diff --git a/pyopencl/__init__.py b/pyopencl/__init__.py
index 45e14ed5f3451dc03a1f10c78b3415ebc49336d2..4f49df0525948329621365c5c0759bdaa065c938 100644
--- a/pyopencl/__init__.py
+++ b/pyopencl/__init__.py
@@ -648,11 +648,7 @@ def _add_functionality():
err_gen = PythonCodeGenerator()
- err_gen("try:")
- with Indentation(err_gen):
- err_gen.extend(body)
- err_gen("except TypeError as e:")
- with Indentation(err_gen):
+ def gen_error_handler():
err_gen("""
if current_arg is not None:
args = [{args}]
@@ -671,6 +667,16 @@ def _add_functionality():
.format(args=", ".join(arg_names)))
err_gen("")
+ err_gen("try:")
+ with Indentation(err_gen):
+ err_gen.extend(body)
+ err_gen("except TypeError as e:")
+ with Indentation(err_gen):
+ gen_error_handler()
+ err_gen("except _cl.LogicError as e:")
+ with Indentation(err_gen):
+ gen_error_handler()
+
# }}}
def add_preamble(gen):
@@ -715,10 +721,10 @@ def _add_functionality():
# }}}
- def kernel__generate_buffer_arg_setter(self, gen, arg_idx, buf_var):
+ def kernel__generate_buffer_arg_setter(self, gen, arg_idx, buf_var, could_be_numpy_scalar):
from pytools.py_codegen import Indentation
- if _CPY2:
+ if _CPY2 and could_be_numpy_scalar:
# https://github.com/numpy/numpy/issues/5381
gen("if isinstance({buf_var}, np.generic):".format(buf_var=buf_var))
with Indentation(gen):
@@ -747,7 +753,8 @@ def _add_functionality():
gen("else:")
with Indentation(gen):
- self._generate_buffer_arg_setter(gen, arg_idx, arg_var)
+ self._generate_buffer_arg_setter(gen, arg_idx, arg_var,
+ could_be_numpy_scalar=True)
def kernel__generate_naive_call(self):
num_args = self.num_args
@@ -783,9 +790,10 @@ def _add_functionality():
has_struct_arg_count_bug(dev)
for dev in self.context.devices]
+ from pytools import single_valued
if any(count_bug_per_dev):
if all(count_bug_per_dev):
- work_around_arg_count_bug = True
+ work_around_arg_count_bug = single_valued(count_bug_per_dev)
else:
warn_about_arg_count_bug = True
@@ -834,25 +842,38 @@ def _add_functionality():
else:
raise TypeError("unexpected complex type: %s" % arg_dtype)
- if (work_around_arg_count_bug
+ if (work_around_arg_count_bug == "pocl"
and arg_dtype == np.complex128
and fp_arg_count + 2 <= 8):
gen(
"buf = pack('{arg_char}', {arg_var}.real)"
.format(arg_char=arg_char, arg_var=arg_var))
- self._generate_buffer_arg_setter(gen, cl_arg_idx, "buf")
+ self._generate_buffer_arg_setter(gen, cl_arg_idx, "buf",
+ could_be_numpy_scalar=False)
cl_arg_idx += 1
+ gen("current_arg = current_arg + 1000")
gen(
"buf = pack('{arg_char}', {arg_var}.imag)"
.format(arg_char=arg_char, arg_var=arg_var))
- self._generate_buffer_arg_setter(gen, cl_arg_idx, "buf")
+ self._generate_buffer_arg_setter(gen, cl_arg_idx, "buf",
+ could_be_numpy_scalar=False)
cl_arg_idx += 1
+
+ elif (work_around_arg_count_bug == "apple"
+ and arg_dtype == np.complex128
+ and fp_arg_count + 2 <= 8):
+ raise NotImplementedError("No work-around to "
+ "Apple's broken structs-as-kernel arg "
+ "handling has been found. "
+ "Cannot pass complex numbers to kernels.")
