Remove redundant ROCPROFSYS_TRACE_CACHED variable from the code (#2434)

This commit is contained in:
marantic-amd
2025-12-25 13:36:04 +01:00
committed by GitHub
parent c3132773c8
commit bb83791b17
23 changed files with 88 additions and 111 deletions
@@ -176,18 +176,14 @@ Primary collection modes
Trace mode (default)
^^^^^^^^^^^^^^^^^^^^^^^^
Tracing mode generates comprehensive, deterministic traces of every event and measurement during application execution. This mode can be enabled using ``ROCPROFSYS_MODE=trace`` or by enabling one of the trace backend options.
Tracing mode generates comprehensive, deterministic traces of every event and measurement during application execution. This mode can be enabled using ``ROCPROFSYS_TRACE=true``, ``ROCPROFSYS_MODE=trace``, or by using the ``--trace`` / ``-T`` CLI flag.
ROCm Systems Profiler provides two trace backend modes:
ROCm Systems Profiler provides two tracing implementations:
- **Cached Mode (default, recommended)**: ``ROCPROFSYS_TRACE_CACHED=true`` or ``--trace`` / ``-T`` enables deferred trace generation with minimal runtime overhead. Trace data is buffered during execution and written after the application completes, significantly reducing performance impact during profiling.
- **Cached Mode (default)**: By default, when tracing is enabled, ROCm Systems Profiler uses deferred trace generation with minimal runtime overhead. Trace data is buffered during execution and written after the application completes, significantly reducing performance impact during profiling.
- **Legacy Mode**: ``ROCPROFSYS_TRACE_LEGACY=true`` or ``--trace-legacy`` / ``-L`` enables direct mode where trace data is written immediately during execution. This mode provides real-time trace generation but has higher runtime overhead compared to cached mode.
.. note::
The ``ROCPROFSYS_TRACE`` environment variable is deprecated and has been renamed to ``ROCPROFSYS_TRACE_LEGACY``. For new workflows, use ``ROCPROFSYS_TRACE_CACHED`` (default) or ``ROCPROFSYS_TRACE_LEGACY`` explicitly.
Additional configuration options to control the tracing behavior include:
- ``ROCPROFSYS_TRACE_DELAY`` (``--trace-wait``): Delay before starting trace collection (in seconds).
@@ -34,8 +34,8 @@ and tweak the default sampling values.
.. code-block:: shell
# ...
ROCPROFSYS_TRACE_CACHED = true # Recommended: deferred trace generation for minimal overhead
# ROCPROFSYS_TRACE_LEGACY = false # Alternative: direct mode with higher overhead
ROCPROFSYS_TRACE = true
# ROCPROFSYS_TRACE_LEGACY = false # Set to true for direct mode (higher overhead)
ROCPROFSYS_PROFILE = true
ROCPROFSYS_USE_SAMPLING = true
ROCPROFSYS_USE_PROCESS_SAMPLING = true
@@ -340,7 +340,7 @@ Generating a default configuration file
ROCPROFSYS_CONFIG_FILE =
ROCPROFSYS_MODE = trace
ROCPROFSYS_TRACE_CACHED = true
ROCPROFSYS_TRACE = true
ROCPROFSYS_TRACE_LEGACY = false
ROCPROFSYS_PROFILE = false
ROCPROFSYS_USE_SAMPLING = false
@@ -499,9 +499,8 @@ Viewing the setting descriptions
| ROCPROFSYS_USE_CODE_COVERAGE | Enable support for code coverage |
| ROCPROFSYS_USE_KOKKOSP | Enable support for Kokkos Tools |
| ROCPROFSYS_USE_OMPT | Enable support for OpenMP-Tools |
| ROCPROFSYS_TRACE_CACHED | Enable perfetto backend with deferred...|
| ROCPROFSYS_TRACE_LEGACY | Enable perfetto backend (legacy, dir... |
| ROCPROFSYS_TRACE | [DEPRECATED] Renamed to ROCPROFSYS_T... |
| ROCPROFSYS_TRACE | Enable perfetto backend for tracing |
| ROCPROFSYS_TRACE_LEGACY | Use legacy direct mode for tracing |
| ROCPROFSYS_USE_PID | Enable tagging filenames with proces... |
| ROCPROFSYS_USE_AMD_SMI | Enable sampling GPU power, temp, uti... |
| ROCPROFSYS_USE_ROCM | Enable ROCM tracing |
@@ -1349,8 +1348,8 @@ but do not override an existing value for the environment variable.
