Back last year I made copyprop also propagate constants (if a copy
chain ultimatively ends in a constant). At least I thought so, but
it turns out I missed a piece.
Bootstrapped and tested on x86_64-unknown-linux-gnu, applied to trunk.
Richard.
2011-03-25 Richard Guenther <rguent...@suse.de>
* tree-ssa-copy.c (stmt_may_generate_copy): Copies from constants
also generate copies.
(fini_copy_prop): Handle constant values properly.
Index: gcc/tree-ssa-copy.c
===================================================================
*** gcc/tree-ssa-copy.c (revision 171450)
--- gcc/tree-ssa-copy.c (working copy)
*************** stmt_may_generate_copy (gimple stmt)
*** 315,322 ****
/* Otherwise, the only statements that generate useful copies are
assignments whose RHS is just an SSA name that doesn't flow
through abnormal edges. */
! return (gimple_assign_rhs_code (stmt) == SSA_NAME
! && !SSA_NAME_OCCURS_IN_ABNORMAL_PHI (gimple_assign_rhs1 (stmt)));
}
--- 315,323 ----
/* Otherwise, the only statements that generate useful copies are
assignments whose RHS is just an SSA name that doesn't flow
through abnormal edges. */
! return ((gimple_assign_rhs_code (stmt) == SSA_NAME
! && !SSA_NAME_OCCURS_IN_ABNORMAL_PHI (gimple_assign_rhs1 (stmt)))
! || is_gimple_min_invariant (gimple_assign_rhs1 (stmt)));
}
*************** fini_copy_prop (void)
*** 771,776 ****
--- 772,778 ----
of the representative to the first solution we find if
it doesn't have one already. */
if (copy_of[i].value != var
+ && TREE_CODE (copy_of[i].value) == SSA_NAME
&& POINTER_TYPE_P (TREE_TYPE (var))
&& SSA_NAME_PTR_INFO (var)
&& !SSA_NAME_PTR_INFO (copy_of[i].value))
