libavcodec/dsputil.h

   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  */
  19 #ifndef DSPUTIL_H
  20 #define DSPUTIL_H
  21
  22 #include "common.h"
  23 #include "avcodec.h"
  24
  25 //#define DEBUG
  26 /* dct code */
  27 typedef short DCTELEM;
  28
  29 void fdct_ifast (DCTELEM *data);
  30 void ff_jpeg_fdct_islow (DCTELEM *data);
  31
  32 void j_rev_dct (DCTELEM *data);
  33
  34 void ff_fdct_mmx(DCTELEM *block);
  35
  36 /* encoding scans */
  37 extern const UINT8 ff_alternate_horizontal_scan[64];
  38 extern const UINT8 ff_alternate_vertical_scan[64];
  39 extern const UINT8 ff_zigzag_direct[64];
  40
  41 /* pixel operations */
  42 #define MAX_NEG_CROP 384
  43
  44 /* temporary */
  45 extern UINT32 squareTbl[512];
  46 extern UINT8 cropTbl[256 + 2 * MAX_NEG_CROP];
  47
  48
  49 /* minimum alignment rules ;)
  50 if u notice errors in the align stuff, need more alignment for some asm code for some cpu
  51 or need to use a function with less aligned data then send a mail to the ffmpeg-dev list, ...
  52
  53 !warning these alignments might not match reallity, (missing attribute((align)) stuff somewhere possible)
  54 i (michael) didnt check them, these are just the alignents which i think could be reached easily ...
  55
  56 !future video codecs might need functions with less strict alignment
  57 */
  58
  59 /*
  60 void get_pixels_c(DCTELEM *block, const UINT8 *pixels, int line_size);
  61 void diff_pixels_c(DCTELEM *block, const UINT8 *s1, const UINT8 *s2, int stride);
  62 void put_pixels_clamped_c(const DCTELEM *block, UINT8 *pixels, int line_size);
  63 void add_pixels_clamped_c(const DCTELEM *block, UINT8 *pixels, int line_size);
  64 void clear_blocks_c(DCTELEM *blocks);
  65 */
  66
  67 /* add and put pixel (decoding) */
  68 // blocksizes for op_pixels_func are 8x4,8x8 16x8 16x16
  69 typedef void (*op_pixels_func)(UINT8 *block/*align width (8 or 16)*/, const UINT8 *pixels/*align 1*/, int line_size, int h);
  70 typedef void (*qpel_mc_func)(UINT8 *dst/*align width (8 or 16)*/, UINT8 *src/*align 1*/, int stride);
  71
  72
  73
  74 #define CALL_2X_PIXELS(a, b, n)\
  75 static void a(uint8_t *block, const uint8_t *pixels, int line_size, int h){\
  76     b(block  , pixels  , line_size, h);\
  77     b(block+n, pixels+n, line_size, h);\
  78 }
  79
  80 /* motion estimation */
  81
  82 typedef int (*op_pixels_abs_func)(UINT8 *blk1/*align width (8 or 16)*/, UINT8 *blk2/*align 1*/, int line_size)/* __attribute__ ((const))*/;
  83
  84 typedef int (*me_cmp_func)(void /*MpegEncContext*/ *s, UINT8 *blk1/*align width (8 or 16)*/, UINT8 *blk2/*align 1*/, int line_size)/* __attribute__ ((const))*/;
  85
  86 typedef struct DSPContext {
  87     /* pixel ops : interface with DCT */
  88     void (*get_pixels)(DCTELEM *block/*align 16*/, const UINT8 *pixels/*align 8*/, int line_size);
  89     void (*diff_pixels)(DCTELEM *block/*align 16*/, const UINT8 *s1/*align 8*/, const UINT8 *s2/*align 8*/, int stride);
  90     void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, UINT8 *pixels/*align 8*/, int line_size);
  91     void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, UINT8 *pixels/*align 8*/, int line_size);
  92     void (*gmc1)(UINT8 *dst/*align 8*/, UINT8 *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
  93     void (*gmc )(UINT8 *dst/*align 8*/, UINT8 *src/*align 1*/, int stride, int h, int ox, int oy,
  94                     int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
