Sync staging with mainline (#524)
* External CI: rename pipeline to rocprofiler-compute (#463) Signed-off-by: Daniel Su <danielsu@amd.com> * Update webui branding (#459) * Update name and icon for browser tab to rocprofiler-compute. Signed-off-by: xuchen-amd <xuchen@amd.com> * Update name and icon for browser tab to rocprofiler-compute. Signed-off-by: xuchen-amd <xuchen@amd.com> --------- Signed-off-by: xuchen-amd <xuchen@amd.com> * Update branding in documentation (#442) * find/replace Omniperf to ROCm Compute Profiler Signed-off-by: Peter Park <peter.park@amd.com> * update name in Sphinx conf Signed-off-by: Peter Park <peter.park@amd.com> * mv what-is-omniperf.rst -> what-is-rocprof-compute.rst Signed-off-by: Peter Park <peter.park@amd.com> * update Tutorials section Signed-off-by: Peter Park <peter.park@amd.com> * add Omniperf as keyword to Conceptual section for internal search Signed-off-by: Peter Park <peter.park@amd.com> * update Reference section Signed-off-by: Peter Park <peter.park@amd.com> * black fmt conf.py Signed-off-by: Peter Park <peter.park@amd.com> * update profile mode and basic usage subsections Signed-off-by: Peter Park <peter.park@amd.com> * update how to use analyze mode subsection Signed-off-by: Peter Park <peter.park@amd.com> * update install section Signed-off-by: Peter Park <peter.park@amd.com> * fix sphinx warnings Signed-off-by: Peter Park <peter.park@amd.com> * fix cmd line examples in profile/mode.rst Signed-off-by: Peter Park <peter.park@amd.com> * update install decision tree image Signed-off-by: Peter Park <peter.park@amd.com> * fix TOC and index Signed-off-by: Peter Park <peter.park@amd.com> fix weird wording * fix cli text: deriving rocprofiler-compute metrics... Signed-off-by: Peter Park <peter.park@amd.com> * update standalone-gui.rst Signed-off-by: Peter Park <peter.park@amd.com> * restore removed doc updates from #428 Signed-off-by: Peter Park <peter.park@amd.com> * update ref to Omniperf in index.rst Signed-off-by: Peter Park <peter.park@amd.com> * fix grafana connection name to match image Signed-off-by: Peter Park <peter.park@amd.com> * update cmds in tutorials Signed-off-by: Peter Park <peter.park@amd.com> --------- Signed-off-by: Peter Park <peter.park@amd.com> * MI300 roofline enablement in rocprofiler-compute (#470) * MI300 roofline enablement in rocprofiler-compute requirements.txt - running some modules complained about numpy version too new, adding extra requirement that numpy be 1.x pmc_roof_perf.txt - adding TCC_BUBBLE_sum counter to profile soc_gfx940.py soc_gfx941.py soc_gfx942.py - remove console logs reading that roofline is temporarily disabled, uncommenting blocks that check for roofline csv and run roofline post-processing roofline_calc.py - add mi300 to supported soc - add new calculation for hbm_data for MI300 using tcc_bubble_sum, checks if counter > 0 to use - add to a few comments roofline-ubuntu-20_04-mi300-rocm6 - binary for the ubuntu systems to enable mi300 roofline calculations from rocm-amdgpu-bench Note- other distros will get roofline bins to enable mi300, but need to be further tested before putting into branch. Signed-off-by: Carrie Fallows <carrie.fallows@amd.com> * Reformatting roofline_calc.py Signed-off-by: Carrie Fallows <carrie.fallows@amd.com> --------- Signed-off-by: Carrie Fallows <carrie.fallows@amd.com> * Update Python format checker (#471) * Add pre commit hook for Python formatting Signed-off-by: coleramos425 <colramos@amd.com> * Update formatting workflow to run on latest Python and add isort formatter Signed-off-by: coleramos425 <colramos@amd.com> * Fix caught yaml formatting issues * Update pyproject file * Add pre-commit hook instruction to CONTRIBUTING guide * Remove target-version from black pyproject.toml * Fixed formatting errors found with black and isort Signed-off-by: David Galiffi <David.Galiffi@amd.com> * Run hook: Whitespaces, fix end of file spaces --------- Signed-off-by: coleramos425 <colramos@amd.com> Signed-off-by: David Galiffi <David.Galiffi@amd.com> Co-authored-by: David Galiffi <David.Galiffi@amd.com> * Bump cryptography from 43.0.0 to 43.0.1 in /docs/sphinx (#473) Bumps [cryptography](https://github.com/pyca/cryptography) from 43.0.0 to 43.0.1. - [Changelog](https://github.com/pyca/cryptography/blob/main/CHANGELOG.rst) - [Commits](https://github.com/pyca/cryptography/compare/43.0.0...43.0.1) --- updated-dependencies: - dependency-name: cryptography dependency-type: indirect ... Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> * Fix file permission on MI300 roofline binary (#477) Signed-off-by: David Galiffi <David.Galiffi@amd.com> * Removing numpy requirements of <2 (#478) Checks are failing if version too high and no need for lower version Signed-off-by: Carrie Fallows <Carrie.Fallows@amd.com> * Fix crash when loading web UI roofline for gfx942 (#479) * Fix crash when loading web UI roofline for gfx942 * Fix formatting Signed-off-by: benrichard-amd <ben.richard@amd.com> * Make same changs for gfx940, gfx942. Signed-off-by: benrichard-amd <ben.richard@amd.com> * Fix formatting in soc_gfx940 and soc_gfx941. Signed-off-by: benrichard-amd <ben.richard@amd.com> --------- Signed-off-by: benrichard-amd <ben.richard@amd.com> * Rebranding name change patch (#469) * Patch in missed name change for rebranding. Signed-off-by: xuchen-amd <xuchen@amd.com> * Patch in missed name change for rebranding. Signed-off-by: xuchen-amd <xuchen@amd.com> --------- Signed-off-by: xuchen-amd <xuchen@amd.com> * Move dependabot.yml to .github/ and bump rocm-docs-core (#481) * Move dependabot.yml to .github/ * Bump rocm-docs-core to 1.8.5 * Bump rocm-docs-core to 1.9.0 * Fix packaging for upgrading (#486) Specify that "rocprofiler-compute" replaces / obsoletes the "omniperf" package. * Renamed extension path from omniperf to rocprofiler_compute (#487) Signed-off-by: Tim Gu <Tim.Gu@amd.com> * MI300 rhel and sles roofline binaries (#480) * Roofline bins for MI300 on rhel and sles distributions Built from rocm-amdgpu-bench, tested on respective distro systems with MI300 hardware. Signed-off-by: Carrie Fallows <Carrie.Fallows@amd.com> * Minor modifications removing hardcoded variables in roofline files. Signed-off-by: Carrie Fallows <Carrie.Fallows@amd.com> --------- Signed-off-by: Carrie Fallows <Carrie.Fallows@amd.com> * Modify test_profile_general.py ctest to include MI300 enablement (#498) Signed-off-by: Carrie Fallows <Carrie.Fallows@amd.com> * part 1 to support rocprofv3 (#492) * rocprofv3 support initial commit -Can run rocprofv3 but ultimately fails. rocprofv3 says the counter capacity is exceeded and the output CSV file format is different from v1/v2. * Add rocprofv3 detection so v2 can still be used It's hacky but it'll do for now. * Add code path to convert rocprofv3 JSON output into CSV * Grab correct value for Queue ID * Use _sum suffix to sum TCC counters Previously we were specifying each channel for TCC counters. rocprofv3 does not support specifing each TCC channel, and instead will auto sum given the TCC counter name. The counter name with the _sum suffix is also supported and is also supported in v1 and v2. So we will use the TCC counter name with the _sum suffix. * Fix incorrect counter outputs when using rocprofv3 In the JSON output some counters appear multime times and must be summed to get the correct value. These summed values match the rocprofv3 output in CSV mode and also match the rocprofv2 output. * Remove duplicate Correlation_ID and Wave_Size in output * Handle json output that does not contain any dispatches Omniperf was assuming each JSON output from rocprofv3 would always contain dispatches. This is not the case. For example, in a multi-process workload where one of the processes does not dispatch any kernels. A JSON file will still be output for this process but it will not contain any dispatches. * Code cleanup * Update search path for rocprofv3 results Rocprofv3 was updated to include the hostname in the path where it outputs results. * Handle accumulate counters In v1/v2 rocprof uses the SQ_ACCUM_PREV_HIRES counter for the accumualte counters. v3 does not have this. So we need to define our own counters in counter_defs.yaml. For this we use the counter name + _ACCUM, for example SQ_INSTR_LEVEL_SMEM_ACCUM. To use rocprofv3 you will need to update counter_defs.yaml to include these new counter definitions. * Use correct GPU ID When converting JSON -> CSV we were assigning node_id to GPU_ID. Since the JSON contains non-GPU devices, the node_id for GPUs might not start at 0 as expected. This commit maps the agent ID to the appropriate GPU ID. * Parse scratch memory per work item from JSON * Support rocprofv3 CSV parsing JSON decoding is very slow for large files. Include support for parsing rocprofv3 CSV output and make that the default. CSV/JSON can be toggled via the ROCPROF_OUTPUT_FORMAT environment variable e.g. ROCPROF_OUTPUT_FORMAT=csv or ROCPROF_OUTPUT_FORMAT=json * black format after merge * format isort * change return of rocprof_cmd to try to resolve test's error * hack to pick last part of rocminfo's name * debug log of hacks * Modify test_profile_general.py ctest to include MI300 enablement. Currently failing because of explicitly excluded roofline files for the soc and autofailed asserts for roof-only tests- originally in place because roofline was not enabled on mi300 yet. Signed-off-by: Carrie Fallows <Carrie.Fallows@amd.com> * black and isort formated * corrected line of copyright --------- Signed-off-by: Carrie Fallows <Carrie.Fallows@amd.com> Co-authored-by: benrichard-amd <ben.richard@amd.com> Co-authored-by: YANG WANG <ywang@ywang-ubuntu.amd.com> Co-authored-by: Carrie Fallows <Carrie.Fallows@amd.com> * fix for crash of timestamp of part 1 for rocprofv3 (#499) * fix the error caused by ignoring the lack of counter csv file from rocprofv3 for timestamp * isort and black formated * quick fix for gfx906 roofline (#505) * Multi node support (#503) * [CTest] Pipeline failures for MI300 (#483) * Propagate new chip_id logic to testing workflow Signed-off-by: coleramos425 <colramos@amd.com> * Add a debug line to tests Signed-off-by: coleramos425 <colramos@amd.com> * Trying to set rocprofv2 generally in CTest module Signed-off-by: coleramos425 <colramos@amd.com> * Remove temp debugging lines from CI Signed-off-by: coleramos425 <colramos@amd.com> * Add roofline entry for MI300 expected files in CI tests Signed-off-by: coleramos425 <colramos@amd.com> * Make num_devices modifier global in scope Signed-off-by: coleramos425 <colramos@amd.com> * Change kernel name in PyTest to confirm rocprofv2 bug Related to https://ontrack-internal.amd.com/browse/SWDEV-503453 Signed-off-by: coleramos425 <colramos@amd.com> --------- Signed-off-by: coleramos425 <colramos@amd.com> * Spatial-multiplexing: part 1 profiling stage (#465) * rocprofv3 support initial commit -Can run rocprofv3 but ultimately fails. rocprofv3 says the counter capacity is exceeded and the output CSV file format is different from v1/v2. * Add rocprofv3 detection so v2 can still be used It's hacky but it'll do for now. * Add code path to convert rocprofv3 JSON output into CSV * Grab correct value for Queue ID * Use _sum suffix to sum TCC counters Previously we were specifying each channel for TCC counters. rocprofv3 does not support specifing each TCC channel, and instead will auto sum given the TCC counter name. The counter name with the _sum suffix is also supported and is also supported in v1 and v2. So we will use the TCC counter name with the _sum suffix. * Fix incorrect counter outputs when using rocprofv3 In the JSON output some counters appear multime times and must be summed to get the correct value. These summed values match the rocprofv3 output in CSV mode and also match the rocprofv2 output. * Remove duplicate Correlation_ID and Wave_Size in output * Handle json output that does not contain any dispatches Omniperf was assuming each JSON output from rocprofv3 would always contain dispatches. This is not the case. For example, in a multi-process workload where one of the processes does not dispatch any kernels. A JSON file will still be output for this process but it will not contain any dispatches. * Code cleanup * Update search path for rocprofv3 results Rocprofv3 was updated to include the hostname in the path where it outputs results. * Handle accumulate counters In v1/v2 rocprof uses the SQ_ACCUM_PREV_HIRES counter for the accumualte counters. v3 does not have this. So we need to define our own counters in counter_defs.yaml. For this we use the counter name + _ACCUM, for example SQ_INSTR_LEVEL_SMEM_ACCUM. To use rocprofv3 you will need to update counter_defs.yaml to include these new counter definitions. * debug code * add logic code for multiplexing * minor fix * more fixes * rocprofv3 support initial commit -Can run rocprofv3 but ultimately fails. rocprofv3 says the counter capacity is exceeded and the output CSV file format is different from v1/v2. * Add rocprofv3 detection so v2 can still be used It's hacky but it'll do for now. * Add code path to convert rocprofv3 JSON output into CSV * Grab correct value for Queue ID * Use _sum suffix to sum TCC counters Previously we were specifying each channel for TCC counters. rocprofv3 does not support specifing each TCC channel, and instead will auto sum given the TCC counter name. The counter name with the _sum suffix is also supported and is also supported in v1 and v2. So we will use the TCC counter name with the _sum suffix. * Fix incorrect counter outputs when using rocprofv3 In the JSON output some counters appear multime times and must be summed to get the correct value. These summed values match the rocprofv3 output in CSV mode and also match the rocprofv2 output. * Remove duplicate Correlation_ID and Wave_Size in output * Handle json output that does not contain any dispatches Omniperf was assuming each JSON output from rocprofv3 would always contain dispatches. This is not the case. For example, in a multi-process workload where one of the processes does not dispatch any kernels. A JSON file will still be output for this process but it will not contain any dispatches. * Code cleanup * Update search path for rocprofv3 results Rocprofv3 was updated to include the hostname in the path where it outputs results. * Handle accumulate counters In v1/v2 rocprof uses