Merge Release_1.1.2

CMakeLists.txt

@@ -10,8 +10,8 @@ endif()
 project(Caffe C CXX)
 
 # ---[ Caffe version
-set(CAFFE_TARGET_VERSION "1.1.0" CACHE STRING "Caffe logical version")
-set(CAFFE_TARGET_SOVERSION "1.1.0" CACHE STRING "Caffe soname version")
+set(CAFFE_TARGET_VERSION "1.1.2" CACHE STRING "Caffe logical version")
+set(CAFFE_TARGET_SOVERSION "1.1.2" CACHE STRING "Caffe soname version")
 add_definitions(-DCAFFE_VERSION=${CAFFE_TARGET_VERSION})
 
 # ---[ Using cmake scripts and modules
@@ -47,6 +47,7 @@ caffe_option(PERFORMANCE_MONITORING "Build Caffe with PERFORMANCE_MONITORING " O
 caffe_option(CODE_COVERAGE "Build with code coverage instrumentation" OFF)
 caffe_option(CO_SIM "Build readonly cpu blob data/diff pycaffe interface" OFF)
 caffe_option(USE_MLSL "Build Caffe with Intel(r) Machine Learning Scaling Library for multinode training" ON)
+caffe_option(ICC_STATIC_BUILD "Build Intel Caffe static link with Intel compiler" OFF)
 
 # ---[ Dependencies
 include(cmake/MKLDNN.cmake)
@@ -93,7 +94,11 @@ if(UNIX OR APPLE)
     # shows that it generates correct PIE anyway if only -pie is provided.
     set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fstack-protector -wd2196")
     # Auto enable SGD Fusion if use intel compiler
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -DENABLE_SGD_FUSION")
+    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -DENABLE_SGD_FUSION -DENABLE_NMS_OPTIMIZATION")
+    # ICC static build
+    if(ICC_STATIC_BUILD)
+      set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} -static-intel")
+    endif()
   endif()
 
   # Generic flags.

Makefile

@@ -102,7 +102,7 @@ LIB_BUILD_DIR := $(BUILD_DIR)/lib
 STATIC_NAME := $(LIB_BUILD_DIR)/lib$(LIBRARY_NAME).a
 DYNAMIC_VERSION_MAJOR 		:= 1
 DYNAMIC_VERSION_MINOR 		:= 1
-DYNAMIC_VERSION_REVISION 	:= 0
+DYNAMIC_VERSION_REVISION 	:= 2
 DYNAMIC_NAME_SHORT := lib$(LIBRARY_NAME).so
 #DYNAMIC_SONAME_SHORT := $(DYNAMIC_NAME_SHORT).$(DYNAMIC_VERSION_MAJOR)
 DYNAMIC_VERSIONED_NAME_SHORT := $(DYNAMIC_NAME_SHORT).$(DYNAMIC_VERSION_MAJOR).$(DYNAMIC_VERSION_MINOR).$(DYNAMIC_VERSION_REVISION)
@@ -389,7 +389,11 @@ endif
 ifneq (,$(findstring icpc,$(CXX)))
 	CXX_HARDENING_FLAGS += -fstack-protector -wd2196
 	#Enable SGD FUSION if use intel compiler
-	COMMON_FLAGS += -DENABLE_SGD_FUSION
+	COMMON_FLAGS += -DENABLE_SGD_FUSION -DENABLE_NMS_OPTIMIZATION
+	#Intel compiler static build flag
+	ifeq ($(ICC_STATIC_BUILD), 1)
+		LDFLAGS += -static-intel
+	endif
 
 else ifneq (,$(findstring clang++,$(CXX)))
 	CXX_HARDENING_FLAGS += -fPIE -fstack-protector
@@ -447,7 +451,7 @@ ifeq ($(DEBUG), 1)
 	COMMON_FLAGS += -DDEBUG -g -O0
 	NVCCFLAGS += -G
 else ifneq (,$(findstring icpc,$(CXX)))
-	COMMON_FLAGS += -DNDEBUG -O3 -xCORE-AVX2 -no-prec-div -fp-model fast=2
+	COMMON_FLAGS += -DNDEBUG -O3 -xHost -no-prec-div -fp-model fast=2
 else
 	COMMON_FLAGS += -DNDEBUG -O3
 endif

