This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [PATCH][libgcc-math] Vectorized intrinsics for x86_64
- From: Richard Henderson <rth at redhat dot com>
- To: Richard Guenther <rguenther at suse dot de>
- Cc: gcc-patches at gcc dot gnu dot org
- Date: Fri, 31 Mar 2006 07:57:57 -0800
- Subject: Re: [PATCH][libgcc-math] Vectorized intrinsics for x86_64
- References: <Pine.LNX.4.64.0603281022000.3982@t148.fhfr.qr>
On Tue, Mar 28, 2006 at 10:23:47AM +0200, Richard Guenther wrote:
> The intrinsics implementation was contributed by AMD to be licensed
> as GPL + libgcc execption.
It's a shame they were written in hand-coded assembly; otherwise
we could use them for 32-bit as well. I don't have any trouble
adding these routines, but I think the data sections need some work.
First, none of the data put in .data is writable. At bare minimum
this should be going into .rodata. But I also see that there is
quite a bit of overlap between the various routines, so it would be
Much Better if we could make use of the constant pooling featues of
the linker. So the tables should go into .rodata, but the individual
double-precision values should go into
.section .rodata.cst8, "M", @progbits, 8
.align 8
:REVIEWMAIL:
r~