- readOnly ISA allocations must be written through CPU pointer
- command buffer allocation in singleAddressSpaceSbaTracking mode cannot
be readonly - it is written by SBA tracking commands
- this change is fixing ZET_ENABLE_PROGRAM_DEBUGGING mode 2
Related-To: GSD-10359
Signed-off-by: Mateusz Hoppe <mateusz.hoppe@intel.com>
Allow the application to force storageInfo.localOnly and get the
out-of-memory returned if not possible.
This is a windows-only feature supported on discrete platforms.
Related-To: NEO-13428
Signed-off-by: Maciej Bielski <maciej.bielski@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>
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>
Addresses regressions from the reverted merge
of the tbx fault manager for host memory.
Recursive locking of mutex caused deadlock.
To fix, separate tbx fault data from base
cpu fault data, allowing separate mutexes
for each, eliminating recursive locks on
the same mutex.
By separating, we also help ensure that tbx-related
changes don't affect the original cpu fault manager code
paths.
As an added safe guard preventing critical regressions
and avoiding another auto-revert, the tbx fault manager
is hidden behind a new debug flag which is disabled by default.
Related-To: NEO-12268
Signed-off-by: Jack Myers <jack.myers@intel.com>
Addresses regressions from the reverted merge
of the tbx fault manager for host memory.
This fixes attempts by the tbx fault manager
to protect/unprotect host buffer memory, even
if the host ptr was not driver-allocated.
In the case of the smoke test that triggered
the critical regression, clCreateBuffer was
called with the CL_MEM_USE_HOST_PTR flag.
The subsequent `mprotect` calls on the
provided host ptr then failed.
Related-To: NEO-12268
Signed-off-by: Jack Myers <jack.myers@intel.com>
In TBX mode, the host could not write to host buffers after access from device
code due to the lack of a migration mechanism post-initial TBX upload.
Migration is unnecessary with real hardware, but required for TBX.
This patch introduces a new page fault manager type that extends the original
CPU fault manager, enabling automatic migration of host buffers in TBX mode.
Refactoring was necessary to avoid diamond inheritance, achieved by using a
template parameter as the base class for OS-specific fault managers.
Related-To: NEO-12268
Signed-off-by: Jack Myers <jack.myers@intel.com>
This PR handles the situation in which a component
has reserved a front window space for itself in the external heap,
so that the Compute Runtime cannot access this area.
In such a situation, we perform the following steps:
1. reserve 4GB chunk in heapStandard
2. split our chunk into 2 parts: heapFrontWindow, heapRegular
3. from this point on, map all linearStream allocations in reserved 4GB
chunk
Patch applies to Windows and WSL.
Patch only applies when the bindless global allocator is enabled.
Related-To: HSD-16025889919
Signed-off-by: Fabian Zwoliński <fabian.zwolinski@intel.com>
Add tracking of memory used for usm reuse mechanism when multiple cl
contexts are used.
Tracking for device added to NEO::Device, for host added to
NEO::MemoryManager.
This fixes usm reuse using x% of memory per each context instead of
globally.
Related-To: NEO-13308
Signed-off-by: Dominik Dabek <dominik.dabek@intel.com>
- bitset is 64 bit in size, context ids may go beyond that limit
when multiple devices are available
- this change subtracts contextId of first context for a given root
device - tracked state dirty contexts ids are now zero-based
Resolves: GSD-10025
Signed-off-by: Mateusz Hoppe <mateusz.hoppe@intel.com>
Track memory used by memory allocations. System and local per device.
Will be used for heuristics in memory pooling.
Related-To: NEO-11356
Signed-off-by: Dominik Dabek <dominik.dabek@intel.com>
Move the code determining the memoryBanks parameter into separate
virtual function for further memory-manager-specific customization.
Related-To: NEO-9754
Signed-off-by: Maciej Bielski <maciej.bielski@intel.com>
Get the local memory regions count from the primary source (MemoryInfo)
and store for further use when using DrmMemoryManager.
Add a point of dispatch (virtual `createStorageInfoFromProperties`) for
further Drm-specific customizations related to StorageInfo. As the
function became virtual, move one of its callers
(`isLocalMemoryUsedForIsa()`) from the constructor of `MemoryManager` to
respective constructors of all derivative classes. This prevents
bypassing the virtual call dispatch.
Related-To: NEO-9754
Signed-off-by: Maciej Bielski <maciej.bielski@intel.com>
