A quick word about the pthreadrope7 fails.

[Dhruv Matani]

On Fri, 2004-10-15 at 00:02, Paolo Carlini wrote:
> Dhruv Matani wrote:
> 
> >Except for the first case(first allocation), all of them will be even!
> >  
> >
> Good. Therefore let's just start with 2 longs as bitmap, everywhere, and 
> we are
> done once and for all!
> 
> >I have a beta patch in which I have used long. I figured that this
> >conversion from unsigned int to long can be only better for the
> >allocator, because now(64-bit) the CPU can manipulate 64-bits directly!
> >Hence, 64-bits in a bitmap can be checked for emptiness/fullness.
> >
> >Would you be interested in testing that out on a 64-bit machine?
> >  
> >
> Yes, of course, but see above. Also, please document that 8 in struct 
> _Alloc_block:
> it must be very clear that raising to 16 guarantees alignment 16.
> 
> Basically, the patch, besides the 2 above, boils down to a search and 
> replace
> s/int/long/, right? No need of additional weird computations and casts!

I have attached a patch, which works with current(unpatched mainline).
The Makefiles are not changed in this patch.

> 
> Paolo.
-- 
        -Dhruv Matani.
http://www.geocities.com/dhruvbird/

The price of freedom is responsibility, but it's a bargain, because
freedom is priceless. ~ Hugh Downs
-------------- next part --------------
A non-text attachment was scrubbed...
Name: patch4_bitalloc_alignment_14102004_dhruv.diff.tar.bz2
Type: application/x-bzip
Size: 13261 bytes
Desc: not available
URL: <http://gcc.gnu.org/pipermail/libstdc++/attachments/20041014/82ecc01e/attachment.bin>