about summary refs log tree commit diff
path: root/sysdeps/ia64/fpu/s_expm1.S
blob: f0b911e29556e8b71ed725c7c7e57ab0d2149549 (plain) (blame)
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
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
.file "exp_m1.s"


// Copyright (c) 2000 - 2005, Intel Corporation
// All rights reserved.
//
// Contributed 2000 by the Intel Numerics Group, Intel Corporation
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions are
// met:
//
// * Redistributions of source code must retain the above copyright
// notice, this list of conditions and the following disclaimer.
//
// * Redistributions in binary form must reproduce the above copyright
// notice, this list of conditions and the following disclaimer in the
// documentation and/or other materials provided with the distribution.
//
// * The name of Intel Corporation may not be used to endorse or promote
// products derived from this software without specific prior written
// permission.

// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL INTEL OR ITS
// CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
// EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
// PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
// PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
// OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY OR TORT (INCLUDING
// NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
// SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
//
// Intel Corporation is the author of this code, and requests that all
// problem reports or change requests be submitted to it directly at
// http://www.intel.com/software/products/opensource/libraries/num.htm.
//
// History
//==============================================================
// 02/02/00 Initial Version
// 04/04/00 Unwind support added
// 08/15/00 Bundle added after call to __libm_error_support to properly
//          set [the previously overwritten] GR_Parameter_RESULT.
// 07/07/01 Improved speed of all paths
// 05/20/02 Cleaned up namespace and sf0 syntax
// 11/20/02 Improved speed, algorithm based on exp
// 03/31/05 Reformatted delimiters between data tables

// API
//==============================================================
// double expm1(double)

// Overview of operation
//==============================================================
// 1. Inputs of Nan, Inf, Zero, NatVal handled with special paths
//
// 2. |x| < 2^-60
//    Result = x, computed by x + x*x to handle appropriate flags and rounding
//
// 3. 2^-60 <= |x| < 2^-2
//    Result determined by 13th order Taylor series polynomial
//    expm1f(x) = x + Q2*x^2 + ... + Q13*x^13
//
// 4. x < -48.0
//    Here we know result is essentially -1 + eps, where eps only affects
//    rounded result.  Set I.
//
// 5. x >= 709.7827
//    Result overflows.  Set I, O, and call error support
//
// 6. 2^-2 <= x < 709.7827  or  -48.0 <= x < -2^-2
//    This is the main path.  The algorithm is described below:

// Take the input x. w is "how many log2/128 in x?"
//  w = x * 128/log2
//  n = int(w)
//  x = n log2/128 + r + delta

//  n = 128M + index_1 + 2^4 index_2
//  x = M log2 + (log2/128) index_1 + (log2/8) index_2 + r + delta

//  exp(x) = 2^M  2^(index_1/128)  2^(index_2/8) exp(r) exp(delta)
//       Construct 2^M
//       Get 2^(index_1/128) from table_1;
//       Get 2^(index_2/8)   from table_2;
//       Calculate exp(r) by series by 5th order polynomial
//          r = x - n (log2/128)_high
//          delta = - n (log2/128)_low
//       Calculate exp(delta) as 1 + delta


// Special values
//==============================================================
// expm1(+0)    = +0.0
// expm1(-0)    = -0.0

// expm1(+qnan) = +qnan
// expm1(-qnan) = -qnan
// expm1(+snan) = +qnan
// expm1(-snan) = -qnan

// expm1(-inf)  = -1.0
// expm1(+inf)  = +inf

// Overflow and Underflow
//=======================
// expm1(x) = largest double normal when
//     x = 709.7827 = 40862e42fefa39ef
//
// Underflow is handled as described in case 2 above.


