Skip to content

Commit

Permalink
merge with 38f7c55
Browse files Browse the repository at this point in the history
compiles on GPU

update check alloc:

Checkpoint. Pass elem-sum gpu test

bug fix for copyfromto. sparse sgd test pass on gpu

inefficient implementation for csr copy

update submodule

fix lint

Simple bind with infer storage type (#32)

* Symbol binding for sparse tensor development. (#31)

* Initial checkin

* Add init functions for simple bind in graph_executor

* Add simple_bind c_api

* Add simple bind c-api

* Assign zeros to in_args, arg_grads, and aux_states

* Add simple_bind2 python interface

* Fix python interface bugs

* Interface changes

* Fix

* Fix core dump

* Add bind_ith_exec c_api

* Change simple_bind2

* Fix seg fault

* Finish simple_bind

* Change _bind_ith_exec

* Refactor simple_bind initialization flow for bind

* Consolidate bind and simple_bind graph init flow

* Fix bug

* Clean up

* Add comments

* Clean up

* Clean up

* Minor correction

* Rename APIs in graph executor

* Refactor

* Rebase

* Delete deprecated functions

* Move more front-end work to backend

* Bug fix

* Fix failed tests

* Minor fix

* Fix lint

* Fix lint

* Revert unnecessary changes

* Revert

* Revert

* Clean up

* Fix lint

Conflicts:
	python/mxnet/symbol.py
	src/executor/graph_executor.cc

* Add inferstorage to graph executor

* re-enable tests for sparse embedding with simple_bind

* type switch fix in sparse embedding"
;

change `default` to `default_storage` for cast storage op (#33)

* change default to default_storage

* disable cpp test build temporarily

attempt to fix windows build error, and fix lint (#34)

update nnvm submodule (#37)

Scipy build (#38)

* update nnvm submodule

* add scipy pip install for dockerfile

Python3 unit tests (#39)

* change xrange to range for python3 compatiblity"

* remove more xrange from tests

replace long with int for python3 (#40)

fix the rest of TShape constructor errors (#41)

fix lint (#42)

fix wrong usage of mshadow::Shape1" (#43)

implementation for Csr slice on cpu (#36)

* CPU implementation for CSR

remove seg_len from csr slice

add some docs for slice csr

change indptr, values, etc to be private member

bug fix in sparse embedding

update nnvm submoduel

fix lint

update unit test for sparse nd"

* add const for SliceCsrIndPtr kernel

Fix sparse dot according to the new RSP definition (#35)

* Fix csr dot dns

* Fix sparse dot

* Add fallback and test cases for dot(csr, dns)=dns

* Add int type switch

* Fix

* Fix

* Fix

update mshadow submodule (#44)

Fix dns to rsp (#46)

fix lint (#47)

add runtime storage fallback detection" (#48)

* add runtime storage fallback detection"

* replace cast storage ex with cast storage impl

Fm example (#45)

* update csr slice logic to avoid confusion. add more exmaples.

* add hint to module.update

* more testcases(fallback) for sparse_nd

* add to_csr() and to_rsp() method. More unit test (fallback now)

* add fm test. fix lint

* register sparse sgd under Optim.SGD

* update dmlc-core submoduel

* change indptr to _indptr temporarily. add const ref to fname
  • Loading branch information
eric-haibin-lin committed May 24, 2017
1 parent cde5361 commit 2367445
Show file tree
Hide file tree
Showing 76 changed files with 5,894 additions and 523 deletions.
6 changes: 3 additions & 3 deletions Jenkinsfile
Original file line number Diff line number Diff line change
Expand Up @@ -204,17 +204,17 @@ del /Q *.7z
// Python unittest for CPU
def python_ut(docker_type) {
timeout(time: max_time, unit: 'MINUTES') {
sh "${docker_run} ${docker_type} PYTHONPATH=./python/ nosetests --with-timer --verbose tests/python/unittest"
sh "${docker_run} ${docker_type} PYTHONPATH=./python/ nosetests-2.7 --with-timer --verbose tests/python/unittest"
sh "${docker_run} ${docker_type} PYTHONPATH=./python/ nosetests-3.4 --with-timer --verbose tests/python/unittest"
sh "${docker_run} ${docker_type} PYTHONPATH=./python/ nosetests --with-timer --verbose tests/python/train"
sh "${docker_run} ${docker_type} PYTHONPATH=./python/ nosetests-2.7 --with-timer --verbose tests/python/train"
}
}

