1.file "powf.s"
2
3
4// Copyright (c) 2000 - 2005, Intel Corporation
5// All rights reserved.
6//
7//
8// Redistribution and use in source and binary forms, with or without
9// modification, are permitted provided that the following conditions are
10// met:
11//
12// * Redistributions of source code must retain the above copyright
13// notice, this list of conditions and the following disclaimer.
14//
15// * Redistributions in binary form must reproduce the above copyright
16// notice, this list of conditions and the following disclaimer in the
17// documentation and/or other materials provided with the distribution.
18//
19// * The name of Intel Corporation may not be used to endorse or promote
20// products derived from this software without specific prior written
21// permission.
22
23// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
24// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
25// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
26// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL INTEL OR ITS
27// CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
28// EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
29// PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
30// PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY
31// OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY OR TORT (INCLUDING
32// NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
33// SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34//
35// Intel Corporation is the author of this code, and requests that all
36// problem reports or change requests be submitted to it directly at
37// http://www.intel.com/software/products/opensource/libraries/num.htm.
38//
39// History
40//==============================================================
41// 02/02/00 Initial version
42// 02/03/00 Added p12 to definite over/under path. With odd power we did not
43//          maintain the sign of x in this path.
44// 04/04/00 Unwind support added
45// 04/19/00 pow(+-1,inf) now returns NaN
46//          pow(+-val, +-inf) returns 0 or inf, but now does not call error
47//          support
48//          Added s1 to fcvt.fx because invalid flag was incorrectly set.
49// 08/15/00 Bundle added after call to __libm_error_support to properly
50//          set [the previously overwritten] GR_Parameter_RESULT.
51// 09/07/00 Improved performance by eliminating bank conflicts and other stalls,
52//          and tweaking the critical path
53// 09/08/00 Per c99, pow(+-1,inf) now returns 1, and pow(+1,nan) returns 1
54// 09/28/00 Updated NaN**0 path
55// 01/20/01 Fixed denormal flag settings.
56// 02/13/01 Improved speed.
57// 03/19/01 Reordered exp polynomial to improve speed and eliminate monotonicity
58//          problem in round up, down, and to zero modes.  Also corrected
59//          overflow result when x negative, y odd in round up, down, zero.
60// 06/14/01 Added brace missing from bundle
61// 12/10/01 Corrected case where x negative, 2^23 <= |y| < 2^24, y odd integer.
62// 02/08/02 Fixed overflow/underflow cases that were not calling error support.
63// 05/20/02 Cleaned up namespace and sf0 syntax
64// 08/29/02 Improved Itanium 2 performance
65// 02/10/03 Reordered header: .section, .global, .proc, .align
66// 10/09/03 Modified algorithm to improve performance, reduce table size, and
67//          fix boundary case powf(2.0,-150.0)
68// 03/31/05 Reformatted delimiters between data tables
69//
70// API
71//==============================================================
72// float powf(float x, float y)
73//
74// Overview of operation
75//==============================================================
76//
77// Three steps...
78// 1. Log(x)
79// 2. y Log(x)
80// 3. exp(y log(x))
81//
82// This means we work with the absolute value of x and merge in the sign later.
83//      Log(x) = G + delta + r -rsq/2 + p
84// G,delta depend on the exponent of x and table entries. The table entries are
85// indexed by the exponent of x, called K.
86//
87// The G and delta come out of the reduction; r is the reduced x.
88//
89// B = frcpa(x)
90// xB-1 is small means that B is the approximate inverse of x.
91//
92//      Log(x) = Log( (1/B)(Bx) )
93//             = Log(1/B) + Log(Bx)
94//             = Log(1/B) + Log( 1 + (Bx-1))
95//
96//      x  = 2^K 1.x_1x_2.....x_52
97//      B= frcpa(x) = 2^-k Cm
98//      Log(1/B) = Log(1/(2^-K Cm))
99//      Log(1/B) = Log((2^K/ Cm))
100//      Log(1/B) = K Log(2) + Log(1/Cm)
101//
102//      Log(x)   = K Log(2) + Log(1/Cm) + Log( 1 + (Bx-1))
103//
104// If you take the significand of x, set the exponent to true 0, then Cm is
105// the frcpa. We tabulate the Log(1/Cm) values. There are 256 of them.
106// The frcpa table is indexed by 8 bits, the x_1 thru x_8.
107// m = x_1x_2...x_8 is an 8-bit index.
108//
109//      Log(1/Cm) = log(1/frcpa(1+m/256)) where m goes from 0 to 255.
110//
111// We tabluate as one double, T for single precision power
112//
113//      Log(x)   = (K Log(2)_hi + T) + (K Log(2)_lo) + Log( 1 + (Bx-1))
114//      Log(x)   =  G                +     delta     + Log( 1 + (Bx-1))
115//
116// The Log( 1 + (Bx-1)) can be calculated as a series in r = Bx-1.
117//
118//      Log( 1 + (Bx-1)) = r - rsq/2 + p
119//        where p = r^3(P0 + P1*r + P2*r^2)
120//
121// Then,
122//
123//      yLog(x) = yG + y delta + y(r-rsq/2) + yp
124//      yLog(x) = Z1 + e3      + Z2         + Z3
125//
126//
127//     exp(yLog(x)) = exp(Z1 + Z2) exp(Z3) exp(e3)
128//
129//
130//       exp(Z3) is another series.
131//       exp(e3) is approximated as f3 = 1 +  e3
132//
133//       exp(Z1 + Z2) = exp(Z)
134//       Z (128/log2) = number of log2/128 in Z is N
135//
136//       s = Z - N log2/128
137//
138//       exp(Z)       = exp(s) exp(N log2/128)
139//
140//       exp(r)       = exp(Z - N log2/128)
141//
142//      r = s + d = (Z - N (log2/128)_hi) -N (log2/128)_lo
143//                =  Z - N (log2/128)
144//
145//      Z         = s+d +N (log2/128)
146//
147//      exp(Z)    = exp(s) (1+d) exp(N log2/128)
148//
149//      N = M 128 + n
150//
151//      N log2/128 = M log2 + n log2/128
152//
153//      n is 8 binary digits = n_7n_6...n_1
154//
155//      n log2/128 = n_7n_6n_5 16 log2/128 + n_4n_3n_2n_1 log2/128
156//      n log2/128 = n_7n_6n_5 log2/8 + n_4n_3n_2n_1 log2/128
157//      n log2/128 = I2 log2/8 + I1 log2/128
158//
159//      N log2/128 = M log2 + I2 log2/8 + I1 log2/128
160//
161//      exp(Z)    = exp(s) (1+d) exp(log(2^M) + log(2^I2/8) + log(2^I1/128))
162//      exp(Z)    = exp(s) f12 (2^M) 2^I2/8 2^I1/128
163//
164// I1, I2 are table indices. Use a series for exp(s).
165// Then get exp(Z)
166//
167//     exp(yLog(x)) = exp(Z) exp(Z3) f3
168//     exp(yLog(x)) = exp(Z)f3 exp(Z3)
169//     exp(yLog(x)) = A exp(Z3)
170//
171// We actually calculate exp(Z3) -1.
172// Then,
173//     exp(yLog(x)) = A + A( exp(Z3)   -1)
174//
175
176// Table Generation
177//==============================================================
178
179// The log values
180// ==============
181// The operation (K*log2_hi) must be exact. K is the true exponent of x.
182// If we allow gradual underflow (denormals), K can be represented in 12 bits
183// (as a two's complement number). We assume 13 bits as an engineering
184// precaution.
185//
186//           +------------+----------------+-+
187//           |  13 bits   | 50 bits        | |
188//           +------------+----------------+-+
189//           0            1                66
190//                        2                34
191//
192// So we want the lsb(log2_hi) to be 2^-50
193// We get log2 as a quad-extended (15-bit exponent, 128-bit significand)
194//
195//      0 fffe b17217f7d1cf79ab c9e3b39803f2f6af (4...)
196//
197// Consider numbering the bits left to right, starting at 0 thru 127.
198// Bit 0 is the 2^-1 bit; bit 49 is the 2^-50 bit.
199//
200//  ...79ab
201//     0111 1001 1010 1011
202//     44
203//     89
204//
205// So if we shift off the rightmost 14 bits, then (shift back only
206// the top half) we get
207//
208//      0 fffe b17217f7d1cf4000 e6af278ece600fcb dabc000000000000
209//
210// Put the right 64-bit signficand in an FR register, convert to double;
211// it is exact. Put the next 128 bits into a quad register and round to double.
212// The true exponent of the low part is -51.
213//
214// hi is 0 fffe b17217f7d1cf4000
215// lo is 0 ffcc e6af278ece601000
216//
217// Convert to double memory format and get
218//
219// hi is 0x3fe62e42fefa39e8
220// lo is 0x3cccd5e4f1d9cc02
221//
222// log2_hi + log2_lo is an accurate value for log2.
223//
224//
225// The T and t values
226// ==================
227// A similar method is used to generate the T and t values.
228//
229// K * log2_hi + T  must be exact.
230//
231// Smallest T,t
232// ----------
233// The smallest T,t is
234//       T                   t
235// 0x3f60040155d58800, 0x3c93bce0ce3ddd81  log(1/frcpa(1+0/256))=  +1.95503e-003
236//
237// The exponent is 0x3f6 (biased)  or -9 (true).
238// For the smallest T value, what we want is to clip the significand such that
239// when it is shifted right by 9, its lsb is in the bit for 2^-51. The 9 is the
240// specific for the first entry. In general, it is 0xffff - (biased 15-bit
241// exponent).
242
243// Independently, what we have calculated is the table value as a quad
244// precision number.
245// Table entry 1 is
246// 0 fff6 80200aaeac44ef38 338f77605fdf8000
247//
248// We store this quad precision number in a data structure that is
249//    sign:           1
250//    exponent:      15
251//    signficand_hi: 64 (includes explicit bit)
252//    signficand_lo: 49
253// Because the explicit bit is included, the significand is 113 bits.
254//
255// Consider significand_hi for table entry 1.
256//
257//
258// +-+--- ... -------+--------------------+
259// | |
260// +-+--- ... -------+--------------------+
261// 0 1               4444444455555555556666
262//                   2345678901234567890123
263//
264// Labeled as above, bit 0 is 2^0, bit 1 is 2^-1, etc.
265// Bit 42 is 2^-42. If we shift to the right by 9, the bit in
266// bit 42 goes in 51.
267//
268// So what we want to do is shift bits 43 thru 63 into significand_lo.
269// This is shifting bit 42 into bit 63, taking care to retain shifted-off bits.
270// Then shifting (just with signficaand_hi) back into bit 42.
271//
272// The shift_value is 63-42 = 21. In general, this is
273//      63 - (51 -(0xffff - 0xfff6))
274// For this example, it is
275//      63 - (51 - 9) = 63 - 42  = 21
276//
277// This means we are shifting 21 bits into significand_lo. We must maintain more
278// that a 128-bit signficand not to lose bits. So before the shift we put the
279// 128-bit significand into a 256-bit signficand and then shift.
280// The 256-bit significand has four parts: hh, hl, lh, and ll.
281//
282// Start off with
283//      hh         hl         lh         ll
284//      <64>       <49><15_0> <64_0>     <64_0>
285//
286// After shift by 21 (then return for significand_hi),
287//      <43><21_0> <21><43>   <6><58_0>  <64_0>
288//
289// Take the hh part and convert to a double. There is no rounding here.
290// The conversion is exact. The true exponent of the high part is the same as
291// the true exponent of the input quad.
292//
293// We have some 64 plus significand bits for the low part. In this example, we
294// have 70 bits. We want to round this to a double. Put them in a quad and then
295// do a quad fnorm.