// Registers used
//==============================================================
// Floating Point registers used:
// f8, input
// f9 -> f15,  f32 -> f75

// General registers used:
// r14 -> r40

// Predicate registers used:
// p6 -> p15

// Assembly macros
//==============================================================

rRshf                  = r14
rAD_TB1                = r15
rAD_T1                 = r15
rAD_TB2                = r16
rAD_T2                 = r16
rAD_Ln2_lo             = r17
rAD_P                  = r17

rN                     = r18
rIndex_1               = r19
rIndex_2_16            = r20

rM                     = r21
rBiased_M              = r21
rIndex_1_16            = r22
rSignexp_x             = r23
rExp_x                 = r24
rSig_inv_ln2           = r25

rAD_Q1                 = r26
rAD_Q2                 = r27
rTmp                   = r27
rExp_bias              = r28
rExp_mask              = r29
rRshf_2to56            = r30

rGt_ln                 = r31
rExp_2tom56            = r31


GR_SAVE_B0             = r33
GR_SAVE_PFS            = r34
GR_SAVE_GP             = r35
GR_SAVE_SP             = r36

GR_Parameter_X         = r37
GR_Parameter_Y         = r38
GR_Parameter_RESULT    = r39
GR_Parameter_TAG       = r40


FR_X                   = f10
FR_Y                   = f1
FR_RESULT              = f8

fRSHF_2TO56            = f6
fINV_LN2_2TO63         = f7
fW_2TO56_RSH           = f9
f2TOM56                = f11
fP5                    = f12
fP54                   = f50
fP5432                 = f50
fP4                    = f13
fP3                    = f14
fP32                   = f14
fP2                    = f15

fLn2_by_128_hi         = f33
fLn2_by_128_lo         = f34

fRSHF                  = f35
fNfloat                = f36
fW                     = f37
fR                     = f38
fF                     = f39

fRsq                   = f40
fRcube                 = f41

f2M                    = f42
fS1                    = f43
fT1                    = f44

fMIN_DBL_OFLOW_ARG     = f45
fMAX_DBL_MINUS_1_ARG   = f46
fMAX_DBL_NORM_ARG      = f47
fP_lo                  = f51
fP_hi                  = f52
fP                     = f53
fS                     = f54

fNormX                 = f56

fWre_urm_f8            = f57

fGt_pln                = f58
fTmp                   = f58

fS2                    = f59
fT2                    = f60
fSm1                   = f61

fXsq                   = f62
fX6                    = f63
fX4                    = f63
fQ7                    = f64
fQ76                   = f64
fQ7654                 = f64
fQ765432               = f64
fQ6                    = f65
fQ5                    = f66
fQ54                   = f66
fQ4                    = f67
fQ3                    = f68
fQ32                   = f68
fQ2                    = f69
fQD                    = f70
fQDC                   = f70
fQDCBA                 = f70
fQDCBA98               = f70
fQDCBA98765432         = f70
fQC                    = f71
fQB                    = f72
fQBA                   = f72
fQA                    = f73
fQ9                    = f74
fQ98                   = f74
fQ8                    = f75

// Data tables
//==============================================================

RODATA
.align 16

// ************* DO NOT CHANGE ORDER OF THESE TABLES ********************

// double-extended 1/ln(2)
// 3fff b8aa 3b29 5c17 f0bb be87fed0691d3e88
// 3fff b8aa 3b29 5c17 f0bc
// For speed the significand will be loaded directly with a movl and setf.sig
//   and the exponent will be bias+63 instead of bias+0.  Thus subsequent
//   computations need to scale appropriately.
// The constant 128/ln(2) is needed for the computation of w.  This is also
//   obtained by scaling the computations.
//
// Two shifting constants are loaded directly with movl and setf.d.
//   1. fRSHF_2TO56 = 1.1000..00 * 2^(63-7)
//        This constant is added to x*1/ln2 to shift the integer part of
//        x*128/ln2 into the rightmost bits of the significand.
//        The result of this fma is fW_2TO56_RSH.
//   2. fRSHF       = 1.1000..00 * 2^(63)
//        This constant is subtracted from fW_2TO56_RSH * 2^(-56) to give
//        the integer part of w, n, as a floating-point number.
//        The result of this fms is fNfloat.


