Moved the bit_rate and max_bandwidth back into links in the
amdsmi_link_metrics_t struct as this change was impacting
other teams. Modified the C and python API's, wrapper, and
CLI accordingly.
Signed-off-by: Bindhiya Kanangot Balakrishnan <Bindhiya.KanangotBalakrishnan@amd.com>
Moved the bit_rate and max_bandwidth back into links in the
amdsmi_link_metrics_t struct as this change was impacting
other teams. Modified the C and python API's, wrapper, and
CLI accordingly.
Signed-off-by: Bindhiya Kanangot Balakrishnan <Bindhiya.KanangotBalakrishnan@amd.com>
[ROCm/amdsmi commit: 645c313f00]
* Add gpu metrics caching defaulted to 100ms
* AMDSMI_GPU_METRICS_CACHE_MS is used to set the caching rate limits
---------
Signed-off-by: Maisam Arif <Maisam.Arif@amd.com>
* Add gpu metrics caching defaulted to 100ms
* AMDSMI_GPU_METRICS_CACHE_MS is used to set the caching rate limits
---------
Signed-off-by: Maisam Arif <Maisam.Arif@amd.com>
[ROCm/amdsmi commit: 42096c1398]
add hip table version to 13, API_ID_LAST is not changed from version=12 since no new struct has been added.
Co-authored-by: Venkateshwar Reddy Kandula <vkandula@amd.com>
[ROCm/rocprofiler-sdk commit: d2393c97f8]
add hip table version to 13, API_ID_LAST is not changed from version=12 since no new struct has been added.
Co-authored-by: Venkateshwar Reddy Kandula <vkandula@amd.com>
Cache packet creation in all cases to reduce the number of allocations/
destruction operations made down to KFD. There is a bug that we
encounter after a period of runtime in KFD where allocations fail to be
visable to the GPU (suspect this is a FW issue, similar to other FW
issues they have had along the same lines). This sidesteps that issue in
rocprof (and likely should be done regardless)
Co-authored-by: Benjamin Welton <bewelton@amd.com>
[ROCm/rocprofiler-sdk commit: 7e3ea0c58e]
Cache packet creation in all cases to reduce the number of allocations/
destruction operations made down to KFD. There is a bug that we
encounter after a period of runtime in KFD where allocations fail to be
visable to the GPU (suspect this is a FW issue, similar to other FW
issues they have had along the same lines). This sidesteps that issue in
rocprof (and likely should be done regardless)
Co-authored-by: Benjamin Welton <bewelton@amd.com>
A hang would occur when a memory error occurs because the
AQLQueue destructor would be waiting for a signal that
wouldn't come. This change allows it to break out of the
wait loop.
A hang would occur when a memory error occurs because the
AQLQueue destructor would be waiting for a signal that
wouldn't come. This change allows it to break out of the
wait loop.
[ROCm/ROCR-Runtime commit: c065d9a7e2]
Replace ROCm SMI (rsmi) API calls with AMDSMI (amdsmi) API calls
in rocrtst.
Signed-off-by: Alysa Liu <Alysa.Liu@amd.com>
[ROCm/ROCR-Runtime commit: 4fab4d70e6]
Improve support for DirectNIC (CX8)
* Add support for XDR speed detection.
* When DirectNIC is enabled, report only the RDMA interfaces.
Extend the P2C (PXN over C2C) support to send/receive operations.
Support compilation with GCC 14 (Issues #1743, #1751).
Fix the unloading of network plugins that also provide tuner capability.
Fix the change of the current device across the calls to ncclCommDestroy()
and ncclCommAbort().
A note for users on MNNVL systems: please ensure an adequate stack size for
NCCL threads. While the default Linux stack size limit of 8192 KB is known
to be sufficient, we've seen crashes if the limit is changed to
"unlimited", as it causes the glibc library to unexpectedly *decrease* the
stack size of NCCL's background threads to just 2048 KB. Use "ulimit -s"
in bash to print the current limit; if needed, reset it to 8192 KB using
"ulimit -s 8192" (one also needs to ensure that the new setting is
propagated to other nodes when launching a multi-node NCCL job).
darren-amd
Sajina PK
David Galiffi
Fei Zheng
Bindhiya Kanangot Balakrishnan
zichguan-amd
xuchen-amd
Pratik Basyal
Maisam Arif
dependabot[bot]
Wenkai Du
Stojiljkovic, Vladana
Pryor, Adam
Manocha, Rahul
Kandula, Venkateshwar reddy
Sam Wu
Welton, Benjamin
Chris Freehill
Nikhil-Nunna
Jobbins
Xie, Jiabao(Jimbo)
Alysa Liu
Sang, Tao
Kamil Iskra