rs_cl.c revision 5a47020542c52af3e879c1cd67674ca979ff0a18
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) {
516    return pow(v, (float)p);
517}
518extern float2 __attribute__((overloadable)) pown(float2 v, int2 p) {
519    float2 f2 = convert_float2(p);
520    return pow(v, f2);
521}
522extern float3 __attribute__((overloadable)) pown(float3 v, int3 p) {
523    float3 f3 = convert_float3(p);
524    return pow(v, f3);
525}
526extern float4 __attribute__((overloadable)) pown(float4 v, int4 p) {
527    float4 f4 = convert_float4(p);
528    return pow(v, f4);
529}
530
531extern float __attribute__((overloadable)) powr(float v, float p) {
532    return pow(v, p);
533}
534extern float2 __attribute__((overloadable)) powr(float2 v, float2 p) {
535    return pow(v, p);
536}
537extern float3 __attribute__((overloadable)) powr(float3 v, float3 p) {
538    return pow(v, p);
539}
540extern float4 __attribute__((overloadable)) powr(float4 v, float4 p) {
541    return pow(v, p);
542}
543
544extern float __attribute__((overloadable)) remainder(float, float);
545FN_FUNC_FN_FN(remainder)
546
547extern float __attribute__((overloadable)) remquo(float, float, int *);
548FN_FUNC_FN_FN_PIN(remquo)
549
550extern float __attribute__((overloadable)) rint(float);
551FN_FUNC_FN(rint)
552
553extern float __attribute__((overloadable)) rootn(float v, int r) {
554    if (r == 0) {
555        return nan(0);
556    }
557
558    if (iszero(v)) {
559        if (r < 0) {
560            if (r & 1) {
561                return copysign(posinf(), v);
562            } else {
563                return posinf();
564            }
565        } else {
566            if (r & 1) {
567                return copysign(0.f, v);
568            } else {
569                return 0.f;
570            }
571        }
572    }
573
574    if (!isinf(v) && !isnan(v) && (v < 0.f)) {
575        if (r & 1) {
576            return (-1.f * pow(-1.f * v, 1.f / r));
577        } else {
578            return nan(0);
579        }
580    }
581
582    return pow(v, 1.f / r);
583}
584FN_FUNC_FN_IN(rootn);
585
586extern float __attribute__((overloadable)) round(float);
587FN_FUNC_FN(round)
588
589
590extern float __attribute__((overloadable)) sqrt(float);
591extern float __attribute__((overloadable)) rsqrt(float v) {
592    return 1.f / sqrt(v);
593}
594FN_FUNC_FN(rsqrt)
595
596extern float __attribute__((overloadable)) sin(float);
597FN_FUNC_FN(sin)
598
599extern float __attribute__((overloadable)) sincos(float v, float *cosptr) {
600    *cosptr = cos(v);
601    return sin(v);
602}
603extern float2 __attribute__((overloadable)) sincos(float2 v, float2 *cosptr) {
604    *cosptr = cos(v);
605    return sin(v);
606}
607extern float3 __attribute__((overloadable)) sincos(float3 v, float3 *cosptr) {
608    *cosptr = cos(v);
609    return sin(v);
610}
611extern float4 __attribute__((overloadable)) sincos(float4 v, float4 *cosptr) {
612    *cosptr = cos(v);
613    return sin(v);
614}
615
616extern float __attribute__((overloadable)) sinh(float);
617FN_FUNC_FN(sinh)
618
619extern float __attribute__((overloadable)) sinpi(float v) {
620    return sin(v * M_PI);
621}
622FN_FUNC_FN(sinpi)
623
624extern float __attribute__((overloadable)) tan(float);
625FN_FUNC_FN(tan)
626
627extern float __attribute__((overloadable)) tanh(float);
628FN_FUNC_FN(tanh)
629
630extern float __attribute__((overloadable)) tanpi(float v) {
