summaryrefslogtreecommitdiffstats
path: root/vec_sse_float1.h
blob: 9cee891906c9eb7a36cdf71cea53a8f07b38ede3 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
// -*-C++-*-

#ifndef VEC_SSE_FLOAT1_H
#define VEC_SSE_FLOAT1_H

#include "floatprops.h"
#include "mathfuncs.h"
#include "vec_base.h"

#include <cmath>
#include <climits>

// SSE2 intrinsics
#include <emmintrin.h>
#ifdef __SSE3__                 // Intel's SSE 3
#  include <pmmintrin.h>
#endif
#ifdef __SSE4_1__               // Intel's SSE 4.1
#  include <smmintrin.h>
#endif
#ifdef __SSE4A__                // AMD's SSE 4a
#  include <ammintrin.h>
#endif
#if defined __AVX__             // Intel's AVX
#  include <immintrin.h>
#endif



namespace vecmathlib {
  
#define VECMATHLIB_HAVE_VEC_FLOAT_1
  template<> struct boolvec<float,1>;
  template<> struct intvec<float,1>;
  template<> struct realvec<float,1>;
  
  
  
  template<>
  struct boolvec<float,1>: floatprops<float>
  {
    static int const size = 1;
    typedef bool scalar_t;
    typedef uint_t bvector_t;
    static int const alignment = sizeof(bvector_t);
    
    static_assert(size * sizeof(real_t) == sizeof(bvector_t),
                  "vector size is wrong");
    
    // true values are non-zero, false values are zero
    
    typedef boolvec boolvec_t;
    typedef intvec<real_t, size> intvec_t;
    typedef realvec<real_t, size> realvec_t;
    
    // Short names for type casts
    typedef real_t R;
    typedef int_t I;
    typedef uint_t U;
    typedef realvec_t RV;
    typedef intvec_t IV;
    typedef boolvec_t BV;
    typedef floatprops<real_t> FP;
    typedef mathfuncs<realvec_t> MF;
    
    
    
    bvector_t v;
    
    boolvec() {}
    // Can't have a non-trivial copy constructor; if so, objects won't
    // be passed in registers
    // boolvec(boolvec const& x): v(x.v) {}
    // boolvec& operator=(boolvec const& x) { return v=x.v, *this; }
    boolvec(bvector_t x): v(x) {}
    boolvec(bool a): v(a) {}
    boolvec(bool const* as): v(as[0]) {}
    
    operator bvector_t() const { return v; }
    bool operator[](int n) const { return v; }
    boolvec_t& set_elt(int n, bool a) { return v=a, *this; }
    
    
    
    intvec_t as_int() const;      // defined after intvec
    intvec_t convert_int() const; // defined after intvec
    
    
    
    boolvec_t operator!() const { return !v; }
    
    boolvec_t operator&&(boolvec_t x) const { return v && x.v; }
    boolvec_t operator||(boolvec_t x) const { return v || x.v; }
    boolvec_t operator==(boolvec_t x) const { return bool(v) == bool(x.v); }
    boolvec_t operator!=(boolvec_t x) const { return bool(v) != bool(x.v); }
    
    bool all() const { return *this; }
    bool any() const { return *this; }
    
    
    
    // ifthen(condition, then-value, else-value)
    boolvec_t ifthen(boolvec_t x, boolvec_t y) const;
    intvec_t ifthen(intvec_t x, intvec_t y) const; // defined after intvec
    realvec_t ifthen(realvec_t x, realvec_t y) const; // defined after realvec
  };
  
  
  
  template<>
  struct intvec<float,1>: floatprops<float>
  {
    static int const size = 1;
    typedef int_t scalar_t;
    typedef int_t ivector_t;
    static int const alignment = sizeof(ivector_t);
    
    static_assert(size * sizeof(real_t) == sizeof(ivector_t),
                  "vector size is wrong");
    
    typedef boolvec<real_t, size> boolvec_t;
    typedef intvec intvec_t;
    typedef realvec<real_t, size> realvec_t;
    
    // Short names for type casts
    typedef real_t R;
    typedef int_t I;
    typedef uint_t U;
    typedef realvec_t RV;
    typedef intvec_t IV;
    typedef boolvec_t BV;
    typedef floatprops<real_t> FP;
    typedef mathfuncs<realvec_t> MF;
    
