Re: clisp vs. SBCL

To: Chavdar Ivanov <ci4ic4%gmail.com@localhost>
Subject: Re: clisp vs. SBCL
From: Yorick Hardy <yorickhardy%gmail.com@localhost>
Date: Mon, 9 Aug 2021 09:46:02 +0200
Dear Chavdar,

On 2021-08-08, Chavdar Ivanov wrote:
> On Sun, 8 Aug 2021 at 19:21, Yorick Hardy <yorickhardy%gmail.com@localhost> wrote:
> >
> > Dear Chavdar,
> >
> > On 2021-08-08, Chavdar Ivanov wrote:
> > > Hi,
> > >
> > > If one wants to use xMaxima for some maths, it is strongly advisable
> > > to make sure it is using SBCL, as opposed to the default clisp. Here
> > > are the results of the tests performed on the same machine, same
> > > version of xMaxima, first with clisp, then with SBCL:
> > > .....
> > >
> > > xMaxima test, clisp
> > >
> > > Error summary:
> > > Error(s) found:
> > >   /usr/pkg/share/maxima/5.44.0/tests/rtest_rules.mac problems:
> > >     (11 13)
> > >   /usr/pkg/share/maxima/5.44.0/tests/rtest9a.mac problems:
> > >     (24 27 30 31 35 36 47 48 51 52 55 56 59 60 63 64 67 68 71 72)
> > >   /usr/pkg/share/maxima/5.44.0/tests/rtest11.mac problems:
> > >     (136 137)
> > >   /usr/pkg/share/maxima/5.44.0/tests/rtest16.mac problem:
> > >     (50)
> > >   /usr/pkg/share/maxima/5.44.0/tests/rtest3.mac problems:
> > >     (161 162 163 164 165 166 167 168 169 170)
> > > 35 tests failed out of 11,711 total tests.
> > > Real time: 2261.4172f0 sec.
> > > Run time: 2258.7747f0 sec.
> > > Space: 38127589208 Bytes
> > > GC: 9020, GC time: 736.08234f0 sec.
> > >
> > > xMaxima test, SBCL:
> > >
> > > No unexpected errors found out of 11,711 tests.
> > > Evaluation took:
> > >   131.133 seconds of real time
> > >   131.023201 seconds of total run time (127.417664 user, 3.605537 system)
> > >   [ Run times consist of 4.800 seconds GC time, and 126.224 seconds
> > > non-GC time. ]
> > >   99.92% CPU
> > >   21,598 forms interpreted
> > >   15,783 lambdas converted
> > >   339,898,359,702 processor cycles
> > >   30,489,795,840 bytes consed
> > >
> > > ...
> > >
> > > That's about 17 times faster and with no unexpected errors...
> > >
> > >
> > > --
> > > ----
> >
> > Interesting. I thought a comparison between clisp and ecl might
> > also be useful. So here are the numbers on my (much slower!) PC.
> >
> > Xmaxima, clisp:
> >
> > Error summary:
> > Error(s) found:
> >   /usr/pkg/share/maxima/5.44.0/tests/rtest_rules.mac problems:
> >     (11 13)
> >   /usr/pkg/share/maxima/5.44.0/tests/rtest9a.mac problems:
> >     (24 27 30 31 35 36 47 48 51 52 55 56 59 60 63 64 67 68 71 72)
> >   /usr/pkg/share/maxima/5.44.0/tests/rtest11.mac problems:
> >     (136 137)
> >   /usr/pkg/share/maxima/5.44.0/tests/rtest16.mac problem:
> >     (50)
> >   /usr/pkg/share/maxima/5.44.0/tests/rtest3.mac problems:
> >     (161 162 163 164 165 166 167 168 169 170)
> > 35 tests failed out of 11,711 total tests.
> > Real time: 5683.2637f0 sec.
> > Run time: 5575.9287f0 sec.
> > Space: 37989447000 Bytes
> > GC: 9113, GC time: 2085.8396f0 sec.
> >
> >
