Merge branch 'master' of github.com:KernelTuner/kernel_tuner into pyatf-eval-count-fix

Author: benvanwerkhoven

.github/workflows/test-python-package.yml

@@ -21,7 +21,7 @@ jobs:
 
         strategy:
             matrix:
-                os: [ubuntu-latest, macos-13]
+                os: [ubuntu-latest, macos-latest]
 
         steps:
             - uses: actions/checkout@v4

doc/requirements.txt

@@ -82,7 +82,7 @@ tornado==6.5.1 ; python_version >= "3.9" and python_version < "3.15"
 traitlets==5.14.3 ; python_version >= "3.9" and python_version < "3.15"
 typing-extensions==4.12.2 ; python_version >= "3.9" and python_version < "3.15"
 tzdata==2025.1 ; python_version >= "3.9" and python_version < "3.15"
-urllib3==2.5.0 ; python_version >= "3.9" and python_version < "3.15"
+urllib3==2.6.0 ; python_version >= "3.9" and python_version < "3.15"
 wcwidth==0.2.13 ; python_version >= "3.9" and python_version < "3.15"
 webencodings==0.5.1 ; python_version >= "3.9" and python_version < "3.15"
 xmltodict==0.14.2 ; python_version >= "3.9" and python_version < "3.15"

kernel_tuner/backends/hip/__init__.py

@@ -0,0 +1 @@
+from .hip import *

kernel_tuner/backends/hip.py kernel_tuner/backends/hip/hip.pykernel_tuner/backends/hip.py renamed to kernel_tuner/backends/hip/hip.py

@@ -8,6 +8,7 @@
 
 from kernel_tuner.backends.backend import GPUBackend
 from kernel_tuner.observers.hip import HipRuntimeObserver
+from kernel_tuner.backends.hip.util import hip_check
 
 try:
     from hip import hip, hiprtc
@@ -31,20 +32,6 @@
 
 hipSuccess = 0
 
-
-def hip_check(call_result):
-    """helper function to check return values of hip calls"""
-    err = call_result[0]
-    result = call_result[1:]
-    if len(result) == 1:
-        result = result[0]
-    if isinstance(err, hip.hipError_t) and err != hip.hipError_t.hipSuccess:
-        _, error_name = hip.hipGetErrorName(err)
-        _, error_str = hip.hipGetErrorString(err)
-        raise RuntimeError(f"{error_name}: {error_str}")
-    return result
-
-
 class HipFunctions(GPUBackend):
     """Class that groups the HIP functions on maintains state about the device."""
 

kernel_tuner/backends/hip/util.py

@@ -0,0 +1,18 @@
+try:
+    from hip import hip, hiprtc
+except (ImportError, RuntimeError):
+    hip = None
+
+
+def hip_check(call_result):
+    """helper function to check return values of hip calls"""
+    err = call_result[0]
+    result = call_result[1:]
+    if len(result) == 1:
+        result = result[0]
+    if isinstance(err, hip.hipError_t) and err != hip.hipError_t.hipSuccess:
+        _, error_name = hip.hipGetErrorName(err)
+        _, error_str = hip.hipGetErrorString(err)
+        raise RuntimeError(f"{error_name}: {error_str}")
+    return result
+

kernel_tuner/observers/hip.py

@@ -1,6 +1,7 @@
 import numpy as np
 
 from kernel_tuner.observers.observer import BenchmarkObserver
+from kernel_tuner.backends.hip.util import hip_check
 
 try:
     from hip import hip, hiprtc
@@ -24,7 +25,7 @@ def __init__(self, dev):
 
     def after_finish(self):
         # Time is measured in milliseconds
-        EventElapsedTime = hip.hipEventElapsedTime(self.start, self.end)
+        EventElapsedTime = hip_check(hip.hipEventElapsedTime(self.start, self.end))
         self.times.append(EventElapsedTime)
 
     def get_results(self):

kernel_tuner/runners/simulation.py

@@ -7,6 +7,7 @@
 from kernel_tuner import util
 from kernel_tuner.runners.runner import Runner
 
+
 _SimulationDevice = namedtuple("_SimulationDevice", ["max_threads", "env", "quiet"])
 
 
@@ -91,6 +92,11 @@ def run(self, parameter_space, tuning_options):
             if tuning_options.cache and x_int in tuning_options.cache:
                 result = tuning_options.cache[x_int].copy()
 
+                # only compute metrics on configs that have not errored
+                if tuning_options.metrics and not isinstance(result.get(tuning_options.objective), util.ErrorConfig):
+                    result = util.process_metrics(result, tuning_options.metrics)
+
+
                 # Simulate behavior of sequential runner that when a configuration is
                 # served from the cache by the sequential runner, the compile_time,
                 # verification_time, and benchmark_time are set to 0.

noxfile.py

@@ -15,7 +15,7 @@
 
 # set the test parameters
 verbose = False
-python_versions_to_test = ["3.10", "3.11", "3.12", "3.13"] # 3.14 has not yet been officially released so is not tested against, but is allowed by the pyproject.toml
+python_versions_to_test = ["3.11", "3.12", "3.13", "3.14"]
 nox.options.stop_on_first_error = True
 nox.options.error_on_missing_interpreters = True
 nox.options.default_venv_backend = 'virtualenv'