I got a  list of paired end bam files, and want to do DESeq2 analysis. The first thing is to count the reads at gene level. I tried feature counts. However, it requires reorder of the bam files if I specified paired end is True, which takes infinite time for my huge dataset. I also tried summarizeOverlaps, which also takes very long time. 

Is there any tools could quickly count the paired end reads for DESeq2 at gene level? Thanks.

If you're looking for speed, you might try using kallisto or salmon to quantitate expression from your data, then use the tximport package to roll the transcript-level estimates up into gene level estimates so you can feed those numbers into DESeq2 (or edgeR (or voom)).

Note that, technically, you will have to be running the development version of Bioconductor to get tximport ...