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
36 extern const uint8_t ff_alternate_horizontal_scan[64];
37 extern const uint8_t ff_alternate_vertical_scan[64];
38 extern const uint8_t ff_zigzag248_direct[64];
40 extern uint32_t ff_square_tab[512];
42 void ff_put_pixels8x8_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
43 void ff_avg_pixels8x8_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
44 void ff_put_pixels16x16_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
45 void ff_avg_pixels16x16_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
48 void ff_put_rv40_qpel16_mc33_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
49 void ff_avg_rv40_qpel16_mc33_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
50 void ff_put_rv40_qpel8_mc33_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
51 void ff_avg_rv40_qpel8_mc33_c(uint8_t *dst, uint8_t *src, ptrdiff_t stride);
53 void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
54 int dxx, int dxy, int dyx, int dyy, int shift, int r,
55 int width, int height);
57 /* minimum alignment rules ;)
58 * If you notice errors in the align stuff, need more alignment for some ASM code
59 * for some CPU or need to use a function with less aligned data then send a mail
60 * to the ffmpeg-devel mailing list, ...
62 * !warning These alignments might not match reality, (missing attribute((align))
63 * stuff somewhere possible).
64 * I (Michael) did not check them, these are just the alignments which I think
65 * could be reached easily ...
67 * !future video codecs might need functions with less strict alignment
70 /* add and put pixel (decoding)
71 * Block sizes for op_pixels_func are 8x4,8x8 16x8 16x16.
72 * h for op_pixels_func is limited to { width / 2, width },
73 * but never larger than 16 and never smaller than 4. */
74 typedef void (*qpel_mc_func)(uint8_t *dst /* align width (8 or 16) */,
75 uint8_t *src /* align 1 */, ptrdiff_t stride);
77 typedef void (*op_fill_func)(uint8_t *block /* align width (8 or 16) */,
78 uint8_t value, int line_size, int h);
80 #define DEF_OLD_QPEL(name) \
81 void ff_put_ ## name(uint8_t *dst /* align width (8 or 16) */, \
82 uint8_t *src /* align 1 */, ptrdiff_t stride); \
83 void ff_put_no_rnd_ ## name(uint8_t *dst /* align width (8 or 16) */, \
84 uint8_t *src /* align 1 */, ptrdiff_t stride); \
85 void ff_avg_ ## name(uint8_t *dst /* align width (8 or 16) */, \
86 uint8_t *src /* align 1 */, ptrdiff_t stride);
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)
101 struct MpegEncContext;
102 /* Motion estimation:
103 * h is limited to { width / 2, width, 2 * width },
104 * but never larger than 16 and never smaller than 2.
105 * Although currently h < 4 is not used as functions with
106 * width < 8 are neither used nor implemented. */
107 typedef int (*me_cmp_func)(struct MpegEncContext *c,
108 uint8_t *blk1 /* align width (8 or 16) */,
109 uint8_t *blk2 /* align 1 */, int line_size, int h);
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 *permutation, ScanTable *st,
121 const uint8_t *src_scantable);
122 void ff_init_scantable_permutation(uint8_t *idct_permutation,
123 int idct_permutation_type);
128 typedef struct DSPContext {
129 /* pixel ops : interface with DCT */
130 void (*get_pixels)(int16_t *block /* align 16 */,
131 const uint8_t *pixels /* align 8 */,
133 void (*diff_pixels)(int16_t *block /* align 16 */,
134 const uint8_t *s1 /* align 8 */,
135 const uint8_t *s2 /* align 8 */,
137 void (*put_pixels_clamped)(const int16_t *block /* align 16 */,
138 uint8_t *pixels /* align 8 */,
140 void (*put_signed_pixels_clamped)(const int16_t *block /* align 16 */,
141 uint8_t *pixels /* align 8 */,
143 void (*add_pixels_clamped)(const int16_t *block /* align 16 */,
144 uint8_t *pixels /* align 8 */,
146 void (*add_pixels8)(uint8_t *pixels, int16_t *block, int line_size);
147 int (*sum_abs_dctelem)(int16_t *block /* align 16 */);
149 * translational global motion compensation.
151 void (*gmc1)(uint8_t *dst /* align 8 */, uint8_t *src /* align 1 */,
152 int srcStride, int h, int x16, int y16, int rounder);
154 * global motion compensation.
156 void (*gmc)(uint8_t *dst /* align 8 */, uint8_t *src /* align 1 */,
157 int stride, int h, int ox, int oy,
158 int dxx, int dxy, int dyx, int dyy,
159 int shift, int r, int width, int height);
160 void (*clear_block)(int16_t *block /* align 16 */);
161 void (*clear_blocks)(int16_t *blocks /* align 16 */);
162 int (*pix_sum)(uint8_t *pix, int line_size);
163 int (*pix_norm1)(uint8_t *pix, int line_size);
165 me_cmp_func sad[6]; /* identical to pix_absAxA except additional void * */
167 me_cmp_func hadamard8_diff[6];
168 me_cmp_func dct_sad[6];
169 me_cmp_func quant_psnr[6];
177 me_cmp_func dct_max[6];
178 me_cmp_func dct264_sad[6];
180 me_cmp_func me_pre_cmp[6];
181 me_cmp_func me_cmp[6];
182 me_cmp_func me_sub_cmp[6];
183 me_cmp_func mb_cmp[6];
184 me_cmp_func ildct_cmp[6]; // only width 16 used
185 me_cmp_func frame_skip_cmp[6]; // only width 8 used
187 int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
190 qpel_mc_func put_qpel_pixels_tab[2][16];
191 qpel_mc_func avg_qpel_pixels_tab[2][16];
192 qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
193 qpel_mc_func put_mspel_pixels_tab[8];
195 me_cmp_func pix_abs[2][4];
197 /* HuffYUV specific */
198 void (*add_bytes)(uint8_t *dst /* align 16 */,
199 uint8_t *src /* align 16 */,
201 void (*diff_bytes)(uint8_t *dst /* align 16 */,
202 const uint8_t *src1 /* align 16 */,
203 const uint8_t *src2 /* align 1 */,
206 * Subtract HuffYUV's variant of median prediction.
207 * Note, this might read from src1[-1], src2[-1].
209 void (*sub_hfyu_median_prediction)(uint8_t *dst, const uint8_t *src1,
210 const uint8_t *src2, int w,
211 int *left, int *left_top);
212 void (*add_hfyu_median_prediction)(uint8_t *dst, const uint8_t *top,
213 const uint8_t *diff, int w,
214 int *left, int *left_top);
215 int (*add_hfyu_left_prediction)(uint8_t *dst, const uint8_t *src,
217 void (*add_hfyu_left_prediction_bgr32)(uint8_t *dst, const uint8_t *src,
218 int w, int *red, int *green,
219 int *blue, int *alpha);
220 /* this might write to dst[w] */
221 void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
222 void (*bswap16_buf)(uint16_t *dst, const uint16_t *src, int len);
224 /* assume len is a multiple of 8, and arrays are 16-byte aligned */
225 void (*vector_clipf)(float *dst /* align 16 */,
226 const float *src /* align 16 */,
227 float min, float max, int len /* align 16 */);
230 void (*fdct)(int16_t *block /* align 16 */);
231 void (*fdct248)(int16_t *block /* align 16 */);
234 void (*idct)(int16_t *block /* align 16 */);
237 * block -> idct -> clip to unsigned 8 bit -> dest.
238 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
239 * @param line_size size in bytes of a horizontal line of dest
241 void (*idct_put)(uint8_t *dest /* align 8 */,
242 int line_size, int16_t *block /* align 16 */);
245 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
246 * @param line_size size in bytes of a horizontal line of dest
248 void (*idct_add)(uint8_t *dest /* align 8 */,
249 int line_size, int16_t *block /* align 16 */);
252 * IDCT input permutation.
253 * Several optimized IDCTs need a permutated input (relative to the
254 * normal order of the reference IDCT).
255 * This permutation must be performed before the idct_put/add.
256 * Note, normally this can be merged with the zigzag/alternate scan<br>
257 * An example to avoid confusion:
258 * - (->decode coeffs -> zigzag reorder -> dequant -> reference IDCT -> ...)
259 * - (x -> reference DCT -> reference IDCT -> x)
260 * - (x -> reference DCT -> simple_mmx_perm = idct_permutation
261 * -> simple_idct_mmx -> x)
262 * - (-> decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant
263 * -> simple_idct_mmx -> ...)
265 uint8_t idct_permutation[64];
266 int idct_permutation_type;
267 #define FF_NO_IDCT_PERM 1
268 #define FF_LIBMPEG2_IDCT_PERM 2
269 #define FF_SIMPLE_IDCT_PERM 3
270 #define FF_TRANSPOSE_IDCT_PERM 4
271 #define FF_PARTTRANS_IDCT_PERM 5
272 #define FF_SSE2_IDCT_PERM 6
274 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64],
275 int16_t basis[64], int scale);
276 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
277 #define BASIS_SHIFT 16
278 #define RECON_SHIFT 6
280 void (*draw_edges)(uint8_t *buf, int wrap, int width, int height,
281 int w, int h, int sides);
282 #define EDGE_WIDTH 16
284 #define EDGE_BOTTOM 2
286 void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src,
287 int src_wrap, int width, int height);
290 * Calculate scalar product of two vectors.
291 * @param len length of vectors, should be multiple of 16
293 int32_t (*scalarproduct_int16)(const int16_t *v1,
294 const int16_t *v2 /* align 16 */, int len);
297 * Calculate scalar product of v1 and v2,
298 * and v1[i] += v3[i] * mul
299 * @param len length of vectors, should be multiple of 16
301 int32_t (*scalarproduct_and_madd_int16)(int16_t *v1 /* align 16 */,
307 * Clip each element in an array of int32_t to a given minimum and
309 * @param dst destination array
310 * constraints: 16-byte aligned
311 * @param src source array
312 * constraints: 16-byte aligned
313 * @param min minimum value
314 * constraints: must be in the range [-(1 << 24), 1 << 24]
315 * @param max maximum value
316 * constraints: must be in the range [-(1 << 24), 1 << 24]
317 * @param len number of elements in the array
318 * constraints: multiple of 32 greater than zero
320 void (*vector_clip_int32)(int32_t *dst, const int32_t *src, int32_t min,
321 int32_t max, unsigned int len);
323 op_fill_func fill_block_tab[2];
326 void ff_dsputil_static_init(void);
327 void ff_dsputil_init(DSPContext *p, AVCodecContext *avctx);
328 void avpriv_dsputil_init(DSPContext* p, AVCodecContext *avctx);
329 attribute_deprecated void dsputil_init(DSPContext* c, AVCodecContext *avctx);
331 int ff_check_alignment(void);
333 void ff_set_cmp(DSPContext *c, me_cmp_func *cmp, int type);
335 void ff_dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
336 void ff_dsputil_init_arm(DSPContext *c, AVCodecContext *avctx,
337 unsigned high_bit_depth);
338 void ff_dsputil_init_bfin(DSPContext *c, AVCodecContext *avctx,
339 unsigned high_bit_depth);
340 void ff_dsputil_init_ppc(DSPContext *c, AVCodecContext *avctx,
341 unsigned high_bit_depth);
342 void ff_dsputil_init_x86(DSPContext *c, AVCodecContext *avctx,
343 unsigned high_bit_depth);
345 void ff_dsputil_init_dwt(DSPContext *c);
347 #endif /* AVCODEC_DSPUTIL_H */