Merge pull request #176 from loostrum/allow_bool_arg

Support boolean datatype in PyCUDA backend.

Author: benvanwerkhoven

CHANGELOG.md

@@ -9,6 +9,7 @@ This project adheres to [Semantic Versioning](http://semver.org/).
 - Support for locked clocks in NVMLObserver
 - Support for measuring core voltages using NVML
 - Support for custom preprocessor definitions
+- Support for boolean scalar arguments in PyCUDA backend
 
 ### Changed
 - Migrated from github.com/benvanwerkhoven to github.com/KernelTuner

kernel_tuner/pycuda.py

@@ -188,6 +188,9 @@ def ready_argument_list(self, arguments):
                     gpu_args.append(Holder(arg))
                 else:
                     gpu_args.append(Holder(arg.cuda()))
+            # pycuda does not support bool, convert to uint8 instead
+            elif isinstance(arg, np.bool_):
+                gpu_args.append(arg.astype(np.uint8))
             else:    # if not an array, just pass argument along
                 gpu_args.append(arg)
         return gpu_args

kernel_tuner/util.py

@@ -60,6 +60,7 @@ class StopCriterionReached(Exception):
 def check_argument_type(dtype, kernel_argument):
     """check if the numpy.dtype matches the type used in the code"""
     types_map = {
+        "bool": ["bool"],
         "uint8": ["uchar", "unsigned char", "uint8_t"],
         "int8": ["char", "int8_t"],
         "uint16": ["ushort", "unsigned short", "uint16_t"],

test/test_pycuda_functions.py

@@ -17,16 +17,18 @@ def test_ready_argument_list():
     size = 1000
     a = np.int32(75)
     b = np.random.randn(size).astype(np.float32)
-    c = np.zeros_like(b)
+    c = np.bool_(True)
+    d = np.zeros_like(b)
 
-    arguments = [c, a, b]
+    arguments = [d, a, b, c]
 
     dev = kt_pycuda.PyCudaFunctions(0)
     gpu_args = dev.ready_argument_list(arguments)
 
     assert isinstance(gpu_args[0], pycuda.driver.DeviceAllocation)
     assert isinstance(gpu_args[1], np.int32)
     assert isinstance(gpu_args[2], pycuda.driver.DeviceAllocation)
+    assert isinstance(gpu_args[3], np.uint8)
 
 
 @skip_if_no_pycuda