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 "p_debug.h"
  33 #include <math.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 MALLOC_STRUCT(T)   (struct T *) MALLOC(sizeof(struct T))
 113
 114 #define CALLOC_STRUCT(T)   (struct T *) CALLOC(1, sizeof(struct T))
 115
 116
 117 /**
 118  * Return a pointer aligned to next multiple of N bytes.
 119  */
 120 static INLINE void *
 121 align_pointer( void *unaligned, uint alignment )
 122 {
 123    if (sizeof(void *) == 64) {
 124       union {
 125          void *p;
 126          uint64 u;
 127       } pu;
 128       pu.p = unaligned;
 129       pu.u = (pu.u + alignment - 1) & ~(uint64) (alignment - 1);
 130       return pu.p;
 131    }
 132    else {
 133       /* 32-bit pointers */
 134       union {
 135          void *p;
 136          uint u;
 137       } pu;
 138       pu.p = unaligned;
 139       pu.u = (pu.u + alignment - 1) & ~(alignment - 1);
 140       return pu.p;
 141    }
 142 }
 143
 144 /**
 145  * Return memory on given byte alignment
 146  */
 147 static INLINE void *
 148 align_malloc(size_t bytes, uint alignment)
 149 {
 150 #if defined(HAVE_POSIX_MEMALIGN)
 151    void *mem;
 152    (void) posix_memalign(& mem, alignment, bytes);
 153    return mem;
 154 #else
 155    char *ptr, *buf;
 156
 157    assert( alignment > 0 );
 158
 159    ptr = (char *) MALLOC(bytes + alignment + sizeof(void *));
 160    if (!ptr)
 161       return NULL;
 162
 163    buf = (char *) align_pointer( ptr + sizeof(void *), alignment );
 164    *(char **)(buf - sizeof(void *)) = ptr;
 165
 166    return buf;
 167 #endif /* defined(HAVE_POSIX_MEMALIGN) */
 168 }
 169
 170 /**
 171  * Free memory returned by align_malloc().
 172  */
 173 static INLINE void
 174 align_free(void *ptr)
 175 {
 176 #if defined(HAVE_POSIX_MEMALIGN)
 177    FREE(ptr);
 178 #else
 179    void **cubbyHole = (void **) ((char *) ptr - sizeof(void *));
 180    void *realAddr = *cubbyHole;
 181    FREE(realAddr);
 182 #endif /* defined(HAVE_POSIX_MEMALIGN) */
 183 }
 184
 185
 186
 187 /**
 188  * Duplicate of a block of memory
 189  */
 190 static INLINE void *
 191 mem_dup(const void *src, uint size)
 192 {
 193    void *dup = malloc(size);
 194    if (dup)
 195       memcpy(dup, src, size);
 196    return dup;
 197 }
 198
 199
 200
 201 #define CLAMP( X, MIN, MAX )  ( (X)<(MIN) ? (MIN) : ((X)>(MAX) ? (MAX) : (X)) )
 202 #define MIN2( A, B )   ( (A)<(B) ? (A) : (B) )
 203 #define MAX2( A, B )   ( (A)>(B) ? (A) : (B) )
 204
 205 #define Elements(x) sizeof(x)/sizeof(*(x))
 206 #define Offset(TYPE, MEMBER) ((unsigned)&(((TYPE *)NULL)->MEMBER))
 207
 208 /**
 209  * Return a pointer aligned to next multiple of 16 bytes.
 210  */
 211 static INLINE void *
 212 align16( void *unaligned )
 213 {
 214    return align_pointer( unaligned, 16 );
 215 }
 216
 217
 218 static INLINE int align_int(int x, int align)
 219 {
 220    return (x + align - 1) & ~(align - 1);
 221 }
 222
 223
 224
 225 #if defined(__MSC__) && defined(__WIN32__)
 226 static INLINE unsigned ffs( unsigned u )
 227 {
 228    unsigned i;
 229
 230    if( u == 0 ) {
 231       return 0;
 232    }
 233
 234    __asm bsf eax, [u]
 235    __asm inc eax
 236    __asm mov [i], eax
 237
 238    return i;
 239 }
 240 #endif
 241
 242 union fi {
 243    float f;
 244    int i;
 245    unsigned ui;
 246 };
 247
 248 #define UBYTE_TO_FLOAT( ub ) ((float)(ub) / 255.0F)
 249
 250 #define IEEE_0996 0x3f7f0000    /* 0.996 or so */
 251
 252 /* This function/macro is sensitive to precision.  Test very carefully
 253  * if you change it!
