I have two questions regarding the setup of the activity buffer.
- What is the relationship between the user-allocated activity buffer size and the CUPTI-maintained device buffer size
CUPTI_ACTIVITY_ATTR_DEVICE_BUFFER_SIZE
? Are they completely independent of each other, or the former should be the same with the latter? - Is it recommended to make the user-allocated activity buffer pinned? I would imagine that CUPTI copies the results from the device buffer to the host activity buffer, and making the activity buffer pinned gives bandwidth advantage.
Thanks!
PS: CUDA 12.4 on Ubuntu 22.04.