LOCAL_OBJECT_START(exp_Table_1)
data8 0x40862e42fefa39f0 // smallest dbl overflow arg
data8 0xc048000000000000 // approx largest arg for minus one result
data8 0x40862e42fefa39ef // largest dbl arg to give normal dbl result
data8 0x0                // pad
data8 0xb17217f7d1cf79ab , 0x00003ff7 // ln2/128 hi
data8 0xc9e3b39803f2f6af , 0x00003fb7 // ln2/128 lo
//
// Table 1 is 2^(index_1/128) where
// index_1 goes from 0 to 15
//
data8 0x8000000000000000 , 0x00003FFF
data8 0x80B1ED4FD999AB6C , 0x00003FFF
data8 0x8164D1F3BC030773 , 0x00003FFF
data8 0x8218AF4373FC25EC , 0x00003FFF
data8 0x82CD8698AC2BA1D7 , 0x00003FFF
data8 0x8383594EEFB6EE37 , 0x00003FFF
data8 0x843A28C3ACDE4046 , 0x00003FFF
data8 0x84F1F656379C1A29 , 0x00003FFF
data8 0x85AAC367CC487B15 , 0x00003FFF
data8 0x8664915B923FBA04 , 0x00003FFF
data8 0x871F61969E8D1010 , 0x00003FFF
data8 0x87DB357FF698D792 , 0x00003FFF
data8 0x88980E8092DA8527 , 0x00003FFF
data8 0x8955EE03618E5FDD , 0x00003FFF
data8 0x8A14D575496EFD9A , 0x00003FFF
data8 0x8AD4C6452C728924 , 0x00003FFF
LOCAL_OBJECT_END(exp_Table_1)

// Table 2 is 2^(index_1/8) where
// index_2 goes from 0 to 7
LOCAL_OBJECT_START(exp_Table_2)
data8 0x8000000000000000 , 0x00003FFF
data8 0x8B95C1E3EA8BD6E7 , 0x00003FFF
data8 0x9837F0518DB8A96F , 0x00003FFF
data8 0xA5FED6A9B15138EA , 0x00003FFF
data8 0xB504F333F9DE6484 , 0x00003FFF
data8 0xC5672A115506DADD , 0x00003FFF
data8 0xD744FCCAD69D6AF4 , 0x00003FFF
data8 0xEAC0C6E7DD24392F , 0x00003FFF
LOCAL_OBJECT_END(exp_Table_2)


LOCAL_OBJECT_START(exp_p_table)
data8 0x3f8111116da21757 //P5
data8 0x3fa55555d787761c //P4
data8 0x3fc5555555555414 //P3
data8 0x3fdffffffffffd6a //P2
LOCAL_OBJECT_END(exp_p_table)

LOCAL_OBJECT_START(exp_Q1_table)
data8 0x3de6124613a86d09 // QD = 1/13!
data8 0x3e21eed8eff8d898 // QC = 1/12!
data8 0x3ec71de3a556c734 // Q9 = 1/9!
data8 0x3efa01a01a01a01a // Q8 = 1/8!
data8 0x8888888888888889,0x3ff8 // Q5 = 1/5!
data8 0xaaaaaaaaaaaaaaab,0x3ffc // Q3 = 1/3!
data8 0x0,0x0            // Pad to avoid bank conflicts
LOCAL_OBJECT_END(exp_Q1_table)

LOCAL_OBJECT_START(exp_Q2_table)
data8 0x3e5ae64567f544e4 // QB = 1/11!
data8 0x3e927e4fb7789f5c // QA = 1/10!
data8 0x3f2a01a01a01a01a // Q7 = 1/7!
data8 0x3f56c16c16c16c17 // Q6 = 1/6!
data8 0xaaaaaaaaaaaaaaab,0x3ffa // Q4 = 1/4!
data8 0x8000000000000000,0x3ffe // Q2 = 1/2!
LOCAL_OBJECT_END(exp_Q2_table)


.section .text
GLOBAL_IEEE754_ENTRY(expm1)

{ .mlx
      getf.exp        rSignexp_x = f8  // Must recompute if x unorm
      movl            rSig_inv_ln2 = 0xb8aa3b295c17f0bc  // signif of 1/ln2
}
{ .mlx
      addl            rAD_TB1    = @ltoff(exp_Table_1), gp
      movl            rRshf_2to56 = 0x4768000000000000   // 1.10000 2^(63+56)
}
;;

