Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Sat, 21 Nov 2020 21:46:26 +0100
From:   Luc Van Oostenryck <luc.vanoostenryck@gmail.com>
To:     Jakub Jelinek <jakub@redhat.com>
Cc:     Peter Zijlstra <peterz@infradead.org>,
        Andrew Morton <akpm@linux-foundation.org>,
        Linus Torvalds <torvalds@linux-foundation.org>,
        linux-kernel@vger.kernel.org, christophe.leroy@csgroup.eu,
        rdunlap@infradead.org
Subject: Re: [PATCH] ilog2: Improve ilog2 for constant arguments
Message-ID: <20201121204626.tw3jribsz326e3ag@ltop.local>
References: <20201120125154.GB3040@hirez.programming.kicks-ass.net>
 <20201121202310.lx2iqqebal3fmmmo@ltop.local>
 <20201121202954.GW3788@tucnak>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20201121202954.GW3788@tucnak>
Precedence: bulk

On Sat, Nov 21, 2020 at 09:29:54PM +0100, Jakub Jelinek wrote:
> On Sat, Nov 21, 2020 at 09:23:10PM +0100, Luc Van Oostenryck wrote:
> > On Fri, Nov 20, 2020 at 01:51:54PM +0100, Peter Zijlstra wrote:
> > > 
> > > Other option would be to change the const_ilog2 macro, though as the
> > > description says it is meant to be used also in C constant expressions,
> > > and while GCC will fold it to constant with constant argument even in
> > > those, perhaps it is better to avoid using extensions in that case.
> > 
> > Just for info, the description is outdated and Sparse is just fine with
> > __builtin_clzll() and friends in constant expressions (since Feb 2017)
> 
> Why is the description outdated?  It is still an extension that not every
> compiler might fold in constant expressions.  And, the large expressions
> aren't really a problem in constant expressions, they will be folded there
> to constant or error.
> The problem the patch was trying to solve is that the large expressions are
> a problem at least for GCC in runtime code when guarded by
> __builtin_constant_p, because __builtin_constant_p is folded quite late
> (intentionally so, so that more constants can be propagated into it, e.g.
> after inlining etc.), and the large expressions might confuse inliner
> heuristics.

I was only talking about the part "Use this where *sparse* expects a true
constant expression, e.g. for array 75 indices." and wanted to say that 
__builtin_clzll() with a constant argument is now also expanded to a
constant, like GCC does (it wasn't the case before 2017 and I think
it was the main reason why const_ilog2() is written as it is).

-- Luc