pyopencl 2024.3__tar.gz → 2025.1__tar.gz

{pyopencl-2024.3 → pyopencl-2025.1}/CMakeLists.txt

@@ -167,6 +167,8 @@ find_package(OpenCL REQUIRED)
 nanobind_add_module(
   _cl
   NB_STATIC # Build static libnanobind (the extension module itself remains a shared library)
+  LTO
+  NOMINSIZE
   src/wrap_constants.cpp
   src/wrap_cl.cpp
   src/wrap_cl_part_1.cpp

{pyopencl-2024.3 → pyopencl-2025.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: pyopencl
-Version: 2024.3
+Version: 2025.1
 Summary: Python wrapper for OpenCL
 Author-Email: Andreas Kloeckner <inform@tiker.net>
 Classifier: Development Status :: 5 - Production/Stable

{pyopencl-2024.3 → pyopencl-2025.1}/contrib/cldis.py

@@ -64,7 +64,7 @@ def main(ctx, tmp_dir, cl_str, output=None, build_options=()):
             if output == "sass":
                 with open(os.path.join(tmp_dir, "cl.ptx"), "w") as f:
                     f.write(res)
-                tgt = re.findall(".target sm_[0-9]*", res, re.MULTILINE)[0]
+                tgt = re.findall(r".target sm_[0-9]*", res, re.MULTILINE)[0]
                 gpu_name = tgt[8:]
                 subprocess.check_call(["ptxas", "cl.ptx", "--verbose",
                     f"--gpu-name={gpu_name}", "--warn-on-spills"], cwd=tmp_dir)

{pyopencl-2024.3 → pyopencl-2025.1}/contrib/fortran-to-opencl/translate.py

@@ -624,7 +624,7 @@ class Scope:
 
             if self.implicit_types is None:
                 raise TranslationError(
-                        "no type for '%s' found in implict none routine"
+                        "no type for '%s' found in implicit none routine"
                         % name)
 
             return self.implicit_types.get(name[0], np.dtype(np.int32))

{pyopencl-2024.3 → pyopencl-2025.1}/doc/runtime_memory.rst

@@ -238,7 +238,7 @@ Operations on SVM
 Image
 -----
 
-.. class:: ImageFormat([channel_order, channel_type])
+.. class:: ImageFormat(channel_order, channel_type)
 
     .. attribute:: channel_order
 

{pyopencl-2024.3 → pyopencl-2025.1}/doc/subst.rst

@@ -8,7 +8,7 @@
     (e.g. by being an :class:`numpy.ndarray`)
 .. |explain-waitfor| replace:: *wait_for*
     may either be *None* or a list of :class:`pyopencl.Event` instances for
-    whose completion this command waits before starting exeuction.
+    whose completion this command waits before starting execution.
 .. |std-enqueue-blurb| replace:: Returns a new :class:`pyopencl.Event`. |explain-waitfor|
 
 .. |copy-depr| replace:: **Note:** This function is deprecated as of PyOpenCL 2011.1.

{pyopencl-2024.3 → pyopencl-2025.1}/examples/n-body.py

@@ -317,7 +317,7 @@ __kernel void InBallSplutterPoints(__global MYFLOAT4* clDataX,
         Heat=MWCfp;
     }
 
-// More accurate distribution based on spherical coordonates
+// More accurate distribution based on spherical coordinates
 // Disactivated because of AMD Oland GPU crash on launch
 //     private MYFLOAT Radius,Theta,Phi,PosX,PosY,PosZ,SinTheta;
 //     Radius=MWCfp*diameter/2.e0f;
@@ -375,7 +375,7 @@ __kernel void SplutterStress(__global MYFLOAT4* clDataX,__global MYFLOAT4* clDat
     }
 
     // cast to float for sin,cos are NEEDED by Mesa FP64 implementation!
-    // Implemention on AMD Oland are probably broken in float
+    // Implementation on AMD Oland are probably broken in float
 