    
    
    ivector_t v;
    
    intvec() {}
    // Can't have a non-trivial copy constructor; if so, objects won't
    // be passed in registers
    // intvec(intvec const& x): v(x.v) {}
    // intvec& operator=(intvec const& x) { return v=x.v, *this; }
    intvec(int_t a): v(a) {}
    intvec(int_t const* as): v(as[0]) {}
    static intvec_t iota() { return intvec(I(0)); }
    
    operator ivector_t() const { return v; }
    int_t operator[](int n) const { return v; }
    intvec_t& set_elt(int n, int_t a) { return v=a, *this; }
    
    
    
    boolvec_t as_bool() const { return U(v); }
    boolvec_t convert_bool() const { return bool(v); }
    realvec_t as_float() const;      // defined after realvec
    realvec_t convert_float() const; // defined after realvec
    
    
    
    intvec_t operator+() const { return +v; }
    intvec_t operator-() const { return -v; }
    
    intvec_t operator+(intvec_t x) const { return v+x.v; }
    intvec_t operator-(intvec_t x) const { return v-x.v; }
    intvec_t operator*(intvec_t x) const { return v*x.v; }
    intvec_t operator/(intvec_t x) const { return v/x.v; }
    intvec_t operator%(intvec_t x) const { return v%x.v; }
    
    intvec_t& operator+=(intvec_t const& x) { return *this=*this+x; }
    intvec_t& operator-=(intvec_t const& x) { return *this=*this-x; }
    intvec_t& operator*=(intvec_t const& x) { return *this=*this*x; }
    intvec_t& operator/=(intvec_t const& x) { return *this=*this/x; }
    intvec_t& operator%=(intvec_t const& x) { return *this=*this%x; }
    
    
    
    intvec_t operator~() const { return ~v; }
    
    intvec_t operator&(intvec_t x) const { return v&x.v; }
    intvec_t operator|(intvec_t x) const { return v|x.v; }
    intvec_t operator^(intvec_t x) const { return v^x.v; }
    
    intvec_t& operator&=(intvec_t const& x) { return *this=*this&x; }
    intvec_t& operator|=(intvec_t const& x) { return *this=*this|x; }
    intvec_t& operator^=(intvec_t const& x) { return *this=*this^x; }
    
    intvec_t bitifthen(intvec_t x, intvec_t y) const;
    
    
    
    intvec_t lsr(int_t n) const { return U(v) >> U(n); }
    intvec_t rotate(int_t n) const;
    intvec_t operator>>(int_t n) const { return v>>n; }
    intvec_t operator<<(int_t n) const { return v<<n; }
    
    intvec_t& operator>>=(int_t n) { return *this=*this>>n; }
    intvec_t& operator<<=(int_t n) { return *this=*this<<n; }
    
    intvec_t lsr(intvec_t n) const { return U(v) >> U(n); }
    intvec_t rotate(intvec_t n) const;
    intvec_t operator>>(intvec_t n) const { return v>>n; }
    intvec_t operator<<(intvec_t n) const { return v<<n; }
    
    intvec_t& operator>>=(intvec_t n) { return *this=*this>>n; }
    intvec_t& operator<<=(intvec_t n) { return *this=*this<<n; }
    
    intvec_t clz() const { return __builtin_clz(v); }
    intvec_t popcount() const { return __builtin_popcount(v); }
    
    
    
    boolvec_t operator==(intvec_t const& x) const { return v==x.v; }
    boolvec_t operator!=(intvec_t const& x) const { return v!=x.v; }
    boolvec_t operator<(intvec_t const& x) const { return v<x.v; }
    boolvec_t operator<=(intvec_t const& x) const { return v<=x.v; }
    boolvec_t operator>(intvec_t const& x) const { return v>x.v; }
    boolvec_t operator>=(intvec_t const& x) const { return v>=x.v; }
    
    intvec_t abs() const { return std::abs(v); }
    boolvec_t isignbit() const { return v<0; }
    intvec_t max(intvec_t x) const { return std::max(v, x.v); }
    intvec_t min(intvec_t x) const { return std::min(v, x.v); }
  };
  
