miscellaneous doc updates (#86)
* miscellaneous doc updates
* updated deprecartion message
* Updated memory allocation tracking documentation
* Update comparing-with-legacy-tools.rst
Co-authored-by: Elwazir, Ammar <Ammar.Elwazir@amd.com>
* Update comparing-with-legacy-tools.rst
Co-authored-by: Elwazir, Ammar <Ammar.Elwazir@amd.com>
* Update comparing-with-legacy-tools.rst
---------
Co-authored-by: Ian Trowbridge <ian.trowbridge@amd.com>
Co-authored-by: Elwazir, Ammar <Ammar.Elwazir@amd.com>
[ROCm/rocprofiler-sdk commit: 81f600e3ba]
This commit is contained in:
@@ -86,7 +86,14 @@ ROCprofiler-SDK introduces a new command-line tool, `rocprofv3`, which is a more
|
||||
- Part of HIP and HSA Traces
|
||||
- `--memory-copy-trace`
|
||||
- Provides granularity for memory move operations
|
||||
-
|
||||
-
|
||||
* - Basic tracing options
|
||||
- Memory allocation Trace
|
||||
- *Not Available*
|
||||
- *Not Available*
|
||||
- `--memory-allocation-trace`
|
||||
- New option for collecting Memory Allocation Traces. Displays starting address, allocation size, and agent where allocation occurred.
|
||||
-
|
||||
* - Basic tracing options
|
||||
- Kernel Trace
|
||||
- `--kernel-trace`
|
||||
@@ -136,6 +143,20 @@ ROCprofiler-SDK introduces a new command-line tool, `rocprofv3`, which is a more
|
||||
- `--hsa-finalizer-trace`
|
||||
- New option for collecting HSA API Traces (Finalizer-extension API), e.g. HSA functions prefixed with only `hsa_ext_program_` (i.e. hsa_ext_program_create)
|
||||
-
|
||||
* - Advanced tracing options
|
||||
- Kokkos trace
|
||||
- *Not Available*
|
||||
- *Not Available*
|
||||
- `--kokkos-trace`
|
||||
- New option to enable built-in Kokkos Tools support (implies --marker-trace and --kernel-rename)
|
||||
-
|
||||
* - Advanced tracing options
|
||||
- RCCL trace
|
||||
- *Not Available*
|
||||
- *Not Available*
|
||||
- `--rccl-trace`
|
||||
- For collecting RCCL(ROCm Communication Collectives Library. Also pronounced as 'Rickle' ) Traces
|
||||
-
|
||||
* - Aggregate tracing options
|
||||
- Sys Trace
|
||||
- `--sys-trace` [hip-trace|hsa-trace|roctx-trace|kernel-trace]
|
||||
@@ -300,6 +321,13 @@ ROCprofiler-SDK introduces a new command-line tool, `rocprofv3`, which is a more
|
||||
| # YAML and JSON formats are more readable and easy to maintain.
|
||||
| # Allows flexibility to add more features for the tool input
|
||||
-
|
||||
* - I/O options
|
||||
- Command-line Counter Collection
|
||||
- *Not Available*
|
||||
- *Not Available*
|
||||
- `--pmc`
|
||||
- New option to collect performance counters from command line. Counters should be comma OR space separated in case of more than 1 counters
|
||||
-
|
||||
* - I/O options
|
||||
- Providing Custom metrics file
|
||||
- `-m` <metric file>
|
||||
@@ -318,9 +346,9 @@ ROCprofiler-SDK introduces a new command-line tool, `rocprofv3`, which is a more
|
||||
- Trace Period
|
||||
- `--trace-period`
|
||||
- `-tp | --trace-period`
|
||||
- *Not available*
|
||||
- Not yet in rocprofv3
|
||||
-
|
||||
- `-p |--collection-period`,`--collection-period-unit`
|
||||
- Users can specify multiple configurations, each defined by a triplet in the format `start_delay:collection_time:repeat`, with the ability to change the unit of time in the given configurations.
|
||||
-
|
||||
* - Trace Control options
|
||||
- Trace start
|
||||
- `--trace-start <on|off>`
|
||||
@@ -341,6 +369,13 @@ ROCprofiler-SDK introduces a new command-line tool, `rocprofv3`, which is a more
|
||||
- *Not available*
|
||||
- *Not available*
|
||||
- Not yet in rocprofv3
|
||||
-
|
||||
* - PC Sampling options
|
||||
- PC Sampling`
|
||||
- *Not available*
|
||||
- *Not available*
|
||||
- `--pc-sampling-beta-enabled`
|
||||
- Enable pc sampling support; beta version.
|
||||
-
|
||||
* - Legacy options
|
||||
- Timestamp On/Off
|
||||
|
||||
@@ -520,6 +520,17 @@ For the description of the fields in the output file, see :ref:`output-file-fiel
|
||||
Memory allocation trace
|
||||
+++++++++++++++++++++++++
|
||||
|
||||
Memory allocation traces track the HSA functions ``hsa_memory_allocate``,
|
||||
``hsa_amd_memory_pool_allocate``, and ``hsa_amd_vmem_handle_create```. The function
|
||||
``hipMalloc`` calls these underlying HSA functions allowing memory allocations to be
|
||||
tracked.
|
||||
|
||||
In addition to the HSA memory allocation functions listed above, the corresponding HSA
|
||||
free functions ``hsa_memory_free``, ``hsa_amd_memory_pool_free``, and ``hsa_amd_vmem_handle_release``
|
||||
are also tracked. Unlike the allocation functions, however, only the address of the freed memory
|
||||
is recorded. As such, the agent id and size of the freed memory are recorded as 0 in the CSV and
|
||||
JSON outputs.
|
||||
|
||||
To trace memory allocations during the application run, use:
|
||||
|
||||
.. code-block:: shell
|
||||
|
||||
Reference in New Issue
Block a user