SimdAsHWIntrinsic improvements and cleanup (#80134)

* Merge common code paths in simdashwintrinsic.cpp

* Fixing a bug in the WithElement node for CreateFromVector

* Ensure simdashwintrinsic paths consistently use the gtNewSimd*Node helpers where one path was already

* Ensure remaining simdashwintrinsic paths use the gtNewSimd*Node helpers

* Consistently use fgMakeMultiUse in the gtNewSimd*Node APIs

* Applying formatting patch

* Ensure isSimdAsHWIntrinsic gets passed through

* Revert "Consistently use fgMakeMultiUse in the gtNewSimd*Node APIs"

This reverts commit edb2ce8.

* Handle SN_op_UnaryNegation for Vector2/3/4 on Mono

* Ensure short, ushort, float, and double don't assert for Vector128_Dot on downlevel hardware

Author: tannergooding

src/coreclr/jit/gentree.cpp

@@ -20566,8 +20566,7 @@ GenTree* Compiler::gtNewSimdCmpOpAllNode(genTreeOps  op,
                 intrinsic = NI_Vector128_op_Equality;
             }
 
-            op1 = gtNewSimdCmpOpNode(op, simdType, op1, op2, simdBaseJitType, simdSize,
-                                     /* isSimdAsHWIntrinsic */ false);
+            op1 = gtNewSimdCmpOpNode(op, simdType, op1, op2, simdBaseJitType, simdSize, isSimdAsHWIntrinsic);
             op2 = gtNewAllBitsSetConNode(simdType);
 
             if (simdBaseType == TYP_FLOAT)
@@ -20606,8 +20605,7 @@ GenTree* Compiler::gtNewSimdCmpOpAllNode(genTreeOps  op,
                 intrinsic = NI_Vector128_op_Equality;
             }
 
-            op1 = gtNewSimdCmpOpNode(op, simdType, op1, op2, simdBaseJitType, simdSize,
-                                     /* isSimdAsHWIntrinsic */ false);
+            op1 = gtNewSimdCmpOpNode(op, simdType, op1, op2, simdBaseJitType, simdSize, isSimdAsHWIntrinsic);
             op2 = gtNewAllBitsSetConNode(simdType);
 
             if (simdBaseType == TYP_FLOAT)
@@ -20688,8 +20686,7 @@ GenTree* Compiler::gtNewSimdCmpOpAnyNode(genTreeOps  op,
                 intrinsic = NI_Vector128_op_Inequality;
             }
 
-            op1 = gtNewSimdCmpOpNode(op, simdType, op1, op2, simdBaseJitType, simdSize,
-                                     /* isSimdAsHWIntrinsic */ false);
+            op1 = gtNewSimdCmpOpNode(op, simdType, op1, op2, simdBaseJitType, simdSize, isSimdAsHWIntrinsic);
             op2 = gtNewZeroConNode(simdType);
 
             if (simdBaseType == TYP_FLOAT)
@@ -20732,8 +20729,7 @@ GenTree* Compiler::gtNewSimdCmpOpAnyNode(genTreeOps  op,
 
             intrinsic = (simdSize == 8) ? NI_Vector64_op_Inequality : NI_Vector128_op_Inequality;
 
-            op1 = gtNewSimdCmpOpNode(op, simdType, op1, op2, simdBaseJitType, simdSize,
-                                     /* isSimdAsHWIntrinsic */ false);
+            op1 = gtNewSimdCmpOpNode(op, simdType, op1, op2, simdBaseJitType, simdSize, isSimdAsHWIntrinsic);
             op2 = gtNewZeroConNode(simdType);
 
             if (simdBaseType == TYP_FLOAT)

src/coreclr/jit/hwintrinsicxarch.cpp

@@ -609,6 +609,9 @@ GenTree* Compiler::impBaseIntrinsic(NamedIntrinsic        intrinsic,
         {
             assert(sig->numArgs == 2);
 
+            impSpillSideEffect(true,
+                               verCurrentState.esStackDepth - 2 DEBUGARG("Spilling op1 side effects for HWIntrinsic"));
+
             op2 = impSIMDPopStack(retType);
             op1 = impSIMDPopStack(retType);