  95     void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
  96     int (*pix_sum)(UINT8 * pix, int line_size);
  97     int (*pix_norm1)(UINT8 * pix, int line_size);
  98     me_cmp_func sad[2]; /* identical to pix_absAxA except additional void * */
  99     me_cmp_func sse[2];
 100     me_cmp_func hadamard8_diff[2];
 101     me_cmp_func dct_sad[2];
 102     me_cmp_func quant_psnr[2];
 103     int (*hadamard8_abs )(uint8_t *src, int stride, int mean);
 104
 105     me_cmp_func me_pre_cmp[11];
 106     me_cmp_func me_cmp[11];
 107     me_cmp_func me_sub_cmp[11];
 108     me_cmp_func mb_cmp[11];
 109
 110     /* maybe create an array for 16/8 functions */
 111     op_pixels_func put_pixels_tab[2][4];
 112     op_pixels_func avg_pixels_tab[2][4];
 113     op_pixels_func put_no_rnd_pixels_tab[2][4];
 114     op_pixels_func avg_no_rnd_pixels_tab[2][4];
 115     qpel_mc_func put_qpel_pixels_tab[2][16];
 116     qpel_mc_func avg_qpel_pixels_tab[2][16];
 117     qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
 118     qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
 119     qpel_mc_func put_mspel_pixels_tab[8];
 120
 121     op_pixels_abs_func pix_abs16x16;
 122     op_pixels_abs_func pix_abs16x16_x2;
 123     op_pixels_abs_func pix_abs16x16_y2;
 124     op_pixels_abs_func pix_abs16x16_xy2;
 125     op_pixels_abs_func pix_abs8x8;
 126     op_pixels_abs_func pix_abs8x8_x2;
 127     op_pixels_abs_func pix_abs8x8_y2;
 128     op_pixels_abs_func pix_abs8x8_xy2;
 129
 130     /* huffyuv specific */
 131     void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
 132     void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 1*/,int w);
 133 } DSPContext;
 134
 135 void dsputil_init(DSPContext* p, unsigned mask);
 136
 137 /**
 138  * permute block according to permuatation.
 139  * @param last last non zero element in scantable order
 140  */
 141 void ff_block_permute(INT16 *block, UINT8 *permutation, const UINT8 *scantable, int last);
 142
 143 #define emms_c()
 144
 145 #if defined(HAVE_MMX)
 146
 147 #undef emms_c
 148
 149 #define MM_MMX    0x0001 /* standard MMX */
 150 #define MM_3DNOW  0x0004 /* AMD 3DNOW */
 151 #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
 152 #define MM_SSE    0x0008 /* SSE functions */
 153 #define MM_SSE2   0x0010 /* PIV SSE2 functions */
 154
 155 extern int mm_flags;
 156
 157 int mm_support(void);
 158 void add_pixels_clamped_mmx(const DCTELEM *block, UINT8 *pixels, int line_size);
 159 void put_pixels_clamped_mmx(const DCTELEM *block, UINT8 *pixels, int line_size);
 160
 161 static inline void emms(void)
 162 {
 163     __asm __volatile ("emms;":::"memory");
 164 }
 165
 166
 167 #define emms_c() \
 168 {\
 169     if (mm_flags & MM_MMX)\
 170         emms();\
 171 }
 172
 173 #define __align8 __attribute__ ((aligned (8)))
 174
 175 void dsputil_init_mmx(DSPContext* c, unsigned mask);
 176 void dsputil_set_bit_exact_mmx(DSPContext* c, unsigned mask);
 177
 178 #elif defined(ARCH_ARMV4L)
 179
 180 /* This is to use 4 bytes read to the IDCT pointers for some 'zero'
 181    line ptimizations */
 182 #define __align8 __attribute__ ((aligned (4)))
 183
 184 void dsputil_init_armv4l(DSPContext* c, unsigned mask);
 185
 186 #elif defined(HAVE_MLIB)
 187
 188 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
 189 #define __align8 __attribute__ ((aligned (8)))
 190
 191 void dsputil_init_mlib(DSPContext* c, unsigned mask);
 192
 193 #elif defined(ARCH_ALPHA)
 194
 195 #define __align8 __attribute__ ((aligned (8)))
 196