631    return tan(v * M_PI);
632}
633FN_FUNC_FN(tanpi)
634
635
636extern float __attribute__((overloadable)) tgamma(float);
637FN_FUNC_FN(tgamma)
638
639extern float __attribute__((overloadable)) trunc(float);
640FN_FUNC_FN(trunc)
641
642// Int ops (partial), 6.11.3
643
644#define XN_FUNC_YN(typeout, fnc, typein)                                \
645extern typeout __attribute__((overloadable)) fnc(typein);               \
646extern typeout##2 __attribute__((overloadable)) fnc(typein##2 v) {  \
647    typeout##2 r;                                                       \
648    r.x = fnc(v.x);                                                     \
649    r.y = fnc(v.y);                                                     \
650    return r;                                                           \
651}                                                                       \
652extern typeout##3 __attribute__((overloadable)) fnc(typein##3 v) {  \
653    typeout##3 r;                                                       \
654    r.x = fnc(v.x);                                                     \
655    r.y = fnc(v.y);                                                     \
656    r.z = fnc(v.z);                                                     \
657    return r;                                                           \
658}                                                                       \
659extern typeout##4 __attribute__((overloadable)) fnc(typein##4 v) {  \
660    typeout##4 r;                                                       \
661    r.x = fnc(v.x);                                                     \
662    r.y = fnc(v.y);                                                     \
663    r.z = fnc(v.z);                                                     \
664    r.w = fnc(v.w);                                                     \
665    return r;                                                           \
666}
667
668
669#define UIN_FUNC_IN(fnc)          \
670XN_FUNC_YN(uchar, fnc, char)      \
671XN_FUNC_YN(ushort, fnc, short)    \
672XN_FUNC_YN(uint, fnc, int)
673
674#define IN_FUNC_IN(fnc)           \
675XN_FUNC_YN(uchar, fnc, uchar)     \
676XN_FUNC_YN(char, fnc, char)       \
677XN_FUNC_YN(ushort, fnc, ushort)   \
678XN_FUNC_YN(short, fnc, short)     \
679XN_FUNC_YN(uint, fnc, uint)       \
680XN_FUNC_YN(int, fnc, int)
681
682
683#define XN_FUNC_XN_XN_BODY(type, fnc, body)         \
684extern type __attribute__((overloadable))       \
685        fnc(type v1, type v2) {                     \
686    return body;                                    \
687}                                                   \
688extern type##2 __attribute__((overloadable))    \
689        fnc(type##2 v1, type##2 v2) {               \
690    type##2 r;                                      \
691    r.x = fnc(v1.x, v2.x);                          \
692    r.y = fnc(v1.y, v2.y);                          \
693    return r;                                       \
694}                                                   \
695extern type##3 __attribute__((overloadable))    \
696        fnc(type##3 v1, type##3 v2) {               \
697    type##3 r;                                      \
698    r.x = fnc(v1.x, v2.x);                          \
699    r.y = fnc(v1.y, v2.y);                          \
