Python tensorflow.python.platform.test.is_gpu_available() Examples

def testOutputSizeWithStrideOneValidPaddingNCHW(self):
    if test.is_gpu_available(cuda_only=True):
      with self.session(use_gpu=True) as sess:
        num_filters = 32
        input_size = [5, 3, 10, 12]
        expected_size = [5, num_filters, 12, 14]

        images = random_ops.random_uniform(input_size, seed=1)
        output = layers_lib.conv2d_transpose(
            images,
            num_filters, [3, 3],
            stride=1,
            padding='VALID',
            data_format='NCHW')
        self.assertEqual(output.op.name, 'Conv2d_transpose/Relu')

        sess.run(variables_lib.global_variables_initializer())
        self.assertListEqual(list(output.eval().shape), expected_size) 

def testCompareBilinear(self):
    if test.is_gpu_available():
      input_shape = [1, 5, 6, 3]
      target_height = 8
      target_width = 12
      for nptype in [np.float32, np.float64]:
        for align_corners in [True, False]:
          img_np = np.arange(
              0, np.prod(input_shape), dtype=nptype).reshape(input_shape)
          value = {}
          for use_gpu in [True, False]:
            with self.test_session(use_gpu=use_gpu):
              image = constant_op.constant(img_np, shape=input_shape)
              new_size = constant_op.constant([target_height, target_width])
              out_op = image_ops.resize_images(
                  image, new_size,
                  image_ops.ResizeMethod.BILINEAR,
                  align_corners=align_corners)
              value[use_gpu] = out_op.eval()
          self.assertAllClose(value[True], value[False], rtol=1e-5, atol=1e-5) 

def testOutputSizeWithStride2x5NCHW(self):
    if test.is_gpu_available(cuda_only=True):
      with self.session(use_gpu=True) as sess:
        num_filters = 1
        input_size = [1, 1, 3, 2]
        expected_size = [1, num_filters, 6, 10]

        images = random_ops.random_uniform(input_size, seed=1)
        output = layers_lib.conv2d_transpose(
            images,
            num_filters, [2, 4],
            stride=[2, 5],
            padding='VALID',
            data_format='NCHW')
        sess.run(variables_lib.global_variables_initializer())
        self.assertEqual(output.op.name, 'Conv2d_transpose/Relu')
        self.assertListEqual(list(output.eval().shape), expected_size) 

def testOutputSizeWithStride2x4NCHW(self):
    if test.is_gpu_available(cuda_only=True):
      with self.session(use_gpu=True) as sess:
        num_filters = 1
        input_size = [1, 1, 3, 2]
        expected_size = [1, num_filters, 6, 8]

        images = random_ops.random_uniform(input_size, seed=1)
        output = layers_lib.conv2d_transpose(
            images,
            num_filters, [2, 4],
            stride=[2, 4],
            padding='VALID',
            data_format='NCHW')
        sess.run(variables_lib.global_variables_initializer())
        self.assertEqual(output.op.name, 'Conv2d_transpose/Relu')
        self.assertListEqual(list(output.eval().shape), expected_size) 

def testOutputSizeWithStride2x1NCHW(self):
    if test.is_gpu_available(cuda_only=True):
      with self.session(use_gpu=True) as sess:
        num_filters = 1
        input_size = [1, 1, 3, 2]
        expected_size = [1, num_filters, 6, 5]

        images = random_ops.random_uniform(input_size, seed=1)
        output = layers_lib.conv2d_transpose(
            images,
            num_filters, [2, 4],
            stride=[2, 1],
            padding='VALID',
            data_format='NCHW')
        sess.run(variables_lib.global_variables_initializer())
        self.assertEqual(output.op.name, 'Conv2d_transpose/Relu')
        self.assertListEqual(list(output.eval().shape), expected_size) 

def testOutputSizeWith2x2StrideTwoValidPaddingNCHW(self):
    if test.is_gpu_available(cuda_only=True):
      with self.session(use_gpu=True) as sess:
        num_filters = 1
        input_size = [1, 1, 2, 2]
        expected_size = [1, num_filters, 4, 4]

        images = random_ops.random_uniform(input_size, seed=1)
        output = layers_lib.conv2d_transpose(
            images,
            num_filters, [2, 2],
            stride=[2, 2],
            padding='VALID',
            data_format='NCHW')
        sess.run(variables_lib.global_variables_initializer())
        self.assertEqual(output.op.name, 'Conv2d_transpose/Relu')
        self.assertListEqual(list(output.eval().shape), expected_size) 

