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
|
Copyright 1999, 2001-2019 Free Software Foundation, Inc.
Contributed by the AriC and Caramba projects, INRIA.
This file is part of the GNU MPFR Library.
The GNU MPFR Library is free software; you can redistribute it and/or modify
it under the terms of the GNU Lesser General Public License as published by
the Free Software Foundation; either version 3 of the License, or (at your
option) any later version.
The GNU MPFR Library is distributed in the hope that it will be useful, but
WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
or FITNESS FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public
License for more details.
You should have received a copy of the GNU Lesser General Public License
along with the GNU MPFR Library; see the file COPYING.LESSER. If not, see
https://www.gnu.org/licenses/ or write to the Free Software Foundation, Inc.,
51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA.
##############################################################################
Known bugs:
* The overflow/underflow exceptions may be badly handled in some functions;
specially when the intermediary internal results have exponent which
exceeds the hardware limit (2^30 for a 32 bits CPU, and 2^62 for a 64 bits
CPU) or the exact result is close to an overflow/underflow threshold.
* Under Linux/x86 with the traditional FPU, some functions do not work
if the FPU rounding precision has been changed to single (this is a
bad practice and should be useless, but one never knows what other
software will do).
* Some functions do not use MPFR_SAVE_EXPO_* macros, thus do not behave
correctly in a reduced exponent range.
* Function hypot gives incorrect result when on the one hand the difference
between parameters' exponents is near 2*MPFR_EMAX_MAX and on the other hand
the output precision or the precision of the parameter with greatest
absolute value is greater than 2*MPFR_EMAX_MAX-4.
Potential bugs:
* Possible incorrect results due to internal underflow, which can lead to
a huge loss of accuracy while the error analysis doesn't take that into
account. If the underflow occurs at the last function call (just before
the MPFR_CAN_ROUND), the result should be correct (or MPFR gets into an
infinite loop). TODO: check the code and the error analysis.
* Possible bugs with huge precisions (> 2^30) and a 32-bit ABI, in particular
undetected integer overflows. TODO: use the MPFR_ADD_PREC macro.
* Possible bugs if the chosen exponent range does not allow to represent
the range [1/16, 16].
* Possible infinite loop in some functions for particular cases: when
the exact result is an exactly representable number or the middle of
consecutive two such numbers. However for non-algebraic functions, it is
believed that no such case exists, except the well-known cases like cos(0)=1,
exp(0)=1, and so on, and the x^y function when y is an integer or y=1/2^k.
* The mpfr_set_ld function may be quite slow if the long double type has an
exponent of more than 15 bits.
* mpfr_set_d may give wrong results on some non-IEEE architectures.
* Error analysis for some functions may be incorrect (out-of-date due
to modifications in the code?).
* Possible use of non-portable feature (pre-C99) of the integer division
with negative result.
|
