Gitiles
Code Review Sign In
gerrit-public.fairphone.software / platform / frameworks / rs / bcd5b9af756d10317faf54fa3742f89dfacef152 / . / driver / runtime / rs_cl.c
blob: d054bcf1d6670dadf1506b841c123f418a46fd4b [file] [log] [blame]
Stephen Hines5a470202013-05-29 15:36:18 -0700[diff] [blame]1#include "rs_types.rsh"
2
3extern float2 __attribute__((overloadable)) convert_float2(int2 c);
4extern float3 __attribute__((overloadable)) convert_float3(int3 c);
5extern float4 __attribute__((overloadable)) convert_float4(int4 c);
6
7extern int2 __attribute__((overloadable)) convert_int2(float2 c);
8extern int3 __attribute__((overloadable)) convert_int3(float3 c);
9extern int4 __attribute__((overloadable)) convert_int4(float4 c);
10
11
12extern float __attribute__((overloadable)) fmin(float v, float v2);
13extern float2 __attribute__((overloadable)) fmin(float2 v, float v2);
14extern float3 __attribute__((overloadable)) fmin(float3 v, float v2);
15extern float4 __attribute__((overloadable)) fmin(float4 v, float v2);
16
17extern float __attribute__((overloadable)) fmax(float v, float v2);
18extern float2 __attribute__((overloadable)) fmax(float2 v, float v2);
19extern float3 __attribute__((overloadable)) fmax(float3 v, float v2);
20extern float4 __attribute__((overloadable)) fmax(float4 v, float v2);
21
22// Float ops, 6.11.2
23
24#define FN_FUNC_FN(fnc) \
25extern float2 __attribute__((overloadable)) fnc(float2 v) { \
26 float2 r; \
27 r.x = fnc(v.x); \
28 r.y = fnc(v.y); \
29 return r; \
30} \
31extern float3 __attribute__((overloadable)) fnc(float3 v) { \
32 float3 r; \
33 r.x = fnc(v.x); \
34 r.y = fnc(v.y); \
35 r.z = fnc(v.z); \
36 return r; \
37} \
38extern float4 __attribute__((overloadable)) fnc(float4 v) { \
39 float4 r; \
40 r.x = fnc(v.x); \
41 r.y = fnc(v.y); \
42 r.z = fnc(v.z); \
43 r.w = fnc(v.w); \
44 return r; \
45}
46
47#define IN_FUNC_FN(fnc) \
48extern int2 __attribute__((overloadable)) fnc(float2 v) { \
49 int2 r; \
50 r.x = fnc(v.x); \
51 r.y = fnc(v.y); \
52 return r; \
53} \
54extern int3 __attribute__((overloadable)) fnc(float3 v) { \
55 int3 r; \
56 r.x = fnc(v.x); \
57 r.y = fnc(v.y); \
58 r.z = fnc(v.z); \
59 return r; \
60} \
61extern int4 __attribute__((overloadable)) fnc(float4 v) { \
62 int4 r; \
63 r.x = fnc(v.x); \
64 r.y = fnc(v.y); \
65 r.z = fnc(v.z); \
66 r.w = fnc(v.w); \
67 return r; \
68}
69
70#define FN_FUNC_FN_FN(fnc) \
71extern float2 __attribute__((overloadable)) fnc(float2 v1, float2 v2) { \
72 float2 r; \
73 r.x = fnc(v1.x, v2.x); \
74 r.y = fnc(v1.y, v2.y); \
75 return r; \
76} \
77extern float3 __attribute__((overloadable)) fnc(float3 v1, float3 v2) { \
78 float3 r; \
79 r.x = fnc(v1.x, v2.x); \
80 r.y = fnc(v1.y, v2.y); \
81 r.z = fnc(v1.z, v2.z); \
82 return r; \
83} \
84extern float4 __attribute__((overloadable)) fnc(float4 v1, float4 v2) { \
85 float4 r; \
86 r.x = fnc(v1.x, v2.x); \
87 r.y = fnc(v1.y, v2.y); \
88 r.z = fnc(v1.z, v2.z); \
89 r.w = fnc(v1.w, v2.w); \
90 return r; \
91}
92
93#define FN_FUNC_FN_F(fnc) \
94extern float2 __attribute__((overloadable)) fnc(float2 v1, float v2) { \
95 float2 r; \
96 r.x = fnc(v1.x, v2); \
97 r.y = fnc(v1.y, v2); \
98 return r; \
99} \
100extern float3 __attribute__((overloadable)) fnc(float3 v1, float v2) { \
101 float3 r; \
102 r.x = fnc(v1.x, v2); \
103 r.y = fnc(v1.y, v2); \
104 r.z = fnc(v1.z, v2); \
105 return r; \
106} \
107extern float4 __attribute__((overloadable)) fnc(float4 v1, float v2) { \
108 float4 r; \
109 r.x = fnc(v1.x, v2); \
110 r.y = fnc(v1.y, v2); \
111 r.z = fnc(v1.z, v2); \
112 r.w = fnc(v1.w, v2); \
113 return r; \
114}
115
116#define FN_FUNC_FN_IN(fnc) \
117extern float2 __attribute__((overloadable)) fnc(float2 v1, int2 v2) { \
118 float2 r; \
119 r.x = fnc(v1.x, v2.x); \
120 r.y = fnc(v1.y, v2.y); \
121 return r; \
122} \
123extern float3 __attribute__((overloadable)) fnc(float3 v1, int3 v2) { \
124 float3 r; \
125 r.x = fnc(v1.x, v2.x); \
126 r.y = fnc(v1.y, v2.y); \
127 r.z = fnc(v1.z, v2.z); \
128 return r; \
129} \
130extern float4 __attribute__((overloadable)) fnc(float4 v1, int4 v2) { \
131 float4 r; \
132 r.x = fnc(v1.x, v2.x); \
133 r.y = fnc(v1.y, v2.y); \
134 r.z = fnc(v1.z, v2.z); \
135 r.w = fnc(v1.w, v2.w); \
136 return r; \
137}
138
139#define FN_FUNC_FN_I(fnc) \
140extern float2 __attribute__((overloadable)) fnc(float2 v1, int v2) { \
141 float2 r; \
142 r.x = fnc(v1.x, v2); \
143 r.y = fnc(v1.y, v2); \
144 return r; \
145} \
146extern float3 __attribute__((overloadable)) fnc(float3 v1, int v2) { \
147 float3 r; \
148 r.x = fnc(v1.x, v2); \
149 r.y = fnc(v1.y, v2); \
150 r.z = fnc(v1.z, v2); \
151 return r; \
152} \
153extern float4 __attribute__((overloadable)) fnc(float4 v1, int v2) { \
154 float4 r; \
155 r.x = fnc(v1.x, v2); \
156 r.y = fnc(v1.y, v2); \
157 r.z = fnc(v1.z, v2); \
158 r.w = fnc(v1.w, v2); \
159 return r; \
160}
161
162#define FN_FUNC_FN_PFN(fnc) \
