Adding support for the SSE Set scalar intrinsic

Author: tannergooding

src/jit/hwintrinsiccodegenxarch.cpp

@@ -602,6 +602,21 @@ void CodeGen::genSSEIntrinsic(GenTreeHWIntrinsic* node)
             emit->emitIns_SIMD_R_R_R_I(INS_shufps, targetReg, op1Reg, op1Reg, 0, TYP_SIMD16);
             break;
 
+        case NI_SSE_SetScalar:
+            assert(baseType == TYP_FLOAT);
+            assert(op2 == nullptr);
+
+            if (op1Reg == targetReg)
+            {
+                regNumber tmpReg = node->GetSingleTempReg();
+                emit->emitIns_SIMD_R_R(INS_movaps, tmpReg, op1Reg, TYP_SIMD16);
+                op1Reg = tmpReg;
+            }
+
+            emit->emitIns_SIMD_R_R_R(INS_xorps, targetReg, targetReg, targetReg, TYP_SIMD16);
+            emit->emitIns_SIMD_R_R_R(INS_movss, targetReg, targetReg, op1Reg, TYP_SIMD16);
+            break;
+
         case NI_SSE_SetZeroVector128:
             assert(baseType == TYP_FLOAT);
             assert(op1 == nullptr);

src/jit/hwintrinsicxarch.cpp

@@ -635,6 +635,7 @@ GenTree* Compiler::impSSEIntrinsic(NamedIntrinsic        intrinsic,
         }
 
         case NI_SSE_SetAllVector128:
+        case NI_SSE_SetScalar:
             assert(sig->numArgs == 1);
             assert(getBaseTypeOfSIMDType(sig->retTypeSigClass) == TYP_FLOAT);
             op1     = impPopStack().val;

src/jit/lsraxarch.cpp

@@ -2550,6 +2550,12 @@ void LinearScan::TreeNodeInfoInitHWIntrinsic(GenTreeHWIntrinsic* intrinsicTree,
             info->setInternalCandidates(this, allRegs(TYP_INT));
             break;
 
+        case NI_SSE_SetScalar:
+            // Need an internal register to stitch together all the values into a single vector in a SIMD reg.
+            info->internalFloatCount = 1;
+            info->setInternalCandidates(this, allSIMDRegs());
+            break;
+
         case NI_SSE_Shuffle:
         {
             assert(op1->OperIsList());