Mercurial > libavcodec.hg
annotate dsputil.h @ 1217:eb2affe57a2a libavcodec
10l
author | michaelni |
---|---|
date | Fri, 25 Apr 2003 20:03:17 +0000 |
parents | a71fe70b428b |
children | ca979193f8e1 |
rev | line source |
---|---|
429 | 1 /* |
2 * DSP utils | |
3 * Copyright (c) 2000, 2001, 2002 Fabrice Bellard. | |
4 * | |
5 * This library is free software; you can redistribute it and/or | |
6 * modify it under the terms of the GNU Lesser General Public | |
7 * License as published by the Free Software Foundation; either | |
8 * version 2 of the License, or (at your option) any later version. | |
9 * | |
10 * This library is distributed in the hope that it will be useful, | |
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
13 * Lesser General Public License for more details. | |
14 * | |
15 * You should have received a copy of the GNU Lesser General Public | |
16 * License along with this library; if not, write to the Free Software | |
17 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA | |
18 */ | |
1102 | 19 |
20 /** | |
21 * @file dsputil.h | |
1106 | 22 * DSP utils. |
1213 | 23 * note, many functions in here may use MMX which trashes the FPU state, it is |
24 * absolutely necessary to call emms_c() between dsp & float/double code | |
1102 | 25 */ |
26 | |
0 | 27 #ifndef DSPUTIL_H |
28 #define DSPUTIL_H | |
29 | |
30 #include "common.h" | |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
214
diff
changeset
|
31 #include "avcodec.h" |
0 | 32 |
1102 | 33 |
255 | 34 //#define DEBUG |
0 | 35 /* dct code */ |
36 typedef short DCTELEM; | |
1008 | 37 //typedef int DCTELEM; |
0 | 38 |
474
11dbd00682fc
avoid name clash with libjpeg - added missing externs
bellard
parents:
429
diff
changeset
|
39 void fdct_ifast (DCTELEM *data); |
625
bb6a69f9d409
slow but accurate integer dct from IJG (should be ok with the LGPL as the old DCT is the fast integer DCT from IJG)
michaelni
parents:
623
diff
changeset
|
40 void ff_jpeg_fdct_islow (DCTELEM *data); |
0 | 41 |
42 void j_rev_dct (DCTELEM *data); | |
43 | |
687
9abb13c21fbe
fdct_mmx -> ff_fdct_mmx (renamed to avoid namespace conflict with xvid)
arpi_esp
parents:
675
diff
changeset
|
44 void ff_fdct_mmx(DCTELEM *block); |
0 | 45 |
34 | 46 /* encoding scans */ |
1064 | 47 extern const uint8_t ff_alternate_horizontal_scan[64]; |
48 extern const uint8_t ff_alternate_vertical_scan[64]; | |
49 extern const uint8_t ff_zigzag_direct[64]; | |
190
9e0e56869d05
fix for non-mmx runtimedetect encoding bugs - patch by Michael Niedermayer <michaelni@gmx.at>
uid46427
parents:
174
diff
changeset
|
50 |
0 | 51 /* pixel operations */ |
52 #define MAX_NEG_CROP 384 | |
53 | |
54 /* temporary */ | |
1064 | 55 extern uint32_t squareTbl[512]; |
56 extern uint8_t cropTbl[256 + 2 * MAX_NEG_CROP]; | |
0 | 57 |
58 | |
675 | 59 /* minimum alignment rules ;) |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
60 if u notice errors in the align stuff, need more alignment for some asm code for some cpu |
675 | 61 or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ... |
62 | |
63 !warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible) | |
64 i (michael) didnt check them, these are just the alignents which i think could be reached easily ... | |
65 | |
66 !future video codecs might need functions with less strict alignment | |
67 */ | |
0 | 68 |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
69 /* |
1064 | 70 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size); |
71 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride); | |
72 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size); | |
73 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size); | |
296 | 74 void clear_blocks_c(DCTELEM *blocks); |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
75 */ |
0 | 76 |
77 /* add and put pixel (decoding) */ | |
675 | 78 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16 |
1064 | 79 typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h); |
80 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride); | |
1168 | 81 typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y); |
0 | 82 |