+
else:
gen(
"buf = pack('{arg_char}{arg_char}', "
"{arg_var}.real, {arg_var}.imag)"
.format(arg_char=arg_char, arg_var=arg_var))
- self._generate_buffer_arg_setter(gen, cl_arg_idx, "buf")
+ self._generate_buffer_arg_setter(gen, cl_arg_idx, "buf",
+ could_be_numpy_scalar=False)
cl_arg_idx += 1
fp_arg_count += 2
@@ -864,7 +885,8 @@ def _add_functionality():
gen(
"buf = pack('{arg_char}', long({arg_var}))"
.format(arg_char=arg_dtype.char, arg_var=arg_var))
- self._generate_buffer_arg_setter(gen, cl_arg_idx, "buf")
+ self._generate_buffer_arg_setter(gen, cl_arg_idx, "buf",
+ could_be_numpy_scalar=False)
cl_arg_idx += 1
else:
@@ -878,7 +900,8 @@ def _add_functionality():
.format(
arg_char=arg_char,
arg_var=arg_var))
- self._generate_buffer_arg_setter(gen, cl_arg_idx, "buf")
+ self._generate_buffer_arg_setter(gen, cl_arg_idx, "buf",
+ could_be_numpy_scalar=False)
cl_arg_idx += 1
gen("")
diff --git a/pyopencl/characterize/__init__.py b/pyopencl/characterize/__init__.py
index e4676a68d4aacc06ba4073664ab0925cbe30c366..175a951d42f2e3e531ead13bafe0128e3fc6a092 100644
--- a/pyopencl/characterize/__init__.py
+++ b/pyopencl/characterize/__init__.py
@@ -328,7 +328,7 @@ def has_struct_arg_count_bug(dev):
"""
if dev.platform.name == "Apple" and dev.type & cl.device_type.CPU:
- return True
+ return "apple"
if dev.platform.name == "Portable Computing Language":
- return True
+ return "pocl"
return False
diff --git a/test/test_algorithm.py b/test/test_algorithm.py
index 8f44cc594bab73126af98849811b35d99eebd059..43e88ca8c25c2b5594bddbab8df8f12cd96d2aef 100644
--- a/test/test_algorithm.py
+++ b/test/test_algorithm.py
@@ -37,7 +37,7 @@ import pyopencl as cl
import pyopencl.array as cl_array # noqa
from pyopencl.tools import ( # noqa
pytest_generate_tests_for_pyopencl as pytest_generate_tests)
-from pyopencl.characterize import has_double_support
+from pyopencl.characterize import has_double_support, has_struct_arg_count_bug
from pyopencl.scan import InclusiveScanKernel, ExclusiveScanKernel
@@ -326,9 +326,14 @@ def test_dot(ctx_factory):
dev = context.devices[0]
+ import faulthandler
+ faulthandler.enable()
dtypes = [np.float32, np.complex64]
if has_double_support(dev):
- dtypes.extend([np.float64, np.complex128])
+ if has_struct_arg_count_bug(dev) == "apple":
+ dtypes.extend([np.float64])
+ else:
+ dtypes.extend([np.float64, np.complex128])
for a_dtype in dtypes:
for b_dtype in dtypes:
@@ -853,7 +858,11 @@ def test_bitonic_sort(ctx_factory, size, dtype):
ctx = cl.create_some_context()
queue = cl.CommandQueue(ctx)
- if (ctx.devices[0].platform.name == "Portable Computing Language"
+ dev = ctx.devices[0]
+ if (dev.platform.name == "Apple" and dev.type & cl.device_type.CPU):
+ pytest.xfail("Bitonic sort won't work on Apple CPU: no workgroup "
+ "parallelism")
+ if (dev.platform.name == "Portable Computing Language"
and dtype == np.float64):
pytest.xfail("Double precision bitonic sort doesn't work on POCL")
@@ -882,7 +891,11 @@ def test_bitonic_argsort(ctx_factory, size, dtype):
ctx = cl.create_some_context()
queue = cl.CommandQueue(ctx)
- if (ctx.devices[0].platform.name == "Portable Computing Language"
+ dev = ctx.devices[0]
+ if (dev.platform.name == "Apple" and dev.type & cl.device_type.CPU):
+ pytest.xfail("Bitonic sort won't work on Apple CPU: no workgroup "
+ "parallelism")
+ if (dev.platform.name == "Portable Computing Language"
and dtype == np.float64):
pytest.xfail("Double precision bitonic sort doesn't work on POCL")
diff --git a/test/test_array.py b/test/test_array.py
index e183a4de0d16aed400bdd7b146f8c85031380921..f3b6a6687379ea84b4962d2eba38f7f832a13007 100644
--- a/test/test_array.py
+++ b/test/test_array.py
@@ -35,7 +35,7 @@ import pyopencl.array as cl_array
import pyopencl.tools as cl_tools
from pyopencl.tools import ( # noqa
pytest_generate_tests_for_pyopencl as pytest_generate_tests)
-from pyopencl.characterize import has_double_support
+from pyopencl.characterize import has_double_support, has_struct_arg_count_bug
from pyopencl.cffi_cl import _PYPY
@@ -102,7 +102,12 @@ def test_mix_complex(ctx_factory):
#(np.int32, np.complex64),
]
- if has_double_support(context.devices[0]):
+ dev = context.devices[0]
+ if has_double_support(dev) and has_struct_arg_count_bug(dev) == "apple":
+ dtypes.extend([
+ (np.float32, np.float64),
+ ])
+ elif has_double_support(dev):
dtypes.extend([
(np.float32, np.float64),
