[WIP] Second pass at LaunchConfig #727
Conversation
|
Auto-sync is disabled for draft pull requests in this repository. Workflows must be run manually. Contributors can view more details about this message here. |
cpcloud
left a comment
cpcloud
left a comment
There was a problem hiding this comment.
Thanks for the PR!
This really needs some more explicit motivation in the PR description, as well as some real justification for all the duplicated tooling.
There was a problem hiding this comment.
Not entirely sure what the purpose of this file is beyond what's happening in the existing test_kernel_launch.py benchmarks.
There was a problem hiding this comment.
I think all of this can be done with the existing pytest-benchmark plugin.
There was a problem hiding this comment.
I think all or most of this functionality can be done with the existing pytest-benchmark plugin.
Really would like to avoid duplicating functionality, especially if it's AI generated duplication.
| - `bench-launch-overhead` | ||
| - `bench` | ||
| - `benchcmp` | ||
| - `bench-against` |
There was a problem hiding this comment.
While this script doesn't do a three way comparison, it also doesn't require writing any new code to run it.
Can we try to reuse bench-against instead of reinventing a lot of what that already does?
| def some_kernel_1(): | ||
| return | ||
|
|
||
| @cuda.jit("void(float32[:])") | ||
| def some_kernel_2(arr1): | ||
| return | ||
|
|
||
| @cuda.jit("void(float32[:],float32[:])") | ||
| def some_kernel_3(arr1, arr2): | ||
| return | ||
|
|
||
| @cuda.jit("void(float32[:],float32[:],float32[:])") | ||
| def some_kernel_4(arr1, arr2, arr3): | ||
| return | ||
|
|
||
| @cuda.jit("void(float32[:],float32[:],float32[:],float32[:])") | ||
| def some_kernel_5(arr1, arr2, arr3, arr4): |
There was a problem hiding this comment.
These are nearly identical to the existing benchmarks. Let's avoid repeating existing benchmarks and tools that run them.
2 participants
No description provided.