700    r.z = fnc(v1.z, v2.z);                          \
701    return r;                                       \
702}                                                   \
703extern type##4 __attribute__((overloadable))    \
704        fnc(type##4 v1, type##4 v2) {               \
705    type##4 r;                                      \
706    r.x = fnc(v1.x, v2.x);                          \
707    r.y = fnc(v1.y, v2.y);                          \
708    r.z = fnc(v1.z, v2.z);                          \
709    r.w = fnc(v1.w, v2.w);                          \
710    return r;                                       \
711}
712
713#define IN_FUNC_IN_IN_BODY(fnc, body) \
714XN_FUNC_XN_XN_BODY(uchar, fnc, body)  \
715XN_FUNC_XN_XN_BODY(char, fnc, body)   \
716XN_FUNC_XN_XN_BODY(ushort, fnc, body) \
717XN_FUNC_XN_XN_BODY(short, fnc, body)  \
718XN_FUNC_XN_XN_BODY(uint, fnc, body)   \
719XN_FUNC_XN_XN_BODY(int, fnc, body)    \
720XN_FUNC_XN_XN_BODY(float, fnc, body)
721
722
723/**
724 * abs
725 */
726extern uint32_t __attribute__((overloadable)) abs(int32_t v) {
727    if (v < 0)
728        return -v;
729    return v;
730}
731extern uint16_t __attribute__((overloadable)) abs(int16_t v) {
732    if (v < 0)
733        return -v;
734    return v;
735}
736extern uint8_t __attribute__((overloadable)) abs(int8_t v) {
737    if (v < 0)
738        return -v;
739    return v;
740}
741
742/**
743 * clz
744 */
745extern uint32_t __attribute__((overloadable)) clz(uint32_t v) {
746    return __builtin_clz(v);
747}
748extern uint16_t __attribute__((overloadable)) clz(uint16_t v) {
749    return (uint16_t)__builtin_clz(v);
750}
751extern uint8_t __attribute__((overloadable)) clz(uint8_t v) {
752    return (uint8_t)__builtin_clz(v);
753}
754extern int32_t __attribute__((overloadable)) clz(int32_t v) {
755    return (int32_t)__builtin_clz((uint32_t)v);
756}
757extern int16_t __attribute__((overloadable)) clz(int16_t v) {
758    return (int16_t)__builtin_clz(v);
759}
760extern int8_t __attribute__((overloadable)) clz(int8_t v) {
761    return (int8_t)__builtin_clz(v);
762}
763
764
765UIN_FUNC_IN(abs)
766IN_FUNC_IN(clz)
767
768
769// 6.11.4
770
771
772extern float __attribute__((overloadable)) degrees(float radians) {
773    return radians * (180.f / M_PI);
774}
775extern float2 __attribute__((overloadable)) degrees(float2 radians) {
776    return radians * (180.f / M_PI);
777}
778extern float3 __attribute__((overloadable)) degrees(float3 radians) {
779    return radians * (180.f / M_PI);
780}
781extern float4 __attribute__((overloadable)) degrees(float4 radians) {
782    return radians * (180.f / M_PI);
783}
784
785extern float __attribute__((overloadable)) mix(float start, float stop, float amount) {
786    return start + (stop - start) * amount;
787}
788extern float2 __attribute__((overloadable)) mix(float2 start, float2 stop, float2 amount) {
789    return start + (stop - start) * amount;
790}
791extern float3 __attribute__((overloadable)) mix(float3 start, float3 stop, float3 amount) {
792    return start + (stop - start) * amount;
793}
794extern float4 __attribute__((overloadable)) mix(float4 start, float4 stop, float4 amount) {
795    return start + (stop - start) * amount;
796}
797extern float2 __attribute__((overloadable)) mix(float2 start, float2 stop, float amount) {
798    return start + (stop - start) * amount;
799}
