Hi.
I've been working on transition of cond expressions to match.pd.
With my changes I noticed there's one wrong pattern that leads to:
Transforming _6 > _7 & _6 < _7 into 0
...
/home/marxin/Programming/gcc/gcc/testsuite/c-c++-common/vector-compare-3.c:20:1:
error: the first argument of a ‘vec_cond_expr’ must be of a boolean vector
type of the same number of elements as the result
20 | g (v4i *x, v4i const *y, v4i *z, v4i *t)
| ^
vector(4) int
_Bool
_9 = VEC_COND_EXPR <0, { -1, -1, -1, -1 }, { 0, 0, 0, 0 }>;
which is wrong.
Proper simlification is:
Transforming _6 > _7 & _6 < _7 into { 0, 0, 0, 0 }
Patch can bootstrap on x86_64-linux-gnu and survives regression tests.
Ready to be installed?
Thanks,
Martin
gcc/ChangeLog:
2019-09-06 Martin Liska <mli...@suse.cz>
* match.pd: For vector types, take type from a vector argument
of the expression.
---
gcc/match.pd | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/gcc/match.pd b/gcc/match.pd
index 1d13543a615..66e3c25ee87 100644
--- a/gcc/match.pd
+++ b/gcc/match.pd
@@ -3769,7 +3769,7 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
(bit_and:c (test1 @0 @1) (test2 @0 @1))
(if (INTEGRAL_TYPE_P (TREE_TYPE (@0))
|| VECTOR_INTEGER_TYPE_P (TREE_TYPE (@0)))
- { constant_boolean_node (false, type); })))
+ { constant_boolean_node (false, INTEGRAL_TYPE_P (TREE_TYPE (@0)) ? type : TREE_TYPE (@0)); })))
/* A & (2**N - 1) <= 2**K - 1 -> A & (2**N - 2**K) == 0
A & (2**N - 1) > 2**K - 1 -> A & (2**N - 2**K) != 0