// GPU test has two parts. 1) run unittest on GPU, 2) compare the results on
// both CPU and GPU
def python_gpu_ut(docker_type) {
timeout(time: max_time, unit: 'MINUTES') {
sh "${docker_run} ${docker_type} PYTHONPATH=./python/ nosetests --with-timer --verbose tests/python/gpu"
sh "${docker_run} ${docker_type} PYTHONPATH=./python/ nosetests-2.7 --with-timer --verbose tests/python/gpu"
sh "${docker_run} ${docker_type} PYTHONPATH=./python/ nosetests-3.4 --with-timer --verbose tests/python/gpu"
}
}
Expand Down
3 changes: 2 additions & 1 deletion Makefile
Original file line number Diff line number Diff line change
Expand Up @@ -44,8 +44,9 @@ ifeq ($(DEV), 1)
endif

# CFLAGS for debug
# FIXME(haibin) temporarily turn on -DDMLC_LOG_FATAL_THROW for debug
ifeq ($(DEBUG), 1)
CFLAGS += -g -O0
CFLAGS += -g -O0 -DDMLC_LOG_FATAL_THROW=1
else
CFLAGS += -O3
endif
Expand Down
124 changes: 124 additions & 0 deletions include/mxnet/c_api.h
Original file line number Diff line number Diff line change
Expand Up @@ -241,6 +241,38 @@ MXNET_DLL int MXNDArrayCreateEx(const mx_uint *shape,
int delay_alloc,
int dtype,
NDArrayHandle *out);


/*!
* \brief create an empty sparse NDArray with specified shape and data type
* \param storage_type the storage type of the ndarray
* \param shape the pointer to the shape
* \param ndim the dimension of the shape
* \param dev_type device type, specify device we want to take
* \param dev_id the device id of the specific device
* \param delay_alloc whether to delay allocation until
* the narray is first mutated
* \param dtype data type of created array
* \param num_aux the number of aux data to support this ndarray
* \param aux_type data type of the aux data for the created array
* \param aux_ndims the dimension of the shapes of aux data
* \param aux_shape the shapes of aux data
* \param out the returning handle
* \return 0 when success, -1 when failure happens
*/
MXNET_DLL int MXNDArrayCreateSparseEx(int storage_type,
const mx_uint *shape,
mx_uint ndim,
int dev_type,
int dev_id,
int delay_alloc,
int dtype,
mx_uint num_aux,
int *aux_type,
mx_uint *aux_ndims,
const mx_uint *aux_shape,
NDArrayHandle *out);

/*!
* \brief create a NDArray handle that is loaded from raw bytes.
* \param buf the head of the raw bytes
Expand Down Expand Up @@ -353,6 +385,19 @@ MXNET_DLL int MXNDArraySlice(NDArrayHandle handle,
mx_uint slice_begin,
mx_uint slice_end,
NDArrayHandle *out);

/*!
* \brief Slice the NDArray with non-default storage along axis 0.
* \param handle the handle to the NDArray
* \param slice_begin The beginning index of slice
* \param slice_end The ending index of slice
* \param out The NDArrayHandle of sliced NDArray
* \return 0 when success, -1 when failure happens
*/
MXNET_DLL int MXNDArraySliceEx(NDArrayHandle handle,
mx_uint slice_begin,
mx_uint slice_end,
NDArrayHandle out);
/*!
* \brief Index the NDArray along axis 0.
* \param handle the handle to the NDArray
Expand All @@ -363,6 +408,13 @@ MXNET_DLL int MXNDArraySlice(NDArrayHandle handle,
MXNET_DLL int MXNDArrayAt(NDArrayHandle handle,
mx_uint idx,
NDArrayHandle *out);

/*!
* \brief get the storage type of the array
*/
MXNET_DLL int MXNDArrayGetStorageType(NDArrayHandle handle,
int *out_storage_type);

/*!
* \brief Reshape the NDArray.
* \param handle the handle to the narray
Expand Down Expand Up @@ -401,6 +453,26 @@ MXNET_DLL int MXNDArrayGetData(NDArrayHandle handle,
*/
MXNET_DLL int MXNDArrayGetDType(NDArrayHandle handle,
int *out_dtype);

/*!
* \brief get the type of the ith aux data in NDArray
* \param handle the handle to the narray
* \param i the index of the aux data
* \param out_type pointer holder to get type of aux data
* \return 0 when success, -1 when failure happens
*/
MXNET_DLL int MXNDArrayGetAuxType(NDArrayHandle handle,
mx_uint i,
int *out_type);

// Get the ith aux data blob wrapped in an NDArray
MXNET_DLL int MXNDArrayGetAuxNDArray(NDArrayHandle handle,
mx_uint i,
NDArrayHandle *out);

