-
-
Notifications
You must be signed in to change notification settings - Fork 79
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Make "cooperativity" be part of the launch configuration #258
Comments
As far as I understand your point you want to discuss how much 'kernel launch configuration' should be saved in class I agree with your point of having less functionality in class |
No... I'm wondering whether to add |
To me this sounds reasonable. |
@codecircuit : Reasonable, and better than what we have currently? |
…on. Several `launch()` functions and methods removed.
@codecircuit : Done. |
Yes, the commit looks good 👍. |
@codecircuit : I've deleted 63 more lines than I've added, so it has to be good :-P Reminding you to possibly have a look at the driver-wrappers branch. |
…on. Several `launch()` functions and methods removed.
…on. Several `launch()` functions and methods removed.
…on. Several `launch()` functions and methods removed.
…on. Several `launch()` functions and methods removed.
…on. Several `launch()` functions and methods removed.
At the moment, the user must specify, when launching a kernel, whether thread blocks should be allowed to cooperate or not. This is a bit cumbersome, and also increases the number of kernel launch functions and methods, which is already not so small.
On the other hand, our wrapped kernel class does indicate whether thread blocks should be allowed to cooperate, and putting that flag into a launch config may remove the ability to hard-code that requirement.
Yet again on the first hand - why should we not have the ability to hard-code requirements for other aspects of the launch, into a
kernel_t
? Dynamic shared memory size requirements, and even block and grid size requirements? A kernel can certainly have those.I would lean in favor of having less functionality in
kernel_t
and making the move, since this library is just for wrapping, not for beefier abstractions.Why do you think @codecircuit , @raplonu, @quxflux et al. ?
The text was updated successfully, but these errors were encountered: