This is the mail archive of the
gcc@gcc.gnu.org
mailing list for the GCC project.
Re: gcc/resource.c mark_target_live_regs Question
- To: gcc at gcc dot gnu dot org
- Subject: Re: gcc/resource.c mark_target_live_regs Question
- From: Brad Lucier <lucier at math dot purdue dot edu>
- Date: Fri, 16 Feb 2001 00:29:15 -0500 (EST)
- Cc: lucier at math dot purdue dot edu (Brad Lucier)
Re:
> On Thu, Feb 15, 2001 at 09:43:46AM -0800, Mark Mitchell wrote:
> > We're making some assumption that reorg confuses everything, but not
> > across BARRIERs? Is that it?
>
> Exactly.
>
> > mark_target_live_regs mark all registers live if find_basic_block goes
> > back, say, more than 1000 instructions ...
>
> 50 or 100 is probably a better default. Remember, that
> if you have N insns in the block you're still going to
> scan N**2/2 insns.
>
>
> r~
I'm not exactly understanding all the implications of this discussion,
but one easily gets > 200 instructions in the inner basic block of
a 8-way butterfly FFT code used for bignum multiplication, and I wouldn't
want all registers marked as live in that block.
Brad