Replies: 1 comment 1 reply
-
Hi @zjin-lcf, questions about the SYCL spec itself should be better asked in KhronosGroup/SYCL-Docs repo, where the spec source is published. But anyway, regarding your question: there is an existing mechanism for measuring kernel performance: you can query profiling information through events, but you should create your command queue with profiling enabled. So, if you just want to time your kernels, that should be enough. If you want to precisely time a part of your kernel, then apparently existing API won't be enough for you. Tagging @gmlueck, @Pennycook and @mkinsner for awareness |
Beta Was this translation helpful? Give feedback.
1 reply
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
This example shows how to use the clock function to measure the performance of a kernel accurately.
https://github.com/zchee/cuda-sample/blob/master/0_Simple/clock/clock.cu
Is it useful for SYCL to have a clock() function ?
Thanks
Beta Was this translation helpful? Give feedback.
All reactions