Avoid using properties().max_shared_memory_per_block() #549
Assignees
Labels
Comments
eyalroz
added a commit
that referenced
this issue
Oct 7, 2023
…code revamp * Validation code is now consolidated in `launch_config.hpp` and the multi-pwrapper implementations file * Now performing more validation at more opportunities before kernel launch * Now ensuring the launch config dimensions don't result in 0-volumne
eyalroz
added a commit
that referenced
this issue
Oct 7, 2023
…code revamp * Validation code is now consolidated in `launch_config.hpp` and the multi-pwrapper implementations file * Now performing more validation at more opportunities before kernel launch * Now ensuring the launch config dimensions don't result in 0-volumne * Now using `get_attribute()` for the device-level maximum shared memory per block, rather than getting all properties
eyalroz
added a commit
that referenced
this issue
Oct 7, 2023
…code revamp * Validation code is now consolidated in `launch_config.hpp` and the multi-pwrapper implementations file * Now performing more validation at more opportunities before kernel launch * Now ensuring the launch config dimensions don't result in 0-volumne * Now using `get_attribute()` for the device-level maximum shared memory per block, rather than getting all properties
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
In our code for validating the shared memory size, we use
.properties().max_shared_memory_per_block()on a device_t object. This is wasteful - let's just get the single attribute. Plus, let's be explicit about it being the opt-in value.The text was updated successfully, but these errors were encountered: