On Wed, Jun 30, 2021 at 04:54:22PM +0000, Van Haaren, Harry wrote: > > -----Original Message----- > > From: Eelco Chaudron <echau...@redhat.com> > > Sent: Wednesday, June 30, 2021 3:35 PM > > To: Van Haaren, Harry <harry.van.haa...@intel.com> > > Cc: Amber, Kumar <kumar.am...@intel.com>; d...@openvswitch.org; > > i.maxim...@ovn.org; Flavio Leitner <f...@sysclose.org>; Stokes, Ian > > <ian.sto...@intel.com> > > Subject: Re: [ovs-dev] [v4 10/12] dpif-netdev/mfex: Add AVX512 based > > optimized > > miniflow extract > > > > > > > > On 30 Jun 2021, at 15:30, Van Haaren, Harry wrote: > > > > >> -----Original Message----- > > >> From: Eelco Chaudron <echau...@redhat.com> > > >> Sent: Wednesday, June 30, 2021 2:12 PM > > >> To: Amber, Kumar <kumar.am...@intel.com>; Van Haaren, Harry > > >> <harry.van.haa...@intel.com> > > >> Cc: d...@openvswitch.org; i.maxim...@ovn.org; Flavio Leitner > > <f...@sysclose.org>; > > >> Stokes, Ian <ian.sto...@intel.com> > > >> Subject: Re: [ovs-dev] [v4 10/12] dpif-netdev/mfex: Add AVX512 based > > >> optimized > > >> miniflow extract > > >> > > >> This patch was an interesting patch to review and being reminded about > > endianness, > > >> and this site, > > >> > > https://software.intel.com/sites/landingpage/IntrinsicsGuide/#text=_mm512_maskz > > >> _permutexvar_epi8&expand=4315, got me through it ;) > > > > > > Hah, yes the Intrinsics Guide is very useful for reading/investigating > > > what/how > > instructions can do. > > > Its... almost always open in a browser in some tab here! :) > > > > > > > > >> Some comments below... > > >> > > >> //Eelco > > > > > > Thanks for review, I'll snip away large chunks of code to reduce > > > verbosity. > > > > > > Regards, -Harry > > > > > > > > >> On 17 Jun 2021, at 18:27, Kumar Amber wrote: > > >> > > >>> From: Harry van Haaren <harry.van.haa...@intel.com> > > > > > > <snip> > > > > > >>> +/* AVX512-BW level permutex2var_epi8 emulation. */ > > >>> +static inline __m512i > > >>> +__attribute__((target("avx512bw"))) > > >> > > >> Are these targets universal enough for all supported compilers, if not > > >> we might > > need > > >> to move them to individual macros in compile.h. > > > > > > Yes, these are the standard gcc/clang etc compiler -m <isa level> > > > switches. > > > > > > Search for "-mavx512bw" on e.g. this GCC page, lists them all; > > > https://gcc.gnu.org/onlinedocs/gcc/x86-Options.html > > > > > > If a compiler does not understand them, we will have to #ifdef that > > > compiler out, > > > as it just doesn't support the ISA. > > > > Guess my concern is with the windows/Microsoft compiler, as I have no > > windows > > setup, I can not verify this. > > Me neither. Flavio you mentioned a windows compiler issue on the DPIF > patchset, > would you test compile here please?
I used the DPIF v13 and MFEX v4 with a quick fix replacing __builtin_ctz() with raw_ctz() and it builds fine on AppVeyor: https://ci.appveyor.com/project/fleitner/ovs/build/job/svobt835gl8wrm8q < snipped some other unrelated points > Thanks, -- fbl _______________________________________________ dev mailing list d...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-dev
