def take(a, indices, out=None, stream=None):
if out is None:
out = GPUArray(indices.shape, a.dtype, a.allocator)
assert len(indices.shape) == 1
func, tex_src = elementwise.get_take_kernel(a.dtype, indices.dtype)
a.bind_to_texref_ext(tex_src[0], allow_double_hack=True)
func.prepared_async_call(out._grid, out._block, stream, indices.gpudata, out.gpudata, indices.size)
return out
def take(a, indices, out=None, stream=None):
if out is None:
out = GPUArray(indices.shape, a.dtype, a.allocator)
assert len(indices.shape) == 1
func, tex_src = elementwise.get_take_kernel(a.dtype, indices.dtype)
a.bind_to_texref_ext(tex_src[0], allow_double_hack=True, allow_complex_hack=True)
func.prepared_async_call(out._grid, out._block, stream,
indices.gpudata, out.gpudata, indices.size)
return out
def make_func_for_chunk_size(chunk_size):
return elementwise.get_take_kernel(a_dtype, indices.dtype,
vec_count=chunk_size)
def make_func_for_chunk_size(chunk_size):
return elementwise.get_take_kernel(a_dtype, indices.dtype,
vec_count=chunk_size)
def make_func_for_chunk_size(chunk_size):
func, tex_src = elementwise.get_take_kernel(a_dtype, indices.dtype,
vec_count=chunk_size)
func.set_block_shape(*indices._block)
return func, tex_src