163extern float2 __attribute__((overloadable)) \
164 fnc(float2 v1, float2 *v2) { \
165 float2 r; \
166 float t[2]; \
167 r.x = fnc(v1.x, &t[0]); \
168 r.y = fnc(v1.y, &t[1]); \
169 v2->x = t[0]; \
170 v2->y = t[1]; \
171 return r; \
172} \
173extern float3 __attribute__((overloadable)) \
174 fnc(float3 v1, float3 *v2) { \
175 float3 r; \
176 float t[3]; \
177 r.x = fnc(v1.x, &t[0]); \
178 r.y = fnc(v1.y, &t[1]); \
179 r.z = fnc(v1.z, &t[2]); \
180 v2->x = t[0]; \
181 v2->y = t[1]; \
182 v2->z = t[2]; \
183 return r; \
184} \
185extern float4 __attribute__((overloadable)) \
186 fnc(float4 v1, float4 *v2) { \
187 float4 r; \
188 float t[4]; \
189 r.x = fnc(v1.x, &t[0]); \
190 r.y = fnc(v1.y, &t[1]); \
191 r.z = fnc(v1.z, &t[2]); \
192 r.w = fnc(v1.w, &t[3]); \
193 v2->x = t[0]; \
194 v2->y = t[1]; \
195 v2->z = t[2]; \
196 v2->w = t[3]; \
197 return r; \
198}
199
200#define FN_FUNC_FN_PIN(fnc) \
201extern float2 __attribute__((overloadable)) fnc(float2 v1, int2 *v2) { \
202 float2 r; \
203 int t[2]; \
204 r.x = fnc(v1.x, &t[0]); \
205 r.y = fnc(v1.y, &t[1]); \
206 v2->x = t[0]; \
207 v2->y = t[1]; \
208 return r; \
209} \
210extern float3 __attribute__((overloadable)) fnc(float3 v1, int3 *v2) { \
211 float3 r; \
212 int t[3]; \
213 r.x = fnc(v1.x, &t[0]); \
214 r.y = fnc(v1.y, &t[1]); \
215 r.z = fnc(v1.z, &t[2]); \
216 v2->x = t[0]; \
217 v2->y = t[1]; \
218 v2->z = t[2]; \
219 return r; \
220} \
221extern float4 __attribute__((overloadable)) fnc(float4 v1, int4 *v2) { \
222 float4 r; \
223 int t[4]; \
224 r.x = fnc(v1.x, &t[0]); \
225 r.y = fnc(v1.y, &t[1]); \
226 r.z = fnc(v1.z, &t[2]); \
227 r.w = fnc(v1.w, &t[3]); \
228 v2->x = t[0]; \
229 v2->y = t[1]; \
230 v2->z = t[2]; \
231 v2->w = t[3]; \
232 return r; \
233}
234
235#define FN_FUNC_FN_FN_FN(fnc) \
236extern float2 __attribute__((overloadable)) \
237 fnc(float2 v1, float2 v2, float2 v3) { \
238 float2 r; \
239 r.x = fnc(v1.x, v2.x, v3.x); \
240 r.y = fnc(v1.y, v2.y, v3.y); \
241 return r; \
242} \
243extern float3 __attribute__((overloadable)) \
244 fnc(float3 v1, float3 v2, float3 v3) { \
245 float3 r; \
246 r.x = fnc(v1.x, v2.x, v3.x); \
247 r.y = fnc(v1.y, v2.y, v3.y); \
248 r.z = fnc(v1.z, v2.z, v3.z); \
249 return r; \
250} \
251extern float4 __attribute__((overloadable)) \
252 fnc(float4 v1, float4 v2, float4 v3) { \
253 float4 r; \
254 r.x = fnc(v1.x, v2.x, v3.x); \
255 r.y = fnc(v1.y, v2.y, v3.y); \
256 r.z = fnc(v1.z, v2.z, v3.z); \
257 r.w = fnc(v1.w, v2.w, v3.w); \
258 return r; \
259}
260
261#define FN_FUNC_FN_FN_PIN(fnc) \
262extern float2 __attribute__((overloadable)) \
263 fnc(float2 v1, float2 v2, int2 *v3) { \
264 float2 r; \
265 int t[2]; \
266 r.x = fnc(v1.x, v2.x, &t[0]); \
267 r.y = fnc(v1.y, v2.y, &t[1]); \
268 v3->x = t[0]; \
269 v3->y = t[1]; \
270 return r; \
271} \
272extern float3 __attribute__((overloadable)) \
273 fnc(float3 v1, float3 v2, int3 *v3) { \
274 float3 r; \
275 int t[3]; \
276 r.x = fnc(v1.x, v2.x, &t[0]); \
277 r.y = fnc(v1.y, v2.y, &t[1]); \
278 r.z = fnc(v1.z, v2.z, &t[2]); \
279 v3->x = t[0]; \
280 v3->y = t[1]; \
281 v3->z = t[2]; \
282 return r; \
283} \
284extern float4 __attribute__((overloadable)) \
285 fnc(float4 v1, float4 v2, int4 *v3) { \
286 float4 r; \
287 int t[4]; \
288 r.x = fnc(v1.x, v2.x, &t[0]); \
289 r.y = fnc(v1.y, v2.y, &t[1]); \
290 r.z = fnc(v1.z, v2.z, &t[2]); \
291 r.w = fnc(v1.w, v2.w, &t[3]); \
292 v3->x = t[0]; \
293 v3->y = t[1]; \
294 v3->z = t[2]; \
295 v3->w = t[3]; \
296 return r; \
297}
298
299static const int iposinf = 0x7f800000;
300static const int ineginf = 0xff800000;
301
302static const float posinf() {
303 float f = *((float*)&iposinf);
304 return f;
305}
306
307static const float neginf() {
308 float f = *((float*)&ineginf);
309 return f;
310}
311
312static bool isinf(float f) {
313 int i = *((int*)(void*)&f);
314 return (i == iposinf) || (i == ineginf);
315}
316
317static bool isnan(float f) {
318 int i = *((int*)(void*)&f);
319 return (((i & 0x7f800000) == 0x7f800000) && (i & 0x007fffff));
320}
321
322static bool isposzero(float f) {
323 int i = *((int*)(void*)&f);
324 return (i == 0x00000000);
325}
326
327static bool isnegzero(float f) {
328 int i = *((int*)(void*)&f);
329 return (i == 0x80000000);
330}
331
332static bool iszero(float f) {
333 return isposzero(f) || isnegzero(f);
334}
335
336
337extern float __attribute__((overloadable)) acos(float);
338FN_FUNC_FN(acos)
339
340extern float __attribute__((overloadable)) acosh(float);
341FN_FUNC_FN(acosh)
342
343
344extern float __attribute__((overloadable)) acospi(float v) {
345 return acos(v) / M_PI;
346}
347FN_FUNC_FN(acospi)
348
349extern float __attribute__((overloadable)) asin(float);
350FN_FUNC_FN(asin)
351
352extern float __attribute__((overloadable)) asinh(float);
353FN_FUNC_FN(asinh)
354
355extern float __attribute__((overloadable)) asinpi(float v) {
356 return asin(v) / M_PI;
357}