 197 void dsputil_init_alpha(DSPContext* c, unsigned mask);
 198
 199 #elif defined(ARCH_POWERPC)
 200
 201 #define MM_ALTIVEC    0x0001 /* standard AltiVec */
 202
 203 extern int mm_flags;
 204
 205 #define __align8 __attribute__ ((aligned (16)))
 206
 207 void dsputil_init_ppc(DSPContext* c, unsigned mask);
 208
 209 #elif defined(HAVE_MMI)
 210
 211 #define __align8 __attribute__ ((aligned (16)))
 212
 213 void dsputil_init_mmi(DSPContext* c, unsigned mask);
 214
 215 #else
 216
 217 #define __align8
 218
 219 #endif
 220
 221 #ifdef __GNUC__
 222
 223 struct unaligned_64 { uint64_t l; } __attribute__((packed));
 224 struct unaligned_32 { uint32_t l; } __attribute__((packed));
 225
 226 #define LD32(a) (((const struct unaligned_32 *) (a))->l)
 227 #define LD64(a) (((const struct unaligned_64 *) (a))->l)
 228
 229 #define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)
 230
 231 #else /* __GNUC__ */
 232
 233 #define LD32(a) (*((uint32_t*)(a)))
 234 #define LD64(a) (*((uint64_t*)(a)))
 235
 236 #define ST32(a, b) *((uint32_t*)(a)) = (b)
 237
 238 #endif /* !__GNUC__ */
 239
 240 /* PSNR */
 241 void get_psnr(UINT8 *orig_image[3], UINT8 *coded_image[3],
 242               int orig_linesize[3], int coded_linesize,
 243               AVCodecContext *avctx);
 244
 245 /* FFT computation */
 246
 247 /* NOTE: soon integer code will be added, so you must use the
 248    FFTSample type */
 249 typedef float FFTSample;
 250
 251 typedef struct FFTComplex {
 252     FFTSample re, im;
 253 } FFTComplex;
 254
 255 typedef struct FFTContext {
 256     int nbits;
 257     int inverse;
 258     uint16_t *revtab;
 259     FFTComplex *exptab;
 260     FFTComplex *exptab1; /* only used by SSE code */
 261     void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
 262 } FFTContext;
 263
 264 int fft_init(FFTContext *s, int nbits, int inverse);
 265 void fft_permute(FFTContext *s, FFTComplex *z);
 266 void fft_calc_c(FFTContext *s, FFTComplex *z);
 267 void fft_calc_sse(FFTContext *s, FFTComplex *z);
 268 static inline void fft_calc(FFTContext *s, FFTComplex *z)
 269 {
 270     s->fft_calc(s, z);
 271 }
 272 void fft_end(FFTContext *s);
 273
 274 /* MDCT computation */
 275
 276 typedef struct MDCTContext {
 277     int n;  /* size of MDCT (i.e. number of input data * 2) */
 278     int nbits; /* n = 2^nbits */
 279     /* pre/post rotation tables */
 280     FFTSample *tcos;
 281     FFTSample *tsin;
 282     FFTContext fft;
 283 } MDCTContext;
 284
 285 int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
 286 void ff_imdct_calc(MDCTContext *s, FFTSample *output,
 287                 const FFTSample *input, FFTSample *tmp);
 288 void ff_mdct_calc(MDCTContext *s, FFTSample *out,
 289                const FFTSample *input, FFTSample *tmp);
 290 void ff_mdct_end(MDCTContext *s);
 291
 292 #define WARPER88_1616(name8, name16)\
 293 static int name16(void /*MpegEncContext*/ *s, uint8_t *dst, uint8_t *src, int stride){\
 294     return name8(s, dst           , src           , stride)\
 295           +name8(s, dst+8         , src+8         , stride)\
 296           +name8(s, dst  +8*stride, src  +8*stride, stride)\
 297           +name8(s, dst+8+8*stride, src+8+8*stride, stride);\
 298 }
 299
 300 #ifndef HAVE_LRINTF
 301 /* XXX: add ISOC specific test to avoid specific BSD testing. */
 302 /* better than nothing implementation. */
 303 /* btw, rintf() is existing on fbsd too -- alex */
 304 static inline long int lrintf(float x)
 305 {
 306     return (int)(rint(x));
 307 }
 308 #endif
 309
 310 #endif