1 | /* |
---|
2 | Copyright (c) 2003-2006 Gino van den Bergen / Erwin Coumans http://continuousphysics.com/Bullet/ |
---|
3 | |
---|
4 | This software is provided 'as-is', without any express or implied warranty. |
---|
5 | In no event will the authors be held liable for any damages arising from the use of this software. |
---|
6 | Permission is granted to anyone to use this software for any purpose, |
---|
7 | including commercial applications, and to alter it and redistribute it freely, |
---|
8 | subject to the following restrictions: |
---|
9 | |
---|
10 | 1. The origin of this software must not be misrepresented; you must not claim that you wrote the original software. If you use this software in a product, an acknowledgment in the product documentation would be appreciated but is not required. |
---|
11 | 2. Altered source versions must be plainly marked as such, and must not be misrepresented as being the original software. |
---|
12 | 3. This notice may not be removed or altered from any source distribution. |
---|
13 | */ |
---|
14 | |
---|
15 | |
---|
16 | |
---|
17 | #ifndef SIMD___SCALAR_H |
---|
18 | #define SIMD___SCALAR_H |
---|
19 | |
---|
20 | #include <math.h> |
---|
21 | |
---|
22 | #include <stdlib.h>//size_t for MSVC 6.0 |
---|
23 | |
---|
24 | #include <cstdlib> |
---|
25 | #include <cfloat> |
---|
26 | #include <float.h> |
---|
27 | |
---|
28 | #define BT_BULLET_VERSION 274 |
---|
29 | |
---|
30 | inline int btGetVersion() |
---|
31 | { |
---|
32 | return BT_BULLET_VERSION; |
---|
33 | } |
---|
34 | |
---|
35 | #if defined(DEBUG) || defined (_DEBUG) |
---|
36 | #define BT_DEBUG |
---|
37 | #endif |
---|
38 | |
---|
39 | |
---|
40 | #ifdef WIN32 |
---|
41 | |
---|
42 | #if defined(__MINGW32__) || defined(__CYGWIN__) || (defined (_MSC_VER) && _MSC_VER < 1300) |
---|
43 | |
---|
44 | #define SIMD_FORCE_INLINE inline |
---|
45 | #define ATTRIBUTE_ALIGNED16(a) a |
---|
46 | #define ATTRIBUTE_ALIGNED128(a) a |
---|
47 | #else |
---|
48 | //#define BT_HAS_ALIGNED_ALLOCATOR |
---|
49 | #pragma warning(disable : 4324) // disable padding warning |
---|
50 | // #pragma warning(disable:4530) // Disable the exception disable but used in MSCV Stl warning. |
---|
51 | // #pragma warning(disable:4996) //Turn off warnings about deprecated C routines |
---|
52 | // #pragma warning(disable:4786) // Disable the "debug name too long" warning |
---|
53 | |
---|
54 | #define SIMD_FORCE_INLINE __forceinline |
---|
55 | #define ATTRIBUTE_ALIGNED16(a) __declspec(align(16)) a |
---|
56 | #define ATTRIBUTE_ALIGNED128(a) __declspec (align(128)) a |
---|
57 | #ifdef _XBOX |
---|
58 | #define BT_USE_VMX128 |
---|
59 | |
---|
60 | #include <ppcintrinsics.h> |
---|
61 | #define BT_HAVE_NATIVE_FSEL |
---|
62 | #define btFsel(a,b,c) __fsel((a),(b),(c)) |
---|
63 | #else |
---|
64 | |
---|
65 | #if (defined (WIN32) && (_MSC_VER) && _MSC_VER >= 1400) && (!defined (BT_USE_DOUBLE_PRECISION)) |
---|
66 | #define BT_USE_SSE |
---|
67 | #include <emmintrin.h> |
---|
68 | #endif |
---|
69 | |
---|
70 | #endif//_XBOX |
---|
71 | |
---|
72 | #endif //__MINGW32__ |
---|
73 | |
---|
74 | #include <assert.h> |
