Odd Target Behavior in PGI 13.5

All,

I was wondering if anyone else is seeing this. I was doing some compiling for both Westmere and Sandybridge and noticed an oddity. I have a make that is essentially:

mpif90 -fast -Kieee -Minfo -fpic -Mbackslash -Ktrap=fp -tp=nehalem-64,sandybridge-64 -Iincludedirs sorad.F90

Now, with -V13.4, part of the -Minfo says:

deledd:
   1689, PGI Unified Binary version for -tp=sandybridge-64
deledd:
   1689, PGI Unified Binary version for -tp=nehalem-64

Okay. But with -V13.5:

deledd:
   1689, PGI Unified Binary version for -tp=ivybridge-64
deledd:
   1689, PGI Unified Binary version for -tp=bulldozer-64

Bwuh? I mean, I guess I can see ivybridge-64 being sort of equivalent to sandybridge-64, but is bulldozer-64 really equivalent to nehalem-64?

Matt

Hi Matt,

Known issue that was caught internally (TPR#19431). Looks like they have it fixed with the fix being available in 13.7.

Thanks!
Mat

Mat,

Thanks for the reply. One question: is it actually making bulldozer code, say, on Westmeres, or just printing that it is?

Matt

It should be just printing. I looked at profile and it looked like the correct code generation and run time paths were being generated.

  • Mat