Re: How early are COND_EXEC patterns created?

[Michael Meissner <meissner at cygnus dot com>]

On Mon, May 01, 2000 at 10:24:07PM -0700, Richard Henderson wrote:
> On Tue, May 02, 2000 at 12:10:06AM -0400, John Wehle wrote:
> > What's the earliest that COND_EXEC patterns are created?
> 
> At present, after reload.  We'd need to teach reload to
> create conditional spills before we could move it earlier.
> I suspect it'll never go further back than just before sched1,
> as too few optimization passes would be able to do anything
> useful with it.

Unfortunately the machine I'm working on needs to create new pseudos for
optimal use of COND_EXEC (it doesn't do conditional execution of instructions
involving constants, and I had hoped to load the constant up into a scratch
register for the basic block.  I'll have to think on this.

> > BTW, can COND_EXEC appear inside a PARALLEL?
> 
> I would expect not, but did make provision for it in the bits
> that scan for stores.

As Richard knows, the reverse certain occurs (having a PARALLEL inside of a
COND_EXEC) in the d30v.  Look at cond_exec_si_addsub.

-- 
Michael Meissner, Cygnus Solutions, a Red Hat company.
PMB 198, 174 Littleton Road #3, Westford, Massachusetts 01886, USA
Work:	  meissner@redhat.com		phone: +1 978-486-9304
Non-work: meissner@spectacle-pond.org	fax:   +1 978-692-4482