358FN_FUNC_FN(asinpi)
359
360extern float __attribute__((overloadable)) atan(float);
361FN_FUNC_FN(atan)
362
363extern float __attribute__((overloadable)) atan2(float, float);
364FN_FUNC_FN_FN(atan2)
365
366extern float __attribute__((overloadable)) atanh(float);
367FN_FUNC_FN(atanh)
368
369extern float __attribute__((overloadable)) atanpi(float v) {
370 return atan(v) / M_PI;
371}
372FN_FUNC_FN(atanpi)
373
374
375extern float __attribute__((overloadable)) atan2pi(float y, float x) {
376 return atan2(y, x) / M_PI;
377}
378FN_FUNC_FN_FN(atan2pi)
379
380extern float __attribute__((overloadable)) cbrt(float);
381FN_FUNC_FN(cbrt)
382
383extern float __attribute__((overloadable)) ceil(float);
384FN_FUNC_FN(ceil)
385
386extern float __attribute__((overloadable)) copysign(float, float);
387FN_FUNC_FN_FN(copysign)
388
389extern float __attribute__((overloadable)) cos(float);
390FN_FUNC_FN(cos)
391
392extern float __attribute__((overloadable)) cosh(float);
393FN_FUNC_FN(cosh)
394
395extern float __attribute__((overloadable)) cospi(float v) {
396 return cos(v * M_PI);
397}
398FN_FUNC_FN(cospi)
399
400extern float __attribute__((overloadable)) erfc(float);
401FN_FUNC_FN(erfc)
402
403extern float __attribute__((overloadable)) erf(float);
404FN_FUNC_FN(erf)
405
406extern float __attribute__((overloadable)) exp(float);
407FN_FUNC_FN(exp)
408
409extern float __attribute__((overloadable)) exp2(float);
410FN_FUNC_FN(exp2)
411
412extern float __attribute__((overloadable)) pow(float, float);
413
414extern float __attribute__((overloadable)) exp10(float v) {
415 return exp2(v * 3.321928095f);
416}
417FN_FUNC_FN(exp10)
418
419extern float __attribute__((overloadable)) expm1(float);
420FN_FUNC_FN(expm1)
421
422extern float __attribute__((overloadable)) fabs(float v) {
423 int i = *((int*)(void*)&v) & 0x7fffffff;
424 return *((float*)(void*)&i);
425}
426FN_FUNC_FN(fabs)
427
428extern float __attribute__((overloadable)) fdim(float, float);
429FN_FUNC_FN_FN(fdim)
430
431extern float __attribute__((overloadable)) floor(float);
432FN_FUNC_FN(floor)
433
434extern float __attribute__((overloadable)) fma(float, float, float);
435FN_FUNC_FN_FN_FN(fma)
436
437extern float __attribute__((overloadable)) fmin(float, float);
438
439extern float __attribute__((overloadable)) fmod(float, float);
440FN_FUNC_FN_FN(fmod)
441
442extern float __attribute__((overloadable)) fract(float v, float *iptr) {
443 int i = (int)floor(v);
444 if (iptr) {
445 iptr[0] = i;
446 }
447 return fmin(v - i, 0x1.fffffep-1f);
448}
449FN_FUNC_FN_PFN(fract)
450
451extern float __attribute__((overloadable)) frexp(float, int *);
452FN_FUNC_FN_PIN(frexp)
453
454extern float __attribute__((overloadable)) hypot(float, float);
455FN_FUNC_FN_FN(hypot)
456
457extern int __attribute__((overloadable)) ilogb(float);
458IN_FUNC_FN(ilogb)
459
460extern float __attribute__((overloadable)) ldexp(float, int);
461FN_FUNC_FN_IN(ldexp)
462FN_FUNC_FN_I(ldexp)
463
464extern float __attribute__((overloadable)) lgamma(float);
465FN_FUNC_FN(lgamma)
466extern float __attribute__((overloadable)) lgamma(float, int*);
467FN_FUNC_FN_PIN(lgamma)
468
469extern float __attribute__((overloadable)) log(float);
470FN_FUNC_FN(log)
471
472extern float __attribute__((overloadable)) log10(float);
473FN_FUNC_FN(log10)
474
475
476extern float __attribute__((overloadable)) log2(float v) {
477 return log10(v) * 3.321928095f;
478}
479FN_FUNC_FN(log2)
480
481extern float __attribute__((overloadable)) log1p(float);
482FN_FUNC_FN(log1p)
483
484extern float __attribute__((overloadable)) logb(float);
485FN_FUNC_FN(logb)
486
487extern float __attribute__((overloadable)) mad(float a, float b, float c) {
488 return a * b + c;
489}
490extern float2 __attribute__((overloadable)) mad(float2 a, float2 b, float2 c) {
491 return a * b + c;
492}
493extern float3 __attribute__((overloadable)) mad(float3 a, float3 b, float3 c) {
494 return a * b + c;
495}
496extern float4 __attribute__((overloadable)) mad(float4 a, float4 b, float4 c) {
497 return a * b + c;
498}
499
500extern float __attribute__((overloadable)) modf(float, float *);
501FN_FUNC_FN_PFN(modf);
502
503extern float __attribute__((overloadable)) nan(uint v) {
504 float f[1];
505 uint32_t *ip = (uint32_t *)f;
506 *ip = v | 0x7fc00000;
507 return f[0];
508}
509
510extern float __attribute__((overloadable)) nextafter(float, float);
511FN_FUNC_FN_FN(nextafter)
512
513FN_FUNC_FN_FN(pow)
514
515extern float __attribute__((overloadable)) pown(float v, int p) {
Jean-Luc Brouillet0b0bcff2014-02-25 19:36:01 -0800[diff] [blame]516 /* The mantissa of a float has fewer bits than an int (24 effective vs. 31).
517 * For very large ints, we'll lose whether the exponent is even or odd, making
Jean-Luc Brouilletbcd5b9a2014-03-07 18:00:57 -0800[diff] [blame^]518 * the selection of a correct sign incorrect. We correct this. Use copysign
519 * to handle the negative zero case.
