8320276: Improve class initialization barrier in TemplateTable::_new

Reviewed-by: dholmes, fparain

src/hotspot/cpu/aarch64/templateTable_aarch64.cpp

@@ -3603,11 +3603,9 @@ void TemplateTable::_new() {
   // get InstanceKlass
   __ load_resolved_klass_at_offset(r4, r3, r4, rscratch1);
 
-  // make sure klass is initialized & doesn't have finalizer
-  // make sure klass is fully initialized
-  __ ldrb(rscratch1, Address(r4, InstanceKlass::init_state_offset()));
-  __ cmp(rscratch1, (u1)InstanceKlass::fully_initialized);
-  __ br(Assembler::NE, slow_case);
+  // make sure klass is initialized
+  assert(VM_Version::supports_fast_class_init_checks(), "Optimization requires support for fast class initialization checks");
+  __ clinit_barrier(r4, rscratch1, nullptr /*L_fast_path*/, &slow_case);
 
   // get instance_size in InstanceKlass (scaled to a count of bytes)
   __ ldrw(r3,

src/hotspot/cpu/aarch64/vm_version_aarch64.hpp

@@ -165,6 +165,7 @@ enum Ampere_CPU_Model {
   static int dcache_line_size() { return _dcache_line_size; }
   static int get_initial_sve_vector_length()  { return _initial_sve_vector_length; };
 
+  // Aarch64 supports fast class initialization checks
   static bool supports_fast_class_init_checks() { return true; }
   constexpr static bool supports_stack_watermark_barrier() { return true; }
 

src/hotspot/cpu/ppc/vm_version_ppc.hpp

@@ -91,7 +91,7 @@ public:
   // Override Abstract_VM_Version implementation
   static void print_platform_virtualization_info(outputStream*);
 
-  // PPC64 supports fast class initialization checks for static methods.
+  // PPC64 supports fast class initialization checks
   static bool supports_fast_class_init_checks() { return true; }
   constexpr static bool supports_stack_watermark_barrier() { return true; }
 

src/hotspot/cpu/s390/vm_version_s390.hpp

@@ -410,7 +410,7 @@ class VM_Version: public Abstract_VM_Version {
   // Override Abstract_VM_Version implementation
   static void print_platform_virtualization_info(outputStream*);
 
-  // s390 supports fast class initialization checks for static methods.
+  // s390 supports fast class initialization checks
   static bool supports_fast_class_init_checks() { return true; }
 
   // CPU feature query functions

src/hotspot/cpu/x86/templateTable_x86.cpp

@@ -4048,11 +4048,16 @@ void TemplateTable::_new() {
   __ load_resolved_klass_at_index(rcx, rcx, rdx);
   __ push(rcx);  // save the contexts of klass for initializing the header
 
-  // make sure klass is initialized & doesn't have finalizer
-  // make sure klass is fully initialized
+  // make sure klass is initialized
+#ifdef _LP64
+  assert(VM_Version::supports_fast_class_init_checks(), "must support fast class initialization checks");
+  __ clinit_barrier(rcx, r15_thread, nullptr /*L_fast_path*/, &slow_case);
+#else
   __ cmpb(Address(rcx, InstanceKlass::init_state_offset()), InstanceKlass::fully_initialized);
   __ jcc(Assembler::notEqual, slow_case);
+#endif
 
+  // make sure klass doesn't have finalizer
   // get instance_size in InstanceKlass (scaled to a count of bytes)
   __ movl(rdx, Address(rcx, Klass::layout_helper_offset()));
   // test to see if it has a finalizer or is malformed in some way

src/hotspot/cpu/x86/vm_version_x86.hpp

@@ -756,7 +756,7 @@ public:
   // the intrinsic for java.lang.Thread.onSpinWait()
   static bool supports_on_spin_wait() { return supports_sse2(); }
 
-  // x86_64 supports fast class initialization checks for static methods.
+  // x86_64 supports fast class initialization checks
   static bool supports_fast_class_init_checks() {
     return LP64_ONLY(true) NOT_LP64(false); // not implemented on x86_32
   }