lkml.org 
[lkml]   [2008]   [Mar]   [13]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
From
SubjectRe: [PATCH] x86: Change x86 to use generic find_next_bit
Date

On Thu, 13 Mar 2008 18:14:24 +0530, "Aneesh Kumar K.V"
<aneesh.kumar@linux.vnet.ibm.com> said:
> On Sun, Mar 09, 2008 at 09:01:04PM +0100, Alexander van Heukelum wrote:
> > x86: Change x86 to use the generic find_next_bit implementation
> >
> > The versions with inline assembly are in fact slower on the machines I
> > tested them on (in userspace) (Athlon XP 2800+, p4-like Xeon 2.8GHz, AMD
> > Opteron 270). The i386-version needed a fix similar to 06024f21 to avoid
> > crashing the benchmark.
> >
> > Benchmark using: gcc -fomit-frame-pointer -Os. For each bitmap size
> > 1...512, for each possible bitmap with one bit set, for each possible
> > offset: find the position of the first bit starting at offset. If you
> > follow ;). Times include setup of the bitmap and checking of the
> > results.
> >
> > Athlon Xeon Opteron 32/64bit
> > x86-specific: 0m3.692s 0m2.820s 0m3.196s / 0m2.480s
> > generic: 0m2.622s 0m1.662s 0m2.100s / 0m1.572s
> >
> > If the bitmap size is not a multiple of BITS_PER_LONG, and no set
> > (cleared) bit is found, find_next_bit (find_next_zero_bit) returns a
> > value outside of the range [0,size]. The generic version always returns
> > exactly size. The generic version also uses unsigned long everywhere,
> > while the x86 versions use a mishmash of int, unsigned (int), long and
> > unsigned long.
>
> This problem is observed on x86_64 and powerpc also.

I'm not entirely sure if it is a problem. In some cases it
certainly is an inconvenience, though ;). I mentioned the
difference between the old and generic versions, because
of the possibility of dependence of this behaviour.

Indeed I see for example (in fs/ext4/mballoc.c).

bit = mb_find_next_zero_bit(bitmap_bh->b_data, end, bit);
if (bit >= end)
break;
next = mb_find_next_bit(bitmap_bh->b_data, end, bit);
if (next > end)
next = end;
free += next - bit;

So here it needed to adjust the value.

> We need a long
> aligned address for test_bit, set_bit find_bit etc. In ext4 we have
> to make sure we align the address passed to
>
> ext4_test_bit
> ext4_set_bit
> ext4_find_next_zero_bit
> ext4_find_next_bit
>
> fs/ext4/mballoc.c have some examples.

This is a different 'problem'. find_next_bit works on arrays of long,
while the bitmaps in ext4_find_next_bit are of type void * and seem
not to have any alignment restrictions. ext4 implements wrappers
around find_next_bit to solve that 'problem'.

The question that arises is: do we want find_first_bit, find_next_bit.
etc. to always return a value in the range [0, size], or do we want
to allow implementations that return [0, size-1] if there is a bit
found and something else (roundup(size,bitsperlong) or ulongmax, for
example if, none were found?

The current x86_64 versions of find_first_bit and find_next_bit return
roundup(size,bitsperlong) if no bits were found, but on the other hand
I guess most bitmaps are a multiple of bitsperlong bits in size, which
hides the difference.

Greetings,
Alexander

> -aneesh
--
Alexander van Heukelum
heukelum@fastmail.fm

--
http://www.fastmail.fm - Or how I learned to stop worrying and
love email again



\
 
 \ /
  Last update: 2008-03-13 15:51    [W:0.212 / U:0.600 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site