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
|
<HTML>
<HEAD>
<TITLE>Berkeley TestFloat General Documentation</TITLE>
</HEAD>
<BODY>
<H1>Berkeley TestFloat Release 3b: General Documentation</H1>
<P>
John R. Hauser<BR>
2016 July 22<BR>
</P>
<H2>Contents</H2>
<BLOCKQUOTE>
<TABLE BORDER=0 CELLSPACING=0 CELLPADDING=0>
<COL WIDTH=25>
<COL WIDTH=*>
<TR><TD COLSPAN=2>1. Introduction</TD></TR>
<TR><TD COLSPAN=2>2. Limitations</TD></TR>
<TR><TD COLSPAN=2>3. Acknowledgments and License</TD></TR>
<TR><TD COLSPAN=2>4. What TestFloat Does</TD></TR>
<TR><TD COLSPAN=2>5. Executing TestFloat</TD></TR>
<TR><TD COLSPAN=2>6. Operations Tested by TestFloat</TD></TR>
<TR><TD></TD><TD>6.1. Conversion Operations</TD></TR>
<TR><TD></TD><TD>6.2. Basic Arithmetic Operations</TD></TR>
<TR><TD></TD><TD>6.3. Fused Multiply-Add Operations</TD></TR>
<TR><TD></TD><TD>6.4. Remainder Operations</TD></TR>
<TR><TD></TD><TD>6.5. Round-to-Integer Operations</TD></TR>
<TR><TD></TD><TD>6.6. Comparison Operations</TD></TR>
<TR><TD COLSPAN=2>7. Interpreting TestFloat Output</TD></TR>
<TR>
<TD COLSPAN=2>8. Variations Allowed by the IEEE Floating-Point Standard</TD>
</TR>
<TR><TD></TD><TD>8.1. Underflow</TD></TR>
<TR><TD></TD><TD>8.2. NaNs</TD></TR>
<TR><TD></TD><TD>8.3. Conversions to Integer</TD></TR>
<TR><TD COLSPAN=2>9. Contact Information</TD></TR>
</TABLE>
</BLOCKQUOTE>
<H2>1. Introduction</H2>
<P>
Berkeley TestFloat is a small collection of programs for testing that an
implementation of binary floating-point conforms to the IEEE Standard for
Floating-Point Arithmetic.
All operations required by the original 1985 version of the IEEE Floating-Point
Standard can be tested, except for conversions to and from decimal.
With the current release, the following binary formats can be tested:
<NOBR>16-bit</NOBR> half-precision, <NOBR>32-bit</NOBR> single-precision,
<NOBR>64-bit</NOBR> double-precision, <NOBR>80-bit</NOBR>
double-extended-precision, and/or <NOBR>128-bit</NOBR> quadruple-precision.
TestFloat cannot test decimal floating-point.
</P>
<P>
Included in the TestFloat package are the <CODE>testsoftfloat</CODE> and
<CODE>timesoftfloat</CODE> programs for testing the Berkeley SoftFloat software
implementation of floating-point and for measuring its speed.
Information about SoftFloat can be found at the SoftFloat Web page,
<A HREF="http://www.jhauser.us/arithmetic/SoftFloat.html"><NOBR><CODE>http://www.jhauser.us/arithmetic/SoftFloat.html</CODE></NOBR></A>.
The <CODE>testsoftfloat</CODE> and <CODE>timesoftfloat</CODE> programs are
expected to be of interest only to people compiling the SoftFloat sources.
</P>
<P>
This document explains how to use the TestFloat programs.
It does not attempt to define or explain much of the IEEE Floating-Point
Standard.
Details about the standard are available elsewhere.
</P>
<P>
The current version of TestFloat is <NOBR>Release 3b</NOBR>.
This release differs from the previous <NOBR>Release 3a</NOBR> mainly in the
ability to test the <NOBR>16-bit</NOBR> half-precision format.
Compared to Release 2c and earlier, the set of TestFloat programs as well as
the programs’ arguments and behavior changed some with
<NOBR>Release 3</NOBR>.
For more about the evolution of TestFloat releases, see
<A HREF="TestFloat-history.html"><NOBR><CODE>TestFloat-history.html</CODE></NOBR></A>.
</P>
<H2>2. Limitations</H2>
<P>
TestFloat output is not always easily interpreted.
Detailed knowledge of the IEEE Floating-Point Standard and its vagaries is
needed to use TestFloat responsibly.
</P>
<P>
TestFloat performs relatively simple tests designed to check the fundamental
soundness of the floating-point under test.
TestFloat may also at times manage to find rarer and more subtle bugs, but it
will probably only find such bugs by chance.
Software that purposefully seeks out various kinds of subtle floating-point
bugs can be found through links posted on the TestFloat Web page,
<A HREF="http://www.jhauser.us/arithmetic/TestFloat.html"><NOBR><CODE>http://www.jhauser.us/arithmetic/TestFloat.html</CODE></NOBR></A>.
</P>
<H2>3. Acknowledgments and License</H2>
<P>
The TestFloat package was written by me, <NOBR>John R.</NOBR> Hauser.
<NOBR>Release 3</NOBR> of TestFloat was a completely new implementation
supplanting earlier releases.
The project to create <NOBR>Release 3</NOBR> (now <NOBR>through 3b</NOBR>) was
done in the employ of the University of California, Berkeley, within the
Department of Electrical Engineering and Computer Sciences, first for the
Parallel Computing Laboratory (Par Lab) and then for the ASPIRE Lab.
