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
38 typedef short DCTELEM;
40 typedef short IDWTELEM;
42 void fdct_ifast (DCTELEM *data);
43 void fdct_ifast248 (DCTELEM *data);
44 void ff_jpeg_fdct_islow (DCTELEM *data);
45 void ff_fdct248_islow (DCTELEM *data);
47 void j_rev_dct (DCTELEM *data);
48 void j_rev_dct4 (DCTELEM *data);
49 void j_rev_dct2 (DCTELEM *data);
50 void j_rev_dct1 (DCTELEM *data);
52 void ff_fdct_mmx(DCTELEM *block);
53 void ff_fdct_mmx2(DCTELEM *block);
54 void ff_fdct_sse2(DCTELEM *block);
56 void ff_h264_idct8_add_c(uint8_t *dst, DCTELEM *block, int stride);
57 void ff_h264_idct_add_c(uint8_t *dst, DCTELEM *block, int stride);
58 void ff_h264_idct8_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
59 void ff_h264_idct_dc_add_c(uint8_t *dst, DCTELEM *block, int stride);
60 void ff_h264_lowres_idct_add_c(uint8_t *dst, int stride, DCTELEM *block);
61 void ff_h264_lowres_idct_put_c(uint8_t *dst, int stride, DCTELEM *block);
63 void ff_vector_fmul_add_add_c(float *dst, const float *src0, const float *src1,
64 const float *src2, int src3, int blocksize, int step);
65 void ff_float_to_int16_c(int16_t *dst, const float *src, int len);
68 extern const uint8_t ff_alternate_horizontal_scan[64];
69 extern const uint8_t ff_alternate_vertical_scan[64];
70 extern const uint8_t ff_zigzag_direct[64];
71 extern const uint8_t ff_zigzag248_direct[64];
73 /* pixel operations */
74 #define MAX_NEG_CROP 1024
77 extern uint32_t ff_squareTbl[512];
78 extern uint8_t ff_cropTbl[256 + 2 * MAX_NEG_CROP];
80 /* VP3 DSP functions */
81 void ff_vp3_idct_c(DCTELEM *block/* align 16*/);
82 void ff_vp3_idct_put_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
83 void ff_vp3_idct_add_c(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
85 /* 1/2^n downscaling functions from imgconvert.c */
86 void ff_img_copy_plane(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
87 void ff_shrink22(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
88 void ff_shrink44(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
89 void ff_shrink88(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
91 void ff_gmc_c(uint8_t *dst, uint8_t *src, int stride, int h, int ox, int oy,
92 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
94 /* minimum alignment rules ;)
95 if u notice errors in the align stuff, need more alignment for some asm code for some cpu
96 or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ...
98 !warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible)
99 i (michael) didnt check them, these are just the alignents which i think could be reached easily ...
101 !future video codecs might need functions with less strict alignment
105 void get_pixels_c(DCTELEM *block, const uint8_t *pixels, int line_size);
106 void diff_pixels_c(DCTELEM *block, const uint8_t *s1, const uint8_t *s2, int stride);
107 void put_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
108 void add_pixels_clamped_c(const DCTELEM *block, uint8_t *pixels, int line_size);
109 void clear_blocks_c(DCTELEM *blocks);
112 /* add and put pixel (decoding) */
113 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
114 //h for op_pixels_func is limited to {width/2, width} but never larger than 16 and never smaller then 4
115 typedef void (*op_pixels_func)(uint8_t *block/*align width (8 or 16)*/, const uint8_t *pixels/*align 1*/, int line_size, int h);
116 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);
117 typedef void (*qpel_mc_func)(uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
118 typedef void (*h264_chroma_mc_func)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x, int y);
119 typedef void (*h264_weight_func)(uint8_t *block, int stride, int log2_denom, int weight, int offset);
120 typedef void (*h264_biweight_func)(uint8_t *dst, uint8_t *src, int stride, int log2_denom, int weightd, int weights, int offset);
122 #define DEF_OLD_QPEL(name)\
123 void ff_put_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
124 void ff_put_no_rnd_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);\
125 void ff_avg_ ## name (uint8_t *dst/*align width (8 or 16)*/, uint8_t *src/*align 1*/, int stride);
