> On 17 Sep 2019, at 17:12, Christian Hopps <cho...@chopps.org> wrote: > > > >> On Sep 17, 2019, at 3:56 AM, Benoit Ganne (bganne) via Lists.Fd.Io >> <bganne=cisco....@lists.fd.io> wrote: >> >>>> AFAIK the default for x86_64 is to omit frame pointer >>>> for optimized builds so you probably do not have it in most cases in VPP >>>> builds (but I did not check). >>>> So when perf tries to interpret random data as frame pointer, this is >>>> what you get :) >>>> The most reliable (but slowest) should be dwarf. lbr should be the way >>>> to go on Intel Haswell and later. >>> FWIW, this is being run on the Macchiatobin platform (i.e., aarch64 >>> cortex-A72). :) >> >> My bad. Not sure what is going-on but it looks like gcc also omit fp by >> default on aarch64: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84521 >> Did you had a chance to try dwarf instead of fp? > > It doesn't appear to be an option in my perf binary, unfortunately.
Wasn't much simpler to run "perf top" on the specific core and see where cycles are spent?
-=-=-=-=-=-=-=-=-=-=-=- Links: You receive all messages sent to this group. View/Reply Online (#14006): https://lists.fd.io/g/vpp-dev/message/14006 Mute This Topic: https://lists.fd.io/mt/34128898/21656 Group Owner: vpp-dev+ow...@lists.fd.io Unsubscribe: https://lists.fd.io/g/vpp-dev/unsub [arch...@mail-archive.com] -=-=-=-=-=-=-=-=-=-=-=-