Merge branch 'master' of https://github.com/apache/incubator-mxnet in…

…to erf-opt

.gitmodules

@@ -20,12 +20,12 @@
 	path = 3rdparty/mkldnn
 	url = https://github.com/intel/mkl-dnn.git
 	branch = master
-[submodule "3rdparty/cub"]
-	path = 3rdparty/cub
-	url = https://github.com/dmlc/cub
 [submodule "3rdparty/tvm"]
 	path = 3rdparty/tvm
 	url = https://github.com/dmlc/tvm
 [submodule "3rdparty/onnx-tensorrt"]
 	path = 3rdparty/onnx-tensorrt
 	url = https://github.com/onnx/onnx-tensorrt.git
+[submodule "3rdparty/nvidia_cub"]
+	path = 3rdparty/nvidia_cub
+	url = https://github.com/NVlabs/cub.git

3rdparty/sparse-matrix/Makefile

@@ -0,0 +1,21 @@
+CC = g++
+C = gcc
+MKLROOT = /opt/intel/mkl
+
+ifneq ($(USE_INTEL_PATH),)
+	MKLROOT = $(USE_INTEL_PATH)/mkl
+endif
+
+CFLAGS  = -fpic -O2 -I/opt/intel/mkl/include -c -Wall -Werror  -DMKL_ILP64 -m64  -std=c++11 
+LDFLAGS =  -Wl,--start-group -L${MKLROOT}/../compiler/lib/intel64 ${MKLROOT}/lib/intel64/libmkl_intel_ilp64.a ${MKLROOT}/lib/intel64/libmkl_intel_thread.a ${MKLROOT}/lib/intel64/libmkl_core.a -Wl,--end-group -liomp5 -lpthread -lm -ldl
+
+default: libsparse_matrix.so
+
+libsparse_matrix.so:  sparse_matrix.o
+	$(CC) -shared -o libsparse_matrix.so sparse_matrix.o $(LDFLAGS) 
+
+sparse_matrix.o:  sparse_matrix.cc sparse_matrix.h
+	$(CC) $(CFLAGS) sparse_matrix.cc 
+
+clean:
+	$(RM) libsparse_matrix.so *.o *~

3rdparty/sparse-matrix/sparse_matrix.cc

@@ -0,0 +1,45 @@
+#include <iostream>
+#include <string>
+#include <fstream>
+#include <mkl_spblas.h>
+#include "sparse_matrix.h"
+
+
+
+bool mkl_DotCsrDnsDns(SP_INT64* rows_start, SP_INT64* col_indx,
+	float* values, float* X, float* y,
+	int rows, int cols, int X_columns)
+{
+
+	sparse_index_base_t indexing = SPARSE_INDEX_BASE_ZERO;
+	sparse_status_t status;
+	sparse_matrix_t A = NULL;
+	sparse_layout_t layout = SPARSE_LAYOUT_ROW_MAJOR;
+	float one, zero;
+	one = (float)1.0;
+	zero = (float)0.0;
+
+	MKL_INT* rows_end = rows_start + 1;
+	status = mkl_sparse_s_create_csr(&A, indexing, rows, cols, rows_start, rows_end, col_indx, values);
+
+  if (status != SPARSE_STATUS_SUCCESS)
+  {
+    std::cout << "mkl_sparse_s_create_csr status :" << status << std::endl;
+    return false;
+  }
+	sparse_operation_t operation = SPARSE_OPERATION_NON_TRANSPOSE;
+	struct matrix_descr descrA;
+	descrA.type = SPARSE_MATRIX_TYPE_GENERAL;
+
+	status = mkl_sparse_s_mm(operation, one, A, descrA, layout, X, X_columns, X_columns, zero, y, X_columns);
+  if (status != SPARSE_STATUS_SUCCESS)
+  {
+    std::cout << "mkl_sparse_s_create_csr status :" << status << std::endl;
+    return false;
+  }
+
+	mkl_sparse_destroy(A);
+
+	return true;
+
+}

