Related-To: NEO-7361
Currently additional command buffer is
preallocated for all CSRs, even for those which
won't be used by application. This PR changes that
Signed-off-by: Szymon Morek <szymon.morek@intel.com>
This change makes that drm file is opened in nonblocking mode for prelim
kernels. In such case when calling exec buffer ioctl and get
EAGAIN (aka EWOULDBLOCK) we may return error to API level
Related-To: NEO-7144
Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
This change makes that drm file is opened in nonblocking mode for prelim
kernels. In such case when calling exec buffer ioctl and get
EAGAIN (aka EWOULDBLOCK) we may return error to API level
Related-To: NEO-7144
Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
Move Batching code under batching if to not call not required functions.
Update task level only if level is closed.
70ns gain.
Signed-off-by: Michal Mrozek <michal.mrozek@intel.com>
This change is intended to be used in immediate command lists that are
using flush task functionality.
With this change all immediate command list using the same csr will consume
shared allocations for dsh and ssh heaps. This will decrease number of SBA
commands dispatched when multiple command lists coexists and dispatch kernels.
With this change new SBA command should be dispatched only when current heap
allocation is exhausted.
Functionality is currently disabled and available under debug key.
Functionality will be enabled by default for all immediate command lists
with flush task functionality enabled.
Related-To: NEO-7142
Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
Linear stream is always being updated under mutex.
Massive improvement for whole zeCommandListAppendLaunchKernel +20%.
Signed-off-by: Michal Mrozek <michal.mrozek@intel.com>
This optimization removes pipeline select from command list preamble
and presented to command queue for necessary state update.
Code is disabled by default and available under debug key.
Related-To: NEO-5019
Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
We want to return error code to the application instead of aborting when
we are not able to make more memory resident.
Related-To: NEO-7289
Signed-off-by: Fabian Zwolinski <fabian.zwolinski@intel.com>
In case of debuggable context device should be additionally
initialized by early empty submission issue.
Signed-off-by: Igor Venevtsev <igor.venevtsev@intel.com>
Another step towards cleaner callers of
StateBaseAddressHelper<>::programStateBaseAddress.
Export programming state base address into a separate function to
improve code reuse and reduce copy-pasted fragments, which make code
modifications or maintenance more and more difficult over time. Use
specialization for gen-specific variations.
Related-To: NEO-6774
Signed-off-by: Maciej Bielski <maciej.bielski@intel.com>
This change prepares infrastructure for pipeline select handling in
command lists and queues by optimization of number of commands dispatched.
State is synchronized between flush-task immediate and regular command lists.
Next step is to add optimization itself which disables legacy hw command
dispatch algorithm.
This change corrects ADL-P support for systolic mode changes.
Related-To: NEO-5019
Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
Adding new interface to cooperate with hw context state
Simplify programming removing unnecessary functions
Code optimization that stop using expensive call and instead
stores configuration parameter
Related-To: NEO-5019
Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
So far captureStateBaseAddress() was a wrapper around
programSbaTrackingCommands(), doing an additional checking before
calling the latter. The checking is apparently no longer relevant, so
unify the distinction and remove part of the code which is no longer
needed.
In practice, keep the captureStateBaseAddress() while moving the body of
programSbaTrackingCommands() into it. This imposes lower diff-impact
onto the class hierarchy. Remove the second function. Simplify the
caller which had to distinct these two functions previously.
Related-To: NEO-6774
Signed-off-by: Maciej Bielski <maciej.bielski@intel.com>
Fixes found out while working on the StateBaseAddress adaptation to
StreamProperties. Removing unused parameters, improving code reuse
(further improvements come with following commits).
Related-To: NEO-6774
Signed-off-by: Maciej Bielski <maciej.bielski@intel.com>
It was only needed for LOAD_BALANCED scenarios, so with recent disabling
of this feature in KMD, it is no longer required.
Signed-off-by: Michal Mrozek <michal.mrozek@intel.com>
Discarding RAII lock returned from function almost always
is a bug. This change introduces usage of [[no_discard]]
attribute from C++17 to prevent such misues.
Signed-off-by: Patryk Wrobel <patryk.wrobel@intel.com>
With compiler LSC WAs this gives better performance.
If debugger is active, policy will not be changed ie.
will be WBP.
Related-To: NEO-7003
Signed-off-by: Dominik Dabek <dominik.dabek@intel.com>
This change introduces usage of forward declarations
and removes unneeded includes from the mentioned files.
Signed-off-by: Patryk Wrobel <patryk.wrobel@intel.com>
With compiler LSC WAs this gives better performance.
If debugger is active, policy will not be changed ie.
will be WBP.
