Statistics
| Branch: | Revision:

ffmpeg / libavcodec / dsputil.h @ e42a152b

History | View | Annotate | Download (21.3 KB)

1 ff4ec49e Fabrice Bellard
/*
2
 * DSP utils
3
 * Copyright (c) 2000, 2001, 2002 Fabrice Bellard.
4 8f2ab833 Michael Niedermayer
 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
5 ff4ec49e Fabrice Bellard
 *
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 24641185 Michael Niedermayer
21
/**
22
 * @file dsputil.h
23 983e3246 Michael Niedermayer
 * DSP utils.
24 5755c27f Michael Niedermayer
 * 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 24641185 Michael Niedermayer
 */
27
28 de6d9b64 Fabrice Bellard
#ifndef DSPUTIL_H
29
#define DSPUTIL_H
30
31
#include "common.h"
32 43f1708f Juanjo
#include "avcodec.h"
33 de6d9b64 Fabrice Bellard
34 24641185 Michael Niedermayer
35 44eb4951 Michael Niedermayer
//#define DEBUG
36 de6d9b64 Fabrice Bellard
/* dct code */
37
typedef short DCTELEM;
38
39 03c94ede Fabrice Bellard
void fdct_ifast (DCTELEM *data);
40 48b1f800 Roman Shaposhnik
void fdct_ifast248 (DCTELEM *data);
41 28db7fce Michael Niedermayer
void ff_jpeg_fdct_islow (DCTELEM *data);
42 10acc479 Roman Shaposhnik
void ff_fdct248_islow (DCTELEM *data);
43 de6d9b64 Fabrice Bellard
44
void j_rev_dct (DCTELEM *data);
45 178fcca8 Michael Niedermayer
void j_rev_dct4 (DCTELEM *data);
46 9ca358b9 Michael Niedermayer
void j_rev_dct2 (DCTELEM *data);
47 1aa8c57b Michael Niedermayer
void j_rev_dct1 (DCTELEM *data);
48 de6d9b64 Fabrice Bellard
49 3f09f52a Arpi
void ff_fdct_mmx(DCTELEM *block);
50 cf3bf5bb Michael Niedermayer
void ff_fdct_mmx2(DCTELEM *block);
51 8fd19ab2 Michael Niedermayer
void ff_fdct_sse2(DCTELEM *block);
52 de6d9b64 Fabrice Bellard
53 0fa8158d Michael Niedermayer
void ff_h264_idct_add_c(uint8_t *dst, DCTELEM *block, int stride);
54
void ff_h264_lowres_idct_add_c(uint8_t *dst, int stride, DCTELEM *block);
55
void ff_h264_lowres_idct_put_c(uint8_t *dst, int stride, DCTELEM *block);
56
57 e0eac44e Fabrice Bellard
/* encoding scans */
58 0c1a9eda Zdenek Kabelac
extern const uint8_t ff_alternate_horizontal_scan[64];
59
extern const uint8_t ff_alternate_vertical_scan[64];
60
extern const uint8_t ff_zigzag_direct[64];
61 10acc479 Roman Shaposhnik
extern const uint8_t ff_zigzag248_direct[64];
62 5a240838 Michael Niedermayer
63 de6d9b64 Fabrice Bellard
/* pixel operations */
64 f2e92ef2 Michael Niedermayer
#define MAX_NEG_CROP 1024
65 de6d9b64 Fabrice Bellard
66
/* temporary */
67 0c1a9eda Zdenek Kabelac
extern uint32_t squareTbl[512];
68
extern uint8_t cropTbl[256 + 2 * MAX_NEG_CROP];
69 de6d9b64 Fabrice Bellard
70 44cb64ee Mike Melanson
/* VP3 DSP functions */
71
void vp3_dsp_init_c(void);
72 116824d0 Mike Melanson
void vp3_idct_c(int16_t *input_data, int16_t *dequant_matrix,
73
    int coeff_count, DCTELEM *output_data);
74 44cb64ee Mike Melanson
75
void vp3_dsp_init_mmx(void);
76 116824d0 Mike Melanson
void vp3_idct_mmx(int16_t *input_data, int16_t *dequant_matrix,
77
    int coeff_count, DCTELEM *output_data);
78 44cb64ee Mike Melanson
79 38acbc3c Mike Melanson
void vp3_dsp_init_sse2(void);
80 116824d0 Mike Melanson
void vp3_idct_sse2(int16_t *input_data, int16_t *dequant_matrix,
81
    int coeff_count, DCTELEM *output_data);
82 de6d9b64 Fabrice Bellard
83 b7c27ee6 Michael Niedermayer
/* minimum alignment rules ;)
84 eb4b3dd3 Zdenek Kabelac
if u notice errors in the align stuff, need more alignment for some asm code for some cpu
85 b7c27ee6 Michael Niedermayer
or 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)
88
i (michael) didnt check them, these are just the alignents which i think could be reached easily ...
89 de6d9b64 Fabrice Bellard