The work was officially overseen by Prof. Krste Asanovic, with funding provided
by these sources:
<BLOCKQUOTE>
<TABLE>
<COL>
<COL WIDTH=10>
<COL>
<TR>
<TD VALIGN=TOP><NOBR>Par Lab:</NOBR></TD>
<TD></TD>
<TD>
Microsoft (Award #024263), Intel (Award #024894), and U.C. Discovery
(Award #DIG07-10227), with additional support from Par Lab affiliates Nokia,
NVIDIA, Oracle, and Samsung.
</TD>
</TR>
<TR>
<TD VALIGN=TOP><NOBR>ASPIRE Lab:</NOBR></TD>
<TD></TD>
<TD>
DARPA PERFECT program (Award #HR0011-12-2-0016), with additional support from
ASPIRE industrial sponsor Intel and ASPIRE affiliates Google, Nokia, NVIDIA,
Oracle, and Samsung.
</TD>
</TR>
</TABLE>
</BLOCKQUOTE>
</P>
<P>
The following applies to the whole of TestFloat <NOBR>Release 3b</NOBR> as well
as to each source file individually.
</P>
<P>
Copyright 2011, 2012, 2013, 2014, 2015, 2016 The Regents of the University of
California.
All rights reserved.
</P>
<P>
Redistribution and use in source and binary forms, with or without
modification, are permitted provided that the following conditions are met:
<OL>
<LI>
<P>
Redistributions of source code must retain the above copyright notice, this
list of conditions, and the following disclaimer.
</P>
<LI>
<P>
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.
</P>
<LI>
<P>
Neither the name of the University nor the names of its contributors may be
used to endorse or promote products derived from this software without specific
prior written permission.
</P>
</OL>
</P>
<P>
THIS SOFTWARE IS PROVIDED BY THE REGENTS 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 THE REGENTS OR 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.
</P>
<H2>4. What TestFloat Does</H2>
<P>
TestFloat is designed to test a floating-point implementation by comparing its
behavior with that of TestFloat’s own internal floating-point implemented
in software.
For each operation to be tested, the TestFloat programs can generate a large
number of test cases, made up of simple pattern tests intermixed with weighted
random inputs.
The cases generated should be adequate for testing carry chain propagations,
plus the rounding of addition, subtraction, multiplication, and simple
operations like conversions.
TestFloat makes a point of checking all boundary cases of the arithmetic,
including underflows, overflows, invalid operations, subnormal inputs, zeros
(positive and negative), infinities, and NaNs.
For the interesting operations like addition and multiplication, millions of
test cases may be checked.
</P>
<P>
TestFloat is not remarkably good at testing difficult rounding cases for
division and square root.
It also makes no attempt to find bugs specific to SRT division and the like
(such as the infamous Pentium division bug).
Software that tests for such failures can be found through links on the
TestFloat Web page,
<A HREF="http://www.jhauser.us/arithmetic/TestFloat.html"><NOBR><CODE>http://www.jhauser.us/arithmetic/TestFloat.html</CODE></NOBR></A>.
</P>
<P>
NOTE!<BR>
It is the responsibility of the user to verify that the discrepancies TestFloat
finds actually represent faults in the implementation being tested.
Advice to help with this task is provided later in this document.
Furthermore, even if TestFloat finds no fault with a floating-point
implementation, that in no way guarantees that the implementation is bug-free.
</P>
<P>
For each operation, TestFloat can test all five rounding modes defined by the
IEEE Floating-Point Standard.
TestFloat verifies not only that the numeric results of an operation are
correct, but also that the proper floating-point exception flags are raised.
All five exception flags are tested, including the <I>inexact</I> flag.
TestFloat does not attempt to verify that the floating-point exception flags
are actually implemented as sticky flags.
</P>
<P>
For the <NOBR>80-bit</NOBR> double-extended-precision format, TestFloat can
test the addition, subtraction, multiplication, division, and square root
operations at all three of the standard rounding precisions.
The rounding precision can be set to <NOBR>32 bits</NOBR>, equivalent to
single-precision, to <NOBR>64 bits</NOBR>, equivalent to double-precision, or
to the full <NOBR>80 bits</NOBR> of the double-extended-precision.
Rounding precision control can be applied only to the double-extended-precision
format and only for the five basic arithmetic operations: addition,
subtraction, multiplication, division, and square root.
Other operations can be tested only at full precision.
</P>
<P>
As a rule, TestFloat is not particular about the bit patterns of NaNs that
appear as operation results.
Any NaN is considered as good a result as another.
This laxness can be overridden so that TestFloat checks for particular bit
patterns within NaN results.
See <NOBR>section 8</NOBR> below, <I>Variations Allowed by the IEEE
Floating-Point Standard</I>, plus the <CODE>-checkNaNs</CODE> option documented
for programs <CODE>testfloat_ver</CODE> and <CODE>testfloat</CODE>.
</P>
<P>
TestFloat normally compares an implementation of floating-point against the
Berkeley SoftFloat software implementation of floating-point, also created by
me.
The SoftFloat functions are linked into each TestFloat program’s
executable.
Information about SoftFloat can be found at the Web page
<A HREF="http://www.jhauser.us/arithmetic/SoftFloat.html"><NOBR><CODE>http://www.jhauser.us/arithmetic/SoftFloat.html</CODE></NOBR></A>.
</P>
<P>
For testing SoftFloat itself, the TestFloat package includes a
<CODE>testsoftfloat</CODE> program that compares SoftFloat’s
floating-point against <EM>another</EM> software floating-point implementation.
The second software floating-point is simpler and slower than SoftFloat, and is
completely independent of SoftFloat.
Although the second software floating-point cannot be guaranteed to be
bug-free, the chance that it would mimic any of SoftFloat’s bugs is low.
