On Tue, Jan 22, 2013 at 10:38:24AM +, Ramana Radhakrishnan wrote:
> Instead I'd just use s_register_operand for operand1 and continue to
> use cmpdi_operand for operand2 and fix it so.
>
> And likewise .
>
> Ok with those changes and if no regressions.
Managed to bootstrap/regtest this (via
On 01/21/13 10:55, Jakub Jelinek wrote:
Hi!
As can be seen on the testcase, this backend bug is still reproduceable even
on trunk, the backend just can't rely on cstoredi4 or cbranchdi4 expansion
not being performed with two constants, unless it has predicates that
disallow it (Steven's patch in
Hi!
As can be seen on the testcase, this backend bug is still reproduceable even
on trunk, the backend just can't rely on cstoredi4 or cbranchdi4 expansion
not being performed with two constants, unless it has predicates that
disallow it (Steven's patch in the PR).
This patch just forces it into r
