hvirtual/quicktime/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);
  83 /*
  84 int pix_abs16x16_c(UINT8 *blk1, UINT8 *blk2, int lx);
  85 int pix_abs16x16_x2_c(UINT8 *blk1, UINT8 *blk2, int lx);
  86 int pix_abs16x16_y2_c(UINT8 *blk1, UINT8 *blk2, int lx);
  87 int pix_abs16x16_xy2_c(UINT8 *blk1, UINT8 *blk2, int lx);
  88 */
  89 typedef struct DSPContext {
  90     /* pixel ops : interface with DCT */
  91     void (*get_pixels)(DCTELEM *block/*align 16*/, const UINT8 *pixels/*align 8*/, int line_size);
  92     void (*diff_pixels)(DCTELEM *block/*align 16*/, const UINT8 *s1/*align 8*/, const UINT8 *s2/*align 8*/, int stride);
  93     void (*put_pixels_clamped)(const DCTELEM *block/*align 16*/, UINT8 *pixels/*align 8*/, int line_size);
  94     void (*add_pixels_clamped)(const DCTELEM *block/*align 16*/, UINT8 *pixels/*align 8*/, int line_size);
  95     void (*gmc1)(UINT8 *dst/*align 8*/, UINT8 *src/*align 1*/, int srcStride, int h, int x16, int y16, int rounder);
  96     void (*gmc )(UINT8 *dst/*align 8*/, UINT8 *src/*align 1*/, int stride, int h, int ox, int oy,
  97                     int dxx, int dxy, int dyx, int dyy, int shift, int r, int width, int height);
  98     void (*clear_blocks)(DCTELEM *blocks/*align 16*/);
  99     int (*pix_sum)(UINT8 * pix, int line_size);
 100     int (*pix_norm1)(UINT8 * pix, int line_size);
 101     int (*pix_norm)(UINT8 * pix1, UINT8 * pix2, int line_size);
 102
 103     /* maybe create an array for 16/8 functions */
 104     op_pixels_func put_pixels_tab[2][4];
 105     op_pixels_func avg_pixels_tab[2][4];
 106     op_pixels_func put_no_rnd_pixels_tab[2][4];
 107     op_pixels_func avg_no_rnd_pixels_tab[2][4];
 108     qpel_mc_func put_qpel_pixels_tab[2][16];
 109     qpel_mc_func avg_qpel_pixels_tab[2][16];
 110     qpel_mc_func put_no_rnd_qpel_pixels_tab[2][16];
 111     qpel_mc_func avg_no_rnd_qpel_pixels_tab[2][16];
 112
 113     op_pixels_abs_func pix_abs16x16;
 114     op_pixels_abs_func pix_abs16x16_x2;
 115     op_pixels_abs_func pix_abs16x16_y2;
 116     op_pixels_abs_func pix_abs16x16_xy2;
 117     op_pixels_abs_func pix_abs8x8;
 118     op_pixels_abs_func pix_abs8x8_x2;
 119     op_pixels_abs_func pix_abs8x8_y2;
 120     op_pixels_abs_func pix_abs8x8_xy2;
 121
 122     /* huffyuv specific */
 123     //FIXME note: alignment isnt guranteed currently but could be if needed
 124     void (*add_bytes)(uint8_t *dst/*align 16*/, uint8_t *src/*align 16*/, int w);
 125     void (*diff_bytes)(uint8_t *dst/*align 16*/, uint8_t *src1/*align 16*/, uint8_t *src2/*align 16*/,int w);
 126 } DSPContext;
 127
 128 void dsputil_init(DSPContext* p, unsigned mask);
 129
 130 /**
 131  * permute block according to permuatation.