  
  
  template<>
  struct realvec<float,1>: floatprops<float>
  {
    static int const size = 1;
    typedef real_t scalar_t;
    typedef float vector_t;
    static int const alignment = sizeof(vector_t);
    
    static char const* name() { return "<SSE2:1*float>"; }
    void barrier() { __asm__("": "+x"(v)); }
    
    static_assert(size * sizeof(real_t) == sizeof(vector_t),
                  "vector size is wrong");
    
  private:
    static __m128 from_float(float a) { return _mm_set_ss(a); }
    static float to_float(__m128 a) { return _mm_cvtss_f32(a); }
  public:
    
    typedef boolvec<real_t, size> boolvec_t;
    typedef intvec<real_t, size> intvec_t;
    typedef realvec realvec_t;
    
    // Short names for type casts
    typedef real_t R;
    typedef int_t I;
    typedef uint_t U;
    typedef realvec_t RV;
    typedef intvec_t IV;
    typedef boolvec_t BV;
    typedef floatprops<real_t> FP;
    typedef mathfuncs<realvec_t> MF;
    
    
    
    vector_t v;
    
    realvec() {}
    // Can't have a non-trivial copy constructor; if so, objects won't
    // be passed in registers
    // realvec(realvec const& x): v(x.v) {}
    // realvec& operator=(realvec const& x) { return v=x.v, *this; }
    realvec(real_t a): v(a) {}
    realvec(real_t const* as): v(as[0]) {}
    
    operator vector_t() const { return v; }
    real_t operator[](int n) const { return v; }
    realvec_t& set_elt(int n, real_t a) { return v=a, *this; }
    
    
    
    typedef vecmathlib::mask_t<realvec_t> mask_t;
    
    static realvec_t loada(real_t const* p)
    {
      VML_ASSERT(intptr_t(p) % alignment == 0);
      return *p;
    }
    static realvec_t loadu(real_t const* p)
    {
      return *p;
    }
    static realvec_t loadu(real_t const* p, std::ptrdiff_t ioff)
    {
      VML_ASSERT(intptr_t(p) % alignment == 0);
      return loada(p+ioff);
    }
    realvec_t loada(real_t const* p, mask_t const& m) const
    {
      VML_ASSERT(intptr_t(p) % alignment == 0);
      if (__builtin_expect(all(m.m), true)) {
        return loada(p);
      } else {
        return *this;
      }
    }
    realvec_t loadu(real_t const* p, mask_t const& m) const
    {
      if (__builtin_expect(m.all_m, true)) {
        return loadu(p);
      } else {
        return *this;
      }
    }
    realvec_t loadu(real_t const* p, std::ptrdiff_t ioff, mask_t const& m) const
    {
      VML_ASSERT(intptr_t(p) % alignment == 0);
      return loada(p+ioff, m);
    }
    
    void storea(real_t* p) const
    {
      VML_ASSERT(intptr_t(p) % alignment == 0);
      *p = v;
    }
    void storeu(real_t* p) const
    {
      *p = v;
    }
    void storeu(real_t* p, std::ptrdiff_t ioff) const
    {
      VML_ASSERT(intptr_t(p) % alignment == 0);
      storea(p+ioff);
    }
    void storea(real_t* p, mask_t const& m) const
    {
      VML_ASSERT(intptr_t(p) % alignment == 0);
      if (__builtin_expect(m.all_m, true)) {
        storea(p);
      }
    }
    void storeu(real_t* p, mask_t const& m) const
    {
      if (__builtin_expect(m.all_m, true)) {
        storeu(p);
      }
    }
    void storeu(real_t* p, std::ptrdiff_t ioff, mask_t const& m) const
    {
      VML_ASSERT(intptr_t(p) % alignment == 0);
      storea(p+ioff, m);
    }
    
    
    
    intvec_t as_int() const { return floatprops::as_int(v); }
    intvec_t convert_int() const {
      // return floatprops::convert_int(v);
      return _mm_cvttss_si32(_mm_set_ss(v));
    }
    
    
    
    realvec_t operator+() const { return +v; }
    realvec_t operator-() const { return -v; }
    
    realvec_t operator+(realvec_t x) const { return v+x.v; }
    realvec_t operator-(realvec_t x) const { return v-x.v; }
    realvec_t operator*(realvec_t x) const { return v*x.v; }
    realvec_t operator/(realvec_t x) const { return v/x.v; }
    
    realvec_t& operator+=(realvec_t const& x) { return *this=*this+x; }
    realvec_t& operator-=(realvec_t const& x) { return *this=*this-x; }
    realvec_t& operator*=(realvec_t const& x) { return *this=*this*x; }
    realvec_t& operator/=(realvec_t const& x) { return *this=*this/x; }
    
    real_t maxval() const { return *this; }
    real_t minval() const { return *this; }
    real_t prod() const { return *this; }
    real_t sum() const { return *this; }
    