the SQ_ACCUM_PREV_HIRES counter for the accumualte counters. v3 does not have this. So we need to define our own counters in counter_defs.yaml. For this we use the counter name + _ACCUM, for example SQ_INSTR_LEVEL_SMEM_ACCUM. To use rocprofv3 you will need to update counter_defs.yaml to include these new counter definitions. * count accu files as well * Use correct GPU ID When converting JSON -> CSV we were assigning node_id to GPU_ID. Since the JSON contains non-GPU devices, the node_id for GPUs might not start at 0 as expected. This commit maps the agent ID to the appropriate GPU ID. * fix error with csv file parse from json and merge during post-processing * implemented parsing of csv files from v3 output for optimization * Parse scratch memory per work item from JSON * Support rocprofv3 CSV parsing JSON decoding is very slow for large files. Include support for parsing rocprofv3 CSV output and make that the default. CSV/JSON can be toggled via the ROCPROF_OUTPUT_FORMAT environment variable e.g. ROCPROF_OUTPUT_FORMAT=csv or ROCPROF_OUTPUT_FORMAT=json * black format after merge * format isort * change return of rocprof_cmd to try to resolve test's error * hack to pick last part of rocminfo's name * debug log of hacks * Modify test_profile_general.py ctest to include MI300 enablement. Currently failing because of explicitly excluded roofline files for the soc and autofailed asserts for roof-only tests- originally in place because roofline was not enabled on mi300 yet. Signed-off-by: Carrie Fallows <Carrie.Fallows@amd.com> * black and isort formated * formated by isort and black * change default rocprof's output to csv * repaired crash caused by missing csv counter file when running for timestamp * change name to spatial-multiplexing from multiplexing * make necessary modification for review * set the value of spatial_multiplexing in argument defautly to None * repair the part that blocks regular pmc files' generation --------- Signed-off-by: Carrie Fallows <Carrie.Fallows@amd.com> Co-authored-by: benrichard-amd <ben.richard@amd.com> Co-authored-by: fei.zheng <fei.zheng@amd.com> Co-authored-by: YANG WANG <ywang@ywang-ubuntu.amd.com> Co-authored-by: Carrie Fallows <Carrie.Fallows@amd.com> * Simple fix for gpu model value. (#508) Signed-off-by: xuchen-amd <xuchen@amd.com> * Add FP64 to plot adhering to pdf name (#507) * Replacing FP32-only plot with an FP32&FP64 combo plot. Results will likely be negligible but the plot name indicates both should be graphed. Signed-off-by: Carrie Fallows <Carrie.Fallows@amd.com> * Remove duplicate AI plot to clean up fp32 fp64 graph Signed-off-by: Carrie Fallows <Carrie.Fallows@amd.com> --------- Signed-off-by: Carrie Fallows <Carrie.Fallows@amd.com> * Add gpu series for roofline (#510) * Add gpu_series for roofline. * Use gpu_series in path names for roofline. * Fix TCC on MI200 when introduce rocprofv3 (#509) * quick fix for v2 * one more fix * revert a bit --------- Co-authored-by: ywang103-amd <ywang103@amd.com> * Bump rocm-docs-core from 1.9.0 to 1.12.0 in /docs/sphinx (#511) Bumps [rocm-docs-core](https://github.com/ROCm/rocm-docs-core) from 1.9.0 to 1.12.0. - [Release notes](https://github.com/ROCm/rocm-docs-core/releases) - [Changelog](https://github.com/ROCm/rocm-docs-core/blob/develop/CHANGELOG.md) - [Commits](https://github.com/ROCm/rocm-docs-core/compare/v1.9.0...v1.12.0) --- updated-dependencies: - dependency-name: rocm-docs-core dependency-type: direct:production update-type: version-update:semver-minor ... Signed-off-by: dependabot[bot] <support@github.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> * Update sample roofline plot img (#516) * Modify path to use gpu_model instead of gpu_series to match other workload directory path creation/search points. Affects manual testing, does not seem to affect ctests. (#513) Signed-off-by: Carrie Fallows <Carrie.Fallows@amd.com> * Improve formatting when displaying rocprof command. (#476) * Improve formatting when displaying rocprof command. Signed-off-by: xuchen-amd <xuchen@amd.com> * Fix python formatting. Signed-off-by: xuchen-amd <xuchen@amd.com> * Strip unwanted characters (rocprofv1 specific) from rocprof commands. Signed-off-by: xuchen-amd <xuchen@amd.com> * Strip unwanted characters (rocprofv1 specific) from rocprof commands. Signed-off-by: xuchen-amd <xuchen@amd.com> * Save the unmodified arguments for rocprof for debug message display. Signed-off-by: xuchen-amd <xuchen@amd.com> --------- Signed-off-by: xuchen-amd <xuchen@amd.com> * quick fix for mpi_support (#518) * Pass accumulate counters to rocprofv3 using -E option (#522) rocprofv3 has a new -E option where extra counters can be passed (see accum_counters.yaml) instead of defining them in counter_defs.yaml. * Unify all file handling with pathlib (#512) * Replace occurences of os.path functions with equivalent functions from pathlib library * Remove unwanted imports of os.path and os * Add coding guidelines for using pathlib instead of os.path * Auto sync staging and mainline on a weekly cadence (#517) Signed-off-by: coleramos425 <colramos@amd.com> --------- Signed-off-by: Daniel Su <danielsu@amd.com> Signed-off-by: xuchen-amd <xuchen@amd.com> Signed-off-by: Peter Park <peter.park@amd.com> Signed-off-by: Carrie Fallows <carrie.fallows@amd.com> Signed-off-by: coleramos425 <colramos@amd.com> Signed-off-by: David Galiffi <David.Galiffi@amd.com> Signed-off-by: dependabot[bot] <support@github.com> Signed-off-by: Carrie Fallows <Carrie.Fallows@amd.com> Signed-off-by: benrichard-amd <ben.richard@amd.com> Signed-off-by: Tim Gu <Tim.Gu@amd.com> Co-authored-by: Daniel Su <danielsu@amd.com> Co-authored-by: xuchen-amd <xuchen@amd.com> Co-authored-by: Peter Park <peter.park@amd.com> Co-authored-by: cfallows-amd <Carrie.Fallows@amd.com> Co-authored-by: David Galiffi <David.Galiffi@amd.com> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com> Co-authored-by: Ben Richard <143630488+benrichard-amd@users.noreply.github.com> Co-authored-by: Tim Gu <Tim.Gu@amd.com> Co-authored-by: ywang103-amd <ywang103@amd.com> Co-authored-by: benrichard-amd <ben.richard@amd.com> Co-authored-by: YANG WANG <ywang@ywang-ubuntu.amd.com> Co-authored-by: Fei Zheng <44449748+feizheng10@users.noreply.github.com> Co-authored-by: fei.zheng <fei.zheng@amd.com> Co-authored-by: vedithal-amd <Vignesh.Edithal@amd.com>
This commit is contained in:
Cole Ramos
gecommit door
GitHub
GitHub
bovenliggende
71bc463921
commit
272e5b6e32
@@ -5,7 +5,7 @@ Infinity Fabric transactions
|
||||
|
||||
For this example, consider the
|
||||
:dev-sample:`Infinity Fabric™ sample <fabric.hip>` distributed as a part of
|
||||
Omniperf.
|
||||
ROCm Compute Profiler.
|
||||
|
||||
This following code snippet launches a simple read-only kernel.
|
||||
|
||||
@@ -36,7 +36,7 @@ is identically false -- and thus we expect no writes.
|
||||
|
||||
.. note::
|
||||
|
||||
The actual sample included with Omniperf also includes the ability to select
|
||||
The actual sample included with ROCm Compute Profiler also includes the ability to select
|
||||
different operation types (such as atomics, writes). This abbreviated version
|
||||
is presented here for reference only.
