On Tue, Mar 10, 2020 at 02:39:23PM -0700, Stephane Eranian wrote: > On Tue, Mar 10, 2020 at 12:59 PM Andi Kleen <ak@xxxxxxxxxxxxxxx> wrote: > > > > On Tue, Mar 10, 2020 at 11:50:03AM -0700, Ian Rogers wrote: > > > This patch links perf with the libpfm4 library. > > > This library contains all the hardware event tables for all > > > processors supported by perf_events. This is a helper library > > > that help convert from a symbolic event name to the event > > > encoding required by the underlying kernel interface. This > > > library is open-source and available from: http://perfmon2.sf.net. > > > > For most CPUs the builtin perf JSON event support should make > > this redundant. > > > We decided to post this patch to propose an alternative to the JSON > file approach. It could be an option during the build. > The libpfm4 library has been around for 15 years now. Therefore, it > supports a lot of processors core and uncore and it is very portable. > The key value add I see is that this is a library that can be, and has > been, used by tool developers directly in their apps. It can > work with more than Linux perf_events interface. It is not tied to the > interface. It has well defined and documented entry points. > We do use libpfm4 extensively at Google in both the perf tool and > applications. The PAPI toolkit also relies on this library. > > I don't see this as competing with the JSON approach. It is just an > option I'd like to offer to users especially those familiar > with it in their apps. I dont mind having it, in fact I found really old email where I'm asking Peter about that ;-) and he wasn't very keen about that: https://lore.kernel.org/lkml/1312806326.10488.30.camel@twins/ not sure what was the actual reason at that time and if anything changed since.. Peter? btw I can't apply even that v2 on latest Arnaldo's branch jirka
