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 #include <stdint.h>
  34
  35
  36 #ifdef WIN32
  37
  38 #ifdef __cplusplus
  39 extern "C"
  40 {
  41 #endif
  42
  43 void * __stdcall
  44 EngAllocMem(
  45     unsigned long Flags,
  46     unsigned long MemSize,
  47     unsigned long Tag );
  48
  49 void __stdcall
  50 EngFreeMem(
  51     void *Mem );
  52
  53 #ifdef __cplusplus
  54 }
  55 #endif
  56
  57 static INLINE void *
  58 MALLOC( unsigned size )
  59 {
  60    return EngAllocMem( 0, size, 'D3AG' );
  61 }
  62
  63 static INLINE void *
  64 CALLOC( unsigned count, unsigned size )
  65 {
  66    void *ptr = MALLOC( count * size );
  67    if( ptr ) {
  68       memset( ptr, 0, count * size );
  69    }
  70    return ptr;
  71 }
  72
  73 static INLINE void
  74 FREE( void *ptr )
  75 {
  76    if( ptr ) {
  77       EngFreeMem( ptr );
  78    }
  79 }
  80
  81 static INLINE void *
  82 REALLOC( void *old_ptr, unsigned old_size, unsigned new_size )
  83 {
  84    void *new_ptr;
  85    if( new_size <= old_size ) {
  86       return old_ptr;
  87    }
  88    new_ptr = MALLOC( new_size );
  89    if( new_ptr ) {
  90       memcpy( new_ptr, old_ptr, old_size );
  91    }
  92    FREE( old_ptr );
  93    return new_ptr;
  94 }
  95
  96 #define GETENV( X )  NULL
  97
  98 #else /* WIN32 */
  99
 100 #define MALLOC( SIZE )  malloc( SIZE )
 101
 102 #define CALLOC( COUNT, SIZE )   calloc( COUNT, SIZE )
 103
 104 #define FREE( PTR )  free( PTR )
 105
 106 #define REALLOC( OLDPTR, OLDSIZE, NEWSIZE )  realloc( OLDPTR, NEWSIZE )
 107
 108 #define GETENV( X )  getenv( X )
 109
 110 #endif /* WIN32 */
 111
 112 #define CALLOC_STRUCT(T)   (struct T *) CALLOC(1, sizeof(struct T))
 113
 114
 115
 116 /**
 117  * Return memory on given byte alignment
 118  */
 119 static INLINE void *
 120 align_malloc(size_t bytes, uint alignment)
 121 {
 122 #if defined(HAVE_POSIX_MEMALIGN)
 123    void *mem;
 124    (void) posix_memalign(& mem, alignment, bytes);
 125    return mem;
 126 #else
 127    uintptr_t ptr, buf;
 128
 129    assert( alignment > 0 );
 130
 131    ptr = (uintptr_t) MALLOC(bytes + alignment + sizeof(void *));
 132    if (!ptr)
 133       return NULL;
 134
 135    buf = (ptr + alignment + sizeof(void *)) & ~(uintptr_t)(alignment - 1);
 136    *(uintptr_t *)(buf - sizeof(void *)) = ptr;
 137
 138    return (void *) buf;
 139 #endif /* defined(HAVE_POSIX_MEMALIGN) */
 140 }
 141
 142 /**
 143  * Free memory returned by align_malloc().
 144  */
 145 static INLINE void
 146 align_free(void *ptr)
 147 {
 148 #if defined(HAVE_POSIX_MEMALIGN)
 149    FREE(ptr);
 150 #else
 151    void **cubbyHole = (void **) ((char *) ptr - sizeof(void *));
 152    void *realAddr = *cubbyHole;
 153    FREE(realAddr);
 154 #endif /* defined(HAVE_POSIX_MEMALIGN) */
 155 }
 156
 157
 158
 159 #define CLAMP( X, MIN, MAX )  ( (X)<(MIN) ? (MIN) : ((X)>(MAX) ? (MAX) : (X)) )
 160 #define MIN2( A, B )   ( (A)<(B) ? (A) : (B) )
 161 #define MAX2( A, B )   ( (A)>(B) ? (A) : (B) )
 162
 163 #define Elements(x) sizeof(x)/sizeof(*(x))
 164 #define Offset(TYPE, MEMBER) ((unsigned)&(((TYPE *)NULL)->MEMBER))
 165
 166 /**
 167  * Return a pointer aligned to next multiple of 16 bytes.
 168  */
 169 static INLINE void *
 170 align16( void *unaligned )
 171 {
 172    if (sizeof(void *) == 64) {
 173       union {
 174          void *p;
 175          uint64 u;
 176       } pu;
 177       pu.p = unaligned;
 178       pu.u = (pu.u + 15) & ~15;
 179       return pu.p;
 180    }
 181    else {
 182       /* 32-bit pointers */
 183       union {
 184          void *p;
 185          uint u;
 186       } pu;
 187       pu.p = unaligned;
 188       pu.u = (pu.u + 15) & ~15;
 189       return pu.p;
 190    }
 191 }
 192
 193
 194 #if defined(__MSC__) && defined(__WIN32__)
 195 static INLINE unsigned ffs( unsigned u )
 196 {
 197    unsigned i;
 198
 199    if( u == 0 ) {
 200       return 0;
 201    }
 202
 203    __asm bsf eax, [u]
 204    __asm inc eax
 205    __asm mov [i], eax
 206
 207    return i;
 208 }
 209 #endif
 210
 211 union fi {
 212    float f;
 213    int i;
 214    unsigned ui;
 215 };
 216
 217 #define UBYTE_TO_FLOAT( ub ) ((float)(ub) / 255.0F)
 218
 219 #define IEEE_0996 0x3f7f0000    /* 0.996 or so */
 220
 221 /* This function/macro is sensitive to precision.  Test very carefully
 222  * if you change it!