Jean-Luc Brouillet0b0bcff2014-02-25 19:36:01 -0800[diff] [blame]520 */
Jean-Luc Brouilletbcd5b9a2014-03-07 18:00:57 -0800[diff] [blame^]521 float sign = (p & 0x1) ? copysign(1.f, v) : 1.f;
Jean-Luc Brouillet0b0bcff2014-02-25 19:36:01 -0800[diff] [blame]522 float f = pow(v, (float)p);
523 return copysign(f, sign);
Stephen Hines5a470202013-05-29 15:36:18 -0700[diff] [blame]524}
Jean-Luc Brouillet0b0bcff2014-02-25 19:36:01 -0800[diff] [blame]525FN_FUNC_FN_IN(pown)
Stephen Hines5a470202013-05-29 15:36:18 -0700[diff] [blame]526
527extern float __attribute__((overloadable)) powr(float v, float p) {
528 return pow(v, p);
529}
530extern float2 __attribute__((overloadable)) powr(float2 v, float2 p) {
531 return pow(v, p);
532}
533extern float3 __attribute__((overloadable)) powr(float3 v, float3 p) {
534 return pow(v, p);
535}
536extern float4 __attribute__((overloadable)) powr(float4 v, float4 p) {
537 return pow(v, p);
538}
539
540extern float __attribute__((overloadable)) remainder(float, float);
541FN_FUNC_FN_FN(remainder)
542
543extern float __attribute__((overloadable)) remquo(float, float, int *);
544FN_FUNC_FN_FN_PIN(remquo)
545
546extern float __attribute__((overloadable)) rint(float);
547FN_FUNC_FN(rint)
548
549extern float __attribute__((overloadable)) rootn(float v, int r) {
550 if (r == 0) {
551 return nan(0);
552 }
553
554 if (iszero(v)) {
555 if (r < 0) {
556 if (r & 1) {
557 return copysign(posinf(), v);
558 } else {
559 return posinf();
560 }
561 } else {
562 if (r & 1) {
563 return copysign(0.f, v);
564 } else {
565 return 0.f;
566 }
567 }
568 }
569
570 if (!isinf(v) && !isnan(v) && (v < 0.f)) {
571 if (r & 1) {
572 return (-1.f * pow(-1.f * v, 1.f / r));
573 } else {
574 return nan(0);
575 }
576 }
577
578 return pow(v, 1.f / r);
579}
580FN_FUNC_FN_IN(rootn);
581
582extern float __attribute__((overloadable)) round(float);
583FN_FUNC_FN(round)
584
585
586extern float __attribute__((overloadable)) sqrt(float);
587extern float __attribute__((overloadable)) rsqrt(float v) {
588 return 1.f / sqrt(v);
589}
Stephen Hines146e1382013-08-20 01:17:01 -0700[diff] [blame]590
Stephen Hines4da42502013-10-03 14:57:51 -0700[diff] [blame]591#if !defined(__i386__)
Stephen Hines146e1382013-08-20 01:17:01 -0700[diff] [blame]592FN_FUNC_FN(sqrt)
Stephen Hines4da42502013-10-03 14:57:51 -0700[diff] [blame]593#endif // !defined(__i386__)
Stephen Hines146e1382013-08-20 01:17:01 -0700[diff] [blame]594
Stephen Hines5a470202013-05-29 15:36:18 -0700[diff] [blame]595FN_FUNC_FN(rsqrt)
596
597extern float __attribute__((overloadable)) sin(float);
598FN_FUNC_FN(sin)
599
600extern float __attribute__((overloadable)) sincos(float v, float *cosptr) {
601 *cosptr = cos(v);
602 return sin(v);
603}
604extern float2 __attribute__((overloadable)) sincos(float2 v, float2 *cosptr) {
605 *cosptr = cos(v);
606 return sin(v);
607}
608extern float3 __attribute__((overloadable)) sincos(float3 v, float3 *cosptr) {
609 *cosptr = cos(v);
610 return sin(v);
611}
612extern float4 __attribute__((overloadable)) sincos(float4 v, float4 *cosptr) {
613 *cosptr = cos(v);
614 return sin(v);
615}
616
617extern float __attribute__((overloadable)) sinh(float);
618FN_FUNC_FN(sinh)
619
620extern float __attribute__((overloadable)) sinpi(float v) {
621 return sin(v * M_PI);
622}
623FN_FUNC_FN(sinpi)
624
625extern float __attribute__((overloadable)) tan(float);
626FN_FUNC_FN(tan)
627
628extern float __attribute__((overloadable)) tanh(float);
629FN_FUNC_FN(tanh)
630
631extern float __attribute__((overloadable)) tanpi(float v) {
632 return tan(v * M_PI);
633}
634FN_FUNC_FN(tanpi)
635
636
637extern float __attribute__((overloadable)) tgamma(float);
638FN_FUNC_FN(tgamma)
639
640extern float __attribute__((overloadable)) trunc(float);
641FN_FUNC_FN(trunc)
642
643// Int ops (partial), 6.11.3
644
645#define XN_FUNC_YN(typeout, fnc, typein) \
646extern typeout __attribute__((overloadable)) fnc(typein); \
647extern typeout##2 __attribute__((overloadable)) fnc(typein##2 v) { \
648 typeout##2 r; \
649 r.x = fnc(v.x); \
650 r.y = fnc(v.y); \
651 return r; \
652} \
653extern typeout##3 __attribute__((overloadable)) fnc(typein##3 v) { \
654 typeout##3 r; \
655 r.x = fnc(v.x); \
656 r.y = fnc(v.y); \
657 r.z = fnc(v.z); \
658 return r; \
659} \
660extern typeout##4 __attribute__((overloadable)) fnc(typein##4 v) { \
661 typeout##4 r; \
662 r.x = fnc(v.x); \
663 r.y = fnc(v.y); \
664 r.z = fnc(v.z); \
665 r.w = fnc(v.w); \
666 return r; \
667}
668
669
670#define UIN_FUNC_IN(fnc) \
671XN_FUNC_YN(uchar, fnc, char) \
672XN_FUNC_YN(ushort, fnc, short) \
673XN_FUNC_YN(uint, fnc, int)
674
675#define IN_FUNC_IN(fnc) \
676XN_FUNC_YN(uchar, fnc, uchar) \
677XN_FUNC_YN(char, fnc, char) \
678XN_FUNC_YN(ushort, fnc, ushort) \
679XN_FUNC_YN(short, fnc, short) \
680XN_FUNC_YN(uint, fnc, uint) \
681XN_FUNC_YN(int, fnc, int)
682
683
684#define XN_FUNC_XN_XN_BODY(type, fnc, body) \
685extern type __attribute__((overloadable)) \
686 fnc(type v1, type v2) { \
687 return body; \
688} \
689extern type##2 __attribute__((overloadable)) \
690 fnc(type##2 v1, type##2 v2) { \
691 type##2 r; \
692 r.x = fnc(v1.x, v2.x); \
693 r.y = fnc(v1.y, v2.y); \
694 return r; \
695} \
696extern type##3 __attribute__((overloadable)) \
697 fnc(type##3 v1, type##3 v2) { \
698 type##3 r; \
699 r.x = fnc(v1.x, v2.x); \
700 r.y = fnc(v1.y, v2.y); \
701 r.z = fnc(v1.z, v2.z); \
702 return r; \
703} \
704extern type##4 __attribute__((overloadable)) \
705 fnc(type##4 v1, type##4 v2) { \
706 type##4 r; \
707 r.x = fnc(v1.x, v2.x); \
708 r.y = fnc(v1.y, v2.y); \
709 r.z = fnc(v1.z, v2.z); \
710 r.w = fnc(v1.w, v2.w); \
711 return r; \
712}
713
714#define IN_FUNC_IN_IN_BODY(fnc, body) \
715XN_FUNC_XN_XN_BODY(uchar, fnc, body) \
716XN_FUNC_XN_XN_BODY(char, fnc, body) \
717XN_FUNC_XN_XN_BODY(ushort, fnc, body) \
718XN_FUNC_XN_XN_BODY(short, fnc, body) \
719XN_FUNC_XN_XN_BODY(uint, fnc, body) \
720XN_FUNC_XN_XN_BODY(int, fnc, body) \
721XN_FUNC_XN_XN_BODY(float, fnc, body)
722
723
724/**
725 * abs
726 */
727extern uint32_t __attribute__((overloadable)) abs(int32_t v) {
728 if (v < 0)
729 return -v;
730 return v;
731}
732extern uint16_t __attribute__((overloadable)) abs(int16_t v) {
733 if (v < 0)
734 return -v;
735 return v;
736}
737extern uint8_t __attribute__((overloadable)) abs(int8_t v) {
738 if (v < 0)
739 return -v;
740 return v;
741}
742
743/**
744 * clz
Stephen Hinesc117d8d2014-02-06 17:17:04 -0800[diff] [blame]745 * __builtin_clz only accepts a 32-bit unsigned int, so every input will be
746 * expanded to 32 bits. For our smaller data types, we need to subtract off
747 * these unused top bits (that will be always be composed of zeros).
