File: e_powf.S

package info (click to toggle)
glibc 2.28-10
  • links: PTS, VCS
  • area: main
  • in suites: buster, experimental, sid
  • size: 272,168 kB
  • sloc: ansic: 1,008,602; asm: 259,607; makefile: 11,271; sh: 10,477; python: 6,910; cpp: 4,992; perl: 2,258; awk: 2,005; yacc: 290; pascal: 182; sed: 73
file content (2072 lines) | stat: -rw-r--r-- 67,045 bytes parent folder | download | duplicates (6)
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
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
.file "powf.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
// 02/03/00 Added p12 to definite over/under path. With odd power we did not
//          maintain the sign of x in this path.
// 04/04/00 Unwind support added
// 04/19/00 pow(+-1,inf) now returns NaN
//          pow(+-val, +-inf) returns 0 or inf, but now does not call error
//          support
//          Added s1 to fcvt.fx because invalid flag was incorrectly set.
// 08/15/00 Bundle added after call to __libm_error_support to properly
//          set [the previously overwritten] GR_Parameter_RESULT.
// 09/07/00 Improved performance by eliminating bank conflicts and other stalls,
//          and tweaking the critical path
// 09/08/00 Per c99, pow(+-1,inf) now returns 1, and pow(+1,nan) returns 1
// 09/28/00 Updated NaN**0 path
// 01/20/01 Fixed denormal flag settings.
// 02/13/01 Improved speed.
// 03/19/01 Reordered exp polynomial to improve speed and eliminate monotonicity
//          problem in round up, down, and to zero modes.  Also corrected
//          overflow result when x negative, y odd in round up, down, zero.
// 06/14/01 Added brace missing from bundle
// 12/10/01 Corrected case where x negative, 2^23 <= |y| < 2^24, y odd integer.
// 02/08/02 Fixed overflow/underflow cases that were not calling error support.
// 05/20/02 Cleaned up namespace and sf0 syntax
// 08/29/02 Improved Itanium 2 performance
// 02/10/03 Reordered header: .section, .global, .proc, .align
// 10/09/03 Modified algorithm to improve performance, reduce table size, and
//          fix boundary case powf(2.0,-150.0)
// 03/31/05 Reformatted delimiters between data tables
//
// API
//==============================================================
// float powf(float x, float y)
//
// Overview of operation
//==============================================================
//
// Three steps...
// 1. Log(x)
// 2. y Log(x)
// 3. exp(y log(x))
//
// This means we work with the absolute value of x and merge in the sign later.
//      Log(x) = G + delta + r -rsq/2 + p
// G,delta depend on the exponent of x and table entries. The table entries are
// indexed by the exponent of x, called K.
//
// The G and delta come out of the reduction; r is the reduced x.
//
// B = frcpa(x)
// xB-1 is small means that B is the approximate inverse of x.
//
//      Log(x) = Log( (1/B)(Bx) )
//             = Log(1/B) + Log(Bx)
//             = Log(1/B) + Log( 1 + (Bx-1))
//
//      x  = 2^K 1.x_1x_2.....x_52
//      B= frcpa(x) = 2^-k Cm
//      Log(1/B) = Log(1/(2^-K Cm))
//      Log(1/B) = Log((2^K/ Cm))
//      Log(1/B) = K Log(2) + Log(1/Cm)
//
//      Log(x)   = K Log(2) + Log(1/Cm) + Log( 1 + (Bx-1))
//
// If you take the significand of x, set the exponent to true 0, then Cm is
// the frcpa. We tabulate the Log(1/Cm) values. There are 256 of them.
// The frcpa table is indexed by 8 bits, the x_1 thru x_8.
// m = x_1x_2...x_8 is an 8-bit index.
//
//      Log(1/Cm) = log(1/frcpa(1+m/256)) where m goes from 0 to 255.
//
// We tabluate as one double, T for single precision power
//
//      Log(x)   = (K Log(2)_hi + T) + (K Log(2)_lo) + Log( 1 + (Bx-1))
//      Log(x)   =  G                +     delta     + Log( 1 + (Bx-1))
//
// The Log( 1 + (Bx-1)) can be calculated as a series in r = Bx-1.
//
//      Log( 1 + (Bx-1)) = r - rsq/2 + p
//        where p = r^3(P0 + P1*r + P2*r^2)
//
// Then,
//
//      yLog(x) = yG + y delta + y(r-rsq/2) + yp
//      yLog(x) = Z1 + e3      + Z2         + Z3
//
//
//     exp(yLog(x)) = exp(Z1 + Z2) exp(Z3) exp(e3)
//
//
//       exp(Z3) is another series.
//       exp(e3) is approximated as f3 = 1 +  e3
//
//       exp(Z1 + Z2) = exp(Z)
//       Z (128/log2) = number of log2/128 in Z is N
//
//       s = Z - N log2/128
//
//       exp(Z)       = exp(s) exp(N log2/128)
//
//       exp(r)       = exp(Z - N log2/128)
//
//      r = s + d = (Z - N (log2/128)_hi) -N (log2/128)_lo
//                =  Z - N (log2/128)
//
//      Z         = s+d +N (log2/128)
//
//      exp(Z)    = exp(s) (1+d) exp(N log2/128)
//
//      N = M 128 + n
//
//      N log2/128 = M log2 + n log2/128
//
//      n is 8 binary digits = n_7n_6...n_1
//
//      n log2/128 = n_7n_6n_5 16 log2/128 + n_4n_3n_2n_1 log2/128
//      n log2/128 = n_7n_6n_5 log2/8 + n_4n_3n_2n_1 log2/128
//      n log2/128 = I2 log2/8 + I1 log2/128
//
//      N log2/128 = M log2 + I2 log2/8 + I1 log2/128
//
//      exp(Z)    = exp(s) (1+d) exp(log(2^M) + log(2^I2/8) + log(2^I1/128))
//      exp(Z)    = exp(s) f12 (2^M) 2^I2/8 2^I1/128
//
// I1, I2 are table indices. Use a series for exp(s).
// Then get exp(Z)
//
//     exp(yLog(x)) = exp(Z) exp(Z3) f3
//     exp(yLog(x)) = exp(Z)f3 exp(Z3)
//     exp(yLog(x)) = A exp(Z3)
//
// We actually calculate exp(Z3) -1.
// Then,
//     exp(yLog(x)) = A + A( exp(Z3)   -1)
//

// Table Generation
//==============================================================

// The log values
// ==============
// The operation (K*log2_hi) must be exact. K is the true exponent of x.
// If we allow gradual underflow (denormals), K can be represented in 12 bits
// (as a two's complement number). We assume 13 bits as an engineering
// precaution.
//
//           +------------+----------------+-+
//           |  13 bits   | 50 bits        | |
//           +------------+----------------+-+
//           0            1                66
//                        2                34
//
// So we want the lsb(log2_hi) to be 2^-50
// We get log2 as a quad-extended (15-bit exponent, 128-bit significand)
//
//      0 fffe b17217f7d1cf79ab c9e3b39803f2f6af (4...)
//
// Consider numbering the bits left to right, starting at 0 thru 127.
// Bit 0 is the 2^-1 bit; bit 49 is the 2^-50 bit.
//
//  ...79ab
//     0111 1001 1010 1011
//     44
//     89
//
// So if we shift off the rightmost 14 bits, then (shift back only
// the top half) we get
//
//      0 fffe b17217f7d1cf4000 e6af278ece600fcb dabc000000000000
//
// Put the right 64-bit signficand in an FR register, convert to double;
// it is exact. Put the next 128 bits into a quad register and round to double.
// The true exponent of the low part is -51.
//
// hi is 0 fffe b17217f7d1cf4000
// lo is 0 ffcc e6af278ece601000
//
// Convert to double memory format and get
//
// hi is 0x3fe62e42fefa39e8
// lo is 0x3cccd5e4f1d9cc02
//
// log2_hi + log2_lo is an accurate value for log2.
//
//
// The T and t values
// ==================
// A similar method is used to generate the T and t values.
//
// K * log2_hi + T  must be exact.
//
// Smallest T,t
// ----------
// The smallest T,t is
//       T                   t
// 0x3f60040155d58800, 0x3c93bce0ce3ddd81  log(1/frcpa(1+0/256))=  +1.95503e-003
//
// The exponent is 0x3f6 (biased)  or -9 (true).
// For the smallest T value, what we want is to clip the significand such that
// when it is shifted right by 9, its lsb is in the bit for 2^-51. The 9 is the
// specific for the first entry. In general, it is 0xffff - (biased 15-bit
// exponent).

// Independently, what we have calculated is the table value as a quad
// precision number.
// Table entry 1 is
// 0 fff6 80200aaeac44ef38 338f77605fdf8000
//
// We store this quad precision number in a data structure that is
//    sign:           1
//    exponent:      15
//    signficand_hi: 64 (includes explicit bit)
//    signficand_lo: 49
// Because the explicit bit is included, the significand is 113 bits.
//
// Consider significand_hi for table entry 1.
//
//
// +-+--- ... -------+--------------------+
// | |
// +-+--- ... -------+--------------------+
// 0 1               4444444455555555556666
//                   2345678901234567890123
//
// Labeled as above, bit 0 is 2^0, bit 1 is 2^-1, etc.
// Bit 42 is 2^-42. If we shift to the right by 9, the bit in
// bit 42 goes in 51.
//
// So what we want to do is shift bits 43 thru 63 into significand_lo.
// This is shifting bit 42 into bit 63, taking care to retain shifted-off bits.
// Then shifting (just with signficaand_hi) back into bit 42.
//
// The shift_value is 63-42 = 21. In general, this is
//      63 - (51 -(0xffff - 0xfff6))
// For this example, it is
//      63 - (51 - 9) = 63 - 42  = 21
//
// This means we are shifting 21 bits into significand_lo. We must maintain more
// that a 128-bit signficand not to lose bits. So before the shift we put the
// 128-bit significand into a 256-bit signficand and then shift.
// The 256-bit significand has four parts: hh, hl, lh, and ll.
//
// Start off with
//      hh         hl         lh         ll
//      <64>       <49><15_0> <64_0>     <64_0>
//
// After shift by 21 (then return for significand_hi),
//      <43><21_0> <21><43>   <6><58_0>  <64_0>
//
// Take the hh part and convert to a double. There is no rounding here.
// The conversion is exact. The true exponent of the high part is the same as
// the true exponent of the input quad.
//
// We have some 64 plus significand bits for the low part. In this example, we
// have 70 bits. We want to round this to a double. Put them in a quad and then
// do a quad fnorm.
// For this example the true exponent of the low part is
//      true_exponent_of_high - 43 = true_exponent_of_high - (64-21)
// In general, this is
//      true_exponent_of_high - (64 - shift_value)
//
//
// Largest T,t
// ----------
// The largest T,t is
// 0x3fe62643fecf9742, 0x3c9e3147684bd37d  log(1/frcpa(1+255/256))=+6.92171e-001
//
// Table entry 256 is
// 0 fffe b1321ff67cba178c 51da12f4df5a0000
//
// The shift value is
//      63 - (51 -(0xffff - 0xfffe)) = 13
//
// The true exponent of the low part is
//      true_exponent_of_high - (64 - shift_value)
//      -1 - (64-13) = -52
// Biased as a double, this is 0x3cb
//
//
//
// So then lsb(T) must be >= 2^-51
// msb(Klog2_hi) <= 2^12
//
//              +--------+---------+
//              |       51 bits    | <== largest T
//              +--------+---------+
//              | 9 bits | 42 bits | <== smallest T
// +------------+----------------+-+
// |  13 bits   | 50 bits        | |
// +------------+----------------+-+
//
// Note: For powf only the table of T is needed