Makefile.config.example

@@ -71,6 +71,8 @@ USE_MKLDNN_AS_DEFAULT_ENGINE := 1
 # To customize your choice of compiler, uncomment and set the following.
 # N.B. the default for Linux is g++ and the default for OSX is clang++
 # CUSTOM_CXX := g++
+# If you use Intel compiler, you can choose uncomment following to enable static build
+# ICC_STATIC_BUILD := 1
 
 # If you use Intel compiler define a path to newer boost if not used
 # already. 
@@ -85,6 +87,9 @@ DISABLE_CONV_RELU_FUSION:= 0
 # Use Bn + ReLU fusion to boost inference
 DISABLE_BN_RELU_FUSION := 0
 
+# Use Conv + Concat  fusion to boost inference.
+ENABLE_CONCAT_FUSION := 0
+
 # Use Conv + Eltwise + Relu layer fusion to boost inference.
 DISABLE_CONV_SUM_FUSION := 0
 

Makefile.dlcp

@@ -1,6 +1,6 @@
 CAFFE_ROOTDIR := $(shell pwd)
 DLCP_IDEEPDIR := $(CAFFE_ROOTDIR)/external/ideep
-DLCP_ROOTDIR := $(DLCP_IDEEPDIR)/ideep/dlcp
+DLCP_ROOTDIR := $(DLCP_IDEEPDIR)/dlcp
 DLCP_INCLDIR := $(DLCP_ROOTDIR)/include
 DLCP_LIBDIR := $(DLCP_ROOTDIR)/lib
 IDEEP_GITHUB := https://github.com/intel/ideep.git
@@ -9,7 +9,7 @@ DLCP_CXX := $(CXX)
 ifeq ("$(wildcard $(DLCP_INCLDIR)/dl_compression.h)", "")
 dlcp_download:
 	rm -rf $(DLCP_IDEEPDIR)
-	git clone $(IDEEP_GITHUB) -b dlcp $(DLCP_IDEEPDIR)
+	git clone $(IDEEP_GITHUB) -b ideep4py $(DLCP_IDEEPDIR)
 
 dlcp_build: dlcp_download
 	make -C $(DLCP_ROOTDIR) -j$(shell cat /proc/cpuinfo |grep 'processor'|wc -l)

Makefile.mkldnn

@@ -22,6 +22,12 @@ ifneq (,$(findstring ccache,$(CC)))
 endif
 
 MKLDNN_GITHUB := https://github.com/01org/mkl-dnn.git
+# Enable MKLDNN intel compiler static build
+ifneq (,$(findstring icpc,$(MKLDNN_CXX)))
+	ifeq ($(ICC_STATIC_BUILD), 1)
+		MKLDNN_CMAKE_FLAGS += $(MKLDNN_SRCDIR) -DCMAKE_INSTALL_PREFIX=$(MKLDNN_INSTALLDIR) -DMKLROOT=${MKL_ROOTDIR} -B$(MKLDNN_BUILDDIR) -DCMAKE_CXX_COMPILER="$(MKLDNN_CXX)" -DCMAKE_C_COMPILER="$(MKLDNN_CC)" -DCMAKE_SHARED_LINKER_FLAGS="$(CMAKE_SHARED_LINKER_FLAGS) -static-intel"
+	endif
+endif
 MKLDNN_CMAKE_FLAGS += $(MKLDNN_SRCDIR) -DCMAKE_INSTALL_PREFIX=$(MKLDNN_INSTALLDIR) -DMKLROOT=${MKL_ROOTDIR} -B$(MKLDNN_BUILDDIR) -DCMAKE_CXX_COMPILER="$(MKLDNN_CXX)" -DCMAKE_C_COMPILER="$(MKLDNN_CC)"
 