800extern float3 __attribute__((overloadable)) mix(float3 start, float3 stop, float amount) {
801    return start + (stop - start) * amount;
802}
803extern float4 __attribute__((overloadable)) mix(float4 start, float4 stop, float amount) {
804    return start + (stop - start) * amount;
805}
806
807extern float __attribute__((overloadable)) radians(float degrees) {
808    return degrees * (M_PI / 180.f);
809}
810extern float2 __attribute__((overloadable)) radians(float2 degrees) {
811    return degrees * (M_PI / 180.f);
812}
813extern float3 __attribute__((overloadable)) radians(float3 degrees) {
814    return degrees * (M_PI / 180.f);
815}
816extern float4 __attribute__((overloadable)) radians(float4 degrees) {
817    return degrees * (M_PI / 180.f);
818}
819
820extern float __attribute__((overloadable)) step(float edge, float v) {
821    return (v < edge) ? 0.f : 1.f;
822}
823extern float2 __attribute__((overloadable)) step(float2 edge, float2 v) {
824    float2 r;
825    r.x = (v.x < edge.x) ? 0.f : 1.f;
826    r.y = (v.y < edge.y) ? 0.f : 1.f;
827    return r;
828}
829extern float3 __attribute__((overloadable)) step(float3 edge, float3 v) {
830    float3 r;
831    r.x = (v.x < edge.x) ? 0.f : 1.f;
832    r.y = (v.y < edge.y) ? 0.f : 1.f;
833    r.z = (v.z < edge.z) ? 0.f : 1.f;
834    return r;
835}
836extern float4 __attribute__((overloadable)) step(float4 edge, float4 v) {
837    float4 r;
838    r.x = (v.x < edge.x) ? 0.f : 1.f;
839    r.y = (v.y < edge.y) ? 0.f : 1.f;
840    r.z = (v.z < edge.z) ? 0.f : 1.f;
841    r.w = (v.w < edge.w) ? 0.f : 1.f;
842    return r;
843}
844extern float2 __attribute__((overloadable)) step(float2 edge, float v) {
845    float2 r;
846    r.x = (v < edge.x) ? 0.f : 1.f;
847    r.y = (v < edge.y) ? 0.f : 1.f;
848    return r;
849}
850extern float3 __attribute__((overloadable)) step(float3 edge, float v) {
851    float3 r;
852    r.x = (v < edge.x) ? 0.f : 1.f;
853    r.y = (v < edge.y) ? 0.f : 1.f;
854    r.z = (v < edge.z) ? 0.f : 1.f;
855    return r;
856}
857extern float4 __attribute__((overloadable)) step(float4 edge, float v) {
858    float4 r;
859    r.x = (v < edge.x) ? 0.f : 1.f;
860    r.y = (v < edge.y) ? 0.f : 1.f;
861    r.z = (v < edge.z) ? 0.f : 1.f;
862    r.w = (v < edge.w) ? 0.f : 1.f;
863    return r;
864}
865
866extern float __attribute__((overloadable)) smoothstep(float, float, float);
867extern float2 __attribute__((overloadable)) smoothstep(float2, float2, float2);
868extern float3 __attribute__((overloadable)) smoothstep(float3, float3, float3);
869extern float4 __attribute__((overloadable)) smoothstep(float4, float4, float4);
870extern float2 __attribute__((overloadable)) smoothstep(float, float, float2);
871extern float3 __attribute__((overloadable)) smoothstep(float, float, float3);
872extern float4 __attribute__((overloadable)) smoothstep(float, float, float4);
873
874extern float __attribute__((overloadable)) sign(float v) {
875    if (v > 0) return 1.f;
876    if (v < 0) return -1.f;
877    return v;
878}
879FN_FUNC_FN(sign)
880
881
882// 6.11.5
883extern float3 __attribute__((overloadable)) cross(float3 lhs, float3 rhs) {
884    float3 r;
885    r.x = lhs.y * rhs.z  - lhs.z * rhs.y;
886    r.y = lhs.z * rhs.x  - lhs.x * rhs.z;
887    r.z = lhs.x * rhs.y  - lhs.y * rhs.x;
888    return r;
889}
890
891extern float4 __attribute__((overloadable)) cross(float4 lhs, float4 rhs) {
892    float4 r;