90 b7c27ee6 Michael Niedermayer
!future video codecs might need functions with less strict alignment
91
*/
92
93 eb4b3dd3 Zdenek Kabelac
/*
94 0c1a9eda Zdenek Kabelac
void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
95
void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
96
void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
97
void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
98 649c00c9 Michael Niedermayer
void clear_blocks_c(DCTELEM *blocks);
99 eb4b3dd3 Zdenek Kabelac
*/
100 de6d9b64 Fabrice Bellard
101
/* add and put pixel (decoding) */
102 b7c27ee6 Michael Niedermayer
// blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
103 7d67aa9b Michael Niedermayer
//h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
104 0c1a9eda Zdenek Kabelac
typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
105 669ac79c Michael Niedermayer
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);
106 0c1a9eda Zdenek Kabelac
typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
107 0da71265 Michael Niedermayer
typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
108 b3184779 Michael Niedermayer
109 db794953 Michael Niedermayer
#define DEF_OLD_QPEL(name)\
110 0c1a9eda Zdenek Kabelac
void ff_put_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
111
void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
112
void ff_avg_        ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
113 db794953 Michael Niedermayer
114
DEF_OLD_QPEL(qpel16_mc11_old_c)
115
DEF_OLD_QPEL(qpel16_mc31_old_c)
116
DEF_OLD_QPEL(qpel16_mc12_old_c)
117
DEF_OLD_QPEL(qpel16_mc32_old_c)
118
DEF_OLD_QPEL(qpel16_mc13_old_c)
119
DEF_OLD_QPEL(qpel16_mc33_old_c)
120
DEF_OLD_QPEL(qpel8_mc11_old_c)
121
DEF_OLD_QPEL(qpel8_mc31_old_c)
122
DEF_OLD_QPEL(qpel8_mc12_old_c)
123
DEF_OLD_QPEL(qpel8_mc32_old_c)
124
DEF_OLD_QPEL(qpel8_mc13_old_c)
125
DEF_OLD_QPEL(qpel8_mc33_old_c)
126 b3184779 Michael Niedermayer
127
#define CALL_2X_PIXELS(a, b, n)\
128
static 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
}
132 44eb4951 Michael Niedermayer
133 de6d9b64 Fabrice Bellard
/* motion estimation */
134 7d67aa9b Michael Niedermayer
// 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
136 bb198e19 Michael Niedermayer
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))*/;
137 1457ab52 Michael Niedermayer
138 0da71265 Michael Niedermayer
139 24641185 Michael Niedermayer
/**
140
 * DSPContext.
141
 */
142 eb4b3dd3 Zdenek Kabelac
typedef struct DSPContext {
143
    /* pixel ops : interface with DCT */
144 0c1a9eda Zdenek Kabelac
    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 f9ed9d85 Mike Melanson
    void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
148 0c1a9eda Zdenek Kabelac
    void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
149 d518aebd Michael Niedermayer
    /**
150
     * translational global motion compensation.
151
     */
152 0c1a9eda Zdenek Kabelac
    void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
153 d518aebd Michael Niedermayer
    /**
154
     * global motion compensation.
155
     */
156 0c1a9eda Zdenek Kabelac
    void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
157 eb4b3dd3 Zdenek Kabelac
                    int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
158
    void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
159 0c1a9eda Zdenek Kabelac
    int (*pix_sum)(uint8_t * pix, int line_size);
160
    int (*pix_norm1)(uint8_t * pix, int line_size);
161 bb198e19 Michael Niedermayer
// 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
162
    