Consequently, an error in one or the other floating-point version should appear
as an unexpected difference between the two implementations.
Note that testing SoftFloat should be necessary only when compiling a new
TestFloat executable or when compiling SoftFloat for some other reason.
</P>
<H2>5. Executing TestFloat</H2>
<P>
The TestFloat package consists of five programs, all intended to be executed
from a command-line interpreter:
<BLOCKQUOTE>
<TABLE>
<TR>
<TD>
<A HREF="testfloat_gen.html"><CODE>testfloat_gen</CODE></A><CODE> </CODE>
</TD>
<TD>
Generates test cases for a specific floating-point operation.
</TD>
</TR>
<TR>
<TD>
<A HREF="testfloat_ver.html"><CODE>testfloat_ver</CODE></A>
</TD>
<TD>
Verifies whether the results from executing a floating-point operation are as
expected.
</TD>
</TR>
<TR>
<TD>
<A HREF="testfloat.html"><CODE>testfloat</CODE></A>
</TD>
<TD>
An all-in-one program that generates test cases, executes floating-point
operations, and verifies whether the results match expectations.
</TD>
</TR>
<TR>
<TD>
<A HREF="testsoftfloat.html"><CODE>testsoftfloat</CODE></A><CODE> </CODE>
</TD>
<TD>
Like <CODE>testfloat</CODE>, but for testing SoftFloat.
</TD>
</TR>
<TR>
<TD>
<A HREF="timesoftfloat.html"><CODE>timesoftfloat</CODE></A><CODE> </CODE>
</TD>
<TD>
A program for measuring the speed of SoftFloat (included in the TestFloat
package for convenience).
</TD>
</TR>
</TABLE>
</BLOCKQUOTE>
Each program has its own page of documentation that can be opened through the
links in the table above.
</P>
<P>
To test a floating-point implementation other than SoftFloat, one of three
different methods can be used.
The first method pipes output from <CODE>testfloat_gen</CODE> to a program
that:
<NOBR>(a) reads</NOBR> the incoming test cases, <NOBR>(b) invokes</NOBR> the
floating-point operation being tested, and <NOBR>(c) writes</NOBR> the
operation results to output.
These results can then be piped to <CODE>testfloat_ver</CODE> to be checked for
correctness.
Assuming a vertical bar (<CODE>|</CODE>) indicates a pipe between programs, the
complete process could be written as a single command like so:
<BLOCKQUOTE>
<PRE>
testfloat_gen ... <type> | <program-that-invokes-op> | testfloat_ver ... <function>
</PRE>
</BLOCKQUOTE>
The program in the middle is not supplied by TestFloat but must be created
independently.
If for some reason this program cannot take command-line arguments, the
<CODE>-prefix</CODE> option of <CODE>testfloat_gen</CODE> can communicate
parameters through the pipe.
</P>
<P>
A second method for running TestFloat is similar but has
<CODE>testfloat_gen</CODE> supply not only the test inputs but also the
expected results for each case.
With this additional information, the job done by <CODE>testfloat_ver</CODE>
can be folded into the invoking program to give the following command:
<BLOCKQUOTE>
<PRE>
testfloat_gen ... <function> | <program-that-invokes-op-and-compares-results>
</PRE>
</BLOCKQUOTE>
Again, the program that actually invokes the floating-point operation is not
supplied by TestFloat but must be created independently.
Depending on circumstance, it may be preferable either to let
<CODE>testfloat_ver</CODE> check and report suspected errors (first method) or
to include this step in the invoking program (second method).
</P>
<P>
The third way to use TestFloat is the all-in-one <CODE>testfloat</CODE>
program.
This program can perform all the steps of creating test cases, invoking the
floating-point operation, checking the results, and reporting suspected errors.
However, for this to be possible, <CODE>testfloat</CODE> must be compiled to
contain the method for invoking the floating-point operations to test.
Each build of <CODE>testfloat</CODE> is therefore capable of testing
<EM>only</EM> the floating-point implementation it was built to invoke.
To test a new implementation of floating-point, a new <CODE>testfloat</CODE>
must be created, linked to that specific implementation.
By comparison, the <CODE>testfloat_gen</CODE> and <CODE>testfloat_ver</CODE>
programs are entirely generic;
one instance is usable for testing any floating-point implementation, because
implementation-specific details are segregated in the custom program that
follows <CODE>testfloat_gen</CODE>.
</P>
<P>
Program <CODE>testsoftfloat</CODE> is another all-in-one program specifically
for testing SoftFloat.
</P>
<P>
Programs <CODE>testfloat_ver</CODE>, <CODE>testfloat</CODE>, and
<CODE>testsoftfloat</CODE> all report status and error information in a common
way.
As it executes, each of these programs writes status information to the
standard error output, which should be the screen by default.
In order for this status to be displayed properly, the standard error stream
should not be redirected to a file.
Any discrepancies that are found are written to the standard output stream,
which is easily redirected to a file if desired.
Unless redirected, reported errors will appear intermixed with the ongoing
status information in the output.
</P>
<H2>6. Operations Tested by TestFloat</H2>
<P>
TestFloat can test all operations required by the original 1985 IEEE
Floating-Point Standard except for conversions to and from decimal.
These operations are:
<UL>
<LI>
conversions among the supported floating-point formats, and also between
integers (<NOBR>32-bit</NOBR> and <NOBR>64-bit</NOBR>, signed and unsigned) and
any of the floating-point formats;
<LI>
for each floating-point format, the usual addition, subtraction,
multiplication, division, and square root operations;
<LI>
for each format, the floating-point remainder operation defined by the IEEE
Standard;
<LI>
for each format, a “round to integer” operation that rounds to the
nearest integer value in the same format; and
<LI>
comparisons between two values in the same floating-point format.