     FromCoM=(MYFLOAT4)(clDataX[gid]-clCoM[0]);
     Length=length(FromCoM);
@@ -387,7 +387,7 @@ __kernel void SplutterStress(__global MYFLOAT4* clDataX,__global MYFLOAT4* clDat
     // Second tangential vector to sphere of length radius
     ThetaB=acos((float)(FromCoM.x/Length));
     PhiB=atan((float)(FromCoM.y/FromCoM.z))+5.e-1f*PI;
-    // (x,y) random coordonates to plane tangential to sphere
+    // (x,y) random coordinates to plane tangential to sphere
     Polar=MWCfp*2.e0f*PI;
     tA=cos((float)Polar);
     tB=sin((float)Polar);
@@ -781,7 +781,7 @@ if __name__ == "__main__":
     Velocity = MyFloat(Velocity)
     Step = MyFloat(Step)
 
-    print("Device choosed : %s" % Device)
+    print("Device chosen : %s" % Device)
     print("Number of particules : %s" % Number)
     print("Size of Shape : %s" % SizeOfShape)
     print("Initial velocity : %s" % Velocity)

{pyopencl-2024.3 → pyopencl-2025.1}/pyopencl/array.py

@@ -32,7 +32,7 @@ import builtins
 from dataclasses import dataclass
 from functools import reduce
 from numbers import Number
-from typing import Any, Dict, Hashable, List, Optional, Tuple, Union
+from typing import Any, Dict, List, Optional, Tuple, Union
 from warnings import warn
 
 import numpy as np
@@ -58,20 +58,14 @@ else:
     _SVMPointer_or_nothing = ()
 
 
-_NUMPY_PRE_2 = np.__version__.startswith("1.")
-
-
 # {{{ _get_common_dtype
 
-_COMMON_DTYPE_CACHE: Dict[Tuple[Hashable, ...], np.dtype] = {}
-
-
 class DoubleDowncastWarning(UserWarning):
     pass
 
 
 _DOUBLE_DOWNCAST_WARNING = (
-        "The operation you requested would result in a double-precisision "
+        "The operation you requested would result in a double-precision "
         "quantity according to numpy semantics. Since your device does not "
         "support double precision, a single-precision quantity is being returned.")
 
@@ -81,78 +75,12 @@ def _get_common_dtype(obj1, obj2, queue):
         raise ValueError("PyOpenCL array has no queue; call .with_queue() to "
                 "add one in order to be able to perform operations")
 
-    allow_double = has_double_support(queue.device)
-    cache_key = None
-    o1_is_array = isinstance(obj1, Array)
-    o2_is_array = isinstance(obj2, Array)
-
-    if o1_is_array and o2_is_array:
-        o1_dtype = obj1.dtype
-        o2_dtype = obj2.dtype
-        cache_key = (obj1.dtype, obj2.dtype, allow_double)
-    else:
-        o1_dtype = getattr(obj1, "dtype", type(obj1))
-        o2_dtype = getattr(obj2, "dtype", type(obj2))
-
-        o1_is_integral = np.issubdtype(o1_dtype, np.integer)
-        o2_is_integral = np.issubdtype(o1_dtype, np.integer)
-
-        o1_key = obj1 if o1_is_integral and not o1_is_array else o1_dtype
-        o2_key = obj2 if o2_is_integral and not o2_is_array else o2_dtype
+    # Note: We are calling np.result_type with pyopencl arrays here.
+    # Luckily, np.result_type only looks at the dtype of input arrays up until
+    # at least numpy v2.1.
+    result = np.result_type(obj1, obj2)
 