 # Disable MKLDNN download by customized setting

cmake/DLCP.cmake

@@ -5,20 +5,20 @@ function(Download_DLCP)
   elseif(HAS_ICPC)
     set(DLCP_CXX "icpc" PARENT_SCOPE)
   else()
-    message(WARNING "weight grad compression is disabled because intel compiler is not found")
+    message("weight grad compression is disabled because intel compiler is not found")
     return()  
   endif()  
 
   set(EXTERNAL_DIR ${CMAKE_SOURCE_DIR}/external)
   set(DLCP_IDEEPDIR ${EXTERNAL_DIR}/ideep)
-  set(DLCP_ROOTDIR ${DLCP_IDEEPDIR}/ideep/dlcp)
+  set(DLCP_ROOTDIR ${DLCP_IDEEPDIR}/dlcp)
   set(DLCP_INCLDIR "${DLCP_ROOTDIR}/include" PARENT_SCOPE)
   set(DLCP_LIBDIR ${DLCP_ROOTDIR}/lib PARENT_SCOPE)
 
   # Download dl compression lib source code if it doesn't exist 
   if (NOT EXISTS ${DLCP_INCLDIR}/dl_compression.h)
     execute_process(COMMAND rm -rf ${DLCP_IDEEPDIR})
-    execute_process(COMMAND git clone https://github.com/intel/ideep.git -b dlcp ${DLCP_IDEEPDIR})
+    execute_process(COMMAND git clone https://github.com/intel/ideep.git -b ideep4py ${DLCP_IDEEPDIR})
   endif()
 
   add_custom_target(DLCP_Build ALL

cmake/Dependencies.cmake

@@ -128,6 +128,7 @@ if(USE_MLSL)
   add_definitions("-DFOUNDED_MLSL_ROOT=${MLSL_ROOT}")
   include_directories(SYSTEM "${MLSL_ROOT}/intel64/include")
   link_directories(SYSTEM "${MLSL_ROOT}/intel64/lib")
+  set(CMAKE_INSTALL_RPATH ${CMAKE_INSTALL_RPATH}:${MLSL_ROOT}/intel64/lib/)
   list(APPEND Caffe_LINKER_LIBS mlsl)
 
   if(CAFFE_MLSL_SHUFFLE)
@@ -151,6 +152,7 @@ if(USE_MLSL)
           message(STATUS "Found DLCP: ${DLCPROOT_DIR}")
           set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -DENABLE_WEIGHT_GRAD_COMPRESSION")
           list(APPEND Caffe_LINKER_LIBS "${DLCPROOT_LIB_DIR}/libdlcomp.so")
+          set(CMAKE_INSTALL_RPATH ${CMAKE_INSTALL_RPATH}:${DLCP_LIBDIR})
           include_directories(SYSTEM ${DLCP_INCLUDE_DIR})
         else()
           message(STATUS "DLCP not found. DLCP_INCLUDE_DIR = ${DLCPROOT_INCLUDE_DIR} DLCP_LIB_DIR = ${DLCPROOT_LIB_DIR}")
@@ -162,6 +164,7 @@ if(USE_MLSL)
       if(DLCP_CXX)
         set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -DENABLE_WEIGHT_GRAD_COMPRESSION")
         list(APPEND Caffe_LINKER_LIBS "${DLCP_LIBDIR}/libdlcomp.so")
+        set(CMAKE_INSTALL_RPATH ${CMAKE_INSTALL_RPATH}:${DLCP_LIBDIR})
         include_directories(SYSTEM ${DLCP_INCLDIR})
       endif()
     endif()

cmake/MKLDNN.cmake

@@ -5,6 +5,12 @@ function(Download_MKLDNN)
   set(MKLDNN_SOURCE_DIR ${MKLDNN_DIR}/src)
   set(MKLDNN_BUILD_DIR ${MKLDNN_DIR}/build)
   set(MKLDNN_INSTALL_DIR ${MKLDNN_DIR}/install CACHE PATH "Installation path of MKLDNN")
+  # Enable MKLDNN intel compiler static build
+  if(${CMAKE_CXX_COMPILER_ID} STREQUAL "Intel" )
+    if(ICC_STATIC_BUILD)
+      set(CMAKE_SHARED_LINKER_FLAGS "${CMAKE_SHARED_LINKER_FLAGS} -static-intel")
+    endif()
+  endif()
   execute_process(COMMAND cat mkldnn.commit
   		  WORKING_DIRECTORY ${CMAKE_SOURCE_DIR}
 		  OUTPUT_VARIABLE MKLDNN_COMMIT)
@@ -18,13 +24,13 @@ function(Download_MKLDNN)
 