</UL>
In addition, TestFloat can also test
<UL>
<LI>
for each floating-point format except <NOBR>80-bit</NOBR>
double-extended-precision, the fused multiply-add operation defined by the 2008
IEEE Standard.
</UL>
</P>
<P>
More information about all these operations is given below.
In the operation names used by TestFloat, <NOBR>16-bit</NOBR> half-precision is
called <CODE>f16</CODE>, <NOBR>32-bit</NOBR> single-precision is
<CODE>f32</CODE>, <NOBR>64-bit</NOBR> double-precision is <CODE>f64</CODE>,
<NOBR>80-bit</NOBR> double-extended-precision is <CODE>extF80</CODE>, and
<NOBR>128-bit</NOBR> quadruple-precision is <CODE>f128</CODE>.
TestFloat generally uses the same names for operations as Berkeley SoftFloat,
except that TestFloat’s names never include the <CODE>M</CODE> that
SoftFloat uses to indicate that values are passed through pointers.
</P>
<H3>6.1. Conversion Operations</H3>
<P>
All conversions among the floating-point formats and all conversions between a
floating-point format and <NOBR>32-bit</NOBR> and <NOBR>64-bit</NOBR> integers
can be tested.
The conversion operations are:
<BLOCKQUOTE>
<PRE>
ui32_to_f16 ui64_to_f16 i32_to_f16 i64_to_f16
ui32_to_f32 ui64_to_f32 i32_to_f32 i64_to_f32
ui32_to_f64 ui64_to_f64 i32_to_f64 i64_to_f64
ui32_to_extF80 ui64_to_extF80 i32_to_extF80 i64_to_extF80
ui32_to_f128 ui64_to_f128 i32_to_f128 i64_to_f128
f16_to_ui32 f32_to_ui32 f64_to_ui32 extF80_to_ui32 f128_to_ui32
f16_to_ui64 f32_to_ui64 f64_to_ui64 extF80_to_ui64 f128_to_ui64
f16_to_i32 f32_to_i32 f64_to_i32 extF80_to_i32 f128_to_i32
f16_to_i64 f32_to_i64 f64_to_i64 extF80_to_i64 f128_to_i64
f16_to_f32 f32_to_f16 f64_to_f16 extF80_to_f16 f128_to_f16
f16_to_f64 f32_to_f64 f64_to_f32 extF80_to_f32 f128_to_f32
f16_to_extF80 f32_to_extF80 f64_to_extF80 extF80_to_f64 f128_to_f64
f16_to_f128 f32_to_f128 f64_to_f128 extF80_to_f128 f128_to_extF80
</PRE>
</BLOCKQUOTE>
Abbreviations <CODE>ui32</CODE> and <CODE>ui64</CODE> indicate
<NOBR>32-bit</NOBR> and <NOBR>64-bit</NOBR> unsigned integer types, while
<CODE>i32</CODE> and <CODE>i64</CODE> indicate their signed counterparts.
These conversions all round according to the current rounding mode as relevant.
Conversions from a smaller to a larger floating-point format are always exact
and so require no rounding.
Likewise, conversions from <NOBR>32-bit</NOBR> integers to <NOBR>64-bit</NOBR>
double-precision or to any larger floating-point format are also exact, as are
conversions from <NOBR>64-bit</NOBR> integers to <NOBR>80-bit</NOBR>
double-extended-precision and <NOBR>128-bit</NOBR> quadruple-precision.
</P>
<P>
For the all-in-one <CODE>testfloat</CODE> program, this list of conversion
operations requires amendment.
For <CODE>testfloat</CODE> only, conversions to an integer type have names that
explicitly specify the rounding mode and treatment of inexactness.
Thus, instead of
<BLOCKQUOTE>
<PRE>
<float>_to_<int>
</PRE>
</BLOCKQUOTE>
as listed above, operations converting to integer type have names of these
forms:
<BLOCKQUOTE>
<PRE>
<float>_to_<int>_r_<round>
<float>_to_<int>_rx_<round>
</PRE>
</BLOCKQUOTE>
The <CODE><round></CODE> component is one of
‘<CODE>near_even</CODE>’, ‘<CODE>near_maxMag</CODE>’,
‘<CODE>minMag</CODE>’, ‘<CODE>min</CODE>’, or
‘<CODE>max</CODE>’, choosing the rounding mode.
Any other indication of rounding mode is ignored.
The operations with ‘<CODE>_r_</CODE>’ in their names never raise
the <I>inexact</I> exception, while those with ‘<CODE>_rx_</CODE>’
raise the <I>inexact</I> exception whenever the result is not exact.
</P>
<P>
TestFloat assumes that conversions from floating-point to an integer type
should raise the <I>invalid</I> exception if the input cannot be rounded to an
integer representable in the result format.
In such a circumstance, if the result type is an unsigned integer, TestFloat
expects the result of the operation to be the type’s largest integer
value;
although, when conversion overflows for a negative input, TestFloat may also
accept a result of zero.
If the result type is a signed integer and conversion overflows, TestFloat
expects the result to be the largest-magnitude integer with the same sign as
the input.
Lastly, when a NaN is converted to a signed integer type, TestFloat allows
either the largest postive or largest-magnitude negative integer to be
returned.
Conversions to integer types are expected never to raise the <I>overflow</I>
exception.
