[PATCH RESEND] perf/util: make TYPE_STATE_MAX_REGS architecture-dependent

[Suchit Karunakaran]

On Mon, 25 Aug 2025 at 22:54, Ian Rogers <irogers at google.com> wrote:
>
> On Mon, Aug 25, 2025 at 7:17 AM Suchit Karunakaran
> <suchitkarunakaran at gmail.com> wrote:
> >
> > On Sat, 16 Aug 2025 at 12:16, Suchit Karunakaran
> > <suchitkarunakaran at gmail.com> wrote:
> > >
> > > Hi Namhyung,
> > > Thanks for reviewing the patch. I'd like to ask if there's anything
> > > else I should do regarding this patch, given that it's supported only
> > > for x86 and powerpc?
> > >
> > > Thanks,
> > > Suchit
> >
> >
> > Hi Namhyung,
> > Following up on this patch. I would appreciate any guidance on further
> > actions needed, considering its current support is only for x86 and
> > PowerPC.
>
> Can we just make  TYPE_STATE_MAX_REGS 32? There's no reason to assume
> the architecture the perf binary is built on matches the perf.data
> file being processed by perf annotate. Given 32 is bigger than 16 then
> this can just be a maximum value (a comment to this effect in the code
> would be grand). In general, the use of the arch directory and things
> like "#ifdef __powerpc__" are code smells that are going to break with
> cross-architecture profiling.
>
> Thanks,
> Ian

Yeah I agree that setting TYPE_STATE_MAX_REGS to 32 is a more
reasonable approach. I will send a v2 patch incorporating this change.
Thanks,
Suchit