Join thread in DllMain (which we are not aware of)
can result in hang occurring in DeferredDeleter, if the library is
freed before FreeLibraryAndExitThread call from within the worker thread,
the thread gets stuck, thus the main thread is stuck on worker->join().
Related-To: NEO-14121
Signed-off-by: Oskar Hubert Weber <oskar.hubert.weber@intel.com>
- add `hasImageWriteArg` flag to KernelDescriptor,
based on image access qualifier
- add `isPostImageWriteFlushRequired` method to productHelper,
to require cache flush only on selected platforms
- if possible, add the `textureCacheInvalidationEnable` arg to the
existing PC, otherwise add a new PC
Related-To: NEO-13427, HSD-18041321008
Signed-off-by: Fabian Zwoliński <fabian.zwolinski@intel.com>
Join thread in DllMain (which we are not aware of)
can result in hang occurring in DeferredDeleter, if the library is
freed before FreeLibraryAndExitThread call from within the worker thread,
the thread gets stuck, thus the main thread is stuck on worker->join().
Related-To: NEO-14121
Signed-off-by: Oskar Hubert Weber <oskar.hubert.weber@intel.com>
Related-To: NEO-13715
Implements basic functionality for zeCommandListAppendMemoryAdvise for
system allocator
Signed-off-by: Chandio, Bibrak Qamar <bibrak.qamar.chandio@intel.com>
Max programmable slm size is stored in RuntimeCapabilityTable as slmSize.
That is misleading name
Rename the variable slmSize to better reflect the actual meaning of the member.
Related-To: NEO-12949
Signed-off-by: Marcel Skierkowski <marcel.skierkowski@intel.com>
Return early and provide justification, otherwise SIGBUS error is
returned.
Related-To: NEO-14506
Signed-off-by: Maciej Bielski <maciej.bielski@intel.com>
Align local memory allocations of tag types to 2MB when
2MB alignment is enabled via the product helper
(is2MBLocalMemAlignmentEnabled flag).
Refactored the allocateGraphicsMemoryInDevicePool function to improve
readability and maintainability. Simplified the logic for
determining base size and final alignment by reducing redundant code.
Related-To: NEO-12287
Signed-off-by: Fabian Zwoliński <fabian.zwolinski@intel.com>
If some operatioins requires ULLS light stop, execute such operations
under mutex in pair with ULLS stop to ensure no other thread will start
ULLS.
Related-To: NEO-14406, NEO-13922
Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@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>
-skip bo::wait if completion fence not needed
-do not restart exec on evict
Related-To: NEO-13922
Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
For the Xe driver, a single tile can have multiple gts. Now the Engine
handles are created for multiple gts available on the single tile as in
the case for the Xe driver.
Related-To: NEO-14401
Signed-off-by: Pratik Bari <pratik.bari@intel.com>
