Make sure runtime waits for the current queue before
synching with device queue
Change-Id: I753b6fc0bb15a3a3d4bf03fef1152842550850c0
[ROCm/clr commit: 86e0f337fc]
TF doesn't reserve all available memory now. If any
client wants to reserve they can explicitly set
HIP_HIDDEN_FREE_MEM env var
Change-Id: Ied3a948b79f49aa7327f6a820e9789e39cec143b
[ROCm/clr commit: d8ca3c632c]
If a offset of the pointer is passed to free it may release
the mem object but may not release from MemObjMap. Erase the map
by getting the parent pointer.
Change-Id: I06b92548de2d49b4029efe6b511329225007cc55
[ROCm/clr commit: 2b6fea4348]
Select cpu in terms of the smallest Numa distance for a GPU device.
This will improve performance of hipMemcpy in the mode of
hipMemcpyHostToDevice or hipMemcpyDeviceToHost for small buffer.
`
Change-Id: I2860f1f83b79be0dff7bf5e64cf68ab4448db0a1
[ROCm/clr commit: aedb9590be]
Optimization for the fence release removed a sync for mem fill.
Add simple const buffer management forr the filled pattern to avoid
pattern overwriting with the async fills.
Change-Id: I63773ac09ceec31d5396d24570e4647ff096326b
[ROCm/clr commit: 2ce6bbebc4]
These are unnecessary and an obstacle to producing a relocatable
package.
Change-Id: I0059bf7a2d11fcece0cd7ab47d7545d0df4d7099
[ROCm/clr commit: 1d267c9c08]
SWDEV-234947
SWDEV-236298
Instead of forcing a barrier packet, just inject system scope on the next packet.
Change-Id: If9bcee23e08dfe5db731235e2fcb30582cbd4c1c
[ROCm/clr commit: 6a5af4056e]
Eliminates most of the global include_directories. The install header
paths are different from the build directory, so we have to separate
those for the exported target include paths.
Change-Id: I13e4c56c1218cb31c29a316422dc5fd1d09d8b1b
[ROCm/clr commit: 3a7f2e3682]
Use ROC_ENABLE_LARGE_BAR (0/1) to toggle. The support is
enabled by default.
Change-Id: I6cb93a46594cb6f5e90bf6057738330225efb553
[ROCm/clr commit: d10d691e76]
If HIP_VISIBLE_DEVICES is active, then make sure the list of HSA
agents contains the valid agents
Change-Id: I584aad999a230ab7f88a0cfe20dcd0abe79c43a5
[ROCm/clr commit: ae4aceb55e]
Apply the optimization to change for OpenCL too.
Clean up some unnecessary checks.
Change-Id: I840261fe35baeeadeba7388e86779d482f509aad
[ROCm/clr commit: 6c5a42b33c]
Remove queue limitation since we loop through HW queues now.
Add a DevLogError if we fail to create the hsa_queue. A ticket showed a regression there.
Change-Id: I4f58e405f88e75600a762f6d6352838c969cdb5e
[ROCm/clr commit: b54c3f7db9]
This workaround is to avoid performance penalty of SDMA engine
taking a while to clock up from a lower DPM state. Add env var
GPU_FORCE_BLIT_COPY_SIZE (1024 by default for HIP in KB). Forcing
Src and Dst agent to be amdgpu makes ROCr take blit copy path for
what otherwise should have been SDMA copy
Change-Id: I222f687155f86000d17d66d25182e490b6710463
[ROCm/clr commit: 5f64e6e7ad]
SWDEV-232580 & SWDEV-232580
Allocate p2p statging buffer when full P2P access is not available between all devices.
p2p staging buffer will eventually be used when required.
Change-Id: If8490ba7b1c52c432c1e942ae95421b9d2ec7097
[ROCm/clr commit: f149fe0803]
German Andryeyev
kjayapra-amd
Aryan Salmanpour
Saleel Kudchadker
Vlad Sytchenko
Tao Sang
Alex Xie
Chauncey Hui
Matt Arsenault
Aakash Sudhanwa
Jason Tang
Christophe Paquot
Michael LIAO
Payam
agodavar