// Special Cases
//==============================================================

//                                   double     float
// overflow                          error 24   30

// underflow                         error 25   31

// X zero  Y zero
//  +0     +0                 +1     error 26   32
//  -0     +0                 +1     error 26   32
//  +0     -0                 +1     error 26   32
//  -0     -0                 +1     error 26   32

// X zero  Y negative
//  +0     -odd integer       +inf   error 27   33  divide-by-zero
//  -0     -odd integer       -inf   error 27   33  divide-by-zero
//  +0     !-odd integer      +inf   error 27   33  divide-by-zero
//  -0     !-odd integer      +inf   error 27   33  divide-by-zero
//  +0     -inf               +inf   error 27   33  divide-by-zero
//  -0     -inf               +inf   error 27   33  divide-by-zero

// X zero  Y positve
//  +0     +odd integer       +0
//  -0     +odd integer       -0
//  +0     !+odd integer      +0
//  -0     !+odd integer      +0
//  +0     +inf               +0
//  -0     +inf               +0
//  +0     Y NaN              quiet Y               invalid if Y SNaN
//  -0     Y NaN              quiet Y               invalid if Y SNaN

// X one
//  -1     Y inf              +1
//  -1     Y NaN              quiet Y               invalid if Y SNaN
//  +1     Y NaN              +1                    invalid if Y SNaN
//  +1     Y any else         +1

// X -     Y not integer      QNAN   error 28   34  invalid

// X NaN   Y 0                +1     error 29   35
// X NaN   Y NaN              quiet X               invalid if X or Y SNaN
// X NaN   Y any else         quiet X               invalid if X SNaN
// X !+1   Y NaN              quiet Y               invalid if Y SNaN


// X +inf  Y >0               +inf
// X -inf  Y >0, !odd integer +inf
// X -inf  Y >0, odd integer  -inf

// X +inf  Y <0               +0
// X -inf  Y <0, !odd integer +0
// X -inf  Y <0, odd integer  -0

// X +inf  Y =0               +1
// X -inf  Y =0               +1

// |X|<1   Y +inf             +0
// |X|<1   Y -inf             +inf
// |X|>1   Y +inf             +inf
// |X|>1   Y -inf             +0

// X any   Y =0               +1

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

// integer registers used

pow_GR_exp_half           = r10
pow_GR_signexp_Xm1        = r11
pow_GR_tmp                = r11

pow_GR_signexp_X          = r14
pow_GR_17ones             = r15
pow_GR_Fpsr               = r15
pow_AD_P                  = r16
pow_GR_rcs0_mask          = r16
pow_GR_exp_2tom8          = r17
pow_GR_rcs0               = r17
pow_GR_sig_X              = r18
pow_GR_10033              = r19
pow_GR_16ones             = r20

pow_AD_Tt                 = r21
pow_GR_exp_X              = r22
pow_AD_Q                  = r23
pow_GR_true_exp_X         = r24
pow_GR_y_zero             = r25

pow_GR_exp_Y              = r26
pow_AD_tbl1               = r27
pow_AD_tbl2               = r28
pow_GR_offset             = r29
pow_GR_exp_Xm1            = r30
pow_GR_xneg_yodd          = r31

pow_GR_int_N              = r38
pow_GR_index1             = r39
pow_GR_index2             = r40

pow_AD_T1                 = r41
pow_AD_T2                 = r42
pow_int_GR_M              = r43
pow_GR_sig_int_Y          = r44
pow_GR_sign_Y_Gpr         = r45

pow_GR_17ones_m1          = r46
pow_GR_one                = r47
pow_GR_sign_Y             = r48
pow_GR_signexp_Y_Gpr      = r49
pow_GR_exp_Y_Gpr          = r50

pow_GR_true_exp_Y_Gpr     = r51
pow_GR_signexp_Y          = r52
pow_GR_x_one              = r53
pow_GR_big_pos            = r55

pow_GR_big_neg            = r56

GR_SAVE_B0                = r50
GR_SAVE_GP                = r51
GR_SAVE_PFS               = r52

GR_Parameter_X            = r53
GR_Parameter_Y            = r54
GR_Parameter_RESULT       = r55
pow_GR_tag                = r56


// floating point registers used

POW_B                     = f32
POW_NORM_X                = f33
POW_Xm1                   = f34
POW_r1                    = f34

POW_NORM_Y                = f37
POW_Q2                    = f38
POW_eps                   = f39
POW_P2                    = f40

POW_P0                    = f42
POW_log2_lo               = f43
POW_r                     = f44
POW_Q0_half               = f45

POW_tmp                   = f47
POW_log2_hi               = f48
POW_Q1                    = f49
POW_P1                    = f50

POW_log2_by_128_hi        = f51
POW_inv_log2_by_128       = f52
POW_rsq                   = f53
POW_Yrcub                 = f54
POW_log2_by_128_lo        = f55

POW_xsq                   = f57
POW_v2                    = f59
POW_T                     = f60

POW_RSHF                  = f62
POW_v210                  = f63
POW_twoV                  = f65

POW_U                     = f66
POW_G                     = f67
POW_delta                 = f68
POW_V                     = f70

POW_p                     = f71
POW_Z                     = f72
POW_e3                    = f73
POW_Z2                    = f75

POW_W1                    = f77
POW_Z3                    = f80

POW_Z3sq                  = f85

POW_Nfloat                = f87
POW_f3                    = f89
POW_q                     = f90

POW_T1                    = f96
POW_T2                    = f97
POW_2M                    = f98
POW_s                     = f99
POW_f12                   = f100

POW_ssq                   = f101
POW_T1T2                  = f102
POW_1ps                   = f103
POW_A                     = f104
POW_es                    = f105

POW_Xp1                   = f106
POW_int_K                 = f107
POW_K                     = f108
POW_f123                  = f109
POW_Gpr                   = f110

POW_Y_Gpr                 = f111
POW_int_Y                 = f112
POW_2Mqp1                 = f113

POW_float_int_Y           = f116
POW_ftz_urm_f8            = f117
POW_wre_urm_f8            = f118
POW_big_neg               = f119
POW_big_pos               = f120

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

RODATA

.align 16

LOCAL_OBJECT_START(pow_table_P)
data8 0x80000000000018E5, 0x0000BFFD  // P_1
data8 0xb8aa3b295c17f0bc, 0x00004006  // inv_ln2_by_128
//
//
data8 0x3FA5555555554A9E // Q_2
data8 0x0000000000000000 // Pad
data8 0x3FC5555555554733 // Q_1
data8 0x43e8000000000000 // Right shift constant for exp
data8 0xc9e3b39803f2f6af, 0x00003fb7  // ln2_by_128_lo
LOCAL_OBJECT_END(pow_table_P)

LOCAL_OBJECT_START(pow_table_Q)
data8 0xCCCCCCCC4ED2BA7F, 0x00003FFC  // P_2
data8 0xAAAAAAAAAAAAB505, 0x00003FFD  // P_0
data8 0x3fe62e42fefa39e8, 0x3cccd5e4f1d9cc02 // log2 hi lo =  +6.93147e-001
data8 0xb17217f7d1cf79ab, 0x00003ff7  // ln2_by_128_hi
LOCAL_OBJECT_END(pow_table_Q)


