PGI User Forum
 SearchSearch   MemberlistMemberlist     RegisterRegister   ProfileProfile    Log inLog in 

Free OpenACC Webinar

fail with linpack benchmark on p4 arch

 
Post new topic   Reply to topic    PGI User Forum Forum Index -> Programming and Compiling
View previous topic :: View next topic  
Author Message
jasonshih



Joined: 03 Aug 2004
Posts: 32

PostPosted: Sun Mar 06, 2005 11:26 pm    Post subject: fail with linpack benchmark on p4 arch Reply with quote

Hi,

I tried to determine the linpack mflops value on p4 arch with latest pgi compiler. However, I got "inf" all the time, no matter how large the matrix size is. hereafter the simple compilation flags adopted:

FC = pgf90 -fastsse -tp p7 -Minfo=all
CC = pgf90 -fastsse -tp p7 -Minfo=all

std log of linpack:
----------------------------
norm. resid resid machep x(1) x(n)
6.49150133E+00 7.20701276E-13 2.22044605E-16 1.00000000E+00 1.00000000E+00


times are reported for matrices of order 1000
factor solve total mflops unit ratio
times for array with leading dimension of1001
0.000E+00 0.000E+00 0.000E+00 inf 0.000E+00 0.000E+00
----------------------------

any suggestion?
Thanks in advance.

BR,
J
Back to top
View user's profile
mkcolg



Joined: 30 Jun 2004
Posts: 6213
Location: The Portland Group Inc.

PostPosted: Mon Mar 07, 2005 11:08 am    Post subject: Reply with quote

Hi Jason,


What happens when you scale back optimization? Try "-O2" and "-O2 -Mscalarsse" to determine if its a x87 vs SSE precision issue and "-fast -Mscalarsse -Mcache_align" to is if its a problem with SSE vectorization ("-Mvect=sse").

Are you using the standard NetLib linpack? or some other package?

- Mat
Back to top
View user's profile
jasonshih



Joined: 03 Aug 2004
Posts: 32

PostPosted: Mon Mar 07, 2005 6:56 pm    Post subject: Reply with quote

Hi Mat,

Thanks for the feedback, however, I am dealing with p4 arch, I am wondering why fastsse fail. I've tried both reducing the optimization level as -O2 as well as adding -Mscalarsse, both are fail with inf mflops values.

BTW,
after tried with couple of compilation flags, I found raising all real variables into double precision will resolve the problem, now I can have clear picture of mflops values versus matrix size using PGI c/f90 compilers.

Thanks for the help Mat. :-)

BR,
J
Back to top
View user's profile
Display posts from previous:   
Post new topic   Reply to topic    PGI User Forum Forum Index -> Programming and Compiling All times are GMT - 7 Hours
Page 1 of 1

 
Jump to:  
You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot vote in polls in this forum


Powered by phpBB © phpBB Group