Use large signal pool if profiler is connected or profiling forced
enabled. This is needed to mitigate signal creation overhead when
profiling as signals are attached to every packet and deeper batch may
show overhead of signal allocation.
Change-Id: I8034b8a20b55328b87d593bf044f59672f9653e8
[ROCm/clr commit: 1ec0ba3537]
This reverts commit cab71e6e00.
Implement the right way to make ExternalSemaphores be signalled
only after prior works on the stream have been finished.
Change-Id: I9d5974e05d5f229170b928db4566c14e40e3cbaa
[ROCm/clr commit: d433df4761]
Let ExternalSemaphores be signalled only after prior works on the
stream have been finished.
Change-Id: I856917db905f68f55fdf484f5267f7fe8ea3117f
[ROCm/clr commit: 44a3935cda]
Add a view bit to avoid original resource destruction when parent
dependency doesn't exist with the image view cache
Change-Id: I8277afd575af8f29951c5d1a9f7d94d784251657
[ROCm/clr commit: b49e8e78e1]
Make sure parent_ field is cleared for the internal image views.
The internal image views don't require dependencies tracking.
The issue appeard only when Navi10 pitch workaround was enabled.
Change-Id: I376d212750085a9391f8c32fc2979dcb5d93c89c
[ROCm/clr commit: 6b89980fb2]
- Enable CUs adjacent pairwise for WGP mode
- In HostQueue::terminate() do not segfault if virtual device hasn't been created
Change-Id: I94402ff333308af5824878086cc238b3993d534d
[ROCm/clr commit: 8c1232124e]
rocclr/platform/external_memory.hpp:93:30: warning: class with destructor marked 'final' cannot be inherited from [-Wfinal-dtor-non-final-class]
93 | virtual ~ExternalImage() final {}
Change-Id: I56d760fa6c08544100e3bc03d35129bd16d8a428
[ROCm/clr commit: e9750de215]
Add missing mipmap Apis’ implementation.
Fix some bugs of mimpmap apis.
Use hipmipmappedArray to differentiate cuda
and driver apis on Nvidia.
Change-Id: I6079d9f3b2ddf4e42b9a6f7f3902322cfca02cfd
[ROCm/clr commit: f03c11491b]
- Make IPC interfaces generic between devices and rely on the IPC buffer
for attach/detach logic
Change-Id: Id3c18d122030329b7ee532bbb6317de9dd6a0bbe
[ROCm/clr commit: 9aa6f25f1c]
The change enables VM support in graphs on Windows. That allows
to avoid caching of all allocations at the cost of map/unmap
overhead during memory create/destroy.
Change-Id: I792be00fba099e5e5d3cd44a963e1dfd6976a86d
[ROCm/clr commit: 04b696abee]
- The implementation in mempool graphs requires refcounting VA object.
That requires release() to update the map only on the actual destruction.
- Add GPU event tracking for paging operation. Otherwise, runtime
may not always flush IB.
Change-Id: Idf99ffb894321a38e04b490116a7ca435635918d
[ROCm/clr commit: 7ef2da5aba]
Rename VK interop to ExternalMemory object, since it should handle
DX interops also
Change-Id: I536ec46d3e53ece35234a2e29030393ad411b96d
[ROCm/clr commit: 3e5803c4c0]
For large bar platform, it's not necessary to mmap
memory allocated on VRAM to cpu again
Change-Id: I0701680476829d4058b3e7b643e8df657d0c6168
Signed-off-by: Ruili Ji <ruiliji2@amd.com>
[ROCm/clr commit: 25fe45bb2a]
OS can terminate unfinished queue thread from default stream at any
time. Potentially leaving the queue lock in a bad state and causing a
deadlock if runtime destroys VirtualGPU later from the host thread.
Change-Id: I247f102ee84e6b4dba947504933395071945c85d
[ROCm/clr commit: 28daf98f1f]
Windows kills threads on exit without any notification. However,
runtime can still destroy VirtualGPU object from the host thread with
HostQueue destruction.
This change also forces RGP trace transfer on the last capture without
any delays.
Change-Id: I768e87e99e1d23a021e63c12f36e450817743759
[ROCm/clr commit: ad33a021cb]
Add lock for HSAIL only in order to fix test failures
in math brute force and integer_ops tests.
Change-Id: I5f14cdcaa4ee9867fdae63fff197a0f21ee5f1d4
[ROCm/clr commit: f29d3bc3ac]
Remove global program lock in order to fix too
long kernel launch overhead with multi-threads
on MGPUs.
This patch depends on a compiler patch that makes
LC thread safe.
Change-Id: Ic8a7374d19112764d6de5d483ec5d07a56661d1b
[ROCm/clr commit: 3b2a8f3c8b]
The ROCclr assigns zero-based IDs to GPUs in the order they are
discovered. That zero-based ID is what is used to identify the GPU
on which the HIP_OPS activity took place.
When multiple ranks are used, each rank's first logical device always
has GPU ID 0, regardless of which physical device is selected with
CUDA_VISIBLE_DEVICES. Because of this, when merging trace files from
multiple ranks, GPU IDs from different processes may overlap.
The long term solution is to use the KFD's gpu_id which is stable
across APIs and processes. Unfortunately the gpu_id is not yet exposed
by the ROCr, so for now use the driver's node id.
Change-Id: Ib78854527d600d175bb76e2df0747c33f898c615
[ROCm/clr commit: 9a82118c85]
Ajay
German
Jaydeep Patel
Saleel Kudchadker
Tao Sang
Ioannis Assiouras
Anusha GodavarthySurya
Todd tiantuo Li
Rakesh Roy
Matt Arsenault
sdashmiz
victzhan
Ruili Ji
kjayapra-amd
Sourabh Betigeri
Juan Manuel MARTINEZ CAAMAÑO
Laurent Morichetti