By reading the webpage PTX ISA 8.5 I have two doubts about “setmaxnreg”:
1、Will setmaxnreg implicitly use block-level synchronization?
2、If “setmaxnreg” encounters “syncthreads”, will there be a deadlock when the total number of registers required by “inc” is greater than the total number of registers released by “dec”? Assuming there is such a scenario, warp A is waiting for the release of registers required by “setmaxnreg.inc”, and warp B is waiting for “syncthreads” of warp A. So I want to ask, when using “setmaxnreg”, is there any quantity constraint between “inc” and “dec”?