-        cache_key = (o1_key, o2_key, o1_is_array, o2_is_array, allow_double)
-
-    try:
-        return _COMMON_DTYPE_CACHE[cache_key]
-    except KeyError:
-        pass
-
-    # Numpy's behavior around integers is a bit bizarre, and definitely value-
-    # and not just type-sensitive when it comes to scalars.  We'll just do our
-    # best to emulate it.
-    #
-    # Some samples that are true as of numpy 1.23.1.
-    #
-    # >>> a = np.zeros(1, dtype=np.int16)
-    # >>> (a + 123123123312).dtype
-    # dtype('int64')
-    # >>> (a + 12312).dtype
-    # dtype('int16')
-    # >>> (a + 12312444).dtype
-    # dtype('int32')
-    # >>> (a + np.int32(12312444)).dtype
-    # dtype('int32')
-    # >>> (a + np.int32(1234)).dtype
-    # dtype('int16')
-    #
-    # Note that np.find_common_type, while appealing, won't be able to tell
-    # the full story.
-
-    if (_NUMPY_PRE_2
-            and not (o1_is_array and o2_is_array)
-            and o1_is_integral and o2_is_integral):
-        if o1_is_array:
-            obj1 = np.zeros(1, dtype=o1_dtype)
-        if o2_is_array:
-            obj2 = np.zeros(1, dtype=o2_dtype)
-
-        result = (obj1 + obj2).dtype
-    else:
-        array_types = []
-        scalars = []
-
-        if o1_is_array:
-            array_types.append(o1_dtype)
-        else:
-            scalars.append(obj1)
-        if o2_is_array:
-            array_types.append(o2_dtype)
-        else:
-            scalars.append(obj2)
-
-        result = np.result_type(*array_types, *scalars)
-
-    if not allow_double:
+    if not has_double_support(queue.device):
         if result == np.float64:
             result = np.dtype(np.float32)
             warn(_DOUBLE_DOWNCAST_WARNING, DoubleDowncastWarning, stacklevel=3)
@@ -160,9 +88,6 @@ def _get_common_dtype(obj1, obj2, queue):
             result = np.dtype(np.complex64)
             warn(_DOUBLE_DOWNCAST_WARNING, DoubleDowncastWarning, stacklevel=3)
 
-    if cache_key is not None:
-        _COMMON_DTYPE_CACHE[cache_key] = result
-
     return result
 
 # }}}
@@ -2518,7 +2443,7 @@ def arange(queue, *args, **kwargs):
         raise TypeError("arange requires a dtype argument")
 
     from math import ceil
-    size = int(ceil((stop-start)/step))
+    size = ceil((stop-start)/step)
 
     result = Array(queue, (size,), dtype, allocator=inf.allocator)
     result.add_event(_arange_knl(result, start, step, queue=queue))

{pyopencl-2024.3 → pyopencl-2025.1}/pyopencl/cltypes.py

@@ -50,7 +50,7 @@ double = np.float64
 # {{{ vector types
 
 def _create_vector_types():
-    _mapping = [(k, globals()[k]) for k in
+    mapping = [(k, globals()[k]) for k in
                 ["char", "uchar", "short", "ushort", "int",
                  "uint", "long", "ulong", "float", "double"]]
 
@@ -64,7 +64,7 @@ def _create_vector_types():
 
     counts = [2, 3, 4, 8, 16]
 
-    for base_name, base_type in _mapping:
+    for base_name, base_type in mapping:
         for count in counts:
             name = "%s%d" % (base_name, count)
 

{pyopencl-2024.3 → pyopencl-2025.1}/pyopencl/version.py

@@ -3,7 +3,7 @@ from importlib import metadata
 
 
 VERSION_TEXT = metadata.version("pyopencl")
-_match = re.match("^([0-9.]+)([a-z0-9]*?)$", VERSION_TEXT)
+_match = re.match(r"^([0-9.]+)([a-z0-9]*?)$", VERSION_TEXT)
 assert _match is not None
 VERSION_STATUS = _match.group(2)
 VERSION = tuple(int(nr) for nr in _match.group(1).split("."))