   ExternalProject_add(MKLDNN_Build
                       SOURCE_DIR ${MKLDNN_SOURCE_DIR}
-                      CMAKE_ARGS -DCMAKE_BUILD_TYPE=${CMAKE_BUILD_TYPE} -DCMAKE_INSTALL_PREFIX=${MKLDNN_INSTALL_DIR} -DMKLROOT=${MKL_ROOT_DIR}
+                      CMAKE_ARGS -DCMAKE_BUILD_TYPE=${CMAKE_BUILD_TYPE} -DCMAKE_INSTALL_PREFIX=${MKLDNN_INSTALL_DIR} -DMKLROOT=${MKL_ROOT_DIR} -DCMAKE_SHARED_LINKER_FLAGS=${CMAKE_SHARED_LINKER_FLAGS}
 #--Download step
                       GIT_REPOSITORY https://github.com/01org/mkl-dnn.git
-		      GIT_TAG ${MKLDNN_COMMIT}
+                      GIT_TAG ${MKLDNN_COMMIT}
 #--Build step
                       BINARY_DIR ${MKLDNN_BUILD_DIR}
-                      BUILD_COMMAND cmake ${MKLDNN_SOURCE_DIR}
+                      BUILD_COMMAND cmake ${MKLDNN_SOURCE_DIR} 
 #--Install step
                       INSTALL_DIR ${MKLDNN_INSTALL_DIR}
                       INSTALL_COMMAND make install -j${NCORE}

cmake/Misc.cmake

@@ -21,6 +21,11 @@ if(DISABLE_BN_RELU_FUSION)
     add_definitions("-DDISABLE_BN_RELU_FUSION")
 endif()
 
+if(ENABLE_CONCAT_FUSION)
+	message(STATUS "conv/concat fusion is enabled!")
+	add_definitions("-DENABLE_CONCAT_FUSION")
+endif()
+
 if(DISABLE_CONV_SUM_FUSION)
 	message(STATUS "conv/eltwise/relu fusion is disabled!")
 	add_definitions("-DDISABLE_CONV_SUM_FUSION")
@@ -71,10 +76,3 @@ if(UNIX OR APPLE)
                                      COMMENT "Adding symlink: <caffe_root>/build -> ${PROJECT_BINARY_DIR}" )
 endif()
 
-# ---[ Set debug postfix
-set(Caffe_DEBUG_POSTFIX "-d")
-
-set(Caffe_POSTFIX "")
-if(CMAKE_BUILD_TYPE MATCHES "Debug")
-  set(Caffe_POSTFIX ${Caffe_DEBUG_POSTFIX})
-endif()

cmake/Summary.cmake

@@ -105,6 +105,9 @@ function(caffe_print_configuration_summary)
   caffe_status("  Git                     :   ${Caffe_GIT_VERSION}")
   caffe_status("  System                  :   ${CMAKE_SYSTEM_NAME}")
   caffe_status("  C++ compiler            :   ${CMAKE_CXX_COMPILER}")
+  if(CMAKE_CXX_COMPILER_ID STREQUAL "Intel")
+    caffe_status("  ICC_STATIC_BUILD        :   ${ICC_STATIC_BUILD}")
+  endif()
   caffe_status("  Release CXX flags       :   ${__flags_rel}")
   caffe_status("  Debug CXX flags         :   ${__flags_deb}")
   caffe_status("  Build type              :   ${CMAKE_BUILD_TYPE}")

cmake/Targets.cmake

@@ -108,7 +108,6 @@ endfunction()
 #   caffe_default_properties(<target>)
 function(caffe_default_properties target)
   set_target_properties(${target} PROPERTIES
-    DEBUG_POSTFIX ${Caffe_DEBUG_POSTFIX}
     ARCHIVE_OUTPUT_DIRECTORY "${PROJECT_BINARY_DIR}/lib"
     LIBRARY_OUTPUT_DIRECTORY "${PROJECT_BINARY_DIR}/lib"
     RUNTIME_OUTPUT_DIRECTORY "${PROJECT_BINARY_DIR}/bin")

