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