def filter(self, data, strict=False, allow_downcast=None):
        if strict:
            if not isinstance(data, gpuarray.GpuArray):
                raise TypeError("%s expected a GpuArray object." % self,
                                data, type(data))
            if self.typecode != data.typecode:
                raise TypeError("%s expected typecode %d (dtype %s), "
                                "got %d (dtype %s)." %
                                (self, self.typecode, self.dtype,
                                 data.typecode, str(data.dtype)))
            # fallthrough to ndim check
        elif allow_downcast:
            data = gpuarray.array(data, dtype=self.typecode, copy=False,
                                  ndmin=len(self.broadcastable))
        else:
            up_dtype = scalar.upcast(self.dtype, data.dtype)
            if up_dtype == self.dtype:
                data = gpuarray.array(data, dtype=self.dtype, copy=False)
            else:
                raise TypeError("%s cannot store a value of dtype %s "
                                "without risking loss of precision." %
                                (self, data.dtype))

        if self.ndim != data.ndim:
            raise TypeError("Wrong number of dimensions: expected %s, "
                            "got %s with shape %s." % (self.ndim, data.ndim,
                                                       data.shape), data)
        shp = data.shape
        for i, b in enumerate(self.broadcastable):
            if b and shp[i] != 1:
                raise TypeError("Non-unit value on shape on a broadcastable"
                                " dimension.", shp, self.broadcastable)
        return data

    def filter(self, data, strict=False, allow_downcast=None):
        if (isinstance(data, gpuarray.GpuArray) and
                data.typecode == self.typecode):
            # This is just to make this condition not enter the
            # following branches
            pass
        elif strict:
            if not isinstance(data, gpuarray.GpuArray):
                raise TypeError("%s expected a GpuArray object." % self,
                                data, type(data))
            if self.typecode != data.typecode:
                raise TypeError("%s expected typecode %d (dtype %s), "
                                "got %d (dtype %s)." %
                                (self, self.typecode, self.dtype,
                                 data.typecode, str(data.dtype)))
            if self.context != data.context:
                raise TypeError("data context does not match type context")
            # fallthrough to ndim check
        elif (allow_downcast or
              (allow_downcast is None and
               type(data) == float and
               self.dtype == config.floatX)):
            data = gpuarray.array(data, dtype=self.typecode, copy=False,
                                  ndmin=len(self.broadcastable),
                                  context=self.context)
        else:
            if not hasattr(data, 'dtype'):
                converted_data = theano._asarray(data, self.dtype)
                # We use the `values_eq` static function from TensorType
                # to handle NaN values.
                if TensorType.values_eq(numpy.asarray(data),
                                        converted_data,
                                        force_same_dtype=False):
                    data = converted_data
                    data = gpuarray.array(data, context=self.context)

            up_dtype = scalar.upcast(self.dtype, data.dtype)
            if up_dtype == self.dtype:
                data = gpuarray.array(data, dtype=self.dtype, copy=False,
                                      context=self.context)
            else:
                raise TypeError("%s cannot store a value of dtype %s "
                                "without risking loss of precision." %
                                (self, data.dtype))

        if self.ndim != data.ndim:
            raise TypeError("Wrong number of dimensions: expected %s, "
                            "got %s with shape %s." % (self.ndim, data.ndim,
                                                       data.shape), data)
        shp = data.shape
        for i, b in enumerate(self.broadcastable):
            if b and shp[i] != 1:
                raise TypeError("Non-unit value on shape on a broadcastable"
                                " dimension.", shp, self.broadcastable)
        return data

    def filter(self, data, strict=False, allow_downcast=None):
        if (isinstance(data, gpuarray.GpuArray) and
                data.typecode == self.typecode):
            # This is just to make this condition not enter the
            # following branches
            pass
        elif strict:
            if not isinstance(data, gpuarray.GpuArray):
                raise TypeError("%s expected a GpuArray object." % self,
                                data, type(data))
            if self.typecode != data.typecode:
                raise TypeError("%s expected typecode %d (dtype %s), "
                                "got %d (dtype %s)." %
                                (self, self.typecode, self.dtype,
                                 data.typecode, str(data.dtype)))
            if self.context != data.context:
                raise TypeError("data context does not match type context")
            # fallthrough to ndim check
            elif (allow_downcast or
                  (allow_downcast is None and
                   type(data) == float and
                   self.dtype == config.floatX)):
                data = gpuarray.array(data, dtype=self.typecode, copy=False,
                                      ndmin=len(self.broadcastable),
                                      context=self.context)
        else:
            if not hasattr(data, 'dtype'):
                # This is to convert objects that don't have a dtype
                # (like lists).  We anticipate that the type below
                # will match and we pass copy=False so it won't make a
                # second object on the GPU.
                data = gpuarray.array(data, copy=False, context=self.context)

