[gambit-list] 64-bit bignum times

[Bradley Lucier]

Marc:

I've run my bignum tests on a 2GHz G5 running MacOS 10.4.1 with  
Apple's compiler

gcc version 4.0.0 (Apple Computer, Inc. build 5018)

from CVS sources, and on a 2.4GHz Opteron running the latest Red Hat  
Enterprise Linux with gcc-4.0.0.

On the Mac, Gambit was configured with

env CC='gcc -m64 -mcpu=970  -force_cpusubtype_ALL' ./configure -- 
enable-single-host

and on the Opteron, it was just

./configure --enable-single-host

but -m64 and -mcpu=whatever are the defaults.

The source and the result files are below.

Basically, cycle for cycle, the Mac beats the Opteron; it's also  
available at higher speeds and is cheaper.  A lot of the time is  
spent in FFTs for the bignum multiplies, so this is not a bad test  
for 64-bit scientific computing capabilities.

Brad

PS:  I Bcc'ed the other people receiving this so their e-mail  
addresses won't show up on the public mail list.



-------------- next part --------------
A non-text attachment was scrubbed...
Name: gambit-trace2.scm
Type: application/octet-stream
Size: 7185 bytes
Desc: not available
URL: <http://mailman.iro.umontreal.ca/pipermail/gambit-list/attachments/20050524/7cc5e5fe/attachment.obj>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: G5-results
Type: application/octet-stream
Size: 6536 bytes
Desc: not available
URL: <http://mailman.iro.umontreal.ca/pipermail/gambit-list/attachments/20050524/7cc5e5fe/attachment-0001.obj>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: X86-64-results
Type: application/octet-stream
Size: 6489 bytes
Desc: not available
URL: <http://mailman.iro.umontreal.ca/pipermail/gambit-list/attachments/20050524/7cc5e5fe/attachment-0002.obj>