c: Fix ICE with -fexcess-precision=standard [PR99136]
authorJakub Jelinek <jakub@redhat.com>
Thu, 18 Feb 2021 21:17:52 +0000 (22:17 +0100)
committerJakub Jelinek <jakub@redhat.com>
Thu, 18 Feb 2021 21:17:52 +0000 (22:17 +0100)
commitd82f829905cfe6cb47d073825f680900274ce764
treeece022c89d1708ed44180958ed3719accb71ce16
parent1021222ee4d291ccb4f49cd0ae3393c83d8ff5d0
c: Fix ICE with -fexcess-precision=standard [PR99136]

The following testcase ICEs on i686-linux, because c_finish_return wraps
c_fully_folded retval back into EXCESS_PRECISION_EXPR, but when the function
return type is void, we don't call convert_for_assignment on it that would
then be fully folded again, but just put the retval into RETURN_EXPR's
operand, so nothing removes it anymore and during gimplification we
ICE as EXCESS_PRECISION_EXPR is not handled.

This patch fixes it by not adding that EXCESS_PRECISION_EXPR in functions
returning void, the return value is ignored and all we need is evaluate any
side-effects of the expression.

2021-02-18  Jakub Jelinek  <jakub@redhat.com>

PR c/99136
* c-typeck.c (c_finish_return): Don't wrap retval into
EXCESS_PRECISION_EXPR in functions that return void.

* gcc.dg/pr99136.c: New test.
gcc/c/c-typeck.c
gcc/testsuite/gcc.dg/pr99136.c [new file with mode: 0644]