984 | 83 #define DEF_OLD_QPEL(name)\ |
1064 | 84 void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ |
85 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\ | |
86 void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride); | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
87 |
984 | 88 DEF_OLD_QPEL(qpel16_mc11_old_c) |
89 DEF_OLD_QPEL(qpel16_mc31_old_c) | |
90 DEF_OLD_QPEL(qpel16_mc12_old_c) | |
91 DEF_OLD_QPEL(qpel16_mc32_old_c) | |
92 DEF_OLD_QPEL(qpel16_mc13_old_c) | |
93 DEF_OLD_QPEL(qpel16_mc33_old_c) | |
94 DEF_OLD_QPEL(qpel8_mc11_old_c) | |
95 DEF_OLD_QPEL(qpel8_mc31_old_c) | |
96 DEF_OLD_QPEL(qpel8_mc12_old_c) | |
97 DEF_OLD_QPEL(qpel8_mc32_old_c) | |
98 DEF_OLD_QPEL(qpel8_mc13_old_c) | |
99 DEF_OLD_QPEL(qpel8_mc33_old_c) | |
651 | 100 |
101 #define CALL_2X_PIXELS(a, b, n)\ | |
102 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\ | |
103 b(block , pixels , line_size, h);\ | |
104 b(block+n, pixels+n, line_size, h);\ | |
105 } | |
255 | 106 |
0 | 107 /* motion estimation */ |
108 | |
1064 | 109 typedef int (*op_pixels_abs_func)(uint8_t *blk1/*align width (8 or 16)*/, uint8_t *blk2/*align 1*/, int line_size)/* __attribute__ ((const))*/; |
936 | 110 |
1064 | 111 typedef int (*me_cmp_func)(void /*MpegEncContext*/ *s, uint8_t *blk1/*align width (8 or 16)*/, uint8_t *blk2/*align 1*/, int line_size)/* __attribute__ ((const))*/; |
936 | 112 |
1168 | 113 |
1102 | 114 /** |
115 * DSPContext. | |
116 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
117 typedef struct DSPContext { |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
118 /* pixel ops : interface with DCT */ |
1064 | 119 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size); |
120 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride); | |
121 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); | |
122 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size); | |
1136 | 123 /** |
124 * translational global motion compensation. | |
125 */ | |
1064 | 126 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder); |
1136 | 127 /** |
128 * global motion compensation. | |
129 */ | |
1064 | 130 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy, |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
131 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height); |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
132 void (*clear_blocks)(DCTELEM *blocks/*align 16*/); |
1064 | 133 int (*pix_sum)(uint8_t * pix, int line_size); |
134 int (*pix_norm1)(uint8_t * pix, int line_size); | |
936 | 135 me_cmp_func sad[2]; /* identical to pix_absAxA except additional void * */ |
136 me_cmp_func sse[2]; | |
137 me_cmp_func hadamard8_diff[2]; | |
138 me_cmp_func dct_sad[2]; | |
139 me_cmp_func quant_psnr[2]; | |
1007 | 140 me_cmp_func bit[2]; |
141 me_cmp_func rd[2]; | |
936 | 142 int (*hadamard8_abs )(uint8_t *src, int stride, int mean); |
143 | |
954 | 144 me_cmp_func me_pre_cmp[11]; |
936 | 145 me_cmp_func me_cmp[11]; |
146 me_cmp_func me_sub_cmp[11]; | |
147 me_cmp_func mb_cmp[11]; | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
148 |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
149 /* maybe create an array for 16/8 functions */ |
1136 | 150 /** |
151 * Halfpel motion compensation with rounding (a+b+1)>>1. | |
1213 | 152 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 153 * @param block destination where the result is stored |
154 * @param pixels source | |
155 * @param line_size number of bytes in a horizontal line of block | |
156 * @param h height | |
157 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
158 op_pixels_func put_pixels_tab[2][4]; |
1136 | 159 |
160 /** | |
161 * Halfpel motion compensation with rounding (a+b+1)>>1. | |
1213 | 162 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 163 * @param block destination into which the result is averaged (a+b+1)>>1 |
164 * @param pixels source | |
165 * @param line_size number of bytes in a horizontal line of block | |
166 * @param h height | |
167 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
168 op_pixels_func avg_pixels_tab[2][4]; |
1136 | 169 |
170 /** | |
