These packages contains only the ICD loader (libopencl.so). This allows
legacy opencl (PAL) to use these packages instead of duplicating these
packages in both builds.
This also resolves a licensing issue where OCL ICD is licensed as
Apache, while opencl is MIT. Having a separate package avoid license
confusion.
Signed-off-by: Jeremy Newton <Jeremy.Newton@amd.com>
Change-Id: I3dd29239841da7f5dc767419a9bc35d9046e9936
Add a threshold for ROCR/SDMA P2P transfers. ROCR copy path
requires extra barriers in compute for synchronization. That costs
extra performance with tiny transfers.
Reduce active wait time to 10us. Tensorflow uses extra thread
per GPU with constant hipEventQuery() calls. Longer active waits
in ROCr affect CPU performance.
Change-Id: I9020358438615fa2d4617f862f00a562f0a588e7
Clang builtins doesnt yet support max/min atomics with float/double
parameters, Hence modify atomicMax/Min to use compare and swap
implementation
Change-Id: Ia351f9b136129fe25f0eddcbbb17549553247182
To fix SWDEV-306293, x86_64-rocm-opencl.conf was renamed to
10-rocm-opencl.conf. This was not changes in the clean up, only in the
creation, so the patch causes a regression by not cleaning up the new
filename.
Signed-off-by: Jeremy Newton <Jeremy.Newton@amd.com>
Change-Id: I4aa456955f6f7c2d52a3560e5e792537f95add17
With SAM on, don't force Persistent for allocations
in HIP. This makes ROCCLR go down paths we don't want
for HIP.
Change-Id: If54cc16fa891d4cfdc761c6ab21ad707627e822a
If the target device doesn't support image,
compiler will send out unavailable error if device texture apis
are called; and host texture apis will return hipErrorNotSupported.
Remove useless include/hip/amd_detail/texture_functions.h.
Add hipDeviceAttributeImageSupport attribute to check image
support in runtime.
Add __HIP_NO_IMAGE_SUPPORT(=1) macro in device code to indicate
texture apis not supported.
Change-Id: Ia891493e681a61980a2444859a90e046a9ebea39
Fix FEAT-39125
Add EMU_ENV option. If it is ON, the building is for
emulation environment, thus some logics can be adjusted
to match emulation environment. If it is OFF, the
building is for regular environment.
Currently only ocltst will use the option. But it can
be used for other modules.
Change-Id: I54e1bc1309e82794b41fca2ae1f01f004138dced
renaming rocm-opencl ldconfig file with "10-" prefix to prioritize
over system icd loader.
Change-Id: I0a54de7a0bc0e256d6470eae743ec712d1e42a36
Change-Id: I4af1d0e3e27f40f8b03b798cae43cfd4ba4a1880
The license file needs to be installed.
RPM also requires the LICENSE field to be set.
Signed-off-by: Jeremy Newton <Jeremy.Newton@amd.com>
Change-Id: I8bcec77e51b88f3f6a6fa5799d6e38db1d1ab566
Stall in the host thread could occur earlier than the app expects.
Make sure rutnime can grow the signals to the queue size without
any stall. Also adding a new signal to the end of the pool could
break the dependency chain on signal reuse. The new logic will
insert the new signal after current to keep the chain intact.
Change-Id: I9c90b98515907db8b677528263c3e88cd9581a14
Chauncey Hui
haoyuan2
German Andryeyev
Ajay
Sarbojit Sarkar
Jeremy Newton
Julia Jiang
Satyanvesh Dittakavi
Christophe Paquot
Tao Sang
Jatin Chaudhary
anusha GodavarthySurya
sdashmiz
Saleel Kudchadker
Freddy Paul
kjayapra-amd
pghafari