def testOutputSizeWith2x2StrideTwoSamePaddingNCHW(self):
    if test.is_gpu_available(cuda_only=True):
      with self.session(use_gpu=True) as sess:
        num_filters = 1
        input_size = [1, 1, 2, 2]
        expected_size = [1, num_filters, 4, 4]

        images = random_ops.random_uniform(input_size, seed=1)
        output = layers_lib.conv2d_transpose(
            images,
            num_filters, [2, 2],
            stride=[2, 2],
            padding='SAME',
            data_format='NCHW')
        sess.run(variables_lib.global_variables_initializer())
        self.assertEqual(output.op.name, 'Conv2d_transpose/Relu')
        self.assertListEqual(list(output.eval().shape), expected_size) 

def testOutputSizeWith1x1StrideTwoSamePaddingNCHW(self):
    if test.is_gpu_available(cuda_only=True):
      with self.session(use_gpu=True) as sess:
        num_filters = 1
        input_size = [1, 1, 1, 1]
        expected_size = [1, num_filters, 2, 2]

        images = random_ops.random_uniform(input_size, seed=1)
        output = layers_lib.conv2d_transpose(
            images,
            num_filters, [2, 2],
            stride=[2, 2],
            padding='SAME',
            data_format='NCHW')
        self.assertListEqual(list(output.get_shape().as_list()), expected_size)

        sess.run(variables_lib.global_variables_initializer())
        self.assertEqual(output.op.name, 'Conv2d_transpose/Relu')
        self.assertListEqual(list(output.eval().shape), expected_size) 

def testOutputSizeWithStrideTwoValidPaddingNCHW(self):
    if test.is_gpu_available(cuda_only=True):
      with self.session(use_gpu=True) as sess:
        num_filters = 32
        input_size = [5, 3, 9, 11]
        expected_size = [5, num_filters, 19, 23]

        images = random_ops.random_uniform(input_size, seed=1)
        output = layers_lib.conv2d_transpose(
            images,
            num_filters, [3, 3],
            stride=[2, 2],
            padding='VALID',
            data_format='NCHW')
        self.assertEqual(output.op.name, 'Conv2d_transpose/Relu')
        self.assertListEqual(list(output.get_shape().as_list()), expected_size)

        sess.run(variables_lib.global_variables_initializer())
        self.assertListEqual(list(output.eval().shape), expected_size) 

def testOutputSizeWithStrideOneSamePaddingNCHW(self):
    # `NCHW` data format is only supported for `GPU` device.
    if test.is_gpu_available(cuda_only=True):
      with self.session(use_gpu=True) as sess:
        num_filters = 32
        input_size = [5, 3, 10, 12]
        expected_size = [5, num_filters, 10, 12]

        images = random_ops.random_uniform(input_size, seed=1)
        output = layers_lib.conv2d_transpose(
            images,
            num_filters, [3, 3],
            stride=1,
            padding='SAME',
            data_format='NCHW')
        self.assertEqual(output.op.name, 'Conv2d_transpose/Relu')

        sess.run(variables_lib.global_variables_initializer())
        self.assertListEqual(list(output.eval().shape), expected_size) 

def testDynamicOutputSizeWithRateOneValidPaddingNCHW(self):
    if test.is_gpu_available(cuda_only=True):
      num_filters = 32
      input_size = [5, 3, 9, 11]
      expected_size = [None, num_filters, None, None]
      expected_size_dynamic = [5, num_filters, 7, 9]

      with self.session(use_gpu=True):
        images = array_ops.placeholder(np.float32,
                                       [None, input_size[1], None, None])
        output = layers_lib.convolution2d(
            images,
            num_filters, [3, 3],
            rate=1,
            padding='VALID',
            data_format='NCHW')
        variables_lib.global_variables_initializer().run()
        self.assertEqual(output.op.name, 'Conv/Relu')
        self.assertListEqual(output.get_shape().as_list(), expected_size)
        eval_output = output.eval({images: np.zeros(input_size, np.float32)})
        self.assertListEqual(list(eval_output.shape), expected_size_dynamic) 

def testFusedBatchNormFloat16MatchesFloat32(self):
    if test.is_gpu_available(cuda_only=True):
      shape = [5, 4, 2, 3]
      res_32 = self._runFusedBatchNorm(shape, np.float32)
      res_16 = self._runFusedBatchNorm(shape, np.float16)
      self.assertAllClose(res_32, res_16, rtol=1e-3) 

def testResizeDown(self):
    # This test is also conducted with int8, so 127 is the maximum
    # value that can be used.
    data = [127, 127, 64, 64,
            127, 127, 64, 64,
            64, 64, 127, 127,
            64, 64, 127, 127,
            50, 50, 100, 100,
            50, 50, 100, 100]
    expected_data = [127, 64,
                     64, 127,
                     50, 100]
    target_height = 3
    target_width = 2

