On 1/15/21 1:12 PM, Stokes, Ian wrote: >> Hi Folks, >> >> <snipped> >>>> >>>> There are structs, defines and inlines in public headers that change >>>> depending on whether or not AVX* are used. I do not know if the >>>> resulting mix-and-match is safe to use, and most importantly if it is >>>> guaranteed to be supported, and I think it should be clarified first. >>> >>> Sure. We definitely need to clarify that. >>> >>> If this is true, and DPDK cannot be used if the application is built with >>> AVX512 support, but the DPDK itself was built without it, this means that >>> OVS >>> 2.14 with DPDK 19.11 is broken if users use the 'make' build system, since >>> it is >>> not possible to determine which options DPDK was built with in this case. >>> And for pkg-config use case we will have to force disable AVX512 classifier >>> in >>> OVS if '-mno-avx512f' found inside $DPDK_INCLUDE. >>> >>> Best regards, Ilya Maximets. >> >> Just checked if the ABI changes when AVX512 enabled/disabled: >> It doesn’t seem to as the check-abi.sh script reports no errors when compared >> for both 20.11 and 19.11.6 versions(make and meson). >> >> For meson builds , it was ensured that >> "Fetching value of define "__AVX512F__" : " was set for enabled case and >> unset >> for disabled case(with c_args="-mno-avx512f"). >> For Make builds , the option CONFIG_RTE_ENABLE_AVX512 was set for enabled >> case and unset for disabled. >> >> Thanks and regards, >> Sunil >> > > Thanks all for testing, have pushed this to master. Will look at backporting > to 2.14 and 2.13 as well, as this affects packaging will send on separate > patches for testing for those branches.
Thanks! _______________________________________________ dev mailing list d...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-dev
