[Backport] 8264104: Eliminate unnecessary vector mask conversion duri…

…ng VectorUnbox for floating point VectorMask

Summary: 8264104: Eliminate unnecessary vector mask conversion during VectorUnbox for floating point VectorMask
         without new files in src/cpu/aarch64/

Test Plan: ci jtreg

Reviewed-by: JoshuaZhuwj

Issue: #602

src/hotspot/share/adlc/formssel.cpp

@@ -4201,7 +4201,8 @@ bool MatchRule::is_vector() const {
     "FmaVD", "FmaVF","PopCountVI",
     // Next are not supported currently.
     "PackB","PackS","PackI","PackL","PackF","PackD","Pack2L","Pack2D",
-    "ExtractB","ExtractUB","ExtractC","ExtractS","ExtractI","ExtractL","ExtractF","ExtractD"
+    "ExtractB","ExtractUB","ExtractC","ExtractS","ExtractI","ExtractL","ExtractF","ExtractD",
+    "VectorMaskCast"
   };
   int cnt = sizeof(vector_list)/sizeof(char*);
   if (_rChild) {

src/hotspot/share/opto/classes.hpp

@@ -430,6 +430,7 @@ macro(VectorBoxAllocate)
 macro(VectorUnbox)
 macro(VectorMaskWrapper)
 macro(VectorMaskCmp)
+macro(VectorMaskCast)
 macro(VectorTest)
 macro(VectorBlend)
 macro(VectorRearrange)

src/hotspot/share/opto/vectornode.cpp

@@ -1095,6 +1095,13 @@ Node* VectorUnboxNode::Ideal(PhaseGVN* phase, bool can_reshape) {
       bool is_vector_mask    = vbox_klass->is_subclass_of(ciEnv::current()->vector_VectorMask_klass());
       bool is_vector_shuffle = vbox_klass->is_subclass_of(ciEnv::current()->vector_VectorShuffle_klass());
       if (is_vector_mask) {
+        if (in_vt->length_in_bytes() == out_vt->length_in_bytes() &&
+            Matcher::match_rule_supported_vector(Op_VectorMaskCast, out_vt->length(), out_vt->element_basic_type())) {
+          // Apply "VectorUnbox (VectorBox vmask) ==> VectorMaskCast (vmask)"
+          // directly. This could avoid the transformation ordering issue from
+          // "VectorStoreMask (VectorLoadMask vmask) => vmask".
+          return new VectorMaskCastNode(value, out_vt);
+        }
         // VectorUnbox (VectorBox vmask) ==> VectorLoadMask (VectorStoreMask vmask)
         value = phase->transform(VectorStoreMaskNode::make(*phase, value, in_vt->element_basic_type(), in_vt->length()));
         return new VectorLoadMaskNode(value, out_vt);

src/hotspot/share/opto/vectornode.hpp

@@ -1215,6 +1215,17 @@ class VectorStoreMaskNode : public VectorNode {
   static VectorStoreMaskNode* make(PhaseGVN& gvn, Node* in, BasicType in_type, uint num_elem);
 };
 
+class VectorMaskCastNode : public VectorNode {
+ public:
+  VectorMaskCastNode(Node* in, const TypeVect* vt) : VectorNode(in, vt) {
+    const TypeVect* in_vt = in->bottom_type()->is_vect();
+    assert(in_vt->length() == vt->length(), "vector length must match");
+    assert(type2aelembytes(in_vt->element_basic_type()) == type2aelembytes(vt->element_basic_type()), "element size must match");
+  }
+
+  virtual int Opcode() const;
+};
+
 // This is intended for use as a simple reinterpret node that has no cast.
 class VectorReinterpretNode : public VectorNode {
  private:

src/hotspot/share/runtime/vmStructs.cpp

@@ -1924,6 +1924,7 @@ typedef PaddedEnd<ObjectMonitor>              PaddedObjectMonitor;
   declare_c2_type(VectorInsertNode, VectorNode)                           \
   declare_c2_type(VectorUnboxNode, VectorNode)                            \
   declare_c2_type(VectorReinterpretNode, VectorNode)                      \
+  declare_c2_type(VectorMaskCastNode, VectorNode)                         \
   declare_c2_type(VectorBoxNode, Node)                                    \
   declare_c2_type(VectorBoxAllocateNode, CallStaticJavaNode)              \
   declare_c2_type(VectorTestNode, Node)                                   \