Merge pull request #95 from sony/fix/20180914-misspelling

Fix misspell

CMakeLists.txt

@@ -95,7 +95,7 @@ if(BUILD_CPP_LIB)
 
   if (NOT CMAKE_BUILD_TYPE)
     set(CMAKE_BUILD_TYPE "Release" CACHE STRING
-      "Build type release is default on single-configuration build system like GnuMake."
+      "Build type release is default on single-configuration build system like GNU make."
       FORCE)
   endif()
 
@@ -192,7 +192,7 @@ if(BUILD_CPP_LIB)
       message(FATAL_ERROR
         "Python build_ext compiler inference is only supported on Win, Unix or Apple.")
     endif()
-    message("Python build_ext compiler is infered as '${NBLA_PYTHON_BUILD_EXT_COMPILER}'.")
+    message("Python build_ext compiler is inferred as '${NBLA_PYTHON_BUILD_EXT_COMPILER}'.")
     message("You can specify a compiler manually setting a variable"
       " NBLA_PYTHON_BUILD_EXT_COMPILER. You can see a list of supported"
       " compiler by `python setup.py build_ext --help-compiler`.")

CONTRIBUTING.md

@@ -5,4 +5,4 @@ CUDA extension follows [the instruction found in NNabla](https://github.com/sony
 ## Development Guide
 
 * [Adding a new function (layer implementation)](doc/contributing/add_function.md).
-* [Adding a new solver (gradient descent algorighm implemenation)](doc/contributing/add_solver.md).
+* [Adding a new solver (gradient descent algorithm implementation)](doc/contributing/add_solver.md).

build-tools/code_generator/generate.py

@@ -78,7 +78,7 @@ def generate():
         base, 'python/src/nnabla_ext/cudnn/_version.py.tmpl'),
         rootdir=base)
 
-    # Generate function skeltons
+    # Generate function skeletons
     func_src_template = join(
         base,
         'src/nbla/cuda/function/generic/function_impl.cu.tmpl')
@@ -91,16 +91,16 @@ def generate():
     func_header_template_cudnn = join(
         base,
         'include/nbla/cuda/cudnn/function/function_impl.hpp.tmpl')
-    utils.generate_skelton_function_impl(
+    utils.generate_skeleton_function_impl(
         function_info, function_types, ext_info={},
         template=func_src_template, output_format='%s.cu')
-    utils.generate_skelton_function_impl(
+    utils.generate_skeleton_function_impl(
         function_info, function_types, ext_info={},
         template=func_header_template, output_format='%s.hpp')
-    utils.generate_skelton_function_impl(
+    utils.generate_skeleton_function_impl(
         function_info, function_types_cudnn, ext_info={},
         template=func_src_template_cudnn, output_format='%s.cu')
-    utils.generate_skelton_function_impl(
+    utils.generate_skeleton_function_impl(
         function_info, function_types_cudnn, ext_info={},
         template=func_header_template_cudnn, output_format='%s.hpp')
 

build-tools/make/build.mk

@@ -40,7 +40,7 @@ nnabla-ext-cuda-clean-all:
 
 ########################################################################################################################
 # cpplib
-.PHNOY: nnabla-ext-cuda-cpplib
+.PHONY: nnabla-ext-cuda-cpplib
 nnabla-ext-cuda-cpplib:
 	mkdir -p $(BUILD_EXT_CUDA_DIRECTORY_CPPLIB)
 	cd $(BUILD_EXT_CUDA_DIRECTORY_CPPLIB) \
@@ -57,7 +57,7 @@ nnabla-ext-cuda-cpplib:
 		$(NNABLA_EXT_CUDA_DIRECTORY)
 	$(MAKE) -C $(BUILD_EXT_CUDA_DIRECTORY_CPPLIB) -j$(PARALLEL_BUILD_NUM)
 
-.PHNOY: nnabla-ext-cuda-cpplib-multi-gpu
+.PHONY: nnabla-ext-cuda-cpplib-multi-gpu
 nnabla-ext-cuda-cpplib-multi-gpu:
 	mkdir -p $(BUILD_EXT_CUDA_DIRECTORY_CPPLIB_MULTI_GPU)
 	cd $(BUILD_EXT_CUDA_DIRECTORY_CPPLIB_MULTI_GPU) \
@@ -77,15 +77,15 @@ nnabla-ext-cuda-cpplib-multi-gpu:
 
