1 #ifndef _DEINT32_H
2 #define _DEINT32_H
3 /*-------------------------------------------------------------------------
4  * drawElements Base Portability Library
5  * -------------------------------------
6  *
7  * Copyright 2014 The Android Open Source Project
8  *
9  * Licensed under the Apache License, Version 2.0 (the "License");
10  * you may not use this file except in compliance with the License.
11  * You may obtain a copy of the License at
12  *
13  *      http://www.apache.org/licenses/LICENSE-2.0
14  *
15  * Unless required by applicable law or agreed to in writing, software
16  * distributed under the License is distributed on an "AS IS" BASIS,
17  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
18  * See the License for the specific language governing permissions and
19  * limitations under the License.
20  *
21  *//*!
22  * \file
23  * \brief 32-bit integer math.
24  *//*--------------------------------------------------------------------*/
25 
26 #include "deDefs.h"
27 
28 #if (DE_COMPILER == DE_COMPILER_MSC)
29 #	include <intrin.h>
30 #endif
31 
32 DE_BEGIN_EXTERN_C
33 
34 enum
35 {
36 	DE_RCP_FRAC_BITS	= 30		/*!< Number of fractional bits in deRcp32() result. */
37 };
38 
39 void	deRcp32				(deUint32 a, deUint32* rcp, int* exp);
40 void	deInt32_computeLUTs	(void);
41 void	deInt32_selfTest	(void);
42 
43 /*--------------------------------------------------------------------*//*!
44  * \brief Compute the absolute of an int.
45  * \param a	Input value.
46  * \return Absolute of the input value.
47  *
48  * \note The input 0x80000000u (for which the abs value cannot be
49  * represented), is asserted and returns the value itself.
50  *//*--------------------------------------------------------------------*/
deAbs32(int a)51 DE_INLINE int deAbs32 (int a)
52 {
53 	DE_ASSERT((unsigned int) a != 0x80000000u);
54 	return (a < 0) ? -a : a;
55 }
56 
57 /*--------------------------------------------------------------------*//*!
58  * \brief Compute the signed minimum of two values.
59  * \param a	First input value.
60  * \param b Second input value.
61  * \return The smallest of the two input values.
62  *//*--------------------------------------------------------------------*/
deMin32(int a,int b)63 DE_INLINE int deMin32 (int a, int b)
64 {
65 	return (a <= b) ? a : b;
66 }
67 
68 /*--------------------------------------------------------------------*//*!
69  * \brief Compute the signed maximum of two values.
70  * \param a	First input value.
71  * \param b Second input value.
72  * \return The largest of the two input values.
73  *//*--------------------------------------------------------------------*/
deMax32(int a,int b)74 DE_INLINE int deMax32 (int a, int b)
75 {
76 	return (a >= b) ? a : b;
77 }
78 
79 /*--------------------------------------------------------------------*//*!
80  * \brief Compute the unsigned minimum of two values.
81  * \param a	First input value.
82  * \param b Second input value.
83  * \return The smallest of the two input values.
84  *//*--------------------------------------------------------------------*/
deMinu32(deUint32 a,deUint32 b)85 DE_INLINE deUint32 deMinu32 (deUint32 a, deUint32 b)
86 {
87 	return (a <= b) ? a : b;
88 }
89 
90 /*--------------------------------------------------------------------*//*!
91  * \brief Compute the unsigned maximum of two values.
92  * \param a	First input value.
93  * \param b Second input value.
94  * \return The largest of the two input values.
95  *//*--------------------------------------------------------------------*/
deMaxu32(deUint32 a,deUint32 b)96 DE_INLINE deUint32 deMaxu32 (deUint32 a, deUint32 b)
97 {
98 	return (a >= b) ? a : b;
99 }
100 
101 /*--------------------------------------------------------------------*//*!
102  * \brief Check if a value is in the <b>inclusive<b> range [mn, mx].
103  * \param a		Value to check for range.
104  * \param mn	Range minimum value.
105  * \param mx	Range maximum value.
106  * \return True if (a >= mn) and (a <= mx), false otherwise.
