I only want to analyse some genes different RNAseq analyse ,if i analyse the whole transcriptome genes ,it will cost a long time ,and i found if the gene numbers i analysed are different ,the p value and FDR results are also different.

You can theoretically analysis just a handful of genes, but normalization as well as the empirical Bayes strategies that make edgeR and DESeq2 so powerful work best if you analyse the whole genome. The usual procedure is therefore to analyse everything and then just look at the results of interest at the end.

We find that edgeR works acceptably even on tens of thousands of samples. But, if you find edgeR too slow, then use limma.

If you have a huge number of samples then empirical Bayes becomes far less important. In that case, you could normalize using the whole genome and then run the DE analysis just for your genes of interest.