Hi @Ling!
You'll want to use qiime demux emp-paired.
Which primers were used doesn't matter to this step, what's more important is that your data has the non-biological data already stripped. In other words, the linker, primer, barcode, etc, are not in R1 or R2 (barcodes should be in R3). It will use the order of R3 to work out which sequences from R1 and R2 belongs to which samples.
qiime cutadapt demux-paired is for when your barcodes are still in the sequences and need to both be identified and trimmed out. This shouldn't be the case with an EMP-like protocol.