src/gallium/include/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_config.h"
  32 #include "p_compiler.h"
  33 #include "p_debug.h"
  34 #include "p_pointer.h"
  35 #include <math.h>
  36 #include <stdarg.h>
  37
  38
  39 #ifdef __cplusplus
  40 extern "C" {
  41 #endif
  42
  43
  44 #if defined(PIPE_SUBSYSTEM_WINDOWS_DISPLAY) && defined(DEBUG)
  45
  46 /* memory debugging */
  47
  48 #include "p_debug.h"
  49
  50 #define MALLOC( _size ) \
  51    debug_malloc( __FILE__, __LINE__, __FUNCTION__, _size )
  52 #define CALLOC( _count, _size ) \
  53    debug_calloc(__FILE__, __LINE__, __FUNCTION__, _count, _size )
  54 #define FREE( _ptr ) \
  55    debug_free( __FILE__, __LINE__, __FUNCTION__,  _ptr )
  56 #define REALLOC( _ptr, _old_size, _size ) \
  57    debug_realloc( __FILE__, __LINE__, __FUNCTION__,  _ptr, _old_size, _size )
  58
  59 #elif defined(PIPE_SUBSYSTEM_WINDOWS_DISPLAY)
  60
  61 void * __stdcall
  62 EngAllocMem(
  63     unsigned long Flags,
  64     unsigned long MemSize,
  65     unsigned long Tag );
  66
  67 void __stdcall
  68 EngFreeMem(
  69     void *Mem );
  70
  71 #define MALLOC( _size ) EngAllocMem( 0, _size, 'D3AG' )
  72 #define _FREE( _ptr ) EngFreeMem( _ptr )
  73
  74 #elif defined(PIPE_SUBSYSTEM_WINDOWS_MINIPORT)
  75
  76 void *
  77 ExAllocatePool(
  78     unsigned long PoolType,
  79     size_t NumberOfBytes);
  80
  81 void
  82 ExFreePool(void *P);
  83
  84 #define MALLOC(_size) ExAllocatePool(0, _size)
  85 #define _FREE(_ptr) ExFreePool(_ptr)
  86
  87 #else
  88
  89 #define MALLOC( SIZE )  malloc( SIZE )
  90 #define CALLOC( COUNT, SIZE )   calloc( COUNT, SIZE )
  91 #define FREE( PTR )  free( PTR )
  92 #define REALLOC( OLDPTR, OLDSIZE, NEWSIZE )  realloc( OLDPTR, NEWSIZE )
  93
  94 #endif
  95
  96
  97 #ifndef CALLOC
  98 static INLINE void *
  99 CALLOC( unsigned count, unsigned size )
 100 {
 101    void *ptr = MALLOC( count * size );
 102    if( ptr ) {
 103       memset( ptr, 0, count * size );
 104    }
 105    return ptr;
 106 }
 107 #endif /* !CALLOC */
 108
 109 #ifndef FREE
 110 static INLINE void
 111 FREE( void *ptr )
 112 {
 113    if( ptr ) {
 114       _FREE( ptr );
 115    }
 116 }
 117 #endif /* !FREE */
 118
 119 #ifndef REALLOC
 120 static INLINE void *
 121 REALLOC( void *old_ptr, unsigned old_size, unsigned new_size )
 122 {
 123    void *new_ptr = NULL;
 124
 125    if (new_size != 0) {
 126       unsigned copy_size = old_size < new_size ? old_size : new_size;
 127       new_ptr = MALLOC( new_size );
 128       if (new_ptr && old_ptr && copy_size) {
 129          memcpy( new_ptr, old_ptr, copy_size );
 130       }
 131    }
 132
 133    FREE( old_ptr );
 134    return new_ptr;
 135 }
 136 #endif /* !REALLOC */
 137
 138
 139 #define MALLOC_STRUCT(T)   (struct T *) MALLOC(sizeof(struct T))
 140
 141 #define CALLOC_STRUCT(T)   (struct T *) CALLOC(1, sizeof(struct T))
 142
 143 #define GETENV( X ) debug_get_option( X, NULL )
 144
 145
 146 /**
 147  * Return memory on given byte alignment
 148  */
 149 static INLINE void *
 150 align_malloc(size_t bytes, uint alignment)
 151 {
 152 #if defined(HAVE_POSIX_MEMALIGN)
 153    void *mem;
 154    (void) posix_memalign(& mem, alignment, bytes);
 155    return mem;
 156 #else
 157    char *ptr, *buf;
 158
 159    assert( alignment > 0 );
 160
 161    ptr = (char *) MALLOC(bytes + alignment + sizeof(void *));
 162    if (!ptr)
 163       return NULL;
 164
 165    buf = (char *) align_pointer( ptr + sizeof(void *), alignment );
 166    *(char **)(buf - sizeof(void *)) = ptr;
 167
 168    return buf;
 169 #endif /* defined(HAVE_POSIX_MEMALIGN) */
 170 }
 171
 172 /**
 173  * Free memory returned by align_malloc().
