I run a sample program, set one breakpoint only; cuda-gdb will show twice that there is a certain warp has hit the breakpoint;
When I was first prompted that a certain warp hit the breakpoint, I used the Info cuda warps command to observe the pc, which is the current position.
See that all sm/warps show the same location.
My question is, if they are all at the same location, after the continue command, there will be a new warp hitting the breakpoint, which is obviously a conflict.
Hi @204540478,
I run a sample program, set one breakpoint only; cuda-gdb will show twice that there is a certain warp has hit the breakpoint;
In general, it’s expected. There is a detailed explanation in No matter how many warps there are, only one warp hits the breakpoint, right? - #9 by AKravets but it can be summarized as follows:
- All warps (on all SMs) will hit the breakpoint (unless it’s deleted).
- If multiple warps are hitting it at the same time, only one hit will be reported (see my previous answer)
- If multiple warps (e.g. on different SMs) are hitting it not at the same time (what seems to be the case here), the same breakpoint might be reported multiple times (for different warps/SMs).
Now I fully understand. Thank you so much.
