[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20070502002643.GJ4095@ftp.linux.org.uk>
Date: Wed, 2 May 2007 01:26:43 +0100
From: Al Viro <viro@....linux.org.uk>
To: Josh Triplett <josh@...edesktop.org>
Cc: Linus Torvalds <torvalds@...ux-foundation.org>,
linux-sparse@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: sparse -Wptr-subtraction-blows: still needed?
On Tue, May 01, 2007 at 04:59:57PM -0700, Josh Triplett wrote:
> Linus Torvalds wrote:
> >
> > On Tue, 1 May 2007, Josh Triplett wrote:
> >> Does this still apply? Do current versions of GCC still have this problem?
> >> If not, can the option and warning go away?
> >
> > Even if current versions of gcc don't triple the build time (and for the
> > kernel, I suspect it doesn't, because we've tried to clean up our header
> > files), the generated _code_ will invariably suck.
>
> "invariably"?
>
> Do you know whether the current version of GCC generates poor code for pointer
> subtraction?
>
> If so, does anything in particular make this an unfixable problem?
Just the fact that calculation itself is nasty. In the best case you
get (on x86) something like
sarl $2, %eax
imull $-1431655765, %eax, %eax
(that one is with object size equal to 12). On other targets it can get
considerably uglier - e.g. on alpha with -O2 the same will result in
sra $17,2,$17
s4subq $17,$17,$0
s8subq $0,$0,$0
s4addq $0,$17,$0
sll $0,8,$1
addq $0,$1,$0
sll $0,16,$2
addq $0,$2,$0
sll $0,32,$1
addq $0,$1,$0
addq $0,$0,$0
addl $0,$17,$0
With -Os it's
ldah $1,$LC0($29) !gprelhigh
sra $0,2,$0
ldq $1,$LC0($1) !gprellow
mull $0,$1,$0
with LC0 in rodata:
$LC0:
.quad -6148914691236517205
Now imagine the joy of having a bunch of such wonders in a hot path...
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists