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