3rdparty/sparse-matrix/sparse_matrix.h

@@ -0,0 +1,48 @@
+#ifndef MXNET_OPERATOR_SPARSE_MATRIX_INL_H_
+#define MXNET_OPERATOR_SPARSE_MATRIX_INL_H_
+
+
+#if (!defined(__INTEL_COMPILER)) & defined(_MSC_VER)
+#define SP_INT64 __int64
+#define SP_UINT64 unsigned __int64
+#else
+#define SP_INT64 long long int
+#define SP_UINT64 unsigned long long int
+#endif
+
+
+#if defined _WIN32 || defined __CYGWIN__
+  #ifdef BUILDING_DLL
+    #ifdef __GNUC__
+      #define SPM_API_PUBLIC __attribute__ ((dllexport))
+    #else
+      #define SPM_API_PUBLIC __declspec(dllexport) // Note: actually gcc seems to also supports this syntax.
+    #endif
+  #else
+    #ifdef __GNUC__
+      #define SPM_API_PUBLIC __attribute__ ((dllimport))
+    #else
+      #define SPM_API_PUBLIC __declspec(dllimport) // Note: actually gcc seems to also supports this syntax.
+    #endif
+  #endif
+  #define SPM_API_LOCAL
+#else
+  #if __GNUC__ >= 4
+    #define SPM_API_PUBLIC __attribute__ ((visibility ("default")))
+    #define SPM_API_LOCAL  __attribute__ ((visibility ("hidden")))
+  #else
+    #define SPM_API_PUBLIC
+    #define SPM_API_LOCAL
+  #endif
+#endif
+
+
+
+extern "C"
+{
+	extern SPM_API_PUBLIC bool mkl_DotCsrDnsDns(SP_INT64* rows_start, SP_INT64* col_indx,
+		float* values, float* X, float* y, int rows, int cols, int X_columns);
+
+}
+
+#endif //MXNET_OPERATOR_SPARSE_MATRIX_INL_H_

CMakeLists.txt

@@ -32,7 +32,7 @@ mxnet_option(USE_MKL_IF_AVAILABLE "Use MKL if found" ON)
 mxnet_option(USE_MKLML_MKL        "Use MKLDNN variant of MKL (if MKL found)" ON IF USE_MKL_IF_AVAILABLE AND (NOT APPLE))
 mxnet_option(USE_MKLDNN           "Use MKLDNN variant of MKL (if MKL found)" ON IF USE_MKL_IF_AVAILABLE AND (NOT APPLE) AND (NOT MSVC) AND (CMAKE_HOST_SYSTEM_PROCESSOR STREQUAL "x86_64") AND (NOT CMAKE_CROSSCOMPILING))
 mxnet_option(USE_OPERATOR_TUNING  "Enable auto-tuning of operators" ON IF NOT MSVC)
-mxnet_option(USE_GPERFTOOLS       "Build with GPerfTools support (if found)" ON)
+mxnet_option(USE_GPERFTOOLS       "Build with GPerfTools support" OFF)
 mxnet_option(USE_JEMALLOC         "Build with Jemalloc support"   ON)
 mxnet_option(USE_PROFILER         "Build with Profiler support"   ON)
 mxnet_option(USE_DIST_KVSTORE     "Build with DIST_KVSTORE support" OFF)
@@ -255,6 +255,7 @@ if(USE_MKLDNN)
   add_subdirectory(3rdparty/mkldnn)
 
   include_directories(3rdparty/mkldnn/include)
+  include_directories(${PROJECT_BINARY_DIR}/3rdparty/mkldnn/include)
   add_definitions(-DUSE_MKL=1)
   add_definitions(-DCUB_MKL=1)
   add_definitions(-DMXNET_USE_MKLDNN=1)
@@ -327,7 +328,7 @@ endforeach()
 