$SAMPLE = OFF
# use fields
ROCPROFSYS_TRACE_CACHED = $ENABLE # Recommended: deferred trace generation
ROCPROFSYS_TRACE_LEGACY = OFF # Legacy direct mode (higher overhead)
ROCPROFSYS_TRACE = $ENABLE
# ROCPROFSYS_TRACE_LEGACY = OFF # Set to ON for direct mode (higher overhead)
ROCPROFSYS_PROFILE = $ENABLE
ROCPROFSYS_USE_SAMPLING = $SAMPLE
ROCPROFSYS_USE_PROCESS_SAMPLING = $SAMPLE
@@ -75,7 +75,7 @@ The configuration parameter settings can be saved in a configuration file. Here
ROCPROFSYS_SAMPLING_DELAY=0.05
ROCPROFSYS_SAMPLING_CPUS=0-9
ROCPROFSYS_SAMPLING_GPUS=$env:HIP_VISIBLE_DEVICES
ROCPROFSYS_TRACE_CACHED=ON
ROCPROFSYS_TRACE=ON
ROCPROFSYS_PROFILE=ON
ROCPROFSYS_USE_SAMPLING=ON
ROCPROFSYS_USE_PROCESS_SAMPLING=OFF
@@ -284,7 +284,7 @@ The following snippets show how ``rocprof-sys-sample`` runs with various environ
LD_PRELOAD=/opt/rocprofiler-systems/lib/librocprof-sys-dl.so.1.7.1
ROCPROFSYS_CPU_FREQ_ENABLED=false
ROCPROFSYS_PROFILE=true
ROCPROFSYS_TRACE_CACHED=true
ROCPROFSYS_TRACE=true
ROCPROFSYS_USE_AMD_SMI=true
ROCPROFSYS_USE_PROCESS_SAMPLING=true
ROCPROFSYS_USE_SAMPLING=true
@@ -307,7 +307,7 @@ The following snippets show how ``rocprof-sys-sample`` runs with various environ
ROCPROFSYS_USE_KOKKOSP=true
ROCPROFSYS_USE_MPIP=true
ROCPROFSYS_USE_OMPT=true
ROCPROFSYS_TRACE_CACHED=true
ROCPROFSYS_TRACE=true
ROCPROFSYS_USE_PROCESS_SAMPLING=true
ROCPROFSYS_USE_RCCLP=true
ROCPROFSYS_USE_AMD_SMI=true
@@ -337,7 +337,7 @@ The following snippets show how ``rocprof-sys-sample`` runs with various environ
ROCPROFSYS_USE_KOKKOSP=false
ROCPROFSYS_USE_MPIP=false
ROCPROFSYS_USE_OMPT=false
ROCPROFSYS_TRACE_CACHED=true
ROCPROFSYS_TRACE=true
ROCPROFSYS_USE_PROCESS_SAMPLING=true
ROCPROFSYS_USE_RCCLP=false
ROCPROFSYS_USE_AMD_SMI=false
@@ -362,7 +362,7 @@ Here is the full output from the previous
ROCPROFSYS_OUTPUT_PATH=rocprof-sys-output
ROCPROFSYS_OUTPUT_PREFIX=%tag%
ROCPROFSYS_PROFILE=true
ROCPROFSYS_TRACE_CACHED=true
ROCPROFSYS_TRACE=true
ROCPROFSYS_TRACE_THREAD_LOCKS=false
ROCPROFSYS_TRACE_THREAD_RW_LOCKS=false
ROCPROFSYS_TRACE_THREAD_SPIN_LOCKS=false
@@ -17,7 +17,7 @@ For example, starting with the following base configuration:
export ROCPROFSYS_TIME_OUTPUT=ON
export ROCPROFSYS_USE_PID=OFF
export ROCPROFSYS_PROFILE=ON
export ROCPROFSYS_TRACE_CACHED=ON
export ROCPROFSYS_TRACE=ON
.. code-block:: shell
@@ -192,7 +192,7 @@ First, instrument and run the program.
ROCPROFSYS: LD_PRELOAD=/home/rocm-dev/code/rocprofiler-systems/build/ubuntu/22.04/lib/librocprof-sys-dl.so.1.0.0
ROCPROFSYS: OMP_TOOL_LIBRARIES=/home/rocm-dev/code/rocprofiler-systems/build/ubuntu/22.04/lib/librocprof-sys-dl.so.1.0.0
ROCPROFSYS: ROCPROFSYS_PROFILE=true
ROCPROFSYS: ROCPROFSYS_TRACE_CACHED=true
ROCPROFSYS: ROCPROFSYS_TRACE=true
ROCPROFSYS: ROCPROFSYS_VERBOSE=0
[rocprof-sys][dl][1827155] rocprofsys_main
[rocprof-sys][1827155][rocprofsys_init_tooling] Instrumentation mode: Trace