// We do this fnorm right at the beginning to normalize
// any input unnormals so that SWA is not taken.
{ .mfi
      ld8             rAD_TB1    = [rAD_TB1]
      fclass.m        p6,p0 = f8,0x0b  // Test for x=unorm
      mov             rExp_mask = 0x1ffff
}
{ .mfi
      mov             rExp_bias = 0xffff
      fnorm.s1        fNormX   = f8
      mov             rExp_2tom56 = 0xffff-56
}
;;

// Form two constants we need
//  1/ln2 * 2^63  to compute  w = x * 1/ln2 * 128
//  1.1000..000 * 2^(63+63-7) to right shift int(w) into the significand

{ .mfi
      setf.sig        fINV_LN2_2TO63 = rSig_inv_ln2 // form 1/ln2 * 2^63
      fclass.m        p8,p0 = f8,0x07  // Test for x=0
      nop.i           0
}
{ .mlx
      setf.d          fRSHF_2TO56 = rRshf_2to56 // Form 1.100 * 2^(63+56)
      movl            rRshf = 0x43e8000000000000   // 1.10000 2^63 for rshift
}
;;

{ .mfi
      setf.exp        f2TOM56 = rExp_2tom56 // form 2^-56 for scaling Nfloat
      fclass.m        p9,p0 = f8,0x22  // Test for x=-inf
      add             rAD_TB2 = 0x140, rAD_TB1 // Point to Table 2
}
{ .mib
      add             rAD_Q1 = 0x1e0, rAD_TB1 // Point to Q table for small path
      add             rAD_Ln2_lo = 0x30, rAD_TB1 // Point to ln2_by_128_lo
(p6)  br.cond.spnt    EXPM1_UNORM // Branch if x unorm
}
;;

EXPM1_COMMON:
{ .mfi
      ldfpd           fMIN_DBL_OFLOW_ARG, fMAX_DBL_MINUS_1_ARG = [rAD_TB1],16
      fclass.m        p10,p0 = f8,0x1e1  // Test for x=+inf, NaN, NaT
      add             rAD_Q2 = 0x50, rAD_Q1   // Point to Q table for small path
}
{ .mfb
      nop.m           0
      nop.f           0
(p8)  br.ret.spnt     b0                        // Exit for x=0, return x
}
;;

{ .mfi
      ldfd            fMAX_DBL_NORM_ARG = [rAD_TB1],16
      nop.f           0
      and             rExp_x = rExp_mask, rSignexp_x // Biased exponent of x
}
{ .mfb
      setf.d          fRSHF = rRshf // Form right shift const 1.100 * 2^63
(p9)  fms.d.s0        f8 = f0,f0,f1            // quick exit for x=-inf
(p9)  br.ret.spnt     b0
}
;;

{ .mfi
      ldfpd           fQD, fQC = [rAD_Q1], 16  // Load coeff for small path
      nop.f           0
      sub             rExp_x = rExp_x, rExp_bias // True exponent of x
}
{ .mfb
      ldfpd           fQB, fQA = [rAD_Q2], 16  // Load coeff for small path
(p10) fma.d.s0        f8 = f8, f1, f0          // For x=+inf, NaN, NaT
(p10) br.ret.spnt     b0                       // Exit for x=+inf, NaN, NaT
}
;;

{ .mfi
      ldfpd           fQ9, fQ8 = [rAD_Q1], 16  // Load coeff for small path
      fma.s1          fXsq = fNormX, fNormX, f0  // x*x for small path
      cmp.gt          p7, p8 = -2, rExp_x      // Test |x| < 2^(-2)
}
{ .mfi
      ldfpd           fQ7, fQ6 = [rAD_Q2], 16  // Load coeff for small path
      nop.f           0
      nop.i           0
}
;;

{ .mfi
      ldfe            fQ5 = [rAD_Q1], 16       // Load coeff for small path
      nop.f           0
      nop.i           0
}
{ .mib
      ldfe            fQ4 = [rAD_Q2], 16       // Load coeff for small path
(p7)  cmp.gt.unc      p6, p7 = -60, rExp_x     // Test |x| < 2^(-60)
(p7)  br.cond.spnt    EXPM1_SMALL              // Branch if 2^-60 <= |x| < 2^-2
}
;;

// W = X * Inv_log2_by_128
// By adding 1.10...0*2^63 we shift and get round_int(W) in significand.
// We actually add 1.10...0*2^56 to X * Inv_log2 to do the same thing.

