x86_64/isel: fix edge case with branch on float add result

In `if (x + 1)` the implicit `!= 0` can be omitted for integers, because
the x86 `add` instruction sets the zero flag accordingly. But for floats
it doesn't, so applying that optimization there was wrong. Luckily it
was caught by a nullptr dereference later in `emit::Xjcc()` for the
(missing) unordered branch target, instead of miscompiling.

1 files changed, 5 insertions, 0 deletions

diff --git a/test/12-flt.c b/test/12-flt.c
index 36f4e90..5077026 100644
--- a/test/12-flt.c
+++ b/test/12-flt.c
@@ -40,6 +40,11 @@ double nummod(double a, double b) {
    return m;
 }
 
+double branch(double a) {
+   if (a + 1.1) return 0.0;
+   return -a;
+}
+
 int main() {
    NAN/=0.0;
    assert(_(1.0) < _(2.0));