3 * Copyright (c) 2000, 2001, 2002 Fabrice Bellard
4 * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>
6 * This file is part of Libav.
8 * Libav 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 * Libav 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 Libav; 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);
47 void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
48 int dxx, int dxy, int dyx, int dyy, int shift, int r,
49 int width, int height);
51 /* minimum alignment rules ;)
52 * If you notice errors in the align stuff, need more alignment for some ASM code
53 * for some CPU or need to use a function with less aligned data then send a mail
54 * to the libav-devel mailing list, ...
56 * !warning These alignments might not match reality, (missing attribute((align))
57 * stuff somewhere possible).
58 * I (Michael) did not check them, these are just the alignments which I think
59 * could be reached easily ...
61 * !future video codecs might need functions with less strict alignment
64 /* add and put pixel (decoding)
65 * Block sizes for op_pixels_func are 8x4,8x8 16x8 16x16.
66 * h for op_pixels_func is limited to { width / 2, width },
67 * but never larger than 16 and never smaller than 4. */
68 typedef void (*qpel_mc_func)(uint8_t *dst /* align width (8 or 16) */,
69 uint8_t *src /* align 1 */, ptrdiff_t stride);
71 typedef void (*op_fill_func)(uint8_t *block /* align width (8 or 16) */,
72 uint8_t value, int line_size, int h);
74 #define DEF_OLD_QPEL(name) \
75 void ff_put_ ## name(uint8_t *dst /* align width (8 or 16) */, \
76 uint8_t *src /* align 1 */, ptrdiff_t stride); \
77 void ff_put_no_rnd_ ## name(uint8_t *dst /* align width (8 or 16) */, \
78 uint8_t *src /* align 1 */, ptrdiff_t stride); \
79 void ff_avg_ ## name(uint8_t *dst /* align width (8 or 16) */, \
80 uint8_t *src /* align 1 */, ptrdiff_t stride);
82 DEF_OLD_QPEL(qpel16_mc11_old_c)
83 DEF_OLD_QPEL(qpel16_mc31_old_c)
84 DEF_OLD_QPEL(qpel16_mc12_old_c)
85 DEF_OLD_QPEL(qpel16_mc32_old_c)
86 DEF_OLD_QPEL(qpel16_mc13_old_c)
87 DEF_OLD_QPEL(qpel16_mc33_old_c)
88 DEF_OLD_QPEL(qpel8_mc11_old_c)
89 DEF_OLD_QPEL(qpel8_mc31_old_c)
90 DEF_OLD_QPEL(qpel8_mc12_old_c)
91 DEF_OLD_QPEL(qpel8_mc32_old_c)
92 DEF_OLD_QPEL(qpel8_mc13_old_c)
93 DEF_OLD_QPEL(qpel8_mc33_old_c)
95 struct MpegEncContext;
97 * h is limited to { width / 2, width, 2 * width },
98 * but never larger than 16 and never smaller than 2.
99 * Although currently h < 4 is not used as functions with
100 * width < 8 are neither used nor implemented. */
101 typedef int (*me_cmp_func)(struct MpegEncContext *c,
102 uint8_t *blk1 /* align width (8 or 16) */,
103 uint8_t *blk2 /* align 1 */, int line_size, int h);
108 typedef struct ScanTable {
109 const uint8_t *scantable;
110 uint8_t permutated[64];
111 uint8_t raster_end[64];
114 void ff_init_scantable(uint8_t *permutation, ScanTable *st,
115 const uint8_t *src_scantable);
116 void ff_init_scantable_permutation(uint8_t *idct_permutation,
117 int idct_permutation_type);
122 typedef struct DSPContext {
123 /* pixel ops : interface with DCT */
124 void (*get_pixels)(int16_t *block /* align 16 */,
125 const uint8_t *pixels /* align 8 */,
127 void (*diff_pixels)(int16_t *block /* align 16 */,
128 const uint8_t *s1 /* align 8 */,
129 const uint8_t *s2 /* align 8 */,
131 void (*put_pixels_clamped)(const int16_t *block /* align 16 */,
132 uint8_t *pixels /* align 8 */,
134 void (*put_signed_pixels_clamped)(const int16_t *block /* align 16 */,
135 uint8_t *pixels /* align 8 */,
137 void (*add_pixels_clamped)(const int16_t *block /* align 16 */,
138 uint8_t *pixels /* align 8 */,
140 void (*add_pixels8)(uint8_t *pixels, int16_t *block, int line_size);
141 int (*sum_abs_dctelem)(int16_t *block /* align 16 */);
143 * translational global motion compensation.