LOCAL_OBJECT_START(pow_Tt)
data8 0x3f60040155d58800 // log(1/frcpa(1+0/256))=  +1.95503e-003
data8 0x3f78121214586a00 // log(1/frcpa(1+1/256))=  +5.87661e-003
data8 0x3f841929f9683200 // log(1/frcpa(1+2/256))=  +9.81362e-003
data8 0x3f8c317384c75f00 // log(1/frcpa(1+3/256))=  +1.37662e-002
data8 0x3f91a6b91ac73380 // log(1/frcpa(1+4/256))=  +1.72376e-002
data8 0x3f95ba9a5d9ac000 // log(1/frcpa(1+5/256))=  +2.12196e-002
data8 0x3f99d2a807432580 // log(1/frcpa(1+6/256))=  +2.52177e-002
data8 0x3f9d6b2725979800 // log(1/frcpa(1+7/256))=  +2.87291e-002
data8 0x3fa0c58fa19dfa80 // log(1/frcpa(1+8/256))=  +3.27573e-002
data8 0x3fa2954c78cbce00 // log(1/frcpa(1+9/256))=  +3.62953e-002
data8 0x3fa4a94d2da96c40 // log(1/frcpa(1+10/256))=  +4.03542e-002
data8 0x3fa67c94f2d4bb40 // log(1/frcpa(1+11/256))=  +4.39192e-002
data8 0x3fa85188b630f040 // log(1/frcpa(1+12/256))=  +4.74971e-002
data8 0x3faa6b8abe73af40 // log(1/frcpa(1+13/256))=  +5.16017e-002
data8 0x3fac441e06f72a80 // log(1/frcpa(1+14/256))=  +5.52072e-002
data8 0x3fae1e6713606d00 // log(1/frcpa(1+15/256))=  +5.88257e-002
data8 0x3faffa6911ab9300 // log(1/frcpa(1+16/256))=  +6.24574e-002
data8 0x3fb0ec139c5da600 // log(1/frcpa(1+17/256))=  +6.61022e-002
data8 0x3fb1dbd2643d1900 // log(1/frcpa(1+18/256))=  +6.97605e-002
data8 0x3fb2cc7284fe5f00 // log(1/frcpa(1+19/256))=  +7.34321e-002
data8 0x3fb3bdf5a7d1ee60 // log(1/frcpa(1+20/256))=  +7.71173e-002
data8 0x3fb4b05d7aa012e0 // log(1/frcpa(1+21/256))=  +8.08161e-002
data8 0x3fb580db7ceb5700 // log(1/frcpa(1+22/256))=  +8.39975e-002
data8 0x3fb674f089365a60 // log(1/frcpa(1+23/256))=  +8.77219e-002
data8 0x3fb769ef2c6b5680 // log(1/frcpa(1+24/256))=  +9.14602e-002
data8 0x3fb85fd927506a40 // log(1/frcpa(1+25/256))=  +9.52125e-002
data8 0x3fb9335e5d594980 // log(1/frcpa(1+26/256))=  +9.84401e-002
data8 0x3fba2b0220c8e5e0 // log(1/frcpa(1+27/256))=  +1.02219e-001
data8 0x3fbb0004ac1a86a0 // log(1/frcpa(1+28/256))=  +1.05469e-001
data8 0x3fbbf968769fca00 // log(1/frcpa(1+29/256))=  +1.09274e-001
data8 0x3fbccfedbfee13a0 // log(1/frcpa(1+30/256))=  +1.12548e-001
data8 0x3fbda727638446a0 // log(1/frcpa(1+31/256))=  +1.15832e-001
data8 0x3fbea3257fe10f60 // log(1/frcpa(1+32/256))=  +1.19677e-001
data8 0x3fbf7be9fedbfde0 // log(1/frcpa(1+33/256))=  +1.22985e-001
data8 0x3fc02ab352ff25f0 // log(1/frcpa(1+34/256))=  +1.26303e-001
data8 0x3fc097ce579d2040 // log(1/frcpa(1+35/256))=  +1.29633e-001
data8 0x3fc1178e8227e470 // log(1/frcpa(1+36/256))=  +1.33531e-001
data8 0x3fc185747dbecf30 // log(1/frcpa(1+37/256))=  +1.36885e-001
data8 0x3fc1f3b925f25d40 // log(1/frcpa(1+38/256))=  +1.40250e-001
data8 0x3fc2625d1e6ddf50 // log(1/frcpa(1+39/256))=  +1.43627e-001
data8 0x3fc2d1610c868130 // log(1/frcpa(1+40/256))=  +1.47015e-001
data8 0x3fc340c597411420 // log(1/frcpa(1+41/256))=  +1.50414e-001
data8 0x3fc3b08b6757f2a0 // log(1/frcpa(1+42/256))=  +1.53825e-001
data8 0x3fc40dfb08378000 // log(1/frcpa(1+43/256))=  +1.56677e-001
data8 0x3fc47e74e8ca5f70 // log(1/frcpa(1+44/256))=  +1.60109e-001
data8 0x3fc4ef51f6466de0 // log(1/frcpa(1+45/256))=  +1.63553e-001
data8 0x3fc56092e02ba510 // log(1/frcpa(1+46/256))=  +1.67010e-001
data8 0x3fc5d23857cd74d0 // log(1/frcpa(1+47/256))=  +1.70478e-001
data8 0x3fc6313a37335d70 // log(1/frcpa(1+48/256))=  +1.73377e-001
data8 0x3fc6a399dabbd380 // log(1/frcpa(1+49/256))=  +1.76868e-001
data8 0x3fc70337dd3ce410 // log(1/frcpa(1+50/256))=  +1.79786e-001
data8 0x3fc77654128f6120 // log(1/frcpa(1+51/256))=  +1.83299e-001
data8 0x3fc7e9d82a0b0220 // log(1/frcpa(1+52/256))=  +1.86824e-001
data8 0x3fc84a6b759f5120 // log(1/frcpa(1+53/256))=  +1.89771e-001
data8 0x3fc8ab47d5f5a300 // log(1/frcpa(1+54/256))=  +1.92727e-001
data8 0x3fc91fe490965810 // log(1/frcpa(1+55/256))=  +1.96286e-001
data8 0x3fc981634011aa70 // log(1/frcpa(1+56/256))=  +1.99261e-001
data8 0x3fc9f6c407089660 // log(1/frcpa(1+57/256))=  +2.02843e-001
data8 0x3fca58e729348f40 // log(1/frcpa(1+58/256))=  +2.05838e-001
data8 0x3fcabb55c31693a0 // log(1/frcpa(1+59/256))=  +2.08842e-001
data8 0x3fcb1e104919efd0 // log(1/frcpa(1+60/256))=  +2.11855e-001
data8 0x3fcb94ee93e367c0 // log(1/frcpa(1+61/256))=  +2.15483e-001
data8 0x3fcbf851c0675550 // log(1/frcpa(1+62/256))=  +2.18516e-001
data8 0x3fcc5c0254bf23a0 // log(1/frcpa(1+63/256))=  +2.21558e-001
data8 0x3fccc000c9db3c50 // log(1/frcpa(1+64/256))=  +2.24609e-001
data8 0x3fcd244d99c85670 // log(1/frcpa(1+65/256))=  +2.27670e-001
data8 0x3fcd88e93fb2f450 // log(1/frcpa(1+66/256))=  +2.30741e-001
data8 0x3fcdedd437eaef00 // log(1/frcpa(1+67/256))=  +2.33820e-001
data8 0x3fce530effe71010 // log(1/frcpa(1+68/256))=  +2.36910e-001
data8 0x3fceb89a1648b970 // log(1/frcpa(1+69/256))=  +2.40009e-001
data8 0x3fcf1e75fadf9bd0 // log(1/frcpa(1+70/256))=  +2.43117e-001
data8 0x3fcf84a32ead7c30 // log(1/frcpa(1+71/256))=  +2.46235e-001
data8 0x3fcfeb2233ea07c0 // log(1/frcpa(1+72/256))=  +2.49363e-001
data8 0x3fd028f9c7035c18 // log(1/frcpa(1+73/256))=  +2.52501e-001
data8 0x3fd05c8be0d96358 // log(1/frcpa(1+74/256))=  +2.55649e-001
data8 0x3fd085eb8f8ae790 // log(1/frcpa(1+75/256))=  +2.58174e-001
data8 0x3fd0b9c8e32d1910 // log(1/frcpa(1+76/256))=  +2.61339e-001
data8 0x3fd0edd060b78080 // log(1/frcpa(1+77/256))=  +2.64515e-001
data8 0x3fd122024cf00638 // log(1/frcpa(1+78/256))=  +2.67701e-001
data8 0x3fd14be2927aecd0 // log(1/frcpa(1+79/256))=  +2.70257e-001
data8 0x3fd180618ef18ad8 // log(1/frcpa(1+80/256))=  +2.73461e-001
data8 0x3fd1b50bbe2fc638 // log(1/frcpa(1+81/256))=  +2.76675e-001
data8 0x3fd1df4cc7cf2428 // log(1/frcpa(1+82/256))=  +2.79254e-001
data8 0x3fd214456d0eb8d0 // log(1/frcpa(1+83/256))=  +2.82487e-001
data8 0x3fd23ec5991eba48 // log(1/frcpa(1+84/256))=  +2.85081e-001
data8 0x3fd2740d9f870af8 // log(1/frcpa(1+85/256))=  +2.88333e-001
data8 0x3fd29ecdabcdfa00 // log(1/frcpa(1+86/256))=  +2.90943e-001
data8 0x3fd2d46602adcce8 // log(1/frcpa(1+87/256))=  +2.94214e-001
data8 0x3fd2ff66b04ea9d0 // log(1/frcpa(1+88/256))=  +2.96838e-001
data8 0x3fd335504b355a30 // log(1/frcpa(1+89/256))=  +3.00129e-001
data8 0x3fd360925ec44f58 // log(1/frcpa(1+90/256))=  +3.02769e-001
data8 0x3fd38bf1c3337e70 // log(1/frcpa(1+91/256))=  +3.05417e-001
data8 0x3fd3c25277333180 // log(1/frcpa(1+92/256))=  +3.08735e-001
data8 0x3fd3edf463c16838 // log(1/frcpa(1+93/256))=  +3.11399e-001
data8 0x3fd419b423d5e8c0 // log(1/frcpa(1+94/256))=  +3.14069e-001
data8 0x3fd44591e0539f48 // log(1/frcpa(1+95/256))=  +3.16746e-001
data8 0x3fd47c9175b6f0a8 // log(1/frcpa(1+96/256))=  +3.20103e-001
data8 0x3fd4a8b341552b08 // log(1/frcpa(1+97/256))=  +3.22797e-001
data8 0x3fd4d4f390890198 // log(1/frcpa(1+98/256))=  +3.25498e-001
data8 0x3fd501528da1f960 // log(1/frcpa(1+99/256))=  +3.28206e-001
data8 0x3fd52dd06347d4f0 // log(1/frcpa(1+100/256))=  +3.30921e-001
data8 0x3fd55a6d3c7b8a88 // log(1/frcpa(1+101/256))=  +3.33644e-001
data8 0x3fd5925d2b112a58 // log(1/frcpa(1+102/256))=  +3.37058e-001
data8 0x3fd5bf406b543db0 // log(1/frcpa(1+103/256))=  +3.39798e-001
data8 0x3fd5ec433d5c35a8 // log(1/frcpa(1+104/256))=  +3.42545e-001
data8 0x3fd61965cdb02c18 // log(1/frcpa(1+105/256))=  +3.45300e-001
data8 0x3fd646a84935b2a0 // log(1/frcpa(1+106/256))=  +3.48063e-001
data8 0x3fd6740add31de90 // log(1/frcpa(1+107/256))=  +3.50833e-001
data8 0x3fd6a18db74a58c0 // log(1/frcpa(1+108/256))=  +3.53610e-001
data8 0x3fd6cf31058670e8 // log(1/frcpa(1+109/256))=  +3.56396e-001
data8 0x3fd6f180e852f0b8 // log(1/frcpa(1+110/256))=  +3.58490e-001
data8 0x3fd71f5d71b894e8 // log(1/frcpa(1+111/256))=  +3.61289e-001
data8 0x3fd74d5aefd66d58 // log(1/frcpa(1+112/256))=  +3.64096e-001
data8 0x3fd77b79922bd378 // log(1/frcpa(1+113/256))=  +3.66911e-001
data8 0x3fd7a9b9889f19e0 // log(1/frcpa(1+114/256))=  +3.69734e-001
data8 0x3fd7d81b037eb6a0 // log(1/frcpa(1+115/256))=  +3.72565e-001
data8 0x3fd8069e33827230 // log(1/frcpa(1+116/256))=  +3.75404e-001
data8 0x3fd82996d3ef8bc8 // log(1/frcpa(1+117/256))=  +3.77538e-001
data8 0x3fd85855776dcbf8 // log(1/frcpa(1+118/256))=  +3.80391e-001
data8 0x3fd8873658327cc8 // log(1/frcpa(1+119/256))=  +3.83253e-001
data8 0x3fd8aa75973ab8c8 // log(1/frcpa(1+120/256))=  +3.85404e-001
data8 0x3fd8d992dc8824e0 // log(1/frcpa(1+121/256))=  +3.88280e-001
data8 0x3fd908d2ea7d9510 // log(1/frcpa(1+122/256))=  +3.91164e-001
data8 0x3fd92c59e79c0e50 // log(1/frcpa(1+123/256))=  +3.93332e-001
data8 0x3fd95bd750ee3ed0 // log(1/frcpa(1+124/256))=  +3.96231e-001
data8 0x3fd98b7811a3ee58 // log(1/frcpa(1+125/256))=  +3.99138e-001
data8 0x3fd9af47f33d4068 // log(1/frcpa(1+126/256))=  +4.01323e-001
data8 0x3fd9df270c1914a0 // log(1/frcpa(1+127/256))=  +4.04245e-001
data8 0x3fda0325ed14fda0 // log(1/frcpa(1+128/256))=  +4.06442e-001
