I looks to me that I have pretty much the same problem as this person. Basically I added thrust to my project to sort data on the GPU. Sorting now works fine but the compile time is like 100 times longer… Since there was no answer to the post above, I repost it here.
#pragma unroll takes much time when there are many loop cycles to unroll. Maybe it is used somewhere.
Ok it does not take that long… but I can’t it under 4 minutes (when recompiling). So I wonder what makes it take so much time? Would it help to put the thrust files on a ram disk?
You may put kernel to separate file, if it’s not nesseccary to recompile it each time.
BTW I tried to compile example - it compiles fast - less than a minute.
Are those methods templated for different data types? I have seen templated code take forever to compile before. Just throwing a random idea out there…