 174  */
 175 static INLINE void
 176 align_free(void *ptr)
 177 {
 178 #if defined(HAVE_POSIX_MEMALIGN)
 179    FREE(ptr);
 180 #else
 181    void **cubbyHole = (void **) ((char *) ptr - sizeof(void *));
 182    void *realAddr = *cubbyHole;
 183    FREE(realAddr);
 184 #endif /* defined(HAVE_POSIX_MEMALIGN) */
 185 }
 186
 187
 188
 189 /**
 190  * Duplicate a block of memory.
 191  */
 192 static INLINE void *
 193 mem_dup(const void *src, uint size)
 194 {
 195    void *dup = MALLOC(size);
 196    if (dup)
 197       memcpy(dup, src, size);
 198    return dup;
 199 }
 200
 201
 202
 203 #define CLAMP( X, MIN, MAX )  ( (X)<(MIN) ? (MIN) : ((X)>(MAX) ? (MAX) : (X)) )
 204 #define MIN2( A, B )   ( (A)<(B) ? (A) : (B) )
 205 #define MAX2( A, B )   ( (A)>(B) ? (A) : (B) )
 206
 207 /* JB stop warnings */
 208 #ifndef Elements
 209 #define Elements(x) (sizeof(x)/sizeof((x)[0]))
 210 #endif
 211 #define Offset(TYPE, MEMBER) ((unsigned)&(((TYPE *)NULL)->MEMBER))
 212
 213 /**
 214  * Return a pointer aligned to next multiple of 16 bytes.
 215  */
 216 static INLINE void *
 217 align16( void *unaligned )
 218 {
 219    return align_pointer( unaligned, 16 );
 220 }
 221
 222
 223 static INLINE int align_int(int x, int align)
 224 {
 225    return (x + align - 1) & ~(align - 1);
 226 }
 227
 228
 229
 230 #if defined(__MSC__) && defined(__WIN32__)
 231 static INLINE unsigned ffs( unsigned u )
 232 {
 233    unsigned i;
 234
 235    if( u == 0 ) {
 236       return 0;
 237    }
 238
 239    __asm bsf eax, [u]
 240    __asm inc eax
 241    __asm mov [i], eax
 242
 243    return i;
 244 }
 245 #endif
 246
 247 union fi {
 248    float f;
 249    int i;
 250    unsigned ui;
 251 };
 252
 253 #define UBYTE_TO_FLOAT( ub ) ((float)(ub) / 255.0F)
 254
 255 #define IEEE_0996 0x3f7f0000    /* 0.996 or so */
 256
 257 /* This function/macro is sensitive to precision.  Test very carefully
 258  * if you change it!
 259  */
 260 #define UNCLAMPED_FLOAT_TO_UBYTE(UB, F)                                 \
 261         do {                                                            \
 262            union fi __tmp;                                              \
 263            __tmp.f = (F);                                               \
 264            if (__tmp.i < 0)                                             \
 265               UB = (ubyte) 0;                                           \
 266            else if (__tmp.i >= IEEE_0996)                               \
 267               UB = (ubyte) 255;                                 \
 268            else {                                                       \
 269               __tmp.f = __tmp.f * (255.0f/256.0f) + 32768.0f;           \
 270               UB = (ubyte) __tmp.i;                                     \
 271            }                                                            \
 272         } while (0)
 273
 274
 275
 276 static INLINE unsigned pack_ub4( unsigned char b0,
 277                                  unsigned char b1,
 278                                  unsigned char b2,
 279                                  unsigned char b3 )
 280 {
 281    return ((((unsigned int)b0) << 0) |
 282            (((unsigned int)b1) << 8) |
 283            (((unsigned int)b2) << 16) |
 284            (((unsigned int)b3) << 24));
 285 }
 286
 287 static INLINE unsigned fui( float f )
 288 {
 289    union fi fi;
 290    fi.f = f;
 291    return fi.ui;
 292 }
 293
 294 static INLINE unsigned char float_to_ubyte( float f )
 295 {
 296    unsigned char ub;
 297    UNCLAMPED_FLOAT_TO_UBYTE(ub, f);
 298    return ub;
 299 }
 300
 301 static INLINE unsigned pack_ui32_float4( float a,
 302                                          float b,
 303                                          float c,
 304                                          float d )
 305 {
 306    return pack_ub4( float_to_ubyte(a),