 include_directories("include")
 include_directories("3rdparty/mshadow")
-include_directories("3rdparty/cub")
+include_directories("3rdparty/nvidia_cub")
 include_directories("3rdparty/tvm/nnvm/include")
 include_directories("3rdparty/tvm/include")
 include_directories("3rdparty/dmlc-core/include")
@@ -511,7 +512,7 @@ list(APPEND CUDA ${MSHADOW_CUDASOURCE})
 FILE(GLOB_RECURSE GROUP_SOURCE "src/*.cc" "3rdparty/tvm/nnvm/*.cc" "plugin/*.cc")
 FILE(GLOB_RECURSE GROUP_Include "src/*.h" "3rdparty/tvm/nnvm/*.h" "3rdparty/mshadow/mshadow/*.h" "plugin/*.h")
 FILE(GLOB_RECURSE GROUP_CUDA "src/*.cu" "src/*.cuh" "3rdparty/mshadow/mshadow/*.cuh" "plugin/*.cu"
-  "plugin/*.cuh" "3rdparty/cub/cub/*.cuh")
+  "plugin/*.cuh" "3rdparty/nvidia_cub/cub/*.cuh")
 assign_source_group("Source" ${GROUP_SOURCE})
 assign_source_group("Include" ${GROUP_Include})
 assign_source_group("CUDA" ${GROUP_CUDA})

CONTRIBUTORS.md

@@ -65,6 +65,10 @@ The committers are the granted write access to the project.
  - Marco is the creator of the current MXNet CI.
 * [Carin Meier](https://github.com/gigasquid)
   - Carin created and is the current maintainer for the Clojure interface.
+* [Patric Zhao](https://github.com/pengzhao-intel)
+  - Patric is a parallel computing expert and a major contributor to the MXNet MKL-DNN backend.
+* [Tao Lv](https://github.com/TaoLv)
+  - Tao is a major contributor to the MXNet MKL-DNN backend and performance on CPU.
 
 
 ### Become a Committer
@@ -211,10 +215,27 @@ List of Contributors
 * [Harsh Patel](https://github.com/harshp8l)
 * [Xiao Wang](https://github.com/BeyonderXX)
 * [Piyush Ghai](https://github.com/piyushghai)
+* [Dang Trung Kien](https://github.com/kiendang)
 * [Zach Boldyga](https://github.com/zboldyga)
 * [Gordon Reid](https://github.com/gordon1992)
-
 * [Ming Yang](http://ufoym.com)
+* [Satya Krishna Gorti](https://github.com/satyakrishnagorti)
+* [Neo Chien](https://github.com/cchung100m)
+* [Wujie Zhou](https://github.com/eureka7mt)
+* [Ciyong Chen](https://github.com/ciyongch)
+* [Hao Li](https://github.com/lihaofd)
+* [Jin Huang](https://github.com/jinhuang415)
+* [Luobao Zou](https://github.com/luobao-intel)
+* [Pengxin Yuan](https://github.com/pengxin99)
+* [Rong Zhang](https://github.com/rongzha1/)
+* [Shu Zhang](https://github.com/Sherry-Zhang)
+* [Shufan Wu](https://github.com/juliusshufan)
+* [Wenting Jiang](https://github.com/wentingj)
+* [Xiaotao Chen](https://github.com/XiaotaoChen)
+* [Xinyu Chen](https://github.com/xinyu-intel)
+* [Zhennan Qin](https://github.com/ZhennanQin)
+* [Zhiyuan Huang](https://github.com/huangzhiyuan)
+* [Zak Jost](https://github.com/zjost)
 
 Label Bot
 ---------

LICENSE

@@ -315,10 +315,10 @@
          Copyright 2005-2008, Google Inc.
     3. Moderngpu - For details, see, 3rdparty/ctc_include/contrib/moderngpu/LICENSE
          Copyright (c) 2013, NVIDIA CORPORATION.  All rights reserved.
-    4. CUB Library - For details, see, 3rdparty/cub/LICENSE.TXT
+    4. CUB Library - For details, see, 3rdparty/nvidia_cub/LICENSE.TXT
          Copyright (c) 2010-2011, Duane Merrill.  All rights reserved.
          Copyright (c) 2011-2016, NVIDIA CORPORATION.  All rights reserved.
-    5. CUB mersenne.h - For details, see 3rdparty/cub/test/mersenne.h
+    5. CUB mersenne.h - For details, see 3rdparty/nvidia_cub/test/mersenne.h
          Copyright (C) 1997 - 2002, Makoto Matsumoto and Takuji Nishimura,
     6. Googlemock - For details, see, 3rdparty/googletest/googlemock/LICENSE
          Copyright 2006-2015, Google Inc.