8336464: C2: Force CastX2P to be a two-address instruction

src/hotspot/cpu/aarch64/aarch64.ad

@@ -2321,6 +2321,10 @@ OptoRegPair Matcher::vector_return_value(uint ideal_reg) {
   return OptoRegPair(0, 0);
 }
 
+bool Matcher::use_same_src_and_dest_reg_for_CastX2P(void) {
+  return true;
+}
+
 // Is this branch offset short enough that a short branch can be used?
 //
 // NOTE: If the platform does not provide any short branch variants, then
@@ -7906,31 +7910,31 @@ instruct membar_volatile() %{
 instruct castX2P(iRegPNoSp dst, iRegL src) %{
   match(Set dst (CastX2P src));
 
+  // We still assign cost here because two-address instructions may bring
+  // side impact, i.e., spilling.
   ins_cost(INSN_COST);
-  format %{ "mov $dst, $src\t# long -> ptr" %}
-
+  format %{ "castX2P $src\t# long -> ptr" %}
+  size(0);
   ins_encode %{
-    if ($dst$$reg != $src$$reg) {
-      __ mov(as_Register($dst$$reg), as_Register($src$$reg));
-    }
+    /* empty encoding */
+    assert($dst$$reg == $src$$reg, "We can't remove it when dst != src");
   %}
-
-  ins_pipe(ialu_reg);
+  ins_pipe(pipe_class_empty);
 %}
 
 instruct castP2X(iRegLNoSp dst, iRegP src) %{
   match(Set dst (CastP2X src));
 
+  // We still assign cost here because two-address instructions may bring
+  // side impact, i.e., spilling.
   ins_cost(INSN_COST);
-  format %{ "mov $dst, $src\t# ptr -> long" %}
-
+  format %{ "castP2X $src\t# ptr -> long" %}
+  size(0);
   ins_encode %{
-    if ($dst$$reg != $src$$reg) {
-      __ mov(as_Register($dst$$reg), as_Register($src$$reg));
-    }
+    /* empty encoding */
+    assert($dst$$reg == $src$$reg, "We can't remove it when dst != src");
   %}
-
-  ins_pipe(ialu_reg);
+  ins_pipe(pipe_class_empty);
 %}
 
 // Convert oop into int for vectors alignment masking

src/hotspot/cpu/arm/arm.ad

@@ -1017,6 +1017,10 @@ OptoRegPair Matcher::vector_return_value(uint ideal_reg) {
   return OptoRegPair(0, 0);
 }
 
+bool Matcher::use_same_src_and_dest_reg_for_CastX2P(void) {
+  return false;
+}
+
 // Vector width in bytes
 int Matcher::vector_width_in_bytes(BasicType bt) {
   return MaxVectorSize;

src/hotspot/cpu/ppc/ppc.ad

@@ -2164,6 +2164,10 @@ OptoRegPair Matcher::vector_return_value(uint ideal_reg) {
   return OptoRegPair(0, 0);
 }
 
+bool Matcher::use_same_src_and_dest_reg_for_CastX2P(void) {
+  return false;
+}
+
 // Vector width in bytes.
 int Matcher::vector_width_in_bytes(BasicType bt) {
   if (SuperwordUseVSX) {

src/hotspot/cpu/riscv/riscv.ad

@@ -1958,6 +1958,10 @@ OptoRegPair Matcher::vector_return_value(uint ideal_reg) {
   return OptoRegPair(0, 0);
 }
 
+bool Matcher::use_same_src_and_dest_reg_for_CastX2P(void) {
+  return false;
+}
+
 // Is this branch offset short enough that a short branch can be used?
 //
 // NOTE: If the platform does not provide any short branch variants, then

src/hotspot/cpu/s390/s390.ad

@@ -1489,6 +1489,10 @@ OptoRegPair Matcher::vector_return_value(uint ideal_reg) {
   return OptoRegPair(0, 0);
 }
 
+bool Matcher::use_same_src_and_dest_reg_for_CastX2P(void) {
+  return false;
+}
+
 //----------SUPERWORD HELPERS----------------------------------------
 
 // Vector width in bytes.

src/hotspot/cpu/x86/x86_32.ad

@@ -1398,6 +1398,10 @@ OptoRegPair Matcher::vector_return_value(uint ideal_reg) {
   return OptoRegPair(0, 0);
 }
 
+bool Matcher::use_same_src_and_dest_reg_for_CastX2P(void) {
+  return false;
+}
+
 // Is this branch offset short enough that a short branch can be used?
 //
 // NOTE: If the platform does not provide any short branch variants, then

src/hotspot/cpu/x86/x86_64.ad

@@ -1597,6 +1597,10 @@ OptoRegPair Matcher::vector_return_value(uint ideal_reg) {
   return OptoRegPair(hi, lo);
 }
 
+bool Matcher::use_same_src_and_dest_reg_for_CastX2P(void) {
+  return false;
+}
+
 // Is this branch offset short enough that a short branch can be used?
 //
 // NOTE: If the platform does not provide any short branch variants, then

src/hotspot/share/adlc/output_h.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1998, 2023, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1998, 2024, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -1688,6 +1688,19 @@ void ArchDesc::declareClasses(FILE *fp) {
     // Identify which input register matches the input register.
     uint  matching_input = instr->two_address(_globalNames);
 
+    // Allocate the same src and dest register for CastX2P and CastP2X
+    // on some target platforms, then we can remove the instructions in
+    // the final assembly.
+    if (strcmp("CastX2P", instr->ideal_Opcode(_globalNames)) == 0 ||
+        strcmp("CastP2X", instr->ideal_Opcode(_globalNames)) == 0) {
+      assert(matching_input == 0, "");
+      fprintf(fp,"  virtual uint           two_adr() const  {\n");
+      fprintf(fp,"    if (Matcher::use_same_src_and_dest_reg_for_CastX2P())\n");
+      fprintf(fp,"      return oper_input_base();\n");
+      fprintf(fp,"    return 0;\n");
+      fprintf(fp,"  }\n");
+    }
+
     // Generate the method if it returns != 0 otherwise use MachNode::two_adr()
     if( matching_input != 0 ) {
       fprintf(fp,"  virtual uint           two_adr() const  ");

src/hotspot/share/opto/matcher.hpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 1997, 2023, Oracle and/or its affiliates. All rights reserved.
+ * Copyright (c) 1997, 2024, Oracle and/or its affiliates. All rights reserved.
  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  *
  * This code is free software; you can redistribute it and/or modify it
@@ -456,6 +456,10 @@ class Matcher : public PhaseTransform {
   static bool supports_vector_calling_convention();
   static OptoRegPair vector_return_value(uint ideal_reg);
 
+  // Is it preferred to use the same src and dest register for CastX2P and
+  // CastP2X? If true, we can remove the instructions in the final assembly.
+  static bool use_same_src_and_dest_reg_for_CastX2P();
+
   // Is this branch offset small enough to be addressed by a short branch?
   bool is_short_branch_offset(int rule, int br_size, int offset);