So there's more patches in the pipeline to land, then? On Sun, Jun 25, 2017 at 10:36 PM, Das, Dibyendu <Dibyendu.Das at amd.com> wrote:> Scheduler - yes. > > Sent from my Windows Phone > ________________________________ > From: Carsten Mattner via llvm-dev > Sent: 6/25/2017 3:47 PM > To: llvm-dev > Subject: [llvm-dev] Zen arch in 5.0? > > Will 5.0 have scheduler, reg alloc etc. bits for explicit > support of AMD's 1st gen Zen (Ryzen, Epyc) arch? > > Is it safe to assume that the changes for Zen's 2nd gen > will land prior to the hardware release, once the 1st > gen is added? I think zen2 will arrive next year.
We are preparing the patches for the scheduler. Regarding register allocator I don’t think there's anything big. Maybe minor tweaks. We need to work a bit more on the TTI and cost model though as there are differences between Intel and AMD on latencies/micro-coding of certain instructions. Cant really comment on zen2 specifics at this point. -----Original Message----- From: Carsten Mattner [mailto:carstenmattner at gmail.com] Sent: Monday, June 26, 2017 7:23 AM To: Das, Dibyendu <Dibyendu.Das at amd.com> Cc: llvm-dev <llvm-dev at lists.llvm.org> Subject: Re: [llvm-dev] Zen arch in 5.0? So there's more patches in the pipeline to land, then? On Sun, Jun 25, 2017 at 10:36 PM, Das, Dibyendu <Dibyendu.Das at amd.com> wrote:> Scheduler - yes. > > Sent from my Windows Phone > ________________________________ > From: Carsten Mattner via llvm-dev > Sent: 6/25/2017 3:47 PM > To: llvm-dev > Subject: [llvm-dev] Zen arch in 5.0? > > Will 5.0 have scheduler, reg alloc etc. bits for explicit support of > AMD's 1st gen Zen (Ryzen, Epyc) arch? > > Is it safe to assume that the changes for Zen's 2nd gen will land > prior to the hardware release, once the 1st gen is added? I think zen2 > will arrive next year.
On Fri, Jun 30, 2017 at 6:14 AM, Das, Dibyendu <Dibyendu.Das at amd.com> wrote:> We are preparing the patches for the scheduler. Regarding register allocator I > don’t think there's anything big. Maybe minor tweaks. We need to work a bit > more on the TTI and cost model though as there are differences between Intel > and AMD on latencies/micro-coding of certain instructions.Thanks for the info. In fact, I haven't checked gcc at all and since I have your ear, I'm thinking out loud wondering how complete gcc 7.1's zen awareness is. I remember diffs landing in gcc around mid-2015.> Cant really comment on zen2 specifics at this point.No need to talk details. I'm only hoping that the diffs will be applied early enough to land in a release that happens before zen2 reaches customers. This is important for Linux distros but of utmost use for FreeBSD and other systems that have clang as the default host compiler. Apple is special in that they coordinate hardware and toolchains and are irrelevant there, but the ability to use optimal march=native on Gentoo etc. is more than nice to have. Hope this explain where I was trying to get with the question. Thanks and keep up the good work at AMD! This is all I'm asking/wondering and thanks for answering so far. I won't bother/annoy with more questions :).