Hi,
As it stands, potential_constant_expression_1 does not handle
REDUC_PLUS_EXPR trees.
For some cross-lane add-reduce neon intrinsics we would like to
use the TARGET_FOLD_BUILTIN hook to fold these calls to a
REDUC_PLUS_EXPR. As an example, see Tejas Belagod's patch at:
http://gcc.gnu.org/ml/gcc-patches/2013-03/msg00502.html
If we do that, the C++11 front-end is not able to evaluate
them as a potential constant expression and we get an
"unexpected AST of kind REDUC_PLUS_EXPR" error.
I am about as far from a C++11 expert as you can get, so I
don't know whether this is the correct fix, but to my naieve
mind I don't see why REDUC_PLUS_EXPR is different from any
other unary operator...
I have tested the patch on aarch64-none-elf with no regressions,
and it fixed the problem I was seeing previously.
Is this OK to commit to 4.9 when stage 1 opens up?
Thanks,
James Greenhalgh
---
gcc/cp/
* semantics.c
(potential_constant_expression_1): Add case for REDUC_PLUS_EXPR.diff --git a/gcc/cp/semantics.c b/gcc/cp/semantics.c
index 9446f83..232057a 100644
--- a/gcc/cp/semantics.c
+++ b/gcc/cp/semantics.c
@@ -8600,6 +8600,7 @@ potential_constant_expression_1 (tree t, bool want_rval, tsubst_flags_t flags)
case ABS_EXPR:
case TRUTH_NOT_EXPR:
case FIXED_CONVERT_EXPR:
+ case REDUC_PLUS_EXPR:
case UNARY_PLUS_EXPR:
return potential_constant_expression_1 (TREE_OPERAND (t, 0), rval,
flags);