127 DEF_OLD_QPEL(qpel16_mc11_old_c)
128 DEF_OLD_QPEL(qpel16_mc31_old_c)
129 DEF_OLD_QPEL(qpel16_mc12_old_c)
130 DEF_OLD_QPEL(qpel16_mc32_old_c)
131 DEF_OLD_QPEL(qpel16_mc13_old_c)
132 DEF_OLD_QPEL(qpel16_mc33_old_c)
133 DEF_OLD_QPEL(qpel8_mc11_old_c)
134 DEF_OLD_QPEL(qpel8_mc31_old_c)
135 DEF_OLD_QPEL(qpel8_mc12_old_c)
136 DEF_OLD_QPEL(qpel8_mc32_old_c)
137 DEF_OLD_QPEL(qpel8_mc13_old_c)
138 DEF_OLD_QPEL(qpel8_mc33_old_c)
140 #define CALL_2X_PIXELS(a, b, n)\
141 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
142 b(block , pixels , line_size, h);\
143 b(block+n, pixels+n, line_size, h);\
146 /* motion estimation */
147 // h is limited to {width/2, width, 2*width} but never larger than 16 and never smaller then 2
148 // although currently h<4 is not used as functions with width <8 are neither used nor implemented
149 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))*/;
153 typedef struct slice_buffer_s slice_buffer;
158 typedef struct DSPContext {
159 /* pixel ops : interface with DCT */
160 void (*get_pixels)(DCTELEM *block/*align 16*/, const uint8_t *pixels/*align 8*/, int line_size);
161 void (*diff_pixels)(DCTELEM *block/*align 16*/, const uint8_t *s1/*align 8*/, const uint8_t *s2/*align 8*/, int stride);
162 void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
163 void (*put_signed_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
164 void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, uint8_t *pixels/*align 8*/, int line_size);
165 void (*add_pixels8)(uint8_t *pixels, DCTELEM *block, int line_size);
166 void (*add_pixels4)(uint8_t *pixels, DCTELEM *block, int line_size);
167 int (*sum_abs_dctelem)(DCTELEM *block/*align 16*/);
169 * translational global motion compensation.
171 void (*gmc1)(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
173 * global motion compensation.
175 void (*gmc )(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int ox, int oy,
176 int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
177 void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
178 int (*pix_sum)(uint8_t * pix, int line_size);
179 int (*pix_norm1)(uint8_t * pix, int line_size);
180 // 16x16 8x8 4x4 2x2 16x8 8x4 4x2 8x16 4x8 2x4
182 me_cmp_func sad[5]; /* identical to pix_absAxA except additional void * */
184 me_cmp_func hadamard8_diff[5];
185 me_cmp_func dct_sad[5];
186 me_cmp_func quant_psnr[5];
194 me_cmp_func dct_max[5];
195 me_cmp_func dct264_sad[5];
197 me_cmp_func me_pre_cmp[5];
198 me_cmp_func me_cmp[5];
199 me_cmp_func me_sub_cmp[5];
200 me_cmp_func mb_cmp[5];
201 me_cmp_func ildct_cmp[5]; //only width 16 used
202 me_cmp_func frame_skip_cmp[5]; //only width 8 used
204 int (*ssd_int8_vs_int16)(const int8_t *pix1, const int16_t *pix2,
208 * Halfpel motion compensation with rounding (a+b+1)>>1.
209 * this is an array[4][4] of motion compensation functions for 4
210 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
211 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
212 * @param block destination where the result is stored
213 * @param pixels source
214 * @param line_size number of bytes in a horizontal line of block
217 op_pixels_func put_pixels_tab[4][4];
220 * Halfpel motion compensation with rounding (a+b+1)>>1.
221 * This is an array[4][4] of motion compensation functions for 4
222 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
223 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
224 * @param block destination into which the result is averaged (a+b+1)>>1
225 * @param pixels source
226 * @param line_size number of bytes in a horizontal line of block
229 op_pixels_func avg_pixels_tab[4][4];
232 * Halfpel motion compensation with no rounding (a+b)>>1.
233 * this is an array[2][4] of motion compensation functions for 2
234 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
235 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
236 * @param block destination where the result is stored
237 * @param pixels source
238 * @param line_size number of bytes in a horizontal line of block
241 op_pixels_func put_no_rnd_pixels_tab[4][4];
244 * Halfpel motion compensation with no rounding (a+b)>>1.
