This fixes the previous adjustment to operation_could_trap_helper_p where I failed to realize that RDIV_EXPR is also used for fixed-point types. It also fixes that handling by properly checking for a fixed_zerop divisor.
Bootstrapped and tested on x86_64-unknown-linux-gnu, OK? Thanks, Richard. 2021-11-16 Richard Biener <rguent...@suse.de> PR middle-end/103248 * tree-eh.c (operation_could_trap_helper_p): Properly handle fixed-point RDIV_EXPR. * gcc.dg/pr103248.c: New testcase. --- gcc/testsuite/gcc.dg/pr103248.c | 8 ++++++++ gcc/tree-eh.c | 12 +++++++++--- 2 files changed, 17 insertions(+), 3 deletions(-) create mode 100644 gcc/testsuite/gcc.dg/pr103248.c diff --git a/gcc/testsuite/gcc.dg/pr103248.c b/gcc/testsuite/gcc.dg/pr103248.c new file mode 100644 index 00000000000..da6232d21ee --- /dev/null +++ b/gcc/testsuite/gcc.dg/pr103248.c @@ -0,0 +1,8 @@ +/* { dg-do compile } */ +/* { dg-require-effective-target fixed_point } */ +/* { dg-options "-fnon-call-exceptions" } */ + +_Accum sa; +int c; + +void div_csa() { c /= sa; } diff --git a/gcc/tree-eh.c b/gcc/tree-eh.c index 3eff07fc8fe..916da85af2e 100644 --- a/gcc/tree-eh.c +++ b/gcc/tree-eh.c @@ -2474,10 +2474,16 @@ operation_could_trap_helper_p (enum tree_code op, return false; case RDIV_EXPR: - if (honor_snans) + if (fp_operation) + { + if (honor_snans) + return true; + return flag_trapping_math; + } + /* Fixed point operations also use RDIV_EXPR. */ + if (!TREE_CONSTANT (divisor) || fixed_zerop (divisor)) return true; - gcc_assert (fp_operation); - return flag_trapping_math; + return false; case LT_EXPR: case LE_EXPR: -- 2.31.1