[petsc-dev] KNL MatMult performance and unrolling.

Jed Brown jed at jedbrown.org
Thu Sep 29 23:36:53 CDT 2016

Richard Mills <richardtmills at gmail.com> writes:
> My guess is that, with the Intel compiler, at least, we are better off
> letting it do the unrolling.  I'm not sure about other compilers out there.

GCC and Clang seem to generate decent code so long as loop bounds are
known statically, though they are less aggressive about vectorizing than
Intel (and typically produce significantly smaller binaries as a
result).  I've had a hell of a time getting XL compilers to do anything
without nonsensical shenanigans.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 800 bytes
Desc: not available
URL: <http://lists.mcs.anl.gov/pipermail/petsc-dev/attachments/20160929/3dff1499/attachment.sig>

More information about the petsc-dev mailing list