---|
75 | #ifdef BT_DEBUG |
---|
76 | #define btAssert assert |
---|
77 | #else |
---|
78 | #define btAssert(x) |
---|
79 | #endif |
---|
80 | //btFullAssert is optional, slows down a lot |
---|
81 | #define btFullAssert(x) |
---|
82 | |
---|
83 | #define btLikely(_c) _c |
---|
84 | #define btUnlikely(_c) _c |
---|
85 | |
---|
86 | #else |
---|
87 | |
---|
88 | #if defined (__CELLOS_LV2__) |
---|
89 | #define SIMD_FORCE_INLINE inline |
---|
90 | #define ATTRIBUTE_ALIGNED16(a) a __attribute__ ((aligned (16))) |
---|
91 | #define ATTRIBUTE_ALIGNED128(a) a __attribute__ ((aligned (128))) |
---|
92 | #ifndef assert |
---|
93 | #include <assert.h> |
---|
94 | #endif |
---|
95 | #ifdef BT_DEBUG |
---|
96 | #define btAssert assert |
---|
97 | #else |
---|
98 | #define btAssert(x) |
---|
99 | #endif |
---|
100 | //btFullAssert is optional, slows down a lot |
---|
101 | #define btFullAssert(x) |
---|
102 | |
---|
103 | #define btLikely(_c) _c |
---|
104 | #define btUnlikely(_c) _c |
---|
105 | |
---|
106 | #else |
---|
107 | |
---|
108 | #ifdef USE_LIBSPE2 |
---|
109 | |
---|
110 | #define SIMD_FORCE_INLINE __inline |
---|
111 | #define ATTRIBUTE_ALIGNED16(a) a __attribute__ ((aligned (16))) |
---|
112 | #define ATTRIBUTE_ALIGNED128(a) a __attribute__ ((aligned (128))) |
---|
113 | #ifndef assert |
---|
114 | #include <assert.h> |
---|
115 | #endif |
---|
116 | #ifdef BT_DEBUG |
---|
117 | #define btAssert assert |
---|
118 | #else |
---|
119 | #define btAssert(x) |
---|
120 | #endif |
---|
121 | //btFullAssert is optional, slows down a lot |
---|
122 | #define btFullAssert(x) |
---|
123 | |
---|
124 | |
---|
125 | #define btLikely(_c) __builtin_expect((_c), 1) |
---|
126 | #define btUnlikely(_c) __builtin_expect((_c), 0) |
---|
127 | |
---|
128 | |
---|
129 | #else |
---|
130 | //non-windows systems |
---|
131 | |
---|
132 | #define SIMD_FORCE_INLINE inline |
---|
133 | ///@todo: check out alignment methods for other platforms/compilers |
---|
134 | ///#define ATTRIBUTE_ALIGNED16(a) a __attribute__ ((aligned (16))) |
---|
135 | ///#define ATTRIBUTE_ALIGNED128(a) a __attribute__ ((aligned (128))) |
---|
136 | #define ATTRIBUTE_ALIGNED16(a) a |
---|
137 | #define ATTRIBUTE_ALIGNED128(a) a |
---|
138 | #ifndef assert |
---|
139 | #include <assert.h> |
---|
140 | #endif |
---|
141 | |
---|
142 | #if defined(DEBUG) || defined (_DEBUG) |
---|
143 | #define btAssert assert |
---|
144 | #else |
---|
145 | #define btAssert(x) |
---|
146 | #endif |
---|
147 | |
---|
148 | //btFullAssert is optional, slows down a lot |
---|
149 | #define btFullAssert(x) |
---|
150 | #define btLikely(_c) _c |
---|
151 | #define btUnlikely(_c) _c |
---|
152 | |
---|
153 | |
---|
154 | #endif // LIBSPE2 |
---|
155 | |
---|
156 | #endif //__CELLOS_LV2__ |
---|
157 | #endif |
---|
158 | |
---|
159 | /// older compilers (gcc 3.x) and Sun needs double version of sqrt etc. |
---|
160 | /// exclude Apple Intel (i's assumed to be a Macbook or new Intel Dual Core Processor) |
---|
161 | #if defined (__sun) || defined (__sun__) || defined (__sparc) || (defined (__APPLE__) && ! defined (__i386__)) |
---|