 ########################################################################################################################
 # wheel
-.PHNOY: nnabla-ext-cuda-wheel
+.PHONY: nnabla-ext-cuda-wheel
 nnabla-ext-cuda-wheel:
 	$(call with-virtualenv, \
 		$(NNABLA_EXT_CUDA_DIRECTORY), \
 		$(BUILD_EXT_CUDA_DIRECTORY_WHEEL)/env, \
 		-f build-tools/make/build.mk, \
 		nnabla-ext-cuda-wheel-local)
 
-.PHNOY: nnabla-ext-cuda-wheel-local
+.PHONY: nnabla-ext-cuda-wheel-local
 nnabla-ext-cuda-wheel-local: nnabla-install \
 		$(BUILD_DIRECTORY_CPPLIB)/lib/libnnabla.so \
 		$(BUILD_EXT_CUDA_DIRECTORY_CPPLIB)/lib/libnnabla_cuda.so
@@ -105,7 +105,7 @@ nnabla-ext-cuda-wheel-local: nnabla-install \
 		$(NNABLA_EXT_CUDA_DIRECTORY) \
 	&& $(MAKE) -C $(BUILD_EXT_CUDA_DIRECTORY_WHEEL) wheel
 
-.PHNOY: nnabla-ext-cuda-wheel-multi-gpu
+.PHONY: nnabla-ext-cuda-wheel-multi-gpu
 nnabla-ext-cuda-wheel-multi-gpu: \
 			nnabla-cpplib \
 			nnabla-wheel \
@@ -161,7 +161,7 @@ nnabla-ext-cuda-test-local: nnabla-install nnabla-ext-cuda-install
 	&& PYTHONPATH=$(NNABLA_EXT_CUDA_DIRECTORY)/python/test \
 		python -m pytest $(NNABLA_DIRECTORY)/python/test
 
-.PHNOY: nnabla-ext-cuda-multi-gpu-test-local
+.PHONY: nnabla-ext-cuda-multi-gpu-test-local
 nnabla-ext-cuda-multi-gpu-test-local: nnabla-ext-cuda-multi-gpu-install
 	cd $(BUILD_EXT_CUDA_DIRECTORY_WHEEL_MULTI_GPU) \
 	&& PYTHONPATH=$(NNABLA_EXT_CUDA_DIRECTORY)/python/test:$(NNABLA_DIRECTORY)/python/test \

doc/build/build.md

@@ -7,9 +7,9 @@ This document shows how to install CUDA extension on Ubuntu 16.04 LTS. This proc
 
 ## Prerequisites
 
-In addition to NNabla's requirements, CUDA extension requires CUDA setup has done on your system. If you don't have CUDA on your system, follow the procedure desribed below.
+In addition to NNabla's requirements, CUDA extension requires CUDA setup has done on your system. If you don't have CUDA on your system, follow the procedure described below.
 
-Download and install CUDA and cuDNN library (both runtime library and developement library). Please follow the instruction in the document provided by NVIDIA. Do NOT see any instruction provided by any third party. They are often incorrect or based on old instructions, that could destroy your system.
+Download and install CUDA and cuDNN library (both runtime library and development library). Please follow the instruction in the document provided by NVIDIA. Do NOT see any instruction provided by any third party. They are often incorrect or based on old instructions, that could destroy your system.
 
 * [CUDA toolkit](https://developer.nvidia.com/cuda-downloads)
 * [cuDNN library](https://developer.nvidia.com/rdp/cudnn-download) (Registration required)

doc/build/build_distributed.md

@@ -15,7 +15,7 @@ In addition to [requirements of NNabla without distributed execution](build.md),
 In order to use the distributed training, the only difference, when building, is
 the procedure described here.
 
-Download `nccl <https://developer.nvidia.com/nccl/nccl-download>`_ according to your environemnt,
+Download `nccl <https://developer.nvidia.com/nccl/nccl-download>`_ according to your environment,
 then install it manually in case of ubuntu16.04,
 
 ```shell
@@ -25,13 +25,13 @@ sudo apt-get install libnccl2 libnccl-dev
 ```
 
 For developer, if you want to use another nccl not publicly distributed,
-specify **NCCL_HOME** environment variable as the folloing.
+specify **NCCL_HOME** environment variable as the following.
 
 ```shell
 export NCCL_HOME=${path}/build
 ```
 
-Here, we assume the directry structure,
+Here, we assume the directory structure,
 
 * ${path}/build/include
 * ${path}/build/lib
@@ -66,7 +66,7 @@ CUDA includes: /usr/local/cuda-8.0/include;/usr/lib/openmpi/include/openmpi/opal
 ## Unit test
 
