AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Add single kernel filtering to roofline plots (#757)

Browse Source 
* Add single kernel filtering for roofline
* Add --kernel to documentation
* Add kernel labels to roofline pdfs

Signed-off-by: Carrie Fallows <Carrie.Fallows@amd.com>

* Add test cases

Signed-off-by: Carrie Fallows <Carrie.Fallows@amd.com>

* Add autodetect for mode (profile or analyze) during roof validate and filter
Prevent --kernel from affecting roofline in gui mode- although this may be broken in develop branch anyways

Signed-off-by: Carrie Fallows <Carrie.Fallows@amd.com>

* Add note about roof-only usage checking for existing profiling files in the dir. If roof-only is not provided, rocprof-compute currently assumes it has to profile in full regardless. Will look into this another day.

Signed-off-by: Carrie Fallows <Carrie.Fallows@amd.com>

* Update CHANGELOG.md

Add line in resolved issues section to highlight that kernel filtering is now working for roofline plots

* Apply changes suggested by docs team

Signed-off-by: Carrie Fallows <Carrie.Fallows@amd.com>

* Update projects/rocprofiler-compute/CHANGELOG.md

Co-authored-by: Pratik Basyal <pratik.basyal@amd.com>

---------

Signed-off-by: Carrie Fallows <Carrie.Fallows@amd.com>
Co-authored-by: Pratik Basyal <pratik.basyal@amd.com>
This commit is contained in:
cfallows-amd
2025-08-27 13:41:07 -04:00
committed by GitHub
parent 2e50d88fe6
commit c68ba44e72
5 changed files with 129 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rocprofiler-compute/docs/how-to/profile/mode.rst
+7 -1
View File
@@ -477,7 +477,9 @@ Standalone roofline
Roofline analysis occurs on any profile mode run, provided ``--no-roof`` option is not included.
You don't need to include any additional roofline-specific options for roofline analysis.
If you want to focus only on roofline-specific performance data and reduce the time it takes to profile, you can use the ``--roof-only`` option.
This option limits the profiling to just the roofline performance counters.
This option checks if there is existing profiling data in the workload directory (``pmc_perf.csv`` and ``roofline.csv``):
a) If found, uses the data files with the provided arguments to create another roofline PDF output; otherwise,
b) Profile mode runs but is limited to collecting only roofline performance counters.
Roofline options
----------------
@@ -494,6 +496,10 @@ Roofline options
Allows you to specify a device ID to collect performance data from when
running a roofline benchmark on your system.
``-k``, ``--kernel <kernel-substr>``
Allows for kernel filtering. Usage is equivalent with the current ``rocprof``
utility. See :ref:`profiling-kernel-filtering`.
``--roofline-data-type <datatype>``
Allows you to specify data types that you want plotted in the roofline PDF output(s). Selecting more than one data type will overlay the results onto the same plot. Default: FP32