171 * Halfpel motion compensation with no rounding (a+b)>>1. | |
1213 | 172 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 173 * @param block destination where the result is stored |
174 * @param pixels source | |
175 * @param line_size number of bytes in a horizontal line of block | |
176 * @param h height | |
177 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
178 op_pixels_func put_no_rnd_pixels_tab[2][4]; |
1136 | 179 |
180 /** | |
181 * Halfpel motion compensation with no rounding (a+b)>>1. | |
1213 | 182 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ] |
1136 | 183 * @param block destination into which the result is averaged (a+b)>>1 |
184 * @param pixels source | |
185 * @param line_size number of bytes in a horizontal line of block | |
186 * @param h height | |
187 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
188 op_pixels_func avg_no_rnd_pixels_tab[2][4]; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
189 qpel_mc_func put_qpel_pixels_tab[2][16]; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
190 qpel_mc_func avg_qpel_pixels_tab[2][16]; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
191 qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16]; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
192 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16]; |
936 | 193 qpel_mc_func put_mspel_pixels_tab[8]; |
1168 | 194 |
195 /** | |
196 * h264 Chram MC | |
197 */ | |
198 h264_chroma_mc_func put_h264_chroma_pixels_tab[3]; | |
199 h264_chroma_mc_func avg_h264_chroma_pixels_tab[3]; | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
200 |
1168 | 201 qpel_mc_func put_h264_qpel_pixels_tab[3][16]; |
202 qpel_mc_func avg_h264_qpel_pixels_tab[3][16]; | |
203 | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
204 op_pixels_abs_func pix_abs16x16; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
205 op_pixels_abs_func pix_abs16x16_x2; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
206 op_pixels_abs_func pix_abs16x16_y2; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
207 op_pixels_abs_func pix_abs16x16_xy2; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
208 op_pixels_abs_func pix_abs8x8; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
209 op_pixels_abs_func pix_abs8x8_x2; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
210 op_pixels_abs_func pix_abs8x8_y2; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
211 op_pixels_abs_func pix_abs8x8_xy2; |
866 | 212 |
213 /* huffyuv specific */ | |
214 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w); | |
936 | 215 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w); |
1092 | 216 |
217 /* (I)DCT */ | |
218 void (*fdct)(DCTELEM *block/* align 16*/); | |
1102 | 219 |
220 /** | |
1104 | 221 * block -> idct -> clip to unsigned 8 bit -> dest. |
1102 | 222 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...) |
1111 | 223 * @param line_size size in bytes of a horizotal line of dest |
1102 | 224 */ |
1092 | 225 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
1102 | 226 |
227 /** | |
228 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest. | |
1111 | 229 * @param line_size size in bytes of a horizotal line of dest |
1102 | 230 */ |
1092 | 231 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/); |
1102 | 232 |
233 /** | |
1104 | 234 * idct input permutation. |
1102 | 235 * an example to avoid confusion: |
236 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...) | |
237 * - (x -> referece dct -> reference idct -> x) | |
238 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x) | |
239 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...) | |
240 */ | |
1092 | 241 uint8_t idct_permutation[64]; |
242 int idct_permutation_type; | |
243 #define FF_NO_IDCT_PERM 1 | |
244 #define FF_LIBMPEG2_IDCT_PERM 2 | |
245 #define FF_SIMPLE_IDCT_PERM 3 | |
246 #define FF_TRANSPOSE_IDCT_PERM 4 | |
247 | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
248 } DSPContext; |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
249 |
1201 | 250 void dsputil_static_init(void); |
1092 | 251 void dsputil_init(DSPContext* p, AVCodecContext *avctx); |
0 | 252 |
764 | 253 /** |
254 * permute block according to permuatation. | |
255 * @param last last non zero element in scantable order | |
256 */ | |
1064 | 257 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last); |
34 | 258 |
1102 | 259 /** |
1104 | 260 * Empty mmx state. |
1102 | 261 * this must be called between any dsp function and float/double code. |
262 * for example sin(); dsp->idct_put(); emms_c(); cos() | |
263 */ | |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
264 #define emms_c() |
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
265 |
995
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
266 /* should be defined by architectures supporting |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
267 one or more MultiMedia extension */ |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
268 int mm_support(void); |
edc10966b081
altivec jumbo patch by (Romain Dolbeau <dolbeaur at club-internet dot fr>)
michaelni
parents:
984
diff
changeset
|
269 |
62 | 270 #if defined(HAVE_MMX) |
0 | 271 |
862 | 272 #undef emms_c |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
273 |
0 | 274 #define MM_MMX 0x0001 /* standard MMX */ |
275 #define MM_3DNOW 0x0004 /* AMD 3DNOW */ | |
276 #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */ | |
277 #define MM_SSE 0x0008 /* SSE functions */ | |
278 #define MM_SSE2 0x0010 /* PIV SSE2 functions */ | |
279 | |
280 extern int mm_flags; | |
281 | |
1064 | 282 void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size); |
283 void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size); | |
0 | 284 |
285 static inline void emms(void) | |
286 { | |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
287 __asm __volatile ("emms;":::"memory"); |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
288 } |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
289 |
936 | 290 |
6
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
291 #define emms_c() \ |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
292 {\ |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
293 if (mm_flags & MM_MMX)\ |
ec4642daa6fe
added emms_c() macro which should can used in c code in both mmx/non mmx cases
glantau
parents:
2
diff
changeset
|
294 emms();\ |
0 | 295 } |
296 | |
297 #define __align8 __attribute__ ((aligned (8))) | |
298 | |
1092 | 299 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx); |
300 void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx); | |
1065 | 301 |
62 | 302 #elif defined(ARCH_ARMV4L) |
303 | |
304 /* This is to use 4 bytes read to the IDCT pointers for some 'zero' | |
305 line ptimizations */ | |
306 #define __align8 __attribute__ ((aligned (4))) | |
307 | |
1092 | 308 void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx); |
62 | 309 |
88 | 310 #elif defined(HAVE_MLIB) |
311 | |
312 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */ | |
313 #define __align8 __attribute__ ((aligned (8))) | |
314 | |
1092 | 315 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx); |
88 | 316 |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
317 #elif defined(ARCH_ALPHA) |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
318 |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
319 #define __align8 __attribute__ ((aligned (8))) |
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
320 |
1092 | 321 void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx); |
214
73df666cacc7
Alpha optimizations by Falk Hueffner <falk.hueffner@student.uni-tuebingen.de>
nickols_k
parents:
190
diff
changeset
|
322 |
623
92e99e506920
first cut at altivec support on darwin patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
612
diff
changeset
|
323 #elif defined(ARCH_POWERPC) |
92e99e506920
first cut at altivec support on darwin patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
612
diff
changeset
|
324 |
894
a408778eff87
altivec accelerated v-resample patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
884
diff
changeset
|
325 #define MM_ALTIVEC 0x0001 /* standard AltiVec */ |
a408778eff87
altivec accelerated v-resample patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