 254  */
 255 #define UNCLAMPED_FLOAT_TO_UBYTE(UB, F)                                 \
 256         do {                                                            \
 257            union fi __tmp;                                              \
 258            __tmp.f = (F);                                               \
 259            if (__tmp.i < 0)                                             \
 260               UB = (ubyte) 0;                                           \
 261            else if (__tmp.i >= IEEE_0996)                               \
 262               UB = (ubyte) 255;                                 \
 263            else {                                                       \
 264               __tmp.f = __tmp.f * (255.0f/256.0f) + 32768.0f;           \
 265               UB = (ubyte) __tmp.i;                                     \
 266            }                                                            \
 267         } while (0)
 268
 269
 270
 271 static INLINE unsigned pack_ub4( unsigned char b0,
 272                                  unsigned char b1,
 273                                  unsigned char b2,
 274                                  unsigned char b3 )
 275 {
 276    return ((((unsigned int)b0) << 0) |
 277            (((unsigned int)b1) << 8) |
 278            (((unsigned int)b2) << 16) |
 279            (((unsigned int)b3) << 24));
 280 }
 281
 282 static INLINE unsigned fui( float f )
 283 {
 284    union fi fi;
 285    fi.f = f;
 286    return fi.ui;
 287 }
 288
 289 static INLINE unsigned char float_to_ubyte( float f )
 290 {
 291    unsigned char ub;
 292    UNCLAMPED_FLOAT_TO_UBYTE(ub, f);
 293    return ub;
 294 }
 295
 296 static INLINE unsigned pack_ui32_float4( float a,
 297                                          float b,
 298                                          float c,
 299                                          float d )
 300 {
 301    return pack_ub4( float_to_ubyte(a),
 302                     float_to_ubyte(b),
 303                     float_to_ubyte(c),
 304                     float_to_ubyte(d) );
 305 }
 306
 307 #define COPY_4V( DST, SRC )         \
 308 do {                                \
 309    (DST)[0] = (SRC)[0];             \
 310    (DST)[1] = (SRC)[1];             \
 311    (DST)[2] = (SRC)[2];             \
 312    (DST)[3] = (SRC)[3];             \
 313 } while (0)
 314
 315
 316 #define COPY_4FV( DST, SRC )  COPY_4V(DST, SRC)
 317
 318
 319 #define ASSIGN_4V( DST, V0, V1, V2, V3 ) \
 320 do {                                     \
 321    (DST)[0] = (V0);                      \
 322    (DST)[1] = (V1);                      \
 323    (DST)[2] = (V2);                      \
 324    (DST)[3] = (V3);                      \
 325 } while (0)
 326
 327
 328 static INLINE int ifloor(float f)
 329 {
 330    int ai, bi;
 331    double af, bf;
 332    union fi u;
 333
 334    af = (3 << 22) + 0.5 + (double)f;
 335    bf = (3 << 22) + 0.5 - (double)f;
 336    u.f = (float) af;  ai = u.i;
 337    u.f = (float) bf;  bi = u.i;
 338    return (ai - bi) >> 1;
 339 }
 340
 341
 342 #if defined(__GNUC__) && defined(__i386__)
 343 static INLINE int iround(float f)
 344 {
 345    int r;
 346    __asm__ ("fistpl %0" : "=m" (r) : "t" (f) : "st");
 347    return r;
 348 }
 349 #elif defined(__MSC__) && defined(__WIN32__)
 350 static INLINE int iround(float f)
 351 {
 352    int r;
 353    _asm {
 354          fld f
 355          fistp r
 356         }
 357    return r;
 358 }
 359 #else
 360 #define IROUND(f)  ((int) (((f) >= 0.0F) ? ((f) + 0.5F) : ((f) - 0.5F)))
 361 #endif
 362
 363
 364 /* Could maybe have an inline version of this?
 365  */
 366 #if defined(__GNUC__)
 367 #define FABSF(x)   fabsf(x)
 368 #else
 369 #define FABSF(x)   ((float) fabs(x))
 370 #endif
 371
 372 /* Pretty fast, and accurate.
 373  * Based on code from http://www.flipcode.com/totd/
 374  */
 375 static INLINE float LOG2(float val)
 376 {
 377    union fi num;
 378    int log_2;
 379
 380    num.f = val;
 381    log_2 = ((num.i >> 23) & 255) - 128;
 382    num.i &= ~(255 << 23);
 383    num.i += 127 << 23;
 384    num.f = ((-1.0f/3) * num.f + 2) * num.f - 2.0f/3;
 385    return num.f + log_2;
 386 }
 387
 388 #if defined(__GNUC__)
 389 #define CEILF(x)   ceilf(x)
 390 #else
 391 #define CEILF(x)   ((float) ceil(x))
 392 #endif
 393
 394 static INLINE int align(int value, int alignment)
 395 {
 396    return (value + alignment - 1) & ~(alignment - 1);
 397 }
 398
 399
 400 /* util/p_util.c
 401  */
 402 extern void pipe_copy_rect(ubyte * dst, unsigned cpp, unsigned dst_pitch,
 403                            unsigned dst_x, unsigned dst_y, unsigned width,
 404                            unsigned height, const ubyte * src,
 405                            int src_pitch, unsigned src_x, int src_y);
 406
 407
 408 #endif