data8 0x3fda33440224fa78 // log(1/frcpa(1+129/256))=  +4.09379e-001
data8 0x3fda57725e80c380 // log(1/frcpa(1+130/256))=  +4.11587e-001
data8 0x3fda87d0165dd198 // log(1/frcpa(1+131/256))=  +4.14539e-001
data8 0x3fdaac2e6c03f890 // log(1/frcpa(1+132/256))=  +4.16759e-001
data8 0x3fdadccc6fdf6a80 // log(1/frcpa(1+133/256))=  +4.19726e-001
data8 0x3fdb015b3eb1e790 // log(1/frcpa(1+134/256))=  +4.21958e-001
data8 0x3fdb323a3a635948 // log(1/frcpa(1+135/256))=  +4.24941e-001
data8 0x3fdb56fa04462908 // log(1/frcpa(1+136/256))=  +4.27184e-001
data8 0x3fdb881aa659bc90 // log(1/frcpa(1+137/256))=  +4.30182e-001
data8 0x3fdbad0bef3db160 // log(1/frcpa(1+138/256))=  +4.32437e-001
data8 0x3fdbd21297781c28 // log(1/frcpa(1+139/256))=  +4.34697e-001
data8 0x3fdc039236f08818 // log(1/frcpa(1+140/256))=  +4.37718e-001
data8 0x3fdc28cb1e4d32f8 // log(1/frcpa(1+141/256))=  +4.39990e-001
data8 0x3fdc4e19b84723c0 // log(1/frcpa(1+142/256))=  +4.42267e-001
data8 0x3fdc7ff9c74554c8 // log(1/frcpa(1+143/256))=  +4.45311e-001
data8 0x3fdca57b64e9db00 // log(1/frcpa(1+144/256))=  +4.47600e-001
data8 0x3fdccb130a5ceba8 // log(1/frcpa(1+145/256))=  +4.49895e-001
data8 0x3fdcf0c0d18f3268 // log(1/frcpa(1+146/256))=  +4.52194e-001
data8 0x3fdd232075b5a200 // log(1/frcpa(1+147/256))=  +4.55269e-001
data8 0x3fdd490246defa68 // log(1/frcpa(1+148/256))=  +4.57581e-001
data8 0x3fdd6efa918d25c8 // log(1/frcpa(1+149/256))=  +4.59899e-001
data8 0x3fdd9509707ae528 // log(1/frcpa(1+150/256))=  +4.62221e-001
data8 0x3fddbb2efe92c550 // log(1/frcpa(1+151/256))=  +4.64550e-001
data8 0x3fddee2f3445e4a8 // log(1/frcpa(1+152/256))=  +4.67663e-001
data8 0x3fde148a1a2726c8 // log(1/frcpa(1+153/256))=  +4.70004e-001
data8 0x3fde3afc0a49ff38 // log(1/frcpa(1+154/256))=  +4.72350e-001
data8 0x3fde6185206d5168 // log(1/frcpa(1+155/256))=  +4.74702e-001
data8 0x3fde882578823d50 // log(1/frcpa(1+156/256))=  +4.77060e-001
data8 0x3fdeaedd2eac9908 // log(1/frcpa(1+157/256))=  +4.79423e-001
data8 0x3fded5ac5f436be0 // log(1/frcpa(1+158/256))=  +4.81792e-001
data8 0x3fdefc9326d16ab8 // log(1/frcpa(1+159/256))=  +4.84166e-001
data8 0x3fdf2391a21575f8 // log(1/frcpa(1+160/256))=  +4.86546e-001
data8 0x3fdf4aa7ee031928 // log(1/frcpa(1+161/256))=  +4.88932e-001
data8 0x3fdf71d627c30bb0 // log(1/frcpa(1+162/256))=  +4.91323e-001
data8 0x3fdf991c6cb3b378 // log(1/frcpa(1+163/256))=  +4.93720e-001
data8 0x3fdfc07ada69a908 // log(1/frcpa(1+164/256))=  +4.96123e-001
data8 0x3fdfe7f18eb03d38 // log(1/frcpa(1+165/256))=  +4.98532e-001
data8 0x3fe007c053c5002c // log(1/frcpa(1+166/256))=  +5.00946e-001
data8 0x3fe01b942198a5a0 // log(1/frcpa(1+167/256))=  +5.03367e-001
data8 0x3fe02f74400c64e8 // log(1/frcpa(1+168/256))=  +5.05793e-001
data8 0x3fe04360be7603ac // log(1/frcpa(1+169/256))=  +5.08225e-001
data8 0x3fe05759ac47fe30 // log(1/frcpa(1+170/256))=  +5.10663e-001
data8 0x3fe06b5f1911cf50 // log(1/frcpa(1+171/256))=  +5.13107e-001
data8 0x3fe078bf0533c568 // log(1/frcpa(1+172/256))=  +5.14740e-001
data8 0x3fe08cd9687e7b0c // log(1/frcpa(1+173/256))=  +5.17194e-001
data8 0x3fe0a10074cf9018 // log(1/frcpa(1+174/256))=  +5.19654e-001
data8 0x3fe0b5343a234474 // log(1/frcpa(1+175/256))=  +5.22120e-001
data8 0x3fe0c974c89431cc // log(1/frcpa(1+176/256))=  +5.24592e-001
data8 0x3fe0ddc2305b9884 // log(1/frcpa(1+177/256))=  +5.27070e-001
data8 0x3fe0eb524bafc918 // log(1/frcpa(1+178/256))=  +5.28726e-001
data8 0x3fe0ffb54213a474 // log(1/frcpa(1+179/256))=  +5.31214e-001
data8 0x3fe114253da97d9c // log(1/frcpa(1+180/256))=  +5.33709e-001
data8 0x3fe128a24f1d9afc // log(1/frcpa(1+181/256))=  +5.36210e-001
data8 0x3fe1365252bf0864 // log(1/frcpa(1+182/256))=  +5.37881e-001
data8 0x3fe14ae558b4a92c // log(1/frcpa(1+183/256))=  +5.40393e-001
data8 0x3fe15f85a19c7658 // log(1/frcpa(1+184/256))=  +5.42910e-001
data8 0x3fe16d4d38c119f8 // log(1/frcpa(1+185/256))=  +5.44592e-001
data8 0x3fe18203c20dd130 // log(1/frcpa(1+186/256))=  +5.47121e-001
data8 0x3fe196c7bc4b1f38 // log(1/frcpa(1+187/256))=  +5.49656e-001
data8 0x3fe1a4a738b7a33c // log(1/frcpa(1+188/256))=  +5.51349e-001
data8 0x3fe1b981c0c9653c // log(1/frcpa(1+189/256))=  +5.53895e-001
data8 0x3fe1ce69e8bb1068 // log(1/frcpa(1+190/256))=  +5.56447e-001
data8 0x3fe1dc619de06944 // log(1/frcpa(1+191/256))=  +5.58152e-001
data8 0x3fe1f160a2ad0da0 // log(1/frcpa(1+192/256))=  +5.60715e-001
data8 0x3fe2066d7740737c // log(1/frcpa(1+193/256))=  +5.63285e-001
data8 0x3fe2147dba47a390 // log(1/frcpa(1+194/256))=  +5.65001e-001
data8 0x3fe229a1bc5ebac0 // log(1/frcpa(1+195/256))=  +5.67582e-001
data8 0x3fe237c1841a502c // log(1/frcpa(1+196/256))=  +5.69306e-001
data8 0x3fe24cfce6f80d98 // log(1/frcpa(1+197/256))=  +5.71898e-001
data8 0x3fe25b2c55cd5760 // log(1/frcpa(1+198/256))=  +5.73630e-001
data8 0x3fe2707f4d5f7c40 // log(1/frcpa(1+199/256))=  +5.76233e-001
data8 0x3fe285e0842ca380 // log(1/frcpa(1+200/256))=  +5.78842e-001
data8 0x3fe294294708b770 // log(1/frcpa(1+201/256))=  +5.80586e-001
data8 0x3fe2a9a2670aff0c // log(1/frcpa(1+202/256))=  +5.83207e-001
data8 0x3fe2b7fb2c8d1cc0 // log(1/frcpa(1+203/256))=  +5.84959e-001
data8 0x3fe2c65a6395f5f4 // log(1/frcpa(1+204/256))=  +5.86713e-001
data8 0x3fe2dbf557b0df40 // log(1/frcpa(1+205/256))=  +5.89350e-001
data8 0x3fe2ea64c3f97654 // log(1/frcpa(1+206/256))=  +5.91113e-001
data8 0x3fe3001823684d70 // log(1/frcpa(1+207/256))=  +5.93762e-001
data8 0x3fe30e97e9a8b5cc // log(1/frcpa(1+208/256))=  +5.95531e-001
data8 0x3fe32463ebdd34e8 // log(1/frcpa(1+209/256))=  +5.98192e-001
data8 0x3fe332f4314ad794 // log(1/frcpa(1+210/256))=  +5.99970e-001
data8 0x3fe348d90e7464cc // log(1/frcpa(1+211/256))=  +6.02643e-001
data8 0x3fe35779f8c43d6c // log(1/frcpa(1+212/256))=  +6.04428e-001
data8 0x3fe36621961a6a98 // log(1/frcpa(1+213/256))=  +6.06217e-001
data8 0x3fe37c299f3c3668 // log(1/frcpa(1+214/256))=  +6.08907e-001
data8 0x3fe38ae2171976e4 // log(1/frcpa(1+215/256))=  +6.10704e-001
data8 0x3fe399a157a603e4 // log(1/frcpa(1+216/256))=  +6.12504e-001
data8 0x3fe3afccfe77b9d0 // log(1/frcpa(1+217/256))=  +6.15210e-001
data8 0x3fe3be9d503533b4 // log(1/frcpa(1+218/256))=  +6.17018e-001
data8 0x3fe3cd7480b4a8a0 // log(1/frcpa(1+219/256))=  +6.18830e-001
data8 0x3fe3e3c43918f76c // log(1/frcpa(1+220/256))=  +6.21554e-001
data8 0x3fe3f2acb27ed6c4 // log(1/frcpa(1+221/256))=  +6.23373e-001
data8 0x3fe4019c2125ca90 // log(1/frcpa(1+222/256))=  +6.25197e-001
data8 0x3fe4181061389720 // log(1/frcpa(1+223/256))=  +6.27937e-001
data8 0x3fe42711518df544 // log(1/frcpa(1+224/256))=  +6.29769e-001
data8 0x3fe436194e12b6bc // log(1/frcpa(1+225/256))=  +6.31604e-001
data8 0x3fe445285d68ea68 // log(1/frcpa(1+226/256))=  +6.33442e-001
data8 0x3fe45bcc464c8938 // log(1/frcpa(1+227/256))=  +6.36206e-001
data8 0x3fe46aed21f117fc // log(1/frcpa(1+228/256))=  +6.38053e-001
data8 0x3fe47a1527e8a2d0 // log(1/frcpa(1+229/256))=  +6.39903e-001
data8 0x3fe489445efffcc8 // log(1/frcpa(1+230/256))=  +6.41756e-001
data8 0x3fe4a018bcb69834 // log(1/frcpa(1+231/256))=  +6.44543e-001
data8 0x3fe4af5a0c9d65d4 // log(1/frcpa(1+232/256))=  +6.46405e-001
data8 0x3fe4bea2a5bdbe84 // log(1/frcpa(1+233/256))=  +6.48271e-001
data8 0x3fe4cdf28f10ac44 // log(1/frcpa(1+234/256))=  +6.50140e-001
data8 0x3fe4dd49cf994058 // log(1/frcpa(1+235/256))=  +6.52013e-001
data8 0x3fe4eca86e64a680 // log(1/frcpa(1+236/256))=  +6.53889e-001
data8 0x3fe503c43cd8eb68 // log(1/frcpa(1+237/256))=  +6.56710e-001
data8 0x3fe513356667fc54 // log(1/frcpa(1+238/256))=  +6.58595e-001
data8 0x3fe522ae0738a3d4 // log(1/frcpa(1+239/256))=  +6.60483e-001
data8 0x3fe5322e26867854 // log(1/frcpa(1+240/256))=  +6.62376e-001
data8 0x3fe541b5cb979808 // log(1/frcpa(1+241/256))=  +6.64271e-001
data8 0x3fe55144fdbcbd60 // log(1/frcpa(1+242/256))=  +6.66171e-001
data8 0x3fe560dbc45153c4 // log(1/frcpa(1+243/256))=  +6.68074e-001
data8 0x3fe5707a26bb8c64 // log(1/frcpa(1+244/256))=  +6.69980e-001
data8 0x3fe587f60ed5b8fc // log(1/frcpa(1+245/256))=  +6.72847e-001
data8 0x3fe597a7977c8f30 // log(1/frcpa(1+246/256))=  +6.74763e-001
data8 0x3fe5a760d634bb88 // log(1/frcpa(1+247/256))=  +6.76682e-001
data8 0x3fe5b721d295f10c // log(1/frcpa(1+248/256))=  +6.78605e-001
data8 0x3fe5c6ea94431ef8 // log(1/frcpa(1+249/256))=  +6.80532e-001
data8 0x3fe5d6bb22ea86f4 // log(1/frcpa(1+250/256))=  +6.82462e-001
data8 0x3fe5e6938645d38c // log(1/frcpa(1+251/256))=  +6.84397e-001
data8 0x3fe5f673c61a2ed0 // log(1/frcpa(1+252/256))=  +6.86335e-001
data8 0x3fe6065bea385924 // log(1/frcpa(1+253/256))=  +6.88276e-001
data8 0x3fe6164bfa7cc068 // log(1/frcpa(1+254/256))=  +6.90222e-001
data8 0x3fe62643fecf9740 // log(1/frcpa(1+255/256))=  +6.92171e-001
LOCAL_OBJECT_END(pow_Tt)


