[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <1563993952.11067.15.camel@lca.pw>
Date: Wed, 24 Jul 2019 14:45:52 -0400
From: Qian Cai <cai@....pw>
To: David Howells <dhowells@...hat.com>
Cc: akpm@...ux-foundation.org, davem@...emloft.net, arnd@...db.de,
jakub@...hat.com, ndesaulniers@...gle.com, morbo@...gle.com,
jyknight@...gle.com, natechancellor@...il.com,
linux-arch@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH] asm-generic: fix -Wtype-limits compiler warnings
On Wed, 2019-07-24 at 08:49 +0100, David Howells wrote:
> Qian Cai <cai@....pw> wrote:
>
> > Fix it by moving almost all of this multi-line macro into a proper
> > function __get_order(), and leave get_order() as a single-line macro in
> > order to avoid compilation errors.
>
> The idea was that you could compile-time initialise a global variable with
> get_order():
>
> int a = get_order(SOME_MACRO);
>
> This is the same reason that ilog2() is a macro:
>
> int a = ilog2(SOME_MACRO);
>
> See the banner comment on get_order():
>
> * This function may be used to initialise variables with compile time
> * evaluations of constants.
>
> If you're moving the constant branch into __get_order(), an inline function,
> then we'll no longer be able to do this and you need to modify the comment
> too. In fact, would there still be a point in having the get_order() macro?
>
> Also, IIRC, older versions of gcc see __builtin_constant_p(n) == 0 inside an
> function, inline or otherwise, even if the passed-in argument *is* constant.
I have GCC 8.2.1 which works fine.
# cat const.c
#include <stdio.h>
static int i = 0;
static inline void check()
{
if (__builtin_constant_p(i))
printf("i is a const.\n");
}
void main()
{
check();
}
# gcc -O2 const.c -o const
# ./const
i is a const.
Powered by blists - more mailing lists