def test_local_gpualloc_empty():
i = theano.tensor.iscalar()
ii = theano.tensor.iscalar()
# Test with vector
# Should not be moved as the only client is the output
a = tensor.AllocEmpty('float32')(i)
f = theano.function([i], a, mode=mode_with_gpu)
topo = f.maker.fgraph.toposort()
assert len(topo) == 1
assert isinstance(topo[0].op, theano.tensor.AllocEmpty)
# This return not initilized data, so we can only check the shape
assert f(3).shape == (3, )
# Test with vector
# Should be moved
a = tensor.AllocEmpty('float32')(i)
f = theano.function([i], a.cumsum(), mode=mode_with_gpu)
topo = f.maker.fgraph.toposort()
assert len(topo) == 3
assert isinstance(topo[0].op, GpuAllocEmpty)
# This return not initilized data, so we can only check the shape
assert f(3).shape == (3, )
# Test with matrix
a = tensor.AllocEmpty('float32')(i, ii)
f = theano.function([i, ii], a.cumsum(axis=0), mode=mode_with_gpu)
topo = f.maker.fgraph.toposort()
assert len(topo) == 3
assert isinstance(topo[0].op, GpuAllocEmpty)
# This return not initilized data, so we can only check the shape
assert f(3, 4).shape == (3, 4)
def test_gpuallocempty():
f_gpu = theano.function([], tensor.AllocEmpty('float32')(2,3),
mode=mode_with_gpu)
l_gpu = f_gpu.maker.fgraph.toposort()
assert numpy.any([isinstance(x.op, basic_ops.GpuAllocEmpty) for x in l_gpu])
f_cpu = theano.function([], tensor.AllocEmpty('int32')(2,3))
l_cpu = f_cpu.maker.fgraph.toposort()
assert not numpy.any([isinstance(x.op, basic_ops.GpuAllocEmpty) for x in l_cpu])
def test_local_gpualloc_empty():
i = theano.tensor.iscalar()
ii = theano.tensor.iscalar()
# Test with vector
a = tensor.AllocEmpty('float32')(i)
f = theano.function([i], a, mode=mode_with_gpu)
topo = f.maker.fgraph.toposort()
assert len(topo) == 2
assert isinstance(topo[0].op, GpuAllocEmpty)
# This return not initilized data, so we can only check the shape
assert f(3).shape == (3, )
# Test with matrix
a = tensor.AllocEmpty('float32')(i, ii)
f = theano.function([i, ii], a, mode=mode_with_gpu)
topo = f.maker.fgraph.toposort()
assert len(topo) == 2
assert isinstance(topo[0].op, GpuAllocEmpty)
# This return not initilized data, so we can only check the shape
assert f(3, 4).shape == (3, 4)
def expand_empty(tensor_var, size):
"""
Transforms the shape of a tensor from (d1, d2 ... ) to ( d1+size, d2, ..)
by adding uninitialized memory at the end of the tensor.
"""
if size == 0:
return tensor_var
shapes = [tensor_var.shape[x] for x in xrange(tensor_var.ndim)]
new_shape = [size + shapes[0]] + shapes[1:]
empty = tensor.AllocEmpty(tensor_var.dtype)(*new_shape)
return tensor.set_subtensor(empty[:shapes[0]], tensor_var)
def local_gpua_gemmbatch(node, context_name):
a, b = node.inputs
c = tensor.AllocEmpty(a.dtype)(a.shape[0], a.shape[1], b.shape[2])
return gpugemmbatch_no_inplace(c, 1.0, a, b, 0.0)
