On Thu, Jul 2, 2020 at 8:18 AM Van Haaren, Harry <harry.van.haa...@intel.com> wrote: > > > -----Original Message----- > > From: William Tu <u9012...@gmail.com> > > Sent: Wednesday, July 1, 2020 10:53 PM > > To: Van Haaren, Harry <harry.van.haa...@intel.com> > > Cc: ovs-dev <ovs-dev@openvswitch.org>; Stokes, Ian <ian.sto...@intel.com>; > > Ilya > > Maximets <i.maxim...@ovn.org>; Federico Iezzi <fie...@redhat.com> > > Subject: Re: [PATCH v5 0/6] DPCLS Subtable ISA Optimization > > > > On Wed, Jul 1, 2020 at 7:22 AM Harry van Haaren > > <harry.van.haa...@intel.com> wrote: > > > > > > v5 work done: > > > - Add NEWS section > > > - Integrate v4 patches 5 and 6 into a single commit > > > - Address v4 feedback (details in each patch) > > > -- Fixed multiple travis build issues/warnings > > > -- Improved APIs/code, refactored #ifdefs > > > > > > @William Tu: > > > Thanks for pushing v4 commits to travis for testing, and reporting back. > > > There are 2 items that I could not reproduce, and could not reliably > > > state they are real issues, or 2nd order issues. This v5 has addressed > > > the actual errors from Travis output, there is a chance these issues > > > are not resolved by the surrounding fixes: > > > 1) RTE_CPUFLAG_AVX512F flag not defined in a DPDK=1 build? > > > 2) Binutils check for as with --64 parameter is invalid on older as? > > > > > > Would you re-push this branch and we can verify? Thanks in advance! > > Hi Harry, > > This is the v5 on travis. > > https://travis-ci.org/github/williamtu/ovs-travis/builds/704102981 > > William > > Thanks William! > > CPUFLAG was a cross-build issue, resolved with a #ifdef __x86_64__ > Binutils check has been fixed with a simple rm during configure time > > Working on the remaining 2 known issues, and will push a v6 ASAP once > resolved. > > You asked about the "make check-system-userspace" command on the v4, > I see this failing on ovs-master, and (still) failing with the AVX512 > patchset applied. > Hi Harry,
I checked your v6 using "make check-system-userspace" (with autovalidtor enabled) and it works OK! I will continue looking at your v6. Regards, William > The development of AVX512 work was done against "make check" and the datapath > tests there (combined with DPCLS Autovalidator) they provided value in > detecting > real issues during development. I've run an LCOV report on the make check > unit-tests, > and branch/line coverage of the AVX512 work is very high with Autovalidator > enabled. > > Regards, and v6 soon. -Harry _______________________________________________ dev mailing list d...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-dev