893    r.x = lhs.y * rhs.z  - lhs.z * rhs.y;
894    r.y = lhs.z * rhs.x  - lhs.x * rhs.z;
895    r.z = lhs.x * rhs.y  - lhs.y * rhs.x;
896    r.w = 0.f;
897    return r;
898}
899
900extern float __attribute__((overloadable)) length(float v);
901extern float __attribute__((overloadable)) length(float2 v);
902extern float __attribute__((overloadable)) length(float3 v);
903extern float __attribute__((overloadable)) length(float4 v);
904
905extern float __attribute__((overloadable)) distance(float lhs, float rhs) {
906    return length(lhs - rhs);
907}
908extern float __attribute__((overloadable)) distance(float2 lhs, float2 rhs) {
909    return length(lhs - rhs);
910}
911extern float __attribute__((overloadable)) distance(float3 lhs, float3 rhs) {
912    return length(lhs - rhs);
913}
914extern float __attribute__((overloadable)) distance(float4 lhs, float4 rhs) {
915    return length(lhs - rhs);
916}
917
918extern float __attribute__((overloadable)) normalize(float v) {
919    return 1.f;
920}
921extern float2 __attribute__((overloadable)) normalize(float2 v) {
922    return v / length(v);
923}
924extern float3 __attribute__((overloadable)) normalize(float3 v) {
925    return v / length(v);
926}
927extern float4 __attribute__((overloadable)) normalize(float4 v) {
928    return v / length(v);
929}
930
931extern float __attribute__((overloadable)) half_sqrt(float);
932
933extern float __attribute__((overloadable)) fast_length(float v) {
934    return v;
935}
936extern float __attribute__((overloadable)) fast_length(float2 v) {
937    return half_sqrt(v.x*v.x + v.y*v.y);
938}
939extern float __attribute__((overloadable)) fast_length(float3 v) {
940    return half_sqrt(v.x*v.x + v.y*v.y + v.z*v.z);
941}
942extern float __attribute__((overloadable)) fast_length(float4 v) {
943    return half_sqrt(v.x*v.x + v.y*v.y + v.z*v.z + v.w*v.w);
944}
945
946extern float __attribute__((overloadable)) fast_distance(float lhs, float rhs) {
947    return fast_length(lhs - rhs);
948}
949extern float __attribute__((overloadable)) fast_distance(float2 lhs, float2 rhs) {
950    return fast_length(lhs - rhs);
951}
952extern float __attribute__((overloadable)) fast_distance(float3 lhs, float3 rhs) {
953    return fast_length(lhs - rhs);
954}
955extern float __attribute__((overloadable)) fast_distance(float4 lhs, float4 rhs) {
956    return fast_length(lhs - rhs);
957}
958
959extern float __attribute__((overloadable)) half_rsqrt(float);
960
961extern float __attribute__((overloadable)) fast_normalize(float v) {
962    return 1.f;
963}
964extern float2 __attribute__((overloadable)) fast_normalize(float2 v) {
965    return v * half_rsqrt(v.x*v.x + v.y*v.y);
966}
967extern float3 __attribute__((overloadable)) fast_normalize(float3 v) {
968    return v * half_rsqrt(v.x*v.x + v.y*v.y + v.z*v.z);
969}
970extern float4 __attribute__((overloadable)) fast_normalize(float4 v) {
971    return v * half_rsqrt(v.x*v.x + v.y*v.y + v.z*v.z + v.w*v.w);
972}
973
974extern float __attribute__((overloadable)) half_recip(float);
975
976/*
977extern float __attribute__((overloadable)) approx_atan(float x) {
978    if (x == 0.f)
979        return 0.f;
980    if (x < 0.f)
981        return -1.f * approx_atan(-1.f * x);
982    if (x > 1.f)
983        return M_PI_2 - approx_atan(approx_recip(x));
984    return x * approx_recip(1.f + 0.28f * x*x);
985}
986FN_FUNC_FN(approx_atan)
987*/
988
