- Leverage managed buffer that would use chunks for fill pattern. Use a
different chunk for the next fill to avoid wait
Change-Id: I254483c867e112f66564ffd8f55e0a605d8896c9
- flags passed should be only one type at a time
Signed-off-by: sdashmiz <shadi.dashmiz@amd.com>
Change-Id: I3e56f036e51d8cc84fe5c18c06cfa11cf785233f
[ROCm/clr commit: 030e7702f1]
- found when using ROCm-enabled OpenMPI with rocHPL
- when wrapping pthread_rwlock_rdlock, pthread_rwlock_wrlock, and pthread_rwlock_unlock, omnitrace has been found to deadlock for some unknown reason
- New configuration variable: OMNITRACE_TRACE_THREAD_RW_LOCKS which defaults to false
[ROCm/rocprofiler-systems commit: e099c84640]
- found when using ROCm-enabled OpenMPI with rocHPL
- when wrapping pthread_rwlock_rdlock, pthread_rwlock_wrlock, and pthread_rwlock_unlock, omnitrace has been found to deadlock for some unknown reason
- New configuration variable: OMNITRACE_TRACE_THREAD_RW_LOCKS which defaults to false
To improve performance on queue preemption, allocate ctx s/r
area in VRAM instead of system memory, and migrate it back
to system memory when VRAM is full.
Signed-off-by: Eric Huang <jinhuieric.huang@amd.com>
Change-Id: If775782027188dbe84b6868260e429373675434c
To improve performance on queue preemption, allocate ctx s/r
area in VRAM instead of system memory, and migrate it back
to system memory when VRAM is full.
Signed-off-by: Eric Huang <jinhuieric.huang@amd.com>
Change-Id: If775782027188dbe84b6868260e429373675434c
[ROCm/ROCR-Runtime commit: 37be876cad]
It is to add new option for always keeping gpu mapping
and bump KFD version for the feature of unified save
restore memory.
Signed-off-by: Eric Huang <jinhuieric.huang@amd.com>
Change-Id: Iebee35e6de4d52fa29f82dd19f6bbf5640249492
It is to add new option for always keeping gpu mapping
and bump KFD version for the feature of unified save
restore memory.
Signed-off-by: Eric Huang <jinhuieric.huang@amd.com>
Change-Id: Iebee35e6de4d52fa29f82dd19f6bbf5640249492
[ROCm/ROCR-Runtime commit: e1d1a6fbb0]
The path variable was used incorrectly and this is causing issue in spack build
Change-Id: I34890577f1403d0d5efcba80d417ad5f9720aa6e
[ROCm/clr commit: d65aaf2f78]
- Otherwise this consumes the caller's stdin by accident.
Patch by:
- Bolo -- Josef Burger
- https://github.com/bigtrak
Change-Id: I669376c025c0cde2cfdd59b5a4a14a71c5d8e862
[ROCm/clr commit: a0ecfd13a9]
- Otherwise this consumes the caller's stdin by accident.
Patch by:
- Bolo -- Josef Burger
- https://github.com/bigtrak
Change-Id: I669376c025c0cde2cfdd59b5a4a14a71c5d8e862
HIP build scripts updated, so that cmake install prefix and cpack packaging prefix doesn't have destination directory hip
The workaround in source code is not at all required.
Depends-On: I6775407b4bfec84b6b911d333f3725c310539bfc
Change-Id: Ic1f5e170690d6dba0fec493143dea390c435cd42
[ROCm/clr commit: db6eaa35aa]
HIP build scripts updated, so that cmake install prefix and cpack packaging prefix doesn't have destination directory hip
The workaround in source code is not at all required.
Depends-On: I6775407b4bfec84b6b911d333f3725c310539bfc
Change-Id: Ic1f5e170690d6dba0fec493143dea390c435cd42
Optimize CUDA graph launch; avoid launching a CPU callback for
intra-node operations.
Simplify kernel common code to improve the latency of send/recv
operations.
Strengthen CUDA streams semantics.
Change NET API to v6, to add dmabuf support.
Add ncclGetLastError() function.
Add ncclRemoteError code and use it for remote network errors.
Support the use of a different NCCL_NET parameter per communicator.
Add support for SHM and P2P transfers using cudaMemcpy.
[ROCm/rccl commit: 19ab67d172]
Optimize CUDA graph launch; avoid launching a CPU callback for
intra-node operations.
Simplify kernel common code to improve the latency of send/recv
operations.
Strengthen CUDA streams semantics.
Change NET API to v6, to add dmabuf support.
Add ncclGetLastError() function.
Add ncclRemoteError code and use it for remote network errors.
Support the use of a different NCCL_NET parameter per communicator.
Add support for SHM and P2P transfers using cudaMemcpy.
Saleel Kudchadker
agunashe
sdashmiz
Chauncey Hui
Jonathan R. Madsen
Eric Huang
hubertlu-tw
Ranjith Ramakrishnan
Konstantin Zhuravlyov
Sarbojit Sarkar
Sylvain Jeaugey
Finlay
Jatin Chaudhary
Anton Mitkov