163 622348f9 Michael Niedermayer
    me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */
164
    me_cmp_func sse[5];
165
    me_cmp_func hadamard8_diff[5];
166
    me_cmp_func dct_sad[5];
167
    me_cmp_func quant_psnr[5];
168
    me_cmp_func bit[5];
169
    me_cmp_func rd[5];
170
    me_cmp_func vsad[5];
171
    me_cmp_func vsse[5];
172 e6a2ac34 Michael Niedermayer
    me_cmp_func nsse[5];
173 26efc54e Michael Niedermayer
    me_cmp_func w53[5];
174
    me_cmp_func w97[5];
175 0fd6aea1 Michael Niedermayer
    me_cmp_func dct_max[5];
176 1457ab52 Michael Niedermayer
177 bb198e19 Michael Niedermayer
    me_cmp_func me_pre_cmp[5];
178
    me_cmp_func me_cmp[5];
179
    me_cmp_func me_sub_cmp[5];
180
    me_cmp_func mb_cmp[5];
181 622348f9 Michael Niedermayer
    me_cmp_func ildct_cmp[5]; //only width 16 used
182 0fd6aea1 Michael Niedermayer
    me_cmp_func frame_skip_cmp[5]; //only width 8 used
183 eb4b3dd3 Zdenek Kabelac
184 d518aebd Michael Niedermayer
    /**
185
     * Halfpel motion compensation with rounding (a+b+1)>>1.
186 669ac79c Michael Niedermayer
     * this is an array[4][4] of motion compensation funcions for 4 
187 e5771f4f Michael Niedermayer
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
188 5755c27f Michael Niedermayer
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
189 d518aebd Michael Niedermayer
     * @param block destination where the result is stored
190
     * @param pixels source
191
     * @param line_size number of bytes in a horizontal line of block
192
     * @param h height
193
     */
194 669ac79c Michael Niedermayer
    op_pixels_func put_pixels_tab[4][4];
195 d518aebd Michael Niedermayer
196
    /**
197
     * Halfpel motion compensation with rounding (a+b+1)>>1.
198 4e8eed2f Mike Melanson
     * This is an array[4][4] of motion compensation functions for 4 
199 e5771f4f Michael Niedermayer
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
200 5755c27f Michael Niedermayer
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
201 d518aebd Michael Niedermayer
     * @param block destination into which the result is averaged (a+b+1)>>1
202
     * @param pixels source
203
     * @param line_size number of bytes in a horizontal line of block
204
     * @param h height
205
     */
206 da3b9756 Mike Melanson
    op_pixels_func avg_pixels_tab[4][4];
207 d518aebd Michael Niedermayer
208
    /**
209
     * Halfpel motion compensation with no rounding (a+b)>>1.
210 eb14c713 Michael Niedermayer
     * this is an array[2][4] of motion compensation funcions for 2 
211
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
212 5755c27f Michael Niedermayer
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
213 d518aebd Michael Niedermayer
     * @param block destination where the result is stored
214
     * @param pixels source
215
     * @param line_size number of bytes in a horizontal line of block
216
     * @param h height
217
     */
218 dbc56b39 Michael Niedermayer
    op_pixels_func put_no_rnd_pixels_tab[4][4];
219 d518aebd Michael Niedermayer
220
    /**
221
     * Halfpel motion compensation with no rounding (a+b)>>1.
222 eb14c713 Michael Niedermayer
     * this is an array[2][4] of motion compensation funcions for 2 
223
     * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
224 5755c27f Michael Niedermayer
     * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
225 d518aebd Michael Niedermayer
     * @param block destination into which the result is averaged (a+b)>>1
226
     * @param pixels source
227
     * @param line_size number of bytes in a horizontal line of block
228
     * @param h height
229
     */
230 dbc56b39 Michael Niedermayer
    op_pixels_func avg_no_rnd_pixels_tab[4][4];
231 669ac79c Michael Niedermayer
    