</P>
<H3>6.2. Basic Arithmetic Operations</H3>
<P>
The following standard arithmetic operations can be tested:
<BLOCKQUOTE>
<PRE>
f16_add f16_sub f16_mul f16_div f16_sqrt
f32_add f32_sub f32_mul f32_div f32_sqrt
f64_add f64_sub f64_mul f64_div f64_sqrt
extF80_add extF80_sub extF80_mul extF80_div extF80_sqrt
f128_add f128_sub f128_mul f128_div f128_sqrt
</PRE>
</BLOCKQUOTE>
The double-extended-precision (<CODE>extF80</CODE>) operations can be rounded
to reduced precision under rounding precision control.
</P>
<H3>6.3. Fused Multiply-Add Operations</H3>
<P>
For all floating-point formats except <NOBR>80-bit</NOBR>
double-extended-precision, TestFloat can test the fused multiply-add operation
defined by the 2008 IEEE Floating-Point Standard.
The fused multiply-add operations are:
<BLOCKQUOTE>
<PRE>
f16_mulAdd
f32_mulAdd
f64_mulAdd
f128_mulAdd
</PRE>
</BLOCKQUOTE>
</P>
<P>
If one of the multiplication operands is infinite and the other is zero,
TestFloat expects the fused multiply-add operation to raise the <I>invalid</I>
exception even if the third operand is a NaN.
</P>
<H3>6.4. Remainder Operations</H3>
<P>
For each format, TestFloat can test the IEEE Standard’s remainder
operation.
These operations are:
<BLOCKQUOTE>
<PRE>
f16_rem
f32_rem
f64_rem
extF80_rem
f128_rem
</PRE>
</BLOCKQUOTE>
The remainder operations are always exact and so require no rounding.
</P>
<H3>6.5. Round-to-Integer Operations</H3>
<P>
For each format, TestFloat can test the IEEE Standard’s round-to-integer
operation.
For most TestFloat programs, these operations are:
<BLOCKQUOTE>
<PRE>
f16_roundToInt
f32_roundToInt
f64_roundToInt
extF80_roundToInt
f128_roundToInt
</PRE>
</BLOCKQUOTE>
</P>
<P>
Just as for conversions to integer types (<NOBR>section 6.1</NOBR> above), the
all-in-one <CODE>testfloat</CODE> program is again an exception.
For <CODE>testfloat</CODE> only, the round-to-integer operations have names of
these forms:
<BLOCKQUOTE>
<PRE>
<float>_roundToInt_r_<round>
<float>_roundToInt_x
</PRE>
</BLOCKQUOTE>
For the ‘<CODE>_r_</CODE>’ versions, the <I>inexact</I> exception
is never raised, and the <CODE><round></CODE> component specifies the
rounding mode as one of ‘<CODE>near_even</CODE>’,
‘<CODE>near_maxMag</CODE>’, ‘<CODE>minMag</CODE>’,
‘<CODE>min</CODE>’, or ‘<CODE>max</CODE>’.
The usual indication of rounding mode is ignored.
In contrast, the ‘<CODE>_x</CODE>’ versions accept the usual
indication of rounding mode and raise the <I>inexact</I> exception whenever the
result is not exact.
This irregular system follows the IEEE Standard’s particular
specification for the round-to-integer operations.
</P>
<H3>6.6. Comparison Operations</H3>
<P>
The following floating-point comparison operations can be tested:
<BLOCKQUOTE>
<PRE>
f16_eq f16_le f16_lt
f32_eq f32_le f32_lt
f64_eq f64_le f64_lt
extF80_eq extF80_le extF80_lt
f128_eq f128_le f128_lt
</PRE>
</BLOCKQUOTE>
The abbreviation <CODE>eq</CODE> stands for “equal” (=),
<CODE>le</CODE> stands for “less than or equal” (≤), and
<CODE>lt</CODE> stands for “less than” (<).
</P>
<P>
The IEEE Standard specifies that, by default, the less-than-or-equal and
less-than comparisons raise the <I>invalid</I> exception if either input is any
kind of NaN.
The equality comparisons, on the other hand, are defined by default to raise
the <I>invalid</I> exception only for signaling NaNs, not for quiet NaNs.
For completeness, the following additional operations can be tested if
supported:
<BLOCKQUOTE>
<PRE>
f16_eq_signaling f16_le_quiet f16_lt_quiet
f32_eq_signaling f32_le_quiet f32_lt_quiet
f64_eq_signaling f64_le_quiet f64_lt_quiet
extF80_eq_signaling extF80_le_quiet extF80_lt_quiet
f128_eq_signaling f128_le_quiet f128_lt_quiet
</PRE>
</BLOCKQUOTE>
The <CODE>signaling</CODE> equality comparisons are identical to the standard
operations except that the <I>invalid</I> exception should be raised for any
NaN input.
Similarly, the <CODE>quiet</CODE> comparison operations should be identical to
their counterparts except that the <I>invalid</I> exception is not raised for
quiet NaNs.
</P>
<P>
Obviously, no comparison operations ever require rounding.
Any rounding mode is ignored.
</P>
<H2>7. Interpreting TestFloat Output</H2>
<P>
The “errors” reported by TestFloat programs may or may not really
represent errors in the system being tested.
For each test case tried, the results from the floating-point implementation
being tested could differ from the expected results for several reasons:
<UL>
<LI>
The IEEE Floating-Point Standard allows for some variation in how conforming
floating-point behaves.
Two implementations can sometimes give different results without either being
incorrect.
<LI>
The trusted floating-point emulation could be faulty.
This could be because there is a bug in the way the emulation is coded, or
because a mistake was made when the code was compiled for the current system.
<LI>
The TestFloat program may not work properly, reporting differences that do not
exist.
<LI>
Lastly, the floating-point being tested could actually be faulty.
