Dont track the status of fence_dirty_ flag on the host, instead clear it
when we submit a barrier on the respective stream.
Change-Id: I4d98dbf20c81379c9c5da9f5b67629a8f9f6dfcd
This issue is due to the fact that we set fence_dirty_ flag for every
kernel launch. Whenever any next HIP API is caused, the stream logic
assumes fence is dirty and queues a marker, when its not needed.
Change-Id: I10a49ee951daa92221aa6be75e2334849579a45d
Recently some unused compiler options for HSAIL path were removed,
but it affected blit kernels compilation. Hence, remove those options.
Also delete assert for device to device copy in SDMA path for now.
Change-Id: Ib5d7f063af2ab4a3fc5d73d426e39c391b1011ac
Single OCL context with multiple devices should use direct map
even with persistent memory, because cache coherency layer.
Change-Id: I02a48830440203a1dea176acd4dff4452b70f133
- Make sure persistent memory from resource cache is properly adjusted
in free memory calculation.
Change-Id: I74ef68975ccde4694fb1cb904617c418e85dfc9f
This is an initial change before we refactor the build/link paths for
kernel launches for HIP. This current change is needed as compiler was
setting some dump file which needed fs access which has slowdowns for
NFS mounted file systems
Change-Id: I828f9bb04d789b4f8c05c1ed08767f325efeb47c
Add GPU_DEBUG_ENABLE to control ttpm behavior. If enabled,
then HW will collect more debug info at some perf cost
Change-Id: Icee0686b903a7b1bd483710b9d611877cd43c6aa
- For new AccumulateCommand we enqueue nop barrier packet. So during stream sync we need system flush.
Change-Id: I0b97b626bcdae582ef95e7c95030d78df1fa5a54
Recent changes disabled system memory allocation
in the abstraciton layer. That requires memory
allocation/destruction in ROCR. Add destruction logic.
Change-Id: I68fe6b0a620ca743fe5850052ea0efa8bb7931c2
Extra CPU read back will be performed before every submission to make sure
previous writes over PCIE reached GPU. HDP flush is done by CP.
Change-Id: I402d28ca26c8cee4a3920feb3599af8c285d0889
Original logic didn't use pitch because, abstraction layer had
a sysmem copy without pitch. Since extra sysmem copy was
disabled, the code has to accept pitch values from the app.
Change-Id: Ia9fba7b33ddff4e9109b4e63d0d6afa52f501c8f
The new copy kernel can limit the number of launched workgoups.
It can copy in chunks of 16 bytes or 4 bytes.
Workgoup size is increased to 512 or 1024
Change-Id: Ic3fefa2d5bda6afebd1acc4d41ad310b138af6df
It seems that due to removal of vdev()->isHandlerPending(),
Marker queued to ensure finish is not enqueued and that cause
hung at waiting event for kernel enqueue command.
Change-Id: I364abb2dcb4897b11a7eb61b5d85013b69292792
- Add the new fillBuffer kernel, which allows to launch a limited
number of workgroups for memory fill operation
- Switch fill memory to 16 bytes write by default
- Allow to limit the workgroups with DEBUG_CLR_LIMIT_BLIT_WG
Change-Id: Ibad1822f2d42b2fc71bcfc1917c31409c0623e8e
- Report kernel names for optimized graph path
- Refactor code so that we store profiling info in Accumulate command
Change-Id: Ib97735a0239aeb9fc3a50a4bb7126dd0bcadc8af
The "long" type size seems to be platform dependent, causing hash value
overflow on implementations where "long" is 4 bytes. This addresses the
scenario.
Change-Id: I4e3c0df457e35b139dcc496d832210ba2cb849ba
- Do not use extra barrier to detect graph end. If its a kernel node we
can use a completion signal for the last packet. Saves roughly 6us for
Phantom testcase per graph launch.
Change-Id: I5e0c2479d9964fbeda86ed97533f6718f49a7f91
Add support of HIP_FORCE_DEV_KERNARG under PAL.
Fix persistent memory detection for a resource view.
Change-Id: Ifb7db2db14e0c2205a9661cfa53887ec61ab26a4
- Track all captured commands under a new AccumulateCommand
- Add begin() and end() methods to capture commands
- Explicit TS object now passed to certain methods because
profilingBegin() and profilingEnd() now happen separately and thus can
run into threading issues
Change-Id: I171106bdcad72b057836cb2f3fc398db3533119f
Move context allocation into Device::init() method to simplify the logic and handle
HIP_VISIBLE_DEVICES properly
Change-Id: I0fc6f37c7ae39bedbdad0290295d6794c66d6c54
- Support graph with different types of nodes with single
branch when DEBUG_CLR_GRAPH_PACKET_CAPTURE flag is enabled
Change-Id: I149a8629769cd0d5849ffefb04f1352668a685b6
If we submit a systemScope Barrier, we should reset this flag as there
is no need for dispatch AQL again to flush caches/HDP
Change-Id: I55710feb4ba6650852e785b5cadfa64c6b9ce14e
Add hipMemcpyDeviceToDeviceNoCU to force a non blit copy path. This
helps in cases where an app may determine that CU may be busy and copies
with SDMA may be quicker.
Change-Id: I59b415dd8f6022c244e8d75f265464d5c635df1e
Fix wrong logic to get layer index;
Make layered image's layout match cuda spec;
Fix wrong comparision of element size.
Remove amd::BufferRect from ihipMemcpyAtoHCommand()
and ihipMemcpyHtoACommand().
Change-Id: Icc6a4233fbce2e9b2dc6feb79e6bfbd761684c7d
Pass LUID infromation from PAL to HIP device properties.
NodeMask is calculated based on the order of devices in PAL
Change-Id: Ic53707b68b16f392e5284a0105bf1f3941b46653
German
Saleel Kudchadker
ksankisa
Satyanvesh Dittakavi
Anusha GodavarthySurya
Jaydeep Patel
kjayapra-amd
Jatin Chaudhary
taosang2
Vikram
Alex Xie
Ioannis Assiouras