- Added the optimized multi stream path in graph execution. It uses a fixed number of async streams in the execution
- Optimize the launch latency, where commands
creation and execution is done at the same time
- Optimize the scheduling to use less barriers and waiting signals if
the same queue can be detected
- The new path is controlled by DEBUG_HIP_FORCE_GRAPH_QUEUES
environment variable, where 0 will use the original path and any other
value will force the number of asynchronous queues for execution
- DEBUG_HIP_FORCE_ASYNC_QUEUE can force single queue async
execution in graphs(applicable for Navi families only)
Change-Id: I7eb40bc15c45f508d6911868a6f6d4c3598d380e
[ROCm/clr commit: 9db52f9a46]
=> Added support to capture multiple AQL Packets.
=> Added Interface to callback to hip runtime from rocclr to allocate
kernel args from the graph kernel arg pool.
=> Enabled Support to capture memset node.
Change-Id: I7e1c2ba06927459e024653058af142bd82192c43
[ROCm/clr commit: bd3a35bde1]
This issue was happening because of incorrect usage of getStream call,
if we get the null stream first and then typecast it, and call on
getStream again, we lose the advantage of simply passing "nullptr" to
indicate NULL stream. Thus we enter the waitActiveStream call and add
barriers to sync across streams.
Change-Id: I94dc4e3ec927295b9e1ab6dee4b37d7d3e00b0cc
[ROCm/clr commit: cda4b7db1c]
Creation of ReferenceCountedObject will increase reference count by 1.
Clear the commands from Node after capture so that they wont be reference later.
Change-Id: I1cc4085939cf65218ec2aa2e25ab6d737f7cacd3
[ROCm/clr commit: 6ae5d6896c]
Capture AQL packets during GraphInstantiation and enqueue AQL packets during graph launch.
Added support to capture single graph memset node.
Capture support for memset node is currently disabled.
Memset capture will be enabled when capture for multiple packets are supported..
Change-Id: I14dfbc41731025cc3a548a730558915def3fa384
[ROCm/clr commit: 346da4bb40]
For refactoring of childGraph to have its own graphExec,
kernelArgs needs to be separated from the graphExec object.
All the childNodes part of graph should share same kernelArg pool.
Otherwise we endup creating multiple device kernel arg memory chucks
for single graphExec.
Change-Id: I4029a46ebc1fa112d87df64ab1fecbf288fabe5e
[ROCm/clr commit: 35079e834e]
- Introduce a lock when checking isUserObjectValid. We need a lock
here as one can remove the userObject T2, leading to buffer overflow
when checking ranges in T1.
Change-Id: I058144b8cc463c90ab6bf5cf96bf937897742917
[ROCm/clr commit: 6ac67afdd5]
If graph has multiple branches, End command is enqueued on launch stream which
makes sure all the internal parallel streams are finsihed.
When node is removed from the graph, indegree and outdegree are not getting update correctly for parent, child nodes and
resulting in endNode not having deps on parallel commands. Resulting in graph sync issues.
Change-Id: I33cc2f21220e1c017d88099b29b542e05b683f73
[ROCm/clr commit: 9ad7e79e50]
Resolved an issue where a freed virtual buffer was incorrectly
added to the global mapping causing an assertion error during
teardown process.
Change-Id: I4801157a28603ce9be1ca0131982b700ff884f7a
[ROCm/clr commit: 1c6b92627d]
If the graph has kernels that does device side allocation, during packet capture, heap is
allocated because heap pointer has to be added to the AQL packet, and initialized during
graph launch.
Handle race with wait when 2 kernels with device heap are enqueued on multiple streams.
Change-Id: I45933b77fcaf7bc8fdf1bc906462e32b5d8d3688
[ROCm/clr commit: 57156c524d]
Handle GraphExec instance is destroyed before async launch completes
GraphExec instance is destroyed after async launch completes
GraphExec instance is destroyed without a launch
Change-Id: I45a7c82295fea916c7559bd8f796df710513aea1
[ROCm/clr commit: bf4d10ff61]
For all windows allocation on SVM memory tagged with flag ROCCLR_MEM_INTERPROCESS.
hipHostMalloc validation is based on flag. So remove ROCCLR_MEM_INTERPROCESS before check.