// Table 1 is 2^(index_1/128) where
// index_1 goes from 0 to 15
LOCAL_OBJECT_START(pow_tbl1)
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(pow_tbl1)


// Table 2 is 2^(index_1/8) where
// index_2 goes from 0 to 7
LOCAL_OBJECT_START(pow_tbl2)
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(pow_tbl2)

.section .text
WEAK_LIBM_ENTRY(powf)

// Get exponent of x.  Will be used to calculate K.
{ .mfi
          getf.exp     pow_GR_signexp_X = f8
          fms.s1 POW_Xm1 = f8,f1,f1     // Will be used for r1 if x>0
          mov           pow_GR_17ones   = 0x1FFFF
}
{ .mfi
          addl          pow_AD_P        = @ltoff(pow_table_P), gp
          fma.s1 POW_Xp1 = f8,f1,f1     // Will be used for r1 if x<0
          nop.i 999
}
;;

// Get significand of x.  Will be used to get index to fetch T, Tt.
{ .mfi
          getf.sig      pow_GR_sig_X    = f8
          frcpa.s1      POW_B, p6       = f1,f8
          mov           pow_GR_exp_half = 0xFFFE   // Exponent for 0.5
}
{ .mfi
          ld8 pow_AD_P = [pow_AD_P]
          fma.s1        POW_NORM_X      = f8,f1,f0
          mov          pow_GR_exp_2tom8 = 0xFFF7
}
;;

// DOUBLE 0x10033  exponent limit at which y is an integer
{ .mfi
          nop.m 999
          fcmp.lt.s1 p8,p9 = f8, f0     // Test for x<0
          addl pow_GR_10033             = 0x10033, r0
}
{ .mfi
          mov           pow_GR_16ones   = 0xFFFF
          fma.s1        POW_NORM_Y      = f9,f1,f0
          nop.i 999
}
;;

// p13 = TRUE ==> X is unorm
{ .mfi
          setf.exp      POW_Q0_half     = pow_GR_exp_half  // Form 0.5
          fclass.m  p13,p0              = f8, 0x0b  // Test for x unorm
          adds          pow_AD_Tt       = pow_Tt - pow_table_P,  pow_AD_P
}
{ .mfi
          adds          pow_AD_Q        = pow_table_Q - pow_table_P,  pow_AD_P
          nop.f 999
          nop.i 999
}
;;

// p14 = TRUE ==> X is ZERO
{ .mfi
          ldfe          POW_P2          = [pow_AD_Q], 16
          fclass.m  p14,p0              = f8, 0x07
          nop.i 999
}
// Note POW_Xm1 and POW_r1 are used interchangably
{ .mfb
          nop.m 999
(p8)      fnma.s1        POW_Xm1        = POW_Xp1,f1,f0
(p13)     br.cond.spnt POW_X_DENORM
}
;;

// Continue normal and denormal paths here
POW_COMMON:
// p11 = TRUE ==> Y is a NAN
{ .mfi
          and           pow_GR_exp_X    = pow_GR_signexp_X, pow_GR_17ones
          fclass.m  p11,p0              = f9, 0xc3
          nop.i 999
}
{ .mfi
          nop.m 999
          fms.s1        POW_r           = POW_B, POW_NORM_X,f1
          mov pow_GR_y_zero = 0
}
;;

// Get exponent of |x|-1 to use in comparison to 2^-8
{ .mmi
          getf.exp  pow_GR_signexp_Xm1  = POW_Xm1
          sub       pow_GR_true_exp_X   = pow_GR_exp_X, pow_GR_16ones
          extr.u        pow_GR_offset   = pow_GR_sig_X, 55, 8
}
;;

{ .mfi
          alloc         r32=ar.pfs,2,19,4,0
          fcvt.fx.s1   POW_int_Y        = POW_NORM_Y
          shladd pow_AD_Tt = pow_GR_offset, 3, pow_AD_Tt
}
{ .mfi
          setf.sig POW_int_K            = pow_GR_true_exp_X
          nop.f 999
          nop.i 999
}
;;

// p12 = TRUE if Y is ZERO
// Compute xsq to decide later if |x|=1
{ .mfi
          ldfe          POW_P1          = [pow_AD_P], 16
          fclass.m      p12,p0          = f9, 0x07
          nop.i 999
}
{ .mfb
          ldfe          POW_P0          = [pow_AD_Q], 16
          fma.s1        POW_xsq = POW_NORM_X, POW_NORM_X, f0
(p11)     br.cond.spnt  POW_Y_NAN       // Branch if y=nan
}
;;

{ .mmf
          getf.exp  pow_GR_signexp_Y    = POW_NORM_Y
          ldfd  POW_T                   = [pow_AD_Tt]
          fma.s1        POW_rsq         = POW_r, POW_r,f0
}
;;

// p11 = TRUE ==> X is a NAN
{ .mfi
          ldfpd         POW_log2_hi, POW_log2_lo  = [pow_AD_Q], 16
          fclass.m      p11,p0          = POW_NORM_X, 0xc3
          nop.i 999
}
{ .mfi
          ldfe          POW_inv_log2_by_128 = [pow_AD_P], 16
          fma.s1 POW_delta              = f0,f0,f0 // delta=0 in case |x| near 1
(p12)     mov pow_GR_y_zero = 1
}
;;