</UL>
It is the responsibility of the user to determine the causes for the
discrepancies that are reported.
Making this determination can require detailed knowledge about the IEEE
Standard.
Assuming TestFloat is working properly, any differences found will be due to
either the first or last of the reasons above.
Variations in the IEEE Standard that could lead to false error reports are
discussed in <NOBR>section 8</NOBR>, <I>Variations Allowed by the IEEE
Floating-Point Standard</I>.
</P>
<P>
For each reported error (or apparent error), a line of text is written to the
default output.
If a line would be longer than 79 characters, it is divided.
The first part of each error line begins in the leftmost column, and any
subsequent “continuation” lines are indented with a tab.
</P>
<P>
Each error reported is of the form:
<BLOCKQUOTE>
<PRE>
<inputs> => <observed-output> expected: <expected-output>
</PRE>
</BLOCKQUOTE>
The <CODE><inputs></CODE> are the inputs to the operation.
Each output (observed and expected) is shown as a pair: the result value
first, followed by the exception flags.
</P>
<P>
For example, two typical error lines could be
<BLOCKQUOTE>
<PRE>
-00.7FFF00 -7F.000100 => +01.000000 ...ux expected: +01.000000 ....x
+81.000004 +00.1FFFFF => +01.000000 ...ux expected: +01.000000 ....x
</PRE>
</BLOCKQUOTE>
In the first line, the inputs are <CODE>-00.7FFF00</CODE> and
<CODE>-7F.000100</CODE>, and the observed result is <CODE>+01.000000</CODE>
with flags <CODE>...ux</CODE>.
The trusted emulation result is the same but with different flags,
<CODE>....x</CODE>.
Items such as <CODE>-00.7FFF00</CODE> composed of a sign character
<NOBR>(<CODE>+</CODE>/<CODE>-</CODE>)</NOBR>, hexadecimal digits, and a single
period represent floating-point values (here <NOBR>32-bit</NOBR>
single-precision).
The two instances above were reported as errors because the exception flag
results differ.
</P>
<P>
Aside from the exception flags, there are ten data types that may be
represented.
Five are floating-point types: <NOBR>16-bit</NOBR> half-precision,
<NOBR>32-bit</NOBR> single-precision, <NOBR>64-bit</NOBR> double-precision,
<NOBR>80-bit</NOBR> double-extended-precision, and <NOBR>128-bit</NOBR>
quadruple-precision.
The remaining five types are <NOBR>32-bit</NOBR> and <NOBR>64-bit</NOBR>
unsigned integers, <NOBR>32-bit</NOBR> and <NOBR>64-bit</NOBR>
two’s-complement signed integers, and Boolean values (the results of
comparison operations).
Boolean values are represented as a single character, either a <CODE>0</CODE>
(false) or a <CODE>1</CODE> (true).
A <NOBR>32-bit</NOBR> integer is represented as 8 hexadecimal digits.
Thus, for a signed <NOBR>32-bit</NOBR> integer, <CODE>FFFFFFFF</CODE> is
−1, and <CODE>7FFFFFFF</CODE> is the largest positive value.
<NOBR>64-bit</NOBR> integers are the same except with 16 hexadecimal digits.
</P>
<P>
Floating-point values are written decomposed into their sign, encoded exponent,
and encoded significand.
First is the sign character <NOBR>(<CODE>+</CODE> or <CODE>-</CODE>),</NOBR>
followed by the encoded exponent in hexadecimal, then a period
(<CODE>.</CODE>), and lastly the encoded significand in hexadecimal.
</P>
<P>
For <NOBR>16-bit</NOBR> half-precision, notable values include:
<BLOCKQUOTE>
<TABLE CELLSPACING=0 CELLPADDING=0>
<TR><TD><CODE>+00.000 </CODE></TD><TD>+0</TD></TR>
<TR><TD><CODE>+0F.000</CODE></TD><TD> 1</TD></TR>
<TR><TD><CODE>+10.000</CODE></TD><TD> 2</TD></TR>
<TR><TD><CODE>+1E.3FF</CODE></TD><TD>maximum finite value</TD></TR>
<TR><TD><CODE>+1F.000</CODE></TD><TD>+infinity</TD></TR>
<TR><TD> </TD></TR>
<TR><TD><CODE>-00.000</CODE></TD><TD>−0</TD></TR>
<TR><TD><CODE>-0F.000</CODE></TD><TD>−1</TD></TR>
<TR><TD><CODE>-10.000</CODE></TD><TD>−2</TD></TR>
<TR>
<TD><CODE>-1E.3FF</CODE></TD>
<TD>minimum finite value (largest magnitude, but negative)</TD>
</TR>
<TR><TD><CODE>-1F.000</CODE></TD><TD>−infinity</TD></TR>
</TABLE>
</BLOCKQUOTE>
Certain categories are easily distinguished (assuming the <CODE>x</CODE>s are
not all 0):
<BLOCKQUOTE>
<TABLE CELLSPACING=0 CELLPADDING=0>
<TR>
<TD><CODE>+00.xxx </CODE></TD>
<TD>positive subnormal numbers</TD>
