3 * Copyright (c) 2000, 2001, 2002 Fabrice Bellard
4 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
6 * This file is part of FFmpeg.
8 * FFmpeg is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
13 * FFmpeg is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with FFmpeg; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
26 * note, many functions in here may use MMX which trashes the FPU state, it is
27 * absolutely necessary to call emms_c() between dsp & float/double code
30 #ifndef AVCODEC_DSPUTIL_H
31 #define AVCODEC_DSPUTIL_H
33 #include "libavutil/intreadwrite.h"
41 extern const uint8_t ff_alternate_horizontal_scan[64];
42 extern const uint8_t ff_alternate_vertical_scan[64];
43 extern const uint8_t ff_zigzag_direct[64];
44 extern const uint8_t ff_zigzag248_direct[64];
46 /* pixel operations */
47 #define MAX_NEG_CROP 1024
50 extern uint32_t ff_squareTbl[512];
51 extern const uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP];
53 void ff_put_pixels8x8_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
54 void ff_avg_pixels8x8_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
55 void ff_put_pixels16x16_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
56 void ff_avg_pixels16x16_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
59 void ff_put_rv40_qpel16_mc33_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
60 void ff_avg_rv40_qpel16_mc33_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
61 void ff_put_rv40_qpel8_mc33_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
62 void ff_avg_rv40_qpel8_mc33_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
64 void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
65 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
67 /* minimum alignment rules ;)
68 If you notice errors in the align stuff, need more alignment for some ASM code
69 for some CPU or need to use a function with less aligned data then send a mail
70 to the ffmpeg-devel mailing list, ...
72 !warning These alignments might not match reality, (missing attribute((align))
73 stuff somewhere possible).
74 I (Michael) did not check them, these are just the alignments which I think
75 could be reached easily ...
77 !future video codecs might need functions with less strict alignment
80 /* add and put pixel (decoding) */
81 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
82 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller than 4
83 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);
84 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, ptrdiff_t stride);
86 typedef void (*op_fill_func)(uint8_t *block/*align width (8 or 16)*/, uint8_t value, int line_size, int h);
88 #define DEF_OLD_QPEL(name)\
89 void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, ptrdiff_t stride);\
90 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, ptrdiff_t stride);\
91 void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, ptrdiff_t stride);
93 DEF_OLD_QPEL(qpel16_mc11_old_c)
94 DEF_OLD_QPEL(qpel16_mc31_old_c)
95 DEF_OLD_QPEL(qpel16_mc12_old_c)
96 DEF_OLD_QPEL(qpel16_mc32_old_c)
97 DEF_OLD_QPEL(qpel16_mc13_old_c)
98 DEF_OLD_QPEL(qpel16_mc33_old_c)
99 DEF_OLD_QPEL(qpel8_mc11_old_c)
100 DEF_OLD_QPEL(qpel8_mc31_old_c)
101 DEF_OLD_QPEL(qpel8_mc12_old_c)
102 DEF_OLD_QPEL(qpel8_mc32_old_c)
103 DEF_OLD_QPEL(qpel8_mc13_old_c)
104 DEF_OLD_QPEL(qpel8_mc33_old_c)
106 /* motion estimation */
107 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller than 2
108 // although currently h<4 is not used as functions with width <8 are neither used nor implemented
109 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))*/;
114 typedef struct ScanTable{
115 const uint8_t *scantable;
116 uint8_t permutated[64];
117 uint8_t raster_end[64];
120 void ff_init_scantable(uint8_t *, ScanTable *st, const uint8_t *src_scantable);
121 void ff_init_scantable_permutation(uint8_t *idct_permutation,
122 int idct_permutation_type);
127 typedef struct DSPContext {
128 /* pixel ops : interface with DCT */
129 void (*get_pixels)(int16_t *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
130 void (*diff_pixels)(int16_t *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
131 void (*put_pixels_clamped)(const int16_t *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
132 void (*put_signed_pixels_clamped)(const int16_t *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
133 void (*add_pixels_clamped)(const int16_t *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
134 void (*add_pixels8)(uint8_t *pixels, int16_t *block, int line_size);
135 int (*sum_abs_dctelem)(int16_t *block/*align 16*/);
137 * translational global motion compensation.
139 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
141 * global motion compensation.
143 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
144 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
145 void (*clear_block)(int16_t *block/*align 16*/);
146 void (*clear_blocks)(int16_t *blocks/*align 16*/);
147 int (*pix_sum)(uint8_t * pix, int line_size);
148 int (*pix_norm1)(uint8_t * pix, int line_size);
149 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
151 me_cmp_func sad[6]; /* identical to pix_absAxA except additional void * */
153 me_cmp_func hadamard8_diff[6];
154 me_cmp_func dct_sad[6];
155 me_cmp_func quant_psnr[6];
163 me_cmp_func dct_max[6];
164 me_cmp_func dct264_sad[6];
166 me_cmp_func me_pre_cmp[6];
167 me_cmp_func me_cmp[6];
168 me_cmp_func me_sub_cmp[6];
169 me_cmp_func mb_cmp[6];
170 me_cmp_func ildct_cmp[6]; //only width 16 used
171 me_cmp_func frame_skip_cmp[6]; //only width 8 used
173 int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
177 * Thirdpel motion compensation with rounding (a+b+1)>>1.
