Change mx.test_utils.list_gpus to mx.context.num_gpus where possible

docs/tutorials/gluon/datasets.md

@@ -157,7 +157,7 @@ def construct_net():
     return net
 
 # construct and initialize network.
-ctx =  mx.gpu() if mx.test_utils.list_gpus() else mx.cpu()
+ctx =  mx.gpu() if mx.context.num_gpus() else mx.cpu()
 
 net = construct_net()
 net.hybridize()

docs/tutorials/gluon/info_gan.md

@@ -51,7 +51,7 @@ batch_size   = 64
 z_dim        = 100
 n_continuous = 2
 n_categories = 10
-ctx = mx.gpu() if mx.test_utils.list_gpus() else mx.cpu()
+ctx = mx.gpu() if mx.context.num_gpus() else mx.cpu()
 ```
 
 Some functions to load and normalize images.

docs/tutorials/gluon/learning_rate_finder.md

@@ -231,7 +231,7 @@ Using a Pre-activation ResNet-18 from the Gluon model zoo, we instantiate our Le
 
 
 ```python
-ctx = mx.gpu() if mx.test_utils.list_gpus() else mx.cpu()
+ctx = mx.gpu() if mx.context.num_gpus() else mx.cpu()
 net = mx.gluon.model_zoo.vision.resnet18_v2(classes=10)
 learner = Learner(net=net, data_loader=data_loader, ctx=ctx)
 lr_finder = LRFinder(learner)

docs/tutorials/gluon/learning_rate_schedules.md

@@ -140,7 +140,7 @@ As discussed above, the schedule should return a learning rate given an (1-based
 
 ```python
 # Use GPU if one exists, else use CPU
-ctx = mx.gpu() if mx.test_utils.list_gpus() else mx.cpu()
+ctx = mx.gpu() if mx.context.num_gpus() else mx.cpu()
 
 # MNIST images are 28x28. Total pixels in input layer is 28x28 = 784
 num_inputs = 784

docs/tutorials/gluon/save_load_params.md

@@ -50,7 +50,7 @@ Let's define a helper function to build a LeNet model and another helper to trai
 
 ```python
 # Use GPU if one exists, else use CPU
-ctx = mx.gpu() if mx.test_utils.list_gpus() else mx.cpu()
+ctx = mx.gpu() if mx.context.num_gpus() else mx.cpu()
 
 # MNIST images are 28x28. Total pixels in input layer is 28x28 = 784
 num_inputs = 784

docs/tutorials/nlp/cnn.md

@@ -300,7 +300,7 @@ import time
 CNNModel = namedtuple("CNNModel", ['cnn_exec', 'symbol', 'data', 'label', 'param_blocks'])
 
 # Define what device to train/test on, use GPU if available
-ctx = mx.gpu() if mx.test_utils.list_gpus() else mx.cpu()
+ctx = mx.gpu() if mx.context.num_gpus() else mx.cpu()
 
 arg_names = cnn.list_arguments()
 

docs/tutorials/python/kvstore.md

@@ -57,9 +57,9 @@ values and then push the aggregated value:
 
 ```python
 # The numbers used below assume 4 GPUs
-gpus = mx.test_utils.list_gpus()
-if len(gpus) > 1:
-    contexts = [mx.gpu(i) for i in gpus]
+gpus = mx.context.num_gpus()
+if gpus > 0:
+    contexts = [mx.gpu(i) for i in range(gpus)]
 else:
     contexts = [mx.cpu(i) for i in range(4)]
 b = [mx.nd.ones(shape, ctx) for ctx in contexts]
@@ -173,4 +173,4 @@ When the distributed version is ready, we will update this section.
 ## Next Steps
 * [MXNet tutorials index](http://mxnet.io/tutorials/index.html)
 
-<!-- INSERT SOURCE DOWNLOAD BUTTONS -->
+<!-- INSERT SOURCE DOWNLOAD BUTTONS -->

docs/tutorials/python/mnist.md

@@ -50,7 +50,7 @@ mnist = mx.test_utils.get_mnist()
 mx.random.seed(42)
 
 # Set the compute context, GPU is available otherwise CPU
-ctx = mx.gpu() if mx.test_utils.list_gpus() else mx.cpu()
+ctx = mx.gpu() if mx.context.num_gpus() else mx.cpu()
 ```
 