Stephen Hines5a470202013-05-29 15:36:18 -0700[diff] [blame]748 */
749extern uint32_t __attribute__((overloadable)) clz(uint32_t v) {
750 return __builtin_clz(v);
751}
752extern uint16_t __attribute__((overloadable)) clz(uint16_t v) {
Stephen Hinesc117d8d2014-02-06 17:17:04 -0800[diff] [blame]753 return __builtin_clz(v) - 16;
Stephen Hines5a470202013-05-29 15:36:18 -0700[diff] [blame]754}
755extern uint8_t __attribute__((overloadable)) clz(uint8_t v) {
Stephen Hinesc117d8d2014-02-06 17:17:04 -0800[diff] [blame]756 return __builtin_clz(v) - 24;
Stephen Hines5a470202013-05-29 15:36:18 -0700[diff] [blame]757}
758extern int32_t __attribute__((overloadable)) clz(int32_t v) {
Stephen Hinesc117d8d2014-02-06 17:17:04 -0800[diff] [blame]759 return __builtin_clz(v);
Stephen Hines5a470202013-05-29 15:36:18 -0700[diff] [blame]760}
761extern int16_t __attribute__((overloadable)) clz(int16_t v) {
Stephen Hinesc117d8d2014-02-06 17:17:04 -0800[diff] [blame]762 return __builtin_clz(((uint32_t)v) & 0x0000ffff) - 16;
Stephen Hines5a470202013-05-29 15:36:18 -0700[diff] [blame]763}
764extern int8_t __attribute__((overloadable)) clz(int8_t v) {
Stephen Hinesc117d8d2014-02-06 17:17:04 -0800[diff] [blame]765 return __builtin_clz(((uint32_t)v) & 0x000000ff) - 24;
Stephen Hines5a470202013-05-29 15:36:18 -0700[diff] [blame]766}
767
768
769UIN_FUNC_IN(abs)
770IN_FUNC_IN(clz)
771
772
773// 6.11.4
774
775
776extern float __attribute__((overloadable)) degrees(float radians) {
777 return radians * (180.f / M_PI);
778}
779extern float2 __attribute__((overloadable)) degrees(float2 radians) {
780 return radians * (180.f / M_PI);
781}
782extern float3 __attribute__((overloadable)) degrees(float3 radians) {
783 return radians * (180.f / M_PI);
784}
785extern float4 __attribute__((overloadable)) degrees(float4 radians) {
786 return radians * (180.f / M_PI);
787}
788
789extern float __attribute__((overloadable)) mix(float start, float stop, float amount) {
790 return start + (stop - start) * amount;
791}
792extern float2 __attribute__((overloadable)) mix(float2 start, float2 stop, float2 amount) {
793 return start + (stop - start) * amount;
794}
795extern float3 __attribute__((overloadable)) mix(float3 start, float3 stop, float3 amount) {
796 return start + (stop - start) * amount;
797}
798extern float4 __attribute__((overloadable)) mix(float4 start, float4 stop, float4 amount) {
799 return start + (stop - start) * amount;
800}
801extern float2 __attribute__((overloadable)) mix(float2 start, float2 stop, float amount) {
802 return start + (stop - start) * amount;
803}
804extern float3 __attribute__((overloadable)) mix(float3 start, float3 stop, float amount) {
805 return start + (stop - start) * amount;
806}
807extern float4 __attribute__((overloadable)) mix(float4 start, float4 stop, float amount) {
808 return start + (stop - start) * amount;
809}
810
811extern float __attribute__((overloadable)) radians(float degrees) {
812 return degrees * (M_PI / 180.f);
813}
814extern float2 __attribute__((overloadable)) radians(float2 degrees) {
815 return degrees * (M_PI / 180.f);
816}
817extern float3 __attribute__((overloadable)) radians(float3 degrees) {
818 return degrees * (M_PI / 180.f);
819}
820extern float4 __attribute__((overloadable)) radians(float4 degrees) {
821 return degrees * (M_PI / 180.f);
822}
823
824extern float __attribute__((overloadable)) step(float edge, float v) {
825 return (v < edge) ? 0.f : 1.f;
826}
827extern float2 __attribute__((overloadable)) step(float2 edge, float2 v) {
828 float2 r;
829 r.x = (v.x < edge.x) ? 0.f : 1.f;
830 r.y = (v.y < edge.y) ? 0.f : 1.f;
831 return r;
832}
833extern float3 __attribute__((overloadable)) step(float3 edge, float3 v) {
834 float3 r;
835 r.x = (v.x < edge.x) ? 0.f : 1.f;
836 r.y = (v.y < edge.y) ? 0.f : 1.f;
837 r.z = (v.z < edge.z) ? 0.f : 1.f;
838 return r;
839}
840extern float4 __attribute__((overloadable)) step(float4 edge, float4 v) {
841 float4 r;
842 r.x = (v.x < edge.x) ? 0.f : 1.f;
843 r.y = (v.y < edge.y) ? 0.f : 1.f;
844 r.z = (v.z < edge.z) ? 0.f : 1.f;
845 r.w = (v.w < edge.w) ? 0.f : 1.f;
846 return r;
847}
848extern float2 __attribute__((overloadable)) step(float2 edge, float v) {
849 float2 r;
850 r.x = (v < edge.x) ? 0.f : 1.f;
851 r.y = (v < edge.y) ? 0.f : 1.f;
852 return r;
853}
854extern float3 __attribute__((overloadable)) step(float3 edge, float v) {
855 float3 r;
856 r.x = (v < edge.x) ? 0.f : 1.f;
857 r.y = (v < edge.y) ? 0.f : 1.f;
858 r.z = (v < edge.z) ? 0.f : 1.f;
859 return r;
