merge with 38f7c55

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
eric-haibin-lin · May 24, 2017 · 9db62c0 · 9db62c0
1 parent 35d5e54
commit 9db62c0
Show file tree

Hide file tree

Showing 86 changed files with 5,911 additions and 532 deletions.
diff --git a/Jenkinsfile b/Jenkinsfile
@@ -205,17 +205,17 @@ 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"
   }
 }

diff --git a/Makefile b/Makefile
@@ -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 -DNDEBUG=1
 endif

diff --git a/dmlc-core b/dmlc-core
diff --git a/include/mxnet/c_api.h b/include/mxnet/c_api.h
@@ -244,6 +244,38 @@ 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
@@ -356,6 +388,19 @@ 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
@@ -366,6 +411,13 @@ 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
@@ -404,6 +456,26 @@ 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
@@ -935,6 +1007,25 @@ 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
 //--------------------------------------------
@@ -1081,6 +1172,39 @@ 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
  */

diff --git a/include/mxnet/executor.h b/include/mxnet/executor.h
@@ -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.
@@ -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
    */