src/mesa/pipe/p_util.h

   1 /**************************************************************************
   2  *
   3  * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
   4  * All Rights Reserved.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the
   8  * "Software"), to deal in the Software without restriction, including
   9  * without limitation the rights to use, copy, modify, merge, publish,
  10  * distribute, sub license, and/or sell copies of the Software, and to
  11  * permit persons to whom the Software is furnished to do so, subject to
  12  * the following conditions:
  13  *
  14  * The above copyright notice and this permission notice (including the
  15  * next paragraph) shall be included in all copies or substantial portions
  16  * of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  19  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  21  * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
  22  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  23  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  24  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25  *
  26  **************************************************************************/
  27
  28 #ifndef P_UTIL_H
  29 #define P_UTIL_H
  30
  31 #include "p_compiler.h"
  32 #include <math.h>
  33
  34
  35 #ifdef WIN32
  36
  37 #ifdef __cplusplus
  38 extern "C"
  39 {
  40 #endif
  41
  42 void * __stdcall
  43 EngAllocMem(
  44     unsigned long Flags,
  45     unsigned long MemSize,
  46     unsigned long Tag );
  47
  48 void __stdcall
  49 EngFreeMem(
  50     void *Mem );
  51
  52 #ifdef __cplusplus
  53 }
  54 #endif
  55
  56 static INLINE void *
  57 MALLOC( unsigned size )
  58 {
  59    return EngAllocMem( 0, size, 'D3AG' );
  60 }
  61
  62 static INLINE void *
  63 CALLOC( unsigned count, unsigned size )
  64 {
  65    void *ptr = MALLOC( count * size );
  66    if( ptr ) {
  67       memset( ptr, 0, count * size );
  68    }
  69    return ptr;
  70 }
  71
  72 static INLINE void
  73 FREE( void *ptr )
  74 {
  75    if( ptr ) {
  76       EngFreeMem( ptr );
  77    }
  78 }
  79
  80 static INLINE void *
  81 REALLOC( void *old_ptr, unsigned old_size, unsigned new_size )
  82 {
  83    void *new_ptr;
  84    if( new_size <= old_size ) {
  85       return old_ptr;
  86    }
  87    new_ptr = MALLOC( new_size );
  88    if( new_ptr ) {
  89       memcpy( new_ptr, old_ptr, old_size );
  90    }
  91    FREE( old_ptr );
  92    return new_ptr;
  93 }
  94
  95 #define GETENV( X )  NULL
  96
  97 #else /* WIN32 */
  98
  99 #define MALLOC( SIZE )  malloc( SIZE )
 100
 101 #define CALLOC( COUNT, SIZE )   calloc( COUNT, SIZE )
 102
 103 #define FREE( PTR )  free( PTR )
 104
 105 #define REALLOC( OLDPTR, OLDSIZE, NEWSIZE )  realloc( OLDPTR, NEWSIZE )
 106
 107 #define GETENV( X )  getenv( X )
 108
 109 #endif /* WIN32 */
 110
 111 #define MALLOC_STRUCT(T)   (struct T *) MALLOC(sizeof(struct T))
 112
 113 #define CALLOC_STRUCT(T)   (struct T *) CALLOC(1, sizeof(struct T))
 114
 115
 116 /**
 117  * Return a pointer aligned to next multiple of N bytes.
 118  */
 119 static INLINE void *
 120 align_pointer( void *unaligned, uint alignment )
 121 {
 122    if (sizeof(void *) == 64) {
 123       union {
 124          void *p;
 125          uint64 u;
 126       } pu;
 127       pu.p = unaligned;
 128       pu.u = (pu.u + alignment - 1) & ~(uint64) (alignment - 1);
 129       return pu.p;
 130    }
 131    else {
 132       /* 32-bit pointers */
 133       union {
 134          void *p;
 135          uint u;
 136       } pu;
 137       pu.p = unaligned;
 138       pu.u = (pu.u + alignment - 1) & ~(alignment - 1);
 139       return pu.p;
 140    }
 141 }
 142
 143 /**
 144  * Return memory on given byte alignment
 145  */
 146 static INLINE void *
 147 align_malloc(size_t bytes, uint alignment)
 148 {
 149 #if defined(HAVE_POSIX_MEMALIGN)
 150    void *mem;
 151    (void) posix_memalign(& mem, alignment, bytes);
 152    return mem;
 153 #else
 154    char *ptr, *buf;
 155
 156    assert( alignment > 0 );
 157
 158    ptr = (char *) MALLOC(bytes + alignment + sizeof(void *));
 159    if (!ptr)
 160       return NULL;
 161
 162    buf = (char *) align_pointer( ptr + sizeof(void *), alignment );
 163    *(char **)(buf - sizeof(void *)) = ptr;
 164
 165    return buf;
 166 #endif /* defined(HAVE_POSIX_MEMALIGN) */
 167 }
 168
 169 /**
 170  * Free memory returned by align_malloc().
