This is the mail archive of the
libc-alpha@sources.redhat.com
mailing list for the glibc project.
Re: [PATCH] PPC64 Use hardware sqrt.
- From: Paul Mackerras <paulus at samba dot org>
- To: Steve Munroe <sjmunroe at us dot ibm dot com>
- Cc: Kumar Gala <kumar dot gala at freescale dot com>,libc-alpha at sources dot redhat dot com, Tom Gall <tom_gall at vnet dot ibm dot com>
- Date: Wed, 12 May 2004 13:44:37 +1000
- Subject: Re: [PATCH] PPC64 Use hardware sqrt.
- References: <DD8AD76C-9FE4-11D8-B596-000393DBC2E8@freescale.com><OFD2E8C8ED.DAEDA408-ON86256E92.0007E0B6-86256E92.0009005D@us.ibm.com>
Steve Munroe writes:
> Paul do you agree to adding PPC_FEATURE_HAS_FPU_OPT to AT_HWCAP?
Yes, that would be fine.
I assume that means that new glibc running on an old kernel will not
use the optional FP instructions, i.e. the bit will bit 1 to say that
the optional instructions are implemented.
> For current PPC64 processors it will always be true. If/when a PPC64
> processor is shipped that does implement the full set of the Optional FP
> instructions we will need to add logic for the kernel to detect this. I
> suspect the landscape is more complex for PPC32.
I'm not aware of any 32-bit PowerPC processor that implements all of
the optional FP instructions. The 74xx processors implement fsel,
fres and fsqrte. I think we will just leave the bit at 0 in 32-bit
kernels.
I assume you're going to make a 32-bit glibc use the optional
instructions if available, so that we see the speedup for 32-bit
applications running on a 64-bit processor?
Regards,
Paul.