            up_dtype = scalar.upcast(self.dtype, data.dtype)
            if up_dtype == self.dtype:
                data = gpuarray.array(data, dtype=self.dtype, copy=False,
                                      context=self.context)
            else:
                raise TypeError("%s cannot store a value of dtype %s "
                                "without risking loss of precision." %
                                (self, data.dtype))

        if self.ndim != data.ndim:
            raise TypeError("Wrong number of dimensions: expected %s, "
                            "got %s with shape %s." % (self.ndim, data.ndim,
                                                       data.shape), data)
        shp = data.shape
        for i, b in enumerate(self.broadcastable):
            if b and shp[i] != 1:
                raise TypeError("Non-unit value on shape on a broadcastable"
                                " dimension.", shp, self.broadcastable)
        return data

 def setUp(self):
     self.input = gpu_ftensor4()
     self.filters = gpu_ftensor4()
     self.topgrad = gpu_ftensor4()
     self.constant_tensor = gpuarray.array(
         numpy.zeros((3, 5, 7, 11), dtype='float32'),
         context=get_context(test_ctx_name))

def rand_gpuarray(*shape, **kwargs):
    r = rng.rand(*shape) * 2 - 1
    dtype = kwargs.pop("dtype", theano.config.floatX)
    cls = kwargs.pop("cls", None)
    if len(kwargs) != 0:
        raise TypeError("Unexpected argument %s", list(kwargs.keys())[0])
    return gpuarray.array(r, dtype=dtype, cls=cls, context=get_context(test_ctx_name))

def gen_gpuarray(
    shape_orig,
    dtype="float32",
    offseted_outer=False,
    offseted_inner=False,
    sliced=1,
    order="c",
    nozeros=False,
    incr=0,
    ctx=None,
    cls=None,
):
    if sliced is True:
        sliced = 2
    elif sliced is False:
        sliced = 1
    shape = numpy.asarray(shape_orig).copy()
    if sliced != 1 and len(shape) > 0:
        shape[0] *= numpy.absolute(sliced)
    if offseted_outer and len(shape) > 0:
        shape[0] += 1
    if offseted_inner and len(shape) > 0:
        shape[-1] += 1

    low = 0.0
    if nozeros:
        low = 1.0

    a = numpy.random.uniform(low, 10.0, shape)
    a += incr

    a = numpy.asarray(a, dtype=dtype)
    assert order in ["c", "f"]
    if order == "f" and len(shape) > 0:
        a = numpy.asfortranarray(a)
    b = gpuarray.array(a, context=ctx, cls=cls)
    if order == "f" and len(shape) > 0 and b.size > 1:
        assert b.flags["F_CONTIGUOUS"]

    if offseted_outer and len(shape) > 0:
        b = b[1:]
        a = a[1:]
    if offseted_inner and len(shape) > 0:
        # The b[..., 1:] act as the test for this subtensor case.
        b = b[..., 1:]
        a = a[..., 1:]
    if sliced != 1 and len(shape) > 0:
        a = a[::sliced]
        b = b[::sliced]

    if False and shape_orig == ():
        assert a.shape == (1,)
        assert b.shape == (1,)
    else:
        assert a.shape == shape_orig, (a.shape, shape_orig)
        assert b.shape == shape_orig, (b.shape, shape_orig)

    assert numpy.allclose(a, numpy.asarray(b)), (a, numpy.asarray(b))

    return a, b

def transfer_not_contiguous(shp, dtype):
    a = numpy.random.rand(*shp) * 10
    a = a[::-1]
    b = gpu_ndarray.array(a, context=ctx)
    c = numpy.asarray(b)

    assert numpy.allclose(c, a)
    assert a.shape == b.shape == c.shape
    # the result array (c) is C contiguous
    assert a.strides == b.strides == (-c.strides[0],) + c.strides[1:]
    assert a.dtype == b.dtype == c.dtype
    assert c.flags.c_contiguous

def test_transfer_gpu_gpu():
    g = GpuArrayType(dtype='float32', broadcastable=(False, False),
                     context_name=test_ctx_name)()

    av = np.asarray(rng.rand(5, 4), dtype='float32')
    gv = gpuarray.array(av, context=get_context(test_ctx_name))
    mode = mode_with_gpu.excluding('cut_gpua_host_transfers', 'local_cut_gpua_host_gpua')
    f = theano.function([g], GpuToGpu(test_ctx_name)(g), mode=mode)
    topo = f.maker.fgraph.toposort()
    assert len(topo) == 1
    assert isinstance(topo[0].op, GpuToGpu)
    fv = f(gv)
    assert GpuArrayType.values_eq(fv, gv)