107  *
108  * \see deInBounds32()
109  *//*--------------------------------------------------------------------*/
deInRange32(int a,int mn,int mx)110 DE_INLINE deBool deInRange32 (int a, int mn, int mx)
111 {
112 	return (a >= mn) && (a <= mx);
113 }
114 
115 /*--------------------------------------------------------------------*//*!
116  * \brief Check if a value is in the half-inclusive bounds [mn, mx[.
117  * \param a		Value to check for range.
118  * \param mn	Range minimum value.
119  * \param mx	Range maximum value.
120  * \return True if (a >= mn) and (a < mx), false otherwise.
121  *
122  * \see deInRange32()
123  *//*--------------------------------------------------------------------*/
deInBounds32(int a,int mn,int mx)124 DE_INLINE deBool deInBounds32 (int a, int mn, int mx)
125 {
126 	return (a >= mn) && (a < mx);
127 }
128 
129 /*--------------------------------------------------------------------*//*!
130  * \brief Clamp a value into the range [mn, mx].
131  * \param a		Value to clamp.
132  * \param mn	Minimum value.
133  * \param mx	Maximum value.
134  * \return The clamped value in [mn, mx] range.
135  *//*--------------------------------------------------------------------*/
deClamp32(int a,int mn,int mx)136 DE_INLINE int deClamp32 (int a, int mn, int mx)
137 {
138 	DE_ASSERT(mn <= mx);
139 	if (a < mn) return mn;
140 	if (a > mx) return mx;
141 	return a;
142 }
143 
144 /*--------------------------------------------------------------------*//*!
145  * \brief Get the sign of an integer.
146  * \param a	Input value.
147  * \return +1 if a>0, 0 if a==0, -1 if a<0.
148  *//*--------------------------------------------------------------------*/
deSign32(int a)149 DE_INLINE int deSign32 (int a)
150 {
151 	if (a > 0) return +1;
152 	if (a < 0) return -1;
153 	return 0;
154 }
155 
156 /*--------------------------------------------------------------------*//*!
157  * \brief Extract the sign bit of a.
158  * \param a	Input value.
159  * \return 0x80000000 if a<0, 0 otherwise.
160  *//*--------------------------------------------------------------------*/
deSignBit32(int a)161 DE_INLINE int deSignBit32 (int a)
162 {
163 	return (a & 0x80000000);
164 }
165 
166 /*--------------------------------------------------------------------*//*!
167  * \brief Integer rotate right.
168  * \param val	Value to rotate.
169  * \param r		Number of bits to rotate (in range [0, 32]).
170  * \return The rotated value.
171  *//*--------------------------------------------------------------------*/
deRor32(int val,int r)172 DE_INLINE int deRor32 (int val, int r)
173 {
174 	DE_ASSERT(r >= 0 && r <= 32);
175 	if (r == 0 || r == 32)
176 		return val;
177 	else
178 		return (int)(((deUint32)val >> r) | ((deUint32)val << (32-r)));
179 }
180 
181 /*--------------------------------------------------------------------*//*!
182  * \brief Integer rotate left.
183  * \param val	Value to rotate.
184  * \param r		Number of bits to rotate (in range [0, 32]).
185  * \return The rotated value.
186  *//*--------------------------------------------------------------------*/
deRol32(int val,int r)187 DE_INLINE int deRol32 (int val, int r)
188 {
189 	DE_ASSERT(r >= 0 && r <= 32);
190 	if (r == 0 || r == 32)
191 		return val;
192 	else
193 		return (int)(((deUint32)val << r) | ((deUint32)val >> (32-r)));
194 }
195 
196 /*--------------------------------------------------------------------*//*!
197  * \brief Check if a value is a power-of-two.
198  * \param a Input value.
199  * \return True if input is a power-of-two value, false otherwise.
200  *
201  * \note Also returns true for zero.
202  *//*--------------------------------------------------------------------*/
deIsPowerOfTwo32(int a)203 DE_INLINE deBool deIsPowerOfTwo32 (int a)
204 {
205 	return ((a & (a - 1)) == 0);
206 }
207 
208 /*--------------------------------------------------------------------*//*!