|
||||
|
||||
@@ -44,13 +44,13 @@ Finally, this sample code lets the user control the
|
||||
:ref:`granularity of an allocation <memory-type>`, the owner of an allocation
|
||||
(local HBM, CPU DRAM or remote HBM), and the size of an allocation (the default
|
||||
is :math:`\sim4`\ GiB) via command line arguments. In doing so, we can explore
|
||||
the impact of these parameters on the L2-Fabric metrics reported by Omniperf to
|
||||
the impact of these parameters on the L2-Fabric metrics reported by ROCm Compute Profiler to
|
||||
further understand their meaning.
|
||||
|
||||
.. note::
|
||||
|
||||
All results in this section were generated an a node of Infinity
|
||||
Fabric connected MI250 accelerators using ROCm version 5.6.0, and Omniperf
|
||||
Fabric connected MI250 accelerators using ROCm version 5.6.0, and ROCm Compute Profiler
|
||||
version 2.0.0. Although results may vary with ROCm versions and accelerator
|
||||
connectivity, we expect the lessons learned here to be broadly applicable.
|
||||
|
||||
@@ -64,7 +64,7 @@ In our first experiment, we consider the simplest possible case, a
|
||||
|
||||
.. code-block:: shell-session
|
||||
|
||||
$ omniperf profile -n coarse_grained_local --no-roof -- ./fabric -t 1 -o 0
|
||||
$ rocprof-compute profile -n coarse_grained_local --no-roof -- ./fabric -t 1 -o 0
|
||||
Using:
|
||||
mtype:CoarseGrained
|
||||
mowner:Device
|
||||
@@ -73,7 +73,7 @@ In our first experiment, we consider the simplest possible case, a
|
||||
mdata:Unsigned
|
||||
remoteId:-1
|
||||
<...>
|
||||
$ omniperf analyze -p workloads/coarse_grained_local/mi200 -b 17.2.0 17.2.1 17.2.2 17.4.0 17.4.1 17.4.2 17.5.0 17.5.1 17.5.2 17.5.3 17.5.4 -n per_kernel --dispatch 2
|
||||
$ rocprof-compute analyze -p workloads/coarse_grained_local/mi200 -b 17.2.0 17.2.1 17.2.2 17.4.0 17.4.1 17.4.2 17.5.0 17.5.1 17.5.2 17.5.3 17.5.4 -n per_kernel --dispatch 2
|
||||
<...>
|
||||
17. L2 Cache
|
||||
17.2 L2 - Fabric Transactions
|
||||
@@ -163,7 +163,7 @@ accelerator. Our code uses the ``hipExtMallocWithFlag`` API with the
|
||||
|
||||
.. code-block:: shell-session
|
||||
|
||||
$ omniperf profile -n fine_grained_local --no-roof -- ./fabric -t 0 -o 0
|
||||
$ rocprof-compute profile -n fine_grained_local --no-roof -- ./fabric -t 0 -o 0
|
||||
Using:
|
||||
mtype:FineGrained
|
||||
mowner:Device
|
||||
@@ -172,7 +172,7 @@ accelerator. Our code uses the ``hipExtMallocWithFlag`` API with the
|
||||
mdata:Unsigned
|
||||
remoteId:-1
|
||||
<...>
|
||||
$ omniperf analyze -p workloads/fine_grained_local/mi200 -b 17.2.0 17.2.1 17.2.2 17.2.3 17.4.0 17.4.1 17.4.2 17.5.0 17.5.1 17.5.2 17.5.3 17.5.4 -n per_kernel --dispatch 2
|
||||
$ rocprof-compute analyze -p workloads/fine_grained_local/mi200 -b 17.2.0 17.2.1 17.2.2 17.2.3 17.4.0 17.4.1 17.4.2 17.5.0 17.5.1 17.5.2 17.5.3 17.5.4 -n per_kernel --dispatch 2
|
||||
<...>
|
||||
17. L2 Cache
|
||||
17.2 L2 - Fabric Transactions
|
||||
@@ -245,7 +245,7 @@ substantial change in the L2-Fabric metrics:
|
||||
|
||||
.. code-block:: shell-session
|
||||
|
||||
$ omniperf profile -n fine_grained_remote --no-roof -- ./fabric -t 0 -o 2
|
||||
$ rocprof-compute profile -n fine_grained_remote --no-roof -- ./fabric -t 0 -o 2
|
||||
Using:
|
||||
mtype:FineGrained
|
||||
mowner:Remote
|
||||
@@ -254,7 +254,7 @@ substantial change in the L2-Fabric metrics:
|
||||
mdata:Unsigned
|
||||
remoteId:-1
|
||||
<...>
|
||||
$ omniperf analyze -p workloads/fine_grained_remote/mi200 -b 17.2.0 17.2.1 17.2.2 17.2.3 17.4.0 17.4.1 17.4.2 17.5.0 17.5.1 17.5.2 17.5.3 17.5.4 -n per_kernel --dispatch 2
|
||||
$ rocprof-compute analyze -p workloads/fine_grained_remote/mi200 -b 17.2.0 17.2.1 17.2.2 17.2.3 17.4.0 17.4.1 17.4.2 17.5.0 17.5.1 17.5.2 17.5.3 17.5.4 -n per_kernel --dispatch 2
|
||||
<...>
|
||||
17. L2 Cache
|
||||
17.2 L2 - Fabric Transactions
|
||||
@@ -339,7 +339,7 @@ fine-grained memory using the ``hipHostMalloc`` API:
|
||||
|
||||
.. code-block:: shell-session
|
||||
|
||||
$ omniperf profile -n fine_grained_host --no-roof -- ./fabric -t 0 -o 1
|
||||
$ rocprof-compute profile -n fine_grained_host --no-roof -- ./fabric -t 0 -o 1
|
||||
Using:
|
||||
mtype:FineGrained
|
||||
mowner:Host
|
||||
@@ -348,7 +348,7 @@ fine-grained memory using the ``hipHostMalloc`` API:
|
||||
mdata:Unsigned
|
||||
remoteId:-1
|
||||
<...>
|
||||
$ omniperf analyze -p workloads/fine_grained_host/mi200 -b 17.2.0 17.2.1 17.2.2 17.2.3 17.4.0 17.4.1 17.4.2 17.5.0 17.5.1 17.5.2 17.5.3 17.5.4 -n per_kernel --dispatch 2
|
||||
$ rocprof-compute analyze -p workloads/fine_grained_host/mi200 -b 17.2.0 17.2.1 17.2.2 17.2.3 17.4.0 17.4.1 17.4.2 17.5.0 17.5.1 17.5.2 17.5.3 17.5.4 -n per_kernel --dispatch 2
|
||||
<...>
|
||||
17. L2 Cache
|
||||
17.2 L2 - Fabric Transactions
|
||||
@@ -416,7 +416,7 @@ allocation as coarse-grained:
|
||||
|
||||
.. code-block:: shell-session
|
||||
|
||||
$ omniperf profile -n coarse_grained_host --no-roof -- ./fabric -t 1 -o 1
|
||||
$ rocprof-compute profile -n coarse_grained_host --no-roof -- ./fabric -t 1 -o 1
|
||||
Using:
|
||||
mtype:CoarseGrained
|
||||
mowner:Host
|
||||
@@ -425,7 +425,7 @@ allocation as coarse-grained:
|
||||
mdata:Unsigned
|
||||
remoteId:-1
|
||||
<...>
|
||||
$ omniperf analyze -p workloads/coarse_grained_host/mi200 -b 17.2.0 17.2.1 17.2.2 17.2.3 17.4.0 17.4.1 17.4.2 17.5.0 17.5.1 17.5.2 17.5.3 17.5.4 -n per_kernel --dispatch 2
|
||||
$ rocprof-compute analyze -p workloads/coarse_grained_host/mi200 -b 17.2.0 17.2.1 17.2.2 17.2.3 17.4.0 17.4.1 17.4.2 17.5.0 17.5.1 17.5.2 17.5.3 17.5.4 -n per_kernel --dispatch 2
|
||||
<...>
|
||||
17. L2 Cache
|
||||
17.2 L2 - Fabric Transactions
|
||||
@@ -484,7 +484,7 @@ operations to fine-grained memory allocated on the host:
|
||||
|
||||
.. code-block:: shell-session
|
||||
|
||||
$ omniperf profile -n fine_grained_host_write --no-roof -- ./fabric -t 0 -o 1 -p 1
|
||||
$ rocprof-compute profile -n fine_grained_host_write --no-roof -- ./fabric -t 0 -o 1 -p 1
|
||||
Using:
|
||||
mtype:FineGrained
|
||||
mowner:Host
|
||||
@@ -493,7 +493,7 @@ operations to fine-grained memory allocated on the host:
|
||||
mdata:Unsigned
|
||||
remoteId:-1
|
||||
<...>
|
||||
$ omniperf analyze -p workloads/fine_grained_host_writes/mi200 -b 17.2.4 17.2.5 17.2.6 17.2.7 17.2.8 17.4.3 17.4.4 17.4.5 17.4.6 17.5.5 17.5.6 17.5.7 17.5.8 17.5.9 17.5.10 -n per_kernel --dispatch 2
|
||||
$ rocprof-compute analyze -p workloads/fine_grained_host_writes/mi200 -b 17.2.4 17.2.5 17.2.6 17.2.7 17.2.8 17.4.3 17.4.4 17.4.5 17.4.6 17.5.5 17.5.6 17.5.7 17.5.8 17.5.9 17.5.10 -n per_kernel --dispatch 2
|
||||
<...>
|
||||
17. L2 Cache
|
||||
17.2 L2 - Fabric Transactions
|
||||
@@ -576,7 +576,7 @@ operations to the CPU’s DRAM.