860}
861extern float4 __attribute__((overloadable)) step(float4 edge, float v) {
862 float4 r;
863 r.x = (v < edge.x) ? 0.f : 1.f;
864 r.y = (v < edge.y) ? 0.f : 1.f;
865 r.z = (v < edge.z) ? 0.f : 1.f;
866 r.w = (v < edge.w) ? 0.f : 1.f;
867 return r;
868}
Jean-Luc Brouillet0ec16352014-01-22 15:23:30 -0800[diff] [blame]869extern float2 __attribute__((overloadable)) step(float edge, float2 v) {
870 float2 r;
871 r.x = (v.x < edge) ? 0.f : 1.f;
872 r.y = (v.y < edge) ? 0.f : 1.f;
873 return r;
874}
875extern float3 __attribute__((overloadable)) step(float edge, float3 v) {
876 float3 r;
877 r.x = (v.x < edge) ? 0.f : 1.f;
878 r.y = (v.y < edge) ? 0.f : 1.f;
879 r.z = (v.z < edge) ? 0.f : 1.f;
880 return r;
881}
882extern float4 __attribute__((overloadable)) step(float edge, float4 v) {
883 float4 r;
884 r.x = (v.x < edge) ? 0.f : 1.f;
885 r.y = (v.y < edge) ? 0.f : 1.f;
886 r.z = (v.z < edge) ? 0.f : 1.f;
887 r.w = (v.w < edge) ? 0.f : 1.f;
888 return r;
889}
Stephen Hines5a470202013-05-29 15:36:18 -0700[diff] [blame]890
891extern float __attribute__((overloadable)) smoothstep(float, float, float);
892extern float2 __attribute__((overloadable)) smoothstep(float2, float2, float2);
893extern float3 __attribute__((overloadable)) smoothstep(float3, float3, float3);
894extern float4 __attribute__((overloadable)) smoothstep(float4, float4, float4);
895extern float2 __attribute__((overloadable)) smoothstep(float, float, float2);
896extern float3 __attribute__((overloadable)) smoothstep(float, float, float3);
897extern float4 __attribute__((overloadable)) smoothstep(float, float, float4);
898
899extern float __attribute__((overloadable)) sign(float v) {
900 if (v > 0) return 1.f;
901 if (v < 0) return -1.f;
902 return v;
903}
904FN_FUNC_FN(sign)
905
906
907// 6.11.5
908extern float3 __attribute__((overloadable)) cross(float3 lhs, float3 rhs) {
909 float3 r;
910 r.x = lhs.y * rhs.z - lhs.z * rhs.y;
911 r.y = lhs.z * rhs.x - lhs.x * rhs.z;
912 r.z = lhs.x * rhs.y - lhs.y * rhs.x;
913 return r;
914}
915
916extern float4 __attribute__((overloadable)) cross(float4 lhs, float4 rhs) {
917 float4 r;
918 r.x = lhs.y * rhs.z - lhs.z * rhs.y;
919 r.y = lhs.z * rhs.x - lhs.x * rhs.z;
920 r.z = lhs.x * rhs.y - lhs.y * rhs.x;
921 r.w = 0.f;
922 return r;
923}
924
Stephen Hines4da42502013-10-03 14:57:51 -0700[diff] [blame]925#if !defined(__i386__)
Stephen Hines146e1382013-08-20 01:17:01 -0700[diff] [blame]926
927extern float __attribute__((overloadable)) dot(float lhs, float rhs) {
928 return lhs * rhs;
929}
930extern float __attribute__((overloadable)) dot(float2 lhs, float2 rhs) {
931 return lhs.x*rhs.x + lhs.y*rhs.y;
932}
933extern float __attribute__((overloadable)) dot(float3 lhs, float3 rhs) {
934 return lhs.x*rhs.x + lhs.y*rhs.y + lhs.z*rhs.z;
935}
936extern float __attribute__((overloadable)) dot(float4 lhs, float4 rhs) {
937 return lhs.x*rhs.x + lhs.y*rhs.y + lhs.z*rhs.z + lhs.w*rhs.w;
938}
939
940extern float __attribute__((overloadable)) length(float v) {
941 return fabs(v);
942}
943extern float __attribute__((overloadable)) length(float2 v) {
944 return sqrt(v.x*v.x + v.y*v.y);
945}
946extern float __attribute__((overloadable)) length(float3 v) {
947 return sqrt(v.x*v.x + v.y*v.y + v.z*v.z);
948}
949extern float __attribute__((overloadable)) length(float4 v) {
950 return sqrt(v.x*v.x + v.y*v.y + v.z*v.z + v.w*v.w);
951}
952
953#else
954
Stephen Hines5a470202013-05-29 15:36:18 -0700[diff] [blame]955extern float __attribute__((overloadable)) length(float v);
956extern float __attribute__((overloadable)) length(float2 v);
957extern float __attribute__((overloadable)) length(float3 v);
958extern float __attribute__((overloadable)) length(float4 v);
959
Stephen Hines146e1382013-08-20 01:17:01 -0700[diff] [blame]960#endif
961
Stephen Hines5a470202013-05-29 15:36:18 -0700[diff] [blame]962extern float __attribute__((overloadable)) distance(float lhs, float rhs) {
963 return length(lhs - rhs);
964}
965extern float __attribute__((overloadable)) distance(float2 lhs, float2 rhs) {
966 return length(lhs - rhs);
967}
968extern float __attribute__((overloadable)) distance(float3 lhs, float3 rhs) {
969 return length(lhs - rhs);
970}
971extern float __attribute__((overloadable)) distance(float4 lhs, float4 rhs) {
972 return length(lhs - rhs);
973}
974
Jean-Luc Brouillet3e0d1e72014-02-24 16:25:42 -0800[diff] [blame]975/* For the normalization functions, vectors of length 0 should simply be
976 * returned (i.e. all the components of that vector are 0).