> > Xmaxima, ecl (maxima+ecl is correctly passing tests which were
> > expected to fail due to bugs):
> >
> > No unexpected errors found out of 11,711 tests.
> > Tests that were expected to fail but passed:
> >   /usr/pkg/share/maxima/5.44.0/tests/rtest1.mac problem:
> >     (115)
> >   /usr/pkg/share/maxima/5.44.0/tests/rtest6.mac problems:
> >     (43 45)
> >   /usr/pkg/share/maxima/5.44.0/tests/rtest14.mac problems:
> >     (145 201 233 234 249 250 251 252 267 297 298 307 310 312 315 319)
> >   /usr/pkg/share/maxima/5.44.0/tests/rtest16.mac problems:
> >     (50 561)
> >   /usr/pkg/share/maxima/5.44.0/tests/rtest8.mac problem:
> >     (104)
> > real time : 1809.874 secs
> > run time  : 1912.675 secs
> > gc count  : 4325 times
> > consed    : 51671573456 bytes
> 
> I haven't tried this with ecl (I use it only as a bootstrap compiler
> for SBCL, subsequently it rebuilds itself); I ran the test on the same
> machine and got:
> 
> xMaxima test, ECL:
> 
> No unexpected errors found out of 11,711 tests.
> Tests that were expected to fail but passed:
>   /usr/pkg/share/maxima/5.44.0/tests/rtest1.mac problem:
>     (115)
>   /usr/pkg/share/maxima/5.44.0/tests/rtest6.mac problems:
>     (43 45)
>   /usr/pkg/share/maxima/5.44.0/tests/rtest14.mac problems:
>     (145 201 233 234 249 250 251 252 267 297 298 307 310 312 315 319)
>   /usr/pkg/share/maxima/5.44.0/tests/rtest16.mac problems:
>     (50 561)
>   /usr/pkg/share/maxima/5.44.0/tests/rtest8.mac problem:
>     (104)
> real time : 1030.576 secs
> run time  : 1081.703 secs
> gc count  : 3494 times
> consed    : 51897350160 bytes
> 
> (Interesting that the test expects errors when using ecl, but in fact
> all the results are as expected).
> 
> The considerable difference in speed is perhaps due to SBCL being a
> compiler essentially; on top of that, recently threading was enabled -
> only SBCL's *features* contain THREAD:
> 
> SBCL:
> * *features*
> (:X86-64 :GENCGC :64-BIT :ANSI-CL :BSD :COMMON-LISP :ELF :IEEE-FLOATING-POINT
>  :LITTLE-ENDIAN :NETBSD :PACKAGE-LOCAL-NICKNAMES :SB-CORE-COMPRESSION :SB-LDB
>  :SB-PACKAGE-LOCKS :SB-THREAD :SB-UNICODE :SBCL :UNIX)
> 
> ECL:
> > *features*
> 
> (:WALKER :CDR-1 :CDR-5 :NETBSD :FORMATTER :CDR-7 :ECL-WEAK-HASH :LITTLE-ENDIAN
>  :LONG-LONG :UINT64-T :UINT32-T :UINT16-T :LONG-FLOAT :UNICODE :DFFI
>  :CLOS-STREAMS :CMU-FORMAT :UNIX :ECL-PDE :DLOPEN :CLOS :BOEHM-GC :ANSI-CL
>  :COMMON-LISP :FLOATING-POINT-EXCEPTIONS :IEEE-FLOATING-POINT
>  :PACKAGE-LOCAL-NICKNAMES :CDR-14 :PREFIXED-API :FFI :X86_64 :COMMON :ECL)
> 
> CLISP:
> [1]> *features*
> (:READLINE :REGEXP :SYSCALLS :I18N :LOOP :COMPILER :CLOS :MOP :CLISP
> :ANSI-CL :COMMON-LISP :LISP=CL :INTERPRETER :SOCKETS :GENERIC-STREAMS
> :LOGICAL-PATHNAMES :SCREEN :FFI :GETTEXT
>  :UNICODE :BASE-CHAR=CHARACTER :WORD-SIZE=64 :PC386 :UNIX)

Thank you for the information! I also tried with SBCL now and here it is about 13
times faster than CLISP, and about 4 times faster than ECL.

