[RFA] The Integrated Register Allocator

[Kaz Kojima]

Vladimir Makarov <vmakarov@redhat.com> wrote:
> The first test (20030914-1.c) is not big.  So I'll try to find what is 
> wrong with it.

Thanks!

I've tried the current ira branch (revision 133928) for
sh-unknown-elf target with the unified tree and found that
gcc.c-torture/execute/20030914-1.c only fails with -m4 -ml
target options which are default on sh4-unknown-linux-gnu.
I've also noticed that

  double dadd (double x, double y) { return x + y;}

is compiled to a very lengthy code:

dadd:
	fadd	dr6,dr4
	mov.l	r14,@-r15
	mov	r15,r14
	fmov.s	fr4,@-r15
	fmov.s	fr5,@-r15
	mov.l	@r15+,r1
	mov.l	@r15+,r2
	mov.l	r2,@-r15
	mov.l	r1,@-r15
	fmov.s	@r15+,fr1
	fmov.s	@r15+,fr0
	mov	r14,r15
	rts	
	mov.l	@r15+,r14

with -O2 -fira -m4 on sh-unknown-elf.

Regards,
	kaz