{pyopencl-2024.3 → pyopencl-2025.1}/pyproject.toml

@@ -11,7 +11,7 @@ requires = [
 
 [project]
 name = "pyopencl"
-version = "2024.3"
+version = "2025.1"
 description = "Python wrapper for OpenCL"
 readme = "README.rst"
 authors = [
@@ -154,6 +154,14 @@ test-command = "pytest {project}/test"
 test-extras = [
     "test",
 ]
+environment-pass = [
+    "CL_INC_DIR",
+    "CL_LIB_DIR",
+]
+test-skip = [
+    "*-macosx_*:arm64",
+    "*-macosx_arm64",
+]
 
 [tool.cibuildwheel.linux]
 skip = [
@@ -205,3 +213,32 @@ skip = [
 ]
 test-command = ""
 before-all = "bash {package}/scripts/build-ocl-windows.sh"
+
+[tool.typos.default]
+extend-ignore-re = [
+  "(?Rm)^.*(#|//)\\s*spellchecker:\\s*disable-line$"
+]
+
+[tool.typos.default.extend-words]
+# for ND Range
+ND = "ND"
+nd = "nd"
+
+# level-of-detail
+LOD = "LOD"
+
+# short for 'series'
+ "ser" = "ser"
+
+# like the numpy function
+ "arange" = "arange"
+
+[tool.typos.files]
+extend-exclude = [
+# No thanks, hex IDs in JSON should not be spellchecked.
+"examples/*.ipynb",
+# Copied from upstream
+"pyopencl/cl/pyopencl-random123/*",
+# This one has comments in French
+"examples/black-hole-accretion.py"
+]

{pyopencl-2024.3 → pyopencl-2025.1}/scripts/build-ocl-macos.sh

@@ -8,11 +8,11 @@ git clone --branch v2022.01.04 https://github.com/KhronosGroup/OpenCL-Headers
 
 
 
-cmake -D CMAKE_INSTALL_PREFIX=./OpenCL-Headers/install -S ./OpenCL-Headers -B ./OpenCL-Headers/build 
+cmake -D CMAKE_INSTALL_PREFIX=./OpenCL-Headers/install -S ./OpenCL-Headers -B ./OpenCL-Headers/build
 cmake --build ./OpenCL-Headers/build --target install
 
 cmake -D CMAKE_PREFIX_PATH=${PWD}/OpenCL-Headers/install -D OPENCL_ICD_LOADER_HEADERS_DIR=${PWD}/OpenCL-Headers/install/include -D CMAKE_INSTALL_PREFIX=./OpenCL-ICD-Loader/install -S ./OpenCL-ICD-Loader -B ./OpenCL-ICD-Loader/build 
 cmake --build ./OpenCL-ICD-Loader/build --target install --config Release
 
 echo "PyOpenCL wheel includes Khronos Group OpenCL-ICD-Loader which is licensed as below" >> ${SCRIPT_DIR}/../LICENSE
-cat ./OpenCL-ICD-Loader/LICENSE >> ${SCRIPT_DIR}/../LICENSE
+cat ./OpenCL-ICD-Loader/LICENSE >> ${SCRIPT_DIR}/../LICENSE

{pyopencl-2024.3 → pyopencl-2025.1}/src/wrap_cl.hpp

@@ -1893,7 +1893,7 @@ namespace pyopencl
         py::object m_py_event;
         py::object m_py_callback;
 
-        bool m_set_callback_suceeded;
+        bool m_set_callback_succeeded;
 
         bool m_notify_thread_wakeup_is_genuine;
 
@@ -1901,7 +1901,7 @@ namespace pyopencl
         cl_int m_command_exec_status;
 
         event_callback_info_t(py::object py_event, py::object py_callback)
-        : m_py_event(py_event), m_py_callback(py_callback), m_set_callback_suceeded(true),
+        : m_py_event(py_event), m_py_callback(py_callback), m_set_callback_succeeded(true),
         m_notify_thread_wakeup_is_genuine(false)
         {}
       };
@@ -1945,7 +1945,7 @@ namespace pyopencl
               {
                 py::gil_scoped_acquire acquire;
 