296// For this example the true exponent of the low part is
297//      true_exponent_of_high - 43 = true_exponent_of_high - (64-21)
298// In general, this is
299//      true_exponent_of_high - (64 - shift_value)
300//
301//
302// Largest T,t
303// ----------
304// The largest T,t is
305// 0x3fe62643fecf9742, 0x3c9e3147684bd37d  log(1/frcpa(1+255/256))=+6.92171e-001
306//
307// Table entry 256 is
308// 0 fffe b1321ff67cba178c 51da12f4df5a0000
309//
310// The shift value is
311//      63 - (51 -(0xffff - 0xfffe)) = 13
312//
313// The true exponent of the low part is
314//      true_exponent_of_high - (64 - shift_value)
315//      -1 - (64-13) = -52
316// Biased as a double, this is 0x3cb
317//
318//
319//
320// So then lsb(T) must be >= 2^-51
321// msb(Klog2_hi) <= 2^12
322//
323//              +--------+---------+
324//              |       51 bits    | <== largest T
325//              +--------+---------+
326//              | 9 bits | 42 bits | <== smallest T
327// +------------+----------------+-+
328// |  13 bits   | 50 bits        | |
329// +------------+----------------+-+
330//
331// Note: For powf only the table of T is needed
332
333
334// Special Cases
335//==============================================================
336
337//                                   double     float
338// overflow                          error 24   30
339
340// underflow                         error 25   31
341
342// X zero  Y zero
343//  +0     +0                 +1     error 26   32
344//  -0     +0                 +1     error 26   32
345//  +0     -0                 +1     error 26   32
346//  -0     -0                 +1     error 26   32
347
348// X zero  Y negative
349//  +0     -odd integer       +inf   error 27   33  divide-by-zero
350//  -0     -odd integer       -inf   error 27   33  divide-by-zero
351//  +0     !-odd integer      +inf   error 27   33  divide-by-zero
352//  -0     !-odd integer      +inf   error 27   33  divide-by-zero
353//  +0     -inf               +inf   error 27   33  divide-by-zero
354//  -0     -inf               +inf   error 27   33  divide-by-zero
355
356// X zero  Y positve
357//  +0     +odd integer       +0
358//  -0     +odd integer       -0
359//  +0     !+odd integer      +0
360//  -0     !+odd integer      +0
361//  +0     +inf               +0
362//  -0     +inf               +0
363//  +0     Y NaN              quiet Y               invalid if Y SNaN
364//  -0     Y NaN              quiet Y               invalid if Y SNaN
365
366// X one
367//  -1     Y inf              +1
368//  -1     Y NaN              quiet Y               invalid if Y SNaN
369//  +1     Y NaN              +1                    invalid if Y SNaN
370//  +1     Y any else         +1
371
372// X -     Y not integer      QNAN   error 28   34  invalid
373
374// X NaN   Y 0                +1     error 29   35
375// X NaN   Y NaN              quiet X               invalid if X or Y SNaN
376// X NaN   Y any else         quiet X               invalid if X SNaN
377// X !+1   Y NaN              quiet Y               invalid if Y SNaN
378
379
380// X +inf  Y >0               +inf
381// X -inf  Y >0, !odd integer +inf
382// X -inf  Y >0, odd integer  -inf
383
384// X +inf  Y <0               +0
385// X -inf  Y <0, !odd integer +0
386// X -inf  Y <0, odd integer  -0
387
388// X +inf  Y =0               +1
389// X -inf  Y =0               +1
390
391// |X|<1   Y +inf             +0
392// |X|<1   Y -inf             +inf
393// |X|>1   Y +inf             +inf
394// |X|>1   Y -inf             +0
395
396// X any   Y =0               +1
397
398// Assembly macros
399//==============================================================
400
401// integer registers used
402
403pow_GR_exp_half           = r10
404pow_GR_signexp_Xm1        = r11
405pow_GR_tmp                = r11
406
407pow_GR_signexp_X          = r14
408pow_GR_17ones             = r15
409pow_GR_Fpsr               = r15
410pow_AD_P                  = r16
411pow_GR_rcs0_mask          = r16
412pow_GR_exp_2tom8          = r17
413pow_GR_rcs0               = r17
414pow_GR_sig_X              = r18
415pow_GR_10033              = r19
416pow_GR_16ones             = r20
417
418pow_AD_Tt                 = r21
419pow_GR_exp_X              = r22
420pow_AD_Q                  = r23
421pow_GR_true_exp_X         = r24
422pow_GR_y_zero             = r25
423
424pow_GR_exp_Y              = r26
425pow_AD_tbl1               = r27
426pow_AD_tbl2               = r28
427pow_GR_offset             = r29
428pow_GR_exp_Xm1            = r30
429pow_GR_xneg_yodd          = r31
430
431pow_GR_int_N              = r38
432pow_GR_index1             = r39
433pow_GR_index2             = r40
434
435pow_AD_T1                 = r41
436pow_AD_T2                 = r42
437pow_int_GR_M              = r43
438pow_GR_sig_int_Y          = r44
439pow_GR_sign_Y_Gpr         = r45
440
441pow_GR_17ones_m1          = r46
442pow_GR_one                = r47
443pow_GR_sign_Y             = r48
444pow_GR_signexp_Y_Gpr      = r49
445pow_GR_exp_Y_Gpr          = r50
446
447pow_GR_true_exp_Y_Gpr     = r51
448pow_GR_signexp_Y          = r52
449pow_GR_x_one              = r53
450pow_GR_big_pos            = r55
451
452pow_GR_big_neg            = r56
453
454GR_SAVE_B0                = r50
455GR_SAVE_GP                = r51
456GR_SAVE_PFS               = r52
457
458GR_Parameter_X            = r53
459GR_Parameter_Y            = r54
460GR_Parameter_RESULT       = r55
461pow_GR_tag                = r56
462
463
464// floating point registers used
465
466POW_B                     = f32
467POW_NORM_X                = f33
468POW_Xm1                   = f34
469POW_r1                    = f34
470
471POW_NORM_Y                = f37
472POW_Q2                    = f38
473POW_eps                   = f39
474POW_P2                    = f40
475
476POW_P0                    = f42
477POW_log2_lo               = f43
478POW_r                     = f44
479POW_Q0_half               = f45
480
481POW_tmp                   = f47
482POW_log2_hi               = f48
483POW_Q1                    = f49
484POW_P1                    = f50
485
486POW_log2_by_128_hi        = f51
487POW_inv_log2_by_128       = f52
488POW_rsq                   = f53
489POW_Yrcub                 = f54
490POW_log2_by_128_lo        = f55
491
492POW_xsq                   = f57
493POW_v2                    = f59
494POW_T                     = f60
495
496POW_RSHF                  = f62
497POW_v210                  = f63
498POW_twoV                  = f65
499
500POW_U                     = f66
501POW_G                     = f67
502POW_delta                 = f68
503POW_V                     = f70
504
505POW_p                     = f71
506POW_Z                     = f72
507POW_e3                    = f73
508POW_Z2                    = f75
509
510POW_W1                    = f77
511POW_Z3                    = f80
512
513POW_Z3sq                  = f85
514
515POW_Nfloat                = f87
516POW_f3                    = f89
517POW_q                     = f90
518
519POW_T1                    = f96
520POW_T2                    = f97
521POW_2M                    = f98
522POW_s                     = f99
523POW_f12                   = f100
524
525POW_ssq                   = f101
526POW_T1T2                  = f102
527POW_1ps                   = f103
528POW_A                     = f104
529POW_es                    = f105
530
531POW_Xp1                   = f106
532POW_int_K                 = f107
533POW_K                     = f108
534POW_f123                  = f109
535POW_Gpr                   = f110
536
537POW_Y_Gpr                 = f111
538POW_int_Y                 = f112
539POW_2Mqp1                 = f113
540
541POW_float_int_Y           = f116
542POW_ftz_urm_f8            = f117
543POW_wre_urm_f8            = f118
544POW_big_neg               = f119
545POW_big_pos               = f120
546
547// Data tables
548//==============================================================
549
550RODATA
551
552.align 16
553
554LOCAL_OBJECT_START(pow_table_P)
555data8 0x80000000000018E5, 0x0000BFFD  // P_1
556data8 0xb8aa3b295c17f0bc, 0x00004006  // inv_ln2_by_128
557//
558//
559data8 0x3FA5555555554A9E // Q_2
560data8 0x0000000000000000 // Pad
561data8 0x3FC5555555554733 // Q_1
562data8 0x43e8000000000000 // Right shift constant for exp
563data8 0xc9e3b39803f2f6af, 0x00003fb7  // ln2_by_128_lo
564LOCAL_OBJECT_END(pow_table_P)
565
566LOCAL_OBJECT_START(pow_table_Q)
567data8 0xCCCCCCCC4ED2BA7F, 0x00003FFC  // P_2
568data8 0xAAAAAAAAAAAAB505, 0x00003FFD  // P_0
569data8 0x3fe62e42fefa39e8, 0x3cccd5e4f1d9cc02 // log2 hi lo =  +6.93147e-001
570data8 0xb17217f7d1cf79ab, 0x00003ff7  // ln2_by_128_hi
571LOCAL_OBJECT_END(pow_table_Q)
