[Gcl-devel] question about the reduce function

[Bill Six]

Hi,

I'm relatively new to Common Lisp.  GCL has been my preferred CL platform (over CMUCL and CLisp) because it generates portable C code and then calls a compiler.  I recently was trying to sum a large list of numbers using reduce, and GCL took a very long time to compute the result.  I'm trying to understand why.

In GCL, it takes ~852 seconds to sum 200000 integers:

     >(defun range (start end)
             (loop for i from start below end collect i)) 
     RANGE

     >(progn
        (setf *foo* (range 0 200000))
        (length *foo*))
     200000

     (time (reduce #'+ *foo*))

     real time       :    852.200 secs
     run-gbc time    :    839.160 secs
     child run time  :      0.000 secs 
     gbc time        :      0.100 secs
     19999900000

In CLisp, the computation took less than a second:

     [1]> (defun range (start end)
             (loop for i from start below end collect i)) 
     RANGE
     [2]> (progn
        (setf *foo* (range 0 200000))
        (length *foo*))

     200000
     [3]> (time (reduce #'+ *foo*))
     Real time: 0.109255 sec.
     Run time: 0.108007 sec.
     Space: 3107312 Bytes
     GC: 3, GC time: 0.048003 sec.
     19999900000



Why is this?  (I'm running GCL (GNU Common Lisp)  2.6.8 ANSI    Aug  2 2007 17:12:54 on Ubuntu Gutsy).  How would I go about finding out the bottleneck in gcl_seqlib.lsp?

Any advice would be much appreciated.  

Thanks,

Bill Six