Updated CHANGELOG to include the performance fix for
kernel launch latency with increasing number of idle streams.
Change-Id: I509e14cb8f8cd3abe61c6ede78808e96ef8f06e1
If a user requests PC sampling on a system that does not support this feature,
report a fatal error message and stop executing the program.
[ROCm/rocprofiler-sdk commit: 0ce75c1043]
NOPTION is meant for component options or alias runtime options so
the option group must not be OA_RUNTIME or OA_MISC_ALIAS must be set,
otherwise we incorrectly assume that it has an option variable and
attempting to write to it causes corruption of OptionVariables.
Change-Id: Iafb5a8f743e5ed0f87be36061c44578178f6cfde
Add support for device counter colleciton ioctl
Adds support for the device counter collection IOCTL. This IOCTL
allows for device wide counters to be collected even if the queue
is not intercepted by rocprofiler-sdk (required for system profilers).
A test is also included which checks this behavior by creating a queue
that does not have profiling enabled on it and checks to see if SQ
counters can be read from it. Note: this test will be skipped if the KFD
version does not contain this IOCTL.
Right now the check is "soft" in that if the IOCTL is present and there
is an error with permissions, rocprofiler will continue but will print
an error stating that system wide device profiling and collected counter
values may be degraded. This is primarily to avoid breaking existing
users (like PAPI) who may not need the IOCTL's capability and to give
them time to update.
Co-authored-by: Benjamin Welton <ben@amd.com>
[ROCm/rocprofiler-sdk commit: c574881cdb]
Add support for device counter colleciton ioctl
Adds support for the device counter collection IOCTL. This IOCTL
allows for device wide counters to be collected even if the queue
is not intercepted by rocprofiler-sdk (required for system profilers).
A test is also included which checks this behavior by creating a queue
that does not have profiling enabled on it and checks to see if SQ
counters can be read from it. Note: this test will be skipped if the KFD
version does not contain this IOCTL.
Right now the check is "soft" in that if the IOCTL is present and there
is an error with permissions, rocprofiler will continue but will print
an error stating that system wide device profiling and collected counter
values may be degraded. This is primarily to avoid breaking existing
users (like PAPI) who may not need the IOCTL's capability and to give
them time to update.
Co-authored-by: Benjamin Welton <ben@amd.com>
- Added script to merge multiprocess output automatically to one file when there are multiprocess proto files written into output folder
- Execute the merge multiprocess script from the rank 0 process
- Added the scripts folder path to env path, via setup-env.sh
- Installed merge_multiprocess_output.sh to /share/rocprofiler-systems/bin dir
Co-authored-by: David Galiffi <David.Galiffi@amd.com>
[ROCm/rocprofiler-systems commit: 0263e951ff]
- Added script to merge multiprocess output automatically to one file when there are multiprocess proto files written into output folder
- Execute the merge multiprocess script from the rank 0 process
- Added the scripts folder path to env path, via setup-env.sh
- Installed merge_multiprocess_output.sh to /share/rocprofiler-systems/bin dir
Co-authored-by: David Galiffi <David.Galiffi@amd.com>
The vector with all kernels is preallocated on the executable init.
Thus, reduce the scope of global lock to the binary creation only.
Change-Id: I73035013a6562175069137e895bba815f466ee35
Enable VCN activity tracing on different instances from the GPU metrics fetched using rsmi_dev_gpu_metrics_info_get in the ROCm-SMI. library.
The tracing can be controlled with ROCPROFSYS_ROCM_SMI_METRICS by setting the value as vcn_activity, Currently this configuration takes the following values: busy, temp, power, mem_usage, vcn_activity.
By default, all the 5 values will be enabled.
Signed-off-by: Sajina P Kandy <Sajina.PuthalathKandy@amd.com>
Co-authored-by: Sajina Kandy <sputhala-amd@amd.com>
[ROCm/rocprofiler-systems commit: 3fa37c991e]
Enable VCN activity tracing on different instances from the GPU metrics fetched using rsmi_dev_gpu_metrics_info_get in the ROCm-SMI. library.
The tracing can be controlled with ROCPROFSYS_ROCM_SMI_METRICS by setting the value as vcn_activity, Currently this configuration takes the following values: busy, temp, power, mem_usage, vcn_activity.
By default, all the 5 values will be enabled.
Signed-off-by: Sajina P Kandy <Sajina.PuthalathKandy@amd.com>
Co-authored-by: Sajina Kandy <sputhala-amd@amd.com>
Indic, Vladimir
Julia Jiang
taosang2
Jaydeep Patel
Ma, Li
stali
Nagaraj, Sriraksha
Greg Scaffidi
Todd tiantuo Li
Ioannis Assiouras
Adam Pryor
Galantsev, Dmitrii
sonadeem
Baraldi, Giovanni
Chris Freehill
Pankaj Kumar Divedi
Welton, Benjamin
David Galiffi
Maisam Arif
Vladana Stojiljkovic
Flora Cui
Pranjal Swarup
German Andryeyev
Sajina PK