GCC trunk SPEC2000 performance

[Michael Matz]

Hi,

On Fri, 21 Jun 2002, Graham Stott wrote:

> There's also on place in reload () that uses PUT_MODE (reg, MEM)
> on a reg which is an entry in the regno_reg_rtx array.

Yes, that's expected.  This way regs, which hadn't got a hardreg can
easily be put into a stack-ref on all the insns with only one set, and be
"converted" back to registers again in the next pass, if it's worthy.
Reload itself takes care of massaging the invalid insns, unsharing the
rtl, if necessary.  And because after reload there are no pseudos anymore,
it doesn't matter what is in the regno_reg_rtx[] array.  This has been the
case forever (AFAIK).


Ciao,
Michael.