Change-Id: I823bbf228d9a4a9acb4abffc01ac6b3f544c6e12
[ROCm/clr commit: 8179fa98a2]
Under ROCr physical allocations don't have initial VA and require extra
flag in ROCclr. Add an option to have a mempool of physical allocations.
Change-Id: I4d062fe0dd8113d4eaf6e8b51749ed56d8701d1e
[ROCm/clr commit: f296159f62]
Node can be enabled/disabled only for kernel, memcpy and memset nodes.
If the node is disabled it becomes empty node.
To maintain ordering just enqueue marker with respective node dependencies.
Change-Id: I710f3e88ab4e76c81f6f86a40a7dc61fd4c7e440
[ROCm/clr commit: e0e63eb04d]
When kernel does device side malloc, initial heap is allocated with __amd_rocclr_initHeap.
During graph launch kernel __amd_rocclr_initHeap is enqueued followed by actual kernel . So kernel will execute after initHeap kernel.
But with graph optimizations during capture initHeap gets enqueued on device null stream and actual kernel on graph launch stream.
So no proper synchronization. Switch to command creation and enqueue during launch for kernel node with hidden heap.
Change-Id: Iaf600251faef9a448853f19429023c118aa760b9
[ROCm/clr commit: 2dc6ec68a5]
When graph is Instantiate on device 0 graph and launch on device1 switch to command creation and enqueue during launch.
Change-Id: Ied34dc99b2a776130d1354ed3830c6ccab9068e4
[ROCm/clr commit: 7d09e1abed]
Windows path still uses multi threading implementation. Hence, in graphs
all nodes are executed in a queue thread and that requires to manage
mempool in the queue thread. However, the spec allows to destory memory,
allocated in a graph, outside of the graph's execution. That may cause
mempool management to go out of sync.
Change-Id: I0ffb2244b3cb720455ed44d1b3e2487fa8959a77
[ROCm/clr commit: 4f6b0da028]
During hipGraphExecKernelNodeSetParams kernel function can also be updated.
Hence size required for kernel parameters differs from what is allocated during graphInstantiation.
So, create new 128KB kernel pool and allocate kernel args from the pool.
If the pool is full create new 128KB pool. Release kernel pools when graph exec object is destroyed.
Change-Id: I9567946d63400c79cbfd4c5439c654c92557ceae
[ROCm/clr commit: e9957151f3]
- Remove newline from logging as log function internally inserts a new
line
Change-Id: I25eb2242a1f1e87cf811bcc373d1d485b2e027a8
[ROCm/clr commit: 058b2702db]
- Report kernel names for optimized graph path
- Refactor code so that we store profiling info in Accumulate command
Change-Id: Ib97735a0239aeb9fc3a50a4bb7126dd0bcadc8af
[ROCm/clr commit: b056686607]
- Support graph with different types of nodes with single
branch when DEBUG_CLR_GRAPH_PACKET_CAPTURE flag is enabled
Change-Id: I149a8629769cd0d5849ffefb04f1352668a685b6
[ROCm/clr commit: 38d2c56784]
- Address corner cases that can arise with the new
hipMemcpyDeviceToDeviceNoCU enum
- Better log
Change-Id: I6035b901f8d616741054b7a5ff4f67956329ac57
[ROCm/clr commit: 5662d4037c]
Addresses the below scenarios like parameters mismatch
in memcopy node, difference in the count of nodes, difference
in the dependencies of the nodes.
Change-Id: I31c6516fb27cc1007809f1b50306fdb0c2356ccc
[ROCm/clr commit: f16d336e32]
Jaydeep Patel
German Andryeyev
Anusha GodavarthySurya
Saleel Kudchadker
Ioannis Assiouras
shadi
Rahul Manocha
kjayapra-amd
Rahul Manocha
Sourabh Betigeri
Satyanvesh Dittakavi