If you have sequences in both directions for every fragment you plan to use in phylogeny, making consensus before alignment would be better solution. For alignment sequences you should have all sequences in the same directions. If you put all your sequences in latest MEGA version, in most cases times the direction will be automatically recognized and all sampling will organized in one direction - direct or reverse. But this always need to be checked to be sure that all was done properly
More information see here: http://www.megasoftware.net/citations
As already mentioned, when you align all sequences need to be in the same orientation or your results will be useless. If you intend to use any sort of model that incorporates codon information then you must have them in the forward orientation or your results will be faulty.