884
diff
changeset
|
326 |
a408778eff87
altivec accelerated v-resample patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
884
diff
changeset
|
327 extern int mm_flags; |
a408778eff87
altivec accelerated v-resample patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
884
diff
changeset
|
328 |
1033
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
329 #if defined(HAVE_ALTIVEC) && !defined(CONFIG_DARWIN) |
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
330 #include <altivec.h> |
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
331 #endif |
b4172ff70d27
Altivec on non darwin systems patch by Romain Dolbeau
bellard
parents:
1008
diff
changeset
|
332 |
623
92e99e506920
first cut at altivec support on darwin patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
612
diff
changeset
|
333 #define __align8 __attribute__ ((aligned (16))) |
92e99e506920
first cut at altivec support on darwin patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
612
diff
changeset
|
334 |
1092 | 335 void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx); |
623
92e99e506920
first cut at altivec support on darwin patch by (Brian Foley <bfoley at compsoc dot nuigalway dot ie>)
michaelni
parents:
612
diff
changeset
|
336 |
689
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
337 #elif defined(HAVE_MMI) |
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
338 |
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
339 #define __align8 __attribute__ ((aligned (16))) |
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
340 |
1092 | 341 void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx); |
689
efcbfbd18864
ps2 idct patch by (Leon van Stuivenberg <leonvs at iae dot nl>)
michaelni
parents:
687
diff
changeset
|
342 |
0 | 343 #else |
344 | |
345 #define __align8 | |
346 | |
347 #endif | |
348 | |
493
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
349 #ifdef __GNUC__ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
350 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
351 struct unaligned_64 { uint64_t l; } __attribute__((packed)); |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
352 struct unaligned_32 { uint32_t l; } __attribute__((packed)); |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
353 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
354 #define LD32(a) (((const struct unaligned_32 *) (a))->l) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
355 #define LD64(a) (((const struct unaligned_64 *) (a))->l) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
356 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
357 #define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
358 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
359 #else /* __GNUC__ */ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
360 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
361 #define LD32(a) (*((uint32_t*)(a))) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
362 #define LD64(a) (*((uint64_t*)(a))) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
363 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
364 #define ST32(a, b) *((uint32_t*)(a)) = (b) |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
365 |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
366 #endif /* !__GNUC__ */ |
873b9075d853
move unaligned access macros to dsputil.h - added unaligned 32 bit store
bellard
parents:
480
diff
changeset
|
367 |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
214
diff
changeset
|
368 /* PSNR */ |
1064 | 369 void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3], |
252
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
214
diff
changeset
|
370 int orig_linesize[3], int coded_linesize, |
ddb1a0e94cf4
- Added PSNR feature to libavcodec and ffmpeg. By now just Y PSNR until I'm
pulento
parents:
214
diff
changeset
|
371 AVCodecContext *avctx); |
781 | 372 |
373 /* FFT computation */ | |
374 | |
375 /* NOTE: soon integer code will be added, so you must use the | |
376 FFTSample type */ | |
377 typedef float FFTSample; | |
378 | |
379 typedef struct FFTComplex { | |
380 FFTSample re, im; | |