def test_transfer_cpu_gpu():
    a = T.fmatrix('a')
    g = GpuArrayType(dtype='float32', broadcastable=(False, False))('g')

    av = np.asarray(rng.rand(5, 4), dtype='float32')
    gv = gpuarray.array(av, context=get_context(test_ctx_name))

    f = theano.function([a], GpuFromHost(test_ctx_name)(a))
    fv = f(av)
    assert GpuArrayType.values_eq(fv, gv)

    f = theano.function([g], host_from_gpu(g))
    fv = f(gv)
    assert np.all(fv == av)

def transfer_fortran(shp, dtype):
    a = numpy.random.rand(*shp) * 10
    a_ = numpy.asfortranarray(a)
    if len(shp) > 1:
        assert a_.strides != a.strides
    a = a_
    b = gpu_ndarray.array(a, context=ctx)
    c = numpy.asarray(b)

    assert a.shape == b.shape == c.shape
    assert a.dtype == b.dtype == c.dtype
    assert a.flags.f_contiguous
    assert c.flags.f_contiguous
    assert a.strides == b.strides == c.strides
    assert numpy.allclose(c, a)

def test_transfer_strided():
    # This is just to ensure that it works in theano
    # libgpuarray has a much more comprehensive suit of tests to
    # ensure correctness
    a = T.fmatrix('a')
    g = GpuArrayType(dtype='float32', broadcastable=(False, False))('g')

    av = np.asarray(rng.rand(5, 8), dtype='float32')
    gv = gpuarray.array(av, context=get_context(test_ctx_name))

    av = av[:, ::2]
    gv = gv[:, ::2]

    f = theano.function([a], GpuFromHost(test_ctx_name)(a))
    fv = f(av)
    assert GpuArrayType.values_eq(fv, gv)

    f = theano.function([g], host_from_gpu(g))
    fv = f(gv)
    assert np.all(fv == av)

def as_gpuarray(x):
    return gpuarray.array(x, copy=False)

 def perform(self, node, inp, out):
     x, = inp
     z, = out
     z[0] = gpuarray.array(numpy.asarray(x))

 def perform(self, node, inp, out, ctx):
     x, = inp
     z, = out
     z[0] = gpuarray.array(x, context=ctx)

        def thunk():
            context = inputs[0][0].context

            # Size of the matrices to invert.
            z = outputs[0]

            # Matrix.
            A = inputs[0][0]

            # Solution vectors.
            b = inputs[1][0]

            assert(len(A.shape) == 2)
            assert(len(b.shape) == 2)

            if self.trans in ['T', 'C']:
                trans = 1
                l, n = A.shape
                k, m = b.shape
            elif self.trans == 'N':
                trans = 0
                n, l = A.shape
                k, m = b.shape
            else:
                raise ValueError('Invalid value for trans')
            if l != n:
                raise ValueError('A must be a square matrix')
            if n != k:
                raise ValueError('A and b must be aligned.')

            lda = max(1, n)
            ldb = max(1, k, m)

            # We copy A and b as cusolver operates inplace
            b = gpuarray.array(b, copy=True, order='F')
            if not self.inplace:
                A = gpuarray.array(A, copy=True)
            A_ptr = A.gpudata
            b_ptr = b.gpudata

            # cusolver expects a F ordered matrix, but A is not explicitly
            # converted between C and F order, instead we switch the
            # "transpose" flag.
            if A.flags['C_CONTIGUOUS']:
                trans = 1 - trans

            workspace_size = cusolver.cusolverDnSgetrf_bufferSize(
                cusolver_handle, n, n, A_ptr, lda)

            if (thunk.workspace is None or
                    thunk.workspace.size != workspace_size):
                thunk.workspace = gpuarray.zeros((workspace_size,),
                                                 dtype='float32',
                                                 context=context)

            if thunk.pivots is None or thunk.pivots.size != min(n, n):
                thunk.pivots = gpuarray.zeros((min(n, n),),
                                              dtype='float32',
                                              context=context)

            if thunk.dev_info is None:
                thunk.dev_info = gpuarray.zeros((1,),
                                                dtype='float32',
                                                context=context)

            workspace_ptr = thunk.workspace.gpudata
            pivots_ptr = thunk.pivots.gpudata
            dev_info_ptr = thunk.dev_info.gpudata

            cusolver.cusolverDnSgetrf(
                cusolver_handle, n, n, A_ptr, lda, workspace_ptr,
                pivots_ptr, dev_info_ptr)

            cusolver.cusolverDnSgetrs(
                cusolver_handle, trans, n, m, A_ptr, lda,
                pivots_ptr, b_ptr, ldb, dev_info_ptr)

            z[0] = b