 223  */
 224 #define UNCLAMPED_FLOAT_TO_UBYTE(UB, F)                                 \
 225         do {                                                            \
 226            union fi __tmp;                                              \
 227            __tmp.f = (F);                                               \
 228            if (__tmp.i < 0)                                             \
 229               UB = (ubyte) 0;                                           \
 230            else if (__tmp.i >= IEEE_0996)                               \
 231               UB = (ubyte) 255;                                 \
 232            else {                                                       \
 233               __tmp.f = __tmp.f * (255.0f/256.0f) + 32768.0f;           \
 234               UB = (ubyte) __tmp.i;                                     \
 235            }                                                            \
 236         } while (0)
 237
 238
 239
 240 static INLINE unsigned pack_ub4( unsigned char b0,
 241                                  unsigned char b1,
 242                                  unsigned char b2,
 243                                  unsigned char b3 )
 244 {
 245    return ((((unsigned int)b0) << 0) |
 246            (((unsigned int)b1) << 8) |
 247            (((unsigned int)b2) << 16) |
 248            (((unsigned int)b3) << 24));
 249 }
 250
 251 static INLINE unsigned fui( float f )
 252 {
 253    union fi fi;
 254    fi.f = f;
 255    return fi.ui;
 256 }
 257
 258 static INLINE unsigned char float_to_ubyte( float f )
 259 {
 260    unsigned char ub;
 261    UNCLAMPED_FLOAT_TO_UBYTE(ub, f);
 262    return ub;
 263 }
 264
 265 static INLINE unsigned pack_ui32_float4( float a,
 266                                          float b,
 267                                          float c,
 268                                          float d )
 269 {
 270    return pack_ub4( float_to_ubyte(a),
 271                     float_to_ubyte(b),
 272                     float_to_ubyte(c),
 273                     float_to_ubyte(d) );
 274 }
 275
 276 #define COPY_4V( DST, SRC )         \
 277 do {                                \
 278    (DST)[0] = (SRC)[0];             \
 279    (DST)[1] = (SRC)[1];             \
 280    (DST)[2] = (SRC)[2];             \
 281    (DST)[3] = (SRC)[3];             \
 282 } while (0)
 283
 284
 285 #define COPY_4FV( DST, SRC )  COPY_4V(DST, SRC)
 286
 287
 288 #define ASSIGN_4V( DST, V0, V1, V2, V3 ) \
 289 do {                                     \
 290    (DST)[0] = (V0);                      \
 291    (DST)[1] = (V1);                      \
 292    (DST)[2] = (V2);                      \
 293    (DST)[3] = (V3);                      \
 294 } while (0)
 295
 296
 297 static INLINE int ifloor(float f)
 298 {
 299    int ai, bi;
 300    double af, bf;
 301    union fi u;
 302
 303    af = (3 << 22) + 0.5 + (double)f;
 304    bf = (3 << 22) + 0.5 - (double)f;
 305    u.f = (float) af;  ai = u.i;
 306    u.f = (float) bf;  bi = u.i;
 307    return (ai - bi) >> 1;
 308 }
 309
 310
 311 #if defined(__GNUC__) && defined(__i386__)
 312 static INLINE int iround(float f)
 313 {
 314    int r;
 315    __asm__ ("fistpl %0" : "=m" (r) : "t" (f) : "st");
 316    return r;
 317 }
 318 #elif defined(__MSC__) && defined(__WIN32__)
 319 static INLINE int iround(float f)
 320 {
 321    int r;
 322    _asm {
 323          fld f
 324          fistp r
 325         }
 326    return r;
 327 }
 328 #else
 329 #define IROUND(f)  ((int) (((f) >= 0.0F) ? ((f) + 0.5F) : ((f) - 0.5F)))
 330 #endif
 331
 332
 333 /* Could maybe have an inline version of this?
 334  */
 335 #if defined(__GNUC__)
 336 #define FABSF(x)   fabsf(x)
 337 #else
 338 #define FABSF(x)   ((float) fabs(x))
 339 #endif
 340
 341 /* Pretty fast, and accurate.
 342  * Based on code from http://www.flipcode.com/totd/
 343  */
 344 static INLINE float LOG2(float val)
 345 {
 346    union fi num;
 347    int log_2;
 348
 349    num.f = val;
 350    log_2 = ((num.i >> 23) & 255) - 128;
 351    num.i &= ~(255 << 23);
 352    num.i += 127 << 23;
 353    num.f = ((-1.0f/3) * num.f + 2) * num.f - 2.0f/3;
 354    return num.f + log_2;
 355 }
 356
 357 #if defined(__GNUC__)
 358 #define CEILF(x)   ceilf(x)
 359 #else
 360 #define CEILF(x)   ((float) ceil(x))
 361 #endif
 362
 363 static INLINE int align(int value, int alignment)
 364 {
 365    return (value + alignment - 1) & ~(alignment - 1);
 366 }
 367
 368 /* Convenient...
 369  */
 370 extern void _mesa_printf(const char *str, ...);
 371
 372 #endif