{ .mfi
          ldfd   POW_Q2                 = [pow_AD_P], 16
          fnma.s1 POW_twoV              = POW_r, POW_Q0_half,f1
          and       pow_GR_exp_Xm1      = pow_GR_signexp_Xm1, pow_GR_17ones
}
{ .mfi
          nop.m 999
          fma.s1 POW_U                  = POW_NORM_Y,POW_r,f0
          nop.i 999
}
;;

// Determine if we will use the |x| near 1 path (p6) or normal path (p7)
{ .mfi
          nop.m 999
          fcvt.xf POW_K                 = POW_int_K
          cmp.lt p6,p7                  = pow_GR_exp_Xm1, pow_GR_exp_2tom8
}
{ .mfb
          nop.m 999
          fma.s1 POW_G                  = f0,f0,f0  // G=0 in case |x| near 1
(p11)     br.cond.spnt  POW_X_NAN       // Branch if x=nan and y not nan
}
;;

// If on the x near 1 path, assign r1 to r
{ .mfi
          ldfpd  POW_Q1, POW_RSHF       = [pow_AD_P], 16
(p6)      fma.s1    POW_r               = POW_r1, f1, f0
          nop.i 999
}
{ .mfb
          nop.m 999
(p6)      fma.s1    POW_rsq             = POW_r1, POW_r1, f0
(p14)     br.cond.spnt POW_X_0          // Branch if x zero and y not nan
}
;;

{ .mfi
          getf.sig pow_GR_sig_int_Y     = POW_int_Y
(p6)      fnma.s1 POW_twoV              = POW_r1, POW_Q0_half,f1
          and pow_GR_exp_Y              = pow_GR_signexp_Y, pow_GR_17ones
}
{ .mfb
          andcm pow_GR_sign_Y           = pow_GR_signexp_Y, pow_GR_17ones
(p6)      fma.s1 POW_U                  = POW_NORM_Y,POW_r1,f0
(p12)     br.cond.spnt POW_Y_0   // Branch if y=zero, x not zero or nan
}
;;

{ .mfi
          ldfe      POW_log2_by_128_lo  = [pow_AD_P], 16
(p7)      fma.s1 POW_Z2                 = POW_twoV, POW_U, f0
          nop.i 999
}
{ .mfi
          ldfe      POW_log2_by_128_hi  = [pow_AD_Q], 16
          nop.f 999
          nop.i 999
}
;;

{ .mfi
          nop.m 999
          fcvt.xf   POW_float_int_Y     = POW_int_Y
          nop.i 999
}
{ .mfi
          nop.m 999
(p7)      fma.s1 POW_G                  = POW_K, POW_log2_hi, POW_T
          adds          pow_AD_tbl1     = pow_tbl1 - pow_Tt,  pow_AD_Q
}
;;

// p11 = TRUE ==> X is NEGATIVE but not inf
{ .mfi
          nop.m 999
          fclass.m  p11,p0              = POW_NORM_X, 0x1a
          nop.i 999
}
{ .mfi
          nop.m 999
(p7)      fma.s1 POW_delta              = POW_K, POW_log2_lo, f0
          adds pow_AD_tbl2              = pow_tbl2 - pow_tbl1,  pow_AD_tbl1
}
;;

{ .mfi
          nop.m 999
(p6)      fma.s1 POW_Z                  = POW_twoV, POW_U, f0
          nop.i 999
}
{ .mfi
          nop.m 999
          fma.s1 POW_v2                 = POW_P1, POW_r,  POW_P0
          nop.i 999
}
;;

// p11 = TRUE ==> X is NEGATIVE but not inf
//    p12 = TRUE ==> X is NEGATIVE  AND  Y  already even int
//    p13 = TRUE ==> X is NEGATIVE  AND  Y possible int
{ .mfi
          nop.m 999
(p7)      fma.s1 POW_Z                  = POW_NORM_Y, POW_G, POW_Z2
(p11)     cmp.gt.unc  p12,p13           = pow_GR_exp_Y, pow_GR_10033
}
{ .mfi
          nop.m 999
          fma.s1 POW_Gpr                = POW_G, f1, POW_r
          nop.i 999
}
;;

{ .mfi
          nop.m 999
          fma.s1 POW_Yrcub              = POW_rsq, POW_U, f0
          nop.i 999
}
{ .mfi
          nop.m 999
          fma.s1 POW_p                  = POW_rsq, POW_P2, POW_v2
          nop.i 999
}
;;

// Test if x inf
{ .mfi
          nop.m 999
          fclass.m p15,p0 = POW_NORM_X,  0x23
          nop.i 999
}
// By adding RSHF (1.1000...*2^63) we put integer part in rightmost significand
{ .mfi
          nop.m 999
          fma.s1 POW_W1  = POW_Z, POW_inv_log2_by_128, POW_RSHF
          nop.i 999
}
;;

// p13 = TRUE ==> X is NEGATIVE  AND  Y possible int
//     p10 = TRUE ==> X is NEG and Y is an int
//     p12 = TRUE ==> X is NEG and Y is not an int
{ .mfi
          nop.m 999
(p13)     fcmp.eq.unc.s1 p10,p12        = POW_float_int_Y,  POW_NORM_Y
          mov pow_GR_xneg_yodd = 0
}
{ .mfi
          nop.m 999
          fma.s1 POW_Y_Gpr              = POW_NORM_Y, POW_Gpr, f0
          nop.i 999
}
;;

// p11 = TRUE ==> X is +1.0
{ .mfi
          nop.m 999
          fcmp.eq.s1 p11,p0 = POW_NORM_X, f1
          nop.i 999
}
;;

// Extract rounded integer from rightmost significand of POW_W1
// By subtracting RSHF we get rounded integer POW_Nfloat
{ .mfi
          getf.sig pow_GR_int_N        = POW_W1
          fms.s1 POW_Nfloat  = POW_W1, f1, POW_RSHF
          nop.i 999
}
{ .mfb
          nop.m 999
          fma.s1 POW_Z3                 = POW_p, POW_Yrcub, f0
(p12)     br.cond.spnt POW_X_NEG_Y_NONINT  // Branch if x neg, y not integer
}
;;

// p7  = TRUE ==> Y is +1.0
// p12 = TRUE ==> X is NEGATIVE  AND Y is an odd integer
{ .mfi
          getf.exp pow_GR_signexp_Y_Gpr = POW_Y_Gpr
          fcmp.eq.s1 p7,p0 = POW_NORM_Y, f1  // Test for y=1.0
(p10)     tbit.nz.unc  p12,p0           = pow_GR_sig_int_Y,0
}
{ .mfb
          nop.m 999
(p11)     fma.s.s0 f8 = f1,f1,f0    // If x=1, result is +1
(p15)     br.cond.spnt POW_X_INF
}
;;

// Test x and y and flag denormal
{ .mfi
          nop.m 999
          fcmp.eq.s0 p15,p0 = f8,f9
          nop.i 999
}
{ .mfb
          nop.m 999
          fma.s1 POW_e3                 = POW_NORM_Y, POW_delta, f0
(p11)     br.ret.spnt b0            // Early exit if x=1.0, result is +1
}
;;

{ .mfi
(p12)     mov pow_GR_xneg_yodd = 1
          fnma.s1 POW_f12  = POW_Nfloat, POW_log2_by_128_lo, f1
          nop.i 999
}
{ .mfb
          nop.m 999
          fnma.s1 POW_s  = POW_Nfloat, POW_log2_by_128_hi, POW_Z
(p7)      br.ret.spnt b0        // Early exit if y=1.0, result is x
}
;;

{ .mmi
          and pow_GR_index1             = 0x0f, pow_GR_int_N
          and pow_GR_index2             = 0x70, pow_GR_int_N
          shr pow_int_GR_M              = pow_GR_int_N, 7    // M = N/128
}
;;

{ .mfi
          shladd pow_AD_T1              = pow_GR_index1, 4, pow_AD_tbl1
          fma.s1 POW_q                  = POW_Z3, POW_Q1, POW_Q0_half
          add pow_int_GR_M              = pow_GR_16ones, pow_int_GR_M
}
{ .mfi
          add pow_AD_T2                 = pow_AD_tbl2, pow_GR_index2
          fma.s1 POW_Z3sq               = POW_Z3, POW_Z3, f0
          nop.i 999
}
;;

{ .mmi
          ldfe POW_T1                   = [pow_AD_T1]
          ldfe POW_T2                   = [pow_AD_T2]
          nop.i 999
}
;;

// f123 = f12*(e3+1) = f12*e3+f12
{ .mfi
          setf.exp POW_2M               = pow_int_GR_M
          fma.s1 POW_f123               = POW_e3,POW_f12,POW_f12
          nop.i 999
}
{ .mfi
          nop.m 999
          fma.s1 POW_ssq                = POW_s, POW_s, f0
          nop.i 999
}
;;

{ .mfi
          nop.m 999
          fma.s1 POW_v2                 = POW_s, POW_Q2, POW_Q1
          and pow_GR_exp_Y_Gpr          = pow_GR_signexp_Y_Gpr, pow_GR_17ones
}
;;

{ .mfi
          cmp.ne p12,p13 = pow_GR_xneg_yodd, r0
          fma.s1 POW_q                  = POW_Z3sq, POW_q, POW_Z3
          sub pow_GR_true_exp_Y_Gpr     = pow_GR_exp_Y_Gpr, pow_GR_16ones
}
;;

// p8 TRUE ==> |Y(G + r)| >= 7

// single
//     -2^7   -2^6             2^6   2^7
// -----+-----+----+ ... +-----+-----+-----
//  p8  |             p9             |  p8
//      |     |       p10      |     |

// Form signexp of constants to indicate overflow
{ .mfi
          mov         pow_GR_big_pos    = 0x1007f
          nop.f 999
          cmp.le p8,p9                  = 7, pow_GR_true_exp_Y_Gpr
}
{ .mfi
          mov         pow_GR_big_neg    = 0x3007f
          nop.f 999
          andcm pow_GR_sign_Y_Gpr       = pow_GR_signexp_Y_Gpr, pow_GR_17ones
}
;;

// Form big positive and negative constants to test for possible overflow
// Scale both terms of the polynomial by POW_f123
{ .mfi
          setf.exp POW_big_pos          = pow_GR_big_pos
          fma.s1 POW_ssq                = POW_ssq, POW_f123, f0
(p9)      cmp.le.unc p0,p10             = 6, pow_GR_true_exp_Y_Gpr
}
{ .mfb
          setf.exp POW_big_neg          = pow_GR_big_neg
          fma.s1 POW_1ps                = POW_s, POW_f123, POW_f123
(p8)      br.cond.spnt POW_OVER_UNDER_X_NOT_INF
}
;;

