1
RELEASE NOTES FOR 2.5.1:
2
========================
3
The GNU Common Lisp (GCL) development team is pleased to release
4
Version 2.5.1, the first major release since the untimely death of the
5
former maintainer Dr William Schelter over a year ago. This release
6
is dedicated to his memory. The project is now hosted on
7
http://savannah.gnu.org/projects/gcl/ and is maintained and developed
8
by a team of thirteen programmers. Our home page lives at
9
http://www.gnu.org/software/gcl/.
11
This release stabilizes the CLtL1 compliant build of GCL on most major
12
Unices including 11 Debian Linux 64 and 32 bit architectures and modern
13
versions of Microsoft Windows (TM). A rapidly progressing, partially ANSI
14
compliant version is also available on the Linux platforms.
16
GCL plays a substantial role in development of the Maxima computer
17
algebra system (http://maxima.sourceforge.net/), ACL2, a computational
18
logic system (http://www.cs.utexas.edu/users/moore/acl2/), and the
19
forthcoming public release of the Axiom computer algebra system.. The
20
compiler is a descendant of the famous KCL and AKCL Common Lisp
21
compilers and is licensed under version two of the GNU Library General
24
As with any Lisp system GCL is a lot of fun to work with. We welcome all
25
comments and feedback. Developers are particularly welcome too. You will
26
find that the project offers a wide variety of challenges on various
27
platforms to anyone with an interest in compilers, low level C programming
34
* Compiles itself, maxima, and acl2, passing all tests, on 11
35
Debian GNU/Linux platforms (i386, sparc, powerpc, s390, ia64, alpha,
36
mips, mipsel, hppa, arm, and m68k), Sparc Solaris, and recent Windows
39
* Compilation to native object code. Lisp disassembly shows intermediate
40
C source and native assembler.
42
* Native code relocation on all supported platforms except alpha, mips,
43
mipsel, ia64, and hppa.
45
* Can save its running memory image to a file on all systems where native object
46
code relocation is supported, thus producing standalone executables.
48
* Compiles Lisp function calls to C function calls with inlined
49
arguments, when function proclamation/declamations are made.
51
* Quite fast, particularly if one pre-allocates memory to be commensurate
52
with that typically available on modern computer systems. (see below)
54
* A foreign function interface as flexible in principle as the C interface.
56
* Socket support via streams
58
* Support for numbers of arbitrary precision via the GNU
59
Multiprecision Library. If you build GCL on your own system,
60
multiprecision numerical support will make use of ISA extension
61
instructions available on your system for maximum large number
64
* An exact garbage collector with no (known) leaks.
66
* An ANSI mode on Unix systems which passes approximately 97% of
67
the ANSI compliance tests currently developed for the project.
68
On Debian GNU/Linux systems, this mode can be selected by setting the
69
GCL_ANSI environment variable to any non-empty string. See
70
/usr/share/doc/gcl/test_results on Debian GNU/Linux systems.
72
* An MPI extension for cluster computing support. See the website for details.
74
* A long history of leveraging GCC compiler technology for use in
75
production lisp applications.
79
GCL is one of the oldest Lisp systems still in use, and as such has
80
served as the basis for large lisp applications when computers were
81
much more limited than they are today, particularly in terms of
82
available memory. Considerable effort was therefore made in the past
83
to keep the memory image as small as possible. As of the present
84
time, the GCL team has not tuned the default memory allocation scheme
85
to be more in line with modern systems. One can therefore often get
86
significant performance increases by preallocating memory, as in for
90
(si::allocate 'cons 10000 t)
91
(si::allocate 'fixnum 200 t)
92
(si::allocate 'symbol 100 t)
93
(si::allocate-relocatable-pages 2000 t)
94
(si::allocate 'cfun 1000 t))
96
Optimal values will no doubt vary by application and machine. One
97
user/developer reports effects of the following magnitude when
102
Take a look on some funny numbers below. This is time and RAM
103
required to compute ratsimp((x+y+z)^300)$ on Linux AthlonXP 2400+.
104
For GCL run time is in the form T - G = N, where T is the total
105
time as shown by showtime:true; G is total GC tome and N
106
is run time without GC.
108
Lisp Time RAM RAM RAM
109
[sec] before max after
110
T - G = N [Mb] [Mb] [Mb]
111
=====================================================
117
GCL class 5.9 - 5.2 = 0.7 8 24 24
118
GCL ansi 9.5 - 8.9 = 0.6 9.5 29 29
120
GCL class 1.0 - 0.4 = 0.6 24 31 31
121
GCL ansi 1.1 - 0.6 = 0.5 25 32 32
123
GCL class 0.7 - 0.1 = 0.6 48 55 55
124
GCL ansi 0.5 - 0.0 = 0.5 49 56 56
126
====================================================
133
1) Full ANSI compliance
134
2) Native optimized blas support
135
3) Integrate MPI support
136
4) GCL as a suported GCC front end.
137
5) Performance/memory optimization