 307                     float_to_ubyte(b),
 308                     float_to_ubyte(c),
 309                     float_to_ubyte(d) );
 310 }
 311
 312 #define COPY_4V( DST, SRC )         \
 313 do {                                \
 314    (DST)[0] = (SRC)[0];             \
 315    (DST)[1] = (SRC)[1];             \
 316    (DST)[2] = (SRC)[2];             \
 317    (DST)[3] = (SRC)[3];             \
 318 } while (0)
 319
 320
 321 #define COPY_4FV( DST, SRC )  COPY_4V(DST, SRC)
 322
 323
 324 #define ASSIGN_4V( DST, V0, V1, V2, V3 ) \
 325 do {                                     \
 326    (DST)[0] = (V0);                      \
 327    (DST)[1] = (V1);                      \
 328    (DST)[2] = (V2);                      \
 329    (DST)[3] = (V3);                      \
 330 } while (0)
 331
 332
 333 static INLINE int ifloor(float f)
 334 {
 335    int ai, bi;
 336    double af, bf;
 337    union fi u;
 338
 339    af = (3 << 22) + 0.5 + (double)f;
 340    bf = (3 << 22) + 0.5 - (double)f;
 341    u.f = (float) af;  ai = u.i;
 342    u.f = (float) bf;  bi = u.i;
 343    return (ai - bi) >> 1;
 344 }
 345
 346
 347 #if defined(__GNUC__) && defined(__i386__)
 348 static INLINE int iround(float f)
 349 {
 350    int r;
 351    __asm__ ("fistpl %0" : "=m" (r) : "t" (f) : "st");
 352    return r;
 353 }
 354 #elif defined(__MSC__) && defined(__WIN32__)
 355 static INLINE int iround(float f)
 356 {
 357    int r;
 358    _asm {
 359          fld f
 360          fistp r
 361         }
 362    return r;
 363 }
 364 #else
 365 #define IROUND(f)  ((int) (((f) >= 0.0F) ? ((f) + 0.5F) : ((f) - 0.5F)))
 366 #endif
 367
 368
 369 /* Could maybe have an inline version of this?
 370  */
 371 #if defined(__GNUC__)
 372 #define FABSF(x)   fabsf(x)
 373 #else
 374 #define FABSF(x)   ((float) fabs(x))
 375 #endif
 376
 377 /* Pretty fast, and accurate.
 378  * Based on code from http://www.flipcode.com/totd/
 379  */
 380 static INLINE float LOG2(float val)
 381 {
 382    union fi num;
 383    int log_2;
 384
 385    num.f = val;
 386    log_2 = ((num.i >> 23) & 255) - 128;
 387    num.i &= ~(255 << 23);
 388    num.i += 127 << 23;
 389    num.f = ((-1.0f/3) * num.f + 2) * num.f - 2.0f/3;
 390    return num.f + log_2;
 391 }
 392
 393 #if defined(__GNUC__)
 394 #define CEILF(x)   ceilf(x)
 395 #else
 396 #define CEILF(x)   ((float) ceil(x))
 397 #endif
 398
 399 static INLINE int align(int value, int alignment)
 400 {
 401    return (value + alignment - 1) & ~(alignment - 1);
 402 }
 403
 404
 405 /* util/p_util.c
 406  */
 407 extern void pipe_copy_rect(ubyte * dst, unsigned cpp, unsigned dst_pitch,
 408                            unsigned dst_x, unsigned dst_y, unsigned width,
 409                            unsigned height, const ubyte * src,
 410                            int src_pitch, unsigned src_x, int src_y);
 411
 412
 413
 414 #ifdef WIN32
 415
 416 #if !defined(_INC_MATH) || !defined(__cplusplus)
 417
 418 static INLINE float cosf( float f )
 419 {
 420    return (float) cos( (double) f );
 421 }
 422
 423 static INLINE float sinf( float f )
 424 {
 425    return (float) sin( (double) f );
 426 }
 427
 428 static INLINE float ceilf( float f )
 429 {
 430    return (float) ceil( (double) f );
 431 }
 432
 433 static INLINE float floorf( float f )
 434 {
 435    return (float) floor( (double) f );
 436 }
 437
 438 static INLINE float powf( float f, float g )
 439 {
 440    return (float) pow( (double) f, (double) g );
 441 }
 442
 443 static INLINE float sqrtf( float f )
 444 {
 445    return (float) sqrt( (double) f );
 446 }
 447
 448 static INLINE float fabsf( float f )
 449 {
 450    return (float) fabs( (double) f );
 451 }
 452
 453 static INLINE float logf( float f )
 454 {
 455    return (float) cos( (double) f );
 456 }
 457 #endif  /* _INC_MATH */
 458 #endif
 459
 460
 461 #ifdef __cplusplus
 462 }
 463 #endif
 464
 465 #endif