 132  * @param last last non zero element in scantable order
 133  */
 134 void ff_block_permute(INT16 *block, UINT8 *permutation, const UINT8 *scantable, int last);
 135
 136 #define emms_c()
 137
 138 #if defined(HAVE_MMX)
 139
 140 #undef emms_c
 141
 142 #define MM_MMX    0x0001 /* standard MMX */
 143 #define MM_3DNOW  0x0004 /* AMD 3DNOW */
 144 #define MM_MMXEXT 0x0002 /* SSE integer functions or AMD MMX ext */
 145 #define MM_SSE    0x0008 /* SSE functions */
 146 #define MM_SSE2   0x0010 /* PIV SSE2 functions */
 147
 148 extern int mm_flags;
 149
 150 int mm_support(void);
 151 void add_pixels_clamped_mmx(const DCTELEM *block, UINT8 *pixels, int line_size);
 152 void put_pixels_clamped_mmx(const DCTELEM *block, UINT8 *pixels, int line_size);
 153
 154 static inline void emms(void)
 155 {
 156     __asm __volatile ("emms;":::"memory");
 157 }
 158
 159 #define emms_c() \
 160 {\
 161     if (mm_flags & MM_MMX)\
 162         emms();\
 163 }
 164
 165 #define __align8 __attribute__ ((aligned (8)))
 166
 167 void dsputil_init_mmx(DSPContext* c, unsigned mask);
 168 void dsputil_set_bit_exact_mmx(DSPContext* c, unsigned mask);
 169
 170 #elif defined(ARCH_ARMV4L)
 171
 172 /* This is to use 4 bytes read to the IDCT pointers for some 'zero'
 173    line ptimizations */
 174 #define __align8 __attribute__ ((aligned (4)))
 175
 176 void dsputil_init_armv4l(DSPContext* c, unsigned mask);
 177
 178 #elif defined(HAVE_MLIB)
 179
 180 /* SPARC/VIS IDCT needs 8-byte aligned DCT blocks */
 181 #define __align8 __attribute__ ((aligned (8)))
 182
 183 void dsputil_init_mlib(DSPContext* c, unsigned mask);
 184
 185 #elif defined(ARCH_ALPHA)
 186
 187 #define __align8 __attribute__ ((aligned (8)))
 188
 189 void dsputil_init_alpha(DSPContext* c, unsigned mask);
 190
 191 #elif defined(ARCH_POWERPC)
 192
 193 #define MM_ALTIVEC    0x0001 /* standard AltiVec */
 194
 195 extern int mm_flags;
 196
 197 #define __align8 __attribute__ ((aligned (16)))
 198
 199 void dsputil_init_ppc(DSPContext* c, unsigned mask);
 200
 201 #elif defined(HAVE_MMI)
 202
 203 #define __align8 __attribute__ ((aligned (16)))
 204
 205 void dsputil_init_mmi(DSPContext* c, unsigned mask);
 206
 207 #else
 208
 209 #define __align8
 210
 211 #endif
 212
 213 #ifdef __GNUC__
 214
 215 struct unaligned_64 { uint64_t l; } __attribute__((packed));
 216 struct unaligned_32 { uint32_t l; } __attribute__((packed));
 217
 218 #define LD32(a) (((const struct unaligned_32 *) (a))->l)
 219 #define LD64(a) (((const struct unaligned_64 *) (a))->l)
 220
 221 #define ST32(a, b) (((struct unaligned_32 *) (a))->l) = (b)
 222
 223 #else /* __GNUC__ */
 224
 225 #define LD32(a) (*((uint32_t*)(a)))
 226 #define LD64(a) (*((uint64_t*)(a)))
 227
 228 #define ST32(a, b) *((uint32_t*)(a)) = (b)
 229
 230 #endif /* !__GNUC__ */
 231
 232 /* PSNR */
 233 void get_psnr(UINT8 *orig_image[3], UINT8 *coded_image[3],
 234               int orig_linesize[3], int coded_linesize,
 235               AVCodecContext *avctx);
 236
 237 /* FFT computation */
 238
 239 /* NOTE: soon integer code will be added, so you must use the
 240    FFTSample type */
 241 typedef float FFTSample;
 242
 243 typedef struct FFTComplex {
 244     FFTSample re, im;
 245 } FFTComplex;
 246
 247 typedef struct FFTContext {
 248     int nbits;
 249     int inverse;
 250     uint16_t *revtab;
 251     FFTComplex *exptab;
 252     FFTComplex *exptab1; /* only used by SSE code */
 253     void (*fft_calc)(struct FFTContext *s, FFTComplex *z);
 254 } FFTContext;
 255
 256 int fft_init(FFTContext *s, int nbits, int inverse);
 257 void fft_permute(FFTContext *s, FFTComplex *z);
 258 void fft_calc_c(FFTContext *s, FFTComplex *z);
 259 void fft_calc_sse(FFTContext *s, FFTComplex *z);
 260 static inline void fft_calc(FFTContext *s, FFTComplex *z)
 261 {
 262     s->fft_calc(s, z);
 263 }
 264 void fft_end(FFTContext *s);
 265
 266 /* MDCT computation */
 267
 268 typedef struct MDCTContext {
 269     int n;  /* size of MDCT (i.e. number of input data * 2) */
 270     int nbits; /* n = 2^nbits */
 271     /* pre/post rotation tables */
 272     FFTSample *tcos;
 273     FFTSample *tsin;
 274     FFTContext fft;
 275 } MDCTContext;
 276
 277 int ff_mdct_init(MDCTContext *s, int nbits, int inverse);
 278 void ff_imdct_calc(MDCTContext *s, FFTSample *output,
 279                 const FFTSample *input, FFTSample *tmp);
 280 void ff_mdct_calc(MDCTContext *s, FFTSample *out,
 281                const FFTSample *input, FFTSample *tmp);
 282 void ff_mdct_end(MDCTContext *s);
 283
 284 #ifndef HAVE_LRINTF
 285 /* XXX: add ISOC specific test to avoid specific BSD testing. */
 286 /* better than nothing implementation. */
 287 /* btw, rintf() is existing on fbsd too -- alex */
 288 static inline long int lrintf(float x)
 289 {
 290     return (int)(rint(x));
 291 }
 292 #endif
 293
 294 #endif