- Add option to skip stream allocation on access.
- Avoid null stream allocation in ihipFree, so an inactive device
won't be initialized
Change-Id: Id24426640df59a5e7a08b2dd9dcd4d67758b84bf
The build should never touch the source directory, and only write
generated files to the output directory.
Also track the dependency with a custom command and target, as is the
normal method for dependencies on generated files.
Change-Id: I9d835256c643aeef241d26ca05ab390ebba65111
Don't rely on shell commands. On Ubuntu 20.04, /usr/bin/python no
longer exists, so I'm seeing some failures from somewhere assuming the
path.
The top level CMakeLists also repeats exactly this, which should be
fixed.
Change-Id: I56b26742920f0dc40b363b409892bd41cfa485ef
This will update the hipcc script so that it will use HIP_PLATFORM, HIP_COMPILER, and HIP_RUNTIME variables correctly based on updated hipconfig file.
Removing HIP_PLATFORM=clang, only use hcc and nvcc for now. HIP_PLATFORM should eventually switch to amd or nvcc.
Change-Id: Ie527d07a1add974cc3aab8e9f61b0518147d7602
Embedded device binaries should not be cloned if we want the ROCR's
code object URI for this binary to point to the container (host
executable or shared library) instead of a memory location.
Depends-On: I7973bb0243f5a2d1b639b8a88445cfe6af919dd7
Change-Id: Id3afc981e027e5371114d011f2caaa6414f5fc58
Device binaries that are embedded inside the host binary do not
require a copy. Their lifetime is guaranteed to exceed that of the
loaded executable.
Add a 'make_copy' parameter to amd::Program::addDeviceProgram. If
make_copy is false the original image will be used and will not
get freed when the amd::Program is destroyed.
Change-Id: I7973bb0243f5a2d1b639b8a88445cfe6af919dd7
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
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
Object libraries are weird, and producing a library by using the
target objects from them doesn't automatically import the interface
properties of the linked targets. These object libraries only have
single uses, so just directly create the final library from the
sources.
Leaves libelf as an object library, since there seems to be some cmake
oddity when trying to link an unexported target to an exported one.
Change-Id: Ic379612c89340c40085c9862cfe111fa4bbff425
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
Aaron En Ye Shi
German Andryeyev
Aaron Enye Shi
Vlad Sytchenko
kjayapra-amd
Christophe Paquot
Alex Xie
Michael Hong Bin Liao
agodavar
Matt Arsenault
Chauncey Hui
Payam
Laurent Morichetti
Saleel Kudchadker