 After running the above source code, the entire MNIST dataset should be fully loaded into memory. Note that for large datasets it is not feasible to pre-load the entire dataset first like we did here. What is needed is a mechanism by which we can quickly and efficiently stream data directly from the source. MXNet Data iterators come to the rescue here by providing exactly that. Data iterator is the mechanism by which we feed input data into an MXNet training algorithm and they are very simple to initialize and use and are optimized for speed. During training, we typically process training samples in small batches and over the entire training lifetime will end up processing each training example multiple times. In this tutorial, we'll configure the data iterator to feed examples in batches of 100. Keep in mind that each example is a 28x28 grayscale image and the corresponding label.

docs/tutorials/python/profiler.md

@@ -111,7 +111,7 @@ Let's define a method that will run one training iteration given data and label.
 
 ```python
 # Use GPU if available
-if len(mx.test_utils.list_gpus())!=0:
+if mx.context.num_gpus():
     ctx=mx.gpu()
 else:
     ctx=mx.cpu()

docs/tutorials/unsupervised_learning/gan.md

@@ -240,7 +240,7 @@ sigma = 0.02
 lr = 0.0002
 beta1 = 0.5
 # Define the compute context, use GPU if available
-ctx = mx.gpu() if mx.test_utils.list_gpus() else mx.cpu()
+ctx = mx.gpu() if mx.context.num_gpus() else mx.cpu()
 
 #=============Generator Module=============
 generator = mx.mod.Module(symbol=generatorSymbol, data_names=('rand',), label_names=None, context=ctx)

example/adversary/adversary_generation.ipynb

@@ -45,7 +45,7 @@
    },
    "outputs": [],
    "source": [
-    "ctx = mx.gpu() if len(mx.test_utils.list_gpus()) else mx.cpu()\n",
+    "ctx = mx.gpu() if mx.context.num_gpus() else mx.cpu()\n",
     "batch_size = 128"
    ]
   },

example/autoencoder/convolutional_autoencoder.ipynb

@@ -50,7 +50,7 @@
    "outputs": [],
    "source": [
     "batch_size = 512\n",
-    "ctx = mx.gpu() if len(mx.test_utils.list_gpus()) > 0 else mx.cpu()"
+    "ctx = mx.gpu() if mx.context.num_gpus() else mx.cpu()"
    ]
   },
   {

example/bi-lstm-sort/bi-lstm-sort.ipynb

@@ -39,7 +39,7 @@
     "seq_len = 5\n",
     "split = 0.8\n",
     "batch_size = 512\n",
-    "ctx = mx.gpu() if len(mx.test_utils.list_gpus()) > 0 else mx.cpu()"
+    "ctx = mx.gpu() if mx.context.num_gpus() > 0 else mx.cpu()"
    ]
   },
   {

example/distributed_training-horovod/gluon_mnist.py

@@ -45,7 +45,7 @@
 
 if not args.no_cuda:
     # Disable CUDA if there are no GPUs.
-    if not mx.test_utils.list_gpus():
+    if mx.context.num_gpus() == 0:
         args.no_cuda = True
 
 logging.basicConfig(level=logging.INFO)

example/distributed_training-horovod/module_mnist.py

@@ -42,7 +42,7 @@
 
 if not args.no_cuda:
     # Disable CUDA if there are no GPUs.
-    if not mx.test_utils.list_gpus():
+    if mx.context.num_gpus() == 0:
         args.no_cuda = True
 
 logging.basicConfig(level=logging.INFO)

example/image-classification/test_score.py

@@ -51,11 +51,10 @@ def test_imagenet1k_inception_bn(**kwargs):
     assert r > g and r < g + .1
 
 if __name__ == '__main__':