 171  */
 172 static INLINE void
 173 align_free(void *ptr)
 174 {
 175 #if defined(HAVE_POSIX_MEMALIGN)
 176    FREE(ptr);
 177 #else
 178    void **cubbyHole = (void **) ((char *) ptr - sizeof(void *));
 179    void *realAddr = *cubbyHole;
 180    FREE(realAddr);
 181 #endif /* defined(HAVE_POSIX_MEMALIGN) */
 182 }
 183
 184
 185
 186 #define CLAMP( X, MIN, MAX )  ( (X)<(MIN) ? (MIN) : ((X)>(MAX) ? (MAX) : (X)) )
 187 #define MIN2( A, B )   ( (A)<(B) ? (A) : (B) )
 188 #define MAX2( A, B )   ( (A)>(B) ? (A) : (B) )
 189
 190 #define Elements(x) sizeof(x)/sizeof(*(x))
 191 #define Offset(TYPE, MEMBER) ((unsigned)&(((TYPE *)NULL)->MEMBER))
 192
 193 /**
 194  * Return a pointer aligned to next multiple of 16 bytes.
 195  */
 196 static INLINE void *
 197 align16( void *unaligned )
 198 {
 199    return align_pointer( unaligned, 16 );
 200 }
 201
 202
 203 static INLINE int align_int(int x, int align)
 204 {
 205    return (x + align - 1) & ~(align - 1);
 206 }
 207
 208
 209
 210 #if defined(__MSC__) && defined(__WIN32__)
 211 static INLINE unsigned ffs( unsigned u )
 212 {
 213    unsigned i;
 214
 215    if( u == 0 ) {
 216       return 0;
 217    }
 218
 219    __asm bsf eax, [u]
 220    __asm inc eax
 221    __asm mov [i], eax
 222
 223    return i;
 224 }
 225 #endif
 226
 227 union fi {
 228    float f;
 229    int i;
 230    unsigned ui;
 231 };
 232
 233 #define UBYTE_TO_FLOAT( ub ) ((float)(ub) / 255.0F)
 234
 235 #define IEEE_0996 0x3f7f0000    /* 0.996 or so */
 236
 237 /* This function/macro is sensitive to precision.  Test very carefully
 238  * if you change it!
