JIT: Clean up and optimize call arg lowering (#112639)

* Simplify the logic to be based purely off of new ABI info
* Consistently insert bitcast nodes for register file mismatches when
  creating `PUTARG_REG` nodes and for `PUTARG_SPLIT`
* Add a lowering optimization that removes `BITCAST` by changing the
  operand (by changing constants to other constants or the type of
  indirections)
* Stop DNER'ing for register file mismatches in physical promotion and
  morph

Author: jakobbotsch

src/coreclr/jit/compiler.h

@@ -3119,7 +3119,7 @@ class Compiler
     Statement* gtNewStmt(GenTree* expr, const DebugInfo& di);
 
     // For unary opers.
-    GenTree* gtNewOperNode(genTreeOps oper, var_types type, GenTree* op1);
+    GenTreeUnOp* gtNewOperNode(genTreeOps oper, var_types type, GenTree* op1);
 
     // For binary opers.
     GenTreeOp* gtNewOperNode(genTreeOps oper, var_types type, GenTree* op1, GenTree* op2);
@@ -3208,7 +3208,7 @@ class Compiler
 
     GenTree* gtNewPutArgReg(var_types type, GenTree* arg, regNumber argReg);
 
-    GenTree* gtNewBitCastNode(var_types type, GenTree* arg);
+    GenTreeUnOp* gtNewBitCastNode(var_types type, GenTree* arg);
 
 public:
     GenTreeCall* gtNewCallNode(gtCallTypes           callType,

src/coreclr/jit/compiler.hpp

@@ -1411,14 +1411,14 @@ inline Statement* Compiler::gtNewStmt(GenTree* expr, const DebugInfo& di)
     return stmt;
 }
 
-inline GenTree* Compiler::gtNewOperNode(genTreeOps oper, var_types type, GenTree* op1)
+inline GenTreeUnOp* Compiler::gtNewOperNode(genTreeOps oper, var_types type, GenTree* op1)
 {
     assert((GenTree::OperKind(oper) & (GTK_UNOP | GTK_BINOP)) != 0);
     assert((GenTree::OperKind(oper) & GTK_EXOP) == 0); // Can't use this to construct any types that extend unary/binary
                                                        // operator.
     assert(op1 != nullptr || oper == GT_RETFILT || (oper == GT_RETURN && type == TYP_VOID));
 
-    GenTree* node = new (this, oper) GenTreeOp(oper, type, op1, nullptr);
+    GenTreeUnOp* node = new (this, oper) GenTreeOp(oper, type, op1, nullptr);
 
     return node;
 }

src/coreclr/jit/gentree.cpp

@@ -9153,20 +9153,12 @@ GenTree* Compiler::gtNewPutArgReg(var_types type, GenTree* arg, regNumber argReg
 // Notes:
 //    The node is generated as GenTreeMultiRegOp on RyuJIT/arm, as GenTreeOp on all the other archs.
 //
-GenTree* Compiler::gtNewBitCastNode(var_types type, GenTree* arg)
+GenTreeUnOp* Compiler::gtNewBitCastNode(var_types type, GenTree* arg)
 {
     assert(arg != nullptr);
     assert(type != TYP_STRUCT);
 
-    GenTree* node = nullptr;
-#if defined(TARGET_ARM)
-    // A BITCAST could be a MultiRegOp on arm since we could move a double register to two int registers.
-    node = new (this, GT_BITCAST) GenTreeMultiRegOp(GT_BITCAST, type, arg, nullptr);
-#else
-    node = gtNewOperNode(GT_BITCAST, type, arg);
-#endif
-
-    return node;
+    return gtNewOperNode(GT_BITCAST, type, arg);
 }
 
 //------------------------------------------------------------------------