-                if (cb_info->m_set_callback_suceeded)
+                if (cb_info->m_set_callback_succeeded)
                 {
                   try {
                     cb_info->m_py_callback(
@@ -1982,7 +1982,7 @@ namespace pyopencl
           // be woken up. Wake it up to let it know that it can stop.
           {
             std::lock_guard<std::mutex> lg(cb_info->m_mutex);
-            cb_info->m_set_callback_suceeded = false;
+            cb_info->m_set_callback_succeeded = false;
             cb_info->m_notify_thread_wakeup_is_genuine = true;
           }
           cb_info->m_condvar.notify_one();
@@ -3673,7 +3673,7 @@ namespace pyopencl
 #ifdef PYPY_VERSION
         // FIXME: get a read-only buffer
         // Not quite honest, but Pypy doesn't consider numpy arrays
-        // created from objects with the __aray_interface__ writeable.
+        // created from objects with the __array_interface__ writeable.
         ward->get(holder.ptr(), PyBUF_ANY_CONTIGUOUS);
 #else
         ward->get(holder.ptr(), PyBUF_ANY_CONTIGUOUS | PyBUF_WRITABLE);
@@ -4508,7 +4508,7 @@ namespace pyopencl
           py::tuple name_hdr_tup = py::borrow<py::tuple>(name_hdr_tup_py);
           if (py::len(name_hdr_tup) != 2)
             throw error("Program.compile", CL_INVALID_VALUE,
-                "epxected (name, header) tuple in headers list");
+                "expected (name, header) tuple in headers list");
           std::string name = py::cast<std::string>(name_hdr_tup[0]);
           program &prg = py::cast<program &>(name_hdr_tup[1]);
 

{pyopencl-2024.3 → pyopencl-2025.1}/test/test_array.py

@@ -2393,6 +2393,24 @@ def test_xdg_cache_home(ctx_factory):
 # }}}
 
 
+def test_numpy_type_promotion_with_cl_arrays(ctx_factory):
+    ctx = ctx_factory()
+    queue = cl.CommandQueue(ctx)
+
+    class NotReallyAnArray:
+        @property
+        def dtype(self):
+            return np.dtype("float64")
+
+    # Make sure that np.result_type accesses only the dtype attribute of the
+    # class, not (e.g.) its data.
+    assert np.result_type(42, NotReallyAnArray()) == np.float64
+
+    from pyopencl.array import _get_common_dtype
+    assert _get_common_dtype(42, NotReallyAnArray(), queue) == np.float64
+    assert _get_common_dtype(42.0, NotReallyAnArray(), queue) == np.float64
+
+
 if __name__ == "__main__":
     if len(sys.argv) > 1:
         exec(sys.argv[1])

{pyopencl-2024.3 → pyopencl-2025.1}/test/test_wrapper.py

@@ -865,22 +865,22 @@ def test_user_event(ctx_factory):
     evt = cl.UserEvent(ctx)
     Thread(target=event_waiter1, args=(evt, 1)).start()
     sleep(.05)
-    if status.get(1, False):
+    if status.get(1):
         raise RuntimeError("UserEvent triggered before set_status")
     evt.set_status(cl.command_execution_status.COMPLETE)
     sleep(.05)
-    if not status.get(1, False):
+    if not status.get(1):
         raise RuntimeError("UserEvent.wait timeout")
     assert evt.command_execution_status == cl.command_execution_status.COMPLETE
 
     evt = cl.UserEvent(ctx)
     Thread(target=event_waiter2, args=(evt, 2)).start()
     sleep(.05)
-    if status.get(2, False):
+    if status.get(2):
         raise RuntimeError("UserEvent triggered before set_status")
     evt.set_status(cl.command_execution_status.COMPLETE)
     sleep(.05)
-    if not status.get(2, False):
+    if not status.get(2):
         raise RuntimeError("cl.wait_for_events timeout on UserEvent")
     assert evt.command_execution_status == cl.command_execution_status.COMPLETE
 
@@ -1402,7 +1402,7 @@ def test_empty_ndrange(ctx_factory, empty_shape):
     queue = cl.CommandQueue(ctx)
 
     if ctx._get_cl_version() < (1, 2) or cl.get_cl_header_version() < (1, 2):
-        pytest.skip("OpenCL 1.2 required for empty NDRange suuport")
+        pytest.skip("OpenCL 1.2 required for empty NDRange support")
 
     a = cl_array.zeros(queue, empty_shape, dtype=np.float32)