This ICEs in ubsan_instrument_division on the assert that checks whether both operands of the division have the same type. Well, here they didn't, because in cp_build_binary_op we first converted both operands to result_type but then fold_non_dependent_expr changed the type of op0, so we need to catch this case before calling ubsan_instrument_division.
Bootstrapped/regtested on x86_64-linux, ok for trunk? 2017-04-07 Marek Polacek <pola...@redhat.com> PR sanitizer/80348 * typeck.c (cp_build_binary_op): Convert COP[01] to ORIG_TYPE. * g++.dg/ubsan/div-by-zero-2.C: New test. diff --git gcc/cp/typeck.c gcc/cp/typeck.c index 79391c0..65a3435 100644 --- gcc/cp/typeck.c +++ gcc/cp/typeck.c @@ -5218,10 +5218,12 @@ cp_build_binary_op (location_t location, original result_type. */ tree cop0 = op0; tree cop1 = op1; - if (orig_type != NULL && result_type != orig_type) + if (orig_type != NULL_TREE) { - cop0 = cp_convert (orig_type, op0, complain); - cop1 = cp_convert (orig_type, op1, complain); + if (TREE_TYPE (cop0) != orig_type) + cop0 = cp_convert (orig_type, op0, complain); + if (TREE_TYPE (cop1) != orig_type) + cop1 = cp_convert (orig_type, op1, complain); } instrument_expr = ubsan_instrument_division (location, cop0, cop1); } diff --git gcc/testsuite/g++.dg/ubsan/div-by-zero-2.C gcc/testsuite/g++.dg/ubsan/div-by-zero-2.C index e69de29..d500ae6 100644 --- gcc/testsuite/g++.dg/ubsan/div-by-zero-2.C +++ gcc/testsuite/g++.dg/ubsan/div-by-zero-2.C @@ -0,0 +1,10 @@ +// PR sanitizer/80348 +// { dg-do compile } +// { dg-options "-fsanitize=integer-divide-by-zero" } + +void +foo () +{ + if (0) + unsigned ((0 != 60806) > (0 != 0)) / 0; // { dg-warning "division by zero" } +} Marek