145 void (*gmc1)(uint8_t *dst /* align 8 */, uint8_t *src /* align 1 */,
146 int srcStride, int h, int x16, int y16, int rounder);
148 * global motion compensation.
150 void (*gmc)(uint8_t *dst /* align 8 */, uint8_t *src /* align 1 */,
151 int stride, int h, int ox, int oy,
152 int dxx, int dxy, int dyx, int dyy,
153 int shift, int r, int width, int height);
154 void (*clear_block)(int16_t *block /* align 16 */);
155 void (*clear_blocks)(int16_t *blocks /* align 16 */);
156 int (*pix_sum)(uint8_t *pix, int line_size);
157 int (*pix_norm1)(uint8_t *pix, int line_size);
159 me_cmp_func sad[6]; /* identical to pix_absAxA except additional void * */
161 me_cmp_func hadamard8_diff[6];
162 me_cmp_func dct_sad[6];
163 me_cmp_func quant_psnr[6];
169 me_cmp_func dct_max[6];
170 me_cmp_func dct264_sad[6];
172 me_cmp_func me_pre_cmp[6];
173 me_cmp_func me_cmp[6];
174 me_cmp_func me_sub_cmp[6];
175 me_cmp_func mb_cmp[6];
176 me_cmp_func ildct_cmp[6]; // only width 16 used
177 me_cmp_func frame_skip_cmp[6]; // only width 8 used
179 int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
182 qpel_mc_func put_qpel_pixels_tab[2][16];
183 qpel_mc_func avg_qpel_pixels_tab[2][16];
184 qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
185 qpel_mc_func put_mspel_pixels_tab[8];
187 me_cmp_func pix_abs[2][4];
189 /* HuffYUV specific */
190 void (*add_bytes)(uint8_t *dst /* align 16 */,
191 uint8_t *src /* align 16 */,
193 void (*diff_bytes)(uint8_t *dst /* align 16 */,
194 uint8_t *src1 /* align 16 */,
195 uint8_t *src2 /* align 1 */,
198 * Subtract HuffYUV's variant of median prediction.
199 * Note, this might read from src1[-1], src2[-1].
201 void (*sub_hfyu_median_prediction)(uint8_t *dst, const uint8_t *src1,
202 const uint8_t *src2, int w,
203 int *left, int *left_top);
204 void (*add_hfyu_median_prediction)(uint8_t *dst, const uint8_t *top,
205 const uint8_t *diff, int w,
206 int *left, int *left_top);
207 int (*add_hfyu_left_prediction)(uint8_t *dst, const uint8_t *src,
209 void (*add_hfyu_left_prediction_bgr32)(uint8_t *dst, const uint8_t *src,
210 int w, int *red, int *green,
211 int *blue, int *alpha);
212 void (*bswap_buf)(uint32_t *dst, const uint32_t *src, int w);
213 void (*bswap16_buf)(uint16_t *dst, const uint16_t *src, int len);
215 /* assume len is a multiple of 8, and arrays are 16-byte aligned */
216 void (*vector_clipf)(float *dst /* align 16 */,
217 const float *src /* align 16 */,
218 float min, float max, int len /* align 16 */);
221 void (*fdct)(int16_t *block /* align 16 */);
222 void (*fdct248)(int16_t *block /* align 16 */);
225 void (*idct)(int16_t *block /* align 16 */);
228 * block -> idct -> clip to unsigned 8 bit -> dest.
