[SYCL] Support for pooled small allocations. #5438
Conversation
…eters. Signed-off-by: Rajiv Deodhar <rajiv.deodhar@intel.com>
rdeodhar
changed the title
|
Hi Rajiv, Another nitpick. I was wondering why definition of class Bucket is not in the header file. Thanks |
The entire implementation of pooling is in the .cpp file. The usm_allocator.hpp header is only the external interface of pooling from the pi_level_zero files. |
|
Changes look good. Can you please comment on adding a test for this? Thanks |
|
The existing test usm_pooling.cpp will be extended to account for the feature added in this PR. |
|
/verify with intel/llvm-test-suite#980 |
|
/verify with intel/llvm-test-suite#980 |
|
/verify with intel/llvm-test-suite#980 |
|
/verify with intel/llvm-test-suite#980 |
|
/verify with intel/llvm-test-suite#980 |
|
/verify with intel/llvm-test-suite#980 |
|
/verify with intel/llvm-test-suite#980 |
|
/verify with intel/llvm-test-suite#980 |
| if (chunkedBucket) | ||
| NewFreeSlabsInBucket = chunkedSlabsInPool + 1; | ||
| else | ||
| NewFreeSlabsInBucket = AvailableSlabs.size() + 1; |
There was a problem hiding this comment.
Would it be simpler to maintain "chunked" slabs in a separate list, i.e. break the AvailableSlabs into 2?
There was a problem hiding this comment.
I don't know if it would be simpler, but there could be alternative approaches such as adding a third type of list, in effect breaking the AvailableSlabs list into two, with different meaning.
I also tried layering the slab types, to make chunked slabs a specialization of full slabs.
However, the approach implemented here required the least change.
There was a problem hiding this comment.
OK, let's go with what we have ready, and only consider a re-design when we need to changes something again.
|
/verify with intel/llvm-test-suite#980 |
This change enables pooling of USM slabs used for small allocations, i.e. < 32KB. These slabs are used in chunked mode, that is, divided into chunks, and each chunk used for an individual user allocation.
Previously, when all the individual allocations in a chunked slab were freed by the program, the slab was freed from USM. We no longer do that. We retain that slab for future use, in effect, pooling it.
Signed-off-by: Rajiv Deodhar rajiv.deodhar@intel.com