Thread arbitration policy can be programmed in the command list
instead of doing it in executeCommandList(), so move it there.
Related-to: NEO-5187
Change-Id: I518e19e939e0548d9d454738c6deb5e095154422
Signed-off-by: Jaime Arteaga <jaime.a.arteaga.molina@intel.com>
change TdDebugControl Register to enum for over-riding in another
platforms
Related-To: NEO-5203
Change-Id: I17b835a1d4cb43986cfd06741e6ec9a606feb5ae
Signed-off-by: Pranav Tendulkar <pranav.tendulkar@intel.com>
- reuse vector instead of creating one per every function call.
Change-Id: I584bfbd7757e7b8851ae4970b740adf20659bcb1
Signed-off-by: Michal Mrozek <michal.mrozek@intel.com>
- Fixed the 2d/3d copy to correctly pass the aligned allocation data to
the kernel programming of the buffer surface states otherwise unaligned
host ptrs cannot be used.
Change-Id: If62e9321d572a786999a5ead8303f11e2e9a3c8d
Signed-off-by: Spruit, Neil R <neil.r.spruit@intel.com>
This patch does the following:
- Fixes a bug in FsAccess::listDirectory that could return
ZE_RESULT_UNKNOWN_ERROR when no error has occurred.
- Fixes a bug in zesDeviceReset that would reset the device
if force was set to false, even if the device was in use.
- Fixes a bug in zesDeviceReset that would reset the device
if force was set to false without closing the file descriptor.
- Added a releaseResources method method to Device object.
This method does the same thing as the DeviceImp
destructor except it does not free the DeviceImp object
and it does not free the SysmanDeviceImp object.
- Added the releaseResources methods to Mock<Device> object.
- Moved the reset of the debugger out of DriverHandleImp
destructor and into DeviceImp releaseResources.
- Added a releaseEngine method to the EngineHandleContext. This
method frees all the Engine handles.
- On reset, I call the Devcie->releaseResources and
EngineHandleContext->releaseEngines before resetting the device.
- Added a -r (--reset) option to zello_sysman so I
could easily test resets.
With these patches, the L0 Sysman CTS for zesDeviceReset
both pass.
Change-Id: I31fad1b27bc5cc6befe31cd6f9319748e2683424
When performing copy operations to or from buffers allocated by the
user, it could happen that the buffer address is not accepted by
kernel, even though the buffer is valid. In those ocassions, then
allocate a new graphics allocation and copy the user buffer.
Change-Id: I6b1b6f2ef5fea0acf32c868bc87eafe8746f9a79
Signed-off: Jaime Arteaga <jaime.a.arteaga.molina@intel.com>
This resolved regressions on Linux/ Windows for SYCL/L0 applications.
Change-Id: I26bd75a0473a5ceace1321d6b3f9613ce2cc5d9d
Signed-off-by: Vinod Tipparaju <vinod.tipparaju@intel.com>
- Given a nullptr Arg Buffer value, the argument value in cross thread
data must be updated to be a nullptr value especially given the argument
could have been set previously to a buffer that may not exist or should
not be used during this call to the kernel
Change-Id: Ie3a03ce25a4bb0b5e186863cc912a55863d0fb69
Signed-off-by: Spruit, Neil R <neil.r.spruit@intel.com>
This resolves corruptions by flushing ID cache on HW
Change-Id: I56919360ac126cec67c5cbcf6380acfd72e0bcdb
Signed-off-by: Vinod Tipparaju <vinod.tipparaju@intel.com>
Doing so causes hangs when reusing a copy command list
Change-Id: I1bd069942e9caee01636497f992dd35cbedcde1b
Signed-off-by: Aravind Gopalakrishnan <Aravind.Gopalakrishnan@intel.com>
