8252990: Intrinsify Unsafe.storeStoreFence

Reviewed-by: dholmes, thartmann, whuang

src/hotspot/cpu/aarch64/aarch64.ad

@@ -8547,6 +8547,7 @@ instruct membar_release() %{
 
 instruct membar_storestore() %{
   match(MemBarStoreStore);
+  match(StoreStoreFence);
   ins_cost(VOLATILE_REF_COST);
 
   format %{ "MEMBAR-store-store" %}

src/hotspot/cpu/arm/arm.ad

@@ -4517,6 +4517,7 @@ instruct storeF( memoryF mem, regF src) %{
 // pattern-match out unnecessary membars
 instruct membar_storestore() %{
   match(MemBarStoreStore);
+  match(StoreStoreFence);
   ins_cost(4*MEMORY_REF_COST);
 
   size(4);

src/hotspot/cpu/ppc/ppc.ad

@@ -7153,6 +7153,7 @@ instruct membar_release() %{
 
 instruct membar_storestore() %{
   match(MemBarStoreStore);
+  match(StoreStoreFence);
   ins_cost(4*MEMORY_REF_COST);
 
   format %{ "MEMBAR-store-store" %}

src/hotspot/cpu/s390/s390.ad

@@ -5055,6 +5055,7 @@ instruct membar_CPUOrder() %{
 
 instruct membar_storestore() %{
   match(MemBarStoreStore);
+  match(StoreStoreFence);
   ins_cost(0);
   size(0);
   format %{ "MEMBAR-storestore (empty)" %}

src/hotspot/cpu/x86/x86_32.ad

@@ -6659,6 +6659,7 @@ instruct unnecessary_membar_volatile() %{
 
 instruct membar_storestore() %{
   match(MemBarStoreStore);
+  match(StoreStoreFence);
   ins_cost(0);
 
   size(0);

src/hotspot/cpu/x86/x86_64.ad

@@ -6787,6 +6787,7 @@ instruct unnecessary_membar_volatile()
 
 instruct membar_storestore() %{
   match(MemBarStoreStore);
+  match(StoreStoreFence);
   ins_cost(0);
 
   size(0);

src/hotspot/share/adlc/formssel.cpp

@@ -4117,6 +4117,7 @@ bool MatchRule::is_ideal_membar() const {
     !strcmp(_opType,"MemBarReleaseLock") ||
     !strcmp(_opType,"LoadFence" ) ||
     !strcmp(_opType,"StoreFence") ||
+    !strcmp(_opType,"StoreStoreFence") ||
     !strcmp(_opType,"MemBarVolatile") ||
     !strcmp(_opType,"MemBarCPUOrder") ||
     !strcmp(_opType,"MemBarStoreStore") ||

src/hotspot/share/c1/c1_Compiler.cpp

@@ -142,6 +142,7 @@ bool Compiler::is_intrinsic_supported(const methodHandle& method) {
     // since GC can change its value.
   case vmIntrinsics::_loadFence:
   case vmIntrinsics::_storeFence:
+  case vmIntrinsics::_storeStoreFence:
   case vmIntrinsics::_fullFence:
   case vmIntrinsics::_floatToRawIntBits:
   case vmIntrinsics::_intBitsToFloat:

src/hotspot/share/c1/c1_LIRGenerator.cpp

@@ -2984,6 +2984,9 @@ void LIRGenerator::do_Intrinsic(Intrinsic* x) {
   case vmIntrinsics::_storeFence:
     __ membar_release();
     break;
+  case vmIntrinsics::_storeStoreFence:
+    __ membar_storestore();
+    break;
   case vmIntrinsics::_fullFence :
     __ membar();
     break;

src/hotspot/share/classfile/vmIntrinsics.hpp

@@ -523,6 +523,9 @@ class methodHandle;
   do_intrinsic(_storeFence,               jdk_internal_misc_Unsafe,     storeFence_name, storeFence_signature,         F_RN)     \
    do_name(     storeFence_name,                                        "storeFence")                                            \
    do_alias(    storeFence_signature,                                   void_method_signature)                                   \
+  do_intrinsic(_storeStoreFence,          jdk_internal_misc_Unsafe,     storeStoreFence_name, storeStoreFence_signature, F_R)    \
+   do_name(     storeStoreFence_name,                                   "storeStoreFence")                                       \
+   do_alias(    storeStoreFence_signature,                              void_method_signature)                                   \
   do_intrinsic(_fullFence,                jdk_internal_misc_Unsafe,     fullFence_name, fullFence_signature,           F_RN)     \
    do_name(     fullFence_name,                                         "fullFence")                                             \
    do_alias(    fullFence_signature,                                    void_method_signature)                                   \