245 * this is an array[2][4] of motion compensation functions for 2
246 * horizontal blocksizes (8,16) and the 4 halfpel positions<br>
247 * *pixels_tab[ 0->16xH 1->8xH ][ xhalfpel + 2*yhalfpel ]
248 * @param block destination into which the result is averaged (a+b)>>1
249 * @param pixels source
250 * @param line_size number of bytes in a horizontal line of block
253 op_pixels_func avg_no_rnd_pixels_tab[4][4];
255 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);
258 * Thirdpel motion compensation with rounding (a+b+1)>>1.
259 * this is an array[12] of motion compensation functions for the 9 thirdpe
261 * *pixels_tab[ xthirdpel + 4*ythirdpel ]
262 * @param block destination where the result is stored
263 * @param pixels source
264 * @param line_size number of bytes in a horizontal line of block
267 tpel_mc_func put_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
268 tpel_mc_func avg_tpel_pixels_tab[11]; //FIXME individual func ptr per width?
270 qpel_mc_func put_qpel_pixels_tab[2][16];
271 qpel_mc_func avg_qpel_pixels_tab[2][16];
272 qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
273 qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
274 qpel_mc_func put_mspel_pixels_tab[8];
279 h264_chroma_mc_func put_h264_chroma_pixels_tab[3];
280 /* This is really one func used in VC-1 decoding */
281 h264_chroma_mc_func put_no_rnd_h264_chroma_pixels_tab[3];
282 h264_chroma_mc_func avg_h264_chroma_pixels_tab[3];
284 qpel_mc_func put_h264_qpel_pixels_tab[4][16];
285 qpel_mc_func avg_h264_qpel_pixels_tab[4][16];
287 qpel_mc_func put_2tap_qpel_pixels_tab[4][16];
288 qpel_mc_func avg_2tap_qpel_pixels_tab[4][16];
290 h264_weight_func weight_h264_pixels_tab[10];
291 h264_biweight_func biweight_h264_pixels_tab[10];
294 qpel_mc_func put_cavs_qpel_pixels_tab[2][16];
295 qpel_mc_func avg_cavs_qpel_pixels_tab[2][16];
296 void (*cavs_filter_lv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
297 void (*cavs_filter_lh)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
298 void (*cavs_filter_cv)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
299 void (*cavs_filter_ch)(uint8_t *pix, int stride, int alpha, int beta, int tc, int bs1, int bs2);
300 void (*cavs_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
302 me_cmp_func pix_abs[2][4];
304 /* huffyuv specific */
305 void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
306 void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
308 * subtract huffyuv's variant of median prediction
309 * note, this might read from src1[-1], src2[-1]
311 void (*sub_hfyu_median_prediction)(uint8_t *dst, uint8_t *src1, uint8_t *src2, int w, int *left, int *left_top);
312 void (*bswap_buf)(uint32_t *dst, uint32_t *src, int w);
314 void (*h264_v_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
315 void (*h264_h_loop_filter_luma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
316 void (*h264_v_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
317 void (*h264_h_loop_filter_chroma)(uint8_t *pix, int stride, int alpha, int beta, int8_t *tc0);
318 void (*h264_v_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
319 void (*h264_h_loop_filter_chroma_intra)(uint8_t *pix, int stride, int alpha, int beta);
320 // h264_loop_filter_strength: simd only. the C version is inlined in h264.c
321 void (*h264_loop_filter_strength)(int16_t bS[2][4][4], uint8_t nnz[40], int8_t ref[2][40], int16_t mv[2][40][2],
322 int bidir, int edges, int step, int mask_mv0, int mask_mv1);
324 void (*h263_v_loop_filter)(uint8_t *src, int stride, int qscale);
325 void (*h263_h_loop_filter)(uint8_t *src, int stride, int qscale);
327 void (*h261_loop_filter)(uint8_t *src, int stride);
329 /* assume len is a multiple of 4, and arrays are 16-byte aligned */
330 void (*vorbis_inverse_coupling)(float *mag, float *ang, int blocksize);
331 /* no alignment needed */
332 void (*flac_compute_autocorr)(const int32_t *data, int len, int lag, double *autoc);
333 /* assume len is a multiple of 8, and arrays are 16-byte aligned */
334 void (*vector_fmul)(float *dst, const float *src, int len);
335 void (*vector_fmul_reverse)(float *dst, const float *src0, const float *src1, int len);
336 /* assume len is a multiple of 8, and src arrays are 16-byte aligned */
337 void (*vector_fmul_add_add)(float *dst, const float *src0, const float *src1, const float *src2, int src3, int len, int step);
339 /* C version: convert floats from the range [384.0,386.0] to ints in [-32768,32767]
340 * simd versions: convert floats from [-32768.0,32767.0] without rescaling and arrays are 16byte aligned */
341 void (*float_to_int16)(int16_t *dst, const float *src, int len);
344 void (*fdct)(DCTELEM *block/* align 16*/);
345 void (*fdct248)(DCTELEM *block/* align 16*/);
348 void (*idct)(DCTELEM *block/* align 16*/);
351 * block -> idct -> clip to unsigned 8 bit -> dest.
