How dose TVM elimitate calls of conv weights layout_transform?

Yes weight layout transformation should be optimized by constant folding.