// Get the data blob wrapped in an NDArray
MXNET_DLL int MXNDArrayGetDataNDArray(NDArrayHandle handle,
NDArrayHandle *out);
/*!
* \brief get the context of the NDArray
* \param handle the handle to the narray
Expand Down Expand Up @@ -932,6 +1004,25 @@ MXNET_DLL int MXSymbolInferType(SymbolHandle sym,
mx_uint *aux_type_size,
const int **aux_type_data,
int *complete);




/*!
* \brief infer storage type of unknown input types given the known one.
*/
MXNET_DLL int MXSymbolInferStorageType(SymbolHandle sym,
mx_uint num_args,
const char** keys,
const int *arg_storage_type_data,
mx_uint *in_storage_type_size,
const int **in_storage_type_data,
mx_uint *out_storage_type_size,
const int **out_storage_type_data,
mx_uint *aux_storage_type_size,
const int **aux_storage_type_data,
int *complete);

//--------------------------------------------
// Part 4: Executor interface
//--------------------------------------------
Expand Down Expand Up @@ -1078,6 +1169,39 @@ MXNET_DLL int MXExecutorBindEX(SymbolHandle symbol_handle,
NDArrayHandle *aux_states,
ExecutorHandle shared_exec,
ExecutorHandle *out);

MXNET_DLL int MXExecutorSimpleBind(SymbolHandle symbol_handle,
int dev_type,
int dev_id,
const mx_uint num_g2c_keys,
const char** g2c_keys,
const int* g2c_dev_types,
const int* g2c_dev_ids,
const mx_uint provided_grad_req_list_len,
const char** provided_grad_req_names,
const char** provided_grad_req_types,
const mx_uint num_provided_arg_shapes,
const char** provided_arg_shape_names,
const mx_uint* provided_arg_shape_data,
const mx_uint* provided_arg_shape_idx,
const mx_uint num_provided_arg_dtypes,
const char** provided_arg_dtype_names,
const int* provided_arg_dtypes,
const mx_uint num_provided_arg_stypes,
const char** provided_arg_stype_names,
const int* provided_arg_stypes,
const mx_uint num_shared_arg_names,
const char** shared_arg_name_list,
mx_uint* shared_buffer_len,
const char*** shared_buffer_name_list,
NDArrayHandle** shared_buffer_handle_list,
mx_uint* num_in_args,
NDArrayHandle** in_args,
NDArrayHandle** arg_grads,
mx_uint* num_aux_states,
NDArrayHandle** aux_states,
ExecutorHandle shared_exec_handle,
ExecutorHandle* out);
/*!
* \brief set a call back to notify the completion of operation
*/
Expand Down
33 changes: 33 additions & 0 deletions include/mxnet/executor.h
Original file line number Diff line number Diff line change
Expand Up @@ -69,6 +69,21 @@ class Executor {
* \return array of outputs in the executor.
*/
virtual const std::vector<NDArray> &outputs() const = 0;
/*!
* \brief get input argument map, key is arg name, value is arg's NDArray.
* \return input argument map in the executor.
*/
virtual const std::unordered_map<std::string, NDArray>& in_arg_map() const = 0;
/*!
* \brief get input argument graident map, key is arg name, value is gradient's NDArray.
* \return input argument gradient map in the executor.
*/
virtual const std::unordered_map<std::string, NDArray>& arg_grad_map() const = 0;
/*!
* \brief get aux state map, key is arg name, value is aux state's NDArray.
* \return aux state map in the executor.
*/
virtual const std::unordered_map<std::string, NDArray>& aux_state_map() const = 0;
/*!
* \brief Create an operator by bind symbol with context and arguments.
* If user do not want to compute the gradients of i-th argument, grad_req_type[i] can be kNullOp.
Expand All @@ -91,6 +106,24 @@ class Executor {
const std::vector<OpReqType> &grad_req_type,
const std::vector<NDArray> &aux_states,
Executor* shared_exec = NULL);

static Executor* SimpleBind(nnvm::Symbol symbol,
const Context& default_ctx,
const std::map<std::string, Context>& group2ctx,
const std::vector<Context>& in_arg_ctxes,
const std::vector<Context>& arg_grad_ctxes,
const std::vector<Context>& aux_state_ctxes,
const std::unordered_map<std::string, TShape>& arg_shape_map,
const std::unordered_map<std::string, int>& arg_dtype_map,
const std::unordered_map<std::string, int>& arg_stype_map,
const std::vector<OpReqType>& grad_req_types,
const std::unordered_set<std::string>& param_names,
std::vector<NDArray>* in_args,
std::vector<NDArray>* arg_grads,
std::vector<NDArray>* aux_states,
std::unordered_map<std::string, NDArray>*
shared_data_arrays = nullptr,
Executor* shared_exec = nullptr);
/*!
* \brief the prototype of user-defined monitor callback
*/
Expand Down
Loading

0 comments on commit 2367445

Please sign in to comment.