8252372: Check if cloning is required to move loads out of loops in P…

…haseIdealLoop::split_if_with_blocks_post()

Reviewed-by: thartmann, kvn

src/hotspot/cpu/aarch64/aarch64.ad

@@ -8735,6 +8735,61 @@ instruct castLL(iRegL dst)
   ins_pipe(pipe_class_empty);
 %}
 
+instruct castFF(vRegF dst)
+%{
+  match(Set dst (CastFF dst));
+
+  size(0);
+  format %{ "# castFF of $dst" %}
+  ins_encode(/* empty encoding */);
+  ins_cost(0);
+  ins_pipe(pipe_class_empty);
+%}
+
+instruct castDD(vRegD dst)
+%{
+  match(Set dst (CastDD dst));
+
+  size(0);
+  format %{ "# castDD of $dst" %}
+  ins_encode(/* empty encoding */);
+  ins_cost(0);
+  ins_pipe(pipe_class_empty);
+%}
+
+instruct castVVD(vecD dst)
+%{
+  match(Set dst (CastVV dst));
+
+  size(0);
+  format %{ "# castVV of $dst" %}
+  ins_encode(/* empty encoding */);
+  ins_cost(0);
+  ins_pipe(pipe_class_empty);
+%}
+
+instruct castVVX(vecX dst)
+%{
+  match(Set dst (CastVV dst));
+
+  size(0);
+  format %{ "# castVV of $dst" %}
+  ins_encode(/* empty encoding */);
+  ins_cost(0);
+  ins_pipe(pipe_class_empty);
+%}
+
+instruct castVV(vReg dst)
+%{
+  match(Set dst (CastVV dst));
+
+  size(0);
+  format %{ "# castVV of $dst" %}
+  ins_encode(/* empty encoding */);
+  ins_cost(0);
+  ins_pipe(pipe_class_empty);
+%}
+
 // ============================================================================
 // Atomic operation instructions
 //

src/hotspot/cpu/arm/arm.ad

@@ -5182,6 +5182,39 @@ instruct castLL( iRegL dst ) %{
   ins_pipe(empty);
 %}
 
+instruct castFF( regF dst ) %{
+  match(Set dst (CastFF dst));
+  format %{ "! castFF of $dst" %}
+  ins_encode( /*empty encoding*/ );
+  ins_cost(0);
+  ins_pipe(empty);
+%}
+
+instruct castDD( regD dst ) %{
+  match(Set dst (CastDD dst));
+  format %{ "! castDD of $dst" %}
+  ins_encode( /*empty encoding*/ );
+  ins_cost(0);
+  ins_pipe(empty);
+%}
+
+instruct castVVD( vecD dst ) %{
+  match(Set dst (CastVV dst));
+  format %{ "! castVV of $dst" %}
+  ins_encode( /*empty encoding*/ );
+  ins_cost(0);
+  ins_pipe(empty);
+%}
+
+instruct castVVX( vecX dst ) %{
+  match(Set dst (CastVV dst));
+  format %{ "! castVV of $dst" %}
+  ins_encode( /*empty encoding*/ );
+  ins_cost(0);
+  ins_pipe(empty);
+%}
+
+
 //----------Arithmetic Instructions--------------------------------------------
 // Addition Instructions
 // Register Addition

src/hotspot/cpu/ppc/ppc.ad

@@ -10335,6 +10335,38 @@ instruct castLL(iRegLdst dst) %{
   ins_pipe(pipe_class_default);
 %}
 
+instruct castFF(regF dst) %{
+  match(Set dst (CastFF dst));
+  format %{ " -- \t// castFF of $dst" %}
+  size(0);
+  ins_encode( /*empty*/ );
+  ins_pipe(pipe_class_default);
+%}
+
+instruct castDD(regD dst) %{
+  match(Set dst (CastDD dst));
+  format %{ " -- \t// castDD of $dst" %}
+  size(0);
+  ins_encode( /*empty*/ );
+  ins_pipe(pipe_class_default);
+%}
+
+instruct castVV8(iRegLdst dst) %{
+  match(Set dst (CastVV dst));
+  format %{ " -- \t// castVV of $dst" %}
+  size(0);
+  ins_encode( /*empty*/ );
+  ins_pipe(pipe_class_default);
+%}
+
+instruct castVV16(vecX dst) %{
+  match(Set dst (CastVV dst));
+  format %{ " -- \t// castVV of $dst" %}
+  size(0);
+  ins_encode( /*empty*/ );
+  ins_pipe(pipe_class_default);
+%}
+
 instruct checkCastPP(iRegPdst dst) %{
   match(Set dst (CheckCastPP dst));
   format %{ " -- \t// checkcastPP of $dst" %}

src/hotspot/cpu/s390/s390.ad

@@ -5260,6 +5260,30 @@ instruct castLL(iRegL dst) %{
   ins_pipe(pipe_class_dummy);
 %}
 
+instruct castFF(regF dst) %{
+  match(Set dst (CastFF dst));
+  size(0);
+  format %{ "# castFF of $dst" %}
+  ins_encode(/*empty*/);
+  ins_pipe(pipe_class_dummy);
+%}
+
+instruct castDD(regD dst) %{
+  match(Set dst (CastDD dst));
+  size(0);
+  format %{ "# castDD of $dst" %}
+  ins_encode(/*empty*/);
+  ins_pipe(pipe_class_dummy);
+%}
+
+instruct castVV(iRegL dst) %{
+  match(Set dst (CastVV dst));
+  size(0);
+  format %{ "# castVV of $dst" %}
+  ins_encode(/*empty*/);
+  ins_pipe(pipe_class_dummy);
+%}
+
 //----------Conditional_store--------------------------------------------------
 // Conditional-store of the updated heap-top.
 // Used during allocation of the shared heap.

src/hotspot/cpu/x86/x86.ad

@@ -8127,3 +8127,25 @@ instruct vmask_first_or_last_true_avx(rRegI dst, vec mask, rRegL tmp, vec xtmp,
   ins_pipe( pipe_slow );
 %}
 #endif // _LP64
+
+instruct castVV(vec dst)
+%{
+  match(Set dst (CastVV dst));
+
+  size(0);
+  format %{ "# castVV of $dst" %}
+  ins_encode(/* empty encoding */);
+  ins_cost(0);
+  ins_pipe(empty);
+%}
+
+instruct castVVLeg(legVec dst)
+%{
+  match(Set dst (CastVV dst));
+
+  size(0);
+  format %{ "# castVV of $dst" %}
+  ins_encode(/* empty encoding */);
+  ins_cost(0);
+  ins_pipe(empty);
+%}

src/hotspot/cpu/x86/x86_32.ad

@@ -7178,6 +7178,22 @@ instruct castLL( eRegL dst ) %{
   ins_pipe( empty );
 %}
 
+instruct castFF( regF dst ) %{
+  match(Set dst (CastFF dst));
+  format %{ "#castFF of $dst" %}
+  ins_encode( /*empty encoding*/ );
+  ins_cost(0);
+  ins_pipe( empty );
+%}
+
+instruct castDD( regD dst ) %{
+  match(Set dst (CastDD dst));
+  format %{ "#castDD of $dst" %}
+  ins_encode( /*empty encoding*/ );
+  ins_cost(0);
+  ins_pipe( empty );
+%}
+
 // Load-locked - same as a regular pointer load when used with compare-swap
 instruct loadPLocked(eRegP dst, memory mem) %{
   match(Set dst (LoadPLocked mem));

src/hotspot/cpu/x86/x86_64.ad

@@ -7624,6 +7624,28 @@ instruct castLL(rRegL dst)
   ins_pipe(empty);
 %}
 
+instruct castFF(regF dst)
+%{
+  match(Set dst (CastFF dst));
+
+  size(0);
+  format %{ "# castFF of $dst" %}
+  ins_encode(/* empty encoding */);
+  ins_cost(0);
+  ins_pipe(empty);
+%}
+
+instruct castDD(regD dst)
+%{
+  match(Set dst (CastDD dst));
+
+  size(0);
+  format %{ "# castDD of $dst" %}
+  ins_encode(/* empty encoding */);
+  ins_cost(0);
+  ins_pipe(empty);
+%}
+
 // LoadP-locked same as a regular LoadP when used with compare-swap
 instruct loadPLocked(rRegP dst, memory mem)
 %{

src/hotspot/share/adlc/formssel.cpp

@@ -764,6 +764,11 @@ int InstructForm::memory_operand(FormDict &globals) const {
 bool InstructForm::captures_bottom_type(FormDict &globals) const {
   if (_matrule && _matrule->_rChild &&
       (!strcmp(_matrule->_rChild->_opType,"CastPP")       ||  // new result type
+       !strcmp(_matrule->_rChild->_opType,"CastDD")       ||
+       !strcmp(_matrule->_rChild->_opType,"CastFF")       ||
+       !strcmp(_matrule->_rChild->_opType,"CastII")       ||
+       !strcmp(_matrule->_rChild->_opType,"CastLL")       ||
+       !strcmp(_matrule->_rChild->_opType,"CastVV")       ||
        !strcmp(_matrule->_rChild->_opType,"CastX2P")      ||  // new result type
        !strcmp(_matrule->_rChild->_opType,"DecodeN")      ||
        !strcmp(_matrule->_rChild->_opType,"EncodeP")      ||

src/hotspot/share/opto/castnode.hpp

@@ -115,6 +115,37 @@ class CastLLNode: public ConstraintCastNode {
   virtual uint ideal_reg() const { return Op_RegL; }
 };
 
+class CastFFNode: public ConstraintCastNode {
+public:
+  CastFFNode(Node* n, const Type* t, bool carry_dependency = false)
+          : ConstraintCastNode(n, t, carry_dependency){
+    init_class_id(Class_CastFF);
+  }
+  virtual int Opcode() const;
+  virtual uint ideal_reg() const { return Op_RegF; }
+};
+
+class CastDDNode: public ConstraintCastNode {
+public:
+  CastDDNode(Node* n, const Type* t, bool carry_dependency = false)
+          : ConstraintCastNode(n, t, carry_dependency){
+    init_class_id(Class_CastDD);
+  }
+  virtual int Opcode() const;
+  virtual uint ideal_reg() const { return Op_RegD; }
+};
+
+class CastVVNode: public ConstraintCastNode {
+public:
+  CastVVNode(Node* n, const Type* t, bool carry_dependency = false)
+          : ConstraintCastNode(n, t, carry_dependency){
+    init_class_id(Class_CastVV);
+  }
+  virtual int Opcode() const;
+  virtual uint ideal_reg() const { return in(1)->ideal_reg(); }
+};
+
+
 //------------------------------CastPPNode-------------------------------------
 // cast pointer to pointer (different type)
 class CastPPNode: public ConstraintCastNode {

src/hotspot/share/opto/classes.hpp

@@ -62,8 +62,11 @@ macro(CallLeafNoFP)
 macro(CallRuntime)
 macro(CallNative)
 macro(CallStaticJava)
+macro(CastDD)
+macro(CastFF)
 macro(CastII)
 macro(CastLL)
+macro(CastVV)
 macro(CastX2P)
 macro(CastP2X)
 macro(CastPP)