Excellent news.
OK, open in serial then read in the data & save both with a new name. Partition to 32 nodes manually and resave both (new name again). Re-try in parallel.
I've seen it before and the above usually fixes it: it's actually possible sometimes to read into serial, save and then read into parallel but the extra step eliminates other problems. Using a different name is an additional safety step: I don't trust computers not to eat my work!