|
||||
|
||||
.. code-block:: shell-session
|
||||
|
||||
$ omniperf profile -n fine_grained_host_add --no-roof -- ./fabric -t 0 -o 1 -p 2
|
||||
$ rocprof-compute profile -n fine_grained_host_add --no-roof -- ./fabric -t 0 -o 1 -p 2
|
||||
Using:
|
||||
mtype:FineGrained
|
||||
mowner:Host
|
||||
@@ -585,7 +585,7 @@ operations to the CPU’s DRAM.
|
||||
mdata:Unsigned
|
||||
remoteId:-1
|
||||
<...>
|
||||
$ omniperf analyze -p workloads/fine_grained_host_add/mi200 -b 17.2.4 17.2.5 17.2.6 17.2.7 17.2.8 17.4.3 17.4.4 17.4.5 17.4.6 17.5.5 17.5.6 17.5.7 17.5.8 17.5.9 17.5.10 -n per_kernel --dispatch 2
|
||||
$ rocprof-compute analyze -p workloads/fine_grained_host_add/mi200 -b 17.2.4 17.2.5 17.2.6 17.2.7 17.2.8 17.4.3 17.4.4 17.4.5 17.4.6 17.5.5 17.5.6 17.5.7 17.5.8 17.5.9 17.5.10 -n per_kernel --dispatch 2
|
||||
<...>
|
||||
17. L2 Cache
|
||||
17.2 L2 - Fabric Transactions
|
||||
|
||||
@@ -5,7 +5,7 @@ Instructions-per-cycle and utilizations example
|
||||
|
||||
For this example, consider the
|
||||
:dev-sample:`instructions-per-cycle (IPC) example <ipc.hip>` included with
|
||||
Omniperf.
|
||||
ROCm Compute Profiler.
|
||||
|
||||
This example is compiled using ``c++17`` support:
|
||||
|
||||
@@ -17,7 +17,7 @@ and was run on an MI250 CDNA2 accelerator:
|
||||
|
||||
.. code-block:: shell
|
||||
|
||||
$ omniperf profile -n ipc --no-roof -- ./ipc
|
||||
$ rocprof-compute profile -n ipc --no-roof -- ./ipc
|
||||
|
||||
The results shown in this section are *generally* applicable to CDNA
|
||||
accelerators, but may vary between generations and specific products.
|
||||
@@ -64,11 +64,11 @@ operation, i.e., a ``v_mov_b32`` instruction, e.g.:
|
||||
|
||||
This instruction simply copies the contents from the source register
|
||||
(``v1``) to the destination register (``v0``). Investigating this kernel
|
||||
with Omniperf, we see:
|
||||
with ROCm Compute Profiler, we see:
|
||||
|
||||
.. code-block:: shell-session
|
||||
|
||||
$ omniperf analyze -p workloads/ipc/mi200/ --dispatch 7 -b 11.2
|
||||
$ rocprof-compute analyze -p workloads/ipc/mi200/ --dispatch 7 -b 11.2
|
||||
<...>
|
||||
--------------------------------------------------------------------------------
|
||||
0. Top Stat
|
||||
@@ -172,7 +172,7 @@ in the IPC example:
|
||||
|
||||
.. code-block:: shell
|
||||
|
||||
$ omniperf analyze -p workloads/ipc/mi200/ --dispatch 8 -b 11.2 --decimal 4
|
||||
$ rocprof-compute analyze -p workloads/ipc/mi200/ --dispatch 8 -b 11.2 --decimal 4
|
||||
<...>
|
||||
--------------------------------------------------------------------------------
|
||||
0. Top Stat
|
||||
@@ -240,7 +240,7 @@ instructions executed over the total
|
||||
There are further complications of the Issued IPC metric (**11.2.1**) that make
|
||||
its use more complicated. We will be explore that in the
|
||||
:ref:`following section <ipc-internal-instructions>`. For these reasons,
|
||||
Omniperf typically promotes use of the regular IPC metric (**11.2.0**), e.g., in
|
||||
ROCm Compute Profiler typically promotes use of the regular IPC metric (**11.2.0**), e.g., in
|
||||
the top-level Speed-of-Light chart.
|
||||
|
||||
.. _ipc-internal-instructions:
|
||||
@@ -261,11 +261,11 @@ Here we choose to use the following no-op to make our point:
|
||||
|
||||
s_nop 0x0
|
||||
|
||||
Running this kernel through Omniperf yields:
|
||||
Running this kernel through ROCm Compute Profiler yields:
|
||||
|
||||
.. code-block:: shell-session
|
||||
|
||||
$ omniperf analyze -p workloads/ipc/mi200/ --dispatch 9 -b 11.2
|
||||
$ rocprof-compute analyze -p workloads/ipc/mi200/ --dispatch 9 -b 11.2
|
||||
<...>
|
||||
--------------------------------------------------------------------------------
|
||||
0. Top Stat
|
||||
@@ -362,11 +362,11 @@ operation, for instance:
|
||||
|
||||
which, in analogue to our :ref:`v_mov <ipc-valu-utilization>` example, copies the
|
||||
contents of the source scalar register (``s1``) to the destination
|
||||
scalar register (``s0``). Running this kernel through Omniperf yields:
|
||||
scalar register (``s0``). Running this kernel through ROCm Compute Profiler yields:
|
||||
|
||||
.. code-block:: shell-session
|
||||
|
||||
$ omniperf analyze -p workloads/ipc/mi200/ --dispatch 10 -b 11.2
|
||||
$ rocprof-compute analyze -p workloads/ipc/mi200/ --dispatch 10 -b 11.2
|
||||
<...>
|
||||
--------------------------------------------------------------------------------
|
||||
0. Top Stat
|
||||
@@ -426,11 +426,11 @@ of our :ref:`v_mov <ipc-valu-utilization>` example:
|
||||
|
||||
That is, we wrap our :ref:`VALU <desc-valu>` operation inside a conditional
|
||||
where only one lane in our wavefront is active. Running this kernel
|
||||
through Omniperf yields:
|
||||
through ROCm Compute Profiler yields:
|
||||
|
||||
.. code-block:: shell-session
|
||||
|
||||
$ omniperf analyze -p workloads/ipc/mi200/ --dispatch 11 -b 11.2
|
||||
$ rocprof-compute analyze -p workloads/ipc/mi200/ --dispatch 11 -b 11.2
|
||||
<...>
|
||||
--------------------------------------------------------------------------------
|
||||
0. Top Stat
|
||||
|
||||
@@ -4,22 +4,22 @@ LDS examples
|
||||
============
|
||||
|
||||
For this example, consider the
|
||||
:dev-sample:`LDS sample <lds.hip>` distributed as a part of Omniperf. This
|
||||
:dev-sample:`LDS sample <lds.hip>` distributed as a part of ROCm Compute Profiler. This
|
||||
code contains two kernels to explore how both :doc:`LDS </conceptual/local-data-share>` bandwidth and
|
||||
bank conflicts are calculated in Omniperf.
|
||||
bank conflicts are calculated in ROCm Compute Profiler.
|
||||
|
||||
This example was compiled and run on an MI250 accelerator using ROCm
|
||||
v5.6.0, and Omniperf v2.0.0.
|
||||
v5.6.0, and ROCm Compute Profiler v2.0.0.
|
||||
|
||||
.. code-block:: shell-session
|
||||
|
||||
$ hipcc -O3 lds.hip -o lds
|
||||
|
||||
Finally, we generate our ``omniperf profile`` as:
|
||||
Finally, we generate our ``rocprof-compute profile`` as:
|
||||
|
||||
.. code-block:: shell-session
|
||||
|
||||
$ omniperf profile -n lds --no-roof -- ./lds
|
||||
$ rocprof-compute profile -n lds --no-roof -- ./lds
|
||||
|
||||
.. _lds-bandwidth:
|
||||
|
||||
@@ -71,7 +71,7 @@ Next, let’s analyze the first of our bandwidth kernel dispatches:
|
||||
|
||||
.. code-block:: shell
|
||||
|
||||
$ omniperf analyze -p workloads/lds/mi200/ -b 12.2.1 --dispatch 0 -n per_kernel
|
||||
$ rocprof-compute analyze -p workloads/lds/mi200/ -b 12.2.1 --dispatch 0 -n per_kernel
|
||||
<...>
|
||||
12. Local Data Share (LDS)
|
||||
12.2 LDS Stats
|
||||
@@ -93,7 +93,7 @@ Recall our definition of this metric:
|
||||
|
||||
Here we see that this instruction *could* have loaded up to 256 bytes of
|
||||
data (4 bytes for each work-item in the wavefront), and therefore this
|
||||
is the expected value for this metric in Omniperf, hence why this metric
|
||||
is the expected value for this metric in ROCm Compute Profiler, hence why this metric
|
||||
is named the “theoretical” bandwidth.
|
||||
|
||||
To further illustrate this point we plot the relationship of the
|
||||
@@ -104,11 +104,11 @@ launched from 1 to 256:
|
||||
.. figure:: ../data/profiling-by-example/ldsbandwidth.png
|
||||
:align: center
|
||||
:alt: Comparison of effective bandwidth versus the theoretical bandwidth
|
||||
metric in Omniperf for our simple example.
|
||||
metric in ROCm Compute Profiler for our simple example.
|
||||
:width: 800
|
||||
|
||||
Comparison of effective bandwidth versus the theoretical bandwidth
|
||||
metric in Omniperf for our simple example.
|
||||
metric in ROCm Compute Profiler for our simple example.
|
||||
|
||||
Here we see that the theoretical bandwidth metric follows a step-function. It
|
||||
increases only when another wavefront issues an LDS instruction for up to 256
|
||||
@@ -172,7 +172,7 @@ see:
|
||||
|
||||
.. code-block:: shell
|
||||
|
||||
$ omniperf analyze -p workloads/lds/mi200/ -b 12.2.4 12.2.6 --dispatch 256 -n per_kernel
|
||||
$ rocprof-compute analyze -p workloads/lds/mi200/ -b 12.2.4 12.2.6 --dispatch 256 -n per_kernel
|
||||
<...>
|
||||
--------------------------------------------------------------------------------
|
||||
12. Local Data Share (LDS)
|
||||
@@ -196,7 +196,7 @@ Looking at the next ``conflicts`` dispatch (i.e., two work-items) yields:
|
||||
|
||||
.. code-block:: shell
|
||||
|
||||
$ omniperf analyze -p workloads/lds/mi200/ -b 12.2.4 12.2.6 --dispatch 257 -n per_kernel
|
||||
$ rocprof-compute analyze -p workloads/lds/mi200/ -b 12.2.4 12.2.6 --dispatch 257 -n per_kernel
|
||||
<...>
|
||||
--------------------------------------------------------------------------------
|
||||
12. Local Data Share (LDS)
|
||||
|
||||
@@ -4,15 +4,15 @@ Occupancy limiters example
|
||||
==========================
|
||||
|
||||
For this example, consider the
|
||||
:dev-sample:`occupancy <occupancy.hip>` included with Omniperf. We will
|
||||
:dev-sample:`occupancy <occupancy.hip>` included with ROCm Compute Profiler. We will
|
||||
investigate the use of the resource allocation panel in the
|
||||
:ref:`Workgroup Manager <desc-spi>`’s metrics section to determine occupancy
|
||||
limiters. This code contains several kernels to explore how both various
|
||||
kernel resources impact achieved occupancy, and how this is reported in
|
||||
Omniperf.
|
||||
ROCm Compute Profiler.
|
||||
|
||||
This example was compiled and run on a MI250 accelerator using ROCm
|
||||
v5.6.0, and Omniperf v2.0.0:
|
||||
v5.6.0, and ROCm Compute Profiler v2.0.0:
|
||||
|
||||
.. code-block:: shell
|
||||
|
||||
@@ -21,11 +21,11 @@ v5.6.0, and Omniperf v2.0.0:
|
||||
We have again included the ``--save-temps`` flag to get the
|
||||
corresponding assembly.
|
||||
|
||||
Finally, we generate our Omniperf profile as:
|
||||
Finally, we generate our ROCm Compute Profiler profile as:
|
||||
|
||||
.. code-block:: shell
|
||||
|
||||
$ omniperf profile -n occupancy --no-roof -- ./occupancy
|
||||
$ rocprof-compute profile -n occupancy --no-roof -- ./occupancy
|
||||
|
||||
.. _occupancy-experiment-design:
|
||||
|
||||
@@ -88,7 +88,7 @@ depend on the exact ROCm/compiler version.
|
||||
|
||||
We will use various permutations of this kernel to limit occupancy, and
|
||||
more importantly for the purposes of this example, demonstrate how this
|
||||
is reported in Omniperf.