232 c0a0170c Michael Niedermayer
    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);
233
    
234 669ac79c Michael Niedermayer
    /**
235
     * Thirdpel motion compensation with rounding (a+b+1)>>1.
236
     * this is an array[12] of motion compensation funcions for the 9 thirdpel positions<br>
237
     * *pixels_tab[ xthirdpel + 4*ythirdpel ]
238
     * @param block destination where the result is stored
239
     * @param pixels source
240
     * @param line_size number of bytes in a horizontal line of block
241
     * @param h height
242
     */
243
    tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
244 da3b9756 Mike Melanson
    tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
245
246 eb4b3dd3 Zdenek Kabelac
    qpel_mc_func put_qpel_pixels_tab[2][16];
247
    qpel_mc_func avg_qpel_pixels_tab[2][16];
248
    qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
249
    qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
250 1457ab52 Michael Niedermayer
    qpel_mc_func put_mspel_pixels_tab[8];
251 0da71265 Michael Niedermayer
    
252
    /**
253
     * h264 Chram MC
254
     */
255
    h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
256
    h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
257 eb4b3dd3 Zdenek Kabelac
258 0da71265 Michael Niedermayer
    qpel_mc_func put_h264_qpel_pixels_tab[3][16];
259
    qpel_mc_func avg_h264_qpel_pixels_tab[3][16];
260
    
261 bb198e19 Michael Niedermayer
    me_cmp_func pix_abs[2][4];
262 11f18faf Michael Niedermayer
    
263
    /* huffyuv specific */
264
    void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
265 1457ab52 Michael Niedermayer
    void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
266 84705403 Michael Niedermayer
    /**
267
     * subtract huffyuv's variant of median prediction
268
     * note, this might read from src1[-1], src2[-1]
269
     */
270
    void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
271 3d2e8cce Michael Niedermayer
    void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w);
272 b0368839 Michael Niedermayer
    
273 332f9ac4 Michael Niedermayer
    void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
274
    void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
275
276 fdbbf2e0 Michael Niedermayer
    void (*h261_loop_filter)(uint8_t *src, int stride);
277 c6148de2 Michael Niedermayer
278 b0368839 Michael Niedermayer
    /* (I)DCT */
279
    void (*fdct)(DCTELEM *block/* align 16*/);
280 10acc479 Roman Shaposhnik
    void (*fdct248)(DCTELEM *block/* align 16*/);
281 4fb518c3 Michael Niedermayer
    
282
    /* IDCT really*/
283
    void (*idct)(DCTELEM *block/* align 16*/);
284 24641185 Michael Niedermayer
    
285
    /**
286 77c92c2d Michael Niedermayer
     * block -> idct -> clip to unsigned 8 bit -> dest.
287 24641185 Michael Niedermayer
     * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
288 a1adf436 Michael Niedermayer
     * @param line_size size in bytes of a horizotal line of dest
289 24641185 Michael Niedermayer
     */
290 b0368839 Michael Niedermayer
    void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
291 24641185 Michael Niedermayer
    
292
    /**
293
     * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
294 a1adf436 Michael Niedermayer
     * @param line_size size in bytes of a horizotal line of dest
295 24641185 Michael Niedermayer
     */
296 b0368839 Michael Niedermayer
    void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
297 24641185 Michael Niedermayer
    
