Update mlir for CreateShuffleVector with unsigned deprecation

[echristo]

Summary

Unless you can think of a better way to do this conversion.

[jpienaar]

Looks good modulo variable names (e.g., postion and mask) and feel free to use [{ ... ]} form instead and or move it inside and have

let llvmBuilder = [{

as LLVM_ExtractValueOp does. If Alex has different suggestion then this can be followed up on.

mlir/include/mlir/Dialect/LLVMIR/LLVMOps.td

let builders = [OpBuilder<
     [{
       build(b, result, container.getType(), container, value, position);
     }]>];
   let parser = [{ return parseInsertValueOp(parser, result); }];
   let printer = [{ printInsertValueOp(p, *this); }];
 }
 def LLVM_ShuffleVectorOp
     : LLVM_OneResultOp<"shufflevector", [NoSideEffect]>,
-      Arguments<(ins LLVM_Type:$v1, LLVM_Type:$v2, ArrayAttr:$mask)>,
-      LLVM_Builder<
-      "$res = builder.CreateShuffleVector($v1, $v2, extractPosition($mask));"> {
+      Arguments<(ins LLVM_Type:$v1, LLVM_Type:$v2, ArrayAttr:$mask)> {
+  string llvmBuilder = [{
+      SmallVector<unsigned, 4> position = extractPosition($mask);
+      SmallVector<int, 4> mask(position.begin(), position.end());
+      $res = builder.CreateShuffleVector($v1, $v2, mask);
+  }];
   let builders = [OpBuilder<
     "Builder *b, OperationState &result, Value v1, Value v2, "
     "ArrayAttr mask, ArrayRef<NamedAttribute> attrs = {}">];
   let verifier = [{
     auto wrappedVectorType1 = v1().getType().cast<LLVM::LLVMType>();
     auto wrappedVectorType2 = v2().getType().cast<LLVM::LLVMType>();
     if (!wrappedVectorType2.getUnderlyingType()->isVectorTy())
       return emitOpError("expected LLVM IR Dialect vector type for operand #2");