|
|
Ublas : |
From: nobody (rrossi_at_[hidden])
Date: 2007-04-19 07:05:52
hello everyone
we have been doing some attempt to parallelize the axpy product between a
compressed matrix and a vector.
we obtained some speedup but the results are far from being good... in any
case our modified version of the axpy appears to be (a little) more performant
than the one in ublas 1.33.1 which may make it interesting
does anyone have experience in the field?
please take a look to the file attached.
it can be compiled with gcc 4.2 and run with
exec_name number_of_processors size
size should be a power of 2
for example
test 1 16 will run a matrix of size 16 on one processor
test 1 13 will give a segmentation fault as 13 is not a power of 2...
any help is (very) welcome
greetings
Riccardo
-- Open WebMail Project (http://openwebmail.org)