Heehoon Kim

icon indicating an email [email protected]

icon company Seoul National University icon location Seoul, South Korea icon location 노는게 제일 좋아

Results 2 issues of Heehoon Kim

[Warning] Use this repo at your own risk!

Because this was a toy project when I was an undergraduate student, the codes are broken here and there. Please use it for study purposes only.

[Issue]: Why does reducing stack size decrease kernel launch overhead?

4 comment

### Problem Description When we use indirect function call in the kernel function as in below: https://github.com/ROCm/rccl/blob/53dcfcc5e0592fa186da38a41d19082723711043/src/device/common.h#L348 RCCL limits the size of stack to 512: https://github.com/ROCm/rccl/blob/53dcfcc5e0592fa186da38a41d19082723711043/src/init.cc#L1876 which is 1024 by...