▼ 3rdparty | |
▼ dlpack | |
▼ include | |
► dlpack | |
dlpack.h | The common header of DLPack |
▼ dmlc-core | |
▼ include | |
► dmlc | |
any.h | Container to hold any data type |
array_view.h | Read only data structure to reference array |
base.h | |
blockingconcurrentqueue.h | |
build_config_default.h | Default detection logic for fopen64 and other symbols. May be overriden by CMake |
common.h | Defines some common utility function |
concurrency.h | Thread-safe data structures |
concurrentqueue.h | |
config.h | Defines config parser class |
data.h | Defines common input data structure, and interface for handling the input data |
endian.h | Endian testing, need c++11 |
filesystem.h | Utilities to manipulate files |
input_split_shuffle.h | Base class to construct input split with global shuffling |
io.h | |
json.h | Lightweight JSON Reader/Writer that read save into C++ data structs. This includes STL composites and structures |
lua.h | C++11 header only interface to easily interact with Lua and Torch. This code is evolved from torch plugin code for MXNet |
memory.h | Additional memory hanlding utilities |
memory_io.h | Defines binary serialization class to serialize things into/from memory region |
omp.h | Header to handle OpenMP compatibility issues |
optional.h | Container to hold optional data |
parameter.h | Provide lightweight util to do parameter setup and checking |
recordio.h | Recordio that is able to pack binary data into a splittable format, useful to exchange data in binary serialization, such as binary raw data or protobuf |
registry.h | Registry utility that helps to build registry singletons |
serializer.h | Serializer template class that helps serialization. This file do not need to be directly used by most user |
strtonum.h | A faster implementation of strtof and strtod |
thread_group.h | Thread and synchronization primitives and lifecycle management |
thread_local.h | Portable thread local storage |
threadediter.h | Thread backed iterator that can be used to implement general thread-based pipeline such as prefetch and pre-computation To use the functions in this header, C++11 is required |
timer.h | Cross platform timer for timing |
type_traits.h | Type traits information header |
▼ mshadow | |
▼ mshadow | |
► extension | |
broadcast.h | Support for broadcast and repmat |
broadcast_with_axis.h | |
channel_pool.h | Support for chpool |
channel_unpool.h | |
choose.h | Support for implicit array selection operation |
complex.h | Support for complex operations |
concat.h | Support for concatenation |
crop.h | Support for crop |
fill.h | Support for implicit array filling operation |
flip.h | Support for flip a certain dimension |
implicit_gemm.h | Support for implicit GEMM operation |
mask.h | |
mirror.h | Support for mirror |
one_hot.h | Create one-hot indicator array based on the index |
pack_col2patch.h | Support for pack |
pad.h | Support for pad |
range.h | Support generating a range vector |
reduce_with_axis.h | |
reduceto1d.h | Support for sum_rows and sumall_except_dim |
reshape.h | Support for reshape |
slice.h | Support for slice a certain dimension |
slice_ex.h | |
spatial_pool.h | Support for spatial pooling |
spatial_unpool.h | Support for unpool |
spatial_upsampling_nearest.h | |
swapaxis.h | Support for swapaxis |
take.h | |
take_grad.h | |
transpose.h | Support for transpose |
unpack_patch2col.h | Support for unpack |
► packet | |
plain-inl.h | Support of plain packet that use the plain datatype |
sse-inl.h | Support of sse2 packet optimization of some operations |
base.h | |
dot_engine-inl.h | Definitions of how Matrix Multiplications can be evaluated |
expr_engine-inl.h | Definitions of how expressions should be evaluated |
expr_scalar-inl.h | Definitions of operators in expression with respect to scalar this file will be included several times, each time with MACRO MSHADOW_SCALAR_ to be different types |
expression.h | Definitions of abstract expressions and expressions template |
extension.h | Some extension of expressions, used to support something beyond elementwise op |
half.h | Definition of half (float16) type |
half2.h | Definition of vector float16, half2 type |
io.h | |