229 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
230 * @param line_size size in bytes of a horizontal line of dest
232 void (*idct_put)(uint8_t *dest /* align 8 */,
233 int line_size, int16_t *block /* align 16 */);
236 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
237 * @param line_size size in bytes of a horizontal line of dest
239 void (*idct_add)(uint8_t *dest /* align 8 */,
240 int line_size, int16_t *block /* align 16 */);
243 * IDCT input permutation.
244 * Several optimized IDCTs need a permutated input (relative to the
245 * normal order of the reference IDCT).
246 * This permutation must be performed before the idct_put/add.
247 * Note, normally this can be merged with the zigzag/alternate scan<br>
248 * An example to avoid confusion:
249 * - (->decode coeffs -> zigzag reorder -> dequant -> reference IDCT -> ...)
250 * - (x -> reference DCT -> reference IDCT -> x)
251 * - (x -> reference DCT -> simple_mmx_perm = idct_permutation
252 * -> simple_idct_mmx -> x)
253 * - (-> decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant
254 * -> simple_idct_mmx -> ...)
256 uint8_t idct_permutation[64];
257 int idct_permutation_type;
258 #define FF_NO_IDCT_PERM 1
259 #define FF_LIBMPEG2_IDCT_PERM 2
260 #define FF_SIMPLE_IDCT_PERM 3
261 #define FF_TRANSPOSE_IDCT_PERM 4
262 #define FF_PARTTRANS_IDCT_PERM 5
263 #define FF_SSE2_IDCT_PERM 6
265 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64],
266 int16_t basis[64], int scale);
267 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
268 #define BASIS_SHIFT 16
269 #define RECON_SHIFT 6
271 void (*draw_edges)(uint8_t *buf, int wrap, int width, int height,
272 int w, int h, int sides);
273 #define EDGE_WIDTH 16
275 #define EDGE_BOTTOM 2
277 void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src,
278 int src_wrap, int width, int height);
281 * Calculate scalar product of two vectors.
282 * @param len length of vectors, should be multiple of 16
284 int32_t (*scalarproduct_int16)(const int16_t *v1,
285 const int16_t *v2 /* align 16 */, int len);
288 * Calculate scalar product of v1 and v2,
289 * and v1[i] += v3[i] * mul
290 * @param len length of vectors, should be multiple of 16
292 int32_t (*scalarproduct_and_madd_int16)(int16_t *v1 /* align 16 */,
298 * Clip each element in an array of int32_t to a given minimum and
300 * @param dst destination array
301 * constraints: 16-byte aligned
302 * @param src source array
303 * constraints: 16-byte aligned
304 * @param min minimum value
305 * constraints: must be in the range [-(1 << 24), 1 << 24]
306 * @param max maximum value
307 * constraints: must be in the range [-(1 << 24), 1 << 24]
308 * @param len number of elements in the array
309 * constraints: multiple of 32 greater than zero
311 void (*vector_clip_int32)(int32_t *dst, const int32_t *src, int32_t min,
312 int32_t max, unsigned int len);
314 op_fill_func fill_block_tab[2];
317 void ff_dsputil_static_init(void);
318 void ff_dsputil_init(DSPContext *p, AVCodecContext *avctx);
320 void ff_set_cmp(DSPContext *c, me_cmp_func *cmp, int type);
322 void ff_dsputil_init_arm(DSPContext *c, AVCodecContext *avctx,
323 unsigned high_bit_depth);
324 void ff_dsputil_init_bfin(DSPContext *c, AVCodecContext *avctx,
325 unsigned high_bit_depth);
326 void ff_dsputil_init_ppc(DSPContext *c, AVCodecContext *avctx,
327 unsigned high_bit_depth);
328 void ff_dsputil_init_x86(DSPContext *c, AVCodecContext *avctx,
329 unsigned high_bit_depth);
331 #endif /* AVCODEC_DSPUTIL_H */