Currently if we have a range of [0,0] and we set the nonzero bits to 1, the current code pessimizes the range to [0,1] because it assumes the range is [1,1] plus the possibility of 0. This fixes the oversight.
gcc/ChangeLog: * value-range.cc (irange::set_nonzero_bits): Do not pessimize range. (range_tests_nonzero_bits): New test. --- gcc/value-range.cc | 13 +++++++++++++ 1 file changed, 13 insertions(+) diff --git a/gcc/value-range.cc b/gcc/value-range.cc index e1066f4946e..6e196574de9 100644 --- a/gcc/value-range.cc +++ b/gcc/value-range.cc @@ -2934,6 +2934,14 @@ irange::set_nonzero_bits (const wide_int_ref &bits) // range immediately. if (wi::popcount (bits) == 1) { + // Make sure we don't pessimize the range. + tree tbits = wide_int_to_tree (type (), bits); + if (!contains_p (tbits)) + { + set_nonzero_bits (tbits); + return; + } + bool has_zero = contains_p (build_zero_cst (type ())); set (type (), bits, bits); if (has_zero) @@ -3628,6 +3636,11 @@ range_tests_nonzero_bits () r1.set_nonzero_bits (0xff); r0.union_ (r1); ASSERT_TRUE (r0.varying_p ()); + + // Test that setting a nonzero bit of 1 does not pessimize the range. + r0.set_zero (integer_type_node); + r0.set_nonzero_bits (1); + ASSERT_TRUE (r0.zero_p ()); } // Build an frange from string endpoints. -- 2.37.1