162 | //use slow double float precision operation on those platforms |
---|
163 | #ifndef BT_USE_DOUBLE_PRECISION |
---|
164 | #define BT_FORCE_DOUBLE_FUNCTIONS |
---|
165 | #endif |
---|
166 | #endif |
---|
167 | |
---|
168 | ///The btScalar type abstracts floating point numbers, to easily switch between double and single floating point precision. |
---|
169 | #if defined(BT_USE_DOUBLE_PRECISION) |
---|
170 | typedef double btScalar; |
---|
171 | #else |
---|
172 | typedef float btScalar; |
---|
173 | #endif |
---|
174 | |
---|
175 | |
---|
176 | |
---|
177 | #define BT_DECLARE_ALIGNED_ALLOCATOR() \ |
---|
178 | SIMD_FORCE_INLINE void* operator new(size_t sizeInBytes) { return btAlignedAlloc(sizeInBytes,16); } \ |
---|
179 | SIMD_FORCE_INLINE void operator delete(void* ptr) { btAlignedFree(ptr); } \ |
---|
180 | SIMD_FORCE_INLINE void* operator new(size_t, void* ptr) { return ptr; } \ |
---|
181 | SIMD_FORCE_INLINE void operator delete(void*, void*) { } \ |
---|
182 | SIMD_FORCE_INLINE void* operator new[](size_t sizeInBytes) { return btAlignedAlloc(sizeInBytes,16); } \ |
---|
183 | SIMD_FORCE_INLINE void operator delete[](void* ptr) { btAlignedFree(ptr); } \ |
---|
184 | SIMD_FORCE_INLINE void* operator new[](size_t, void* ptr) { return ptr; } \ |
---|
185 | SIMD_FORCE_INLINE void operator delete[](void*, void*) { } \ |
---|
186 | |
---|
187 | |
---|
188 | |
---|
189 | #if defined(BT_USE_DOUBLE_PRECISION) || defined(BT_FORCE_DOUBLE_FUNCTIONS) |
---|
190 | |
---|
191 | SIMD_FORCE_INLINE btScalar btSqrt(btScalar x) { return sqrt(x); } |
---|
192 | SIMD_FORCE_INLINE btScalar btFabs(btScalar x) { return fabs(x); } |
---|
193 | SIMD_FORCE_INLINE btScalar btCos(btScalar x) { return cos(x); } |
---|
194 | SIMD_FORCE_INLINE btScalar btSin(btScalar x) { return sin(x); } |
---|
195 | SIMD_FORCE_INLINE btScalar btTan(btScalar x) { return tan(x); } |
---|
196 | SIMD_FORCE_INLINE btScalar btAcos(btScalar x) { return acos(x); } |
---|
197 | SIMD_FORCE_INLINE btScalar btAsin(btScalar x) { return asin(x); } |
---|
198 | SIMD_FORCE_INLINE btScalar btAtan(btScalar x) { return atan(x); } |
---|
199 | SIMD_FORCE_INLINE btScalar btAtan2(btScalar x, btScalar y) { return atan2(x, y); } |
---|
200 | SIMD_FORCE_INLINE btScalar btExp(btScalar x) { return exp(x); } |
---|
201 | SIMD_FORCE_INLINE btScalar btLog(btScalar x) { return log(x); } |
---|
202 | SIMD_FORCE_INLINE btScalar btPow(btScalar x,btScalar y) { return pow(x,y); } |
---|
203 | |
---|
204 | #else |
---|
205 | |
---|
206 | SIMD_FORCE_INLINE btScalar btSqrt(btScalar y) |
---|
207 | { |
---|
208 | #ifdef USE_APPROXIMATION |
---|
209 | double x, z, tempf; |
---|
210 | unsigned long *tfptr = ((unsigned long *)&tempf) + 1; |
---|
211 | |
---|
212 | tempf = y; |
---|
213 | *tfptr = (0xbfcdd90a - *tfptr)>>1; /* estimate of 1/sqrt(y) */ |
---|
214 | x = tempf; |
---|
215 | z = y*btScalar(0.5); /* hoist out the /2 */ |
---|
216 | x = (btScalar(1.5)*x)-(x*x)*(x*z); /* iteration formula */ |
---|
217 | x = (btScalar(1.5)*x)-(x*x)*(x*z); |
---|
218 | x = (btScalar(1.5)*x)-(x*x)*(x*z); |
---|
219 | x = (btScalar(1.5)*x)-(x*x)*(x*z); |
---|
220 | x = (btScalar(1.5)*x)-(x*x)*(x*z); |
---|
221 | return x*y; |
---|
222 | #else |
---|
223 | return sqrtf(y); |
---|
224 | #endif |
---|
225 | } |
---|
226 | SIMD_FORCE_INLINE btScalar btFabs(btScalar x) { return fabsf(x); } |
---|
227 | SIMD_FORCE_INLINE btScalar btCos(btScalar x) { return cosf(x); } |
---|
228 | SIMD_FORCE_INLINE btScalar btSin(btScalar x) { return sinf(x); } |
---|
229 | SIMD_FORCE_INLINE btScalar btTan(btScalar x) { return tanf(x); } |
---|
230 | SIMD_FORCE_INLINE btScalar btAcos(btScalar x) { |
---|
231 | btAssert(x <= btScalar(1.)); |
---|
232 | return acosf(x); |
---|
233 | } |
---|
234 | SIMD_FORCE_INLINE btScalar btAsin(btScalar x) { return asinf(x); } |
---|
235 | SIMD_FORCE_INLINE btScalar btAtan(btScalar x) { return atanf(x); } |
---|
236 | SIMD_FORCE_INLINE btScalar btAtan2(btScalar x, btScalar y) { return atan2f(x, y); } |
---|
237 | SIMD_FORCE_INLINE btScalar btExp(btScalar x) { return expf(x); } |
---|
238 | SIMD_FORCE_INLINE btScalar btLog(btScalar x) { return logf(x); } |
---|
239 | SIMD_FORCE_INLINE btScalar btPow(btScalar x,btScalar y) { return powf(x,y); } |
---|
240 | |
---|
241 | #endif |
---|
242 | |
---|
243 | #define SIMD_2_PI btScalar(6.283185307179586232) |
---|
244 | #define SIMD_PI (SIMD_2_PI * btScalar(0.5)) |
---|
245 | #define SIMD_HALF_PI (SIMD_2_PI * btScalar(0.25)) |
---|
246 | #define SIMD_RADS_PER_DEG (SIMD_2_PI / btScalar(360.0)) |
---|
247 | #define SIMD_DEGS_PER_RAD (btScalar(360.0) / SIMD_2_PI) |
---|
248 | |
---|
249 | #ifdef BT_USE_DOUBLE_PRECISION |
---|
250 | #define SIMD_EPSILON DBL_EPSILON |
---|
251 | #define SIMD_INFINITY DBL_MAX |
---|
252 | #else |
---|
253 | #define SIMD_EPSILON FLT_EPSILON |
---|
254 | #define SIMD_INFINITY FLT_MAX |
---|
255 | #endif |
---|
256 | |
---|
257 | SIMD_FORCE_INLINE btScalar btAtan2Fast(btScalar y, btScalar x) |
---|
258 | { |
---|
259 | btScalar coeff_1 = SIMD_PI / 4.0f; |
---|
260 | btScalar coeff_2 = 3.0f * coeff_1; |
---|
261 | btScalar abs_y = btFabs(y); |
---|
262 | btScalar angle; |
---|
263 | if (x >= 0.0f) { |
---|
264 | btScalar r = (x - abs_y) / (x + abs_y); |
---|
265 | angle = coeff_1 - coeff_1 * r; |
---|
266 | } else { |
---|
267 | btScalar r = (x + abs_y) / (abs_y - x); |
---|
268 | angle = coeff_2 - coeff_1 * r; |
---|
269 | } |
---|
270 | return (y < 0.0f) ? -angle : angle; |
---|
271 | } |
---|
272 | |
---|
273 | SIMD_FORCE_INLINE bool btFuzzyZero(btScalar x) { return btFabs(x) < SIMD_EPSILON; } |
---|
274 | |
---|
275 | SIMD_FORCE_INLINE bool btEqual(btScalar a, btScalar eps) { |
---|
276 | return (((a) <= eps) && !((a) < -eps)); |
---|
277 | } |
---|
278 | SIMD_FORCE_INLINE bool btGreaterEqual (btScalar a, btScalar eps) { |
---|
279 | return (!((a) <= eps)); |
---|
280 | } |
---|
281 | |
---|
282 | |
---|
283 | SIMD_FORCE_INLINE int btIsNegative(btScalar x) { |
---|
284 | return x < btScalar(0.0) ? 1 : 0; |
---|
285 | } |
---|
286 | |
---|
287 | SIMD_FORCE_INLINE btScalar btRadians(btScalar x) { return x * SIMD_RADS_PER_DEG; } |
---|
288 | SIMD_FORCE_INLINE btScalar btDegrees(btScalar x) { return x * SIMD_DEGS_PER_RAD; } |
---|
289 | |
---|