298
    /**
299 77c92c2d Michael Niedermayer
     * idct input permutation.
300 05493021 Michael Niedermayer
     * several optimized IDCTs need a permutated input (relative to the normal order of the reference
301
     * IDCT)
302
     * this permutation must be performed before the idct_put/add, note, normally this can be merged
303
     * with the zigzag/alternate scan<br>
304 24641185 Michael Niedermayer
     * an example to avoid confusion:
305
     * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
306
     * - (x -> referece dct -> reference idct -> x)
307
     * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
308
     * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
309
     */
310 b0368839 Michael Niedermayer
    uint8_t idct_permutation[64];
311
    int idct_permutation_type;
312
#define FF_NO_IDCT_PERM 1
313
#define FF_LIBMPEG2_IDCT_PERM 2
314
#define FF_SIMPLE_IDCT_PERM 3
315
#define FF_TRANSPOSE_IDCT_PERM 4
316
317 364a1797 Michael Niedermayer
    int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
318
    void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
319
#define BASIS_SHIFT 16
320
#define RECON_SHIFT 6
321
322 44cb64ee Mike Melanson
    /**
323
     * This function handles any initialization for the VP3 DSP functions.
324
     */
325
    void (*vp3_dsp_init)(void);
326
327
    /** 
328
     * This function is responsible for taking a block of zigzag'd,
329 116824d0 Mike Melanson
     * quantized DCT coefficients and reconstructing the original block of
330
     * samples.
331 44cb64ee Mike Melanson
     * @param input_data 64 zigzag'd, quantized DCT coefficients
332
     * @param dequant_matrix 64 zigzag'd quantizer coefficients
333
     * @param coeff_count index of the last coefficient
334 116824d0 Mike Melanson
     * @param output_samples space for 64 DCTELEMs where the transformed
335
     * samples will be stored
336 44cb64ee Mike Melanson
     */
337 116824d0 Mike Melanson
    void (*vp3_idct)(int16_t *input_data, int16_t *dequant_matrix,
338
        int coeff_count, DCTELEM *output_samples);
339 0fa8158d Michael Niedermayer
 
