How to profile the number of global memory transactions for cuda kernels?

妖精的绣舞 提交于 2019-12-13 01:28:17

问题


How to enable profiling for "uncached_global_load_transaction" counter in cuda command-line profiler?


回答1:


The command line profiler is controlled using the following environment variables -

COMPUTE_PROFILE: is set to either 1 or 0 (or unset) to enable or disable profiling.
COMPUTE_PROFILE_CONFIG: is used to specify a config file for enabling performance counters in the GPU and various other options.
COMPUTE_PROFILE_LOG: is set to the desired file path for profiling output.

In your case you can set above environment variables as -

COMPUTE_PROFILE=1
COMPUTE_PROFILE_CONFIG=config.txt
COMPUTE_PROFILE_LOG=profiler_output.txt

config.txt must contain an entry uncached_global_load_transaction.



来源:https://stackoverflow.com/questions/9826948/how-to-profile-the-number-of-global-memory-transactions-for-cuda-kernels

标签
易学教程内所有资源均来自网络或用户发布的内容,如有违反法律规定的内容欢迎反馈
该文章没有解决你所遇到的问题?点击提问,说说你的问题,让更多的人一起探讨吧!