def test_printfloat(self):
jprintfloat = cuda.jit('void()', debug=False)(printfloat)
with captured_cuda_stdout() as stdout:
jprintfloat()
# CUDA and the simulator use different formats for float formatting
self.assertIn(stdout.getvalue(), ["0 23 34.750000 321\n",
"0 23 34.75 321\n"])
def test_printfloat(self):
jprintfloat = cuda.jit('void()', debug=False)(printfloat)
with captured_cuda_stdout() as stdout:
jprintfloat()
# CUDA and the simulator use different formats for float formatting
self.assertIn(stdout.getvalue(), ["0 23 34.750000 321\n",
"0 23 34.75 321\n"])
def test_too_many_args(self):
# Tests that we emit the format string and warn when there are more
# than 32 arguments, in common with CUDA C/C++ printf - this is due to
# a limitation in CUDA vprintf, see:
# https://docs.nvidia.com/cuda/cuda-c-programming-guide/index.html#limitations
cufunc = cuda.jit(print_too_many)
r = np.arange(33)
with captured_cuda_stdout() as stdout:
with warnings.catch_warnings(record=True) as w:
cufunc[1, 1](r)
# Check that the format string was printed instead of formatted garbage
expected_fmt_string = ' '.join(['%lld' for _ in range(33)])
self.assertIn(expected_fmt_string, stdout.getvalue())
# Check for the expected warning about formatting more than 32 items
for warning in w:
warnobj = warning.message
if isinstance(warnobj, NumbaWarning):
expected = ('CUDA print() cannot print more than 32 items. '
'The raw format string will be emitted by the '
'kernel instead.')
if warnobj.msg == expected:
break
else:
self.fail('Expected a warning for printing more than 32 items')
def test_string(self):
cufunc = cuda.jit('void()', debug=False)(printstring)
with captured_cuda_stdout() as stdout:
cufunc[1, 3]()
out = stdout.getvalue()
lines = sorted(out.splitlines(True))
expected = ['%d hop! 999\n' % i for i in range(3)]
self.assertEqual(lines, expected)
def test_cuhello(self):
jcuhello = cuda.jit('void()', debug=False)(cuhello)
with captured_cuda_stdout() as stdout:
jcuhello[2, 3]()
# The output of GPU threads is intermingled, just sanity check it
out = stdout.getvalue()
expected = ''.join('%d 999\n' % i for i in range(6))
self.assertEqual(sorted(out), sorted(expected))
def test_cuhello(self):
jcuhello = cuda.jit('void()', debug=False)(cuhello)
with captured_cuda_stdout() as stdout:
jcuhello[2, 3]()
# The output of GPU threads is intermingled, just sanity check it
out = stdout.getvalue()
expected = ''.join('%d 999\n' % i for i in range(6))
self.assertEqual(sorted(out), sorted(expected))
def test_string(self):
cufunc = cuda.jit("void()", debug=False)(printstring)
with captured_cuda_stdout() as stdout:
cufunc[1, 3]()
out = stdout.getvalue()
lines = sorted(out.splitlines(True))
expected = ["%d hop! 999\n" % i for i in range(3)]
self.assertEqual(lines, expected)
def test_cuhello(self):
jcuhello = cuda.jit('void()', debug=False)(cuhello)
with captured_cuda_stdout() as stdout:
jcuhello[2, 3]()
# The output of GPU threads is intermingled, but each print()
# call is still atomic
out = stdout.getvalue()
lines = sorted(out.splitlines(True))
expected = ['-42\n'] * 6 + ['%d 999\n' % i for i in range(6)]
self.assertEqual(lines, expected)
def test_cuhello(self):
jcuhello = cuda.jit("void()", debug=False)(cuhello)
with captured_cuda_stdout() as stdout:
jcuhello[2, 3]()
# The output of GPU threads is intermingled, but each print()
# call is still atomic
out = stdout.getvalue()
lines = sorted(out.splitlines(True))
expected = ["-42\n"] * 6 + ["%d 999\n" % i for i in range(6)]
self.assertEqual(lines, expected)
def test_printempty(self):
cufunc = cuda.jit('void()', debug=False)(printempty)
with captured_cuda_stdout() as stdout:
cufunc()
self.assertEqual(stdout.getvalue(), "\n")
def test_printempty(self):
cufunc = cuda.jit('void()', debug=False)(printempty)
with captured_cuda_stdout() as stdout:
cufunc()
self.assertEqual(stdout.getvalue(), "\n")
def test_string(self):
cufunc = cuda.jit('void()', debug=False)(printstring)
with captured_cuda_stdout() as stdout:
cufunc()
self.assertEqual(stdout.getvalue(), "0 hop! 999\n")