340
    void (*h264_idct_add)(uint8_t *dst, DCTELEM *block, int stride);
341 eb4b3dd3 Zdenek Kabelac
} DSPContext;
342
343 59cf08ce Fabrice Bellard
void dsputil_static_init(void);
344 b0368839 Michael Niedermayer
void dsputil_init(DSPContext* p, AVCodecContext *avctx);
345 de6d9b64 Fabrice Bellard
346 7801d21d Michael Niedermayer
/**
347
 * permute block according to permuatation.
348
 * @param last last non zero element in scantable order
349
 */
350 0c1a9eda Zdenek Kabelac
void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
351 e0eac44e Fabrice Bellard
352 622348f9 Michael Niedermayer
void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
353
354 d8085ea7 Michael Niedermayer
#define        BYTE_VEC32(c)        ((c)*0x01010101UL)
355
356
static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
357
{
358
    return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
359
}
360
361
static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
362
{
363
    return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
364
}
365
366 26efc54e Michael Niedermayer
static inline int get_penalty_factor(int lambda, int lambda2, int type){
367
    switch(type&0xFF){
368
    default:
369
    case FF_CMP_SAD:
370
        return lambda>>FF_LAMBDA_SHIFT;
371
    case FF_CMP_DCT:
372
        return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
373
    case FF_CMP_W53:
374
        return (4*lambda)>>(FF_LAMBDA_SHIFT);
375
    case FF_CMP_W97:
376
        return (2*lambda)>>(FF_LAMBDA_SHIFT);
377
    case FF_CMP_SATD:
378
        return (2*lambda)>>FF_LAMBDA_SHIFT;
379
    case FF_CMP_RD:
380
    case FF_CMP_PSNR:
381
    case FF_CMP_SSE:
382
    case FF_CMP_NSSE:
383
        return lambda2>>FF_LAMBDA_SHIFT;
384
    case FF_CMP_BIT:
385
        return 1;
386
    }
387
}
388
389 24641185 Michael Niedermayer
/**
390 77c92c2d Michael Niedermayer
 * Empty mmx state.
391 24641185 Michael Niedermayer
 * this must be called between any dsp function and float/double code.
392
 * for example sin(); dsp->idct_put(); emms_c(); cos()
393
 */
394 eb4b3dd3 Zdenek Kabelac
#define emms_c()
395
396 e629ab68 Romain Dolbeau
/* should be defined by architectures supporting
397
   one or more MultiMedia extension */
398
int mm_support(void);
399
400 92a69cf8 Mike Melanson
#define __align16 __attribute__ ((aligned (16)))
401
402 3d03c0a2 Fabrice Bellard
#if defined(HAVE_MMX)
403 de6d9b64 Fabrice Bellard
404 18f77016 Zdenek Kabelac
#undef emms_c
405 eb4b3dd3 Zdenek Kabelac
406 de6d9b64 Fabrice Bellard
#define MM_MMX    0x0001 /* standard MMX */
407
#define MM_3DNOW  0x0004 /* AMD 3DNOW */
408
#define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
409
#define MM_SSE    0x0008 /* SSE functions */
410
#define MM_SSE2   0x0010 /* PIV SSE2 functions */
411 e42a152b Michael Niedermayer
#define MM_3DNOWEXT  0x0020 /* AMD 3DNowExt */
412 de6d9b64 Fabrice Bellard
413
extern int mm_flags;
414
415 0c1a9eda Zdenek Kabelac
void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
416
void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
417 f9ed9d85 Mike Melanson
void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
418 de6d9b64 Fabrice Bellard
419
static inline void emms(void)
420
{
421 fb16b7e7 Fabrice Bellard
    __asm __volatile ("emms;":::"memory");
422
}
423
424 1457ab52 Michael Niedermayer
425 fb16b7e7 Fabrice Bellard
#define emms_c() \
426
{\
427
    if (mm_flags & MM_MMX)\
428
        emms();\
429 de6d9b64 Fabrice Bellard
}
430
431
#define __align8 __attribute__ ((aligned (8)))
432 3237f731 Michael Niedermayer
#define STRIDE_ALIGN 8
433 de6d9b64 Fabrice Bellard
434 b0368839 Michael Niedermayer
void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
435
void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
436 2720569a Zdenek Kabelac
437 3d03c0a2 Fabrice Bellard
#elif defined(ARCH_ARMV4L)
438
439
/* This is to use 4 bytes read to the IDCT pointers for some 'zero'