(np.float32, np.complex128),
@@ -169,6 +174,9 @@ def test_pow_neg1_vs_inv(ctx_factory):
if not has_double_support(device):
from pytest import skip
skip("double precision not supported on %s" % device)
+ if has_struct_arg_count_bug(device) == "apple":
+ from pytest import xfail
+ xfail("apple struct arg counting broken")
a_dev = make_random_array(queue, np.complex128, 20000)
diff --git a/test/test_clmath.py b/test/test_clmath.py
index 4dea8f91963e59f5a2ca2e3018cd73bf9287296f..aaa89daf18740ca09d6d999652a7bad641edf29a 100644
--- a/test/test_clmath.py
+++ b/test/test_clmath.py
@@ -29,22 +29,13 @@ import numpy as np
import pytest
-
-def have_cl():
- try:
- import pyopencl # noqa
- return True
- except:
- return False
-
-if have_cl():
- import pyopencl.array as cl_array
- import pyopencl as cl
- import pyopencl.clmath as clmath
- from pyopencl.tools import ( # noqa
- pytest_generate_tests_for_pyopencl
- as pytest_generate_tests)
- from pyopencl.characterize import has_double_support
+import pyopencl.array as cl_array
+import pyopencl as cl
+import pyopencl.clmath as clmath
+from pyopencl.tools import ( # noqa
+ pytest_generate_tests_for_pyopencl
+ as pytest_generate_tests)
+from pyopencl.characterize import has_double_support, has_struct_arg_count_bug
try:
import faulthandler
@@ -76,8 +67,12 @@ def make_unary_function_test(name, limits=(0, 1), threshold=0, use_complex=False
gpu_func = getattr(clmath, name)
cpu_func = getattr(np, numpy_func_names.get(name, name))
- if has_double_support(context.devices[0]):
- if use_complex:
+ dev = context.devices[0]
+
+ if has_double_support(dev):
+ if use_complex and has_struct_arg_count_bug(dev) == "apple":
+ dtypes = [np.float32, np.float64, np.complex64]
+ elif use_complex:
dtypes = [np.float32, np.float64, np.complex64, np.complex128]
else:
dtypes = [np.float32, np.float64]
@@ -110,26 +105,25 @@ def make_unary_function_test(name, limits=(0, 1), threshold=0, use_complex=False
return test
-if have_cl():
- test_ceil = make_unary_function_test("ceil", (-10, 10))
- test_floor = make_unary_function_test("ceil", (-10, 10))
- test_fabs = make_unary_function_test("fabs", (-10, 10))
- test_exp = make_unary_function_test("exp", (-3, 3), 1e-5, use_complex=True)
- test_log = make_unary_function_test("log", (1e-5, 1), 1e-6, use_complex=True)
- test_log10 = make_unary_function_test("log10", (1e-5, 1), 5e-7)
- test_sqrt = make_unary_function_test("sqrt", (1e-5, 1), 3e-7, use_complex=True)
-
- test_sin = make_unary_function_test("sin", (-10, 10), 2e-7, use_complex=2e-2)
- test_cos = make_unary_function_test("cos", (-10, 10), 2e-7, use_complex=2e-2)
- test_asin = make_unary_function_test("asin", (-0.9, 0.9), 5e-7)
- test_acos = make_unary_function_test("acos", (-0.9, 0.9), 5e-7)
- test_tan = make_unary_function_test("tan",
- (-math.pi/2 + 0.1, math.pi/2 - 0.1), 4e-5, use_complex=True)
- test_atan = make_unary_function_test("atan", (-10, 10), 2e-7)
-
- test_sinh = make_unary_function_test("sinh", (-3, 3), 3e-6, use_complex=2e-3)
- test_cosh = make_unary_function_test("cosh", (-3, 3), 3e-6, use_complex=2e-3)
- test_tanh = make_unary_function_test("tanh", (-3, 3), 2e-6, use_complex=True)
+test_ceil = make_unary_function_test("ceil", (-10, 10))
+test_floor = make_unary_function_test("ceil", (-10, 10))
+test_fabs = make_unary_function_test("fabs", (-10, 10))
+test_exp = make_unary_function_test("exp", (-3, 3), 1e-5, use_complex=True)
+test_log = make_unary_function_test("log", (1e-5, 1), 1e-6, use_complex=True)
+test_log10 = make_unary_function_test("log10", (1e-5, 1), 5e-7)
+test_sqrt = make_unary_function_test("sqrt", (1e-5, 1), 3e-7, use_complex=True)
+
+test_sin = make_unary_function_test("sin", (-10, 10), 2e-7, use_complex=2e-2)
+test_cos = make_unary_function_test("cos", (-10, 10), 2e-7, use_complex=2e-2)
+test_asin = make_unary_function_test("asin", (-0.9, 0.9), 5e-7)
+test_acos = make_unary_function_test("acos", (-0.9, 0.9), 5e-7)
+test_tan = make_unary_function_test("tan",
+ (-math.pi/2 + 0.1, math.pi/2 - 0.1), 4e-5, use_complex=True)
+test_atan = make_unary_function_test("atan", (-10, 10), 2e-7)
+
+test_sinh = make_unary_function_test("sinh", (-3, 3), 3e-6, use_complex=2e-3)
+test_cosh = make_unary_function_test("cosh", (-3, 3), 3e-6, use_complex=2e-3)
+test_tanh = make_unary_function_test("tanh", (-3, 3), 2e-6, use_complex=True)
def test_atan2(ctx_factory):