8007144: Incremental inlining mistakes some call sites for dead ones and doesn't inline them
Summary: wrong detection for dead call sites.
Reviewed-by: kvn
--- a/hotspot/src/share/vm/opto/callGenerator.cpp Mon Feb 04 10:28:39 2013 -0800
+++ b/hotspot/src/share/vm/opto/callGenerator.cpp Mon Feb 04 11:30:37 2013 +0100
@@ -305,11 +305,13 @@
void LateInlineCallGenerator::do_late_inline() {
// Can't inline it
if (call_node() == NULL || call_node()->outcnt() == 0 ||
- call_node()->in(0) == NULL || call_node()->in(0)->is_top())
+ call_node()->in(0) == NULL || call_node()->in(0)->is_top()) {
return;
+ }
+ const TypeTuple *r = call_node()->tf()->domain();
for (int i1 = 0; i1 < method()->arg_size(); i1++) {
- if (call_node()->in(TypeFunc::Parms + i1)->is_top()) {
+ if (call_node()->in(TypeFunc::Parms + i1)->is_top() && r->field_at(TypeFunc::Parms + i1) != Type::HALF) {
assert(Compile::current()->inlining_incrementally(), "shouldn't happen during parsing");
return;
}