{ .mfi
      ldfe            fLn2_by_128_hi  = [rAD_TB1],32
      fma.s1          fW_2TO56_RSH  = fNormX, fINV_LN2_2TO63, fRSHF_2TO56
      nop.i           0
}
{ .mfb
      ldfe            fLn2_by_128_lo  = [rAD_Ln2_lo]
(p6)  fma.d.s0        f8 = f8, f8, f8 // If x < 2^-60, result=x+x*x
(p6)  br.ret.spnt     b0              // Exit if x < 2^-60
}
;;

// Divide arguments into the following categories:
//  Certain minus one       p11 - -inf < x <= MAX_DBL_MINUS_1_ARG
//  Possible Overflow       p14 - MAX_DBL_NORM_ARG < x < MIN_DBL_OFLOW_ARG
//  Certain Overflow        p15 - MIN_DBL_OFLOW_ARG <= x < +inf
//
// If the input is really a double arg, then there will never be "Possible
// Overflow" arguments.
//

// After that last load, rAD_TB1 points to the beginning of table 1

{ .mfi
      nop.m           0
      fcmp.ge.s1      p15,p14 = fNormX,fMIN_DBL_OFLOW_ARG
      nop.i           0
}
;;

{ .mfi
      add             rAD_P = 0x80, rAD_TB2
      fcmp.le.s1      p11,p0 = fNormX,fMAX_DBL_MINUS_1_ARG
      nop.i           0
}
;;

{ .mfb
      ldfpd           fP5, fP4  = [rAD_P] ,16
(p14) fcmp.gt.unc.s1  p14,p0 = fNormX,fMAX_DBL_NORM_ARG
(p15) br.cond.spnt    EXPM1_CERTAIN_OVERFLOW
}
;;

// Nfloat = round_int(W)
// The signficand of fW_2TO56_RSH contains the rounded integer part of W,
// as a twos complement number in the lower bits (that is, it may be negative).
// That twos complement number (called N) is put into rN.

// Since fW_2TO56_RSH is scaled by 2^56, it must be multiplied by 2^-56
// before the shift constant 1.10000 * 2^63 is subtracted to yield fNfloat.
// Thus, fNfloat contains the floating point version of N

{ .mfb
      ldfpd           fP3, fP2  = [rAD_P]
      fms.s1          fNfloat = fW_2TO56_RSH, f2TOM56, fRSHF
(p11) br.cond.spnt    EXPM1_CERTAIN_MINUS_ONE
}
;;

{ .mfi
      getf.sig        rN = fW_2TO56_RSH
      nop.f           0
      nop.i           0
}
;;

// rIndex_1 has index_1
// rIndex_2_16 has index_2 * 16
// rBiased_M has M
// rIndex_1_16 has index_1 * 16

// r = x - Nfloat * ln2_by_128_hi
// f = 1 - Nfloat * ln2_by_128_lo
{ .mfi
      and             rIndex_1 = 0x0f, rN
      fnma.s1         fR   = fNfloat, fLn2_by_128_hi, fNormX
      shr             rM = rN,  0x7
}
{ .mfi
      and             rIndex_2_16 = 0x70, rN
      fnma.s1         fF   = fNfloat, fLn2_by_128_lo, f1
      nop.i           0
}
;;

// rAD_T1 has address of T1
// rAD_T2 has address if T2

{ .mmi
      add             rBiased_M = rExp_bias, rM
      add             rAD_T2 = rAD_TB2, rIndex_2_16
      shladd          rAD_T1 = rIndex_1, 4, rAD_TB1
}
;;

// Create Scale = 2^M
// Load T1 and T2
{ .mmi
      setf.exp        f2M = rBiased_M
      ldfe            fT2  = [rAD_T2]
      nop.i           0
}
;;

{ .mfi
      ldfe            fT1  = [rAD_T1]
      fmpy.s0         fTmp = fLn2_by_128_lo, fLn2_by_128_lo // Force inexact
      nop.i           0
}
;;

{ .mfi
      nop.m           0
      fma.s1          fP54 = fR, fP5, fP4
      nop.i           0
}
{ .mfi
      nop.m           0
      fma.s1          fP32 = fR, fP3, fP2
      nop.i           0
}
;;