352 * (-1392, 0, 0, ...) -> idct -> (-174, -174, ...) -> put -> (0, 0, ...)
353 * @param line_size size in bytes of a horizontal line of dest
355 void (*idct_put)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
358 * block -> idct -> add dest -> clip to unsigned 8 bit -> dest.
359 * @param line_size size in bytes of a horizontal line of dest
361 void (*idct_add)(uint8_t *dest/*align 8*/, int line_size, DCTELEM *block/*align 16*/);
364 * idct input permutation.
365 * several optimized IDCTs need a permutated input (relative to the normal order of the reference
367 * this permutation must be performed before the idct_put/add, note, normally this can be merged
368 * with the zigzag/alternate scan<br>
369 * an example to avoid confusion:
370 * - (->decode coeffs -> zigzag reorder -> dequant -> reference idct ->...)
371 * - (x -> referece dct -> reference idct -> x)
372 * - (x -> referece dct -> simple_mmx_perm = idct_permutation -> simple_idct_mmx -> x)
373 * - (->decode coeffs -> zigzag reorder -> simple_mmx_perm -> dequant -> simple_idct_mmx ->...)
375 uint8_t idct_permutation[64];
376 int idct_permutation_type;
377 #define FF_NO_IDCT_PERM 1
378 #define FF_LIBMPEG2_IDCT_PERM 2
379 #define FF_SIMPLE_IDCT_PERM 3
380 #define FF_TRANSPOSE_IDCT_PERM 4
381 #define FF_PARTTRANS_IDCT_PERM 5
383 int (*try_8x8basis)(int16_t rem[64], int16_t weight[64], int16_t basis[64], int scale);
384 void (*add_8x8basis)(int16_t rem[64], int16_t basis[64], int scale);
385 #define BASIS_SHIFT 16
386 #define RECON_SHIFT 6
389 void (*h264_idct_add)(uint8_t *dst, DCTELEM *block, int stride);
390 void (*h264_idct8_add)(uint8_t *dst, DCTELEM *block, int stride);
391 void (*h264_idct_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
392 void (*h264_idct8_dc_add)(uint8_t *dst, DCTELEM *block, int stride);
393 void (*h264_dct)(DCTELEM block[4][4]);
396 void (*vertical_compose97i)(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2, IDWTELEM *b3, IDWTELEM *b4, IDWTELEM *b5, int width);
397 void (*horizontal_compose97i)(IDWTELEM *b, int width);
398 void (*inner_add_yblock)(const uint8_t *obmc, const int obmc_stride, uint8_t * * block, int b_w, int b_h, int src_x, int src_y, int src_stride, slice_buffer * sb, int add, uint8_t * dst8);
400 void (*prefetch)(void *mem, int stride, int h);
402 void (*shrink[4])(uint8_t *dst, int dst_wrap, const uint8_t *src, int src_wrap, int width, int height);
405 void (*vc1_inv_trans_8x8)(DCTELEM *b);
406 void (*vc1_inv_trans_8x4)(DCTELEM *b, int n);
407 void (*vc1_inv_trans_4x8)(DCTELEM *b, int n);
408 void (*vc1_inv_trans_4x4)(DCTELEM *b, int n);
409 void (*vc1_v_overlap)(uint8_t* src, int stride);
410 void (*vc1_h_overlap)(uint8_t* src, int stride);
411 /* put 8x8 block with bicubic interpolation and quarterpel precision
412 * last argument is actually round value instead of height
414 op_pixels_func put_vc1_mspel_pixels_tab[16];
417 void dsputil_static_init(void);
418 void dsputil_init(DSPContext* p, AVCodecContext *avctx);
420 int ff_check_alignment(void);
423 * permute block according to permuatation.