 
-Follow the unit test section in [Build CUDA extension](build.md). Now you could see the communicater
+Follow the unit test section in [Build CUDA extension](build.md). Now you could see the communicator
 test passed.
 
 ```

doc/build/build_windows.md

@@ -2,10 +2,10 @@
 
 ## Prerequisites
 
-In addition to NNabla's requirements, CUDA extension requires CUDA setup has done on your system. If you don't have CUDA on your system, follow the procedure desribed below.
+In addition to NNabla's requirements, CUDA extension requires CUDA setup has done on your system. If you don't have CUDA on your system, follow the procedure described below.
 
 
-Download and install CUDA and cuDNN library (both runtime library and developement library). Please follow the instruction in the document provided by NVIDIA. Do NOT see any instruction provided by any third party. They are often incorrect or based on old instructions, that could destroy your system.
+Download and install CUDA and cuDNN library (both runtime library and development library). Please follow the instruction in the document provided by NVIDIA. Do NOT see any instruction provided by any third party. They are often incorrect or based on old instructions, that could destroy your system.
 
 * [CUDA toolkit](https://developer.nvidia.com/cuda-downloads)
 * [cuDNN library](https://developer.nvidia.com/rdp/cudnn-download) (Registration required)

doc/build/quick_build_tools.md

@@ -47,7 +47,7 @@ Install CUDA8.0, CUDA9.0, CUDA9.1 from following site.
  - https://developer.nvidia.com/cuda-toolkit-archive
 
 
-Get several versions of cuDNN from following site. (Registration requried)
+Get several versions of cuDNN from following site. (Registration required)
 - cuDNN
  - https://developer.nvidia.com/rdp/cudnn-download
 

docker/README.md

@@ -38,6 +38,6 @@ nvidia-docker run -it --rm -p 8888:8888 nnabla/nnabla-ext-cuda:tutorial jupyter
 You can connect the jupyter server with your browser by accessing
 `http://<Host OS address>:8888`. The login password is `nnabla`.
 
-After logging in, the page lists a directory that contains jupyter `.ipynb` tutorials and the `nnabla-examples/` foler.
+After logging in, the page lists a directory that contains jupyter `.ipynb` tutorials and the `nnabla-examples/` folder.
 You can open any tutorial by clicking a `.ipynb` file.
 A DCGAN in `nnabla-examples` is demonstrated in `run-nnabla-examples.ipynb`.

docker/tutorial/run-nnabla-examples.ipynb

@@ -4,7 +4,7 @@
    "cell_type": "markdown",
    "metadata": {},
    "source": [
-    "The following command executes training of Deep Convolutional GAN on MNIST dataset. It takes a couple of minites on an average GPU."
+    "The following command executes training of Deep Convolutional GAN on MNIST dataset. It takes a couple of minutes on an average GPU."
    ]
   },
   {

examples/cpp/mnist_training/README.md

@@ -17,7 +17,7 @@ If you tried the CPU version of this script, you must have downloaded MNIST data
 
 Please copy them to this directory.
 
-## Create NNP file of an initialized model for MNIST classifcation.
+## Create NNP file of an initialized model for MNIST classification.
 You might also have an NNP file of the initialized model in `nnabla/examples/cpp/mnist_training`.
 Please copy it to this directory.
 
@@ -31,9 +31,9 @@ export NNABLA_DIR='path to your nnabla directory'
 make
 ```
 
-The above command generates an executable `mnist_training_cuda` at the current directry.
+The above command generates an executable `mnist_training_cuda` at the current directory.
 
-The build file `GNUMakefile` is simple.
+The build file `GNUmakefile` is simple.
 It links `libnnabla.so`, `libnnabla_utils.so`, `libnnabla_utils.so` and `libz.so` with the executable generated from `main.cpp`, and compiles with C++11 option `-std=c++11`.
 It also needs to include path to `mnist_training.hpp` which is located in `nnabla/examples/cpp/mnist_training` directory.
 

include/nbla/cuda/array/cuda_array.hpp

@@ -31,7 +31,7 @@ using std::shared_ptr;
 class CudaArray : public Array {
 protected:
   int device_;
-  /* Holding CudaMemory until the instance is destoryed to prevent freeing.
+  /* Holding CudaMemory until the instance is destroyed to prevent freeing.
    */
   shared_ptr<CudaMemory> inuse_memory_;
 
@@ -54,7 +54,7 @@ NBLA_CUDA_API void synchronizer_cpu_array_cuda_array(Array *src, Array *dst);
 
 /** Array allocated on CUDA device with Memory Pool
 
-This is a necessary ingredient for imperative programing interface of
+This is a necessary ingredient for imperative programming interface of
 neural networks (aka define-by-run or dynamic). Memory allocation of
 CUDA is not asynchronous. Hence, allocating memory region between each
 function will lead thread synchronization that will block executions of

include/nbla/cuda/common.hpp

@@ -162,7 +162,7 @@ CUBLAS_TYPE_T(HalfCuda, HALF);
 /** Block size */
 #define NBLA_CUDA_GET_BLOCKS(num) NBLA_CEIL_INT_DIV(num, NBLA_CUDA_NUM_THREADS)
 
-/** Get an appropreate block size given a size of elements.
+/** Get an appropriate block size given a size of elements.
 
