[Runtime] Allow query of available device memory through DeviceAPI #16994
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Prior to this commit, the total device memory could be queried through the `DeviceAPI` interface, but the currently available device memory could not. This functionality may be useful for debugging, or for validating available memory prior to model execution. This commit implements the property `Device.available_global_memory`, which queries the `DeviceAttrKind::kAvailableGlobalMemory`. Support for this query, like all device attribute queries, may vary across different backends, and will return `None` for backends that do not support this query. This commit only currently implements support for `kAvailableGlobalMemory` for TVM's Cuda backend.
masahi
approved these changes
May 13, 2024
python/tvm/_ffi/runtime_ctypes.py
Outdated
| Return None if the device does not support this feature. | ||
| """ | ||
| return self._GetDeviceAttr(self.device_type, self.device_id, 14) | ||
|
|
||
| @property | ||
| def available_global_memory(self): | ||
| """Return size of the total global memory. |
There was a problem hiding this comment.
Thank you, and fixed!
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Prior to this commit, the total device memory could be queried through the
DeviceAPIinterface, but the currently available device memory could not. This functionality may be useful for debugging, or for validating available memory prior to model execution.This commit implements the property
Device.available_global_memory, which queries theDeviceAttrKind::kAvailableGlobalMemory. Support for this query, like all device attribute queries, may vary across different backends, and will returnNonefor backends that do not support this query. This commit only currently implements support forkAvailableGlobalMemoryfor TVM's Cuda backend.