{ .mfi
          nop.m 999
(p12)     fnma.s1 POW_T1T2              = POW_T1, POW_T2, f0
          nop.i 999
}
{ .mfi
          nop.m 999
(p13)     fma.s1 POW_T1T2               = POW_T1, POW_T2, f0
          nop.i 999
}
;;

{ .mfi
          nop.m 999
          fma.s1 POW_v210               = POW_s, POW_v2, POW_Q0_half
          nop.i 999
}
{ .mfi
          nop.m 999
          fma.s1 POW_2Mqp1              = POW_2M, POW_q, POW_2M
          nop.i 999
}
;;

{ .mfi
          nop.m 999
          fma.s1 POW_es                 = POW_ssq, POW_v210, POW_1ps
          nop.i 999
}
{ .mfi
          nop.m 999
          fma.s1 POW_A                  = POW_T1T2, POW_2Mqp1, f0
          nop.i 999
}
;;

// Dummy op to set inexact
{ .mfi
          nop.m 999
          fma.s0 POW_tmp                = POW_2M, POW_q, POW_2M
          nop.i 999
}
;;

{ .mfb
          nop.m 999
          fma.s.s0 f8                   = POW_A, POW_es, f0
(p10)     br.ret.sptk     b0            // Exit main branch if no over/underflow
}
;;

// POSSIBLE_OVER_UNDER
// p6 = TRUE ==> Y_Gpr negative
// Result is already computed.  We just need to know if over/underflow occurred.

{ .mfb
        cmp.eq p0,p6                    = pow_GR_sign_Y_Gpr, r0
        nop.f 999
(p6)    br.cond.spnt POW_POSSIBLE_UNDER
}
;;

// POSSIBLE_OVER
// We got an answer.
// overflow is a possibility, not a certainty


// We define an overflow when the answer with
//    WRE set
//    user-defined rounding mode

// double
// Largest double is 7FE (biased double)
//                   7FE - 3FF + FFFF = 103FE
// Create + largest_double_plus_ulp
// Create - largest_double_plus_ulp
// Calculate answer with WRE set.

// single
// Largest single is FE (biased double)
//                   FE - 7F + FFFF = 1007E
// Create + largest_single_plus_ulp
// Create - largest_single_plus_ulp
// Calculate answer with WRE set.

// Cases when answer is ldn+1  are as follows:
//  ldn                   ldn+1
// --+----------|----------+------------
//              |
//    +inf          +inf      -inf
//                  RN         RN
//                             RZ

// Put in s2 (td set, wre set)
{ .mfi
        nop.m 999
        fsetc.s2 0x7F,0x42
        nop.i 999
}
;;

{ .mfi
        nop.m 999
        fma.s.s2 POW_wre_urm_f8         = POW_A, POW_es, f0
        nop.i 999
}
;;

// Return s2 to default
{ .mfi
        nop.m 999
        fsetc.s2 0x7F,0x40
        nop.i 999
}
;;

// p7 = TRUE ==> yes, we have an overflow
{ .mfi
        nop.m 999
        fcmp.ge.s1 p7, p8               =  POW_wre_urm_f8, POW_big_pos
        nop.i 999
}
;;

{ .mfi
        nop.m 999
(p8)    fcmp.le.s1 p7, p0               =  POW_wre_urm_f8, POW_big_neg
        nop.i 999
}
;;

{ .mbb
(p7)   mov pow_GR_tag                   = 30
(p7)   br.cond.spnt __libm_error_region // Branch if overflow
       br.ret.sptk     b0               // Exit if did not overflow
}
;;


POW_POSSIBLE_UNDER:
// We got an answer. input was < -2^9 but > -2^10 (double)
// We got an answer. input was < -2^6 but > -2^7  (float)
// underflow is a possibility, not a certainty

// We define an underflow when the answer with
//    ftz set
// is zero (tiny numbers become zero)
// Notice (from below) that if we have an unlimited exponent range,
// then there is an extra machine number E between the largest denormal and
// the smallest normal.
// So if with unbounded exponent we round to E or below, then we are
// tiny and underflow has occurred.
// But notice that you can be in a situation where we are tiny, namely
// rounded to E, but when the exponent is bounded we round to smallest
// normal. So the answer can be the smallest normal with underflow.
//                           E
// -----+--------------------+--------------------+-----
//      |                    |                    |
//   1.1...10 2^-3fff    1.1...11 2^-3fff    1.0...00 2^-3ffe
//   0.1...11 2^-3ffe                                   (biased, 1)
//    largest dn                               smallest normal

// Form small constant (2^-170) to correct underflow result near region of
// smallest denormal in round-nearest.

// Put in s2 (td set, ftz set)
.pred.rel "mutex",p12,p13
{ .mfi
        mov pow_GR_Fpsr = ar40          // Read the fpsr--need to check rc.s0
        fsetc.s2 0x7F,0x41
        mov pow_GR_rcs0_mask            = 0x0c00 // Set mask for rc.s0
}
{ .mfi
(p12)   mov pow_GR_tmp                  = 0x2ffff - 170
        nop.f 999
(p13)   mov pow_GR_tmp                  = 0x0ffff - 170
}
;;

{ .mfi
        setf.exp POW_eps                = pow_GR_tmp        // Form 2^-170
        fma.s.s2 POW_ftz_urm_f8         = POW_A, POW_es, f0
        nop.i 999
}
;;

// Return s2 to default
{ .mfi
        nop.m 999
        fsetc.s2 0x7F,0x40
        nop.i 999
}
;;

// p7 = TRUE ==> yes, we have an underflow
{ .mfi
        nop.m 999
        fcmp.eq.s1 p7, p0               =  POW_ftz_urm_f8, f0
        nop.i 999
}
;;

{ .mmi
(p7)    and pow_GR_rcs0  = pow_GR_rcs0_mask, pow_GR_Fpsr  // Isolate rc.s0
;;
(p7)    cmp.eq.unc p6,p0 = pow_GR_rcs0, r0    // Test for round to nearest
        nop.i 999
}
;;

// Tweak result slightly if underflow to get correct rounding near smallest
// denormal if round-nearest
{ .mfi
        nop.m 999
(p6)    fms.s.s0 f8                     = POW_A, POW_es, POW_eps
        nop.i 999
}
{ .mbb
(p7)    mov pow_GR_tag                  = 31
(p7)    br.cond.spnt __libm_error_region // Branch if underflow
        br.ret.sptk     b0               // Exit if did not underflow
}
;;

POW_X_DENORM:
// Here if x unorm. Use the NORM_X for getf instructions, and then back
// to normal path
{ .mfi
        getf.exp      pow_GR_signexp_X  = POW_NORM_X
        nop.f 999
        nop.i 999
}
;;

{ .mib
        getf.sig      pow_GR_sig_X      = POW_NORM_X
        nop.i 999
        br.cond.sptk    POW_COMMON
}
;;

POW_X_0:
// Here if x=0 and y not nan
//
// We have the following cases:
//  p6  x=0  and  y>0 and is an integer (may be even or odd)
//  p7  x=0  and  y>0 and is NOT an integer, return +0
//  p8  x=0  and  y>0 and so big as to always be an even integer, return +0
//  p9  x=0  and  y>0 and may not be integer
//  p10 x=0  and  y>0 and is an odd  integer, return x
//  p11 x=0  and  y>0 and is an even integer, return +0
//  p12 used in dummy fcmp to set denormal flag if y=unorm
//  p13 x=0  and  y>0
//  p14 x=0  and  y=0, branch to code for calling error handling
//  p15 x=0  and  y<0, branch to code for calling error handling
//
{ .mfi
        getf.sig pow_GR_sig_int_Y = POW_int_Y // Get signif of int_Y
        fcmp.lt.s1 p15,p13 = f9, f0           // Test for y<0
        and pow_GR_exp_Y = pow_GR_signexp_Y, pow_GR_17ones
}
{ .mfb
        cmp.ne p14,p0 = pow_GR_y_zero,r0      // Test for y=0
        fcvt.xf   POW_float_int_Y = POW_int_Y
(p14)   br.cond.spnt POW_X_0_Y_0              // Branch if x=0 and y=0
}
;;

// If x=0 and y>0, test y and flag denormal
{ .mfb
(p13)   cmp.gt.unc p8,p9 = pow_GR_exp_Y, pow_GR_10033 // Test y +big = even int
(p13)   fcmp.eq.s0 p12,p0 = f9,f0    // If x=0, y>0 dummy op to flag denormal
(p15)   br.cond.spnt POW_X_0_Y_NEG // Branch if x=0 and y<0
}
;;

// Here if x=0 and y>0
{ .mfi
        nop.m 999
(p9)    fcmp.eq.unc.s1 p6,p7 = POW_float_int_Y,  POW_NORM_Y // Test y=int
        nop.i 999
}
{ .mfi
        nop.m 999
(p8)    fma.s.s0 f8 = f0,f0,f0 // If x=0, y>0 and large even int, return +0
        nop.i 999
}
;;

{ .mfi
        nop.m 999
(p7)    fma.s.s0 f8  = f0,f0,f0   // Result +0 if x=0 and y>0 and not integer
(p6)    tbit.nz.unc p10,p11 = pow_GR_sig_int_Y,0 // If y>0 int, test y even/odd
}
;;

// Note if x=0, y>0 and odd integer, just return x
{ .mfb
        nop.m 999
(p11)   fma.s.s0 f8  = f0,f0,f0   // Result +0 if x=0 and y even integer
        br.ret.sptk b0            // Exit if x=0 and y>0
}
;;

POW_X_0_Y_0:
// When X is +-0 and Y is +-0, IEEE returns 1.0
// We call error support with this value

{ .mfb
        mov pow_GR_tag                  = 32
        fma.s.s0 f8                     = f1,f1,f0
        br.cond.sptk __libm_error_region
}
;;

POW_X_0_Y_NEG:
// When X is +-0 and Y is negative, IEEE returns
// X     Y           answer
// +0    -odd int    +inf
// -0    -odd int    -inf

// +0    !-odd int   +inf
// -0    !-odd int   +inf

// p6 == Y is a floating point number outside the integer.
//       Hence it is an integer and is even.
//       return +inf

// p7 == Y is a floating point number within the integer range.
//      p9  == (int_Y = NORM_Y), Y is an integer, which may be odd or even.
//           p11 odd
//              return (sign_of_x)inf
//           p12 even
//              return +inf
//      p10 == Y is not an integer
//         return +inf
//

{ .mfi
          nop.m 999
          nop.f 999
          cmp.gt  p6,p7                 = pow_GR_exp_Y, pow_GR_10033
}
;;

{ .mfi
          mov pow_GR_tag                = 33
(p7)      fcmp.eq.unc.s1 p9,p10         = POW_float_int_Y,  POW_NORM_Y
          nop.i 999
}
;;

{ .mfb
          nop.m 999
(p6)      frcpa.s0 f8,p13               = f1, f0
(p6)      br.cond.sptk __libm_error_region   // x=0, y<0, y large neg int
}
;;

{ .mfb
          nop.m 999
(p10)     frcpa.s0 f8,p13               = f1, f0
(p10)     br.cond.sptk __libm_error_region   // x=0, y<0, y not int
}
;;

// x=0, y<0, y an int
{ .mib
          nop.m 999
(p9)      tbit.nz.unc p11,p12           = pow_GR_sig_int_Y,0
          nop.b 999
}
;;

{ .mfi
          nop.m 999
(p12)     frcpa.s0 f8,p13               = f1,f0
          nop.i 999
}
;;

{ .mfb
          nop.m 999
(p11)     frcpa.s0 f8,p13               = f1,f8
          br.cond.sptk __libm_error_region
}
;;


POW_Y_0:
// Here for y zero, x anything but zero and nan
// Set flag if x denormal
// Result is +1.0
{ .mfi
        nop.m 999
        fcmp.eq.s0 p6,p0 = f8,f0    // Sets flag if x denormal
        nop.i 999
}
{ .mfb
        nop.m 999
        fma.s.s0 f8 = f1,f1,f0
        br.ret.sptk b0
}
;;


POW_X_INF:
// Here when X is +-inf

// X +inf  Y +inf             +inf
// X -inf  Y +inf             +inf

// X +inf  Y >0               +inf
// X -inf  Y >0, !odd integer +inf     <== (-inf)^0.5 = +inf !!
// X -inf  Y >0,  odd integer -inf

// X +inf  Y -inf             +0
// X -inf  Y -inf             +0

// X +inf  Y <0               +0
// X -inf  Y <0, !odd integer +0
// X -inf  Y <0, odd integer  -0

// X + inf Y=+0                +1
// X + inf Y=-0                +1
// X - inf Y=+0                +1
// X - inf Y=-0                +1

// p13 == Y negative
// p14 == Y positive

// p6 == Y is a floating point number outside the integer.
//       Hence it is an integer and is even.
//       p13 == (Y negative)
//          return +inf
//       p14 == (Y positive)
//          return +0

// p7 == Y is a floating point number within the integer range.
//      p9  == (int_Y = NORM_Y), Y is an integer, which may be odd or even.
//           p11 odd
//              p13 == (Y negative)
//                 return (sign_of_x)inf
//              p14 == (Y positive)
//                 return (sign_of_x)0
//           pxx even
//              p13 == (Y negative)
//                 return +inf
//              p14 == (Y positive)
//                 return +0

//      pxx == Y is not an integer
//           p13 == (Y negative)
//                 return +inf
//           p14 == (Y positive)
//                 return +0
//

// If x=inf, test y and flag denormal
{ .mfi
          nop.m 999
          fcmp.eq.s0 p10,p11 = f9,f0
          nop.i 999
}
;;

{ .mfi
          nop.m 999
          fcmp.lt.s0 p13,p14            = POW_NORM_Y,f0
          cmp.gt  p6,p7                 = pow_GR_exp_Y, pow_GR_10033
}
{ .mfi
          nop.m 999
          fclass.m p12,p0               = f9, 0x23 //@inf
          nop.i 999
}
;;

{ .mfi
          nop.m 999
          fclass.m p15,p0               = f9, 0x07 //@zero
          nop.i 999
}
;;

{ .mfb
          nop.m 999
(p15)     fmerge.s f8 = f1,f1      // Return +1.0 if x=inf, y=0
(p15)     br.ret.spnt b0           // Exit if x=inf, y=0
}
;;

{ .mfi
          nop.m 999
(p14)     frcpa.s1 f8,p10 = f1,f0  // If x=inf, y>0, assume result +inf
          nop.i 999
}
{ .mfb
          nop.m 999
(p13)     fma.s.s0 f8 = f0,f0,f0   // If x=inf, y<0, assume result +0.0
(p12)     br.ret.spnt b0           // Exit if x=inf, y=inf
}
;;

// Here if x=inf, and 0 < |y| < inf.  Need to correct results if y odd integer.
{ .mfi
          nop.m 999
(p7)      fcmp.eq.unc.s1 p9,p0 = POW_float_int_Y,  POW_NORM_Y // Is y integer?
          nop.i 999
}
;;

{ .mfi
          nop.m 999
          nop.f 999
(p9)      tbit.nz.unc p11,p0 = pow_GR_sig_int_Y,0  // Test for y odd integer
}
;;

{ .mfb
          nop.m 999
(p11)     fmerge.s f8 = POW_NORM_X,f8    // If y odd integer use sign of x
          br.ret.sptk b0                 // Exit for x=inf, 0 < |y| < inf
}
;;


POW_X_NEG_Y_NONINT:
// When X is negative and Y is a non-integer, IEEE
// returns a qnan indefinite.
// We call error support with this value

{ .mfb
         mov pow_GR_tag                 = 34
         frcpa.s0 f8,p6                 = f0,f0
         br.cond.sptk __libm_error_region
}
;;

POW_X_NAN:
// Here if x=nan, y not nan
{ .mfi
         nop.m 999
         fclass.m  p9,p13 = f9, 0x07 // Test y=zero
         nop.i 999
}
;;

{ .mfb
         nop.m 999
(p13)    fma.s.s0 f8 = f8,f1,f0
(p13)    br.ret.sptk  b0            // Exit if x nan, y anything but zero or nan
}
;;

POW_X_NAN_Y_0:
// When X is a NAN and Y is zero, IEEE returns 1.
// We call error support with this value.
{ .mfi
         nop.m 999
         fcmp.eq.s0 p6,p0 = f8,f0       // Dummy op to set invalid on snan
         nop.i 999
}
{ .mfb
         mov pow_GR_tag                 = 35
         fma.s.s0 f8 = f0,f0,f1
         br.cond.sptk __libm_error_region
}
;;


POW_OVER_UNDER_X_NOT_INF:

// p8 is TRUE for overflow
// p9 is TRUE for underflow

// if y is infinity, we should not over/underflow

{ .mfi
          nop.m 999
          fcmp.eq.s1     p14, p13       = POW_xsq,f1  // Test |x|=1
          cmp.eq p8,p9                  = pow_GR_sign_Y_Gpr, r0
}
;;

{ .mfi
          nop.m 999
(p14)     fclass.m.unc       p15, p0    = f9, 0x23 // If |x|=1, test y=inf
          nop.i 999
}
{ .mfi
          nop.m 999
(p13)     fclass.m.unc       p11,p0     = f9, 0x23 // If |x| not 1, test y=inf
          nop.i 999
}
;;

// p15 = TRUE if |x|=1, y=inf, return +1
{ .mfb
          nop.m 999
(p15)     fma.s.s0          f8          = f1,f1,f0 // If |x|=1, y=inf, result +1
(p15)     br.ret.spnt b0                // Exit if |x|=1, y=inf
}
;;

.pred.rel "mutex",p8,p9
{  .mfb
(p8)      setf.exp           f8 = pow_GR_17ones // If exp(+big), result inf
(p9)      fmerge.s           f8 = f0,f0         // If exp(-big), result 0
(p11)     br.ret.sptk b0                // Exit if |x| not 1, y=inf
}
;;

{ .mfb
          nop.m 999
          nop.f 999
          br.cond.sptk POW_OVER_UNDER_ERROR // Branch if y not inf
}
;;


POW_Y_NAN:
// Here if y=nan, x anything
// If x = +1 then result is +1, else result is quiet Y
{ .mfi
       nop.m 999
       fcmp.eq.s1         p10,p9        = POW_NORM_X, f1
       nop.i 999
}
;;

{ .mfi
       nop.m 999
(p10)  fcmp.eq.s0 p6,p0 = f9,f1   // Set invalid, even if x=+1
       nop.i 999
}
;;

{ .mfi
       nop.m 999
(p10)  fma.s.s0 f8 = f1,f1,f0
       nop.i 999
}
{ .mfb
       nop.m 999
(p9)   fma.s.s0 f8 = f9,f8,f0
       br.ret.sptk b0             // Exit y=nan
}
;;


POW_OVER_UNDER_ERROR:
// Here if we have overflow or underflow.
// Enter with p12 true if x negative and y odd int to force -0 or -inf

{ .mfi
         sub   pow_GR_17ones_m1         = pow_GR_17ones, r0, 1
         nop.f 999
         mov pow_GR_one                 = 0x1
}
;;

// overflow, force inf with O flag
{ .mmb
(p8)     mov pow_GR_tag                 = 30
(p8)     setf.exp POW_tmp               = pow_GR_17ones_m1
         nop.b 999
}
;;

// underflow, force zero with I, U flags
{ .mmi
(p9)    mov pow_GR_tag                  = 31
(p9)    setf.exp POW_tmp                = pow_GR_one
        nop.i 999
}
;;

{ .mfi
        nop.m 999
        fma.s.s0 f8                     = POW_tmp, POW_tmp, f0
        nop.i 999
}
;;

// p12 x is negative and y is an odd integer, change sign of result
{ .mfi
        nop.m 999
(p12)   fnma.s.s0 f8                    = POW_tmp, POW_tmp, f0
        nop.i 999
}
;;

WEAK_LIBM_END(powf)
libm_alias_float_other (__pow, pow)
#ifdef SHARED
.symver powf,powf@@GLIBC_2.27
.weak __powf_compat
.set __powf_compat,__powf
.symver __powf_compat,powf@GLIBC_2.2
#endif


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
        stfs [GR_Parameter_Y] = POW_NORM_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
        stfs [GR_Parameter_X] = POW_NORM_X // STORE Parameter 1 on stack
        add   GR_Parameter_RESULT = 0,GR_Parameter_Y    // Parameter 3 address
        nop.b 0
}
{ .mib
        stfs [GR_Parameter_Y] = f8      // 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
        ldfs  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#