977 */
Stephen Hines5a470202013-05-29 15:36:18 -0700[diff] [blame]978extern float __attribute__((overloadable)) normalize(float v) {
Jean-Luc Brouillet3e0d1e72014-02-24 16:25:42 -0800[diff] [blame]979 if (v == 0.0f) {
980 return 0.0f;
981 } else if (v < 0.0f) {
982 return -1.0f;
983 } else {
984 return 1.0f;
985 }
Stephen Hines5a470202013-05-29 15:36:18 -0700[diff] [blame]986}
987extern float2 __attribute__((overloadable)) normalize(float2 v) {
Jean-Luc Brouillet3e0d1e72014-02-24 16:25:42 -0800[diff] [blame]988 float l = length(v);
989 return l == 0.0f ? v : v / l;
Stephen Hines5a470202013-05-29 15:36:18 -0700[diff] [blame]990}
991extern float3 __attribute__((overloadable)) normalize(float3 v) {
Jean-Luc Brouillet3e0d1e72014-02-24 16:25:42 -0800[diff] [blame]992 float l = length(v);
993 return l == 0.0f ? v : v / l;
Stephen Hines5a470202013-05-29 15:36:18 -0700[diff] [blame]994}
995extern float4 __attribute__((overloadable)) normalize(float4 v) {
Jean-Luc Brouillet3e0d1e72014-02-24 16:25:42 -0800[diff] [blame]996 float l = length(v);
997 return l == 0.0f ? v : v / l;
Stephen Hines5a470202013-05-29 15:36:18 -0700[diff] [blame]998}
999
1000extern float __attribute__((overloadable)) half_sqrt(float);
1001
1002extern float __attribute__((overloadable)) fast_length(float v) {
Stephen Hines582b3642013-06-01 09:58:24 -0700[diff] [blame]1003 return fabs(v);
Stephen Hines5a470202013-05-29 15:36:18 -0700[diff] [blame]1004}
1005extern float __attribute__((overloadable)) fast_length(float2 v) {
1006 return half_sqrt(v.x*v.x + v.y*v.y);
1007}
1008extern float __attribute__((overloadable)) fast_length(float3 v) {
1009 return half_sqrt(v.x*v.x + v.y*v.y + v.z*v.z);
1010}
1011extern float __attribute__((overloadable)) fast_length(float4 v) {
1012 return half_sqrt(v.x*v.x + v.y*v.y + v.z*v.z + v.w*v.w);
1013}
1014
1015extern float __attribute__((overloadable)) fast_distance(float lhs, float rhs) {
1016 return fast_length(lhs - rhs);
1017}
1018extern float __attribute__((overloadable)) fast_distance(float2 lhs, float2 rhs) {
1019 return fast_length(lhs - rhs);
1020}
1021extern float __attribute__((overloadable)) fast_distance(float3 lhs, float3 rhs) {
1022 return fast_length(lhs - rhs);
1023}
1024extern float __attribute__((overloadable)) fast_distance(float4 lhs, float4 rhs) {
1025 return fast_length(lhs - rhs);
1026}
1027
1028extern float __attribute__((overloadable)) half_rsqrt(float);
1029
Jean-Luc Brouillet3e0d1e72014-02-24 16:25:42 -0800[diff] [blame]1030/* For the normalization functions, vectors of length 0 should simply be
1031 * returned (i.e. all the components of that vector are 0).
1032 */
Stephen Hines5a470202013-05-29 15:36:18 -0700[diff] [blame]1033extern float __attribute__((overloadable)) fast_normalize(float v) {
Jean-Luc Brouillet3e0d1e72014-02-24 16:25:42 -0800[diff] [blame]1034 if (v == 0.0f) {
1035 return 0.0f;
1036 } else if (v < 0.0f) {
1037 return -1.0f;
1038 } else {
1039 return 1.0f;
1040 }
Stephen Hines5a470202013-05-29 15:36:18 -0700[diff] [blame]1041}
Jean-Luc Brouillet3e0d1e72014-02-24 16:25:42 -0800[diff] [blame]1042// If the length is 0, then rlength should be NaN.
Stephen Hines5a470202013-05-29 15:36:18 -0700[diff] [blame]1043extern float2 __attribute__((overloadable)) fast_normalize(float2 v) {
Jean-Luc Brouillet3e0d1e72014-02-24 16:25:42 -0800[diff] [blame]1044 float rlength = half_rsqrt(v.x*v.x + v.y*v.y);
1045 return (rlength == rlength) ? v * rlength : v;
Stephen Hines5a470202013-05-29 15:36:18 -0700[diff] [blame]1046}
1047extern float3 __attribute__((overloadable)) fast_normalize(float3 v) {
Jean-Luc Brouillet3e0d1e72014-02-24 16:25:42 -0800[diff] [blame]1048 float rlength = half_rsqrt(v.x*v.x + v.y*v.y + v.z*v.z);
1049 return (rlength == rlength) ? v * rlength : v;
Stephen Hines5a470202013-05-29 15:36:18 -0700[diff] [blame]1050}
1051extern float4 __attribute__((overloadable)) fast_normalize(float4 v) {
Jean-Luc Brouillet3e0d1e72014-02-24 16:25:42 -0800[diff] [blame]1052 float rlength = half_rsqrt(v.x*v.x + v.y*v.y + v.z*v.z + v.w*v.w);
1053 return (rlength == rlength) ? v * rlength : v;
Stephen Hines5a470202013-05-29 15:36:18 -0700[diff] [blame]1054}
1055
1056extern float __attribute__((overloadable)) half_recip(float);
1057
1058/*
1059extern float __attribute__((overloadable)) approx_atan(float x) {
1060 if (x == 0.f)
1061 return 0.f;
1062 if (x < 0.f)
1063 return -1.f * approx_atan(-1.f * x);
1064 if (x > 1.f)
1065 return M_PI_2 - approx_atan(approx_recip(x));
1066 return x * approx_recip(1.f + 0.28f * x*x);
1067}
1068FN_FUNC_FN(approx_atan)
1069*/
1070
1071typedef union
1072{
1073 float fv;
1074 int32_t iv;
1075} ieee_float_shape_type;
1076
1077/* Get a 32 bit int from a float. */
1078
1079#define GET_FLOAT_WORD(i,d) \
1080do { \
1081 ieee_float_shape_type gf_u; \
1082 gf_u.fv = (d); \
1083 (i) = gf_u.iv; \
1084} while (0)
1085
1086/* Set a float from a 32 bit int. */
1087
1088#define SET_FLOAT_WORD(d,i) \
1089do { \
1090 ieee_float_shape_type sf_u; \
1091 sf_u.iv = (i); \
1092 (d) = sf_u.fv; \
1093} while (0)
1094
1095
1096
1097// Valid -125 to 125
1098extern float __attribute__((overloadable)) native_exp2(float v) {
1099 int32_t iv = (int)v;
1100 int32_t x = iv + (iv >> 31); // ~floor(v)
1101 float r = (v - x);
1102
1103 float fo;
1104 SET_FLOAT_WORD(fo, (x + 127) << 23);
1105
1106 r *= 0.694f; // ~ log(e) / log(2)
1107 float r2 = r*r;