572
573
574LOCAL_OBJECT_START(pow_Tt)
575data8 0x3f60040155d58800 // log(1/frcpa(1+0/256))=  +1.95503e-003
576data8 0x3f78121214586a00 // log(1/frcpa(1+1/256))=  +5.87661e-003
577data8 0x3f841929f9683200 // log(1/frcpa(1+2/256))=  +9.81362e-003
578data8 0x3f8c317384c75f00 // log(1/frcpa(1+3/256))=  +1.37662e-002
579data8 0x3f91a6b91ac73380 // log(1/frcpa(1+4/256))=  +1.72376e-002
580data8 0x3f95ba9a5d9ac000 // log(1/frcpa(1+5/256))=  +2.12196e-002
581data8 0x3f99d2a807432580 // log(1/frcpa(1+6/256))=  +2.52177e-002
582data8 0x3f9d6b2725979800 // log(1/frcpa(1+7/256))=  +2.87291e-002
583data8 0x3fa0c58fa19dfa80 // log(1/frcpa(1+8/256))=  +3.27573e-002
584data8 0x3fa2954c78cbce00 // log(1/frcpa(1+9/256))=  +3.62953e-002
585data8 0x3fa4a94d2da96c40 // log(1/frcpa(1+10/256))=  +4.03542e-002
586data8 0x3fa67c94f2d4bb40 // log(1/frcpa(1+11/256))=  +4.39192e-002
587data8 0x3fa85188b630f040 // log(1/frcpa(1+12/256))=  +4.74971e-002
588data8 0x3faa6b8abe73af40 // log(1/frcpa(1+13/256))=  +5.16017e-002
589data8 0x3fac441e06f72a80 // log(1/frcpa(1+14/256))=  +5.52072e-002
590data8 0x3fae1e6713606d00 // log(1/frcpa(1+15/256))=  +5.88257e-002
591data8 0x3faffa6911ab9300 // log(1/frcpa(1+16/256))=  +6.24574e-002
592data8 0x3fb0ec139c5da600 // log(1/frcpa(1+17/256))=  +6.61022e-002
593data8 0x3fb1dbd2643d1900 // log(1/frcpa(1+18/256))=  +6.97605e-002
594data8 0x3fb2cc7284fe5f00 // log(1/frcpa(1+19/256))=  +7.34321e-002
595data8 0x3fb3bdf5a7d1ee60 // log(1/frcpa(1+20/256))=  +7.71173e-002
596data8 0x3fb4b05d7aa012e0 // log(1/frcpa(1+21/256))=  +8.08161e-002
597data8 0x3fb580db7ceb5700 // log(1/frcpa(1+22/256))=  +8.39975e-002
598data8 0x3fb674f089365a60 // log(1/frcpa(1+23/256))=  +8.77219e-002
599data8 0x3fb769ef2c6b5680 // log(1/frcpa(1+24/256))=  +9.14602e-002
600data8 0x3fb85fd927506a40 // log(1/frcpa(1+25/256))=  +9.52125e-002
601data8 0x3fb9335e5d594980 // log(1/frcpa(1+26/256))=  +9.84401e-002
602data8 0x3fba2b0220c8e5e0 // log(1/frcpa(1+27/256))=  +1.02219e-001
603data8 0x3fbb0004ac1a86a0 // log(1/frcpa(1+28/256))=  +1.05469e-001
604data8 0x3fbbf968769fca00 // log(1/frcpa(1+29/256))=  +1.09274e-001
605data8 0x3fbccfedbfee13a0 // log(1/frcpa(1+30/256))=  +1.12548e-001
606data8 0x3fbda727638446a0 // log(1/frcpa(1+31/256))=  +1.15832e-001
607data8 0x3fbea3257fe10f60 // log(1/frcpa(1+32/256))=  +1.19677e-001
608data8 0x3fbf7be9fedbfde0 // log(1/frcpa(1+33/256))=  +1.22985e-001
609data8 0x3fc02ab352ff25f0 // log(1/frcpa(1+34/256))=  +1.26303e-001
610data8 0x3fc097ce579d2040 // log(1/frcpa(1+35/256))=  +1.29633e-001
611data8 0x3fc1178e8227e470 // log(1/frcpa(1+36/256))=  +1.33531e-001
612data8 0x3fc185747dbecf30 // log(1/frcpa(1+37/256))=  +1.36885e-001
613data8 0x3fc1f3b925f25d40 // log(1/frcpa(1+38/256))=  +1.40250e-001
614data8 0x3fc2625d1e6ddf50 // log(1/frcpa(1+39/256))=  +1.43627e-001
615data8 0x3fc2d1610c868130 // log(1/frcpa(1+40/256))=  +1.47015e-001
616data8 0x3fc340c597411420 // log(1/frcpa(1+41/256))=  +1.50414e-001
617data8 0x3fc3b08b6757f2a0 // log(1/frcpa(1+42/256))=  +1.53825e-001
618data8 0x3fc40dfb08378000 // log(1/frcpa(1+43/256))=  +1.56677e-001
619data8 0x3fc47e74e8ca5f70 // log(1/frcpa(1+44/256))=  +1.60109e-001
620data8 0x3fc4ef51f6466de0 // log(1/frcpa(1+45/256))=  +1.63553e-001
621data8 0x3fc56092e02ba510 // log(1/frcpa(1+46/256))=  +1.67010e-001
622data8 0x3fc5d23857cd74d0 // log(1/frcpa(1+47/256))=  +1.70478e-001
623data8 0x3fc6313a37335d70 // log(1/frcpa(1+48/256))=  +1.73377e-001
624data8 0x3fc6a399dabbd380 // log(1/frcpa(1+49/256))=  +1.76868e-001
625data8 0x3fc70337dd3ce410 // log(1/frcpa(1+50/256))=  +1.79786e-001
626data8 0x3fc77654128f6120 // log(1/frcpa(1+51/256))=  +1.83299e-001
627data8 0x3fc7e9d82a0b0220 // log(1/frcpa(1+52/256))=  +1.86824e-001
628data8 0x3fc84a6b759f5120 // log(1/frcpa(1+53/256))=  +1.89771e-001
629data8 0x3fc8ab47d5f5a300 // log(1/frcpa(1+54/256))=  +1.92727e-001
630data8 0x3fc91fe490965810 // log(1/frcpa(1+55/256))=  +1.96286e-001
631data8 0x3fc981634011aa70 // log(1/frcpa(1+56/256))=  +1.99261e-001
632data8 0x3fc9f6c407089660 // log(1/frcpa(1+57/256))=  +2.02843e-001
633data8 0x3fca58e729348f40 // log(1/frcpa(1+58/256))=  +2.05838e-001
634data8 0x3fcabb55c31693a0 // log(1/frcpa(1+59/256))=  +2.08842e-001
635data8 0x3fcb1e104919efd0 // log(1/frcpa(1+60/256))=  +2.11855e-001
636data8 0x3fcb94ee93e367c0 // log(1/frcpa(1+61/256))=  +2.15483e-001
637data8 0x3fcbf851c0675550 // log(1/frcpa(1+62/256))=  +2.18516e-001
638data8 0x3fcc5c0254bf23a0 // log(1/frcpa(1+63/256))=  +2.21558e-001
639data8 0x3fccc000c9db3c50 // log(1/frcpa(1+64/256))=  +2.24609e-001
640data8 0x3fcd244d99c85670 // log(1/frcpa(1+65/256))=  +2.27670e-001
641data8 0x3fcd88e93fb2f450 // log(1/frcpa(1+66/256))=  +2.30741e-001
642data8 0x3fcdedd437eaef00 // log(1/frcpa(1+67/256))=  +2.33820e-001
643data8 0x3fce530effe71010 // log(1/frcpa(1+68/256))=  +2.36910e-001
644data8 0x3fceb89a1648b970 // log(1/frcpa(1+69/256))=  +2.40009e-001
645data8 0x3fcf1e75fadf9bd0 // log(1/frcpa(1+70/256))=  +2.43117e-001
646data8 0x3fcf84a32ead7c30 // log(1/frcpa(1+71/256))=  +2.46235e-001
647data8 0x3fcfeb2233ea07c0 // log(1/frcpa(1+72/256))=  +2.49363e-001
648data8 0x3fd028f9c7035c18 // log(1/frcpa(1+73/256))=  +2.52501e-001
649data8 0x3fd05c8be0d96358 // log(1/frcpa(1+74/256))=  +2.55649e-001
650data8 0x3fd085eb8f8ae790 // log(1/frcpa(1+75/256))=  +2.58174e-001
651data8 0x3fd0b9c8e32d1910 // log(1/frcpa(1+76/256))=  +2.61339e-001
652data8 0x3fd0edd060b78080 // log(1/frcpa(1+77/256))=  +2.64515e-001
653data8 0x3fd122024cf00638 // log(1/frcpa(1+78/256))=  +2.67701e-001
654data8 0x3fd14be2927aecd0 // log(1/frcpa(1+79/256))=  +2.70257e-001
655data8 0x3fd180618ef18ad8 // log(1/frcpa(1+80/256))=  +2.73461e-001
656data8 0x3fd1b50bbe2fc638 // log(1/frcpa(1+81/256))=  +2.76675e-001
657data8 0x3fd1df4cc7cf2428 // log(1/frcpa(1+82/256))=  +2.79254e-001
658data8 0x3fd214456d0eb8d0 // log(1/frcpa(1+83/256))=  +2.82487e-001
659data8 0x3fd23ec5991eba48 // log(1/frcpa(1+84/256))=  +2.85081e-001
660data8 0x3fd2740d9f870af8 // log(1/frcpa(1+85/256))=  +2.88333e-001
661data8 0x3fd29ecdabcdfa00 // log(1/frcpa(1+86/256))=  +2.90943e-001
662data8 0x3fd2d46602adcce8 // log(1/frcpa(1+87/256))=  +2.94214e-001
663data8 0x3fd2ff66b04ea9d0 // log(1/frcpa(1+88/256))=  +2.96838e-001
664data8 0x3fd335504b355a30 // log(1/frcpa(1+89/256))=  +3.00129e-001
665data8 0x3fd360925ec44f58 // log(1/frcpa(1+90/256))=  +3.02769e-001
666data8 0x3fd38bf1c3337e70 // log(1/frcpa(1+91/256))=  +3.05417e-001
667data8 0x3fd3c25277333180 // log(1/frcpa(1+92/256))=  +3.08735e-001
668data8 0x3fd3edf463c16838 // log(1/frcpa(1+93/256))=  +3.11399e-001
669data8 0x3fd419b423d5e8c0 // log(1/frcpa(1+94/256))=  +3.14069e-001
670data8 0x3fd44591e0539f48 // log(1/frcpa(1+95/256))=  +3.16746e-001
671data8 0x3fd47c9175b6f0a8 // log(1/frcpa(1+96/256))=  +3.20103e-001
672data8 0x3fd4a8b341552b08 // log(1/frcpa(1+97/256))=  +3.22797e-001
673data8 0x3fd4d4f390890198 // log(1/frcpa(1+98/256))=  +3.25498e-001
674data8 0x3fd501528da1f960 // log(1/frcpa(1+99/256))=  +3.28206e-001
675data8 0x3fd52dd06347d4f0 // log(1/frcpa(1+100/256))=  +3.30921e-001
676data8 0x3fd55a6d3c7b8a88 // log(1/frcpa(1+101/256))=  +3.33644e-001
677data8 0x3fd5925d2b112a58 // log(1/frcpa(1+102/256))=  +3.37058e-001
678data8 0x3fd5bf406b543db0 // log(1/frcpa(1+103/256))=  +3.39798e-001
679data8 0x3fd5ec433d5c35a8 // log(1/frcpa(1+104/256))=  +3.42545e-001
680data8 0x3fd61965cdb02c18 // log(1/frcpa(1+105/256))=  +3.45300e-001
681data8 0x3fd646a84935b2a0 // log(1/frcpa(1+106/256))=  +3.48063e-001
682data8 0x3fd6740add31de90 // log(1/frcpa(1+107/256))=  +3.50833e-001
683data8 0x3fd6a18db74a58c0 // log(1/frcpa(1+108/256))=  +3.53610e-001
684data8 0x3fd6cf31058670e8 // log(1/frcpa(1+109/256))=  +3.56396e-001
685data8 0x3fd6f180e852f0b8 // log(1/frcpa(1+110/256))=  +3.58490e-001
686data8 0x3fd71f5d71b894e8 // log(1/frcpa(1+111/256))=  +3.61289e-001
687data8 0x3fd74d5aefd66d58 // log(1/frcpa(1+112/256))=  +3.64096e-001
688data8 0x3fd77b79922bd378 // log(1/frcpa(1+113/256))=  +3.66911e-001
689data8 0x3fd7a9b9889f19e0 // log(1/frcpa(1+114/256))=  +3.69734e-001
690data8 0x3fd7d81b037eb6a0 // log(1/frcpa(1+115/256))=  +3.72565e-001
691data8 0x3fd8069e33827230 // log(1/frcpa(1+116/256))=  +3.75404e-001
692data8 0x3fd82996d3ef8bc8 // log(1/frcpa(1+117/256))=  +3.77538e-001
693data8 0x3fd85855776dcbf8 // log(1/frcpa(1+118/256))=  +3.80391e-001
694data8 0x3fd8873658327cc8 // log(1/frcpa(1+119/256))=  +3.83253e-001
695data8 0x3fd8aa75973ab8c8 // log(1/frcpa(1+120/256))=  +3.85404e-001
696data8 0x3fd8d992dc8824e0 // log(1/frcpa(1+121/256))=  +3.88280e-001
697data8 0x3fd908d2ea7d9510 // log(1/frcpa(1+122/256))=  +3.91164e-001
