[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <46C38156.8060502@oracle.com>
Date: Wed, 15 Aug 2007 18:42:30 -0400
From: Chuck Lever <chuck.lever@...cle.com>
To: Andi Kleen <andi@...stfloor.org>
CC: linux-kernel@...r.kernel.org
Subject: Re: [PATCH] Eliminate result signage problem in asm-x86_64/bitops.h
I apologize for sending a separate cover letter for a single patch.
Andi Kleen wrote:
> On Wed, Aug 15, 2007 at 05:02:47PM -0400, Chuck Lever wrote:
>> The return type of __scanbit() doesn't match the return type of
>> find_{first,next}_bit(). Thus when you construct something like
>> this:
>>
>> boolean ? __scanbit() : find_first_bit()
>
> Why would you want to write this? What is boolean?
> Do they have different arguments?
So here's the definition of the x86_64 find_first_bit() macro, straight
from include/x86_64/bitops.h:
#define find_first_bit(addr,size) \
((__builtin_constant_p(size) && (size) <= BITS_PER_LONG ? \
(__scanbit(*(unsigned long *)addr,(size))) : \
find_first_bit(addr,size)))
In this case "boolean" is:
__builtin_constant_p(size) && (size) <= BITS_PER_LONG
the first arm of the conditional is:
__scanbit(*(unsigned long *)addr,(size))
the second arm of the conditional is:
find_first_bit(addr,size)
(this is the "function" version of find_first_bit, not the macro that's
being defined. The naming here is unfortunately confusing).
Thus, roughly speaking, when the type of "size" is smaller than a long,
the macro's return type evaluates to unsigned long. If "size" is larger
than a long, the macro's return type evaluates to signed long.
By making the return type of __scanbit() an unsigned long, both arms of
the conditional evaluate to the same result type.
> It's on my todo list for some time to special case
> f_f_b() and friends for smaller arguments. Would
> that eliminate this construct?
Well, I can only assume what you mean by this, but I think that would
address the problem.
My real interest here is to eliminate a whole lot of compiler noise when
I enable -Wsign-compare for certain parts of the kernel.
View attachment "chuck.lever.vcf" of type "text/x-vcard" (291 bytes)
Powered by blists - more mailing lists