    # Test out 3-D and 4-D image shapes.
    img_shapes = [[1, 6, 4, 1], [6, 4, 1]]
    target_shapes = [[1, target_height, target_width, 1],
                     [target_height, target_width, 1]]

    for target_shape, img_shape in zip(target_shapes, img_shapes):

      for nptype in self.TYPES:
        img_np = np.array(data, dtype=nptype).reshape(img_shape)

        for opt in self.OPTIONS:
          if test.is_gpu_available() and self.shouldRunOnGPU(opt, nptype):
            with self.test_session(use_gpu=True):
              image = constant_op.constant(img_np, shape=img_shape)
              y = image_ops.resize_images(
                  image, [target_height, target_width], opt)
              expected = np.array(expected_data).reshape(target_shape)
              resized = y.eval()
              self.assertAllClose(resized, expected, atol=1e-5) 

def testNoOp(self):
    img_shape = [1, 6, 4, 1]
    single_shape = [6, 4, 1]
    # This test is also conducted with int8, so 127 is the maximum
    # value that can be used.
    data = [127, 127, 64, 64,
            127, 127, 64, 64,
            64, 64, 127, 127,
            64, 64, 127, 127,
            50, 50, 100, 100,
            50, 50, 100, 100]
    target_height = 6
    target_width = 4

    for nptype in self.TYPES:
      img_np = np.array(data, dtype=nptype).reshape(img_shape)

      for opt in self.OPTIONS:
        if test.is_gpu_available() and self.shouldRunOnGPU(opt, nptype):
          with self.test_session(use_gpu=True) as sess:
            image = constant_op.constant(img_np, shape=img_shape)
            y = image_ops.resize_images(
                image, [target_height, target_width], opt)
            yshape = array_ops.shape(y)
            resized, newshape = sess.run([y, yshape])
            self.assertAllEqual(img_shape, newshape)
            self.assertAllClose(resized, img_np, atol=1e-5)

      # Resizing with a single image must leave the shape unchanged also.
      with self.test_session(use_gpu=True):
        img_single = img_np.reshape(single_shape)
        image = constant_op.constant(img_single, shape=single_shape)
        y = image_ops.resize_images(image, [target_height, target_width],
                                    self.OPTIONS[0])
        yshape = array_ops.shape(y)
        newshape = yshape.eval()
        self.assertAllEqual(single_shape, newshape) 

def setUp(self):
    self._dump_root = tempfile.mkdtemp()

    if test.is_gpu_available():
      self._expected_partition_graph_count = 2
      self._expected_num_devices = 2
      self._main_device = "/job:localhost/replica:0/task:0/gpu:0"
    else:
      self._expected_partition_graph_count = 1
      self._expected_num_devices = 1
      self._main_device = "/job:localhost/replica:0/task:0/cpu:0" 

def testChannelsFirst(self):
    # `bias_add` doesn't support NCHW on CPU.
    if test.is_gpu_available(cuda_only=True):
      for ndim in [3, 4, 5]:
        x = np.random.uniform(size=(4, 3, 2, 1)[:ndim])
        y = self._runGDN(x, x.shape, False, 'channels_first')
        self.assertEqual(x.shape, y.shape)
        self.assertAllClose(y, x / np.sqrt(1 + .1 * (x**2)), rtol=0, atol=1e-6) 

def gpu_availability():
    """
    Detect gpu on user system

    :return: Whether at least a CUDA compatible GPU is detected and usable
    :rtype: bool
    :History: 2018-Apr-25 - Written - Henry Leung (University of Toronto)
    """
    # assume if using tensorflow-gpu, then Nvidia GPU is available
    if is_built_with_cuda():
        return is_gpu_available()
    else:
        return is_built_with_cuda() 

def data_formats(self):
    # On CPU, many ops don't support the channels first data format. Hence, if
    # no GPU is available, we skip these tests.
    if test.is_gpu_available(cuda_only=True):
      return ("channels_first", "channels_last")
    else:
      return ("channels_last",) 

def _check_has_gpu():
  if not test.is_gpu_available(cuda_only=True):
    raise ValueError(
        """You have asked to run part or all of this on GPU, but it appears
        that no GPU is available. If your machine has GPUs it is possible you
        do not have a version of TensorFlow with GPU support. To build with GPU
        support, add --config=cuda to the build flags.\n """) 