290 | #define BT_DECLARE_HANDLE(name) typedef struct name##__ { int unused; } *name |
---|
291 | |
---|
292 | #ifndef btFsel |
---|
293 | SIMD_FORCE_INLINE btScalar btFsel(btScalar a, btScalar b, btScalar c) |
---|
294 | { |
---|
295 | return a >= 0 ? b : c; |
---|
296 | } |
---|
297 | #endif |
---|
298 | #define btFsels(a,b,c) (btScalar)btFsel(a,b,c) |
---|
299 | |
---|
300 | |
---|
301 | SIMD_FORCE_INLINE bool btMachineIsLittleEndian() |
---|
302 | { |
---|
303 | long int i = 1; |
---|
304 | const char *p = (const char *) &i; |
---|
305 | if (p[0] == 1) // Lowest address contains the least significant byte |
---|
306 | return true; |
---|
307 | else |
---|
308 | return false; |
---|
309 | } |
---|
310 | |
---|
311 | |
---|
312 | |
---|
313 | ///btSelect avoids branches, which makes performance much better for consoles like Playstation 3 and XBox 360 |
---|
314 | ///Thanks Phil Knight. See also http://www.cellperformance.com/articles/2006/04/more_techniques_for_eliminatin_1.html |
---|
315 | SIMD_FORCE_INLINE unsigned btSelect(unsigned condition, unsigned valueIfConditionNonZero, unsigned valueIfConditionZero) |
---|
316 | { |
---|
317 | // Set testNz to 0xFFFFFFFF if condition is nonzero, 0x00000000 if condition is zero |
---|
318 | // Rely on positive value or'ed with its negative having sign bit on |
---|
319 | // and zero value or'ed with its negative (which is still zero) having sign bit off |
---|
320 | // Use arithmetic shift right, shifting the sign bit through all 32 bits |
---|
321 | unsigned testNz = (unsigned)(((int)condition | -(int)condition) >> 31); |
---|
322 | unsigned testEqz = ~testNz; |
---|
323 | return ((valueIfConditionNonZero & testNz) | (valueIfConditionZero & testEqz)); |
---|
324 | } |
---|
325 | SIMD_FORCE_INLINE int btSelect(unsigned condition, int valueIfConditionNonZero, int valueIfConditionZero) |
---|
326 | { |
---|
327 | unsigned testNz = (unsigned)(((int)condition | -(int)condition) >> 31); |
---|
328 | unsigned testEqz = ~testNz; |
---|
329 | return static_cast<int>((valueIfConditionNonZero & testNz) | (valueIfConditionZero & testEqz)); |
---|
330 | } |
---|
331 | SIMD_FORCE_INLINE float btSelect(unsigned condition, float valueIfConditionNonZero, float valueIfConditionZero) |
---|
332 | { |
---|
333 | #ifdef BT_HAVE_NATIVE_FSEL |
---|
334 | return (float)btFsel((btScalar)condition - btScalar(1.0f), valueIfConditionNonZero, valueIfConditionZero); |
---|
335 | #else |
---|
336 | return (condition != 0) ? valueIfConditionNonZero : valueIfConditionZero; |
---|
337 | #endif |
---|
338 | } |
---|
339 | |
---|
340 | template<typename T> SIMD_FORCE_INLINE void btSwap(T& a, T& b) |
---|
341 | { |
---|
342 | T tmp = a; |
---|
343 | a = b; |
---|
344 | b = tmp; |
---|
345 | } |
---|
346 | |
---|
347 | |
---|
348 | //PCK: endian swapping functions |
---|
349 | SIMD_FORCE_INLINE unsigned btSwapEndian(unsigned val) |
---|
350 | { |
---|
351 | return (((val & 0xff000000) >> 24) | ((val & 0x00ff0000) >> 8) | ((val & 0x0000ff00) << 8) | ((val & 0x000000ff) << 24)); |
---|
352 | } |
---|
353 | |
---|
354 | SIMD_FORCE_INLINE unsigned short btSwapEndian(unsigned short val) |
---|
355 | { |
---|
