- updates coming from regular list are updated in csr last sent variables
- all per context and per kernel transitions kept in single place
- state updates from intermediate to regular are set in csr properties
- global atomics support duplicates removed
Related-To: NEO-5055
Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
- add correct stateless mocs state update in immediate command lists
- disallow stateless mocs dirty sba command dispatch when sba tracking enabled
- checks support first, only then do the dirty state check in csr
Related-To: NEO-5055
Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
Propogate error codes from ioctl failure properly up the layers
so that we skip exposing bad root devices.
Related-To: NEO-7709
Signed-off-by: Aravind Gopalakrishnan <aravind.gopalakrishnan@intel.com>
When using ReturnSubDevicesAsApiDevices=1 to have
sub-devices-as-root-devices, then the driver should read the values
passed in the mask as those corresponding to the physical
sub-devices.
For instance, in a dual system with multi-tile device, we would have:
card 0, tile 0
card 0, tile 1
card 1, tile 0
card 1, tile 1
With:
ReturnSubDevicesAsApiDevices=0
ZE_AFFINITY_MASK=0,1
Then all tiles in card 0 and card 1 need to be exposed.
With:
ReturnSubDevicesAsApiDevices=1
ZE_AFFINITY_MASK=0,3
Then card 0 tile 0, and card 1 tile 1 need to be exposed.
Related-To: NEO-7137
Signed-off-by: Jaime Arteaga <jaime.a.arteaga.molina@intel.com>
- stateless mocs is present in all state base address commands
- select GPGPU pipeline is present in all pipeline select commands
Related-To: NEO-5055
Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
Related-To: LOCI-3871
- Enabled allocation of specified base address in the targeted heap.
- Enabled virtual memory reservations to grow by allocating at the start
of the heap vs the end of the heap.
Signed-off-by: Spruit, Neil R <neil.r.spruit@intel.com>
multi storage memory migration needs to wait until memory is no longer used
Related-To: NEO-5735
Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
When one process had exported and then opened IPC handle
of memory, then close function was called twice for the
same BO handle. It caused debugBreak() and aborted
an application.
This change allows multiple separate BOs to share one
handle. The last shared handle owner calls close() function.
Related-To: NEO-7200
Signed-off-by: Wrobel, Patryk <patryk.wrobel@intel.com>
-start async thread at device initialization which initializes selected
builtins and exits
-share module across builtins using same binary
Resolves: NEO-7644
Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
If on clCreateProgramWithSource() call we detect a nGen dummy kernel usage,
then enforce fallback to the patchtokens format (only for this kernel).
Extend this logic to all platforms (previously - only for TGL/ICL).
Signed-off-by: Kacper Nowak <kacper.nowak@intel.com>
Allow for use of device binary in ocloc concat. Previously only
AR files could be concatenated.
This feature only works for zebin with AOT Product Config note.
Signed-off-by: Krystian Chmielewski <krystian.chmielewski@intel.com>
This commit adds support for decoding SHT_ZEBIN_GTPIN_INFO type
sections. For each section, passed data will be stored in kernel info
(for corresponding kernel).
Related-To: NEO-7689
Signed-off-by: Kacper Nowak <kacper.nowak@intel.com>
This commit adds support for parsing SHT_NOBITS zebin's ELF sections
(containing global/constant zero-initialized data).
- Correction: in CTNI path, do not add related symbol if surface has not
been allocated.
Related-To: NEO-7196
Signed-off-by: Kacper Nowak <kacper.nowak@intel.com>
-start async thread at device initialization which initializes selected
builtins and exits
-share module across builtins using same binary
Resolves: NEO-7644
Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
Sizing context (PVC):
When using LargeGRF (a.k.a GRF256) there are only 4 HW threads per EU
(instead of default 8). Together with SIMD16 that means that there can
be max 64 work-items per EU. With 8 EU per subslice this gives 512
work-items on a single subslice. For correct intra-WG synchronization
all its WIs must be executed on the same subslice (to access the same
SLM, where the synchronization primitives are stored). Thus, with SIMD16
and LargeGRF the work-group size must not exceed 512 (PVC example).
So far `maxWorkGroupSize` is taken solely from a DeviceInfo structure
both in `ModuleTranslationUnit::processUnpackedBinary()` and
`ModuleImp::initialize()`. This method does not take kernel parameters
(LargeGRF) into account. It allows to submit a kernel using LargeGRF
with SIMD16 with the work-group size set to 1024. That leads to a hang.
Fix the `.maxWorkGroupSize` computation so that it takes the kernel
parameters into consideration.
Add new (for discrete platforms >= XeHP) and adapt existing tests, fix
cosmetics by the way.
Similar check for OCL:
https://github.com/intel/compute-runtime/blob/master/opencl/source/comma
nd_queue/enqueue_kernel.h#L130
Related-To: NEO-7684
Signed-off-by: Maciej Bielski <maciej.bielski@intel.com>