178 * this is an array[12] of motion compensation functions for the 9 thirdpe
180 * *pixels_tab[ xthirdpel + 4*ythirdpel ]
181 * @param block destination where the result is stored
182 * @param pixels source
183 * @param line_size number of bytes in a horizontal line of block
186 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
187 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
189 qpel_mc_func put_qpel_pixels_tab[2][16];
190 qpel_mc_func avg_qpel_pixels_tab[2][16];
191 qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
192 qpel_mc_func put_mspel_pixels_tab[8];
194 me_cmp_func pix_abs[2][4];
196 /* huffyuv specific */
197 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
198 void (*diff_bytes)(uint8_t *dst/*align 16*/, const uint8_t *src1/*align 16*/, const uint8_t *src2/*align 1*/,int w);
200 * subtract huffyuv's variant of median prediction
201 * note, this might read from src1[-1], src2[-1]
203 void (*sub_hfyu_median_prediction)(uint8_t *dst, const uint8_t *src1, const uint8_t *src2, int w, int *left, int *left_top);
204 void (*add_hfyu_median_prediction)(uint8_t *dst, const uint8_t *top, const uint8_t *diff, int w, int *left, int *left_top);
205 int (*add_hfyu_left_prediction)(uint8_t *dst, const uint8_t *src, int w, int left);
206 void (*add_hfyu_left_prediction_bgr32)(uint8_t *dst, const uint8_t *src, int w, int *red, int *green, int *blue, int *alpha);
207 /* this might write to dst[w] */
208 void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
209 void (*bswap16_buf)(uint16_t *dst, const uint16_t *src, int len);
211 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
212 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
214 void (*h261_loop_filter)(uint8_t *src, int stride);
216 /* assume len is a multiple of 8, and arrays are 16-byte aligned */
217 void (*vector_clipf)(float *dst /* align 16 */, const float *src /* align 16 */, float min, float max, int len /* align 16 */);
220 void (*fdct)(int16_t *block/* align 16*/);
221 void (*fdct248)(int16_t *block/* align 16*/);
224 void (*idct)(int16_t *block/* align 16*/);
227 * block -> idct -> clip to unsigned 8 bit -> dest.
228 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
229 * @param line_size size in bytes of a horizontal line of dest
231 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, int16_t *block/*align 16*/);
234 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
235 * @param line_size size in bytes of a horizontal line of dest
237 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, int16_t *block/*align 16*/);
240 * idct input permutation.
241 * several optimized IDCTs need a permutated input (relative to the normal order of the reference
243 * this permutation must be performed before the idct_put/add, note, normally this can be merged
244 * with the zigzag/alternate scan<br>
245 * an example to avoid confusion:
246 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
247 * - (x -> reference dct -> reference idct -> x)
248 * - (x -> reference dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
249 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
251 uint8_t idct_permutation[64];
252 int idct_permutation_type;
253 #define FF_NO_IDCT_PERM 1
254 #define FF_LIBMPEG2_IDCT_PERM 2
255 #define FF_SIMPLE_IDCT_PERM 3
256 #define FF_TRANSPOSE_IDCT_PERM 4
257 #define FF_PARTTRANS_IDCT_PERM 5
258 #define FF_SSE2_IDCT_PERM 6
260 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
261 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
262 #define BASIS_SHIFT 16
263 #define RECON_SHIFT 6
265 void (*draw_edges)(uint8_t *buf, int wrap, int width, int height, int w, int h, int sides);
266 #define EDGE_WIDTH 16
268 #define EDGE_BOTTOM 2
270 void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
273 * Calculate scalar product of two vectors.
274 * @param len length of vectors, should be multiple of 16
276 int32_t (*scalarproduct_int16)(const int16_t *v1, const int16_t *v2/*align 16*/, int len);
279 * Calculate scalar product of v1 and v2,
280 * and v1[i] += v3[i] * mul
281 * @param len length of vectors, should be multiple of 16
283 int32_t (*scalarproduct_and_madd_int16)(int16_t *v1/*align 16*/, const int16_t *v2, const int16_t *v3, int len, int mul);
286 * Apply symmetric window in 16-bit fixed-point.
287 * @param output destination array
288 * constraints: 16-byte aligned
289 * @param input source array
290 * constraints: 16-byte aligned
291 * @param window window array
292 * constraints: 16-byte aligned, at least len/2 elements
293 * @param len full window length
294 * constraints: multiple of ? greater than zero
296 void (*apply_window_int16)(int16_t *output, const int16_t *input,
297 const int16_t *window, unsigned int len);
300 * Clip each element in an array of int32_t to a given minimum and maximum value.
301 * @param dst destination array
302 * constraints: 16-byte aligned
303 * @param src source array
304 * constraints: 16-byte aligned
305 * @param min minimum value
306 * constraints: must be in the range [-(1 << 24), 1 << 24]
307 * @param max maximum value
308 * constraints: must be in the range [-(1 << 24), 1 << 24]
309 * @param len number of elements in the array
310 * constraints: multiple of 32 greater than zero
312 void (*vector_clip_int32)(int32_t *dst, const int32_t *src, int32_t min,
313 int32_t max, unsigned int len);
315 op_fill_func fill_block_tab[2];
318 void ff_dsputil_static_init(void);
319 void ff_dsputil_init(DSPContext* p, AVCodecContext *avctx);
320 attribute_deprecated void dsputil_init(DSPContext* c, AVCodecContext *avctx);
322 int ff_check_alignment(void);
324 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
326 void ff_dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
327 void ff_dsputil_init_arm(DSPContext* c, AVCodecContext *avctx);
328 void ff_dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
329 void ff_dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
330 void ff_dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
331 void ff_dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
332 void ff_dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
334 void ff_dsputil_init_dwt(DSPContext *c);
336 #endif /* AVCODEC_DSPUTIL_H */