From: Pan Li <[email protected]>
During the match pattern of SAT_U_MUL form 7, we found there is
a pattern like below:
(nop_convert)(a bit_op (convert b))
which result in the pattern match of SAT_U_MUL complicated and
unintuitive. According to the suggestion of Richard, we would
like to simply it to blew:
(convert a) bit_op (convert b)
which is more friendly for reading and bit_op. There are three
bit_op here, aka bit_ior, bit_and and bit_xor.
gcc/ChangeLog:
* match.pd: Add simplfy to fold outer convert of bit_op
to inner captures.
Signed-off-by: Pan Li <[email protected]>
---
gcc/match.pd | 14 +++++++++++++-
1 file changed, 13 insertions(+), 1 deletion(-)
diff --git a/gcc/match.pd b/gcc/match.pd
index 63d56b08192..22b1bd054b0 100644
--- a/gcc/match.pd
+++ b/gcc/match.pd
@@ -2310,7 +2310,19 @@ DEFINE_INT_AND_FLOAT_ROUND_FN (RINT)
&& types_match (type, @0)
&& !POINTER_TYPE_P (TREE_TYPE (@0))
&& TREE_CODE (TREE_TYPE (@0)) != OFFSET_TYPE)
- (bitop @0 (convert @1)))))
+ (bitop @0 (convert @1))))
+ /* Similar as above, but @0 has a widen type. */
+ (simplify
+ (convert (bitop:cs@2 (convert:s @0) @1))
+ (if (GIMPLE
+ && INTEGRAL_TYPE_P (type)
+ && INTEGRAL_TYPE_P (TREE_TYPE (@0))
+ && TREE_CODE (@1) != INTEGER_CST
+ && tree_nop_conversion_p (type, TREE_TYPE (@2))
+ && !POINTER_TYPE_P (TREE_TYPE (@0))
+ && TREE_CODE (TREE_TYPE (@0)) != OFFSET_TYPE
+ && TYPE_PRECISION (TREE_TYPE (@0)) > TYPE_PRECISION (type))
+ (bitop:type (convert @0) (convert @1)))))
(for bitop (bit_and bit_ior)
rbitop (bit_ior bit_and)
--
2.43.0
