Linearize COND_EXPRs involving a single SSA_NAME
Richard Henderson
rth@redhat.com
Tue Jul 13 07:48:00 GMT 2004
On Mon, Jul 12, 2004 at 06:27:41PM -0400, Diego Novillo wrote:
> Again. fold_stmt() has the same problem. We don't call it when we copy
> propagate. Perhaps we should, but this is somewhat specialized to
> control flow linearization.
Well, *definitely* do not replicate the logic to do the actual
folding. At minimum do
if (TREE_CODE_CLASS (TREE_CODE (cond)) == 'r'
&& TREE_OPERAND (cond, 0) == TREE_OPERAND (cond, 1))
cond = fold (cond);
at the place you're tweaking things now.
r~
More information about the Gcc-patches
mailing list