209  * \brief Check if an integer is aligned to given power-of-two size.
210  * \param a		Input value.
211  * \param align	Alignment to check for.
212  * \return True if input is aligned, false otherwise.
213  *//*--------------------------------------------------------------------*/
deIsAligned32(int a,int align)214 DE_INLINE deBool deIsAligned32 (int a, int align)
215 {
216 	DE_ASSERT(deIsPowerOfTwo32(align));
217 	return ((a & (align-1)) == 0);
218 }
219 
220 /*--------------------------------------------------------------------*//*!
221  * \brief Check if a pointer is aligned to given power-of-two size.
222  * \param ptr	Input pointer.
223  * \param align	Alignment to check for (power-of-two).
224  * \return True if input is aligned, false otherwise.
225  *//*--------------------------------------------------------------------*/
deIsAlignedPtr(const void * ptr,deUintptr align)226 DE_INLINE deBool deIsAlignedPtr (const void* ptr, deUintptr align)
227 {
228 	DE_ASSERT((align & (align-1)) == 0); /* power of two */
229 	return (((deUintptr)ptr & (align-1)) == 0);
230 }
231 
232 /*--------------------------------------------------------------------*//*!
233  * \brief Align an integer to given power-of-two size.
234  * \param val	Input to align.
235  * \param align	Alignment to check for (power-of-two).
236  * \return The aligned value (larger or equal to input).
237  *//*--------------------------------------------------------------------*/
deAlign32(deInt32 val,deInt32 align)238 DE_INLINE deInt32 deAlign32 (deInt32 val, deInt32 align)
239 {
240 	DE_ASSERT(deIsPowerOfTwo32(align));
241 	return (val + align - 1) & ~(align - 1);
242 }
243 
244 /*--------------------------------------------------------------------*//*!
245  * \brief Align a pointer to given power-of-two size.
246  * \param ptr	Input pointer to align.
247  * \param align	Alignment to check for (power-of-two).
248  * \return The aligned pointer (larger or equal to input).
249  *//*--------------------------------------------------------------------*/
deAlignPtr(void * ptr,deUintptr align)250 DE_INLINE void* deAlignPtr (void* ptr, deUintptr align)
251 {
252 	deUintptr val = (deUintptr)ptr;
253 	DE_ASSERT((align & (align-1)) == 0); /* power of two */
254 	return (void*)((val + align - 1) & ~(align - 1));
255 }
256 
257 /*--------------------------------------------------------------------*//*!
258  * \brief Compute number of leading zeros in an integer.
259  * \param a	Input value.
260  * \return The number of leading zero bits in the input.
261  *//*--------------------------------------------------------------------*/
262 extern const deInt8 g_clzLUT[256];
263 
deClz32(deUint32 a)264 DE_INLINE int deClz32 (deUint32 a)
265 {
266 #if (DE_COMPILER == DE_COMPILER_MSC)
267 	unsigned long i;
268 	if (_BitScanReverse(&i, (unsigned long)a) == 0)
269 		return 32;
270 	else
271 		return 31-i;
272 #elif (DE_COMPILER == DE_COMPILER_GCC) || (DE_COMPILER == DE_COMPILER_CLANG)
273 	if (a == 0)
274 		return 32;
275 	else
276 		return __builtin_clz((unsigned int)a);
277 #else
278 	if ((a & 0xFF000000u) != 0)
279 		return (int)g_clzLUT[a >> 24];
280 	if ((a & 0x00FF0000u) != 0)
281 		return 8 + (int)g_clzLUT[a >> 16];
282 	if ((a & 0x0000FF00u) != 0)
283 		return 16 + (int)g_clzLUT[a >> 8];
284 	return 24 + (int)g_clzLUT[a];
285 #endif
286 }
287 
288 /*--------------------------------------------------------------------*//*!
289  * \brief Compute integer 'floor' of 'log2' for a positive integer.
290  * \param a	Input value.
291  * \return floor(log2(a)).
292  *//*--------------------------------------------------------------------*/
deLog2Floor32(deInt32 a)293 DE_INLINE int deLog2Floor32 (deInt32 a)
294 {
295 	DE_ASSERT(a > 0);
296 	return 31 - deClz32(a);
297 }
298 
299 /*--------------------------------------------------------------------*//*!
300  * \brief Compute integer 'ceil' of 'log2' for a positive integer.
301  * \param a	Input value.
302  * \return ceil(log2(a)).
303  *//*--------------------------------------------------------------------*/
deLog2Ceil32(deInt32 a)304 DE_INLINE int deLog2Ceil32 (deInt32 a)
305 {
306 	int log2floor = deLog2Floor32(a);
307 	if (deIsPowerOfTwo32(a))
308 		return log2floor;
309 	else
310 		return log2floor+1;
311 }
312 
313 /* \todo [2012-04-28 pyry] Badly named, deprecated variant of deLog2Ceil32(). Remove once code has been fixed. */
deLog2Clz(deInt32 a)314 DE_INLINE deUint32 deLog2Clz(deInt32 a)
315 {
316 	return (deUint32)deLog2Ceil32(a);
317 }
318 
319 /*--------------------------------------------------------------------*//*!
320  * \brief Compute the bit population count of an integer.
321  * \param a	Input value.
322  * \return The number of one bits in the input.
323  *//*--------------------------------------------------------------------*/
dePop32(deUint32 a)324 DE_INLINE int dePop32 (deUint32 a)
325 {
326 	deUint32 mask0 = 0x55555555; /* 1-bit values. */
327 	deUint32 mask1 = 0x33333333; /* 2-bit values. */
328 	deUint32 mask2 = 0x0f0f0f0f; /* 4-bit values. */
329 	deUint32 mask3 = 0x00ff00ff; /* 8-bit values. */
330 	deUint32 mask4 = 0x0000ffff; /* 16-bit values. */
331 	deUint32 t = (deUint32)a;
332 	t = (t & mask0) + ((t>>1) & mask0);
333 	t = (t & mask1) + ((t>>2) & mask1);
334 	t = (t & mask2) + ((t>>4) & mask2);
335 	t = (t & mask3) + ((t>>8) & mask3);
336 	t = (t & mask4) + (t>>16);
337 	return (int)t;
338 }
339 
dePop64(deUint64 a)340 DE_INLINE int dePop64 (deUint64 a)
341 {
342 	return dePop32((deUint32)(a & 0xffffffffull)) + dePop32((deUint32)(a >> 32));
343 }
344 
345 /*--------------------------------------------------------------------*//*!
346  * \brief Reverse bytes in 32-bit integer (for example MSB -> LSB).
347  * \param a	Input value.
348  * \return The input with bytes reversed
349  *//*--------------------------------------------------------------------*/
deReverseBytes32(deUint32 v)350 DE_INLINE deUint32 deReverseBytes32 (deUint32 v)
351 {
352 	deUint32 b0 = v << 24;
353 	deUint32 b1 = (v & 0x0000ff00) << 8;
354 	deUint32 b2 = (v & 0x00ff0000) >> 8;
355 	deUint32 b3 = v >> 24;
356 	return b0|b1|b2|b3;
357 }
358 
deSafeMul32(deInt32 a,deInt32 b)359 DE_INLINE deInt32 deSafeMul32 (deInt32 a, deInt32 b)
360 {
361 	deInt32 res = a * b;
362 	DE_ASSERT((deInt64)res == ((deInt64)a * (deInt64)b));
363 	return res;
364 }
365 
deSafeAdd32(deInt32 a,deInt32 b)366 DE_INLINE deInt32 deSafeAdd32 (deInt32 a, deInt32 b)
367 {
368 	DE_ASSERT((deInt64)a + (deInt64)b == (deInt64)(a + b));
369 	return (a + b);
370 }
371 
372 /* \todo [petri] Move to deInt64.h? */
373 
deMulAsr32(deInt32 a,deInt32 b,int shift)374 DE_INLINE deInt32 deMulAsr32 (deInt32 a, deInt32 b, int shift)
375 {
376 	return (deInt32)(((deInt64)a * (deInt64)b) >> shift);
377 }
378 
deSafeMulAsr32(deInt32 a,deInt32 b,int shift)379 DE_INLINE deInt32 deSafeMulAsr32 (deInt32 a, deInt32 b, int shift)