src/hotspot/share/opto/c2compiler.cpp

@@ -604,6 +604,7 @@ bool C2Compiler::is_intrinsic_supported(const methodHandle& method, bool is_virt
   case vmIntrinsics::_putLongUnaligned:
   case vmIntrinsics::_loadFence:
   case vmIntrinsics::_storeFence:
+  case vmIntrinsics::_storeStoreFence:
   case vmIntrinsics::_fullFence:
   case vmIntrinsics::_currentThread:
 #ifdef JFR_HAVE_INTRINSICS

src/hotspot/share/opto/classes.hpp

@@ -219,6 +219,7 @@ macro(MemBarAcquireLock)
 macro(MemBarCPUOrder)
 macro(MemBarRelease)
 macro(StoreFence)
+macro(StoreStoreFence)
 macro(MemBarReleaseLock)
 macro(MemBarVolatile)
 macro(MemBarStoreStore)

src/hotspot/share/opto/library_call.cpp

@@ -467,6 +467,7 @@ bool LibraryCallKit::try_to_inline(int predicate) {
 
   case vmIntrinsics::_loadFence:
   case vmIntrinsics::_storeFence:
+  case vmIntrinsics::_storeStoreFence:
   case vmIntrinsics::_fullFence:                return inline_unsafe_fence(intrinsic_id());
 
   case vmIntrinsics::_onSpinWait:               return inline_onspinwait();
@@ -2695,6 +2696,9 @@ bool LibraryCallKit::inline_unsafe_fence(vmIntrinsics::ID id) {
     case vmIntrinsics::_storeFence:
       insert_mem_bar(Op_StoreFence);
       return true;
+    case vmIntrinsics::_storeStoreFence:
+      insert_mem_bar(Op_StoreStoreFence);
+      return true;
     case vmIntrinsics::_fullFence:
       insert_mem_bar(Op_MemBarVolatile);
       return true;

src/hotspot/share/opto/memnode.cpp

@@ -1080,7 +1080,8 @@ Node* MemNode::can_see_stored_value(Node* st, PhaseTransform* phase) const {
           opc == Op_MemBarRelease ||
           opc == Op_StoreFence ||
           opc == Op_MemBarReleaseLock ||
-          opc == Op_MemBarStoreStore) {
+          opc == Op_MemBarStoreStore ||
+          opc == Op_StoreStoreFence) {
         Node* mem = current->in(0)->in(TypeFunc::Memory);
         if (mem->is_MergeMem()) {
           MergeMemNode* merge = mem->as_MergeMem();
@@ -3300,13 +3301,14 @@ MemBarNode* MemBarNode::make(Compile* C, int opcode, int atp, Node* pn) {
   case Op_LoadFence:         return new LoadFenceNode(C, atp, pn);
   case Op_MemBarRelease:     return new MemBarReleaseNode(C, atp, pn);
   case Op_StoreFence:        return new StoreFenceNode(C, atp, pn);
+  case Op_MemBarStoreStore:  return new MemBarStoreStoreNode(C, atp, pn);
+  case Op_StoreStoreFence:   return new StoreStoreFenceNode(C, atp, pn);
   case Op_MemBarAcquireLock: return new MemBarAcquireLockNode(C, atp, pn);
   case Op_MemBarReleaseLock: return new MemBarReleaseLockNode(C, atp, pn);
   case Op_MemBarVolatile:    return new MemBarVolatileNode(C, atp, pn);
   case Op_MemBarCPUOrder:    return new MemBarCPUOrderNode(C, atp, pn);
   case Op_OnSpinWait:        return new OnSpinWaitNode(C, atp, pn);
   case Op_Initialize:        return new InitializeNode(C, atp, pn);
-  case Op_MemBarStoreStore:  return new MemBarStoreStoreNode(C, atp, pn);
   case Op_Blackhole:         return new BlackholeNode(C, atp, pn);
   default: ShouldNotReachHere(); return NULL;
   }

src/hotspot/share/opto/memnode.hpp

@@ -1309,6 +1309,13 @@ public:
   virtual int Opcode() const;
 };
 
+class StoreStoreFenceNode: public MemBarNode {
+public:
+  StoreStoreFenceNode(Compile* C, int alias_idx, Node* precedent)
+    : MemBarNode(C, alias_idx, precedent) {}
+  virtual int Opcode() const;
+};
+
 // Ordering between a volatile store and a following volatile load.
 // Requires multi-CPU visibility?
 class MemBarVolatileNode: public MemBarNode {

src/java.base/share/classes/jdk/internal/misc/Unsafe.java

@@ -3441,16 +3441,14 @@ public final class Unsafe {
      * Ensures that stores before the fence will not be reordered with
      * stores after the fence.
      *
-     * @implNote
-     * This method is operationally equivalent to {@link #storeFence()}.
-     *
      * @since 9
      */
+    @IntrinsicCandidate
     public final void storeStoreFence() {
+        // If storeStoreFence intrinsic is not available, fall back to storeFence.
         storeFence();
     }
 
-
     /**
      * Throws IllegalAccessError; for use by the VM for access control
      * error support.