fortran compiler

John Casu casuj at valinux.com
Fri Oct 27 23:11:13 EDT 2000


> On Fri, 27 Oct 2000, Robert G. Brown wrote:
> > 

..snip..

> The same should work with Absoft.
> It would be a shame to build the Fortran interface with g77:
> You gain a factor of 4 or more in speed with ATLAS and you loose a factor
> 2-4 with g77 ...
> 

I stand corrected wrt the c compilers:

For PGCC (Portland compiler) I get the following extracted numbers from SUMMARY.LOG

         Apparent peak=299.28MFLOPS
         Apparent peak=327.75MFLOPS
      This gave performance of 157.20MFLOPS (52.53% of apparent peak)
      This gave performance of 186.68MFLOPS (56.96% of apparent peak)
      This gave performance of 161.51MFLOPS (53.97% of apparent peak)
      This gave performance of 190.77MFLOPS (58.21% of apparent peak)

For PGCC/EGCS I get:

         Apparent peak=499.79MFLOPS
         Apparent peak=499.79MFLOPS
      This gave performance of 355.84MFLOPS (71.20% of apparent peak)
      This gave performance of 416.01MFLOPS (83.24% of apparent peak)
      This gave performance of 379.84MFLOPS (76.00% of apparent peak)
      This gave performance of 411.84MFLOPS (82.40% of apparent peak)

Both runs were on the same 500MHz PIII, using Atlas3.0 beta with no patches.

For Portland I compiled with -O3. 
For PGCC/EGCS I used the default config flags which are : 
		-funroll_all_loops -O3 -fomit-frame-pointers

_______________________________________________
Beowulf mailing list
Beowulf at beowulf.org
http://www.beowulf.org/mailman/listinfo/beowulf



More information about the Beowulf mailing list