380 {
381 	deInt64 res = ((deInt64)a * (deInt64)b) >> shift;
382 	DE_ASSERT(res == (deInt64)(deInt32)res);
383 	return (deInt32)res;
384 }
385 
deSafeMuluAsr32(deUint32 a,deUint32 b,int shift)386 DE_INLINE deUint32 deSafeMuluAsr32 (deUint32 a, deUint32 b, int shift)
387 {
388 	deUint64 res = ((deUint64)a * (deUint64)b) >> shift;
389 	DE_ASSERT(res == (deUint64)(deUint32)res);
390 	return (deUint32)res;
391 }
392 
deMul32_32_64(deInt32 a,deInt32 b)393 DE_INLINE deInt64 deMul32_32_64 (deInt32 a, deInt32 b)
394 {
395 	return ((deInt64)a * (deInt64)b);
396 }
397 
deAbs64(deInt64 a)398 DE_INLINE deInt64 deAbs64 (deInt64 a)
399 {
400 	DE_ASSERT((deUint64) a != 0x8000000000000000LL);
401 	return (a >= 0) ? a : -a;
402 }
403 
deClz64(deInt64 a)404 DE_INLINE int deClz64 (deInt64 a)
405 {
406 	if (((deUint64)a >> 32) != 0)
407 		return deClz32((deInt32)(a >> 32));
408 	return deClz32((deInt32)a) + 32;
409 }
410 
411 /* Common hash & compare functions. */
412 
deInt32Hash(deInt32 a)413 DE_INLINE deUint32 deInt32Hash (deInt32 a)
414 {
415 	/* From: http://www.concentric.net/~Ttwang/tech/inthash.htm */
416 	deUint32 key = a;
417 	key = (key ^ 61) ^ (key >> 16);
418 	key = key + (key << 3);
419 	key = key ^ (key >> 4);
420 	key = key * 0x27d4eb2d; /* prime/odd constant */
421 	key = key ^ (key >> 15);
422 	return key;
423 }
424 
deInt64Hash(deInt64 a)425 DE_INLINE deUint32 deInt64Hash (deInt64 a)
426 {
427 	/* From: http://www.concentric.net/~Ttwang/tech/inthash.htm */
428 	deUint64 key = a;
429 	key = (~key) + (key << 21); /* key = (key << 21) - key - 1; */
430 	key = key ^ (key >> 24);
431 	key = (key + (key << 3)) + (key << 8); /* key * 265 */
432 	key = key ^ (key >> 14);
433 	key = (key + (key << 2)) + (key << 4); /* key * 21 */
434 	key = key ^ (key >> 28);
435 	key = key + (key << 31);
436 	return (deUint32)key;
437 }
438 
deInt16Hash(deInt16 v)439 DE_INLINE int		deInt16Hash		(deInt16 v)					{ return deInt32Hash(v);			}
deUint16Hash(deUint16 v)440 DE_INLINE int		deUint16Hash	(deUint16 v)				{ return deInt32Hash((deInt32)v);	}
deUint32Hash(deUint32 v)441 DE_INLINE int		deUint32Hash	(deUint32 v)				{ return deInt32Hash((deInt32)v);	}
deUint64Hash(deUint64 v)442 DE_INLINE int		deUint64Hash	(deUint64 v)				{ return deInt64Hash((deInt64)v);	}
443 
deInt16Equal(deInt16 a,deInt16 b)444 DE_INLINE deBool	deInt16Equal	(deInt16 a, deInt16 b)		{ return (a == b);	}
deUint16Equal(deUint16 a,deUint16 b)445 DE_INLINE deBool	deUint16Equal	(deUint16 a, deUint16 b)	{ return (a == b);	}
deInt32Equal(deInt32 a,deInt32 b)446 DE_INLINE deBool	deInt32Equal	(deInt32 a, deInt32 b)		{ return (a == b);	}
deUint32Equal(deUint32 a,deUint32 b)447 DE_INLINE deBool	deUint32Equal	(deUint32 a, deUint32 b)	{ return (a == b);	}
deInt64Equal(deInt64 a,deInt64 b)448 DE_INLINE deBool	deInt64Equal	(deInt64 a, deInt64 b)		{ return (a == b);	}
deUint64Equal(deUint64 a,deUint64 b)449 DE_INLINE deBool	deUint64Equal	(deUint64 a, deUint64 b)	{ return (a == b);	}
450 
dePointerHash(const void * ptr)451 DE_INLINE int dePointerHash (const void* ptr)
452 {
453 	deUintptr val = (deUintptr)ptr;
454 #if (DE_PTR_SIZE == 4)
455 	return deInt32Hash((int)val);
456 #elif (DE_PTR_SIZE == 8)
457 	return deInt64Hash((deInt64)val);
458 #else
459 #	error Unsupported pointer size.
