When trimming old allocations in usm reuse start from largest
allocations.
This will reduce memory usage more quickly once max hold time is hit.
Related-To: NEO-6893, NEO-14429
Signed-off-by: Dominik Dabek <dominik.dabek@intel.com>
Real allocation size should be used to properly apply limits and allow
more usm reuse hits.
Related-To: NEO-6893
Signed-off-by: Dominik Dabek <dominik.dabek@intel.com>
Related-To: NEO-14538
It's valid for 3D image to copy 2D region.
Current checks for mip map do not consider that.
This change correctly checks for mip mapped 3D image.
Signed-off-by: Szymon Morek <szymon.morek@intel.com>
Save svmData on putting into reuse, instead of searching each time.
Change UNRECOVERABLE_IF to DEBUG_BREAK_IF.
Related-To: NEO-6893
Signed-off-by: Dominik Dabek <dominik.dabek@intel.com>
Related-To: NEO-14538
If user passes slice pitch which is larger than region
to copy, do not override memory beyond region but within
that slice pitch.
Signed-off-by: Szymon Morek <szymon.morek@intel.com>
Pass hwInfo to isHeaplessModeEnabled and isForceBindlessRequired functions.
Related-To: NEO-14526
Signed-off-by: Filip Hazubski <filip.hazubski@intel.com>
The patch applies to Level Zero.
Only allocations < 2MB will be fetched from the pool.
Allocations are shared and reused within a given device.
Additionally, I added a new debug flag to control the allocator:
EnableTimestampPoolAllocator
Related-To: NEO-12287
Signed-off-by: Fabian Zwoliński <fabian.zwolinski@intel.com>
use memcmp instead of comparing dereferenced pointer
when address is programmed within Walker's inline data the memory location
address is 4B aligned and is not 8B aligned
Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
Related-To: NEO-14056
No need to explicitly wait on Windows KMD during make resident as it has
a while loop that does it nevertheless. The KMD wait affects the API
overhead of zeCommandQueueExecuteCommandLists some platforms (MTL, and ARL).
Signed-off-by: Chandio, Bibrak Qamar <bibrak.qamar.chandio@intel.com>
Lock on csr is needed before lock on residency controller to prevent
incorrect lock order. Csr may be locked in waitOnCpu called from trimToBudget,
which may lead to deadlocks
Signed-off-by: Maciej Plewka <maciej.plewka@intel.com>
instead of printf use makro that make flush after printf
Related-To: HSD-14024170600
Signed-off-by: Katarzyna Cencelewska <katarzyna.cencelewska@intel.com>
Host usm and device usm for igfx checks system memory usage.
Device usm for dgfx checks local memory usage.
If used memory is above limit threshold:
- no new allocations will be saved for reuse
- cleaner will use shorter hold time of 2 seconds
- cleaner will free all eligible allocations, regardless of async
deleter thread having work
Motivation: in case of gfx memory being full, making resident new
allocations will require evictions which leads to massive slowdown on
enqueue calls.
This change aims to minimize cases where extra memory usage from usm
reuse mechanism leads to above situation.
Related-To: NEO-6893, NEO-14160
Signed-off-by: Dominik Dabek <dominik.dabek@intel.com>
