Add support for KERNEL_TUNER_PARALLEL environment variable

Author: stijnh

doc/source/parallel.rst

@@ -31,7 +31,14 @@ To enable parallel tuning, pass the ``parallel`` argument to ``tune_kernel``:
     )
 
 If ``parallel`` is set to ``True``, Kernel Tuner will use all available Ray workers for tuning.
-Alternatively, ``parallel`` can be set to an integer ``n`` to use exactly ``n`` workers.
+The ``parallel`` option can also be set to an integer ``n`` to use exactly ``n`` workers.
+
+Alternatively, define the environment variable ``KERNEL_TUNER_PARALLEL`` to enable parallel execution without modifying your Python code.
+
+.. code-block:: bash
+
+   $ KERNEL_TUNER_PARALLEL=true python3 my_tuning_script.py
+
 
 
 Parallel tuning and optimization strategies

kernel_tuner/interface.py

@@ -28,6 +28,7 @@
 from argparse import ArgumentParser
 from ast import literal_eval
 from datetime import datetime
+import os
 from pathlib import Path
 from time import perf_counter
 from copy import deepcopy
@@ -67,6 +68,8 @@
 )
 from kernel_tuner.strategies.wrapper import OptAlgWrapper
 
+environment_key_parallel = "KERNEL_TUNER_PARALLEL"
+
 strategy_map = {
     "brute_force": brute_force,
     "random_sample": random_sample,
@@ -585,7 +588,7 @@ def tune_kernel(
     strategy_options=None,
     cache=None,
     metrics=None,
-    simulation_mode=False,
+    simulation_mode=None,
     parallel=None,
     observers=None,
     objective=None,
@@ -662,6 +665,12 @@ def tune_kernel(
     # TODO: we could use the "match case" syntax when removing support for 3.9
     tuning_options.simulated_time = 0
 
+    # Get runner from environment if possible
+    if parallel is None:
+        parallel = bool(os.environ.get(environment_key_parallel))
+
+
+    # Create runner
     if parallel and simulation_mode:
         raise ValueError("Enabling `parallel` and `simulation_mode` together is not supported")
     elif simulation_mode: