--- a/src/share/vm/c1/c1_GraphBuilder.cpp	Fri Mar 09 00:28:20 2012 -0500
+++ b/src/share/vm/c1/c1_GraphBuilder.cpp	Fri Mar 09 08:34:36 2012 -0800
@@ -1306,6 +1306,7 @@
       if (sw.dest_offset_at(i) < 0) has_bb = true;
     }
     // add default successor
+    if (sw.default_offset() < 0) has_bb = true;
     sux->at_put(i, block_at(bci() + sw.default_offset()));
     ValueStack* state_before = has_bb ? copy_state_before() : NULL;
     Instruction* res = append(new TableSwitch(ipop(), sux, sw.low_key(), state_before, has_bb));
@@ -1350,6 +1351,7 @@
       keys->at_put(i, pair.match());
     }
     // add default successor
+    if (sw.default_offset() < 0) has_bb = true;
     sux->at_put(i, block_at(bci() + sw.default_offset()));
     ValueStack* state_before = has_bb ? copy_state_before() : NULL;
     Instruction* res = append(new LookupSwitch(ipop(), sux, keys, state_before, has_bb));

--- a/src/share/vm/opto/compile.hpp	Fri Mar 09 00:28:20 2012 -0500
+++ b/src/share/vm/opto/compile.hpp	Fri Mar 09 08:34:36 2012 -0800
@@ -631,7 +631,7 @@
 
   // 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.

--- a/src/share/vm/opto/doCall.cpp	Fri Mar 09 00:28:20 2012 -0500
+++ b/src/share/vm/opto/doCall.cpp	Fri Mar 09 08:34:36 2012 -0800
@@ -61,7 +61,7 @@
 
 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 @@
   // 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 @@
     // 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;

--- a/src/share/vm/opto/library_call.cpp	Fri Mar 09 00:28:20 2012 -0500
+++ b/src/share/vm/opto/library_call.cpp	Fri Mar 09 08:34:36 2012 -0800
@@ -338,8 +338,27 @@
     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 @@
     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 @@
   push(result);
   return true;
 }
-

--- a/src/share/vm/runtime/safepoint.cpp	Fri Mar 09 00:28:20 2012 -0500
+++ b/src/share/vm/runtime/safepoint.cpp	Fri Mar 09 08:34:36 2012 -0800
@@ -219,6 +219,8 @@
 #ifdef ASSERT
   for (JavaThread *cur = Threads::first(); cur != NULL; cur = cur->next()) {
     assert(cur->safepoint_state()->is_running(), "Illegal initial state");
+    // Clear the visited flag to ensure that the critical counts are collected properly.
+    cur->set_visited_for_critical_count(false);
   }
 #endif // ASSERT
 
@@ -378,6 +380,13 @@
 
   OrderAccess::fence();
 
+#ifdef ASSERT
+  for (JavaThread *cur = Threads::first(); cur != NULL; cur = cur->next()) {
+    // make sure all the threads were visited
+    assert(cur->was_visited_for_critical_count(), "missed a thread");
+  }
+#endif // ASSERT
+
   // Update the count of active JNI critical regions
   GC_locker::set_jni_lock_count(_current_jni_active_count);
 
@@ -626,6 +635,7 @@
         _waiting_to_block--;
         thread->safepoint_state()->set_has_called_back(true);
 
+        DEBUG_ONLY(thread->set_visited_for_critical_count(true));
         if (thread->in_critical()) {
           // Notice that this thread is in a critical section
           increment_jni_active_count();
@@ -907,12 +917,8 @@
   // running, but are actually at a safepoint. We will happily
   // agree and update the safepoint state here.
   if (SafepointSynchronize::safepoint_safe(_thread, state)) {
+    SafepointSynchronize::check_for_lazy_critical_native(_thread, state);
     roll_forward(_at_safepoint);
-    SafepointSynchronize::check_for_lazy_critical_native(_thread, state);
-    if (_thread->in_critical()) {
-      // Notice that this thread is in a critical section
-      SafepointSynchronize::increment_jni_active_count();
-    }
     return;
   }
 
@@ -937,6 +943,11 @@
   switch(_type) {
     case _at_safepoint:
       SafepointSynchronize::signal_thread_at_safepoint();
+      DEBUG_ONLY(_thread->set_visited_for_critical_count(true));
+      if (_thread->in_critical()) {
+        // Notice that this thread is in a critical section
+        SafepointSynchronize::increment_jni_active_count();
+      }
       break;
 
     case _call_back:

--- a/src/share/vm/runtime/thread.cpp	Fri Mar 09 00:28:20 2012 -0500
+++ b/src/share/vm/runtime/thread.cpp	Fri Mar 09 08:34:36 2012 -0800
@@ -247,6 +247,10 @@
   omInUseList = NULL ;
   omInUseCount = 0 ;
 
+#ifdef ASSERT
+  _visited_for_critical_count = false;
+#endif
+
   _SR_lock = new Monitor(Mutex::suspend_resume, "SR_lock", true);
   _suspend_flags = 0;
 

--- a/src/share/vm/runtime/thread.hpp	Fri Mar 09 00:28:20 2012 -0500
+++ b/src/share/vm/runtime/thread.hpp	Fri Mar 09 08:34:36 2012 -0800
@@ -268,6 +268,15 @@
   ObjectMonitor* omInUseList;                   // SLL to track monitors in circulation
   int omInUseCount;                             // length of omInUseList
 
+#ifdef ASSERT
+ private:
+  bool _visited_for_critical_count;
+
+ public:
+  void set_visited_for_critical_count(bool z) { _visited_for_critical_count = z; }
+  bool was_visited_for_critical_count() const   { return _visited_for_critical_count; }
+#endif
+
  public:
   enum {
     is_definitely_current_thread = true