{ .mfi
      nop.m           0
      fma.s1          fRsq = fR, fR, f0
      nop.i           0
}
;;

{ .mfi
      nop.m           0
      fma.s1          fP5432  = fRsq, fP54, fP32
      nop.i           0
}
;;

{ .mfi
      nop.m           0
      fma.s1          fS2  = fF,fT2,f0
      nop.i           0
}
{ .mfi
      nop.m           0
      fma.s1          fS1  = f2M,fT1,f0
      nop.i           0
}
;;

{ .mfi
      nop.m           0
      fma.s1          fP = fRsq, fP5432, fR
      nop.i           0
}
;;

{ .mfi
      nop.m           0
      fms.s1          fSm1 = fS1,fS2,f1    // S - 1.0
      nop.i           0
}
{ .mfb
      nop.m           0
      fma.s1          fS   = fS1,fS2,f0
(p14) br.cond.spnt    EXPM1_POSSIBLE_OVERFLOW
}
;;

{ .mfb
      nop.m           0
      fma.d.s0        f8 = fS, fP, fSm1
      br.ret.sptk     b0                // Normal path exit
}
;;

// Here if 2^-60 <= |x| <2^-2
// Compute 13th order polynomial
EXPM1_SMALL:
{ .mmf
      ldfe            fQ3 = [rAD_Q1], 16
      ldfe            fQ2 = [rAD_Q2], 16
      fma.s1          fX4 = fXsq, fXsq, f0
}
;;

{ .mfi
      nop.m           0
      fma.s1          fQDC = fQD, fNormX, fQC
      nop.i           0
}
{ .mfi
      nop.m           0
      fma.s1          fQBA = fQB, fNormX, fQA
      nop.i           0
}
;;

{ .mfi
      nop.m           0
      fma.s1          fQ98 = fQ9, fNormX, fQ8
      nop.i           0
}
{ .mfi
      nop.m           0
      fma.s1          fQ76= fQ7, fNormX, fQ6
      nop.i           0
}
;;

{ .mfi
      nop.m           0
      fma.s1          fQ54 = fQ5, fNormX, fQ4
      nop.i           0
}
;;

{ .mfi
      nop.m           0
      fma.s1          fX6 = fX4, fXsq, f0
      nop.i           0
}
{ .mfi
      nop.m           0
      fma.s1          fQ32= fQ3, fNormX, fQ2
      nop.i           0
}
;;

{ .mfi
      nop.m           0
      fma.s1          fQDCBA = fQDC, fXsq, fQBA
      nop.i           0
}
{ .mfi
      nop.m           0
      fma.s1          fQ7654 = fQ76, fXsq, fQ54
      nop.i           0
}
;;

{ .mfi
      nop.m           0
      fma.s1          fQDCBA98 = fQDCBA, fXsq, fQ98
      nop.i           0
}
{ .mfi
      nop.m           0
      fma.s1          fQ765432 = fQ7654, fXsq, fQ32
      nop.i           0
}
;;

{ .mfi
      nop.m           0
      fma.s1          fQDCBA98765432 = fQDCBA98, fX6, fQ765432
      nop.i           0
}
;;

{ .mfb
      nop.m           0
      fma.d.s0        f8 = fQDCBA98765432, fXsq, fNormX
      br.ret.sptk     b0                   // Exit small branch
}
;;


EXPM1_POSSIBLE_OVERFLOW:

// Here if fMAX_DBL_NORM_ARG < x < fMIN_DBL_OFLOW_ARG
// This cannot happen if input is a double, only if input higher precision.
// Overflow is a possibility, not a certainty.

