6658428: C2 doesn't inline java method if corresponding intrinsic failed to inline.

Summary: Allow fallback to non-intrinsic inline case
Reviewed-by: kvn, jrose, never
Contributed-by: nils.eliasson@oracle.com

src/share/vm/opto/compile.hpp

@@ -631,7 +631,7 @@ class Compile : public Phase {
 
   // Decide how to build a call.
   // The profile factor is a discount to apply to this site's interp. profile.
-  CallGenerator*    call_generator(ciMethod* call_method, int vtable_index, bool call_is_virtual, JVMState* jvms, bool allow_inline, float profile_factor);
+  CallGenerator*    call_generator(ciMethod* call_method, int vtable_index, bool call_is_virtual, JVMState* jvms, bool allow_inline, float profile_factor, bool allow_intrinsics = true);
   bool should_delay_inlining(ciMethod* call_method, JVMState* jvms);
 
   // Report if there were too many traps at a current method and bci.

src/share/vm/opto/doCall.cpp

@@ -61,7 +61,7 @@ void trace_type_profile(ciMethod *method, int depth, int bci, ciMethod *prof_met
 
 CallGenerator* Compile::call_generator(ciMethod* call_method, int vtable_index, bool call_is_virtual,
                                        JVMState* jvms, bool allow_inline,
-                                       float prof_factor) {
+                                       float prof_factor, bool allow_intrinsics) {
   ciMethod*       caller   = jvms->method();
   int             bci      = jvms->bci();
   Bytecodes::Code bytecode = caller->java_code_at_bci(bci);
@@ -108,7 +108,7 @@ CallGenerator* Compile::call_generator(ciMethod* call_method, int vtable_index,
   // then we return it as the inlined version of the call.
   // We do this before the strict f.p. check below because the
   // intrinsics handle strict f.p. correctly.
-  if (allow_inline) {
+  if (allow_inline && allow_intrinsics) {
     CallGenerator* cg = find_intrinsic(call_method, call_is_virtual);
     if (cg != NULL)  return cg;
   }
@@ -455,21 +455,12 @@ void Parse::do_call() {
     // cg->generate(), we are committed.  If it fails, the whole
     // compilation task is compromised.
     if (failing())  return;
-#ifndef PRODUCT
-    if (PrintOpto || PrintOptoInlining || PrintInlining) {
-      // Only one fall-back, so if an intrinsic fails, ignore any bytecodes.
-      if (cg->is_intrinsic() && call_method->code_size() > 0) {
-        tty->print("Bailed out of intrinsic, will not inline: ");
-        call_method->print_name(); tty->cr();
-      }
-    }
-#endif
+
     // This can happen if a library intrinsic is available, but refuses
     // the call site, perhaps because it did not match a pattern the
-    // intrinsic was expecting to optimize.  The fallback position is
-    // to call out-of-line.
-    try_inline = false;  // Inline tactic bailed out.
-    cg = C->call_generator(call_method, vtable_index, call_is_virtual, jvms, try_inline, prof_factor());
+    // intrinsic was expecting to optimize. Should always be possible to
+    // get a normal java call that may inline in that case
+    cg = C->call_generator(call_method, vtable_index, call_is_virtual, jvms, try_inline, prof_factor(), /* allow_intrinsics= */ false);
     if ((new_jvms = cg->generate(jvms)) == NULL) {
       guarantee(failing(), "call failed to generate:  calls should work");
       return;

src/share/vm/opto/library_call.cpp

@@ -338,8 +338,27 @@ CallGenerator* Compile::make_vm_intrinsic(ciMethod* m, bool is_virtual) {
     break;
 
   case vmIntrinsics::_bitCount_i:
+    if (!Matcher::has_match_rule(Op_PopCountI)) return NULL;
+    break;
+
   case vmIntrinsics::_bitCount_l:
-    if (!UsePopCountInstruction)  return NULL;
+    if (!Matcher::has_match_rule(Op_PopCountL)) return NULL;
+    break;
+
+  case vmIntrinsics::_numberOfLeadingZeros_i:
+    if (!Matcher::match_rule_supported(Op_CountLeadingZerosI)) return NULL;
+    break;
+
+  case vmIntrinsics::_numberOfLeadingZeros_l:
+    if (!Matcher::match_rule_supported(Op_CountLeadingZerosL)) return NULL;
+    break;
+
+  case vmIntrinsics::_numberOfTrailingZeros_i:
+    if (!Matcher::match_rule_supported(Op_CountTrailingZerosI)) return NULL;
+    break;
+
+  case vmIntrinsics::_numberOfTrailingZeros_l:
+    if (!Matcher::match_rule_supported(Op_CountTrailingZerosL)) return NULL;
     break;
 
   case vmIntrinsics::_Reference_get:
@@ -416,14 +435,12 @@ JVMState* LibraryIntrinsic::generate(JVMState* jvms) {
     return kit.transfer_exceptions_into_jvms();
   }
 
-  if (PrintIntrinsics) {
+  // The intrinsic bailed out
+  if (PrintIntrinsics || PrintInlining NOT_PRODUCT( || PrintOptoInlining) ) {
     if (jvms->has_method()) {
       // Not a root compile.
-      tty->print("Did not inline intrinsic %s%s at bci:%d in",
-                 vmIntrinsics::name_at(intrinsic_id()),
-                 (is_virtual() ? " (virtual)" : ""), kit.bci());
-      kit.caller()->print_short_name(tty);
-      tty->print_cr(" (%d bytes)", kit.caller()->code_size());
+      const char* msg = is_virtual() ? "failed to inline (intrinsic, virtual)" : "failed to inline (intrinsic)";
+      CompileTask::print_inlining(kit.callee(), jvms->depth() - 1, kit.bci(), msg);
     } else {
       // Root compile
       tty->print("Did not generate intrinsic %s%s at bci:%d in",
@@ -5453,4 +5470,3 @@ bool LibraryCallKit::inline_reference_get() {
   push(result);
   return true;
 }
-