Merge branch 'master' into directives

Author: isazi

INSTALL.rst

@@ -125,7 +125,7 @@ Or you could install Kernel Tuner and PyOpenCL together if you haven't done so a
 If this fails, please see the PyOpenCL installation guide (https://wiki.tiker.net/PyOpenCL/Installation)
 
 HIP and HIP Python
--------------
+------------------
 
 Before we can install HIP Python, you'll need to have the HIP runtime and compiler installed on your system.
 The HIP compiler is included as part of the ROCm software stack. Here is AMD's installation guide:

doc/source/architecture.png

@@ -89,14 +89,8 @@ kernel_tuner.runners.sequential.SimulationRunner
     :members:
 
 
-Device Interfaces
------------------
-
-kernel_tuner.core.DeviceInterface
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-.. autoclass:: kernel_tuner.core.DeviceInterface
-    :special-members: __init__
-    :members:
+Backends
+--------
 
 kernel_tuner.backends.pycuda.PyCudaFunctions
 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

doc/source/design.rst

@@ -104,6 +104,7 @@ def __init__(self, device=0, iterations=7, compiler_options=None, observers=None
 
     def ready_argument_list(self, arguments):
         """Ready argument list to be passed to the HIP function.
+
         :param arguments: List of arguments to be passed to the HIP function.
             The order should match the argument list on the HIP function.
             Allowed values are np.ndarray, and/or np.int32, np.float32, and so on.

kernel_tuner/backends/hip.py

@@ -592,8 +592,8 @@ def tune_kernel(
     # check for forbidden names in tune parameters
     util.check_tune_params_list(tune_params, observers, simulation_mode=simulation_mode)
 
-    # check whether block_size_names are used as expected
-    util.check_block_size_params_names_list(block_size_names, tune_params)
+    # check whether block_size_names are used
+    block_size_names = util.check_block_size_params_names_list(block_size_names, tune_params)
 
     # ensure there is always at least three names
     util.append_default_block_size_names(block_size_names)

kernel_tuner/interface.py

@@ -12,7 +12,7 @@
 class PowerSensorObserver(BenchmarkObserver):
     """Observer that an external PowerSensor2 device to accurately measure power
 
-    Requires PowerSensor2 hardware and powersensor Python bindings.
+    Requires PowerSensor3 hardware and powersensor Python bindings.
 
     :param observables: A list of string, containing any of "ps_energy" or "ps_power".
         To measure energy in Joules or power consumption in Watt.
@@ -51,9 +51,8 @@ def after_finish(self):
             )  # Joules
             self.results["ps_energy"].append(ps_measured_e)
         if "ps_power" in self.observables:
-            ps_measured_t = (
-                    end_state.time_at_read - self.begin_state.time_at_read
-            )  # seconds
+            ps_measured_t = ((end_state.time_at_read - self.begin_state.time_at_read).microseconds / 1e6)    # Seconds
+
             self.results["ps_power"].append(ps_measured_e / ps_measured_t)  # Watt
 
     def get_results(self):

kernel_tuner/observers/powersensor.py

@@ -235,7 +235,7 @@ def get_hyperparam(name: str, default, supported_values=list()):
         self.invalid_value = 1e20
         self.opt_direction = opt_direction
         if opt_direction == "min":
-            self.worst_value = np.PINF
+            self.worst_value = np.inf
             self.argopt = np.argmin
         elif opt_direction == "max":
             self.worst_value = np.NINF
@@ -262,7 +262,7 @@ def get_hyperparam(name: str, default, supported_values=list()):
         self.__visited_num = 0
         self.__visited_valid_num = 0
         self.__visited_searchspace_indices = [False] * self.searchspace_size
-        self.__observations = [np.NaN] * self.searchspace_size
+        self.__observations = [np.nan] * self.searchspace_size
         self.__valid_observation_indices = [False] * self.searchspace_size
         self.__valid_params = list()
         self.__valid_observations = list()
@@ -311,7 +311,7 @@ def is_not_visited(self, index: int) -> bool:
 
     def is_valid(self, observation: float) -> bool:
         """Returns whether an observation is valid."""
-        return not (observation is None or observation == self.invalid_value or observation == np.NaN)
+        return not (observation is None or observation == self.invalid_value or observation == np.nan)
 
     def get_af_by_name(self, name: str):
         """Get the basic acquisition functions by their name."""

kernel_tuner/strategies/bayes_opt.py

@@ -235,11 +235,22 @@ def check_block_size_params_names_list(block_size_names, tune_params):
                     "Block size name " + name + " is not specified in the tunable parameters list!", UserWarning
                 )
     else:  # if default block size names are used
-        if not any([k in default_block_size_names for k in tune_params.keys()]):
+        if not any([k.lower() in default_block_size_names for k in tune_params.keys()]):
             warnings.warn(
                 "None of the tunable parameters specify thread block dimensions!",
                 UserWarning,
             )
+        else:
+            # check for alternative case spelling of defaults such as BLOCK_SIZE_X or block_Size_X etc
+            result = []
+            for k in tune_params.keys():
+                if k.lower() in default_block_size_names and k not in default_block_size_names:
+                    result.append(k)
+            # ensure order of block_size_names is correct regardless of case used
+            block_size_names = sorted(result, key=str.casefold)
+
+    return block_size_names
+
 
 
 def check_restriction(restrict, params: dict) -> bool:

kernel_tuner/util.py

@@ -6,7 +6,7 @@ build-backend = "poetry.core.masonry.api"
 name = "kernel_tuner"
 packages = [{ include = "kernel_tuner", from = "." }]
 description = "An easy to use CUDA/OpenCL kernel tuner in Python"
-version = "1.0" # adhere to PEP440 versioning: https://packaging.python.org/en/latest/guides/distributing-packages-using-setuptools/#id55
+version = "1.1.0" # adhere to PEP440 versioning: https://packaging.python.org/en/latest/guides/distributing-packages-using-setuptools/#id55
 license = "Apache-2.0"
 authors = [
     "Ben van Werkhoven <[email protected]>",
@@ -57,12 +57,17 @@ generate-setup-file = false
 [tool.poetry.scripts]
 kernel_tuner = "kernel_tuner.interface:entry_point"
 
+[[tool.poetry.source]]
+name = "testpypi"
+url = "https://test.pypi.org/simple/"
+priority = "explicit"
+
 # ATTENTION: if anything is changed here, run `poetry update`
 [tool.poetry.dependencies]
-python = ">=3.9,<3.15"    # NOTE when changing the supported Python versions, also change the test versions in the noxfile
-numpy = "^1.26.0"          # Python 3.12 requires numpy at least 1.26
-scipy = ">=1.11.0"         # held back by Python 3.9
-packaging = "*"                 # required by file_utils
+python = ">=3.9,<4"         # <4 is because of hip-python # NOTE when changing the supported Python versions, also change the test versions in the noxfile
+numpy = "^2.0.0"            # Python 3.12 requires numpy at least 1.26
+scipy = ">=1.11.0"          # held back by Python 3.9
+packaging = "*"             # required by file_utils
 jsonschema = "*"
 python-constraint2 = "^2.1.0"
 xmltodict = "*"
@@ -72,6 +77,7 @@ scikit-learn = ">=1.0.2"
 
 # List of optional dependencies for user installation, e.g. `pip install kernel_tuner[cuda]`, used in the below `extras`.
 # Please note that this is different from the dependency groups below, e.g. `docs` and `test`, those are for development.
+# ATTENTION: if anything is changed here, run `poetry update`
 # CUDA
 pycuda = { version = "^2024.1", optional = true }           # Attention: if pycuda is changed here, also change `session.install("pycuda")` in the Noxfile
 nvidia-ml-py = { version = "^12.535.108", optional = true }
@@ -82,7 +88,7 @@ pynvml = { version = "^11.4.1", optional = true }
 # OpenCL
 pyopencl = { version = "*", optional = true } # Attention: if pyopencl is changed here, also change `session.install("pyopencl")` in the Noxfile
 # HIP
-hip-python-fork = { version = "*", optional = true }
+hip-python = { version = "^6.3.3.540.31", source = "testpypi", optional = true }  # Note: when released, switch this package to pypi and remove tool.poetry.source
 # Tutorial (for the notebooks used in the examples)
 jupyter = { version = "^1.0.0", optional = true }
 matplotlib = { version = "^3.5.0", optional = true }
@@ -91,7 +97,7 @@ matplotlib = { version = "^3.5.0", optional = true }
 cuda = ["pycuda", "nvidia-ml-py", "pynvml"]
 opencl = ["pyopencl"]
 cuda_opencl = ["pycuda", "pyopencl"]
-hip = ["hip-python-fork"]
+hip = ["hip-python"]
 tutorial = ["jupyter", "matplotlib", "nvidia-ml-py"]
 
 # ATTENTION: if anything is changed here, run `poetry update` and `poetry export --with docs --without-hashes --format=requirements.txt --output doc/requirements.txt`
@@ -147,4 +153,4 @@ select = [
     "D", # pydocstyle,
 ]
 [tool.ruff.pydocstyle]
-convention = "google"
+convention = "google"

pyproject.toml

@@ -55,8 +55,9 @@
 
 try:
     from hip import hip
+    hip.hipDriverGetVersion()
     hip_present = True
-except ImportError:
+except (ImportError, RuntimeError):
     hip_present = False
 
 skip_if_no_pycuda = pytest.mark.skipif(
@@ -78,7 +79,7 @@
 )
 skip_if_no_openmp = pytest.mark.skipif(not openmp_present, reason="No OpenMP found")
 skip_if_no_openacc = pytest.mark.skipif(not openacc_present, reason="No nvc++ on PATH")
-skip_if_no_hip = pytest.mark.skipif(not hip_present, reason="No HIP Python found")
+skip_if_no_hip = pytest.mark.skipif(not hip_present, reason="No HIP Python found or no HIP device detected")
 
 
 def skip_backend(backend: str):