698data8 0x3fd92c59e79c0e50 // log(1/frcpa(1+123/256))=  +3.93332e-001
699data8 0x3fd95bd750ee3ed0 // log(1/frcpa(1+124/256))=  +3.96231e-001
700data8 0x3fd98b7811a3ee58 // log(1/frcpa(1+125/256))=  +3.99138e-001
701data8 0x3fd9af47f33d4068 // log(1/frcpa(1+126/256))=  +4.01323e-001
702data8 0x3fd9df270c1914a0 // log(1/frcpa(1+127/256))=  +4.04245e-001
703data8 0x3fda0325ed14fda0 // log(1/frcpa(1+128/256))=  +4.06442e-001
704data8 0x3fda33440224fa78 // log(1/frcpa(1+129/256))=  +4.09379e-001
705data8 0x3fda57725e80c380 // log(1/frcpa(1+130/256))=  +4.11587e-001
706data8 0x3fda87d0165dd198 // log(1/frcpa(1+131/256))=  +4.14539e-001
707data8 0x3fdaac2e6c03f890 // log(1/frcpa(1+132/256))=  +4.16759e-001
708data8 0x3fdadccc6fdf6a80 // log(1/frcpa(1+133/256))=  +4.19726e-001
709data8 0x3fdb015b3eb1e790 // log(1/frcpa(1+134/256))=  +4.21958e-001
710data8 0x3fdb323a3a635948 // log(1/frcpa(1+135/256))=  +4.24941e-001
711data8 0x3fdb56fa04462908 // log(1/frcpa(1+136/256))=  +4.27184e-001
712data8 0x3fdb881aa659bc90 // log(1/frcpa(1+137/256))=  +4.30182e-001
713data8 0x3fdbad0bef3db160 // log(1/frcpa(1+138/256))=  +4.32437e-001
714data8 0x3fdbd21297781c28 // log(1/frcpa(1+139/256))=  +4.34697e-001
715data8 0x3fdc039236f08818 // log(1/frcpa(1+140/256))=  +4.37718e-001
716data8 0x3fdc28cb1e4d32f8 // log(1/frcpa(1+141/256))=  +4.39990e-001
717data8 0x3fdc4e19b84723c0 // log(1/frcpa(1+142/256))=  +4.42267e-001
718data8 0x3fdc7ff9c74554c8 // log(1/frcpa(1+143/256))=  +4.45311e-001
719data8 0x3fdca57b64e9db00 // log(1/frcpa(1+144/256))=  +4.47600e-001
720data8 0x3fdccb130a5ceba8 // log(1/frcpa(1+145/256))=  +4.49895e-001
721data8 0x3fdcf0c0d18f3268 // log(1/frcpa(1+146/256))=  +4.52194e-001
722data8 0x3fdd232075b5a200 // log(1/frcpa(1+147/256))=  +4.55269e-001
723data8 0x3fdd490246defa68 // log(1/frcpa(1+148/256))=  +4.57581e-001
724data8 0x3fdd6efa918d25c8 // log(1/frcpa(1+149/256))=  +4.59899e-001
725data8 0x3fdd9509707ae528 // log(1/frcpa(1+150/256))=  +4.62221e-001
726data8 0x3fddbb2efe92c550 // log(1/frcpa(1+151/256))=  +4.64550e-001
727data8 0x3fddee2f3445e4a8 // log(1/frcpa(1+152/256))=  +4.67663e-001
728data8 0x3fde148a1a2726c8 // log(1/frcpa(1+153/256))=  +4.70004e-001
729data8 0x3fde3afc0a49ff38 // log(1/frcpa(1+154/256))=  +4.72350e-001
730data8 0x3fde6185206d5168 // log(1/frcpa(1+155/256))=  +4.74702e-001
731data8 0x3fde882578823d50 // log(1/frcpa(1+156/256))=  +4.77060e-001
732data8 0x3fdeaedd2eac9908 // log(1/frcpa(1+157/256))=  +4.79423e-001
733data8 0x3fded5ac5f436be0 // log(1/frcpa(1+158/256))=  +4.81792e-001
734data8 0x3fdefc9326d16ab8 // log(1/frcpa(1+159/256))=  +4.84166e-001
735data8 0x3fdf2391a21575f8 // log(1/frcpa(1+160/256))=  +4.86546e-001
736data8 0x3fdf4aa7ee031928 // log(1/frcpa(1+161/256))=  +4.88932e-001
737data8 0x3fdf71d627c30bb0 // log(1/frcpa(1+162/256))=  +4.91323e-001
738data8 0x3fdf991c6cb3b378 // log(1/frcpa(1+163/256))=  +4.93720e-001
739data8 0x3fdfc07ada69a908 // log(1/frcpa(1+164/256))=  +4.96123e-001
740data8 0x3fdfe7f18eb03d38 // log(1/frcpa(1+165/256))=  +4.98532e-001
741data8 0x3fe007c053c5002c // log(1/frcpa(1+166/256))=  +5.00946e-001
742data8 0x3fe01b942198a5a0 // log(1/frcpa(1+167/256))=  +5.03367e-001
743data8 0x3fe02f74400c64e8 // log(1/frcpa(1+168/256))=  +5.05793e-001
744data8 0x3fe04360be7603ac // log(1/frcpa(1+169/256))=  +5.08225e-001
745data8 0x3fe05759ac47fe30 // log(1/frcpa(1+170/256))=  +5.10663e-001
746data8 0x3fe06b5f1911cf50 // log(1/frcpa(1+171/256))=  +5.13107e-001
747data8 0x3fe078bf0533c568 // log(1/frcpa(1+172/256))=  +5.14740e-001
748data8 0x3fe08cd9687e7b0c // log(1/frcpa(1+173/256))=  +5.17194e-001
749data8 0x3fe0a10074cf9018 // log(1/frcpa(1+174/256))=  +5.19654e-001
750data8 0x3fe0b5343a234474 // log(1/frcpa(1+175/256))=  +5.22120e-001
751data8 0x3fe0c974c89431cc // log(1/frcpa(1+176/256))=  +5.24592e-001
752data8 0x3fe0ddc2305b9884 // log(1/frcpa(1+177/256))=  +5.27070e-001
753data8 0x3fe0eb524bafc918 // log(1/frcpa(1+178/256))=  +5.28726e-001
754data8 0x3fe0ffb54213a474 // log(1/frcpa(1+179/256))=  +5.31214e-001
755data8 0x3fe114253da97d9c // log(1/frcpa(1+180/256))=  +5.33709e-001
756data8 0x3fe128a24f1d9afc // log(1/frcpa(1+181/256))=  +5.36210e-001
757data8 0x3fe1365252bf0864 // log(1/frcpa(1+182/256))=  +5.37881e-001
758data8 0x3fe14ae558b4a92c // log(1/frcpa(1+183/256))=  +5.40393e-001
759data8 0x3fe15f85a19c7658 // log(1/frcpa(1+184/256))=  +5.42910e-001
760data8 0x3fe16d4d38c119f8 // log(1/frcpa(1+185/256))=  +5.44592e-001
761data8 0x3fe18203c20dd130 // log(1/frcpa(1+186/256))=  +5.47121e-001
762data8 0x3fe196c7bc4b1f38 // log(1/frcpa(1+187/256))=  +5.49656e-001
763data8 0x3fe1a4a738b7a33c // log(1/frcpa(1+188/256))=  +5.51349e-001
764data8 0x3fe1b981c0c9653c // log(1/frcpa(1+189/256))=  +5.53895e-001
765data8 0x3fe1ce69e8bb1068 // log(1/frcpa(1+190/256))=  +5.56447e-001
766data8 0x3fe1dc619de06944 // log(1/frcpa(1+191/256))=  +5.58152e-001
767data8 0x3fe1f160a2ad0da0 // log(1/frcpa(1+192/256))=  +5.60715e-001
768data8 0x3fe2066d7740737c // log(1/frcpa(1+193/256))=  +5.63285e-001
769data8 0x3fe2147dba47a390 // log(1/frcpa(1+194/256))=  +5.65001e-001
770data8 0x3fe229a1bc5ebac0 // log(1/frcpa(1+195/256))=  +5.67582e-001
771data8 0x3fe237c1841a502c // log(1/frcpa(1+196/256))=  +5.69306e-001
772data8 0x3fe24cfce6f80d98 // log(1/frcpa(1+197/256))=  +5.71898e-001
773data8 0x3fe25b2c55cd5760 // log(1/frcpa(1+198/256))=  +5.73630e-001
774data8 0x3fe2707f4d5f7c40 // log(1/frcpa(1+199/256))=  +5.76233e-001
775data8 0x3fe285e0842ca380 // log(1/frcpa(1+200/256))=  +5.78842e-001
776data8 0x3fe294294708b770 // log(1/frcpa(1+201/256))=  +5.80586e-001
777data8 0x3fe2a9a2670aff0c // log(1/frcpa(1+202/256))=  +5.83207e-001
778data8 0x3fe2b7fb2c8d1cc0 // log(1/frcpa(1+203/256))=  +5.84959e-001
779data8 0x3fe2c65a6395f5f4 // log(1/frcpa(1+204/256))=  +5.86713e-001
780data8 0x3fe2dbf557b0df40 // log(1/frcpa(1+205/256))=  +5.89350e-001
781data8 0x3fe2ea64c3f97654 // log(1/frcpa(1+206/256))=  +5.91113e-001
782data8 0x3fe3001823684d70 // log(1/frcpa(1+207/256))=  +5.93762e-001
783data8 0x3fe30e97e9a8b5cc // log(1/frcpa(1+208/256))=  +5.95531e-001
784data8 0x3fe32463ebdd34e8 // log(1/frcpa(1+209/256))=  +5.98192e-001
785data8 0x3fe332f4314ad794 // log(1/frcpa(1+210/256))=  +5.99970e-001
786data8 0x3fe348d90e7464cc // log(1/frcpa(1+211/256))=  +6.02643e-001
787data8 0x3fe35779f8c43d6c // log(1/frcpa(1+212/256))=  +6.04428e-001
788data8 0x3fe36621961a6a98 // log(1/frcpa(1+213/256))=  +6.06217e-001
789data8 0x3fe37c299f3c3668 // log(1/frcpa(1+214/256))=  +6.08907e-001
790data8 0x3fe38ae2171976e4 // log(1/frcpa(1+215/256))=  +6.10704e-001
791data8 0x3fe399a157a603e4 // log(1/frcpa(1+216/256))=  +6.12504e-001
792data8 0x3fe3afccfe77b9d0 // log(1/frcpa(1+217/256))=  +6.15210e-001
793data8 0x3fe3be9d503533b4 // log(1/frcpa(1+218/256))=  +6.17018e-001
794data8 0x3fe3cd7480b4a8a0 // log(1/frcpa(1+219/256))=  +6.18830e-001
795data8 0x3fe3e3c43918f76c // log(1/frcpa(1+220/256))=  +6.21554e-001
796data8 0x3fe3f2acb27ed6c4 // log(1/frcpa(1+221/256))=  +6.23373e-001
797data8 0x3fe4019c2125ca90 // log(1/frcpa(1+222/256))=  +6.25197e-001
798data8 0x3fe4181061389720 // log(1/frcpa(1+223/256))=  +6.27937e-001
799data8 0x3fe42711518df544 // log(1/frcpa(1+224/256))=  +6.29769e-001
800data8 0x3fe436194e12b6bc // log(1/frcpa(1+225/256))=  +6.31604e-001
801data8 0x3fe445285d68ea68 // log(1/frcpa(1+226/256))=  +6.33442e-001
802data8 0x3fe45bcc464c8938 // log(1/frcpa(1+227/256))=  +6.36206e-001
803data8 0x3fe46aed21f117fc // log(1/frcpa(1+228/256))=  +6.38053e-001
804data8 0x3fe47a1527e8a2d0 // log(1/frcpa(1+229/256))=  +6.39903e-001
805data8 0x3fe489445efffcc8 // log(1/frcpa(1+230/256))=  +6.41756e-001
806data8 0x3fe4a018bcb69834 // log(1/frcpa(1+231/256))=  +6.44543e-001
807data8 0x3fe4af5a0c9d65d4 // log(1/frcpa(1+232/256))=  +6.46405e-001
808data8 0x3fe4bea2a5bdbe84 // log(1/frcpa(1+233/256))=  +6.48271e-001
809data8 0x3fe4cdf28f10ac44 // log(1/frcpa(1+234/256))=  +6.50140e-001
810data8 0x3fe4dd49cf994058 // log(1/frcpa(1+235/256))=  +6.52013e-001
811data8 0x3fe4eca86e64a680 // log(1/frcpa(1+236/256))=  +6.53889e-001
812data8 0x3fe503c43cd8eb68 // log(1/frcpa(1+237/256))=  +6.56710e-001
813data8 0x3fe513356667fc54 // log(1/frcpa(1+238/256))=  +6.58595e-001
814data8 0x3fe522ae0738a3d4 // log(1/frcpa(1+239/256))=  +6.60483e-001
815data8 0x3fe5322e26867854 // log(1/frcpa(1+240/256))=  +6.62376e-001
816data8 0x3fe541b5cb979808 // log(1/frcpa(1+241/256))=  +6.64271e-001
817data8 0x3fe55144fdbcbd60 // log(1/frcpa(1+242/256))=  +6.66171e-001
818data8 0x3fe560dbc45153c4 // log(1/frcpa(1+243/256))=  +6.68074e-001
819data8 0x3fe5707a26bb8c64 // log(1/frcpa(1+244/256))=  +6.69980e-001