How did you manage to bootstrap with ECL? I had to patch SBCL as below
(it is a bit messy, especially the libffi part).

-- 
Kind regards,

Yorick Hardy

Index: lang/sbcl/Makefile
===================================================================
RCS file: /cvsroot/pkgsrc/lang/sbcl/Makefile,v
retrieving revision 1.93
diff -u -r1.93 Makefile
--- lang/sbcl/Makefile	7 Aug 2021 14:33:32 -0000	1.93
+++ lang/sbcl/Makefile	8 Aug 2021 22:17:43 -0000
@@ -39,6 +39,7 @@
 #SBCL_BOOT_SYSTEM=	"openmcl --batch"	# OpenMCL
 #SBCL_BOOT_SYSTEM=	sbcl			# SBCL
 #SBCL_BOOT_SYSTEM=	abcl			# ABCL
+#SBCL_BOOT_SYSTEM=	"ecl --norc --shell /dev/stdin"	# ECL
 
 .if !defined(SBCL_BOOT_SYSTEM)
 # Use "clisp -ansi -on-error abort" to make the build more reproducible,
@@ -48,6 +49,10 @@
 UNLIMIT_RESOURCES=	datasize
 .elif ${SBCL_BOOT_SYSTEM} == abcl
 BUILD_DEPENDS+=		abcl-[0-9]*:../../lang/abcl
+.elif ${SBCL_BOOT_SYSTEM:C/ecl .*/ecl/W} == ecl
+.include "../../lang/ecl/buildlink3.mk"
+# needed if compiled with libffi support
+.include "../../devel/libffi/buildlink3.mk"
 .endif
 
 SUBST_CLASSES+=			fix-paths
Index: lang/sbcl/distinfo
===================================================================
RCS file: /cvsroot/pkgsrc/lang/sbcl/distinfo,v
retrieving revision 1.69
diff -u -r1.69 distinfo
--- lang/sbcl/distinfo	7 Aug 2021 14:33:32 -0000	1.69
+++ lang/sbcl/distinfo	8 Aug 2021 22:17:43 -0000
@@ -13,3 +13,4 @@
 SHA1 (patch-make-target-2.sh) = 7b288c3c320adb89cefa78a2fc93223ea4b3b66f
 SHA1 (patch-slam.sh) = 2f4d28b2d1572cd624a54f493cc7e30c3c491a3f
 SHA1 (patch-src_runtime_run-program.c) = 42b3ba2e8644fa32c6b2fcb8064c8178f9da9699
+SHA1 (patch-tools-for-build_grovel-headers.c) = 5edb9bf375a5bfb99bcda9329e14bb2849455965

===================================================================



--- lang/sbcl/patches/patch-tools-for-build_grovel-headers.c.orig	2021-08-09 00:14:55.642708138 +0200
+++ lang/sbcl/patches/patch-tools-for-build_grovel-headers.c	2021-08-09 00:16:07.374143959 +0200
@@ -0,0 +1,15 @@
+$NetBSD$
+
+lang/ecl also provides gc.h, so make sure the correct header is included.
+
+--- tools-for-build/grovel-headers.c.orig	2021-07-30 08:42:10.000000000 +0000
++++ tools-for-build/grovel-headers.c
+@@ -70,7 +70,7 @@
+ #endif
+ 
+ #include "wrap.h"
+-#include "gc.h"
++#include "../runtime/gc.h"
+ 
+ #define DEFTYPE(lispname,cname) { cname foo; \
+     printf("(define-alien-type " lispname " (%s %lu))\n", \
Follow-Ups:
- Re: clisp vs. SBCL
  - From: Chavdar Ivanov
References:
- clisp vs. SBCL
  - From: Chavdar Ivanov
- Re: clisp vs. SBCL
  - From: Yorick Hardy
- Re: clisp vs. SBCL
  - From: Chavdar Ivanov
Prev by Date: Re: building rust
Next by Date: NetBSD openjdk11, ipv6 support?
Previous by Thread: Re: clisp vs. SBCL
Next by Thread: Re: clisp vs. SBCL
Indexes:
Home | Main Index | Thread Index | Old Index