> > Hey Amber, > > > > I'm OK with the API changes in this patch if we need them, but after > > looking at the later 2 patches in the series, I can see that you're > > using another check in the heart of the MFEX AVX512 code > (mfex_avx512_process()): > > /* Dummy pmd dont always pass correct md_is_valid and hence > > * need to check the tunnel data to ensure correct behaviour. > > */ > > bool tunnel = flow_tnl_dst_is_set(&md->tunnel); > > > > If we need to use the flow_tnl_dst_is_set() check, then it makes me > > think we don't need the md_is_valid bool at all. Can you check if we > > can get away with just using the flow_tnl_dst_is_set() check instead? > > If so, we don't need this patch at all as we could use the > > flow_tnl_dst_is_set() check for both the study and AVX512 MFEX > implementations. > > > > Hopefully that makes sense. > > Thanks, > > Cian > > Using md_is_valid is better as it deterministic for the compiler. And I > have removed bool tunnel = flow_tnl_dst_is_set(&md->tunnel); as it is not > required. > > Regards > Amber
Hi Amber, I agree with Cian's assessment. So, I shall wait for the next revision to review patches 7, 8, 9 in the series. Thanks and regards Sunil _______________________________________________ dev mailing list d...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-dev