 239  */
 240 #define UNCLAMPED_FLOAT_TO_UBYTE(UB, F)                                 \
 241         do {                                                            \
 242            union fi __tmp;                                              \
 243            __tmp.f = (F);                                               \
 244            if (__tmp.i < 0)                                             \
 245               UB = (ubyte) 0;                                           \
 246            else if (__tmp.i >= IEEE_0996)                               \
 247               UB = (ubyte) 255;                                 \
 248            else {                                                       \
 249               __tmp.f = __tmp.f * (255.0f/256.0f) + 32768.0f;           \
 250               UB = (ubyte) __tmp.i;                                     \
 251            }                                                            \
 252         } while (0)
 253
 254
 255
 256 static INLINE unsigned pack_ub4( unsigned char b0,
 257                                  unsigned char b1,
 258                                  unsigned char b2,
 259                                  unsigned char b3 )
 260 {
 261    return ((((unsigned int)b0) << 0) |
 262            (((unsigned int)b1) << 8) |
 263            (((unsigned int)b2) << 16) |
 264            (((unsigned int)b3) << 24));
 265 }
 266
 267 static INLINE unsigned fui( float f )
 268 {
 269    union fi fi;
 270    fi.f = f;
 271    return fi.ui;
 272 }
 273
 274 static INLINE unsigned char float_to_ubyte( float f )
 275 {
 276    unsigned char ub;
 277    UNCLAMPED_FLOAT_TO_UBYTE(ub, f);
 278    return ub;
 279 }
 280
 281 static INLINE unsigned pack_ui32_float4( float a,
 282                                          float b,
 283                                          float c,
 284                                          float d )
 285 {
 286    return pack_ub4( float_to_ubyte(a),
 287                     float_to_ubyte(b),
 288                     float_to_ubyte(c),
 289                     float_to_ubyte(d) );
 290 }
 291
 292 #define COPY_4V( DST, SRC )         \
 293 do {                                \
 294    (DST)[0] = (SRC)[0];             \
 295    (DST)[1] = (SRC)[1];             \
 296    (DST)[2] = (SRC)[2];             \
 297    (DST)[3] = (SRC)[3];             \
 298 } while (0)
 299
 300
 301 #define COPY_4FV( DST, SRC )  COPY_4V(DST, SRC)
 302
 303
 304 #define ASSIGN_4V( DST, V0, V1, V2, V3 ) \
 305 do {                                     \
 306    (DST)[0] = (V0);                      \
 307    (DST)[1] = (V1);                      \
 308    (DST)[2] = (V2);                      \
 309    (DST)[3] = (V3);                      \
 310 } while (0)
 311
 312
 313 static INLINE int ifloor(float f)
 314 {
 315    int ai, bi;
 316    double af, bf;
 317    union fi u;
 318
 319    af = (3 << 22) + 0.5 + (double)f;
 320    bf = (3 << 22) + 0.5 - (double)f;
 321    u.f = (float) af;  ai = u.i;
 322    u.f = (float) bf;  bi = u.i;
 323    return (ai - bi) >> 1;
 324 }
 325
 326
 327 #if defined(__GNUC__) && defined(__i386__)
 328 static INLINE int iround(float f)
 329 {
 330    int r;
 331    __asm__ ("fistpl %0" : "=m" (r) : "t" (f) : "st");
 332    return r;
 333 }
 334 #elif defined(__MSC__) && defined(__WIN32__)
 335 static INLINE int iround(float f)
 336 {
 337    int r;
 338    _asm {
 339          fld f
 340          fistp r
 341         }
 342    return r;
 343 }
 344 #else
 345 #define IROUND(f)  ((int) (((f) >= 0.0F) ? ((f) + 0.5F) : ((f) - 0.5F)))
 346 #endif
 347
 348
 349 /* Could maybe have an inline version of this?
 350  */
 351 #if defined(__GNUC__)
 352 #define FABSF(x)   fabsf(x)
 353 #else
 354 #define FABSF(x)   ((float) fabs(x))
 355 #endif
 356
 357 /* Pretty fast, and accurate.
 358  * Based on code from http://www.flipcode.com/totd/
 359  */
 360 static INLINE float LOG2(float val)
 361 {
 362    union fi num;
 363    int log_2;
 364
 365    num.f = val;
 366    log_2 = ((num.i >> 23) & 255) - 128;
 367    num.i &= ~(255 << 23);
 368    num.i += 127 << 23;
 369    num.f = ((-1.0f/3) * num.f + 2) * num.f - 2.0f/3;
 370    return num.f + log_2;
 371 }
 372
 373 #if defined(__GNUC__)
 374 #define CEILF(x)   ceilf(x)
 375 #else
 376 #define CEILF(x)   ((float) ceil(x))
 377 #endif
 378
 379 static INLINE int align(int value, int alignment)
 380 {
 381    return (value + alignment - 1) & ~(alignment - 1);
 382 }
 383
 384 /* Convenient...
 385  */
 386 extern void _mesa_printf(const char *str, ...);
 387
 388
 389 /* util/p_util.c
 390  */
 391 extern void pipe_copy_rect(ubyte * dst, unsigned cpp, unsigned dst_pitch,
 392                            unsigned dst_x, unsigned dst_y, unsigned width,
 393                            unsigned height, const ubyte * src,
 394                            unsigned src_pitch, unsigned src_x, unsigned src_y);
 395
 396
 397 #endif