</TR>
<TR><TD><CODE>+1F.xxx</CODE></TD><TD>positive NaNs</TD></TR>
<TR><TD><CODE>-00.xxx</CODE></TD><TD>negative subnormal numbers</TD></TR>
<TR><TD><CODE>-1F.xxx</CODE></TD><TD>negative NaNs</TD></TR>
</TABLE>
</BLOCKQUOTE>
</P>
<P>
Likewise for other formats:
<BLOCKQUOTE>
<TABLE CELLSPACING=0 CELLPADDING=0>
<TR><TD>32-bit single</TD><TD>64-bit double</TD><TD>128-bit quadruple</TD></TR>
<TR><TD> </TD></TR>
<TR>
<TD><CODE>+00.000000 </CODE></TD>
<TD><CODE>+000.0000000000000 </CODE></TD>
<TD><CODE>+0000.0000000000000000000000000000 </CODE></TD>
<TD>+0</TD>
</TR>
<TR>
<TD><CODE>+7F.000000</CODE></TD>
<TD><CODE>+3FF.0000000000000</CODE></TD>
<TD><CODE>+3FFF.0000000000000000000000000000</CODE></TD>
<TD> 1</TD>
</TR>
<TR>
<TD><CODE>+80.000000</CODE></TD>
<TD><CODE>+400.0000000000000</CODE></TD>
<TD><CODE>+4000.0000000000000000000000000000</CODE></TD>
<TD> 2</TD>
</TR>
<TR>
<TD><CODE>+FE.7FFFFF</CODE></TD>
<TD><CODE>+7FE.FFFFFFFFFFFFF</CODE></TD>
<TD><CODE>+7FFE.FFFFFFFFFFFFFFFFFFFFFFFFFFFF</CODE></TD>
<TD>maximum finite value</TD>
</TR>
<TR>
<TD><CODE>+FF.000000</CODE></TD>
<TD><CODE>+7FF.0000000000000</CODE></TD>
<TD><CODE>+7FFF.0000000000000000000000000000</CODE></TD>
<TD>+infinity</TD>
</TR>
<TR><TD> </TD></TR>
<TR>
<TD><CODE>-00.000000 </CODE></TD>
<TD><CODE>-000.0000000000000 </CODE></TD>
<TD><CODE>-0000.0000000000000000000000000000 </CODE></TD>
<TD>−0</TD>
</TR>
<TR>
<TD><CODE>-7F.000000</CODE></TD>
<TD><CODE>-3FF.0000000000000</CODE></TD>
<TD><CODE>-3FFF.0000000000000000000000000000</CODE></TD>
<TD>−1</TD>
</TR>
<TR>
<TD><CODE>-80.000000</CODE></TD>
<TD><CODE>-400.0000000000000</CODE></TD>
<TD><CODE>-4000.0000000000000000000000000000</CODE></TD>
<TD>−2</TD>
</TR>
<TR>
<TD><CODE>-FE.7FFFFF</CODE></TD>
<TD><CODE>-7FE.FFFFFFFFFFFFF</CODE></TD>
<TD><CODE>-7FFE.FFFFFFFFFFFFFFFFFFFFFFFFFFFF</CODE></TD>
<TD>minimum finite value</TD>
</TR>
<TR>
<TD><CODE>-FF.000000</CODE></TD>
<TD><CODE>-7FF.0000000000000</CODE></TD>
<TD><CODE>-7FFF.0000000000000000000000000000</CODE></TD>
<TD>−infinity</TD>
</TR>
<TR><TD> </TD></TR>
<TR>
<TD><CODE>+00.xxxxxx</CODE></TD>
<TD><CODE>+000.xxxxxxxxxxxxx</CODE></TD>
<TD><CODE>+0000.xxxxxxxxxxxxxxxxxxxxxxxxxxxx</CODE></TD>
<TD>positive subnormals</TD>
</TR>
<TR>
<TD><CODE>+FF.xxxxxx</CODE></TD>
<TD><CODE>+7FF.xxxxxxxxxxxxx</CODE></TD>
<TD><CODE>+7FFF.xxxxxxxxxxxxxxxxxxxxxxxxxxxx</CODE></TD>
<TD>positive NaNs</TD>
</TR>
<TR>
<TD><CODE>-00.xxxxxx</CODE></TD>
<TD><CODE>-000.xxxxxxxxxxxxx</CODE></TD>
<TD><CODE>-0000.xxxxxxxxxxxxxxxxxxxxxxxxxxxx</CODE></TD>
<TD>negative subnormals</TD>
</TR>
<TR>
<TD><CODE>-FF.xxxxxx</CODE></TD>
<TD><CODE>-7FF.xxxxxxxxxxxxx</CODE></TD>
<TD><CODE>-7FFF.xxxxxxxxxxxxxxxxxxxxxxxxxxxx</CODE></TD>
<TD>negative NaNs</TD>
</TR>
</TABLE>
</BLOCKQUOTE>
</P>
<P>
The <NOBR>80-bit</NOBR> double-extended-precision values are a little unusual
in that the leading bit of precision is not hidden as with other formats.
When canonically encoded, the leading significand bit of an <NOBR>80-bit</NOBR>
double-extended-precision value will be 0 if the value is zero or subnormal,
and will be 1 otherwise.
Hence, the same values listed above appear in <NOBR>80-bit</NOBR>
double-extended-precision as follows (note the leading <CODE>8</CODE> digit in
the significands):
<BLOCKQUOTE>
<TABLE CELLSPACING=0 CELLPADDING=0>
<TR>
<TD><CODE>+0000.0000000000000000 </CODE></TD>
<TD>+0</TD>
</TR>
<TR><TD><CODE>+3FFF.8000000000000000</CODE></TD><TD> 1</TD></TR>
<TR><TD><CODE>+4000.8000000000000000</CODE></TD><TD> 2</TD></TR>
<TR>
<TD><CODE>+7FFE.FFFFFFFFFFFFFFFF</CODE></TD>
<TD>maximum finite value</TD>
</TR>
<TR><TD><CODE>+7FFF.8000000000000000</CODE></TD><TD>+infinity</TD></TR>
<TR><TD> </TD></TR>
<TR><TD><CODE>-0000.0000000000000000</CODE></TD><TD>−0</TD></TR>
<TR><TD><CODE>-3FFF.8000000000000000</CODE></TD><TD>−1</TD></TR>
<TR><TD><CODE>-4000.8000000000000000</CODE></TD><TD>−2</TD></TR>
<TR>
<TD><CODE>-7FFE.FFFFFFFFFFFFFFFF</CODE></TD>
<TD>minimum finite value</TD>
</TR>
<TR><TD><CODE>-7FFF.8000000000000000</CODE></TD><TD>−infinity</TD></TR>
</TABLE>
</BLOCKQUOTE>
</P>
<P>
Lastly, exception flag values are represented by five characters, one character
per flag.