820data8 0x3fe587f60ed5b8fc // log(1/frcpa(1+245/256))=  +6.72847e-001
821data8 0x3fe597a7977c8f30 // log(1/frcpa(1+246/256))=  +6.74763e-001
822data8 0x3fe5a760d634bb88 // log(1/frcpa(1+247/256))=  +6.76682e-001
823data8 0x3fe5b721d295f10c // log(1/frcpa(1+248/256))=  +6.78605e-001
824data8 0x3fe5c6ea94431ef8 // log(1/frcpa(1+249/256))=  +6.80532e-001
825data8 0x3fe5d6bb22ea86f4 // log(1/frcpa(1+250/256))=  +6.82462e-001
826data8 0x3fe5e6938645d38c // log(1/frcpa(1+251/256))=  +6.84397e-001
827data8 0x3fe5f673c61a2ed0 // log(1/frcpa(1+252/256))=  +6.86335e-001
828data8 0x3fe6065bea385924 // log(1/frcpa(1+253/256))=  +6.88276e-001
829data8 0x3fe6164bfa7cc068 // log(1/frcpa(1+254/256))=  +6.90222e-001
830data8 0x3fe62643fecf9740 // log(1/frcpa(1+255/256))=  +6.92171e-001
831LOCAL_OBJECT_END(pow_Tt)
832
833
834// Table 1 is 2^(index_1/128) where
835// index_1 goes from 0 to 15
836LOCAL_OBJECT_START(pow_tbl1)
837data8 0x8000000000000000 , 0x00003FFF
838data8 0x80B1ED4FD999AB6C , 0x00003FFF
839data8 0x8164D1F3BC030773 , 0x00003FFF
840data8 0x8218AF4373FC25EC , 0x00003FFF
841data8 0x82CD8698AC2BA1D7 , 0x00003FFF
842data8 0x8383594EEFB6EE37 , 0x00003FFF
843data8 0x843A28C3ACDE4046 , 0x00003FFF
844data8 0x84F1F656379C1A29 , 0x00003FFF
845data8 0x85AAC367CC487B15 , 0x00003FFF
846data8 0x8664915B923FBA04 , 0x00003FFF
847data8 0x871F61969E8D1010 , 0x00003FFF
848data8 0x87DB357FF698D792 , 0x00003FFF
849data8 0x88980E8092DA8527 , 0x00003FFF
850data8 0x8955EE03618E5FDD , 0x00003FFF
851data8 0x8A14D575496EFD9A , 0x00003FFF
852data8 0x8AD4C6452C728924 , 0x00003FFF
853LOCAL_OBJECT_END(pow_tbl1)
854
855
856// Table 2 is 2^(index_1/8) where
857// index_2 goes from 0 to 7
858LOCAL_OBJECT_START(pow_tbl2)
859data8 0x8000000000000000 , 0x00003FFF
860data8 0x8B95C1E3EA8BD6E7 , 0x00003FFF
861data8 0x9837F0518DB8A96F , 0x00003FFF
862data8 0xA5FED6A9B15138EA , 0x00003FFF
863data8 0xB504F333F9DE6484 , 0x00003FFF
864data8 0xC5672A115506DADD , 0x00003FFF
865data8 0xD744FCCAD69D6AF4 , 0x00003FFF
866data8 0xEAC0C6E7DD24392F , 0x00003FFF
867LOCAL_OBJECT_END(pow_tbl2)
868
869.section .text
870WEAK_LIBM_ENTRY(powf)
871
872// Get exponent of x.  Will be used to calculate K.
873{ .mfi
874          getf.exp     pow_GR_signexp_X = f8
875          fms.s1 POW_Xm1 = f8,f1,f1     // Will be used for r1 if x>0
876          mov           pow_GR_17ones   = 0x1FFFF
877}
878{ .mfi
879          addl          pow_AD_P        = @ltoff(pow_table_P), gp
880          fma.s1 POW_Xp1 = f8,f1,f1     // Will be used for r1 if x<0
881          nop.i 999
882}
883;;
884
885// Get significand of x.  Will be used to get index to fetch T, Tt.
886{ .mfi
887          getf.sig      pow_GR_sig_X    = f8
888          frcpa.s1      POW_B, p6       = f1,f8
889          mov           pow_GR_exp_half = 0xFFFE   // Exponent for 0.5
890}
891{ .mfi
892          ld8 pow_AD_P = [pow_AD_P]
893          fma.s1        POW_NORM_X      = f8,f1,f0
894          mov          pow_GR_exp_2tom8 = 0xFFF7
895}
896;;
897
898// DOUBLE 0x10033  exponent limit at which y is an integer
899{ .mfi
900          nop.m 999
901          fcmp.lt.s1 p8,p9 = f8, f0     // Test for x<0
902          addl pow_GR_10033             = 0x10033, r0
903}
904{ .mfi
905          mov           pow_GR_16ones   = 0xFFFF
906          fma.s1        POW_NORM_Y      = f9,f1,f0
907          nop.i 999
908}
909;;
910
911// p13 = TRUE ==> X is unorm
912{ .mfi
913          setf.exp      POW_Q0_half     = pow_GR_exp_half  // Form 0.5
914          fclass.m  p13,p0              = f8, 0x0b  // Test for x unorm
915          adds          pow_AD_Tt       = pow_Tt - pow_table_P,  pow_AD_P
916}
917{ .mfi
918          adds          pow_AD_Q        = pow_table_Q - pow_table_P,  pow_AD_P
919          nop.f 999
920          nop.i 999
921}
922;;
923
924// p14 = TRUE ==> X is ZERO
925{ .mfi
926          ldfe          POW_P2          = [pow_AD_Q], 16
927          fclass.m  p14,p0              = f8, 0x07
928          nop.i 999
929}
930// Note POW_Xm1 and POW_r1 are used interchangably
931{ .mfb
932          nop.m 999
933(p8)      fnma.s1        POW_Xm1        = POW_Xp1,f1,f0
934(p13)     br.cond.spnt POW_X_DENORM
935}
936;;
937
938// Continue normal and denormal paths here
939POW_COMMON:
940// p11 = TRUE ==> Y is a NAN
941{ .mfi
942          and           pow_GR_exp_X    = pow_GR_signexp_X, pow_GR_17ones
943          fclass.m  p11,p0              = f9, 0xc3
944          nop.i 999
945}
946{ .mfi
947          nop.m 999
948          fms.s1        POW_r           = POW_B, POW_NORM_X,f1
949          mov pow_GR_y_zero = 0
950}
951;;
952
953// Get exponent of |x|-1 to use in comparison to 2^-8
954{ .mmi
955          getf.exp  pow_GR_signexp_Xm1  = POW_Xm1
956          sub       pow_GR_true_exp_X   = pow_GR_exp_X, pow_GR_16ones
957          extr.u        pow_GR_offset   = pow_GR_sig_X, 55, 8
958}
959;;
960
961{ .mfi
962          alloc         r32=ar.pfs,2,19,4,0
963          fcvt.fx.s1   POW_int_Y        = POW_NORM_Y
964          shladd pow_AD_Tt = pow_GR_offset, 3, pow_AD_Tt
965}
966{ .mfi
967          setf.sig POW_int_K            = pow_GR_true_exp_X
968          nop.f 999
969          nop.i 999
970}
971;;
972
973// p12 = TRUE if Y is ZERO
974// Compute xsq to decide later if |x|=1
975{ .mfi
976          ldfe          POW_P1          = [pow_AD_P], 16
977          fclass.m      p12,p0          = f9, 0x07
978          nop.i 999
979}
980{ .mfb
981          ldfe          POW_P0          = [pow_AD_Q], 16
982          fma.s1        POW_xsq = POW_NORM_X, POW_NORM_X, f0
983(p11)     br.cond.spnt  POW_Y_NAN       // Branch if y=nan
984}
985;;
986
987{ .mmf
988          getf.exp  pow_GR_signexp_Y    = POW_NORM_Y
989          ldfd  POW_T                   = [pow_AD_Tt]
990          fma.s1        POW_rsq         = POW_r, POW_r,f0
991}
992;;
993
994// p11 = TRUE ==> X is a NAN
995{ .mfi
996          ldfpd         POW_log2_hi, POW_log2_lo  = [pow_AD_Q], 16
997          fclass.m      p11,p0          = POW_NORM_X, 0xc3
998          nop.i 999
999}
1000{ .mfi
1001          ldfe          POW_inv_log2_by_128 = [pow_AD_P], 16
1002          fma.s1 POW_delta              = f0,f0,f0 // delta=0 in case |x| near 1
1003(p12)     mov pow_GR_y_zero = 1
1004}
1005;;
1006
1007{ .mfi
1008          ldfd   POW_Q2                 = [pow_AD_P], 16
1009          fnma.s1 POW_twoV              = POW_r, POW_Q0_half,f1
1010          and       pow_GR_exp_Xm1      = pow_GR_signexp_Xm1, pow_GR_17ones
1011}
1012{ .mfi
1013          nop.m 999
1014          fma.s1 POW_U                  = POW_NORM_Y,POW_r,f0
1015          nop.i 999
1016}
1017;;
1018
1019// Determine if we will use the |x| near 1 path (p6) or normal path (p7)
1020{ .mfi
1021          nop.m 999
1022          fcvt.xf POW_K                 = POW_int_K
1023          cmp.lt p6,p7                  = pow_GR_exp_Xm1, pow_GR_exp_2tom8
1024}
1025{ .mfb
1026          nop.m 999
1027          fma.s1 POW_G                  = f0,f0,f0  // G=0 in case |x| near 1
1028(p11)     br.cond.spnt  POW_X_NAN       // Branch if x=nan and y not nan
1029}
1030;;
1031
1032// If on the x near 1 path, assign r1 to r
1033{ .mfi
1034          ldfpd  POW_Q1, POW_RSHF       = [pow_AD_P], 16
1035(p6)      fma.s1    POW_r               = POW_r1, f1, f0
1036          nop.i 999
1037}
1038{ .mfb
1039          nop.m 999
1040(p6)      fma.s1    POW_rsq             = POW_r1, POW_r1, f0
1041(p14)     br.cond.spnt POW_X_0          // Branch if x zero and y not nan
1042}
1043;;
1044
1045{ .mfi
1046          getf.sig pow_GR_sig_int_Y     = POW_int_Y
1047(p6)      fnma.s1 POW_twoV              = POW_r1, POW_Q0_half,f1
1048          and pow_GR_exp_Y              = pow_GR_signexp_Y, pow_GR_17ones
1049}
1050{ .mfb
1051          andcm pow_GR_sign_Y           = pow_GR_signexp_Y, pow_GR_17ones
1052(p6)      fma.s1 POW_U                  = POW_NORM_Y,POW_r1,f0
1053(p12)     br.cond.spnt POW_Y_0   // Branch if y=zero, x not zero or nan
1054}
1055;;
1056
1057{ .mfi
1058          ldfe      POW_log2_by_128_lo  = [pow_AD_P], 16
1059(p7)      fma.s1 POW_Z2                 = POW_twoV, POW_U, f0
1060          nop.i 999
1061}
1062{ .mfi
1063          ldfe      POW_log2_by_128_hi  = [pow_AD_Q], 16
1064          nop.f 999
1065          nop.i 999
1066}
1067;;
1068
1069{ .mfi
1070          nop.m 999
1071          fcvt.xf   POW_float_int_Y     = POW_int_Y
1072          nop.i 999
1073}
1074{ .mfi
1075          nop.m 999
1076(p7)      fma.s1 POW_G                  = POW_K, POW_log2_hi, POW_T
1077          adds          pow_AD_tbl1     = pow_tbl1 - pow_Tt,  pow_AD_Q
1078}
1079;;
1080
1081// p11 = TRUE ==> X is NEGATIVE but not inf
1082{ .mfi
1083          nop.m 999
1084          fclass.m  p11,p0              = POW_NORM_X, 0x1a
1085          nop.i 999