440 92a69cf8 Mike Melanson
   line optimizations */
441 3d03c0a2 Fabrice Bellard
#define __align8 __attribute__ ((aligned (4)))
442 3237f731 Michael Niedermayer
#define STRIDE_ALIGN 4
443 3d03c0a2 Fabrice Bellard
444 b0368839 Michael Niedermayer
void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
445 3d03c0a2 Fabrice Bellard
446 c34270f5 Fabrice Bellard
#elif defined(HAVE_MLIB)
447
448
/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
449
#define __align8 __attribute__ ((aligned (8)))
450 3237f731 Michael Niedermayer
#define STRIDE_ALIGN 8
451 c34270f5 Fabrice Bellard
452 b0368839 Michael Niedermayer
void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
453 c34270f5 Fabrice Bellard
454 44f54ceb Michael Niedermayer
#elif defined(ARCH_SPARC)
455
456
/* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
457
#define __align8 __attribute__ ((aligned (8)))
458 3237f731 Michael Niedermayer
#define STRIDE_ALIGN 8
459 44f54ceb Michael Niedermayer
void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
460
461 1e98dffb Nick Kurshev
#elif defined(ARCH_ALPHA)
462
463
#define __align8 __attribute__ ((aligned (8)))
464 3237f731 Michael Niedermayer
#define STRIDE_ALIGN 8
465 1e98dffb Nick Kurshev
466 b0368839 Michael Niedermayer
void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
467 1e98dffb Nick Kurshev
468 59925ef2 Brian Foley
#elif defined(ARCH_POWERPC)
469
470 404d2241 Brian Foley
#define MM_ALTIVEC    0x0001 /* standard AltiVec */
471
472
extern int mm_flags;
473
474 3b991c54 Romain Dolbeau
#if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN)
475 9a197a24 Michael Niedermayer
#define pixel altivec_pixel
476 3b991c54 Romain Dolbeau
#include <altivec.h>
477 9a197a24 Michael Niedermayer
#undef pixel
478 3b991c54 Romain Dolbeau
#endif
479
480 59925ef2 Brian Foley
#define __align8 __attribute__ ((aligned (16)))
481 3237f731 Michael Niedermayer
#define STRIDE_ALIGN 16
482 59925ef2 Brian Foley
483 b0368839 Michael Niedermayer
void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
484 59925ef2 Brian Foley
485 d46aba26 Leon van Stuivenberg
#elif defined(HAVE_MMI)
486
487
#define __align8 __attribute__ ((aligned (16)))
488 3237f731 Michael Niedermayer
#define STRIDE_ALIGN 16
489 d46aba26 Leon van Stuivenberg
490 b0368839 Michael Niedermayer
void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
491 d46aba26 Leon van Stuivenberg
492 0c6bd2ea BERO
#elif defined(ARCH_SH4)
493
494
#define __align8 __attribute__ ((aligned (8)))
495 3237f731 Michael Niedermayer
#define STRIDE_ALIGN 8
496 0c6bd2ea BERO
497
void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
498
499 de6d9b64 Fabrice Bellard
#else
500
501 3237f731 Michael Niedermayer
#define __align8 __attribute__ ((aligned (8)))
502
#define STRIDE_ALIGN 8
503 de6d9b64 Fabrice Bellard
504
#endif
505
506 6d4985bb Fabrice Bellard
#ifdef __GNUC__
507
508
struct unaligned_64 { uint64_t l; } __attribute__((packed));
509
struct unaligned_32 { uint32_t l; } __attribute__((packed));
510 669ac79c Michael Niedermayer
struct unaligned_16 { uint16_t l; } __attribute__((packed));
511 6d4985bb Fabrice Bellard
512 669ac79c Michael Niedermayer
#define LD16(a) (((const struct unaligned_16 *) (a))->l)
513 6d4985bb Fabrice Bellard
#define LD32(a) (((const struct unaligned_32 *) (a))->l)
514
#define LD64(a) (((const struct unaligned_64 *) (a))->l)
515
516
#define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)
517
518
#else /* __GNUC__ */
519
520 669ac79c Michael Niedermayer
#define LD16(a) (*((uint16_t*)(a)))
521 6d4985bb Fabrice Bellard
#define LD32(a) (*((uint32_t*)(a)))
522
#define LD64(a) (*((uint64_t*)(a)))
523
524
#define ST32(a, b) *((uint32_t*)(a)) = (b)
525
526
#endif /* !__GNUC__ */
527
528 43f1708f Juanjo
/* PSNR */
529 0c1a9eda Zdenek Kabelac
void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
530 43f1708f Juanjo
              int orig_linesize[3], int coded_linesize,
531
              AVCodecContext *avctx);