packet-inl.h | Generic packet vectorization code |
random.h | Random inline functions for tensor |
stream_gpu-inl.h | Implementation of GPU code |
tensor.h | |
tensor_container.h | Tensor container that does memory allocation and resize like STL |
tensor_cpu-inl.h | Implementation of CPU host code |
tensor_gpu-inl.h | Implementation of GPU host code |
▼ tvm | |
▼ nnvm | |
► include | |
► nnvm | |
► compiler | |
op_attr_types.h | The Expr and related elements in DataFlow construction |
packed_func_ext.h | Extension to enable packed functionn for nnvm types |
util.h | Utility functions for nnvm compiler |
► top | |
nn.h | Auxiliary param for tensor primitive |
tensor.h | Auxiliary param for tensor primitive |
base.h | Configuration of nnvm as well as basic data structure |
c_api.h | C API of NNVM symbolic construction and pass. Enables construction and transformation of Graph in any other host languages |
graph.h | Configuation of nnvm as well as basic data structure |
graph_attr_types.h | Data structures that can appear in graph attributes |
layout.h | Layout expression. The layout is composed of upper cases, lower cases and numbers, where upper case indicates a (super-)dimension and the corresponding lower case with factor size indicates the split (sub-)dimension. For example, NCHW16c can describe a 5-D tensor of [batch_size, channel, height, width, channel_block]. Here sub-dimension channel_block=16 is the split of super-dimension C (channel) |
node.h | Graph node data structure |
op.h | Operator information structor |
op_attr_types.h | Data structures that can appear in operator attributes |
pass.h | Pass that can be applied to a graph |
pass_functions.h | Pass functions that simply redirect the calls to ApplyPass |
symbolic.h | Symbolic graph construction API |
tuple.h | Data structure Tuple and TShape to store dynamic sized shapes |
▼ cpp-package | |
▼ include | |
▼ mxnet-cpp | |
base.h | |
contrib.h | Utility function to enable some contrib features |
executor.h | |
initializer.h | Random initializer |
io.h | |
kvstore.h | |
lr_scheduler.h | Scheduling learning rate |
metric.h | |
model.h | MXNET.cpp model module |
monitor.h | Monitor definition |
MxNetCpp.h | Meta include file for mxnet.cpp |
ndarray.h | |
op_map.h | Definition of OpMap |
op_suppl.h | A supplement and amendment of the operators from op.h |
op_util.h | Operator helper functions |
operator.h | |
optimizer.h | Definition of optimizer |
shape.h | Definition of shape |
symbol.h | Definition of symbol |
▼ include | |
▼ mkldnn | |
mkldnn.h | |
mkldnn_types.h | |
▼ mxnet | |
base.h | |
c_api.h | |
c_api_error.h | Error handling for C API |
c_api_test.h | C API of mxnet for ease of testing backend in Python |
c_predict_api.h | C predict API of mxnet, contains a minimum API to run prediction. This file is self-contained, and do not dependent on any other files |
engine.h | Engine that schedules all the operations according to dependency |
executor.h | |
graph_attr_types.h | |
imperative.h | |
io.h | |
kvstore.h | |
lib_api.h | APIs to interact with libraries |
libinfo.h | Get features of the MXNet library at runtime |
ndarray.h | |
op_attr_types.h | |
operator.h | |
operator_util.h | Utility functions and registries to help quickly build new operators. [Deprecated] Use the register functions in this file when possible to simplify operator creations. Operators registered in this file will be exposed to both NDArray API and symbolic API |
random_generator.h | Parallel random number generator |
resource.h | Global resource allocation handling |
rtc.h | |
storage.h | Storage manager across multiple devices |
tensor_blob.h | TBlob class that holds common representation of arbirary dimension tensor, can be used to transformed to normal fixed dimenson tensor |
tuple.h | Data structure Tuple and TShape to store dynamic sized shapes |
▼ src | |
▼ common | |
cuda_utils.h | Common CUDA utilities |
exec_utils.h | Common utility functions for executors |
lazy_alloc_array.h | An array that lazily allocate elements as First time the cell get visited |
object_pool.h | |
static_array.h | |
tensor_inspector.h | Utility to inspect tensor objects |
utils.h | Basic utilility functions |