On Tue, 18 Feb 2020, Li Jia He wrote:
Also the pattern doing the standalone transform does/* Optimize TRUNC_MOD_EXPR by a power of two into a BIT_AND_EXPR, i.e. "X % C" into "X & (C - 1)", if X and C are positive. Also optimize A % (C << N) where C is a power of 2, to A & ((C << N) - 1). */ (match (power_of_two_cand @1) INTEGER_CST@1) (match (power_of_two_cand @1) (lshift INTEGER_CST@1 @2)) (for mod (trunc_mod floor_mod) (simplify (mod @0 (convert?@3 (power_of_two_cand@1 @2))) (if ((TYPE_UNSIGNED (type) || tree_expr_nonnegative_p (@0)) && tree_nop_conversion_p (type, TREE_TYPE (@3)) && integer_pow2p (@2) && tree_int_cst_sgn (@2) > 0) (bit_and @0 (convert (minus @1 { build_int_cst (TREE_TYPE (@1), 1); })))))) so it also checks whether @2 is not negative, the new pattern lacks this and could make use of power_of_two_cand as well (in fact I'd place it next to the pattern above.Thank you for your suggestions. I have modified the code according to your suggestions. But power_of_two_cand is not used, it will cause pr70354-2.c failed ((0x1234ULL << (i % 54)) will transfer to (0x1234ULL << (i & 54))).
How exactly did you use power_of_two_cand? As shown in the transformation above, it combines with integer_pow2p, it doesn't replace it.
-- Marc Glisse