1086}
1087{ .mfi
1088          nop.m 999
1089(p7)      fma.s1 POW_delta              = POW_K, POW_log2_lo, f0
1090          adds pow_AD_tbl2              = pow_tbl2 - pow_tbl1,  pow_AD_tbl1
1091}
1092;;
1093
1094{ .mfi
1095          nop.m 999
1096(p6)      fma.s1 POW_Z                  = POW_twoV, POW_U, f0
1097          nop.i 999
1098}
1099{ .mfi
1100          nop.m 999
1101          fma.s1 POW_v2                 = POW_P1, POW_r,  POW_P0
1102          nop.i 999
1103}
1104;;
1105
1106// p11 = TRUE ==> X is NEGATIVE but not inf
1107//    p12 = TRUE ==> X is NEGATIVE  AND  Y  already even int
1108//    p13 = TRUE ==> X is NEGATIVE  AND  Y possible int
1109{ .mfi
1110          nop.m 999
1111(p7)      fma.s1 POW_Z                  = POW_NORM_Y, POW_G, POW_Z2
1112(p11)     cmp.gt.unc  p12,p13           = pow_GR_exp_Y, pow_GR_10033
1113}
1114{ .mfi
1115          nop.m 999
1116          fma.s1 POW_Gpr                = POW_G, f1, POW_r
1117          nop.i 999
1118}
1119;;
1120
1121{ .mfi
1122          nop.m 999
1123          fma.s1 POW_Yrcub              = POW_rsq, POW_U, f0
1124          nop.i 999
1125}
1126{ .mfi
1127          nop.m 999
1128          fma.s1 POW_p                  = POW_rsq, POW_P2, POW_v2
1129          nop.i 999
1130}
1131;;
1132
1133// Test if x inf
1134{ .mfi
1135          nop.m 999
1136          fclass.m p15,p0 = POW_NORM_X,  0x23
1137          nop.i 999
1138}
1139// By adding RSHF (1.1000...*2^63) we put integer part in rightmost significand
1140{ .mfi
1141          nop.m 999
1142          fma.s1 POW_W1  = POW_Z, POW_inv_log2_by_128, POW_RSHF
1143          nop.i 999
1144}
1145;;
1146
1147// p13 = TRUE ==> X is NEGATIVE  AND  Y possible int
1148//     p10 = TRUE ==> X is NEG and Y is an int
1149//     p12 = TRUE ==> X is NEG and Y is not an int
1150{ .mfi
1151          nop.m 999
1152(p13)     fcmp.eq.unc.s1 p10,p12        = POW_float_int_Y,  POW_NORM_Y
1153          mov pow_GR_xneg_yodd = 0
1154}
1155{ .mfi
1156          nop.m 999
1157          fma.s1 POW_Y_Gpr              = POW_NORM_Y, POW_Gpr, f0
1158          nop.i 999
1159}
1160;;
1161
1162// p11 = TRUE ==> X is +1.0
1163{ .mfi
1164          nop.m 999
1165          fcmp.eq.s1 p11,p0 = POW_NORM_X, f1
1166          nop.i 999
1167}
1168;;
1169
1170// Extract rounded integer from rightmost significand of POW_W1
1171// By subtracting RSHF we get rounded integer POW_Nfloat
1172{ .mfi
1173          getf.sig pow_GR_int_N        = POW_W1
1174          fms.s1 POW_Nfloat  = POW_W1, f1, POW_RSHF
1175          nop.i 999
1176}
1177{ .mfb
1178          nop.m 999
1179          fma.s1 POW_Z3                 = POW_p, POW_Yrcub, f0
1180(p12)     br.cond.spnt POW_X_NEG_Y_NONINT  // Branch if x neg, y not integer
1181}
1182;;
1183
1184// p7  = TRUE ==> Y is +1.0
1185// p12 = TRUE ==> X is NEGATIVE  AND Y is an odd integer
1186{ .mfi
1187          getf.exp pow_GR_signexp_Y_Gpr = POW_Y_Gpr
1188          fcmp.eq.s1 p7,p0 = POW_NORM_Y, f1  // Test for y=1.0
1189(p10)     tbit.nz.unc  p12,p0           = pow_GR_sig_int_Y,0
1190}
1191{ .mfb
1192          nop.m 999
1193(p11)     fma.s.s0 f8 = f1,f1,f0    // If x=1, result is +1
1194(p15)     br.cond.spnt POW_X_INF
1195}
1196;;
1197
1198// Test x and y and flag denormal
1199{ .mfi
1200          nop.m 999
1201          fcmp.eq.s0 p15,p0 = f8,f9
1202          nop.i 999
1203}
1204{ .mfb
1205          nop.m 999
1206          fma.s1 POW_e3                 = POW_NORM_Y, POW_delta, f0
1207(p11)     br.ret.spnt b0            // Early exit if x=1.0, result is +1
1208}
1209;;
1210
1211{ .mfi
1212(p12)     mov pow_GR_xneg_yodd = 1
1213          fnma.s1 POW_f12  = POW_Nfloat, POW_log2_by_128_lo, f1
1214          nop.i 999
1215}
1216{ .mfb
1217          nop.m 999
1218          fnma.s1 POW_s  = POW_Nfloat, POW_log2_by_128_hi, POW_Z
1219(p7)      br.ret.spnt b0        // Early exit if y=1.0, result is x
1220}
1221;;
1222
1223{ .mmi
1224          and pow_GR_index1             = 0x0f, pow_GR_int_N
1225          and pow_GR_index2             = 0x70, pow_GR_int_N
1226          shr pow_int_GR_M              = pow_GR_int_N, 7    // M = N/128
1227}
1228;;
1229
1230{ .mfi
1231          shladd pow_AD_T1              = pow_GR_index1, 4, pow_AD_tbl1
1232          fma.s1 POW_q                  = POW_Z3, POW_Q1, POW_Q0_half
1233          add pow_int_GR_M              = pow_GR_16ones, pow_int_GR_M
1234}
1235{ .mfi
1236          add pow_AD_T2                 = pow_AD_tbl2, pow_GR_index2
1237          fma.s1 POW_Z3sq               = POW_Z3, POW_Z3, f0
1238          nop.i 999
1239}
1240;;
1241
1242{ .mmi
1243          ldfe POW_T1                   = [pow_AD_T1]
1244          ldfe POW_T2                   = [pow_AD_T2]
1245          nop.i 999
1246}
1247;;
1248
1249// f123 = f12*(e3+1) = f12*e3+f12
1250{ .mfi
1251          setf.exp POW_2M               = pow_int_GR_M
1252          fma.s1 POW_f123               = POW_e3,POW_f12,POW_f12
1253          nop.i 999
1254}
1255{ .mfi
1256          nop.m 999
1257          fma.s1 POW_ssq                = POW_s, POW_s, f0
1258          nop.i 999
1259}
1260;;
1261
1262{ .mfi
1263          nop.m 999
1264          fma.s1 POW_v2                 = POW_s, POW_Q2, POW_Q1
1265          and pow_GR_exp_Y_Gpr          = pow_GR_signexp_Y_Gpr, pow_GR_17ones
1266}
1267;;
1268
1269{ .mfi
1270          cmp.ne p12,p13 = pow_GR_xneg_yodd, r0
1271          fma.s1 POW_q                  = POW_Z3sq, POW_q, POW_Z3
1272          sub pow_GR_true_exp_Y_Gpr     = pow_GR_exp_Y_Gpr, pow_GR_16ones
1273}
1274;;
1275
1276// p8 TRUE ==> |Y(G + r)| >= 7
1277
1278// single
1279//     -2^7   -2^6             2^6   2^7
1280// -----+-----+----+ ... +-----+-----+-----
1281//  p8  |             p9             |  p8
1282//      |     |       p10      |     |
1283
1284// Form signexp of constants to indicate overflow
1285{ .mfi
1286          mov         pow_GR_big_pos    = 0x1007f
1287          nop.f 999
1288          cmp.le p8,p9                  = 7, pow_GR_true_exp_Y_Gpr
1289}
1290{ .mfi
1291          mov         pow_GR_big_neg    = 0x3007f
1292          nop.f 999
1293          andcm pow_GR_sign_Y_Gpr       = pow_GR_signexp_Y_Gpr, pow_GR_17ones
1294}
1295;;
1296
1297// Form big positive and negative constants to test for possible overflow
1298// Scale both terms of the polynomial by POW_f123
1299{ .mfi
1300          setf.exp POW_big_pos          = pow_GR_big_pos
1301          fma.s1 POW_ssq                = POW_ssq, POW_f123, f0
1302(p9)      cmp.le.unc p0,p10             = 6, pow_GR_true_exp_Y_Gpr
1303}
1304{ .mfb
1305          setf.exp POW_big_neg          = pow_GR_big_neg
1306          fma.s1 POW_1ps                = POW_s, POW_f123, POW_f123
1307(p8)      br.cond.spnt POW_OVER_UNDER_X_NOT_INF
1308}
1309;;
1310
1311{ .mfi
1312          nop.m 999
1313(p12)     fnma.s1 POW_T1T2              = POW_T1, POW_T2, f0
1314          nop.i 999
1315}
1316{ .mfi
1317          nop.m 999
1318(p13)     fma.s1 POW_T1T2               = POW_T1, POW_T2, f0
1319          nop.i 999
1320}
1321;;
1322
1323{ .mfi
1324          nop.m 999
1325          fma.s1 POW_v210               = POW_s, POW_v2, POW_Q0_half
1326          nop.i 999
1327}
1328{ .mfi
1329          nop.m 999
1330          fma.s1 POW_2Mqp1              = POW_2M, POW_q, POW_2M
1331          nop.i 999
1332}
1333;;
1334
1335{ .mfi
1336          nop.m 999
1337          fma.s1 POW_es                 = POW_ssq, POW_v210, POW_1ps
1338          nop.i 999
1339}
1340{ .mfi
1341          nop.m 999
1342          fma.s1 POW_A                  = POW_T1T2, POW_2Mqp1, f0
1343          nop.i 999
1344}
1345;;
1346
1347// Dummy op to set inexact
1348{ .mfi
1349          nop.m 999
1350          fma.s0 POW_tmp                = POW_2M, POW_q, POW_2M
1351          nop.i 999
1352}
1353;;
1354
1355{ .mfb
1356          nop.m 999
1357          fma.s.s0 f8                   = POW_A, POW_es, f0
1358(p10)     br.ret.sptk     b0            // Exit main branch if no over/underflow
1359}
1360;;
1361
1362// POSSIBLE_OVER_UNDER
1363// p6 = TRUE ==> Y_Gpr negative
1364// Result is already computed.  We just need to know if over/underflow occurred.
1365
1366{ .mfb
1367        cmp.eq p0,p6                    = pow_GR_sign_Y_Gpr, r0
1368        nop.f 999
1369(p6)    br.cond.spnt POW_POSSIBLE_UNDER
1370}
1371;;
1372
1373// POSSIBLE_OVER
1374// We got an answer.
1375// overflow is a possibility, not a certainty
1376
1377
1378// We define an overflow when the answer with
1379//    WRE set
1380//    user-defined rounding mode
1381
1382// double
1383// Largest double is 7FE (biased double)
1384//                   7FE - 3FF + FFFF = 103FE
1385// Create + largest_double_plus_ulp
1386// Create - largest_double_plus_ulp
1387// Calculate answer with WRE set.