-    gpus = mx.test_utils.list_gpus()
-    assert len(gpus) > 0
-    batch_size = 16 * len(gpus)
-    gpus = ','.join([str(i) for i in gpus])
-
+    num_gpus = mx.context.num_gpus()
+    assert num_gpus > 0
+    batch_size = 16 * num_gpus
+    gpus = ','.join(map(str, range(num_gpus)))
     kwargs = {'gpus':gpus, 'batch_size':batch_size, 'max_num_examples':500}
     download_data()
     test_imagenet1k_resnet(**kwargs)

example/multi-task/multi-task-learning.ipynb

@@ -58,7 +58,7 @@
    "source": [
     "batch_size = 128\n",
     "epochs = 5\n",
-    "ctx = mx.gpu() if len(mx.test_utils.list_gpus()) > 0 else mx.cpu()\n",
+    "ctx = mx.gpu() if mx.context.num_gpus() > 0 else mx.cpu()\n",
     "lr = 0.01"
    ]
   },

example/recommenders/demo2-dssm.ipynb

@@ -41,7 +41,7 @@
     "hidden_units = 128\n",
     "epsilon_proj = 0.25\n",
     "\n",
-    "ctx = mx.gpu() if len(mx.test_utils.list_gpus()) > 0 else mx.cpu()"
+    "ctx = mx.gpu() if mx.context.num_gpus() > 0 else mx.cpu()"
    ]
   },
   {

example/svm_mnist/svm_mnist.py

@@ -82,7 +82,7 @@
 # Article's suggestion on batch size
 batch_size = 200
 
-ctx = mx.gpu() if len(mx.test_utils.list_gpus()) > 0 else mx.cpu()
+ctx = mx.gpu() if mx.context.num_gpus() > 0 else mx.cpu()
 
 results = {}
 for output in [mlp_svm_l2, mlp_svm_l1, mlp_softmax]:
@@ -121,4 +121,4 @@
 
 #svm_l2 97.85 %s
 #svm_l1 98.15 %s
-#softmax 97.69 %s
+#softmax 97.69 %s

python/mxnet/gluon/contrib/nn/basic_layers.py

@@ -24,7 +24,7 @@
            'PixelShuffle3D']
 
 import warnings
-from .... import nd, test_utils
+from .... import nd, context
 from ...block import HybridBlock, Block
 from ...nn import Sequential, HybridSequential, BatchNorm
 
@@ -233,7 +233,7 @@ def _get_num_devices(self):
         warnings.warn("Caution using SyncBatchNorm: "
                       "if not using all the GPUs, please mannually set num_devices",
                       UserWarning)
-        num_devices = len(test_utils.list_gpus())
+        num_devices = context.num_gpus()
         num_devices = num_devices if num_devices > 0 else 1
         return num_devices
 

tests/python/gpu/test_nccl.py

@@ -22,7 +22,7 @@
 
 shapes = [(10), (100), (1000), (10000), (100000), (2,2), (2,3,4,5,6,7,8)]
 keys = [1,2,3,4,5,6,7]
-num_gpus = len(mx.test_utils.list_gpus())
+num_gpus = mx.context.num_gpus()
 
 
 if num_gpus > 8 :

tests/python/profiling/test_nvtx.py

@@ -25,7 +25,7 @@
 
 def test_nvtx_ranges_present_in_profile():
 
-    if not mx.test_utils.list_gpus():
+    if not mx.context.num_gpus():
         unittest.skip('Test only applicable to machines with GPUs')
 
     # Build a system independent wrapper to execute simple_forward with nvprof

tools/caffe_converter/test_converter.py

@@ -90,9 +90,9 @@ def main():
         gpus = [-1]
         default_batch_size = 32
     else:
-        gpus = mx.test_utils.list_gpus()
-        assert gpus, 'At least one GPU is needed to run test_converter in GPU mode'
-        default_batch_size = 32 * len(gpus)
+        num_gpus = mx.context.num_gpus()
+        assert num_gpus, 'At least one GPU is needed to run test_converter in GPU mode'
+        default_batch_size = 32 * num_gpus
 
     models = ['bvlc_googlenet', 'vgg-16', 'resnet-50']