-
Notifications
You must be signed in to change notification settings - Fork 33
Closed
Milestone
Description
Due to mutiple iterations, the the JitKernel class became messy. PR #896
- There are at least two levels of caching for the same unspecialized kernel, i.e.
self._cacheandself._kernel_bundle_cache - For the specialized kernel, the caching is done in three individual levels. i.e.
self._cache,self._kernel_bundle_cacheandself._specialization_cache - These multiple levels of LRU caching is causing unnecessary overhead. Everything in that class needs to be made streamlined and clean. e.g.
self._cacheandself._kernel_bundle_cachecan be consolidated. - Also due to these reasons, the cache-hit/miss counts (
self._cache_hits) became ambiguous. - Either
_check_range()and_check_ndrange()functions need to be moved insideRange,NdRangeclass or moved into thenumba_dpex.core.kernel_interface.utilsmodule.
Metadata
Metadata
Assignees
Labels
No labels