=> hipDeviceSynchronize is not allowed during capture.
=> hipEventSynchronize during capture should return hipErrorCapturedEvent error
=> hipEventQuery during capture should return hipErrorCapturedEvent error
hipStreamSynchronize, hipEventSynchronize, hipStreamWaitEvent, hipStreamQuery
For Side Stream(Stream that is not currently under capture):
=> If current thread is capturing in relaxed mode, calls are allowed
=> If any stream in current/concurrent thread is capturing in global mode, calls are not allowed
=> If any stream in current thread is capturing in ThreadLocal mode, calls are not allowed
For Stream that is currently under capture
=> calls are not allowed
=> Any call that is not allowed during capture invalidates the capture sequence
=> It is invalid to call synchronous APIs during capture. Synchronous APIs,
such as hipMemcpy(), enqueue work to the legacy stream and synchronize it before returning.
Change-Id: I201c6e63e1a5d93fd416a3b520264c0fdbe31237
[ROCm/clr commit: 19b4660cbb]
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]
Added call to hipDeviceSynchronize in __hipUnregisterFatBinary
to ensure that all HSA async signal handlers complete before removeFatBinary
Change-Id: I756fecca1c2a5eae092613d8079de266399e5685
[ROCm/clr commit: ad32e604c7]
Currently we force inlining everything for HIP. Now we'd like to enable function
supports. The first step is to remove uses of `-amdgpu-early-inline-all` in
various places. This patch is to remove all of them from clr.
Change-Id: Ib0cad1f586714c9989778b00746aa4c47a4eec95
[ROCm/clr commit: a09204388a]
Mempool has capability to track dependency between streams for
faster memory reuse. Enable that capability.
Change-Id: I28266a7e38d0fc4c5d027b9542d3719653840821
[ROCm/clr commit: 17d0c166d2]
- Using runtime unbundler, no any gfx device can load fat binary,
if there is any device without available code object.
- Extract available code object to corresponding gfx devices. So
users can work ROCm with those ready devices without segmentation
fault.
Change-Id: I9f14c65ecebf2d3c4b127a007cb434a3ae98c450
[ROCm/clr commit: 6723277ad4]
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]
- enforce incrementing the table versioning number when a table size changes outside of ifdef for ROCPROFILER_REGISTER
- add new HIP_ENFORCE_ABI entries
- update the HipDispatchTable size and bump HIP_RUNTIME_API_TABLE_STEP_VERSION to 1
- re-enable rocprofiler-register
Change-Id: Ie0cc1d8491c5640056e5dd393ea243e4dce4e8a9
[ROCm/clr commit: d84c5ae3af]
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]
Check the pointer if its present in the arrayset before trying to dereference
it as it can cause access violation if the pointer is allocated using malloc
Change-Id: Ida72b9015dc22269fc1fbe0728e66e3de29fda3d
[ROCm/clr commit: 821ae6a103]
- Implement workaround to ensure HDP writes are done by writing and
reading the HDP MMIO register.
- Implement the same workaround for graphs, we no longer need sentinel
write/readback
Change-Id: I0d3027b46a1f61131ec62e3c8c669ff5184fa6b2
[ROCm/clr commit: f138e0d113]
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]
Read and write int bytes sentinal value to dev_ptr or PCIE connected devices at the tail end of the kernarg surface.
Change-Id: I993d552ac872b3cd56aef4746c4d1d92c58d38b4
[ROCm/clr commit: 853abeb75e]
Anusha GodavarthySurya
Jaydeep Patel
kjayapra-amd
German Andryeyev
Ioannis Assiouras
David Salinas
Satyanvesh Dittakavi
jiabaxie
taosang2
sdashmiz
Shilei Tian
Sameer Sahasrabuddhe
Rahul Manocha
Jatin Chaudhary
ChingShihLi
Sourabh Betigeri
Jonathan R. Madsen
Saleel Kudchadker
Ilya Panfilov