424 * @param last last non zero element in scantable order
426 void ff_block_permute(DCTELEM *block, uint8_t *permutation, const uint8_t *scantable, int last);
428 void ff_set_cmp(DSPContext* c, me_cmp_func *cmp, int type);
430 #define BYTE_VEC32(c) ((c)*0x01010101UL)
432 static inline uint32_t rnd_avg32(uint32_t a, uint32_t b)
434 return (a | b) - (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
437 static inline uint32_t no_rnd_avg32(uint32_t a, uint32_t b)
439 return (a & b) + (((a ^ b) & ~BYTE_VEC32(0x01)) >> 1);
442 static inline int get_penalty_factor(int lambda, int lambda2, int type){
446 return lambda>>FF_LAMBDA_SHIFT;
448 return (3*lambda)>>(FF_LAMBDA_SHIFT+1);
450 return (4*lambda)>>(FF_LAMBDA_SHIFT);
452 return (2*lambda)>>(FF_LAMBDA_SHIFT);
455 return (2*lambda)>>FF_LAMBDA_SHIFT;
460 return lambda2>>FF_LAMBDA_SHIFT;
468 * this must be called between any dsp function and float/double code.
469 * for example sin(); dsp->idct_put(); emms_c(); cos()
473 /* should be defined by architectures supporting
474 one or more MultiMedia extension */
475 int mm_support(void);
477 void dsputil_init_alpha(DSPContext* c, AVCodecContext *avctx);
478 void dsputil_init_armv4l(DSPContext* c, AVCodecContext *avctx);
479 void dsputil_init_bfin(DSPContext* c, AVCodecContext *avctx);
480 void dsputil_init_mlib(DSPContext* c, AVCodecContext *avctx);
481 void dsputil_init_mmi(DSPContext* c, AVCodecContext *avctx);
482 void dsputil_init_mmx(DSPContext* c, AVCodecContext *avctx);
483 void dsputil_init_ppc(DSPContext* c, AVCodecContext *avctx);
484 void dsputil_init_sh4(DSPContext* c, AVCodecContext *avctx);
485 void dsputil_init_vis(DSPContext* c, AVCodecContext *avctx);
487 #define DECLARE_ALIGNED_16(t, v) DECLARE_ALIGNED(16, t, v)
489 #if defined(HAVE_MMX)
493 #define MM_MMX 0x0001 /* standard MMX */
494 #define MM_3DNOW 0x0004 /* AMD 3DNOW */
495 #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
496 #define MM_SSE 0x0008 /* SSE functions */
497 #define MM_SSE2 0x0010 /* PIV SSE2 functions */
498 #define MM_3DNOWEXT 0x0020 /* AMD 3DNowExt */
499 #define MM_SSE3 0x0040 /* Prescott SSE3 functions */
500 #define MM_SSSE3 0x0080 /* Conroe SSSE3 functions */
504 void add_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
505 void put_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
506 void put_signed_pixels_clamped_mmx(const DCTELEM *block, uint8_t *pixels, int line_size);
508 static inline void emms(void)
510 __asm __volatile ("emms;":::"memory");
516 if (mm_flags & MM_MMX)\
520 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
522 #define STRIDE_ALIGN 8
524 void dsputil_init_pix_mmx(DSPContext* c, AVCodecContext *avctx);
526 #elif defined(ARCH_ARMV4L)
528 /* This is to use 4 bytes read to the IDCT pointers for some 'zero'
529 line optimizations */
530 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(4, t, v)
531 #define STRIDE_ALIGN 4
533 #define MM_IWMMXT 0x0100 /* XScale IWMMXT */
537 #elif defined(HAVE_MLIB)
539 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
540 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
541 #define STRIDE_ALIGN 8
543 #elif defined(HAVE_VIS)
545 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
546 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
547 #define STRIDE_ALIGN 8
549 #elif defined(ARCH_ALPHA)
551 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
552 #define STRIDE_ALIGN 8
554 #elif defined(ARCH_POWERPC)
556 #define MM_ALTIVEC 0x0001 /* standard AltiVec */
560 #if defined(HAVE_ALTIVEC) && !defined(__APPLE_CC__)
561 #define pixel altivec_pixel
566 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
567 #define STRIDE_ALIGN 16
569 #elif defined(HAVE_MMI)