     The kernel is assumed to contain a grid-strided loop.
  */

include/nbla/cuda/communicator/data_parallel_communicator.hpp

@@ -127,7 +127,7 @@ class NBLA_API DataParallelCommunicatorNccl
 protected:
   void wait_by_devices_synchronization();
   void wait_by_streams_synchronization();
-  void divide_by_num_divices(bool division);
+  void divide_by_num_devices(bool division);
 
   DISABLE_COPY_AND_ASSIGN(DataParallelCommunicatorNccl);
 };

include/nbla/cuda/communicator/multi_process_data_parallel_communicator.hpp

@@ -215,7 +215,7 @@ class NBLA_API MultiProcessDataParallelCommunicatorNccl
 
   void wait_by_device_synchronization();
   void wait_by_streams_synchronization();
-  void divide_by_num_divices(bool division);
+  void divide_by_num_devices(bool division);
 
   DISABLE_COPY_AND_ASSIGN(MultiProcessDataParallelCommunicatorNccl);
 };

include/nbla/cuda/cudnn/cudnn.hpp

@@ -62,7 +62,7 @@ template <> class cudnn_data_type<HalfCuda> {
   static cudnnDataType_t type() { return CUDNN_DATA_HALF; }
 };
 
-/** Convret cuDNN enum dtype to NNabla enum dtype.
+/** Convert cuDNN enum dtype to NNabla enum dtype.
  */
 inline dtypes get_dtype_by_cudnn_data_type(cudnnDataType_t dtype) {
   switch (dtype) {
@@ -135,7 +135,7 @@ inline string cudnn_status_to_string(cudnnStatus_t status) {
 
 http://docs.nvidia.com/deeplearning/sdk/cudnn-developer-guide/index.html#cudnnSetTensorNdDescriptor
 
-According to the doc above, cudnnSetTensorNdDescriptor does not suport a tensor
+According to the doc above, cudnnSetTensorNdDescriptor does not support a tensor
 less than 4 dimensions. This wrapper function adds unused dimensions with a
 value of 1 at last.
 
@@ -203,7 +203,7 @@ struct NBLA_CUDA_API CudnnConvResource {
   cudnnTensorDescriptor_t y_desc;         ///< Output desc.
   cudnnTensorDescriptor_t b_desc;         ///< Bias desc.
   cudnnTensorDescriptor_t b_desc_deconv;  ///< Bias desc for deconvolution.
-  cudnnFilterDescriptor_t w_desc;         ///< Wegiht desc.
+  cudnnFilterDescriptor_t w_desc;         ///< Weight desc.
   cudnnConvolutionDescriptor_t conv_desc; ///< Conv desc.
   cudnnConvolutionFwdAlgo_t fwd_algo;     ///< Best forward algorithm found.
   cudnnConvolutionBwdFilterAlgo_t
@@ -236,7 +236,7 @@ class NBLA_CUDA_API CudnnHandleManager {
   ~CudnnHandleManager();
 
   /**
-     Get cuDNN handle for devive.
+     Get cuDNN handle for device.
    */
   cudnnHandle_t handle(int device = -1);
 

include/nbla/cuda/cudnn/function/convolution.hpp

@@ -39,7 +39,7 @@ template <typename T> class ConvolutionCudaCudnn : public Convolution<T> {
     // NOTE: dilation > 1 is not supported by cudnn. (2016.10.19)
     for (int i = 0; i < dilation.size(); ++i) {
       if (dilation[i] > 1) {
-        // Fall back to origianl CUDA implementation if dilation > 1.
+        // Fall back to original CUDA implementation if dilation > 1.
         // Setting fall_back_func_ overwrites behaviors of setup, forward and
         // backward functions by the specified function class instance.
         std::cout << "Falling back to ConvolutionCuda since dilation > 1 is "

