Thank you very much, Gang!
I tried the contrast under the glf model, but somehow the result looks very patchy (i.e., very high intensity voxels are scattered), so it seems something went wrong.
I ran the same code under glt, then it looks reasonable, but not the same as I would see on the F-map. Due to running time with many contrast, I decided to run the two sessions of the data separately (in the end separate 3dMVM for D and P)..
Thanks, SJ