def __init__(self, shape, strides, dtype, dgpu_data=None):
"""
Args
----
shape
array shape.
strides
array strides.
dtype
data type as numpy.dtype.
dgpu_data
user provided device memory for the ndarray data buffer
"""
if isinstance(shape, (int, long)):
shape = (shape,)
if isinstance(strides, (int, long)):
strides = (strides,)
self.ndim = len(shape)
if len(strides) != self.ndim:
raise ValueError("strides not match ndim")
self._dummy = dummyarray.Array.from_desc(0, shape, strides, dtype.itemsize)
self.shape = tuple(shape)
self.strides = tuple(strides)
self.dtype = np.dtype(dtype)
self.size = int(np.prod(self.shape))
# prepare dgpu memory
if self.size > 0:
if dgpu_data is None:
from numba.roc.api import _memory_size_from_info
self.alloc_size = _memory_size_from_info(
self.shape, self.strides, self.dtype.itemsize
)
# find a coarse region on the dGPU
dgpu_data = devices.get_context().mempoolalloc(self.alloc_size)
else: # we have some preallocated dgpu_memory
sz = getattr(dgpu_data, "_hsa_memsize_", None)
if sz is None:
raise ValueError("dgpu_data as no _hsa_memsize_ attribute")
assert sz >= 0
self.alloc_size = sz
else:
dgpu_data = None
self.alloc_size = 0
self.dgpu_data = dgpu_data
def __init__(self, shape, strides, dtype, dgpu_data=None):
"""
Args
----
shape
array shape.
strides
array strides.
dtype
data type as numpy.dtype.
dgpu_data
user provided device memory for the ndarray data buffer
"""
if isinstance(shape, (int, long)):
shape = (shape,)
if isinstance(strides, (int, long)):
strides = (strides,)
self.ndim = len(shape)
if len(strides) != self.ndim:
raise ValueError('strides not match ndim')
self._dummy = dummyarray.Array.from_desc(0, shape, strides,
dtype.itemsize)
self.shape = tuple(shape)
self.strides = tuple(strides)
self.dtype = np.dtype(dtype)
self.size = int(np.prod(self.shape))
# prepare dgpu memory
if self.size > 0:
if dgpu_data is None:
from numba.roc.api import _memory_size_from_info
self.alloc_size = _memory_size_from_info(self.shape,
self.strides, self.dtype.itemsize)
# find a coarse region on the dGPU
dgpu_data = devices.get_context().mempoolalloc(self.alloc_size)
else: # we have some preallocated dgpu_memory
sz = getattr(dgpu_data, '_hsa_memsize_', None)
if sz is None:
raise ValueError('dgpu_data as no _hsa_memsize_ attribute')
assert sz >= 0
self.alloc_size = sz
else:
dgpu_data = None
self.alloc_size = 0
self.dgpu_data = dgpu_data