diff --git a/ARMeilleure/CodeGen/X86/Assembler.cs b/ARMeilleure/CodeGen/X86/Assembler.cs
index ff81530b62..15007dceda 100644
--- a/ARMeilleure/CodeGen/X86/Assembler.cs
+++ b/ARMeilleure/CodeGen/X86/Assembler.cs
@@ -167,7 +167,7 @@ namespace ARMeilleure.CodeGen.X86
             Add(X86Instruction.Pshufd,     new InstInfo(BadOp,      BadOp,      BadOp,      BadOp,      0x00000f70, InstFlags.Vex | InstFlags.Prefix66));
             Add(X86Instruction.Pslldq,     new InstInfo(BadOp,      0x07000f73, BadOp,      BadOp,      BadOp,      InstFlags.Vex | InstFlags.Prefix66));
             Add(X86Instruction.Psllw,      new InstInfo(BadOp,      0x06000f71, BadOp,      BadOp,      0x00000ff1, InstFlags.Vex | InstFlags.Prefix66));
-            Add(X86Instruction.Psrad,      new InstInfo(BadOp,      0x00000f72, BadOp,      BadOp,      0x00000fe2, InstFlags.Vex | InstFlags.Prefix66));
+            Add(X86Instruction.Psrad,      new InstInfo(BadOp,      0x04000f72, BadOp,      BadOp,      0x00000fe2, InstFlags.Vex | InstFlags.Prefix66));
             Add(X86Instruction.Psraw,      new InstInfo(BadOp,      0x04000f71, BadOp,      BadOp,      0x00000fe1, InstFlags.Vex | InstFlags.Prefix66));
             Add(X86Instruction.Psrld,      new InstInfo(BadOp,      0x02000f72, BadOp,      BadOp,      0x00000fd2, InstFlags.Vex | InstFlags.Prefix66));
             Add(X86Instruction.Psrlq,      new InstInfo(BadOp,      0x02000f73, BadOp,      BadOp,      0x00000fd3, InstFlags.Vex | InstFlags.Prefix66));
diff --git a/ARMeilleure/Instructions/InstEmitSimdArithmetic.cs b/ARMeilleure/Instructions/InstEmitSimdArithmetic.cs
index 8045726fb6..2e81695cf6 100644
--- a/ARMeilleure/Instructions/InstEmitSimdArithmetic.cs
+++ b/ARMeilleure/Instructions/InstEmitSimdArithmetic.cs
@@ -3153,7 +3153,7 @@ namespace ARMeilleure.Instructions
 
             Operand onesMask = X86GetAllElements(context, -1L);
 
-            cmpMask = context.AddIntrinsic(Instruction.X86Pand, cmpMask, onesMask);
+            cmpMask = context.AddIntrinsic(Instruction.X86Pandn, cmpMask, onesMask);
 
             Instruction subInst = X86PsubInstruction[size];
 
diff --git a/ARMeilleure/Instructions/InstEmitSimdMove.cs b/ARMeilleure/Instructions/InstEmitSimdMove.cs
index 0cfbbe49a4..f6e633d3d6 100644
--- a/ARMeilleure/Instructions/InstEmitSimdMove.cs
+++ b/ARMeilleure/Instructions/InstEmitSimdMove.cs
@@ -567,13 +567,13 @@ namespace ARMeilleure.Instructions
 
             if (Optimizations.UseSsse3)
             {
-                long maskE0 = _masksE0_TrnUzpXtn[op.Size];
-                long maskE1 = _masksE1_TrnUzp   [op.Size];
-
                 Operand mask = null;
 
                 if (op.Size < 3)
                 {
+                    long maskE0 = _masksE0_TrnUzpXtn[op.Size];
+                    long maskE1 = _masksE1_TrnUzp   [op.Size];
+
                     mask = X86GetScalar(context, maskE0);
 
                     mask = EmitVectorInsert(context, mask, Const(maskE1), 1, 3);
@@ -599,6 +599,11 @@ namespace ARMeilleure.Instructions
 
                 Operand res = context.AddIntrinsic(punpckInst, n, m);
 
+                if (op.RegisterSize == RegisterSize.Simd64)
+                {
+                    res = context.VectorZeroUpper64(res);
+                }
+
                 context.Copy(GetVec(op.Rd), res);
             }
             else