Each flag is written as either a letter or a period (<CODE>.</CODE>) according
to whether the flag was set or not by the operation.
A period indicates the flag was not set.
The letter used to indicate a set flag depends on the flag:
<BLOCKQUOTE>
<TABLE CELLSPACING=0 CELLPADDING=0>
<TR>
<TD><CODE>v </CODE></TD>
<TD><I>invalid</I> exception</TD>
</TR>
<TR>
<TD><CODE>i</CODE></TD>
<TD><I>infinite</I> exception (“divide by zero”)</TD>
</TR>
<TR><TD><CODE>o</CODE></TD><TD><I>overflow</I> exception</TD></TR>
<TR><TD><CODE>u</CODE></TD><TD><I>underflow</I> exception</TD></TR>
<TR><TD><CODE>x</CODE></TD><TD><I>inexact</I> exception</TD></TR>
</TABLE>
</BLOCKQUOTE>
For example, the notation <CODE>...ux</CODE> indicates that the
<I>underflow</I> and <I>inexact</I> exception flags were set and that the other
three flags (<I>invalid</I>, <I>infinite</I>, and <I>overflow</I>) were not
set.
The exception flags are always written following the value returned as the
result of the operation.
</P>
<H2>8. Variations Allowed by the IEEE Floating-Point Standard</H2>
<P>
The IEEE Floating-Point Standard admits some variation among conforming
implementations.
Because TestFloat expects the two implementations being compared to deliver
bit-for-bit identical results under most circumstances, this leeway in the
standard can result in false errors being reported if the two implementations
do not make the same choices everywhere the standard provides an option.
</P>
<H3>8.1. Underflow</H3>
<P>
The standard specifies that the <I>underflow</I> exception flag is to be raised
when two conditions are met simultaneously:
<NOBR>(1) <I>tininess</I></NOBR> and <NOBR>(2) <I>loss of accuracy</I></NOBR>.
</P>
<P>
A result is tiny when its magnitude is nonzero yet smaller than any normalized
floating-point number.
The standard allows tininess to be determined either before or after a result
is rounded to the destination precision.
If tininess is detected before rounding, some borderline cases will be flagged
as underflows even though the result after rounding actually lies within the
normal floating-point range.
By detecting tininess after rounding, a system can avoid some unnecessary
signaling of underflow.
All the TestFloat programs support options <CODE>-tininessbefore</CODE> and
<CODE>-tininessafter</CODE> to control whether TestFloat expects tininess on
underflow to be detected before or after rounding.
One or the other is selected as the default when TestFloat is compiled, but
these command options allow the default to be overridden.
</P>
<P>
Loss of accuracy occurs when the subnormal format is not sufficient to
represent an underflowed result accurately.
The original 1985 version of the IEEE Standard allowed loss of accuracy to be
detected either as an <I>inexact result</I> or as a
<I>denormalization loss</I>;
however, few if any systems ever chose the latter.
The latest standard requires that loss of accuracy be detected as an inexact
result, and TestFloat can test only for this case.
</P>
<H3>8.2. NaNs</H3>
<P>
The IEEE Standard gives the floating-point formats a large number of NaN
encodings and specifies that NaNs are to be returned as results under certain
conditions.
However, the standard allows an implementation almost complete freedom over
<EM>which</EM> NaN to return in each situation.
</P>
<P>
By default, TestFloat does not check the bit patterns of NaN results.
When the result of an operation should be a NaN, any NaN is considered as good
as another.
This laxness can be overridden with the <CODE>-checkNaNs</CODE> option of
programs <CODE>testfloat_ver</CODE> and <CODE>testfloat</CODE>.
In order for this option to be sensible, TestFloat must have been compiled so
that its internal floating-point implementation (SoftFloat) generates the
proper NaN results for the system being tested.
</P>
<H3>8.3. Conversions to Integer</H3>
<P>
Conversion of a floating-point value to an integer format will fail if the
source value is a NaN or if it is too large.
The IEEE Standard does not specify what value should be returned as the integer
result in these cases.
Moreover, according to the standard, the <I>invalid</I> exception can be raised
or an unspecified alternative mechanism may be used to signal such cases.
</P>
<P>
TestFloat assumes that conversions to integer will raise the <I>invalid</I>
exception if the source value cannot be rounded to a representable integer.
In such cases, TestFloat expects the result value to be the largest-magnitude
positive or negative integer or zero as detailed earlier in
<NOBR>section 6.1</NOBR>, <I>Conversion Operations</I>.
The current version of TestFloat provides no means to alter these expectations.
</P>
<H2>9. Contact Information</H2>
<P>
At the time of this writing, the most up-to-date information about TestFloat
and the latest release can be found at the Web page
<A HREF="http://www.jhauser.us/arithmetic/TestFloat.html"><NOBR><CODE>http://www.jhauser.us/arithmetic/TestFloat.html</CODE></NOBR></A>.
</P>
</BODY>
|