|
||||
is reported in ROCm Compute Profiler.
|
||||
|
||||
.. _vgpr-occupancy:
|
||||
|
||||
@@ -101,7 +101,7 @@ the analyze step on this kernel:
|
||||
|
||||
.. code-block:: shell
|
||||
|
||||
$ omniperf analyze -p workloads/occupancy/mi200/ -b 2.1.15 6.2 7.1.5 7.1.6 7.1.7 --dispatch 1
|
||||
$ rocprof-compute analyze -p workloads/occupancy/mi200/ -b 2.1.15 6.2 7.1.5 7.1.6 7.1.7 --dispatch 1
|
||||
<...>
|
||||
--------------------------------------------------------------------------------
|
||||
0. Top Stat
|
||||
@@ -226,7 +226,7 @@ Analyzing this:
|
||||
|
||||
.. code-block:: shell
|
||||
|
||||
$ omniperf analyze -p workloads/occupancy/mi200/ -b 2.1.15 6.2 7.1.5 7.1.6 7.1.7 7.1.8 --dispatch 3
|
||||
$ rocprof-compute analyze -p workloads/occupancy/mi200/ -b 2.1.15 6.2 7.1.5 7.1.6 7.1.7 7.1.8 --dispatch 3
|
||||
<...>
|
||||
--------------------------------------------------------------------------------
|
||||
2. System Speed-of-Light
|
||||
@@ -351,7 +351,7 @@ Analyzing this workload yields:
|
||||
|
||||
.. code-block:: shell-session
|
||||
|
||||
$ omniperf analyze -p workloads/occupancy/mi200/ -b 2.1.15 6.2 7.1.5 7.1.6 7.1.7 7.1.8 7.1.9 --dispatch 5
|
||||
$ rocprof-compute analyze -p workloads/occupancy/mi200/ -b 2.1.15 6.2 7.1.5 7.1.6 7.1.7 7.1.8 7.1.9 --dispatch 5
|
||||
<...>
|
||||
--------------------------------------------------------------------------------
|
||||
0. Top Stat
|
||||
|
||||
@@ -5,7 +5,7 @@ VALU arithmetic instruction mix
|
||||
|
||||
For this example, consider the
|
||||
:dev-sample:`instruction mix sample <instmix.hip>` distributed as a part
|
||||
of Omniperf.
|
||||
of ROCm Compute Profiler.
|
||||
|
||||
.. note::
|
||||
|
||||
@@ -55,7 +55,7 @@ Instruction mix
|
||||
^^^^^^^^^^^^^^^
|
||||
|
||||
This example was compiled and run on a MI250 accelerator using ROCm
|
||||
v5.6.0, and Omniperf v2.0.0.
|
||||
v5.6.0, and ROCm Compute Profiler v2.0.0.
|
||||
|
||||
.. code-block:: shell
|
||||
|
||||
@@ -65,13 +65,13 @@ Generate the profile for this example using the following command.
|
||||
|
||||
.. code-block:: shell
|
||||
|
||||
$ omniperf profile -n instmix --no-roof -- ./instmix
|
||||
$ rocprof-compute profile -n instmix --no-roof -- ./instmix
|
||||
|
||||
Analyze the instruction mix section.
|
||||
|
||||
.. code-block:: shell
|
||||
|
||||
$ omniperf analyze -p workloads/instmix/mi200/ -b 10.2
|
||||
$ rocprof-compute analyze -p workloads/instmix/mi200/ -b 10.2
|
||||
<...>
|
||||
10. Compute Units - Instruction Mix
|
||||
10.2 VALU Arithmetic Instr Mix
|
||||
|
||||
@@ -10,7 +10,7 @@ Global / Generic (FLAT)
|
||||
|
||||
For this example, consider the
|
||||
:dev-sample:`vector memory sample <vmem.hip>` distributed as a part of
|
||||
Omniperf. This code launches many different versions of a simple
|
||||
ROCm Compute Profiler. This code launches many different versions of a simple
|
||||
read/write/atomic-only kernels targeting various address spaces. For example,
|
||||
below is our simple ``global_write`` kernel:
|
||||
|
||||
@@ -24,7 +24,7 @@ below is our simple ``global_write`` kernel:
|
||||
.. note::
|
||||
|
||||
This example was compiled and run on an MI250 accelerator using ROCm
|
||||
v5.6.0, and Omniperf v2.0.0.
|
||||
v5.6.0, and ROCm Compute Profiler v2.0.0.
|
||||
|
||||
.. code-block:: shell-session
|
||||
|
||||
@@ -34,11 +34,11 @@ We have also chosen to include the ``--save-temps`` flag to save the
|
||||
compiler temporary files, such as the generated CDNA assembly code, for
|
||||
inspection.
|
||||
|
||||
Finally, we generate our ``omniperf profile`` as follows.
|
||||
Finally, we generate our ``rocprof-compute profile`` as follows.
|
||||
|
||||
.. code-block:: shell-session
|
||||
|
||||
$ omniperf profile -n vmem --no-roof -- ./vmem
|
||||
$ rocprof-compute profile -n vmem --no-roof -- ./vmem
|
||||
|
||||
.. _flat-experiment-design:
|
||||
|
||||
@@ -94,7 +94,7 @@ First, we demonstrate our simple ``global_write`` kernel:
|
||||
|
||||
.. code-block:: shell-session
|
||||
|
||||
$ omniperf analyze -p workloads/vmem/mi200/ --dispatch 1 -b 10.3 15.1.4 15.1.5 15.1.6 15.1.7 15.1.8 15.1.9 15.1.10 15.1.11 -n per_kernel
|
||||
$ rocprof-compute analyze -p workloads/vmem/mi200/ --dispatch 1 -b 10.3 15.1.4 15.1.5 15.1.6 15.1.7 15.1.8 15.1.9 15.1.10 15.1.11 -n per_kernel
|
||||
<...>
|
||||
--------------------------------------------------------------------------------
|
||||
0. Top Stat
|
||||
@@ -208,7 +208,7 @@ Examining this kernel in the VMEM Instruction Mix table yields:
|
||||
|
||||
.. code-block:: shell-session
|
||||
|
||||
$ omniperf analyze -p workloads/vmem/mi200/ --dispatch 2 -b 10.3 -n per_kernel
|
||||
$ rocprof-compute analyze -p workloads/vmem/mi200/ --dispatch 2 -b 10.3 -n per_kernel
|
||||
<...>
|
||||
0. Top Stat
|
||||
╒════╤══════════════════════════════════════════╤═════════╤═══════════╤════════════╤══════════════╤════════╕
|
||||
@@ -264,7 +264,7 @@ access.
|
||||
|
||||
.. code-block:: shell-session
|
||||
|
||||
$ omniperf analyze -p workloads/vmem/mi200/ --dispatch 2 -b 12.2.0 -n per_kernel
|
||||
$ rocprof-compute analyze -p workloads/vmem/mi200/ --dispatch 2 -b 12.2.0 -n per_kernel
|
||||
<...>
|
||||
12. Local Data Share (LDS)
|
||||
12.2 LDS Stats
|
||||
@@ -304,11 +304,11 @@ Here we observe a now familiar pattern:
|
||||
the compiler to statically eliminate, but is identically false. In this
|
||||
case, our ``main()`` function initializes the data in ``ptr`` to zero.