989typedef union
990{
991  float fv;
992  int32_t iv;
993} ieee_float_shape_type;
994
995/* Get a 32 bit int from a float.  */
996
997#define GET_FLOAT_WORD(i,d)                 \
998do {                                \
999  ieee_float_shape_type gf_u;                   \
1000  gf_u.fv = (d);                     \
1001  (i) = gf_u.iv;                      \
1002} while (0)
1003
1004/* Set a float from a 32 bit int.  */
1005
1006#define SET_FLOAT_WORD(d,i)                 \
1007do {                                \
1008  ieee_float_shape_type sf_u;                   \
1009  sf_u.iv = (i);                      \
1010  (d) = sf_u.fv;                     \
1011} while (0)
1012
1013
1014
1015// Valid -125 to 125
1016extern float __attribute__((overloadable)) native_exp2(float v) {
1017    int32_t iv = (int)v;
1018    int32_t x = iv + (iv >> 31); // ~floor(v)
1019    float r = (v - x);
1020
1021    float fo;
1022    SET_FLOAT_WORD(fo, (x + 127) << 23);
1023
1024    r *= 0.694f; // ~ log(e) / log(2)
1025    float r2 = r*r;
1026    float adj = 1.f + r + (r2 * 0.5f) + (r2*r * 0.166666f) + (r2*r2 * 0.0416666f);
1027    return fo * adj;
1028}
1029
1030extern float2 __attribute__((overloadable)) native_exp2(float2 v) {
1031    int2 iv = convert_int2(v);
1032    int2 x = iv + (iv >> (int2)31);//floor(v);
1033    float2 r = (v - convert_float2(x));
1034
1035    x += 127;
1036
1037    float2 fo = (float2)(x << (int2)23);
1038
1039    r *= 0.694f; // ~ log(e) / log(2)
1040    float2 r2 = r*r;
1041    float2 adj = 1.f + r + (r2 * 0.5f) + (r2*r * 0.166666f) + (r2*r2 * 0.0416666f);
1042    return fo * adj;
1043}
1044
1045extern float4 __attribute__((overloadable)) native_exp2(float4 v) {
1046    int4 iv = convert_int4(v);
1047    int4 x = iv + (iv >> (int4)31);//floor(v);
1048    float4 r = (v - convert_float4(x));
1049
1050    x += 127;
1051
1052    float4 fo = (float4)(x << (int4)23);
1053
1054    r *= 0.694f; // ~ log(e) / log(2)
1055    float4 r2 = r*r;
1056    float4 adj = 1.f + r + (r2 * 0.5f) + (r2*r * 0.166666f) + (r2*r2 * 0.0416666f);
1057    return fo * adj;
1058}
1059
1060extern float3 __attribute__((overloadable)) native_exp2(float3 v) {
1061    float4 t = 1.f;
1062    t.xyz = v;
1063    return native_exp2(t).xyz;
1064}
1065
1066
1067extern float __attribute__((overloadable)) native_exp(float v) {
1068    return native_exp2(v * 1.442695041f);
1069}
1070extern float2 __attribute__((overloadable)) native_exp(float2 v) {
1071    return native_exp2(v * 1.442695041f);
1072}
1073extern float3 __attribute__((overloadable)) native_exp(float3 v) {
1074    return native_exp2(v * 1.442695041f);
1075}
1076extern float4 __attribute__((overloadable)) native_exp(float4 v) {
1077    return native_exp2(v * 1.442695041f);
1078}
1079
1080extern float __attribute__((overloadable)) native_exp10(float v) {
1081    return native_exp2(v * 3.321928095f);
1082}
1083extern float2 __attribute__((overloadable)) native_exp10(float2 v) {
1084    return native_exp2(v * 3.321928095f);
1085}
1086extern float3 __attribute__((overloadable)) native_exp10(float3 v) {
1087    return native_exp2(v * 3.321928095f);
1088}
1089extern float4 __attribute__((overloadable)) native_exp10(float4 v) {
1090    return native_exp2(v * 3.321928095f);
1091}
1092
1093extern float __attribute__((overloadable)) native_log2(float v) {
1094    int32_t ibits;
1095    GET_FLOAT_WORD(ibits, v);