381 } FFTComplex; | |
382 | |
383 typedef struct FFTContext { | |
384 int nbits; | |
385 int inverse; | |
386 uint16_t *revtab; | |
387 FFTComplex *exptab; | |
388 FFTComplex *exptab1; /* only used by SSE code */ | |
389 void (*fft_calc)(struct FFTContext *s, FFTComplex *z); | |
390 } FFTContext; | |
391 | |
392 int fft_init(FFTContext *s, int nbits, int inverse); | |
393 void fft_permute(FFTContext *s, FFTComplex *z); | |
394 void fft_calc_c(FFTContext *s, FFTComplex *z); | |
395 void fft_calc_sse(FFTContext *s, FFTComplex *z); | |
975
e05d525505c5
fft altivec by Romain Dolbeau - simplified patch, test it on PPC with fft-test and wma decoding
bellard
parents:
954
diff
changeset
|
396 void fft_calc_altivec(FFTContext *s, FFTComplex *z); |
e05d525505c5
fft altivec by Romain Dolbeau - simplified patch, test it on PPC with fft-test and wma decoding
bellard
parents:
954
diff
changeset
|
397 |
781 | 398 static inline void fft_calc(FFTContext *s, FFTComplex *z) |
399 { | |
400 s->fft_calc(s, z); | |
401 } | |
402 void fft_end(FFTContext *s); | |
403 | |
404 /* MDCT computation */ | |
405 | |
406 typedef struct MDCTContext { | |
407 int n; /* size of MDCT (i.e. number of input data * 2) */ | |
408 int nbits; /* n = 2^nbits */ | |
409 /* pre/post rotation tables */ | |
410 FFTSample *tcos; | |
411 FFTSample *tsin; | |
412 FFTContext fft; | |
413 } MDCTContext; | |
414 | |
794 | 415 int ff_mdct_init(MDCTContext *s, int nbits, int inverse); |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
416 void ff_imdct_calc(MDCTContext *s, FFTSample *output, |
781 | 417 const FFTSample *input, FFTSample *tmp); |
853
eacc2dd8fd9d
* using DSPContext - so each codec could use its local (sub)set of CPU extension
kabi
parents:
838
diff
changeset
|
418 void ff_mdct_calc(MDCTContext *s, FFTSample *out, |
781 | 419 const FFTSample *input, FFTSample *tmp); |
794 | 420 void ff_mdct_end(MDCTContext *s); |
781 | 421 |
936 | 422 #define WARPER88_1616(name8, name16)\ |
423 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride){\ | |
424 return name8(s, dst , src , stride)\ | |
425 +name8(s, dst+8 , src+8 , stride)\ | |
426 +name8(s, dst +8*stride, src +8*stride, stride)\ | |
427 +name8(s, dst+8+8*stride, src+8+8*stride, stride);\ | |
428 } | |
429 | |
838
b78812db886f
lrintf detection (based upon a patch by Franois Revol <revol at free dot fr>)
michaelni
parents:
802
diff
changeset
|
430 #ifndef HAVE_LRINTF |
796
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
431 /* XXX: add ISOC specific test to avoid specific BSD testing. */ |
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
432 /* better than nothing implementation. */ |
802 | 433 /* btw, rintf() is existing on fbsd too -- alex */ |
796
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
434 static inline long int lrintf(float x) |
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
435 { |
1040
998d5035b15b
win32: rint() does not seem to be defined with mingw32-gcc 2.95 - do you have a better solution ?
bellard
parents:
1033
diff
changeset
|
436 #ifdef CONFIG_WIN32 |
998d5035b15b
win32: rint() does not seem to be defined with mingw32-gcc 2.95 - do you have a better solution ?
bellard
parents:
1033
diff
changeset
|
437 /* XXX: incorrect, but make it compile */ |
998d5035b15b
win32: rint() does not seem to be defined with mingw32-gcc 2.95 - do you have a better solution ?
bellard
parents:
1033
diff
changeset
|
438 return (int)(x); |
998d5035b15b
win32: rint() does not seem to be defined with mingw32-gcc 2.95 - do you have a better solution ?
bellard
parents:
1033
diff
changeset
|
439 #else |
796
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
440 return (int)(rint(x)); |
1040
998d5035b15b
win32: rint() does not seem to be defined with mingw32-gcc 2.95 - do you have a better solution ?
bellard
parents:
1033
diff
changeset
|
441 #endif |
796
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
442 } |
0 | 443 #endif |
796
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
444 |
8a5b70c68fbd
added lrintf for non ISOC libcs (fixme: find a better test)
bellard
parents:
794
diff
changeset
|
445 #endif |