def setUpClass(cls):
    if test.is_gpu_available():
      cls._expected_partition_graph_count = 2
      cls._expected_num_devices = 2
      gpu_name = test_util.gpu_device_name()
      cls._main_device = "/job:localhost/replica:0/task:0" + gpu_name
    else:
      cls._expected_partition_graph_count = 1
      cls._expected_num_devices = 1
      cls._main_device = "/job:localhost/replica:0/task:0/device:CPU:0" 

def _use_eigen_kernels():
  use_eigen_kernels = False  # Eigen kernels are default
  if test.is_gpu_available(cuda_only=True):
    use_eigen_kernels = False
  return use_eigen_kernels 

def setUpClass(cls):
    if test.is_gpu_available():
      cls._expected_partition_graph_count = 2
      cls._expected_num_devices = 2
      cls._main_device = "/job:localhost/replica:0/task:0/gpu:0"
    else:
      cls._expected_partition_graph_count = 1
      cls._expected_num_devices = 1
      cls._main_device = "/job:localhost/replica:0/task:0/cpu:0" 

def testNHWCAndNCHWTrainingProduceSameOutput(self):
    if test.is_gpu_available(cuda_only=True):
      for shape in [[7, 3, 5], [5, 2, 3, 4], [11, 3, 2, 4, 5]]:
        nhwc = self._runBatchNormalizationWithFormat(
            data_format='NHWC', shape=shape, is_training=True)
        nchw = self._runBatchNormalizationWithFormat(
            data_format='NCHW', shape=shape, is_training=True)
        self.assertAllClose(nhwc, nchw, atol=1e-4, rtol=1e-4) 

def testNHWCAndNCHWInferenceProduceSameOutput(self):
    if test.is_gpu_available(cuda_only=True):
      for shape in [[7, 3, 5], [5, 2, 3, 4], [11, 3, 2, 4, 5]]:
        nhwc = self._runBatchNormalizationWithFormat(
            data_format='NHWC', shape=shape, is_training=False)
        nchw = self._runBatchNormalizationWithFormat(
            data_format='NCHW', shape=shape, is_training=False)
        self.assertAllClose(nhwc, nchw, atol=1e-4, rtol=1e-4) 

def testNoneUpdatesCollectionIsTrainingVariableFusedNCHW(self):
    if test.is_gpu_available(cuda_only=True):
      with tf.Graph().as_default():
        self._testNoneUpdatesCollectionIsTrainingVariable(
            True, data_format='NCHW') 

def testIsTrainingVariableFusedNCHWZeroDebias(self):
    if test.is_gpu_available(cuda_only=True):
      self._testIsTrainingVariable(
          True, data_format='NCHW', zero_debias_moving_mean=True) 

def testIsTrainingVariableFusedNCHW(self):
    if test.is_gpu_available(cuda_only=True):
      self._testIsTrainingVariable(True, data_format='NCHW') 

def testDelayedUpdateMovingVarsFusedNCHW(self):
    if test.is_gpu_available(cuda_only=True):
      self._testDelayedUpdateMovingVars(True, data_format='NCHW') 

def testNoneUpdatesCollectionsFusedNCHWZeroDebias(self):
    if test.is_gpu_available(cuda_only=True):
      self._testNoneUpdatesCollections(
          True, data_format='NCHW', zero_debias_moving_mean=True) 

def testMaxPoolingWithArgmax(self):
        # MaxPoolWithArgMax is implemented only on CUDA.
        if not test.is_gpu_available(cuda_only=True):
            return
        '''[[[[  1.   2.]
              [  3.   4.]
              [  5.   6.]]
             [[  7.   8.]
              [  9.  10.]
              [ 11.  12.]]
             [[ 13.  14.]
              [ 15.  16.]
              [ 17.  18.]]]]'''
        tensor_input = [1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0,
                        10.0, 11.0, 12.0, 13.0, 14.0, 15.0, 16.0, 17.0, 18.0]
        with self.test_session(use_gpu=True) as sess:
            t = constant_op.constant(tensor_input, shape=[1, 3, 3, 2])
            out_op, argmax_op = segnet_vgg.max_pool_with_argmax(t)
            out, argmax = sess.run([out_op, argmax_op])
            self.assertShapeEqual(out, out_op)
            self.assertShapeEqual(argmax, argmax_op)
            '''[[[9, 10]
                 [11, 12]]
                [[15, 16]
                 [17, 18]]]'''
            self.assertAllClose(out.ravel(), [9., 10., 11., 12., 15., 16., 17., 18.])
            self.assertAllEqual(argmax.ravel(), [8, 9, 10, 11, 14, 15, 16, 17])