    
    
    boolvec_t operator==(realvec_t const& x) const { return v==x.v; }
    boolvec_t operator!=(realvec_t const& x) const { return v!=x.v; }
    boolvec_t operator<(realvec_t const& x) const { return v<x.v; }
    boolvec_t operator<=(realvec_t const& x) const { return v<=x.v; }
    boolvec_t operator>(realvec_t const& x) const { return v>x.v; }
    boolvec_t operator>=(realvec_t const& x) const { return v>=x.v; }
    
    
    
    realvec_t acos() const { return MF::vml_acos(*this); }
    realvec_t acosh() const { return MF::vml_acosh(*this); }
    realvec_t asin() const { return MF::vml_asin(*this); }
    realvec_t asinh() const { return MF::vml_asinh(*this); }
    realvec_t atan() const { return MF::vml_atan(*this); }
    realvec_t atan2(realvec_t y) const { return MF::vml_atan2(*this, y); }
    realvec_t atanh() const { return MF::vml_atanh(*this); }
    realvec_t cbrt() const { return MF::vml_cbrt(*this); }
    realvec_t ceil() const
    {
#ifdef __SSE4_1__
      return to_float(_mm_ceil_ss(from_float(v), from_float(v)));
#else
      return vml_std::ceil(v);
#endif
    }
    realvec_t copysign(realvec_t y) const { return vml_std::copysign(v, y.v); }
    realvec_t cos() const { return MF::vml_cos(*this); }
    realvec_t cosh() const { return MF::vml_cosh(*this); }
    realvec_t exp() const { return MF::vml_exp(*this); }
    realvec_t exp10() const { return MF::vml_exp10(*this); }
    realvec_t exp2() const { return MF::vml_exp2(*this); }
    realvec_t expm1() const { return MF::vml_expm1(*this); }
    realvec_t fabs() const { return vml_std::fabs(v); }
    realvec_t fdim(realvec_t y) const { return MF::vml_fdim(*this, y); }
    realvec_t floor() const
    {
#ifdef __SSE4_1__
      return to_float(_mm_floor_ss(from_float(v), from_float(v)));
#else
      return vml_std::floor(v);
#endif
    }
    realvec_t fma(realvec_t y, realvec_t z) const
    {
      return MF::vml_fma(*this, y, z);
    }
    realvec_t fmax(realvec_t y) const
    {
      return to_float(_mm_max_ss(from_float(v), from_float(y.v)));
    }
    realvec_t fmin(realvec_t y) const
    {
      return to_float(_mm_min_ss(from_float(v), from_float(y.v)));
    }
    realvec_t fmod(realvec_t y) const { return vml_std::fmod(v, y.v); }
    realvec_t frexp(intvec_t* irp) const
    {
      int iri;
      realvec_t r = vml_std::frexp(v, &iri);
      int_t ir = iri;
      if (isinf()) ir = std::numeric_limits<int_t>::max();
      if (isnan()) ir = std::numeric_limits<int_t>::min();
      irp->v = ir;
      return r;
    }
    realvec_t hypot(realvec_t y) const { return MF::vml_hypot(*this, y); }
    intvec_t ilogb() const
    {
      int_t r = vml_std::ilogb(v);
      typedef std::numeric_limits<int_t> NL;
      if (FP_ILOGB0 != NL::min() and *this == RV(R(0.0))) {
        r = NL::min();
#if defined VML_HAVE_INF
      } else if (INT_MAX != NL::max() and vml_std::isinf(v)) {
        r = NL::max();
#endif
#if defined VML_HAVE_NAN
      } else if (FP_ILOGBNAN != NL::min() and isnan()) {
        r = NL::min();
#endif
      }
      return r;
    }
    boolvec_t isfinite() const { return vml_std::isfinite(v); }
    boolvec_t isinf() const { return vml_std::isinf(v); }
    boolvec_t isnan() const
    {
#if defined VML_HAVE_NAN
      // This is wrong:
      // return _mm_ucomineq_ss(from_float(v), from_float(v));
      // This works:
      // char r;
      // __asm__("ucomiss %[v],%[v]; setp %[r]": [r]"=q"(r): [v]"x"(v));
      // return boolvec_t::scalar_t(r);
      // This works as well:
      return vml_std::isnan(v);
#else
      return BV(false);
#endif
    }
    boolvec_t isnormal() const { return vml_std::isnormal(v); }
    realvec_t ldexp(int_t n) const { return vml_std::ldexp(v, n); }
    realvec_t ldexp(intvec_t n) const { return vml_std::ldexp(v, n); }
    realvec_t log() const { return MF::vml_log(*this); }
    realvec_t log10() const { return MF::vml_log10(*this); }
    realvec_t log1p() const { return MF::vml_log1p(*this); }
    realvec_t log2() const { return MF::vml_log2(*this); }
    realvec_t mad(realvec_t y, realvec_t z) const
    {
      return MF::vml_mad(*this, y, z);
    }
    realvec_t nextafter(realvec_t y) const
    {
      return MF::vml_nextafter(*this, y);
    }
    realvec_t pow(realvec_t y) const { return MF::vml_pow(*this, y); }
    realvec_t rcp() const { return R(1.0)/v; }
    realvec_t remainder(realvec_t y) const
    {
      return vml_std::remainder(v, y.v);
    }
    realvec_t rint() const
    {
#ifdef __SSE4_1__
      return to_float(_mm_round_ss(from_float(v), from_float(v),
                                   _MM_FROUND_TO_NEAREST_INT));
#else
      return MF::vml_rint(*this);
#endif
    }
    realvec_t round() const { return MF::vml_round(*this); }
    realvec_t rsqrt() const { return MF::vml_rsqrt(*this); }
    boolvec_t signbit() const { return vml_std::signbit(v); }
    realvec_t sin() const { return MF::vml_sin(*this); }
    realvec_t sinh() const { return MF::vml_sinh(*this); }
    realvec_t sqrt() const { return to_float(_mm_sqrt_ss(from_float(v))); }
    realvec_t tan() const { return MF::vml_tan(*this); }
    realvec_t tanh() const { return MF::vml_tanh(*this); }
    realvec_t trunc() const
    {
#ifdef __SSE4_1__
      return to_float(_mm_round_ss(from_float(v), from_float(v),
                                   _MM_FROUND_TO_ZERO));
#else
      return MF::vml_trunc(*this);
#endif
    }
  };
  