460 #endif
461 }
462 
dePointerEqual(const void * a,const void * b)463 DE_INLINE deBool dePointerEqual (const void* a, const void* b)
464 {
465 	return (a == b);
466 }
467 
468 /**
469  *	\brief	Modulo that generates the same sign as divisor and rounds toward
470  *			negative infinity -- assuming c99 %-operator.
471  */
deInt32ModF(deInt32 n,deInt32 d)472 DE_INLINE deInt32 deInt32ModF (deInt32 n, deInt32 d)
473 {
474 	deInt32 r = n%d;
475 	if ((r > 0 && d < 0) || (r < 0 && d > 0)) r = r+d;
476 	return r;
477 }
478 
deInt64InInt32Range(deInt64 x)479 DE_INLINE deBool deInt64InInt32Range (deInt64 x)
480 {
481 	return ((x >= (-1ll<<31)) && (x <= ((1ll<<31)-1)));
482 }
483 
484 
deBitMask32(int leastSignificantBitNdx,int numBits)485 DE_INLINE deUint32 deBitMask32 (int leastSignificantBitNdx, int numBits)
486 {
487 	DE_ASSERT(deInRange32(leastSignificantBitNdx, 0, 32));
488 	DE_ASSERT(deInRange32(numBits, 0, 32));
489 	DE_ASSERT(deInRange32(leastSignificantBitNdx+numBits, 0, 32));
490 
491 	if (numBits < 32 && leastSignificantBitNdx < 32)
492 		return ((1u<<numBits)-1u) << (deUint32)leastSignificantBitNdx;
493 	else if (numBits == 0 && leastSignificantBitNdx == 32)
494 		return 0u;
495 	else
496 	{
497 		DE_ASSERT(numBits == 32 && leastSignificantBitNdx == 0);
498 		return 0xFFFFFFFFu;
499 	}
500 }
501 
deUintMaxValue32(int numBits)502 DE_INLINE deUint32 deUintMaxValue32 (int numBits)
503 {
504 	DE_ASSERT(deInRange32(numBits, 1, 32));
505 	if (numBits < 32)
506 		return ((1u<<numBits)-1u);
507 	else
508 		return 0xFFFFFFFFu;
509 }
510 
deIntMaxValue32(int numBits)511 DE_INLINE deInt32 deIntMaxValue32 (int numBits)
512 {
513 	DE_ASSERT(deInRange32(numBits, 1, 32));
514 	if (numBits < 32)
515 		return ((deInt32)1 << (numBits - 1)) - 1;
516 	else
517 	{
518 		/* avoid undefined behavior of int overflow when shifting */
519 		return 0x7FFFFFFF;
520 	}
521 }
522 
deIntMinValue32(int numBits)523 DE_INLINE deInt32 deIntMinValue32 (int numBits)
524 {
525 	DE_ASSERT(deInRange32(numBits, 1, 32));
526 	if (numBits < 32)
527 		return -((deInt32)1 << (numBits - 1));
528 	else
529 	{
530 		/* avoid undefined behavior of int overflow when shifting */
531 		return (deInt32)(-0x7FFFFFFF - 1);
532 	}
533 }
534 
535 DE_END_EXTERN_C
536 
537 #endif /* _DEINT32_H */
538