Re: [PATCH v2] x86: use POPCNT for hweight<N>() when available

[Roger Pau Monné <roger.pau@xxxxxxxxxx>]

On Fri, May 22, 2020 at 11:58:40AM +0200, Jan Beulich wrote:
> On 20.05.2020 19:18, Roger Pau Monné wrote:
> > I also assume that using no_caller_saved_registers when available or
> > else keeping the current behavior is not an acceptable solution? FWIW,
> > from a FreeBSD PoV I would be OK with that, as I don't think there are
> > any supported targets with clang < 5.
> 
> By "current behavior" do you mean what the patch currently
> does (matching my earlier response that I may try going this
> route) or the unpatched upstream tree?

Sorry this wasn't clear. By current I meant what's in the upstream
tree. IMO having both no_caller_saved_registers, the -ffixed-reg stuff
and the current upstream no popcnt implementation is kind of too much
divergence?

My preference would be to add popcnt support only when
no_caller_saved_registers is supported by the compiler, as I think
that's the cleanest solution. This is purely a performance
optimization, so it doesn't seem that bad to me to request a newish
compiler in order to enable it.

Roger.