kernel-tuner 1.1.2__tar.gz → 1.1.3__tar.gz

{kernel_tuner-1.1.2 → kernel_tuner-1.1.3}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.3
 Name: kernel_tuner
-Version: 1.1.2
+Version: 1.1.3
 Summary: An easy to use CUDA/OpenCL kernel tuner in Python
 License: Apache-2.0
 Keywords: auto-tuning,gpu,computing,pycuda,cuda,pyopencl,opencl

{kernel_tuner-1.1.2 → kernel_tuner-1.1.3}/kernel_tuner/backends/pycuda.py

@@ -106,7 +106,7 @@ class PyCudaFunctions(GPUBackend):
         )
         if cc == "00":
             cc = self.context.get_device().compute_capability()
-        self.cc = str(cc[0]) + str(cc[1])
+        self.cc = str(cc)
         self.iterations = iterations
         self.current_module = None
         self.func = None
@@ -180,6 +180,9 @@ class PyCudaFunctions(GPUBackend):
             # pycuda does not support bool, convert to uint8 instead
             elif isinstance(arg, np.bool_):
                 gpu_args.append(arg.astype(np.uint8))
+            # pycuda does not support 16-bit formats, view them as uint16
+            elif isinstance(arg, np.generic) and str(arg.dtype) in ("float16", "bfloat16"):
+                gpu_args.append(arg.view(np.uint16))
             # if not an array, just pass argument along
             else:
                 gpu_args.append(arg)

{kernel_tuner-1.1.2 → kernel_tuner-1.1.3}/kernel_tuner/core.py

@@ -122,7 +122,7 @@ class KernelSource(object):
         The files beyond the first are considered additional files that may also contain tunable parameters
 
         For each file beyond the first this function creates a temporary file with
-        preprocessors statements inserted. Occurences of the original filenames in the
+        preprocessors statements inserted. Occurrences of the original filenames in the
         first file are replaced with their temporary counterparts.
 
         :param kernel_name: A string specifying the kernel name.
@@ -180,7 +180,7 @@ class KernelSource(object):
             temp_file = util.get_temp_filename(suffix="." + f.split(".")[-1])
             temp_files[f] = temp_file
             util.write_file(temp_file, ks)
-            # replace occurences of the additional file's name in the first kernel_string with the name of the temp file
+            # replace occurrences of the additional file's name in the first kernel_string with the name of the temp file
             kernel_string = kernel_string.replace(f, temp_file)
 
         return name, kernel_string, temp_files
@@ -511,7 +511,7 @@ class DeviceInterface(object):
         # run the kernel
         check = self.run_kernel(func, gpu_args, instance)
         if not check:
-            return  # runtime failure occured that should be ignored, skip correctness check
+            return  # runtime failure occurred that should be ignored, skip correctness check
 
         # retrieve gpu results to host memory
         result_host = []
@@ -595,7 +595,7 @@ class DeviceInterface(object):
                     if kernel_options.texmem_args is not None:
                         self.dev.copy_texture_memory_args(kernel_options.texmem_args)
 
-                # stop compilation stopwatch and convert to miliseconds
+                # stop compilation stopwatch and convert to milliseconds
                 last_compilation_time = 1000 * (time.perf_counter() - start_compilation)
 
                 # test kernel for correctness
@@ -628,7 +628,7 @@ class DeviceInterface(object):
                 )
                 raise e
 
-            # clean up any temporary files, if no error occured
+            # clean up any temporary files, if no error occurred
             instance.delete_temp_files()
 
         result["compile_time"] = last_compilation_time or 0
@@ -815,9 +815,9 @@ def _default_verify_function(instance, answer, result_host, atol, verbose):
             if isinstance(answer[i], (np.ndarray, cp.ndarray)) and isinstance(
                 arg, (np.ndarray, cp.ndarray)
             ):
-                if answer[i].dtype != arg.dtype:
+                if not np.can_cast(arg.dtype, answer[i].dtype):
                     raise TypeError(
-                        f"Element {i} of the expected results list is not of the same dtype as the kernel output: "
+                        f"Element {i} of the expected results list has a dtype that is not compatible with the dtype of the kernel output: "
                         + str(answer[i].dtype)
                         + " != "
                         + str(arg.dtype)
@@ -905,7 +905,7 @@ def _default_verify_function(instance, answer, result_host, atol, verbose):
                     + " detected during correctness check"
                 )
                 print(
-                    "this error occured when checking value of the %oth kernel argument"
+                    "this error occurred when checking value of the %oth kernel argument"
                     % (i,)
                 )
                 print(
@@ -934,7 +934,7 @@ def split_argument_list(argument_list):
         match = re.match(regex, arg, re.S)
         if not match:
             raise ValueError("error parsing templated kernel argument list")
-        type_list.append(re.sub(r"\s+", " ", match.group(1).strip(), re.S))
+        type_list.append(re.sub(r"\s+", " ", match.group(1).strip(), flags=re.S))
         name_list.append(match.group(2).strip())
     return type_list, name_list
 
@@ -956,7 +956,7 @@ def apply_template_typenames(type_list, templated_typenames):
             # if the templated typename occurs as a token in the string, meaning that it is enclosed in
             # beginning of string or whitespace, and end of string, whitespace or star
             regex = r"(^|\s+)(" + k + r")($|\s+|\*)"
-            sub = re.sub(regex, replace_typename_token, arg_type, re.S)
+            sub = re.sub(regex, replace_typename_token, arg_type, flags=re.S)
             type_list[i] = sub
 
 

{kernel_tuner-1.1.2 → kernel_tuner-1.1.3}/pyproject.toml

@@ -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 = "1.1.2" # adhere to PEP440 versioning: https://packaging.python.org/en/latest/guides/distributing-packages-using-setuptools/#id55
+version = "1.1.3" # adhere to PEP440 versioning: https://packaging.python.org/en/latest/guides/distributing-packages-using-setuptools/#id55
 license = "Apache-2.0"
 authors = [
     "Ben van Werkhoven <b.vanwerkhoven@esciencecenter.nl>",