1096
1097    int32_t e = (ibits >> 23) & 0xff;
1098
1099    ibits &= 0x7fffff;
1100    ibits |= 127 << 23;
1101
1102    float ir;
1103    SET_FLOAT_WORD(ir, ibits);
1104
1105    ir -= 1.5f;
1106    float ir2 = ir*ir;
1107    float adj2 = 0.405465108f + // -0.00009f +
1108                 (0.666666667f * ir) -
1109                 (0.222222222f * ir2) +
1110                 (0.098765432f * ir*ir2) -
1111                 (0.049382716f * ir2*ir2) +
1112                 (0.026337449f * ir*ir2*ir2) -
1113                 (0.014631916f * ir2*ir2*ir2);
1114    adj2 *= (1.f / 0.693147181f);
1115
1116    return (float)(e - 127) + adj2;
1117}
1118extern float2 __attribute__((overloadable)) native_log2(float2 v) {
1119    float2 v2 = {native_log2(v.x), native_log2(v.y)};
1120    return v2;
1121}
1122extern float3 __attribute__((overloadable)) native_log2(float3 v) {
1123    float3 v2 = {native_log2(v.x), native_log2(v.y), native_log2(v.z)};
1124    return v2;
1125}
1126extern float4 __attribute__((overloadable)) native_log2(float4 v) {
1127    float4 v2 = {native_log2(v.x), native_log2(v.y), native_log2(v.z), native_log2(v.w)};
1128    return v2;
1129}
1130
1131extern float __attribute__((overloadable)) native_log(float v) {
1132    return native_log2(v) * (1.f / 1.442695041f);
1133}
1134extern float2 __attribute__((overloadable)) native_log(float2 v) {
1135    return native_log2(v) * (1.f / 1.442695041f);
1136}
1137extern float3 __attribute__((overloadable)) native_log(float3 v) {
1138    return native_log2(v) * (1.f / 1.442695041f);
1139}
1140extern float4 __attribute__((overloadable)) native_log(float4 v) {
1141    return native_log2(v) * (1.f / 1.442695041f);
1142}
1143
1144extern float __attribute__((overloadable)) native_log10(float v) {
1145    return native_log2(v) * (1.f / 3.321928095f);
1146}
1147extern float2 __attribute__((overloadable)) native_log10(float2 v) {
1148    return native_log2(v) * (1.f / 3.321928095f);
1149}
1150extern float3 __attribute__((overloadable)) native_log10(float3 v) {
1151    return native_log2(v) * (1.f / 3.321928095f);
1152}
1153extern float4 __attribute__((overloadable)) native_log10(float4 v) {
1154    return native_log2(v) * (1.f / 3.321928095f);
1155}
1156
1157
1158extern float __attribute__((overloadable)) native_powr(float v, float y) {
1159    float v2 = native_log2(v);
1160    v2 = fmax(v2, -125.f);
1161    return native_exp2(v2 * y);
1162}
1163extern float2 __attribute__((overloadable)) native_powr(float2 v, float2 y) {
1164    float2 v2 = native_log2(v);
1165    v2 = fmax(v2, -125.f);
1166    return native_exp2(v2 * y);
1167}
1168extern float3 __attribute__((overloadable)) native_powr(float3 v, float3 y) {
1169    float3 v2 = native_log2(v);
1170    v2 = fmax(v2, -125.f);
1171    return native_exp2(v2 * y);
1172}
1173extern float4 __attribute__((overloadable)) native_powr(float4 v, float4 y) {
1174    float4 v2 = native_log2(v);
1175    v2 = fmax(v2, -125.f);
1176    return native_exp2(v2 * y);
1177}
1178
1179
1180#undef FN_FUNC_FN
1181#undef IN_FUNC_FN
1182#undef FN_FUNC_FN_FN
1183#undef FN_FUNC_FN_F
1184#undef FN_FUNC_FN_IN
1185#undef FN_FUNC_FN_I
1186#undef FN_FUNC_FN_PFN
1187#undef FN_FUNC_FN_PIN
1188#undef FN_FUNC_FN_FN_FN
1189#undef FN_FUNC_FN_FN_PIN
1190#undef XN_FUNC_YN
1191#undef UIN_FUNC_IN
1192#undef IN_FUNC_IN
1193#undef XN_FUNC_XN_XN_BODY
1194#undef IN_FUNC_IN_IN_BODY
1195