- get PageTable entry data hints and address space from AubHelper
based on local memory flag
- add enableLocalMemory flag in CSR HW
Change-Id: I061bda62be8da55d52cff48ecddcf26c4212dc67
- Command Stream Receiver should be used instead for locking.
- Remove not needed synchronization in clSetUserEventStatus
Change-Id: I17050dc70cb0be03b2003043a9666ba8df1a83c9
- resources are dumped in make non resident call
- in order to dump correct data we need to be sure that GPU is done processing
- waiting needs to be unconditional to handle all cases
- remove not needed parameter to makeSurfacePackNonResident
Change-Id: Ib2b065d486cd3a5d86e599c51b24f3c958c3a10b
This code is an infrastructure for special debug purpose that allow measure
execution time of any hardware command.
Change-Id: Id12a7979d204734a0c4a6c4700e427b65ac2397f
- replace createGraphicsAllocationWithRequiredBitness with more general
methodallocateGraphicsMemoryInPreferredPool based on passed
AllocationData
- proper flags for allocation selected based on AllocationType
- remove allocateGraphicsMemory(size_t size, size_t alignment)
and use allocateGraphicsMemory(size_t size) instead where default
alignment is sufficient, otherwise use full options version:
allocateGraphicsMemory(size_t size, size_t alignment,
bool forcePin, bool uncacheable)
Change-Id: I2da891f372ee181253cb840568a61b33c0d71fc9
- makeCoherent should be called after TBX finished processing
- this is when tagAddress is updated with taskCount
makeCoherent is called from makeNonResident which is invoked just
after flush and may happen before TBX server finished processing
leading to invalid data to be read back to CPU accessible memory
- this fix adds waiting for taskCount to blocking calls for TBX CSR
before calling makeNonResident on surfaces to guarantee correct data
from TBX server is ready.
Change-Id: I498a5454e0826eec2a5413a08880af40268550e1
This commit adds a capability to selectively enable/disable AUB capture,
i.e. by toggling the registry key from the outside or specifying the filter
with a kernel name and/or kernel start index and kernel end index.
Change-Id: Ib5d39c21863fbc4a95aa73c949b9779ff993de0f
when SLD is active:
- make Sip Kernel Resident
- program GPGPU_CSR_BASE_ADDRESS
- Disable Preemption
- adjust getDebuggerOption input param,
value passed has to be at least 2 bytes in size
change unit test behaviour accordingly
Change-Id: I4ec87d0e8dfcf02437fdeeffc5363314eea5dd07
- Rename misnamed test function
- Adjust 2 tests, so they use CSR size getters instead of hardcoded values
- Move getSizeRequiredPreambleCS() into CommandStreamReceiverHw class
- Improve PreambleHelper size estimating
Change-Id: I3f292d50e08f3d10d190c9f8722e1f0498481154
- This is to improve battery usage while waiting in busy loop on CPU
- New Kmd Notify helper to maintain dynamic parameters
- Ask OS about battery status on longer waits
- Pick different timeout when using battery and optimization is disabled
Change-Id: I5f9c8c5a9c635652aac27c707f2b55933947a7fb
- Decission about timeout enabling and value moved out of CSR
- Timeout multiplier is no longer Linux specific
Change-Id: I6858fe2f811ef13802b95e0470e310210a9dea8b
- Switch to internal heap for kernel ISA allocations.
- remove IH from various functions
- remove IHState from CSR , IH is never dirty
- ISA is no longer copied on enqueue calls.
Change-Id: I0099cf2a9ebab6192ea03a74dd35f7da963fd5a5
-This is to make sure those functions are not called when gtpin is not used
-This preserves CPU instruction cache pollution.
-Our enqueue path needs to be as thin as possible, even with this small change
there is visible gain in ULT execution time.
Change-Id: I44cc2144754cda95ca1fe058184cd8a151b8d35c
- Measure time between wait calls. If delay is exeeded use QuickKmdSleep
- Kmd Notify helper functions
- Refactor overriding from debug variables
- Refactor Kmd Notify tests
Change-Id: I123c31f492d98fd304184f99ee0bf7d733d06f04
- KmdNotifyProperties struct for CapabilityTable that can be extended by
incoming KmdNotify related optimizations
- Quick KMD sleep optimization that is called from async events handler
- Optimization makes a taskCount check in busy loop with much smaller
delay than basic version of KMD Notify optimization
Change-Id: I60c851c59895f0cf9de1e1f21e755a8b4c2fe900
New debug option FlattenBatchBufferForAUBDump has been added. When set it
modifies AUB dump in such way that commands from main and chained batch
buffer are dumped as single allocation. Commands from chained batch buffer are
dumped directly after commands from main batch buffer without
MI_BATCH_BUFFER_START. This feature also requires ImmediateDispatch mode which
can be forced using debug option CsrDispatchMode = 1.
Change-Id: I730760791693a748e7f4e1463ce8e7af94287b93
- Microseconds offer better precision.
- Some workloads require threshold less then 1 millisecond to work
efficiently.
Change-Id: I1a565049340fb6eeebe5c0a61ededae9959daca8
-If out of order flag was disabled then pipe control was not having dc flush.
-This could led to a batch buffer that doesn't end with dc flush.
-This change adds differentiation between pipe controls that may be erased and
pipe controls that are used as a part of epilogue command
Change-Id: Ic9c970c75c89ff524a0e40506eff6dd097760145
-Do not flush dc for every command in batched mode
-Do that only in immediate mode
-For commands that needs DC do not noop pipe controls
-Ensure that each command buffer in batching mode ends with dc flush.
Change-Id: I3cd9d1831c19b69c66092687922f20df7e330245
- Tag should be updated only as a part of epilogue.
- Level change should only emit pipe control with cs stall
Change-Id: I6e04f794641818b0d046523776d3ce87aec9f606
- Program one PS with gpgpu selection and media sampler
- Program PS only when media sampler requirement changed
or when preamble was not sent
Change-Id: I85ba3f74087733e79d048e120aeb8b4b04796e00
This commit as aimed to add support for batched dispatch,
but doesn't make it the default mode for AubCSR yet.
Change-Id: I4dc366ec5f01adf2c4793009da2100ba0230c60a
Dunajski, Bartosz