maxima (sbcl) arithmetic is very slow

[tornaria]

Maybe coming from slow arithmetic in sbcl:

$ rmaxima
Maxima 5.45.1 https://maxima.sourceforge.io
using Lisp SBCL 2.1.10
Distributed under the GNU Public License. See the file COPYING.
Dedicated to the memory of William Schelter.
The function bug_report() provides bug reporting information.
(%i1) showtime : true $
Evaluation took 0.0001 seconds (0.0000 elapsed) using 0 bytes.
(%i2) a : 10^(10^5) $
Evaluation took 0.0192 seconds (0.0200 elapsed) using 96.047 KB.
(%i3) b : a^10 $
Evaluation took 1.1198 seconds (1.1194 elapsed) using 1023.875 KB.
(%i4) b : a^20 $
Evaluation took 4.5570 seconds (4.5607 elapsed) using 2.147 MB.
(%i5) 

Contrast with maxima-ecl:

$ rmaxima -l ecl
;;; Loading #P"/usr/lib64/ecl-21.2.1/sb-bsd-sockets.fas"
;;; Loading #P"/usr/lib64/ecl-21.2.1/sockets.fas"
Maxima 5.45.1 https://maxima.sourceforge.io
using Lisp ECL 21.2.1
Distributed under the GNU Public License. See the file COPYING.
Dedicated to the memory of William Schelter.
The function bug_report() provides bug reporting information.
(%i1) showtime : true $
Evaluation took 0.0000 seconds (0.0000 elapsed)
(%i2) a : 10^(10^5) $
Evaluation took 0.0030 seconds (0.0020 elapsed)
(%i3) b : a^10 $
Evaluation took 0.0420 seconds (0.0440 elapsed)
(%i4) b : a^20 $
Evaluation took 0.0680 seconds (0.0450 elapsed)
(%i5) b : a^40 $
Evaluation took 0.0750 seconds (0.0780 elapsed)
(%i6) b : a^80 $
Evaluation took 0.2420 seconds (0.2000 elapsed)
(%i7) b : a^160 $
Evaluation took 0.4430 seconds (0.3890 elapsed)
(%i8) b : a^320 $
Evaluation took 0.8530 seconds (0.7970 elapsed)
(%i9) b : a^640 $
Evaluation took 1.7850 seconds (1.7040 elapsed)
(%i10) ) b : a^1280 $
Evaluation took 3.7550 seconds (3.6670 elapsed)
(%i11) 

Note: ecl compiles with gmp so that's probably where the long arithmetic comes from.

[tornaria]

This seems to be due to slow default arithmetic in sbcl:

$ sbcl 
This is SBCL 2.2.0, an implementation of ANSI Common Lisp.
[...]
* (* 0 (expt (expt 10 (expt 10 5)) 30))
[takes ~ 10s]
0

Now

$ sbcl 
This is SBCL 2.2.0, an implementation of ANSI Common Lisp.
[...]
* (require :sb-gmp)
("SB-GMP")
* (sb-gmp:install-gmp-funs)
* (* 0 (expt (expt 10 (expt 10 5)) 30))
[takes < 0.5 s]]
0

The solution might be to somehow link maxima to sb-gmp and run (sb-gmp:install-gmp-funs) at startup.

[tornaria]

Indeed:

$ rmaxima 
Maxima 5.45.1 https://maxima.sourceforge.io
using Lisp SBCL 2.2.0
Distributed under the GNU Public License. See the file COPYING.
Dedicated to the memory of William Schelter.
The function bug_report() provides bug reporting information.
(%i1) :lisp (require :sb-gmp "/usr/lib/sbcl/contrib/sb-gmp.fasl")

(SB-GMP)
(%i1) :lisp (sb-gmp:install-gmp-funs)

(%i1) showtime : true $
Evaluation took 0.0001 seconds (0.0000 elapsed) using 0 bytes.
(%i2) a : 10^(10^5) $
Evaluation took 0.0015 seconds (0.0010 elapsed) using 3.578 KB.
(%i3) b : a^1280 $
Evaluation took 1.5059 seconds (1.5080 elapsed) using 50.689 MB.
(%i4) 

Caveats:

• this only works if sbcl is installed (for the file "/usr/lib/sbcl/contrib/sb-gmp.fasl")
• moreover the version of sbcl used to compile maxima has to match the version of sbcl used to compile sb-gmp.fasl (rn maxima was compiled with 2.1.11 but sbcl is at 2.2.0)
• the "require" statement needs the full path to sb-gmp.fasl (not a real problem)

[tornaria]

I had success with the following patch

--- a/src/init-cl.lisp	2021-05-15 20:20:53.000000000 -0300
+++ b/src/init-cl.lisp	2022-01-04 17:45:03.838273626 -0300
@@ -574,8 +574,13 @@
 			 (delete-file file)))))
 	   *temp-files-list*))
 
+#+sbcl
+(eval-when (:compile-toplevel :load-toplevel :execute)
+  (require 'sb-gmp))
+
 (defun cl-user::run ()
   "Run Maxima in its own package."
+  #+sbcl (sb-gmp:install-gmp-funs)
   (in-package :maxima)
   (initialize-runtime-globals)
   (let ((input-stream *standard-input*)

Now maxima is automatically initialized using sb-gmp and arithmetic is fast:

$ rmaxima 
Maxima 5.45.1 https://maxima.sourceforge.io
using Lisp SBCL 2.2.0
Distributed under the GNU Public License. See the file COPYING.
Dedicated to the memory of William Schelter.
The function bug_report() provides bug reporting information.
(%i1) showtime : true $
Evaluation took 0.0001 seconds (0.0000 elapsed) using 0 bytes.
(%i2) a : 10^(10^5) $
Evaluation took 0.0017 seconds (0.0020 elapsed) using 13.516 KB.
(%i3) b : a^1280 $
Evaluation took 1.5245 seconds (1.5270 elapsed) using 50.689 MB.
(%i4) 

I want to try to fix an (unrelated) bug with the heap on maxima-ecl (#34030 (comment)); I'll make a single PR with both.