571 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(16, t, v)
572 #define STRIDE_ALIGN 16
574 #elif defined(ARCH_SH4)
576 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
577 #define STRIDE_ALIGN 8
579 #elif defined(ARCH_BFIN)
581 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
582 #define STRIDE_ALIGN 8
586 #define DECLARE_ALIGNED_8(t, v) DECLARE_ALIGNED(8, t, v)
587 #define STRIDE_ALIGN 8
592 void get_psnr(uint8_t *orig_image[3], uint8_t *coded_image[3],
593 int orig_linesize[3], int coded_linesize,
594 AVCodecContext *avctx);
596 /* FFT computation */
598 /* NOTE: soon integer code will be added, so you must use the
600 typedef float FFTSample;
604 typedef struct FFTComplex {
608 typedef struct FFTContext {
613 FFTComplex *exptab1; /* only used by SSE code */
614 void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
615 void (*imdct_calc)(struct MDCTContext *s, FFTSample *output,
616 const FFTSample *input, FFTSample *tmp);
619 int ff_fft_init(FFTContext *s, int nbits, int inverse);
620 void ff_fft_permute(FFTContext *s, FFTComplex *z);
621 void ff_fft_calc_c(FFTContext *s, FFTComplex *z);
622 void ff_fft_calc_sse(FFTContext *s, FFTComplex *z);
623 void ff_fft_calc_3dn(FFTContext *s, FFTComplex *z);
624 void ff_fft_calc_3dn2(FFTContext *s, FFTComplex *z);
625 void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z);
627 static inline void ff_fft_calc(FFTContext *s, FFTComplex *z)
631 void ff_fft_end(FFTContext *s);
633 /* MDCT computation */
635 typedef struct MDCTContext {
636 int n; /* size of MDCT (i.e. number of input data * 2) */
637 int nbits; /* n = 2^nbits */
638 /* pre/post rotation tables */
644 int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
645 void ff_imdct_calc(MDCTContext *s, FFTSample *output,
646 const FFTSample *input, FFTSample *tmp);
647 void ff_imdct_calc_3dn2(MDCTContext *s, FFTSample *output,
648 const FFTSample *input, FFTSample *tmp);
649 void ff_imdct_calc_sse(MDCTContext *s, FFTSample *output,
650 const FFTSample *input, FFTSample *tmp);
651 void ff_mdct_calc(MDCTContext *s, FFTSample *out,
652 const FFTSample *input, FFTSample *tmp);
653 void ff_mdct_end(MDCTContext *s);
655 #define WARPER8_16(name8, name16)\
656 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
657 return name8(s, dst , src , stride, h)\
658 +name8(s, dst+8 , src+8 , stride, h);\
661 #define WARPER8_16_SQ(name8, name16)\
662 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride, int h){\
664 score +=name8(s, dst , src , stride, 8);\
665 score +=name8(s, dst+8 , src+8 , stride, 8);\
669 score +=name8(s, dst , src , stride, 8);\
670 score +=name8(s, dst+8 , src+8 , stride, 8);\
676 static inline void copy_block2(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
681 AV_WN16(dst , AV_RN16(src ));
687 static inline void copy_block4(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
692 AV_WN32(dst , AV_RN32(src ));
698 static inline void copy_block8(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
703 AV_WN32(dst , AV_RN32(src ));
704 AV_WN32(dst+4 , AV_RN32(src+4 ));
710 static inline void copy_block9(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
715 AV_WN32(dst , AV_RN32(src ));
716 AV_WN32(dst+4 , AV_RN32(src+4 ));
723 static inline void copy_block16(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
728 AV_WN32(dst , AV_RN32(src ));
729 AV_WN32(dst+4 , AV_RN32(src+4 ));
730 AV_WN32(dst+8 , AV_RN32(src+8 ));
731 AV_WN32(dst+12, AV_RN32(src+12));
737 static inline void copy_block17(uint8_t *dst, uint8_t *src, int dstStride, int srcStride, int h)
742 AV_WN32(dst , AV_RN32(src ));
743 AV_WN32(dst+4 , AV_RN32(src+4 ));
744 AV_WN32(dst+8 , AV_RN32(src+8 ));
745 AV_WN32(dst+12, AV_RN32(src+12));