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