Hi MH, I don't come here often, so sorry for the delayed response.
RetroTS is more fancy and handles RVT too, so it likely has the edge in effectiveness, and as Rick mentioned it's a popular option particularly for resting state fMRI by the sounds of it. However, I still use 3dretroicor often because it's easier for me to use in some cases, although I primarily do relatively straightforward task fMRI analysis.
To answer your question, 3dretroicor wants a column of numbers in a text file, i.e. an AFNI-standard "1D file", for respiration and/or cardiac inputs. The 1D physio data set(s) must cover the same time span as the fMRI data set, with a suitably high sampling frequency (assumed to be a constant sampling rate). It's usually fine to pass the respiration data in without any preprocessing. If the pulse data are quite distorted you should clean it up yourself first (detrend, denoise, etc.), because 3dretroicor uses a simplistic constant threshold (see -threshold) to isolate time periods where separate peaks (local maxima) occur. In fact, instead of the raw pulse, I just use the peripheral gating output from our scanner which does a decent job of peak-finding if the pulse signal is good enough--it's essentially a train of spikes where "5" indicates a peak. You can use the -cardphase output to help determine if the pulse peak detection is working for you.
Regards,
Fred