532 bb6f5690 Fabrice Bellard
533
/* FFT computation */
534
535
/* NOTE: soon integer code will be added, so you must use the
536
   FFTSample type */
537
typedef float FFTSample;
538
539
typedef struct FFTComplex {
540
    FFTSample re, im;
541
} FFTComplex;
542
543
typedef struct FFTContext {
544
    int nbits;
545
    int inverse;
546
    uint16_t *revtab;
547
    FFTComplex *exptab;
548
    FFTComplex *exptab1; /* only used by SSE code */
549
    void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
550
} FFTContext;
551
552 68951ecf Gildas Bazin
int ff_fft_init(FFTContext *s, int nbits, int inverse);
553
void ff_fft_permute(FFTContext *s, FFTComplex *z);
554
void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
555
void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
556
void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
557 8d268a7d Fabrice Bellard
558 68951ecf Gildas Bazin
static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
559 bb6f5690 Fabrice Bellard
{
560
    s->fft_calc(s, z);
561
}
562 68951ecf Gildas Bazin
void ff_fft_end(FFTContext *s);
563 bb6f5690 Fabrice Bellard
564
/* MDCT computation */
565
566
typedef struct MDCTContext {
567
    int n;  /* size of MDCT (i.e. number of input data * 2) */
568
    int nbits; /* n = 2^nbits */
569
    /* pre/post rotation tables */
570
    FFTSample *tcos;
571
    FFTSample *tsin;
572
    FFTContext fft;
573
} MDCTContext;
574
575 82696bee Fabrice Bellard
int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
576 eb4b3dd3 Zdenek Kabelac
void ff_imdct_calc(MDCTContext *s, FFTSample *output,
577 bb6f5690 Fabrice Bellard
                const FFTSample *input, FFTSample *tmp);
578 eb4b3dd3 Zdenek Kabelac
void ff_mdct_calc(MDCTContext *s, FFTSample *out,
579 bb6f5690 Fabrice Bellard
               const FFTSample *input, FFTSample *tmp);
580 82696bee Fabrice Bellard
void ff_mdct_end(MDCTContext *s);
581 bb6f5690 Fabrice Bellard
582 bb198e19 Michael Niedermayer
#define WARPER8_16(name8, name16)\
583
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
584
    return name8(s, dst           , src           , stride, h)\
585
          +name8(s, dst+8         , src+8         , stride, h);\
586
}
587
588
#define WARPER8_16_SQ(name8, name16)\
589
static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
590
    int score=0;\
591
    score +=name8(s, dst           , src           , stride, 8);\
592
    score +=name8(s, dst+8         , src+8         , stride, 8);\
593
    if(h==16){\
594
        dst += 8*stride;\
595
        src += 8*stride;\
596
        score +=name8(s, dst           , src           , stride, 8);\
597
        score +=name8(s, dst+8         , src+8         , stride, 8);\
598
    }\
599
    return score;\
600 1457ab52 Michael Niedermayer
}
601
602 95e2ce4a Michael Niedermayer
#ifndef HAVE_LRINTF
603 9d85cbd9 Fabrice Bellard
/* XXX: add ISOC specific test to avoid specific BSD testing. */
604
/* better than nothing implementation. */
605 6234d753 Alex Beregszaszi
/* btw, rintf() is existing on fbsd too -- alex */
606 91d6655a Michael Niedermayer
static always_inline long int lrintf(float x)
607 9d85cbd9 Fabrice Bellard
{
608 ea937d01 Fabrice Bellard
#ifdef CONFIG_WIN32
609 91d6655a Michael Niedermayer
#  ifdef ARCH_X86
610
    int32_t i;
611
    asm volatile(
612
        "fistpl %0\n\t"
613
        : "=m" (i) : "t" (x) : "st"
614
    );
615
    return i;
616
#  else
617 ea937d01 Fabrice Bellard
    /* XXX: incorrect, but make it compile */
618 91d6655a Michael Niedermayer
    return (int)(x + (x < 0 ? -0.5 : 0.5));
619
#  endif
620 ea937d01 Fabrice Bellard
#else
621 9d85cbd9 Fabrice Bellard
    return (int)(rint(x));
622 ea937d01 Fabrice Bellard
#endif
623 9d85cbd9 Fabrice Bellard
}
624 e96682e6 Michael Niedermayer
#else
625
#ifndef _ISOC9X_SOURCE
626
#define _ISOC9X_SOURCE
627
#endif
628
#include <math.h>
629 9d85cbd9 Fabrice Bellard
#endif
630
631 de6d9b64 Fabrice Bellard
#endif