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
|
Documentation:
- add a description of the algorithms used + proof of correctness
Installation:
- from Kevin Ryde <user42@zip.com.au>:
Determine the exp2/exp3 thresholds using tune/tuneup.c.
Changes in existing functions:
- invert 2nd and 3rd arguments of mpfr_round_prec (to be coherent with
mpfr_set_prec)
- merge mpfr_inp_str and mpfr_set_str (cf glibc sscanf/fscanf)
- remove mpfr_set_machine_rnd_mode from the documentation,
and move it to the test directory
- in mpfr_set_str, make string comparisons case insensitive and use @Inf@
and @NaN@ instead of the possibly ambiguous Inf and NaN in bases > 10.
Modify the other functions to make them consistent with this choice.
Possibly accept other strings, like those accepted by strtod.
- implement ternary flag for mpfr_agm: result is exact when u=v, or u=0,
or v=0.
New functions to implement:
- those from LIA: missing secant, cosecant, cotangent (trigo/hyperbolic)
- nextafter/nextforward : X + epsilon if X < Y, X - epsilon if X > Y
- nextabove/nextbelow ?
- mpfr_get_exp / mpfr_set_exp
- mpfr_greater_p (Kevin, Vincent ok)
= < > unordered
mpfr_greater_p 0 0 1 0
mpfr_greaterequal_p 1 0 1 0
mpfr_less_p 0 1 0 0
mpfr_lessequal_p 1 1 0 0
mpfr_lessgreater_p 0 1 1 0
mpfr_equal_p 1 0 0 0
mpfr_unordered_p 0 0 0 1
- mpfr_set_f (MPF -> float) or mpfr_strtof, mpfr_set_ld or mpfr_strtold
(idem for mpfr_get_*)
Rounding:
- mpfr_pow -> loops when exact (for example 4^(1/2)).
- mpfr_pow isn't completely specified (concerning signed zeros).
Efficiency:
- implement range reduction in sin/cos/tan for large arguments
(currently too slow for 2^1024)
- mpfr_asin/acos are too slow for small values (2^(-1021) for example)
- idem for mpfr_atanh (2^(-1021) for example)
Miscellaneous:
- rename mpf2mpfr.h to gmp-mpf2mpfr.h?
- add mpfr_get_ld for 'long double' [asked by J-C Fauge`re] ?
(exists since K&R, but has not necessarily a greater precision than double)
cf http://anubis.dkuug.dk/jtc1/sc22/wg14/www/docs/n869/
[cf above]
- from Kevin Ryde <user42@zip.com.au>:
Also for pi.c, a pre-calculated compiled-in pi to a few thousand
digits would be good value I think. After all, say 10000 bits using
1250 bytes would still be small compared to the code size!
Store pi in round to zero mode (to recover other modes).
- problem when reading a float followed by a character, for example 1.5*x
[from Fabrice.Rouillier@loria.fr, Mon, 04 Dec 2000]
- rewrite mpfr_get_str and mpfr_set_str to use mpn_get_str and mpn_set_str.
(Torbjorn Granlund <tege@swox.com>, 30 Jan 2002)
- rename (and rewrite) mpfr_isinteger to mpfr_integer_p.
(Kevin Ryde, 05 Mar 2002)
- use GMP rand/srand functions instead for SEED_RAND (time(NULL)). Kevin:
GMP tests/misc.c has a scheme to print the seed chosen and allow for a
re-run with a given seed. If you moved towards _gmp_rand and friends
at some stage you could probably use that scheme almost unchanged.
An advantage of _gmp_rand of course is that it's the same on all
systems, apart from limb size, so you're not at the mercy of random or
lrand48, etc.
|
