b51e58 466#define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (4)))
3237f731 467#define STRIDE_ALIGN 4
3d03c0a2 468
eba9ae3c
GB
469#define MM_IWMMXT 0x0100 /* XScale IWMMXT */
470
471extern int mm_flags;
472
b0368839 473void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
3d03c0a2 474
c34270f5 475#elif defined(HAVE_MLIB)
c34270f5
FB
476
477/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
68b51e58 478#define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8)))
3237f731 479#define STRIDE_ALIGN 8
c34270f5 480
b0368839 481void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
c34270f5 482
44f54ceb
MN
483#elif defined(ARCH_SPARC)
484
485/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
68b51e58 486#define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8)))
3237f731 487#define STRIDE_ALIGN 8
44f54ceb
MN
488void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
489
1e98dffb
NK
490#elif defined(ARCH_ALPHA)
491
68b51e58 492#define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8)))
3237f731 493#define STRIDE_ALIGN 8
1e98dffb 494
b0368839 495void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
1e98dffb 496
59925ef2
BF
497#elif defined(ARCH_POWERPC)
498
404d2241
BF
499#define MM_ALTIVEC 0x0001 /* standard AltiVec */
500
501extern int mm_flags;
502
3b991c54 503#if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN)
9a197a24 504#define pixel altivec_pixel
3b991c54 505#include <altivec.h>
9a197a24 506#undef pixel
3b991c54
RD
507#endif
508
68b51e58 509#define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (16)))
3237f731 510#define STRIDE_ALIGN 16
59925ef2 511
b0368839 512void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
59925ef2 513
d46aba26
LS
514#elif defined(HAVE_MMI)
515
68b51e58 516#define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (16)))
3237f731 517#define STRIDE_ALIGN 16
d46aba26 518
b0368839 519void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
d46aba26 520
0c6bd2ea
B
521#elif defined(ARCH_SH4)
522
68b51e58 523#define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8)))
3237f731 524#define STRIDE_ALIGN 8
0c6bd2ea
B
525
526void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
527
de6d9b64
FB
528#else
529
68b51e58 530#define DECLARE_ALIGNED_8(t,v) t v __attribute__ ((aligned (8)))
3237f731 531#define STRIDE_ALIGN 8
de6d9b64
FB
532
533#endif
534
6d4985bb
FB
535#ifdef __GNUC__
536
537struct unaligned_64 { uint64_t l; } __attribute__((packed));
538struct unaligned_32 { uint32_t l; } __attribute__((packed));
669ac79c 539struct unaligned_16 { uint16_t l; } __attribute__((packed));
6d4985bb 540
669ac79c 541#define LD16(a) (((const struct unaligned_16 *) (a))->l)
6d4985bb
FB
542#define LD32(a) (((const struct unaligned_32 *) (a))->l)
543#define LD64(a) (((const struct unaligned_64 *) (a))->l)
544
80e44bc3 545#define ST16(a, b) (((struct unaligned_16 *) (a))->l) = (b)
6d4985bb
FB
546#define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)
547
548#else /* __GNUC__ */
549
669ac79c 550#define LD16(a) (*((uint16_t*)(a)))
6d4985bb
FB
551#define LD32(a) (*((uint32_t*)(a)))
552#define LD64(a) (*((uint64_t*)(a)))
553
dc2bf20e 554#define ST16(a, b) *((uint16_t*)(a)) = (b)
6d4985bb
FB
555#define ST32(a, b) *((uint32_t*)(a)) = (b)
556
557#endif /* !__GNUC__ */
558
43f1708f 559/* PSNR */
0c1a9eda 560void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
43f1708f
J
561 int orig_linesize[3], int coded_linesize,
562 AVCodecContext *avctx);
bb6f5690
FB
563
564/* FFT computation */
565
566/* NOTE: soon integer code will be added, so you must use the
567 FFTSample type */
568typedef float FFTSample;
569
570typedef struct FFTComplex {
571 FFTSample re, im;
572} FFTComplex;
573
574typedef struct FFTContext {
575 int nbits;
576 int inverse;
577 uint16_t *revtab;
578 FFTComplex *exptab;
579 FFTComplex *exptab1; /* only used by SSE code */
580 void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
581} FFTContext;
582
68951ecf
GB
583int ff_fft_init(FFTContext *s, int nbits, int inverse);
584void ff_fft_permute(FFTContext *s, FFTComplex *z);
585void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
586void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
82eb4b0f
ZM
587void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z);
588void ff_fft_calc_3dn2(FFTContext *s, FFTComplex *z);
68951ecf 589void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
8d268a7d 590
68951ecf 591static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
bb6f5690
FB
592{
593 s->fft_calc(s, z);
594}
68951ecf 595void ff_fft_end(FFTContext *s);
bb6f5690
FB
596
597/* MDCT computation */
598
599typedef struct MDCTContext {
600 int n; /* size of MDCT (i.e. number of input data * 2) */
601 int nbits; /* n = 2^nbits */
602 /* pre/post rotation tables */
603 FFTSample *tcos;
604 FFTSample *tsin;
605 FFTContext fft;
606} MDCTContext;
607
82696bee 608int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
eb4b3dd3 609void ff_imdct_calc(MDCTContext *s, FFTSample *output,
bb6f5690 610 const FFTSample *input, FFTSample *tmp);
eb4b3dd3 611void ff_mdct_calc(MDCTContext *s, FFTSample *out,
bb6f5690 612 const FFTSample *input, FFTSample *tmp);
82696bee 613void ff_mdct_end(MDCTContext *s);
bb6f5690 614
bb198e19
MN
615#define WARPER8_16(name8, name16)\
616static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
617 return name8(s, dst , src , stride, h)\
618 +name8(s, dst+8 , src+8 , stride, h);\
619}
620
621#define WARPER8_16_SQ(name8, name16)\
622static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
623 int score=0;\
624 score +=name8(s, dst , src , stride, 8);\
625 score +=name8(s, dst+8 , src+8 , stride, 8);\
626 if(h==16){\
627 dst += 8*stride;\
628 src += 8*stride;\
629 score +=name8(s, dst , src , stride, 8);\
630 score +=name8(s, dst+8 , src+8 , stride, 8);\
631 }\
632 return score;\
1457ab52
MN
633}
634
de6d9b64 635#endif