  
  
  // boolvec definitions
  
  inline intvec<float,1> boolvec<float,1>::as_int() const
  {
    return I(v);
  }
  
  inline intvec<float,1> boolvec<float,1>::convert_int() const
  {
    return v;
  }
  
  inline
  boolvec<float,1> boolvec<float,1>::ifthen(boolvec_t x, boolvec_t y) const
  {
    return v ? x : y;
  }
  
  inline intvec<float,1> boolvec<float,1>::ifthen(intvec_t x, intvec_t y) const
  {
    return v ? x : y;
  }
  
  inline
  realvec<float,1> boolvec<float,1>::ifthen(realvec_t x, realvec_t y) const
  {
    return v ? x : y;
  }
  
  
  
  // intvec definitions
  
  inline realvec<float,1> intvec<float,1>::as_float() const
  {
    return FP::as_float(v);
  }
  
  inline intvec<float,1> intvec<float,1>::bitifthen(intvec_t x,
                                                    intvec_t y) const
  {
    return MF::vml_bitifthen(*this, x, y);
  }
  
  inline realvec<float,1> intvec<float,1>::convert_float() const
  {
    // return FP::convert_float(v);
    return _mm_cvtss_f32(_mm_cvtsi32_ss(_mm_setzero_ps(), v));
  }
  
  inline intvec<float,1> intvec<float,1>::rotate(int_t n) const
  {
    return MF::vml_rotate(*this, n);
  }
  
  inline intvec<float,1> intvec<float,1>::rotate(intvec_t n) const
  {
    return MF::vml_rotate(*this, n);
  }
  
} // namespace vecmathlib

#endif  // #ifndef VEC_SSE_FLOAT1_H
OpenPOWER on IntegriCloud