include/nbla/cuda/cudnn/function/deconvolution.hpp

@@ -42,7 +42,7 @@ template <typename T> class DeconvolutionCudaCudnn : public Deconvolution<T> {
     // NOTE: dilation > 1 is not supported by cudnn. (2016.10.19)
     for (int i = 0; i < dilation.size(); ++i) {
       if (dilation[i] > 1) {
-        // Fall back to origianl CUDA implementation if dilation > 1.
+        // Fall back to original CUDA implementation if dilation > 1.
         // Setting fall_back_func_ overwrites behaviors of setup, forward and
         // backward functions by the specified function class instance.
         std::cout << "Falling back to DeconvolutionCuda since dilation > 1 is "

include/nbla/cuda/cudnn/function/function_impl.hpp.tmpl

@@ -38,7 +38,7 @@ template <${dec_targs}> class ${name}CudaCudnn : public ${name}<${targs}> {
 public:
   /* TODO: remove this help message.
   Typedef of CUDA scalar types used in source file.
-  This template function class might be instanciated for each CPU scalar types
+  This template function class might be instantiated for each CPU scalar types
   (double, float, nbla::Half), however, for Half, CUDA kernel functions
   must use nbla::HalfCuda in which a bunch of device operator functions are
   overloaded. nbla::CudaType<T>::type will translate nbla::Half
@@ -67,7 +67,7 @@ public:
 % for oname in outputs.keys():
     NBLA_CUDNN_CHECK(cudnnDestroyTensorDescriptor(${oname}_desc_));
 % endfor
-    // TODO: Destoy other descriptors
+    // TODO: Destroy other descriptors
   }
   virtual string name() { return "${name}CudaCudnn"; }
   virtual vector<string> allowed_array_classes() {

include/nbla/cuda/cudnn/function/softmax.hpp

@@ -24,7 +24,7 @@ namespace nbla {
 
 /** @copydoc Softmax
 
-@note The default algrithm is set as ACCURATE. TODO: Set an algorithm by
+@note The default algorithm is set as ACCURATE. TODO: Set an algorithm by
       context.
 */
 template <typename T> class SoftmaxCudaCudnn : public Softmax<T> {

include/nbla/cuda/function/function_impl.hpp.tmpl

@@ -32,7 +32,7 @@ template <${dec_targs}> class ${name}Cuda : public ${name}<${targs}> {
 public:
   /* TODO: remove this help message.
   Typedef of CUDA scalar types used in source file.
-  This template function class might be instanciated for each CPU scalar types
+  This template function class might be instantiated for each CPU scalar types
   (double, float, nbla::Half), however, for Half, CUDA kernel functions
   must use nbla::HalfCuda in which a bunch of device operator functions are
   overloaded. nbla::CudaType<T>::type will translate nbla::Half

include/nbla/cuda/function/interpolate.hpp

@@ -24,7 +24,7 @@ template <typename T> class InterpolateCuda : public Interpolate<T> {
 public:
   /* TODO: remove this help message.
   Typedef of CUDA scalar types used in source file.
-  This template function class might be instanciated for each CPU scalar types
+  This template function class might be instantiated for each CPU scalar types
   (double, float, nbla::Half), however, for Half, CUDA kernel functions
   must use nbla::HalfCuda in which a bunch of device operator functions are
   overloaded. nbla::CudaType<T>::type will translate nbla::Half

include/nbla/cuda/utils/device_reduce.cuh

@@ -21,10 +21,10 @@
 
 namespace nbla {
 
-/** Geric block-wise reduction kernel.
+/** Generic block-wise reduction kernel.
 
 @param[in] N Number of valid input items.
-@param[in,out] op Reduciton operator class. TODO: doc.
+@param[in,out] op Reduction operator class. TODO: doc.
  */
 template <class ReduceOp>
 __global__ void kernel_reduce_per_block(const int N, ReduceOp op,

python/setup.py

@@ -203,7 +203,7 @@ def get_setup_config(root_dir):
         version=__version__,
         author_email=__email__,
         url="https://github.com/sony/nnabla-ext-cuda",
-        license='Apache Licence 2.0',
+        license='Apache License 2.0',
         classifiers=[
                 'Development Status :: 4 - Beta',
                 'Intended Audience :: Developers',

python/src/nnabla_ext/cuda/init.pyx

@@ -44,7 +44,7 @@ def clear_memory_cache():
 
 
 ###############################################################################
-# Array pereference API
+# Array preference API
 # TODO: Move these to C++
 ###############################################################################
 _original_array_classes = cuda_array_classes()
@@ -94,7 +94,7 @@ def device_synchronize(str device):
 def get_device_count():
     """Call ``cudaGetDeviceCount`` in runtime API`.
 
-    Retuns:
+    Returns:
         int: Number of devices available.
 
     """