356 | return static_cast<unsigned short>(((val & 0xff00) >> 8) | ((val & 0x00ff) << 8)); |
---|
357 | } |
---|
358 | |
---|
359 | SIMD_FORCE_INLINE unsigned btSwapEndian(int val) |
---|
360 | { |
---|
361 | return btSwapEndian((unsigned)val); |
---|
362 | } |
---|
363 | |
---|
364 | SIMD_FORCE_INLINE unsigned short btSwapEndian(short val) |
---|
365 | { |
---|
366 | return btSwapEndian((unsigned short) val); |
---|
367 | } |
---|
368 | |
---|
369 | ///btSwapFloat uses using char pointers to swap the endianness |
---|
370 | ////btSwapFloat/btSwapDouble will NOT return a float, because the machine might 'correct' invalid floating point values |
---|
371 | ///Not all values of sign/exponent/mantissa are valid floating point numbers according to IEEE 754. |
---|
372 | ///When a floating point unit is faced with an invalid value, it may actually change the value, or worse, throw an exception. |
---|
373 | ///In most systems, running user mode code, you wouldn't get an exception, but instead the hardware/os/runtime will 'fix' the number for you. |
---|
374 | ///so instead of returning a float/double, we return integer/long long integer |
---|
375 | SIMD_FORCE_INLINE unsigned int btSwapEndianFloat(float d) |
---|
376 | { |
---|
377 | unsigned int a = 0; |
---|
378 | unsigned char *dst = (unsigned char *)&a; |
---|
379 | unsigned char *src = (unsigned char *)&d; |
---|
380 | |
---|
381 | dst[0] = src[3]; |
---|
382 | dst[1] = src[2]; |
---|
383 | dst[2] = src[1]; |
---|
384 | dst[3] = src[0]; |
---|
385 | return a; |
---|
386 | } |
---|
387 | |
---|
388 | // unswap using char pointers |
---|
389 | SIMD_FORCE_INLINE float btUnswapEndianFloat(unsigned int a) |
---|
390 | { |
---|
391 | float d = 0.0f; |
---|
392 | unsigned char *src = (unsigned char *)&a; |
---|
393 | unsigned char *dst = (unsigned char *)&d; |
---|
394 | |
---|
395 | dst[0] = src[3]; |
---|
396 | dst[1] = src[2]; |
---|
397 | dst[2] = src[1]; |
---|
398 | dst[3] = src[0]; |
---|
399 | |
---|
400 | return d; |
---|
401 | } |
---|
402 | |
---|
403 | |
---|
404 | // swap using char pointers |
---|
405 | SIMD_FORCE_INLINE void btSwapEndianDouble(double d, unsigned char* dst) |
---|
406 | { |
---|
407 | unsigned char *src = (unsigned char *)&d; |
---|
408 | |
---|
409 | dst[0] = src[7]; |
---|
410 | dst[1] = src[6]; |
---|
411 | dst[2] = src[5]; |
---|
412 | dst[3] = src[4]; |
---|
413 | dst[4] = src[3]; |
---|
414 | dst[5] = src[2]; |
---|
415 | dst[6] = src[1]; |
---|
416 | dst[7] = src[0]; |
---|
417 | |
---|
418 | } |
---|
419 | |
---|
420 | // unswap using char pointers |
---|
421 | SIMD_FORCE_INLINE double btUnswapEndianDouble(const unsigned char *src) |
---|
422 | { |
---|
423 | double d = 0.0; |
---|
424 | unsigned char *dst = (unsigned char *)&d; |
---|
425 | |
---|
426 | dst[0] = src[7]; |
---|
427 | dst[1] = src[6]; |
---|
428 | dst[2] = src[5]; |
---|
429 | dst[3] = src[4]; |
---|
430 | dst[4] = src[3]; |
---|
431 | dst[5] = src[2]; |
---|
432 | dst[6] = src[1]; |
---|
433 | dst[7] = src[0]; |
---|
434 | |
---|
435 | return d; |
---|
436 | } |
---|
437 | |
---|
438 | |
---|
439 | #endif //SIMD___SCALAR_H |
---|