|
||||
|
||||
Running Omniperf on this kernel yields:
|
||||
Running ROCm Compute Profiler on this kernel yields:
|
||||
|
||||
.. code-block:: shell-session
|
||||
|
||||
$ omniperf analyze -p workloads/vmem/mi200/ --dispatch 3 -b 10.3 -n per_kernel
|
||||
$ rocprof-compute analyze -p workloads/vmem/mi200/ --dispatch 3 -b 10.3 -n per_kernel
|
||||
<...>
|
||||
0. Top Stat
|
||||
╒════╤════════════════════════════════════╤═════════╤═══════════╤════════════╤══════════════╤════════╕
|
||||
@@ -383,11 +383,11 @@ false conditional (both ``zero`` and ``filter`` are set to zero in the
|
||||
kernel launch). Note that this is a *different* conditional from our
|
||||
pointer assignment (to avoid combination of the two).
|
||||
|
||||
Running Omniperf on this kernel reports:
|
||||
Running ROCm Compute Profiler on this kernel reports:
|
||||
|
||||
.. code-block:: shell-session
|
||||
|
||||
$ omniperf analyze -p workloads/vmem/mi200/ --dispatch 4 -b 10.3 12.2.0 16.3.10 -n per_kernel
|
||||
$ rocprof-compute analyze -p workloads/vmem/mi200/ --dispatch 4 -b 10.3 12.2.0 16.3.10 -n per_kernel
|
||||
<...>
|
||||
0. Top Stat
|
||||
╒════╤══════════════════════════════════════════╤═════════╤═══════════╤════════════╤══════════════╤════════╕
|
||||
@@ -468,11 +468,11 @@ to a pointer.
|
||||
}
|
||||
|
||||
|
||||
Running Omniperf on this kernel yields:
|
||||
Running ROCm Compute Profiler on this kernel yields:
|
||||
|
||||
.. code-block:: shell-session
|
||||
|
||||
$ omniperf analyze -p workloads/vmem/mi200/ --dispatch 5 -b 10.3 16.3.12 -n per_kernel
|
||||
$ rocprof-compute analyze -p workloads/vmem/mi200/ --dispatch 5 -b 10.3 16.3.12 -n per_kernel
|
||||
<...>
|
||||
0. Top Stat
|
||||
╒════╤══════════════════════════════════════╤═════════╤═══════════╤════════════╤══════════════╤════════╕
|
||||
@@ -537,11 +537,11 @@ operation targets both LDS and global memory:
|
||||
This assigns every other work-item to atomically update global memory or
|
||||
local memory.
|
||||
|
||||
Running this kernel through Omniperf shows:
|
||||
Running this kernel through ROCm Compute Profiler shows:
|
||||
|
||||
.. code-block:: shell-session
|
||||
|
||||
$ omniperf analyze -p workloads/vmem/mi200/ --dispatch 6 -b 10.3 12.2.0 16.3.12 -n per_kernel
|
||||
$ rocprof-compute analyze -p workloads/vmem/mi200/ --dispatch 6 -b 10.3 12.2.0 16.3.12 -n per_kernel
|
||||
<...>
|
||||
0. Top Stat
|
||||
╒════╤══════════════════════════════════════════╤═════════╤═══════════╤════════════╤══════════════╤════════╕
|
||||
@@ -623,7 +623,7 @@ manner. See
|
||||
for further reading on this instruction type.
|
||||
|
||||
We develop a `simple
|
||||
kernel <https://github.com/ROCm/omniperf/blob/amd-mainline/sample/stack.hip>`__
|
||||
kernel <https://github.com/ROCm/rocprofiler-compute/blob/amd-mainline/sample/stack.hip>`__
|
||||
that uses stack memory:
|
||||
|
||||
.. code-block:: cpp
|
||||
@@ -647,19 +647,19 @@ Our strategy here is to:
|
||||
to global memory to prevent the compiler from optimizing it out.
|
||||
|
||||
This example was compiled and run on an MI250 accelerator using ROCm v5.6.0, and
|
||||
Omniperf v2.0.0.
|
||||
ROCm Compute Profiler v2.0.0.
|
||||
|
||||
.. code-block:: shell-session
|
||||
|
||||
$ hipcc -O3 stack.hip -o stack.hip
|
||||
|
||||
And profiled using Omniperf:
|
||||
And profiled using ROCm Compute Profiler:
|
||||
|
||||
.. code-block:: shell-session
|
||||
|
||||
$ omniperf profile -n stack --no-roof -- ./stack
|
||||
$ rocprof-compute profile -n stack --no-roof -- ./stack
|
||||
<...>
|
||||
$ omniperf analyze -p workloads/stack/mi200/ -b 10.3 16.3.11 -n per_kernel
|
||||
$ rocprof-compute analyze -p workloads/stack/mi200/ -b 10.3 16.3.11 -n per_kernel
|
||||
<...>
|
||||
10. Compute Units - Instruction Mix
|
||||
10.3 VMEM Instr Mix
|
||||
|
||||
@@ -1,22 +1,23 @@
|
||||
.. meta::
|
||||
:description: Omniperf external training resources
|
||||
:keywords: Omniperf, ROCm, profiler, tool, Instinct, accelerator, AMD,
|
||||
training, examples
|
||||
:description: ROCm Compute Profiler external training resources
|
||||
:keywords: Omniperf, examples, tutorials, videos, lesson, lessons, how
|
||||
|
||||
******************
|
||||
Learning resources
|
||||
******************
|
||||
|
||||
This section is a catalog of external resources and third-party content that
|
||||
can help you learn Omniperf. Some areas of the following content might be
|
||||
outdated.
|
||||
This section provides a curated list of external resources and third-party
|
||||
content to support learning the ROCm Compute Profiler. Some information in
|
||||
these materials may be outdated.
|
||||
|
||||
Introduction to Omniperf
|
||||
ROCm Compute Profiler was previously known as Omniperf. Some of the following
|
||||
resources use the earlier name.
|
||||
|
||||
Introduction to ROCm Compute Profiler
|
||||
:fab:`youtube` `AMD profiling workshop (Pawsey Supercomputing Research Centre) <https://www.youtube.com/watch?v=9AkxBCiInCw>`_
|
||||
|
||||
Omniperf example exercises
|
||||
ROCm Compute Profiler example exercises
|
||||
`<https://github.com/amd/HPCTrainingExamples/tree/main/OmniperfExamples>`__
|
||||
|
||||
AMD Instinct™ tuning guides
|
||||
:doc:`rocm:how-to/tuning-guides/mi300x/workload`
|
||||
|
||||
|
||||
@@ -1,14 +1,14 @@
|
||||
.. meta::
|
||||
:description: Omniperf: Profiling by example
|
||||
:keywords: Omniperf, ROCm, profiler, tool, Instinct, accelerator, AMD
|
||||
:description: ROCm Compute Profiler: Profiling by example
|
||||
:keywords: ROCm Compute Profiler, ROCm, profiler, tool, Instinct, accelerator, AMD
|
||||
|
||||
********************
|
||||
Profiling by example
|
||||
********************
|
||||
|
||||
The following examples refer to sample :doc:`HIP <hip:index>` code located in
|
||||
:fab:`github` :dev-sample:`ROCm/omniperf/blob/amd-mainline/sample <>` and distributed
|
||||
as part of Omniperf.
|
||||
:fab:`github` :dev-sample:`ROCm/rocprofiler-compute/blob/amd-mainline/sample <>`
|
||||
and distributed as part of ROCm Compute Profiler.
|
||||
|
||||
.. include:: ./includes/valu-arithmetic-instruction-mix.rst
|
||||
|
||||
|
||||
Verwijs in nieuw issue
Block a user