Related-To: NEO-7003
Signed-off-by: Dominik Dabek <dominik.dabek@intel.com>
This change introduces capability to set surface state base address
when surface state heap or global base address are not available
Related-To: NEO-7187
Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
When multiple command queues use the same context and retain the same state
No preemption programming for copy command queues
Program preemption preamble only for mid thread preemption
Related-To: NEO-7187
Signed-off-by: Zbigniew Zdanowicz <zbigniew.zdanowicz@intel.com>
EOT WA requires allocating last 64KB of kernel heap and putting EOT
signature at the last 16 bytes of kernel heap
Related-To: NEO-7099
Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
Related-To: NEO-7003
Add L1CachePolicyHelper struct.
This struct is resposible for L1 cache policy
in build option, Surface State and stateless
caching. Currently default option for all
platforms is WBP (write by-pass)
Signed-off-by: Szymon Morek <szymon.morek@intel.com>
For all platforms different than XE_HP_SDV (ATS) stop considering the
`useGlobalAtomics` flag as a decisive factor for trigerring the SBA
(StateBaseAddress) programming on the HW. Only XE_HP_SDV supports such
flag.
For consistency of the implementation, keep the related logic in one
place only, that is a helper in `command_encoder` and then just reuse it
in different places (`command_stream_receiver`).
Related-To: NEO-6953
Signed-off-by: Maciej Bielski <maciej.bielski@intel.com>
This change introduces detection of GPU hangs in flushBcsTask()
function. The new code has been covered with ULTs.
Related-To: NEO-6681
Signed-off-by: Patryk Wrobel <patryk.wrobel@intel.com>
Define single .clang-tidy configuration with all used checks and use
NOLINT to selectively silence tool. That way cleanup should be easier.
third_part/ has its own configuration that disables clang-tidy for this
folder.
Signed-off-by: Artur Harasimiuk <artur.harasimiuk@intel.com>
Add debug variable ForceGrfNumProgrammingWithScm.
Do not update large grf value in StreamProperties when unnecessary.
Related-To: NEO-6659
Signed-off-by: Filip Hazubski <filip.hazubski@intel.com>
This is fixed reupload of this commit after auto revert
With this commit OpenCL will track if external host memory is used from
few threads and will secure to update task count in all threads before
destroing allocation.
Resolves: NEO-6807
Signed-off-by: Maciej Plewka <maciej.plewka@intel.com>
With this commit OpenCL will track if external host memory is used from
few threads and will secure to update task count in all threads before
destroing allocation.
Resolves: NEO-6807
Signed-off-by: Maciej Plewka <maciej.plewka@intel.com>
use tag allocation address as a completion address in exec call
wait for completion value before destroying drm direct submission
Related-To: NEO-6643
Signed-off-by: Mateusz Jablonski <mateusz.jablonski@intel.com>
- add debugger support to imm cmd lists
- add debugger support to flushTask
Related-To: NEO-6845
Signed-off-by: Mateusz Hoppe <mateusz.hoppe@intel.com>
- helper sets all SbaAddresses for debugger in
EncodeStateBaseAddress<GfxFamily>::setSbaAddressesForDebugger()
- change DebuggerL0::captureStateBaseAddress() to take
LinearStream
- move getSbaTrackingCommandsSize() to Debugger class
Related-To: NEO-6845
Signed-off-by: Mateusz Hoppe <mateusz.hoppe@intel.com>
Stack vector will not cause dynamic allocations in most circumstances
ie. number of root device indices not more than 16
Related-To: NEO-6837
Signed-off-by: Dominik Dabek <dominik.dabek@intel.com>
Remove parameter requiredThreadArbitrationPolicy
from PreambleHelper::programPreamble function.
Signed-off-by: Filip Hazubski <filip.hazubski@intel.com>
Logic related to programming non coherent and thread arbitration policy for
gens 9 and 11 has been moved to EncodeComputeMode object, where similar
logic for gens gen12lp and newer is located.
Functions PreambleHelper::programThreadArbitration and
PreambleHelper::getThreadArbitrationCommandsSize have been removed.
Redundant setForceNonCoherent call has been removed from XE HPG
Related-To: NEO-6728
Signed-off-by: Filip Hazubski <filip.hazubski@intel.com>
This change introduces the new flag called DisableGpuHangDetection.
By default it is disabled. When someone wants to disable hang checking,
then this flag can be set to true.
Related-To: NEO-6681
Signed-off-by: Patryk Wrobel <patryk.wrobel@intel.com>
Change ThreadArbitrationPolicy::NotPresent value to -1
Update initial values to ThreadArbitrationPolicy::NotPresent
Related-To: NEO-6728
Signed-off-by: Filip Hazubski <filip.hazubski@intel.com>
Setting one of bitfields requires read from local memory which is very slow.
This is not needed for devices that have coherent L3.
Signed-off-by: Michal Mrozek <michal.mrozek@intel.com>
Some devices do not need dcFlush.
Setting it prevents further optimization of pipe controls which
are not needed.
Signed-off-by: Michal Mrozek <michal.mrozek@intel.com>
This change:
- moves NEO::WaitStatus to a separate file
- enables detection of GPU hang in clWaitForEvents
- adjusts most of blocking calls in CommandStreamReceiver to return WaitStatus
- adds ULTs to cover the new code
Related-To: NEO-6681
Signed-off-by: Patryk Wrobel <patryk.wrobel@intel.com>
When running multiple threads, one thread could clear
allocationsForDownload while another was iterating over it.
Signed-off-by: Dominik Dabek <dominik.dabek@intel.com>
This change introduces detection of GPU hangs in
zeEventHostSynchronize and zeFenceHostSynchronize.
Furthermore, if CommandQueueHw::executeCommandLists
uses ZE_COMMAND_QUEUE_MODE_SYNCHRONOUS and hang occurs,
the information about it is propagated to the caller.
Related-To: NEO-6681
Signed-off-by: Patryk Wrobel <patryk.wrobel@intel.com>
Rename:
- debug flag ProgramPipeControlPriorToNonPipelinedStateCommand
to ProgramExtendedPipeControlPriorToNonPipelinedStateCommand
- local variables
Related-To: NEO-6615
Signed-off-by: Krzysztof Gibala <krzysztof.gibala@intel.com>
Add pipe control before state base address, state compute
mode and state sip commands.
Related-To: NEO-6615
Signed-off-by: Krzysztof Gibala <krzysztof.gibala@intel.com>
Add pipe control before state base address, state compute
mode and state sip commands on DG2 and PVC when CCS flow is used.
Signed-off-by: Krzysztof Gibala <krzysztof.gibala@intel.com>
This change uses value of cpuAddress from monitored fence
to detect GPU hang.
Related-To: NEO-5313
Signed-off-by: Patryk Wrobel <patryk.wrobel@intel.com>
-program barrier after global fence allocation is programmed
-do not double barrier timestamp in blit enqueue
-flush GPGPU while submitting to BCS when barrier requested
Signed-off-by: Lukasz Jobczyk <lukasz.jobczyk@intel.com>
When linear stream created for command container has not enough space
for command and BB_END it will program BB_END and allocate new command
buffer allocation. Pointer returned from getSpace in this case will
return storage from new command buffer allocation.
Related-To: NEO-5707
Signed-off-by: Maciej Plewka <maciej.plewka@intel.com>
This change uses DRM_IOCTL_I915_GET_RESET_STATS to detect
GPU hangs. When such situation is encountered, then
zeCommandQueueSynchronize returns ZE_RESULT_ERROR_DEVICE_LOST.
Related-To: NEO-5313
Signed-off-by: Patryk Wrobel <patryk.wrobel@intel.com>
For non-kernel submission, TAM was incorrectly reprogrammed to default
mode. Correct programming should reuse value from previous submission.
Signed-off-by: Bartosz Dunajski <bartosz.dunajski@intel.com>
Rename MemorySynchronizationCommands::isDcFlushAllowed
to MemorySynchronizationCommands::getDcFlushEnable
Signed-off-by: Filip Hazubski <filip.hazubski@intel.com>
Related-To: NEO-6432
This change applies WA that always programs all fields in SCM for
gen12lp. Also for those platforms Force Non-Coherent is set to 0x2.
Signed-off-by: Maciej Plewka <maciej.plewka@intel.com>
Currently only supported scenario is one in order queue.
Instead of resolving dependencies via semaphores, do this with pipe controls.
Signed-off-by: Michal Mrozek <michal.mrozek@intel.com>
- add function for resolving builtins using relocations
- add relocation for __SubDeviceID symbol
Related-To: NEO-6258
Signed-off-by: Kamil Diedrich <kamil.diedrich@intel.com>
Multi tag allocation is useful only in openCL to ensure cross root
device synchronization based on tag address.
It is not required in level zero.
Futher multi tag allocation is causing instability in sysman
device reset. So, for level zero instead of multi tag allocation
create single tag allocation.
Related-To: LOCI-2651
Signed-off-by: Jitendra Sharma <jitendra.sharma@intel.com>
After using clEnqueueVerifyMemoryINTEL api call
MEM_TRACE_MEMORY_COMPARE instruction was not present
in dumped aub file.
Signed-off-by: Rafal Maziejuk <rafal.maziejuk@intel.com>
Resolves issues with coexistance of NEO L0 and NEO OCL libraries
within a single process running in WSL and using WDDM GPU PV
Signed-off-by: Jaroslaw Chodor <jaroslaw.chodor@intel.com>
Before performing gpu device reset, first all level zero resources
and gpu device specific resources have to be cleaned up. Also as
after device reset, state of gpu device would be lost.
Hence after performing gpu device reset, level zero device have
to be reinitialized by querying gpu device again.
This change is aimed at reinitializing the level zero resources
after gpu device reset, so that user could continue using level zero
devices after device reset.
Related-To: LOCI-2627
Signed-off-by: Jitendra Sharma <jitendra.sharma@intel.com>
Dunajski, Bartosz