This is the mail archive of the libc-ports@sources.redhat.com mailing list for the libc-ports project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Fwd: [PATCH 1/2] AArch64 glibc port


On Thu, 8 Nov 2012, Marcus Shawcroft wrote:

> Switched s_fma version.  With the patch to FP_PACK_SEMIRAW discussed
> earlier we are down to 6 errors in test-ldouble and 3 in test-ildoubl
> remain, see attachment.

Of these, the csqrt ones seem AArch64-specific, so you should continue to 
investigate them (I don't know if they are bug in the port, or a bug in 
generic code that for some reason only shows up with this port).  The 
others (spurious underflows from clog10, clog, yn) appear to be generic to 
ldbl-128 architectures (they appear in the list for S/390 that Andreas 
Krebbel posted at 
<http://sourceware.org/ml/libc-alpha/2012-11/txt00003.txt>, for example) - 
so while I hope someone will investigate them for 2.17, they don't 
indicate anything wrong with your port in particular.

> Revised patch and ChangeLog attached.

This is OK with the following changes:

* Add lll_futex_timed_wait_bitset to lowlevellock.h, as recently added in 
libc.

* Remove the definitions of __NR_pread and __NR_pwrite in sysdep.h 
(sysdeps/unix/sysv/linux/generic/sysdep.h already has them, so you 
shouldn't need to duplicate them; just make sure that file gets used at 
some level among sysdep.h files including each other).

You should also list yourself as AArch64 maintainer at 
<http://sourceware.org/glibc/wiki/MAINTAINERS>.

I believe you still need to continue working on getting the following into 
libc for this port to be fully functional:

* The soft-fp change for correct underflow handling (GCC's soft-fp will 
then need updating), revised as I suggested then will need posting to 
libc-alpha.

* The -fexceptions makefile patch, where you need to answer HJ's question 
and follow up as needed on getting this patch in.

(As well as investigating those csqrt ulps further.  And once the main 
port is in and the above issues are resolved, you may of course wish to 
work on getting in the optimized functions that were in the original 
patch.)

-- 
Joseph S. Myers
joseph@codesourcery.com


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]