1388
1389// single
1390// Largest single is FE (biased double)
1391//                   FE - 7F + FFFF = 1007E
1392// Create + largest_single_plus_ulp
1393// Create - largest_single_plus_ulp
1394// Calculate answer with WRE set.
1395
1396// Cases when answer is ldn+1  are as follows:
1397//  ldn                   ldn+1
1398// --+----------|----------+------------
1399//              |
1400//    +inf          +inf      -inf
1401//                  RN         RN
1402//                             RZ
1403
1404// Put in s2 (td set, wre set)
1405{ .mfi
1406        nop.m 999
1407        fsetc.s2 0x7F,0x42
1408        nop.i 999
1409}
1410;;
1411
1412{ .mfi
1413        nop.m 999
1414        fma.s.s2 POW_wre_urm_f8         = POW_A, POW_es, f0
1415        nop.i 999
1416}
1417;;
1418
1419// Return s2 to default
1420{ .mfi
1421        nop.m 999
1422        fsetc.s2 0x7F,0x40
1423        nop.i 999
1424}
1425;;
1426
1427// p7 = TRUE ==> yes, we have an overflow
1428{ .mfi
1429        nop.m 999
1430        fcmp.ge.s1 p7, p8               =  POW_wre_urm_f8, POW_big_pos
1431        nop.i 999
1432}
1433;;
1434
1435{ .mfi
1436        nop.m 999
1437(p8)    fcmp.le.s1 p7, p0               =  POW_wre_urm_f8, POW_big_neg
1438        nop.i 999
1439}
1440;;
1441
1442{ .mbb
1443(p7)   mov pow_GR_tag                   = 30
1444(p7)   br.cond.spnt __libm_error_region // Branch if overflow
1445       br.ret.sptk     b0               // Exit if did not overflow
1446}
1447;;
1448
1449
1450POW_POSSIBLE_UNDER:
1451// We got an answer. input was < -2^9 but > -2^10 (double)
1452// We got an answer. input was < -2^6 but > -2^7  (float)
1453// underflow is a possibility, not a certainty
1454
1455// We define an underflow when the answer with
1456//    ftz set
1457// is zero (tiny numbers become zero)
1458// Notice (from below) that if we have an unlimited exponent range,
1459// then there is an extra machine number E between the largest denormal and
1460// the smallest normal.
1461// So if with unbounded exponent we round to E or below, then we are
1462// tiny and underflow has occurred.
1463// But notice that you can be in a situation where we are tiny, namely
1464// rounded to E, but when the exponent is bounded we round to smallest
1465// normal. So the answer can be the smallest normal with underflow.
1466//                           E
1467// -----+--------------------+--------------------+-----
1468//      |                    |                    |
1469//   1.1...10 2^-3fff    1.1...11 2^-3fff    1.0...00 2^-3ffe
1470//   0.1...11 2^-3ffe                                   (biased, 1)
1471//    largest dn                               smallest normal
1472
1473// Form small constant (2^-170) to correct underflow result near region of
1474// smallest denormal in round-nearest.
1475
1476// Put in s2 (td set, ftz set)
1477.pred.rel "mutex",p12,p13
1478{ .mfi
1479        mov pow_GR_Fpsr = ar40          // Read the fpsr--need to check rc.s0
1480        fsetc.s2 0x7F,0x41
1481        mov pow_GR_rcs0_mask            = 0x0c00 // Set mask for rc.s0
1482}
1483{ .mfi
1484(p12)   mov pow_GR_tmp                  = 0x2ffff - 170
1485        nop.f 999
1486(p13)   mov pow_GR_tmp                  = 0x0ffff - 170
1487}
1488;;
1489
1490{ .mfi
1491        setf.exp POW_eps                = pow_GR_tmp        // Form 2^-170
1492        fma.s.s2 POW_ftz_urm_f8         = POW_A, POW_es, f0
1493        nop.i 999
1494}
1495;;
1496
1497// Return s2 to default
1498{ .mfi
1499        nop.m 999
1500        fsetc.s2 0x7F,0x40
1501        nop.i 999
1502}
1503;;
1504
1505// p7 = TRUE ==> yes, we have an underflow
1506{ .mfi
1507        nop.m 999
1508        fcmp.eq.s1 p7, p0               =  POW_ftz_urm_f8, f0
1509        nop.i 999
1510}
1511;;
1512
1513{ .mmi
1514(p7)    and pow_GR_rcs0  = pow_GR_rcs0_mask, pow_GR_Fpsr  // Isolate rc.s0
1515;;
1516(p7)    cmp.eq.unc p6,p0 = pow_GR_rcs0, r0    // Test for round to nearest
1517        nop.i 999
1518}
1519;;
1520
1521// Tweak result slightly if underflow to get correct rounding near smallest
1522// denormal if round-nearest
1523{ .mfi
1524        nop.m 999
1525(p6)    fms.s.s0 f8                     = POW_A, POW_es, POW_eps
1526        nop.i 999
1527}
1528{ .mbb
1529(p7)    mov pow_GR_tag                  = 31
1530(p7)    br.cond.spnt __libm_error_region // Branch if underflow
1531        br.ret.sptk     b0               // Exit if did not underflow
1532}
1533;;
1534
1535POW_X_DENORM:
1536// Here if x unorm. Use the NORM_X for getf instructions, and then back
1537// to normal path
1538{ .mfi
1539        getf.exp      pow_GR_signexp_X  = POW_NORM_X
1540        nop.f 999
1541        nop.i 999
1542}
1543;;
1544
1545{ .mib
1546        getf.sig      pow_GR_sig_X      = POW_NORM_X
1547        nop.i 999
1548        br.cond.sptk    POW_COMMON
1549}
1550;;
1551
1552POW_X_0:
1553// Here if x=0 and y not nan
1554//
1555// We have the following cases:
1556//  p6  x=0  and  y>0 and is an integer (may be even or odd)
1557//  p7  x=0  and  y>0 and is NOT an integer, return +0
1558//  p8  x=0  and  y>0 and so big as to always be an even integer, return +0
1559//  p9  x=0  and  y>0 and may not be integer
1560//  p10 x=0  and  y>0 and is an odd  integer, return x
1561//  p11 x=0  and  y>0 and is an even integer, return +0
1562//  p12 used in dummy fcmp to set denormal flag if y=unorm
1563//  p13 x=0  and  y>0
1564//  p14 x=0  and  y=0, branch to code for calling error handling
1565//  p15 x=0  and  y<0, branch to code for calling error handling
1566//
1567{ .mfi
1568        getf.sig pow_GR_sig_int_Y = POW_int_Y // Get signif of int_Y
1569        fcmp.lt.s1 p15,p13 = f9, f0           // Test for y<0
1570        and pow_GR_exp_Y = pow_GR_signexp_Y, pow_GR_17ones
1571}
1572{ .mfb
1573        cmp.ne p14,p0 = pow_GR_y_zero,r0      // Test for y=0
1574        fcvt.xf   POW_float_int_Y = POW_int_Y
1575(p14)   br.cond.spnt POW_X_0_Y_0              // Branch if x=0 and y=0
1576}
1577;;
1578
1579// If x=0 and y>0, test y and flag denormal
1580{ .mfb
1581(p13)   cmp.gt.unc p8,p9 = pow_GR_exp_Y, pow_GR_10033 // Test y +big = even int
1582(p13)   fcmp.eq.s0 p12,p0 = f9,f0    // If x=0, y>0 dummy op to flag denormal
1583(p15)   br.cond.spnt POW_X_0_Y_NEG // Branch if x=0 and y<0
1584}
1585;;
1586
1587// Here if x=0 and y>0
1588{ .mfi
1589        nop.m 999
1590(p9)    fcmp.eq.unc.s1 p6,p7 = POW_float_int_Y,  POW_NORM_Y // Test y=int
1591        nop.i 999
1592}
1593{ .mfi
1594        nop.m 999
1595(p8)    fma.s.s0 f8 = f0,f0,f0 // If x=0, y>0 and large even int, return +0
1596        nop.i 999
1597}
1598;;
1599
1600{ .mfi
1601        nop.m 999
1602(p7)    fma.s.s0 f8  = f0,f0,f0   // Result +0 if x=0 and y>0 and not integer
1603(p6)    tbit.nz.unc p10,p11 = pow_GR_sig_int_Y,0 // If y>0 int, test y even/odd
1604}
1605;;
1606
1607// Note if x=0, y>0 and odd integer, just return x
1608{ .mfb
1609        nop.m 999
1610(p11)   fma.s.s0 f8  = f0,f0,f0   // Result +0 if x=0 and y even integer
1611        br.ret.sptk b0            // Exit if x=0 and y>0
1612}
1613;;
1614
1615POW_X_0_Y_0:
1616// When X is +-0 and Y is +-0, IEEE returns 1.0
1617// We call error support with this value
1618
1619{ .mfb
1620        mov pow_GR_tag                  = 32
1621        fma.s.s0 f8                     = f1,f1,f0
1622        br.cond.sptk __libm_error_region
1623}
1624;;
1625
1626POW_X_0_Y_NEG:
1627// When X is +-0 and Y is negative, IEEE returns
1628// X     Y           answer
1629// +0    -odd int    +inf
1630// -0    -odd int    -inf
1631
1632// +0    !-odd int   +inf
1633// -0    !-odd int   +inf
1634
1635// p6 == Y is a floating point number outside the integer.
1636//       Hence it is an integer and is even.
1637//       return +inf
1638
1639// p7 == Y is a floating point number within the integer range.
1640//      p9  == (int_Y = NORM_Y), Y is an integer, which may be odd or even.
1641//           p11 odd
1642//              return (sign_of_x)inf
1643//           p12 even
1644//              return +inf
1645//      p10 == Y is not an integer
1646//         return +inf
1647//
1648
1649{ .mfi
1650          nop.m 999
1651          nop.f 999
1652          cmp.gt  p6,p7                 = pow_GR_exp_Y, pow_GR_10033
1653}
1654;;
1655
1656{ .mfi
1657          mov pow_GR_tag                = 33
1658(p7)      fcmp.eq.unc.s1 p9,p10         = POW_float_int_Y,  POW_NORM_Y
1659          nop.i 999
1660}
1661;;
1662
1663{ .mfb
1664          nop.m 999
1665(p6)      frcpa.s0 f8,p13               = f1, f0
1666(p6)      br.cond.sptk __libm_error_region   // x=0, y<0, y large neg int
1667}
1668;;
1669
1670{ .mfb
1671          nop.m 999
1672(p10)     frcpa.s0 f8,p13               = f1, f0
1673(p10)     br.cond.sptk __libm_error_region   // x=0, y<0, y not int
1674}
1675;;
1676
1677// x=0, y<0, y an int
1678{ .mib
1679          nop.m 999
1680(p9)      tbit.nz.unc p11,p12           = pow_GR_sig_int_Y,0
1681          nop.b 999
1682}
1683;;
1684
1685{ .mfi
1686          nop.m 999
1687(p12)     frcpa.s0 f8,p13               = f1,f0
1688          nop.i 999
1689}
1690;;
1691
1692{ .mfb
1693          nop.m 999
1694(p11)     frcpa.s0 f8,p13               = f1,f8
1695          br.cond.sptk __libm_error_region
1696}
1697;;
1698
1699
1700POW_Y_0:
1701// Here for y zero, x anything but zero and nan
1702// Set flag if x denormal
1703// Result is +1.0
1704{ .mfi
1705        nop.m 999
1706        fcmp.eq.s0 p6,p0 = f8,f0    // Sets flag if x denormal
1707        nop.i 999
1708}
1709{ .mfb
1710        nop.m 999
1711        fma.s.s0 f8 = f1,f1,f0
1712        br.ret.sptk b0
1713}
1714;;
1715
1716
1717POW_X_INF:
1718// Here when X is +-inf
1719
1720// X +inf  Y +inf             +inf
1721// X -inf  Y +inf             +inf
1722
1723// X +inf  Y >0               +inf
1724// X -inf  Y >0, !odd integer +inf     <== (-inf)^0.5 = +inf !!
1725// X -inf  Y >0,  odd integer -inf
1726
1727// X +inf  Y -inf             +0
1728// X -inf  Y -inf             +0
1729
1730// X +inf  Y <0               +0
1731// X -inf  Y <0, !odd integer +0
1732// X -inf  Y <0, odd integer  -0
1733
1734// X + inf Y=+0                +1
1735// X + inf Y=-0                +1
1736// X - inf Y=+0                +1
1737// X - inf Y=-0                +1
1738
1739// p13 == Y negative
1740// p14 == Y positive
1741
1742// p6 == Y is a floating point number outside the integer.
1743//       Hence it is an integer and is even.
1744//       p13 == (Y negative)
1745//          return +inf
1746//       p14 == (Y positive)
1747//          return +0
1748
1749// p7 == Y is a floating point number within the integer range.
1750//      p9  == (int_Y = NORM_Y), Y is an integer, which may be odd or even.
1751//           p11 odd
1752//              p13 == (Y negative)
1753//                 return (sign_of_x)inf
1754//              p14 == (Y positive)
1755//                 return (sign_of_x)0
1756//           pxx even
1757//              p13 == (Y negative)
1758//                 return +inf
1759//              p14 == (Y positive)
1760//                 return +0
1761
1762//      pxx == Y is not an integer
1763//           p13 == (Y negative)
1764//                 return +inf
1765//           p14 == (Y positive)
1766//                 return +0
1767//
1768
1769// If x=inf, test y and flag denormal
1770{ .mfi
1771          nop.m 999
1772          fcmp.eq.s0 p10,p11 = f9,f0
1773          nop.i 999
1774}
1775;;
1776
1777{ .mfi
1778          nop.m 999
1779          fcmp.lt.s0 p13,p14            = POW_NORM_Y,f0
1780          cmp.gt  p6,p7                 = pow_GR_exp_Y, pow_GR_10033
1781}
1782{ .mfi
1783          nop.m 999
1784          fclass.m p12,p0               = f9, 0x23 //@inf
1785          nop.i 999
1786}
1787;;
1788
1789{ .mfi
1790          nop.m 999
1791          fclass.m p15,p0               = f9, 0x07 //@zero
1792          nop.i 999
1793}
1794;;
1795
1796{ .mfb
1797          nop.m 999
1798(p15)     fmerge.s f8 = f1,f1      // Return +1.0 if x=inf, y=0
1799(p15)     br.ret.spnt b0           // Exit if x=inf, y=0
1800}
1801;;
1802
1803{ .mfi
1804          nop.m 999
1805(p14)     frcpa.s1 f8,p10 = f1,f0  // If x=inf, y>0, assume result +inf
1806          nop.i 999
1807}
1808{ .mfb
1809          nop.m 999
1810(p13)     fma.s.s0 f8 = f0,f0,f0   // If x=inf, y<0, assume result +0.0
1811(p12)     br.ret.spnt b0           // Exit if x=inf, y=inf
1812}
1813;;
1814
1815// Here if x=inf, and 0 < |y| < inf.  Need to correct results if y odd integer.
1816{ .mfi
1817          nop.m 999
1818(p7)      fcmp.eq.unc.s1 p9,p0 = POW_float_int_Y,  POW_NORM_Y // Is y integer?
1819          nop.i 999
1820}
1821;;
1822
1823{ .mfi
1824          nop.m 999
1825          nop.f 999
1826(p9)      tbit.nz.unc p11,p0 = pow_GR_sig_int_Y,0  // Test for y odd integer
1827}
1828;;
1829
1830{ .mfb
1831          nop.m 999
1832(p11)     fmerge.s f8 = POW_NORM_X,f8    // If y odd integer use sign of x
1833          br.ret.sptk b0                 // Exit for x=inf, 0 < |y| < inf
1834}
1835;;
1836
1837
1838POW_X_NEG_Y_NONINT:
1839// When X is negative and Y is a non-integer, IEEE
1840// returns a qnan indefinite.
1841// We call error support with this value
1842
1843{ .mfb
1844         mov pow_GR_tag                 = 34
1845         frcpa.s0 f8,p6                 = f0,f0
1846         br.cond.sptk __libm_error_region
1847}
1848;;
1849
1850POW_X_NAN:
1851// Here if x=nan, y not nan
1852{ .mfi
1853         nop.m 999
1854         fclass.m  p9,p13 = f9, 0x07 // Test y=zero
1855         nop.i 999
1856}
1857;;
1858
1859{ .mfb
1860         nop.m 999
1861(p13)    fma.s.s0 f8 = f8,f1,f0
1862(p13)    br.ret.sptk  b0            // Exit if x nan, y anything but zero or nan
1863}
1864;;
1865
1866POW_X_NAN_Y_0:
1867// When X is a NAN and Y is zero, IEEE returns 1.
1868// We call error support with this value.
1869{ .mfi
1870         nop.m 999
1871         fcmp.eq.s0 p6,p0 = f8,f0       // Dummy op to set invalid on snan
1872         nop.i 999
1873}
1874{ .mfb
1875         mov pow_GR_tag                 = 35
1876         fma.s.s0 f8 = f0,f0,f1
1877         br.cond.sptk __libm_error_region
1878}
1879;;
1880
1881
1882POW_OVER_UNDER_X_NOT_INF:
1883
1884// p8 is TRUE for overflow
1885// p9 is TRUE for underflow
1886
1887// if y is infinity, we should not over/underflow
1888
1889{ .mfi
1890          nop.m 999
1891          fcmp.eq.s1     p14, p13       = POW_xsq,f1  // Test |x|=1
1892          cmp.eq p8,p9                  = pow_GR_sign_Y_Gpr, r0
1893}
1894;;
1895
1896{ .mfi
1897          nop.m 999
1898(p14)     fclass.m.unc       p15, p0    = f9, 0x23 // If |x|=1, test y=inf
1899          nop.i 999
1900}
1901{ .mfi
1902          nop.m 999
1903(p13)     fclass.m.unc       p11,p0     = f9, 0x23 // If |x| not 1, test y=inf
1904          nop.i 999
1905}
1906;;
1907
1908// p15 = TRUE if |x|=1, y=inf, return +1
1909{ .mfb
1910          nop.m 999
1911(p15)     fma.s.s0          f8          = f1,f1,f0 // If |x|=1, y=inf, result +1
1912(p15)     br.ret.spnt b0                // Exit if |x|=1, y=inf
1913}
1914;;
1915
1916.pred.rel "mutex",p8,p9
1917{  .mfb
1918(p8)      setf.exp           f8 = pow_GR_17ones // If exp(+big), result inf
1919(p9)      fmerge.s           f8 = f0,f0         // If exp(-big), result 0
1920(p11)     br.ret.sptk b0                // Exit if |x| not 1, y=inf
1921}
1922;;
1923
1924{ .mfb
1925          nop.m 999
1926          nop.f 999
1927          br.cond.sptk POW_OVER_UNDER_ERROR // Branch if y not inf
1928}
1929;;
1930
1931
1932POW_Y_NAN:
1933// Here if y=nan, x anything
1934// If x = +1 then result is +1, else result is quiet Y
1935{ .mfi
1936       nop.m 999
1937       fcmp.eq.s1         p10,p9        = POW_NORM_X, f1
1938       nop.i 999
1939}
1940;;
1941
1942{ .mfi
1943       nop.m 999
1944(p10)  fcmp.eq.s0 p6,p0 = f9,f1   // Set invalid, even if x=+1
1945       nop.i 999
1946}
1947;;
1948
1949{ .mfi
1950       nop.m 999
1951(p10)  fma.s.s0 f8 = f1,f1,f0
1952       nop.i 999
1953}
1954{ .mfb
1955       nop.m 999
1956(p9)   fma.s.s0 f8 = f9,f8,f0
1957       br.ret.sptk b0             // Exit y=nan
1958}
1959;;
1960
1961
1962POW_OVER_UNDER_ERROR:
1963// Here if we have overflow or underflow.
1964// Enter with p12 true if x negative and y odd int to force -0 or -inf
1965
1966{ .mfi
1967         sub   pow_GR_17ones_m1         = pow_GR_17ones, r0, 1
1968         nop.f 999
1969         mov pow_GR_one                 = 0x1
1970}
1971;;
1972
1973// overflow, force inf with O flag
1974{ .mmb
1975(p8)     mov pow_GR_tag                 = 30
1976(p8)     setf.exp POW_tmp               = pow_GR_17ones_m1
1977         nop.b 999
1978}
1979;;
1980
1981// underflow, force zero with I, U flags
1982{ .mmi
1983(p9)    mov pow_GR_tag                  = 31
1984(p9)    setf.exp POW_tmp                = pow_GR_one
1985        nop.i 999
1986}
1987;;
1988
1989{ .mfi
1990        nop.m 999
1991        fma.s.s0 f8                     = POW_tmp, POW_tmp, f0
1992        nop.i 999
1993}
1994;;
1995
1996// p12 x is negative and y is an odd integer, change sign of result
1997{ .mfi
1998        nop.m 999
1999(p12)   fnma.s.s0 f8                    = POW_tmp, POW_tmp, f0
2000        nop.i 999
2001}
2002;;
2003
2004WEAK_LIBM_END(powf)
2005libm_alias_float_other (__pow, pow)
2006#ifdef SHARED
2007.symver powf,powf@@GLIBC_2.27
2008.weak __powf_compat
2009.set __powf_compat,__powf
2010.symver __powf_compat,powf@GLIBC_2.2
2011#endif
2012
2013
2014LOCAL_LIBM_ENTRY(__libm_error_region)
2015
2016.prologue
2017{ .mfi
2018        add   GR_Parameter_Y=-32,sp     // Parameter 2 value
2019        nop.f 0
2020.save   ar.pfs,GR_SAVE_PFS
2021        mov  GR_SAVE_PFS=ar.pfs         // Save ar.pfs
2022}
2023{ .mfi
2024.fframe 64
2025        add sp=-64,sp                   // Create new stack
2026        nop.f 0
2027        mov GR_SAVE_GP=gp               // Save gp
2028};;
2029
2030{ .mmi
2031        stfs [GR_Parameter_Y] = POW_NORM_Y,16 // STORE Parameter 2 on stack
2032        add GR_Parameter_X = 16,sp      // Parameter 1 address
2033.save   b0, GR_SAVE_B0
2034        mov GR_SAVE_B0=b0               // Save b0
2035};;
2036
2037.body
2038{ .mib
2039        stfs [GR_Parameter_X] = POW_NORM_X // STORE Parameter 1 on stack
2040        add   GR_Parameter_RESULT = 0,GR_Parameter_Y    // Parameter 3 address
2041        nop.b 0
2042}
2043{ .mib
2044        stfs [GR_Parameter_Y] = f8      // STORE Parameter 3 on stack
2045        add   GR_Parameter_Y = -16,GR_Parameter_Y
2046        br.call.sptk b0=__libm_error_support# // Call error handling function
2047};;
2048
2049{ .mmi
2050        add   GR_Parameter_RESULT = 48,sp
2051        nop.m 0
2052        nop.i 0
2053};;
2054
2055{ .mmi
2056        ldfs  f8 = [GR_Parameter_RESULT] // Get return result off stack
2057.restore sp
2058        add   sp = 64,sp                 // Restore stack pointer
2059        mov   b0 = GR_SAVE_B0            // Restore return address
2060};;
2061
2062{ .mib
2063        mov   gp = GR_SAVE_GP            // Restore gp
2064        mov   ar.pfs = GR_SAVE_PFS       // Restore ar.pfs
2065        br.ret.sptk     b0               // Return
2066};;
2067
2068LOCAL_LIBM_END(__libm_error_region)
2069
2070.type   __libm_error_support#,@function
2071.global __libm_error_support#
2072