1108 float adj = 1.f + r + (r2 * 0.5f) + (r2*r * 0.166666f) + (r2*r2 * 0.0416666f);
1109 return fo * adj;
1110}
1111
1112extern float2 __attribute__((overloadable)) native_exp2(float2 v) {
1113 int2 iv = convert_int2(v);
1114 int2 x = iv + (iv >> (int2)31);//floor(v);
1115 float2 r = (v - convert_float2(x));
1116
1117 x += 127;
1118
1119 float2 fo = (float2)(x << (int2)23);
1120
1121 r *= 0.694f; // ~ log(e) / log(2)
1122 float2 r2 = r*r;
1123 float2 adj = 1.f + r + (r2 * 0.5f) + (r2*r * 0.166666f) + (r2*r2 * 0.0416666f);
1124 return fo * adj;
1125}
1126
1127extern float4 __attribute__((overloadable)) native_exp2(float4 v) {
1128 int4 iv = convert_int4(v);
1129 int4 x = iv + (iv >> (int4)31);//floor(v);
1130 float4 r = (v - convert_float4(x));
1131
1132 x += 127;
1133
1134 float4 fo = (float4)(x << (int4)23);
1135
1136 r *= 0.694f; // ~ log(e) / log(2)
1137 float4 r2 = r*r;
1138 float4 adj = 1.f + r + (r2 * 0.5f) + (r2*r * 0.166666f) + (r2*r2 * 0.0416666f);
1139 return fo * adj;
1140}
1141
1142extern float3 __attribute__((overloadable)) native_exp2(float3 v) {
1143 float4 t = 1.f;
1144 t.xyz = v;
1145 return native_exp2(t).xyz;
1146}
1147
1148
1149extern float __attribute__((overloadable)) native_exp(float v) {
1150 return native_exp2(v * 1.442695041f);
1151}
1152extern float2 __attribute__((overloadable)) native_exp(float2 v) {
1153 return native_exp2(v * 1.442695041f);
1154}
1155extern float3 __attribute__((overloadable)) native_exp(float3 v) {
1156 return native_exp2(v * 1.442695041f);
1157}
1158extern float4 __attribute__((overloadable)) native_exp(float4 v) {
1159 return native_exp2(v * 1.442695041f);
1160}
1161
1162extern float __attribute__((overloadable)) native_exp10(float v) {
1163 return native_exp2(v * 3.321928095f);
1164}
1165extern float2 __attribute__((overloadable)) native_exp10(float2 v) {
1166 return native_exp2(v * 3.321928095f);
1167}
1168extern float3 __attribute__((overloadable)) native_exp10(float3 v) {
1169 return native_exp2(v * 3.321928095f);
1170}
1171extern float4 __attribute__((overloadable)) native_exp10(float4 v) {
1172 return native_exp2(v * 3.321928095f);
1173}
1174
1175extern float __attribute__((overloadable)) native_log2(float v) {
1176 int32_t ibits;
1177 GET_FLOAT_WORD(ibits, v);
1178
1179 int32_t e = (ibits >> 23) & 0xff;
1180
1181 ibits &= 0x7fffff;
1182 ibits |= 127 << 23;
1183
1184 float ir;
1185 SET_FLOAT_WORD(ir, ibits);
1186
1187 ir -= 1.5f;
1188 float ir2 = ir*ir;
1189 float adj2 = 0.405465108f + // -0.00009f +
1190 (0.666666667f * ir) -
1191 (0.222222222f * ir2) +
1192 (0.098765432f * ir*ir2) -
1193 (0.049382716f * ir2*ir2) +
1194 (0.026337449f * ir*ir2*ir2) -
1195 (0.014631916f * ir2*ir2*ir2);
1196 adj2 *= (1.f / 0.693147181f);
1197
1198 return (float)(e - 127) + adj2;
1199}
1200extern float2 __attribute__((overloadable)) native_log2(float2 v) {
1201 float2 v2 = {native_log2(v.x), native_log2(v.y)};
1202 return v2;
1203}
1204extern float3 __attribute__((overloadable)) native_log2(float3 v) {
1205 float3 v2 = {native_log2(v.x), native_log2(v.y), native_log2(v.z)};
1206 return v2;
1207}
1208extern float4 __attribute__((overloadable)) native_log2(float4 v) {
1209 float4 v2 = {native_log2(v.x), native_log2(v.y), native_log2(v.z), native_log2(v.w)};
1210 return v2;
1211}
1212
1213extern float __attribute__((overloadable)) native_log(float v) {
1214 return native_log2(v) * (1.f / 1.442695041f);
1215}
1216extern float2 __attribute__((overloadable)) native_log(float2 v) {
1217 return native_log2(v) * (1.f / 1.442695041f);
1218}
1219extern float3 __attribute__((overloadable)) native_log(float3 v) {
1220 return native_log2(v) * (1.f / 1.442695041f);
1221}
1222extern float4 __attribute__((overloadable)) native_log(float4 v) {
1223 return native_log2(v) * (1.f / 1.442695041f);
1224}
1225
1226extern float __attribute__((overloadable)) native_log10(float v) {
1227 return native_log2(v) * (1.f / 3.321928095f);
1228}
1229extern float2 __attribute__((overloadable)) native_log10(float2 v) {
1230 return native_log2(v) * (1.f / 3.321928095f);
1231}
1232extern float3 __attribute__((overloadable)) native_log10(float3 v) {
1233 return native_log2(v) * (1.f / 3.321928095f);
1234}
1235extern float4 __attribute__((overloadable)) native_log10(float4 v) {
1236 return native_log2(v) * (1.f / 3.321928095f);
1237}
1238
1239
1240extern float __attribute__((overloadable)) native_powr(float v, float y) {
1241 float v2 = native_log2(v);
1242 v2 = fmax(v2, -125.f);
1243 return native_exp2(v2 * y);
1244}
1245extern float2 __attribute__((overloadable)) native_powr(float2 v, float2 y) {
1246 float2 v2 = native_log2(v);
1247 v2 = fmax(v2, -125.f);
1248 return native_exp2(v2 * y);
1249}
1250extern float3 __attribute__((overloadable)) native_powr(float3 v, float3 y) {
1251 float3 v2 = native_log2(v);
1252 v2 = fmax(v2, -125.f);
1253 return native_exp2(v2 * y);
1254}
1255extern float4 __attribute__((overloadable)) native_powr(float4 v, float4 y) {
1256 float4 v2 = native_log2(v);
1257 v2 = fmax(v2, -125.f);
1258 return native_exp2(v2 * y);
1259}
1260
1261
1262#undef FN_FUNC_FN
1263#undef IN_FUNC_FN
1264#undef FN_FUNC_FN_FN
1265#undef FN_FUNC_FN_F
1266#undef FN_FUNC_FN_IN
1267#undef FN_FUNC_FN_I
1268#undef FN_FUNC_FN_PFN
1269#undef FN_FUNC_FN_PIN
1270#undef FN_FUNC_FN_FN_FN
1271#undef FN_FUNC_FN_FN_PIN
1272#undef XN_FUNC_YN
1273#undef UIN_FUNC_IN
1274#undef IN_FUNC_IN
1275#undef XN_FUNC_XN_XN_BODY
1276#undef IN_FUNC_IN_IN_BODY