// Recompute result using status field 2 with user's rounding mode,
// and wre set.  If result is larger than largest double, then we have
// overflow

{ .mfi
      mov             rGt_ln  = 0x103ff // Exponent for largest dbl + 1 ulp
      fsetc.s2        0x7F,0x42         // Get user's round mode, set wre
      nop.i           0
}
;;

{ .mfi
      setf.exp        fGt_pln = rGt_ln  // Create largest double + 1 ulp
      fma.d.s2        fWre_urm_f8 = fS, fP, fSm1  // Result with wre set
      nop.i           0
}
;;

{ .mfi
      nop.m           0
      fsetc.s2        0x7F,0x40                   // Turn off wre in sf2
      nop.i           0
}
;;

{ .mfi
      nop.m           0
      fcmp.ge.s1      p6, p0 =  fWre_urm_f8, fGt_pln // Test for overflow
      nop.i           0
}
;;

{ .mfb
      nop.m           0
      nop.f           0
(p6)  br.cond.spnt    EXPM1_CERTAIN_OVERFLOW // Branch if overflow
}
;;

{ .mfb
      nop.m           0
      fma.d.s0        f8 = fS, fP, fSm1
      br.ret.sptk     b0                     // Exit if really no overflow
}
;;

EXPM1_CERTAIN_OVERFLOW:
{ .mmi
      sub             rTmp = rExp_mask, r0, 1
;;
      setf.exp        fTmp = rTmp
      nop.i           0
}
;;

{ .mfi
      alloc           r32=ar.pfs,1,4,4,0
      fmerge.s        FR_X = f8,f8
      nop.i           0
}
{ .mfb
      mov             GR_Parameter_TAG = 41
      fma.d.s0        FR_RESULT = fTmp, fTmp, f0    // Set I,O and +INF result
      br.cond.sptk    __libm_error_region
}
;;

// Here if x unorm
EXPM1_UNORM:
{ .mfb
      getf.exp        rSignexp_x = fNormX    // Must recompute if x unorm
      fcmp.eq.s0      p6, p0 = f8, f0        // Set D flag
      br.cond.sptk    EXPM1_COMMON
}
;;

// here if result will be -1 and inexact, x <= -48.0
EXPM1_CERTAIN_MINUS_ONE:
{ .mmi
      mov             rTmp = 1
;;
      setf.exp        fTmp = rTmp
      nop.i           0
}
;;

{ .mfb
      nop.m           0
      fms.d.s0        FR_RESULT = fTmp, fTmp, f1 // Set I, rounded -1+eps result
      br.ret.sptk     b0
}
;;

GLOBAL_IEEE754_END(expm1)


LOCAL_LIBM_ENTRY(__libm_error_region)
.prologue
{ .mfi
        add   GR_Parameter_Y=-32,sp             // Parameter 2 value
        nop.f 0
.save   ar.pfs,GR_SAVE_PFS
        mov  GR_SAVE_PFS=ar.pfs                 // Save ar.pfs
}
{ .mfi
.fframe 64
        add sp=-64,sp                           // Create new stack
        nop.f 0
        mov GR_SAVE_GP=gp                       // Save gp
};;
{ .mmi
        stfd [GR_Parameter_Y] = FR_Y,16         // STORE Parameter 2 on stack
        add GR_Parameter_X = 16,sp              // Parameter 1 address
.save   b0, GR_SAVE_B0
        mov GR_SAVE_B0=b0                       // Save b0
};;
.body
{ .mib
        stfd [GR_Parameter_X] = FR_X            // STORE Parameter 1 on stack
        add   GR_Parameter_RESULT = 0,GR_Parameter_Y  // Parameter 3 address
	nop.b 0
}
{ .mib
        stfd [GR_Parameter_Y] = FR_RESULT       // STORE Parameter 3 on stack
        add   GR_Parameter_Y = -16,GR_Parameter_Y
        br.call.sptk b0=__libm_error_support#   // Call error handling function
};;
{ .mmi
        add   GR_Parameter_RESULT = 48,sp
        nop.m 0
        nop.i 0
};;
{ .mmi
        ldfd  f8 = [GR_Parameter_RESULT]       // Get return result off stack
.restore sp
        add   sp = 64,sp                       // Restore stack pointer
        mov   b0 = GR_SAVE_B0                  // Restore return address
};;
{ .mib
        mov   gp = GR_SAVE_GP                  // Restore gp
        mov   ar.pfs = GR_SAVE_PFS             // Restore ar.pfs
        br.ret.sptk     b0                     // Return
};;

LOCAL_LIBM_END(__libm_error_region)
.type   __libm_error_support#,@function
.global __libm_error_support#