This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
RE: splits after reload
- From: <tm_gccmail at kloo dot net>
- To: Jim Wilson <wilson at specifixinc dot com>
- Cc: "Rakesh Kumar - Software, Noida" <rakeshku at noida dot hcltech dot com>,gcc at gcc dot gnu dot org
- Date: Wed, 17 Dec 2003 11:37:04 -0800 (PST)
- Subject: RE: splits after reload
On 16 Dec 2003, Jim Wilson wrote:
> On Tue, 2003-12-16 at 02:15, Rakesh Kumar - Software, Noida wrote:
> > I agree, but I was thinking about postreload splits. Is there any
> > probability of throwing away the splitted instructions after reload?
>
> I don't know. Look at the souces.
>
> > I thought about peephole2. But it seems making the optimization too
> > local.
>
> Some good points. It isn't clear to me whether this occurs often enough
> to be worth the trouble though. You will probably have to try it and
> see. There are few targets nowadays that can't load a double with a
> single instruction, so this addresss arithmetic from the splitter is
> going to be rare outside the SH, and this is probably only the early SH
> architectures, maybe 1-3?
Nope. The SH4 is the only released SH CPU which has an FPU which handles
doubles (excluding the SH5). Future members of the family may require this
as well due to the 32-bit alignment of data and stack.
The SH1-SH3 use software fp, and the SH3E forces doubles to floats.
Toshi