def test_set_constant_array_errors(mock_4_device_context):
context = mock_4_device_context
api = API.from_api_id(mock_4_device_context.api.id)
other_context = Context.from_criteria(api)
other_queue = Queue(other_context.devices[0])
# Contexts don't know about each other and can't interact with stack in a consistent manner.
# So we deactivate the other context if we're on CUDA API.
if api.id == cuda_api_id():
other_context.deactivate()
cm1 = numpy.arange(16).astype(numpy.int32)
src = MockDefTemplate(kernels=[
MockKernel('kernel', [],
max_total_local_sizes={
0: 1024,
1: 1024,
2: 1024,
3: 1024
})
],
constant_mem={'cm1': cm1.size * cm1.dtype.itemsize})
queue = Queue(context.devices[0])
if context.api.id == cuda_api_id():
program = Program(context.devices, src, constant_arrays=dict(cm1=cm1))
with pytest.raises(
ValueError,
match=
"The provided queue must belong to the same context as this program uses"
):
program.set_constant_array(other_queue, 'cm1', cm1)
with pytest.raises(TypeError, match="Unsupported array type"):
program.set_constant_array(queue, 'cm1', [1])
with pytest.raises(ValueError,
match="Incorrect size of the constant buffer;"):
program.set_constant_array(queue, 'cm1', cm1[:8])
with pytest.raises(TypeError,
match="Unknown constant array metadata type"):
program = Program(context.devices[[0, 1, 2]],
src,
constant_arrays=dict(cm1=1))
program = Program(context.devices[[0, 1, 2]],
src,
constant_arrays=dict(cm1=cm1))
queue3 = Queue(context.devices[3])
with pytest.raises(
ValueError,
match=
"The program was not compiled for the device this queue uses"):
program.set_constant_array(queue3, 'cm1', cm1)
else:
with pytest.raises(
ValueError,
match=
"Compile-time constant arrays are only supported by CUDA API"):
program = Program(context.devices,
src,
constant_arrays=dict(cm1=cm1))
program = Program(context.devices, src)
with pytest.raises(
ValueError,
match="Constant arrays are only supported for CUDA API"):
program.set_constant_array(queue, 'cm1', cm1)
with pytest.raises(
ValueError,
match=
"Compile-time constant arrays are only supported by CUDA API"):
sk = StaticKernel(context.devices,
src,
'kernel',
1024,
constant_arrays=dict(cm1=cm1))
sk = StaticKernel(context.devices, src, 'kernel', 1024)
with pytest.raises(
ValueError,
match="Constant arrays are only supported for CUDA API"):
sk.set_constant_array(queue, 'cm1', cm1)
def _test_constant_memory(context, mocked, is_static):
cm1 = numpy.arange(16).astype(numpy.int32)
cm2 = numpy.arange(16).astype(numpy.int32) * 2 + 1
cm3 = numpy.arange(16).astype(numpy.int32) * 3 + 2
if mocked:
kernel = MockKernel('copy_from_cm', [None] if context.api.id
== cuda_api_id() else [None, None, None, None],
max_total_local_sizes={0: 1024})
src = MockDefTemplate(constant_mem={
'cm1': cm1.size * cm1.dtype.itemsize,
'cm2': cm2.size * cm2.dtype.itemsize,
'cm3': cm3.size * cm3.dtype.itemsize
},
kernels=[kernel])
else:
src = SRC_CONSTANT_MEM_STATIC if is_static else SRC_CONSTANT_MEM
queue = Queue(context.device)
cm1_dev = Array.from_host(queue, cm1)
cm2_dev = Array.from_host(queue, cm2)
cm3_dev = Array.from_host(queue, cm3)
res_dev = Array.empty(context.device, 16, numpy.int32)
if context.api.id == cuda_api_id():
# Use different forms of constant array representation
constant_arrays = dict(
cm1=cm1, # as an array(-like) object
cm2=(cm2.shape, cm2.dtype), # as a tuple of shape and dtype
cm3=cm3_dev) # as a device array
if is_static:
copy_from_cm = StaticKernel(context.device,
src,
'copy_from_cm',
global_size=16,
constant_arrays=constant_arrays)
copy_from_cm.set_constant_array(
queue, 'cm1', cm1_dev) # setting from a device array
copy_from_cm.set_constant_array(queue, 'cm2',
cm2) # setting from a host array
copy_from_cm.set_constant_array(
queue, 'cm3', cm3_dev.data) # setting from a host buffer
else:
program = Program(context.device,
src,
constant_arrays=constant_arrays)
program.set_constant_array(queue, 'cm1',
cm1_dev) # setting from a device array
program.set_constant_array(queue, 'cm2',
cm2) # setting from a host array
program.set_constant_array(
queue, 'cm3', cm3_dev.data) # setting from a host buffer
copy_from_cm = lambda queue, *args: program.kernel.copy_from_cm(
queue, 16, None, *args)
copy_from_cm(queue, res_dev)
else:
if is_static:
copy_from_cm = StaticKernel(context.device,
src,
'copy_from_cm',
global_size=16)
else:
program = Program(context.device, src)
copy_from_cm = lambda queue, *args: program.kernel.copy_from_cm(
queue, 16, None, *args)
copy_from_cm(queue, res_dev, cm1_dev, cm2_dev, cm3_dev)
res = res_dev.get(queue)
if not mocked:
assert (res == cm1 + cm2 + cm3).all()