docker/README.md

@@ -17,7 +17,7 @@ docker run -ti caffe:cpu caffe --version
 ```
 which should show a message like:
 ```
-caffe version 1.1.0
+caffe version 1.1.2
 ```
 
 One can also build and run the Caffe tests in the image using:

docker/standalone/cpu-centos/Dockerfile

@@ -4,20 +4,28 @@ MAINTAINER [email protected]
 #ENV http_proxy proxy:port
 #ENV https_proxy proxy:port
 
-RUN rpm -iUvh http://download.fedoraproject.org/pub/epel/7/x86_64/e/epel-release-7-10.noarch.rpm
-
-RUN yum upgrade -y
-
-RUN yum install -y \
+RUN rpm -iUvh https://dl.fedoraproject.org/pub/epel/7/x86_64/Packages/e/epel-release-7-11.noarch.rpm && \
+    yum upgrade -y && \
+    yum install -y \
+        bzip2 \
+        numactl \
         redhat-rpm-config \
         tar \
         findutils \
         make \
         gcc-c++ \
         cmake \
         git \
+        vim \
         wget \
         ssh \
+        openssh.x86_64 \
+        openssh-server.x86_64 \
+        openssh-clients.x86_64 \
+        initscripts  \
+        net-tools \
+        ufw \
+        iptables \
         atlas-devel \
         boost-devel \
         gflags-devel \
@@ -36,24 +44,36 @@ RUN yum install -y \
         python-pip \
         python-scipy \
         gcc-gfortran \
-        libjpeg-turbo-devel
+        libjpeg-turbo-devel && \
+    yum clean all
 
-RUN yum clean all 
-ENV CAFFE_ROOT=/opt/caffe
-WORKDIR $CAFFE_ROOT
+# Install conda and Intel Caffe conda package
+WORKDIR /root/
+RUN wget -c https://repo.continuum.io/miniconda/Miniconda2-latest-Linux-x86_64.sh && \ 
+    bash Miniconda2-latest-Linux-x86_64.sh -b && \ 
+    ./miniconda2/bin/conda config --add channels intel && \ 
+    ./miniconda2/bin/conda install -c intel caffe && \ 
+    rm -rf /root/miniconda2/pkgs/* && \
+    rm ~/Miniconda2-latest-Linux-x86_64.sh -f && \
+    echo "export PATH=/root/miniconda2/bin:$PATH" >> /root/.bashrc
+WORKDIR /root/miniconda2/caffe
 
-# FIXME: clone a specific git tag and use ARG instead of ENV once DockerHub supports this.
-ENV CLONE_TAG=master
+# If you want to deploy container in HOST network mode, Install SSH service and config it to Non-standard Port. Or you needn’t below rows.
+RUN mkdir /var/run/sshd && \
+    echo "root:intelcaffe@123" | chpasswd
+RUN sed -i 's/#PermitRootLogin prohibit-password/PermitRootLogin yes/;s/#Port 22/Port 10086/' /etc/ssh/sshd_config
 
-RUN git clone -b ${CLONE_TAG} --depth 1 https://github.com/intel/caffe.git . && \
-    for req in $(cat python/requirements.txt) pydot; do pip --no-cache-dir install $req; done && \
-    mkdir build && cd build && \
-    cmake -DCPU_ONLY=1 -DCMAKE_BUILD_TYPE=Release .. && \
-    make all -j"$(nproc)"
+# SSH login fix. Otherwise user is kicked off after login
+RUN sed 's@session\s*required\s*pam_loginuid.so@session optional pam_loginuid.so@g' -i /etc/pam.d/sshd
 
-ENV PYCAFFE_ROOT $CAFFE_ROOT/python
-ENV PYTHONPATH $PYCAFFE_ROOT:$PYTHONPATH
-ENV PATH $CAFFE_ROOT/build/tools:$PYCAFFE_ROOT:$PATH
-RUN echo "$CAFFE_ROOT/build/lib" >> /etc/ld.so.conf.d/caffe.conf && ldconfig
+ENV NOTVISIBLE "in users profile"
+RUN echo "export VISIBLE=now" >> /etc/profile
 
-WORKDIR /workspace
+EXPOSE 10086
+RUN ssh-keygen -t rsa -A && \
+    mkdir ~/.ssh && \
+    touch ~/.ssh/config && \
+    echo "Host *" > ~/.ssh/config && \
+    echo "Port 10086" >> ~/.ssh/config && \
+    echo "StrictHostKeyChecking no" >> ~/.ssh/config
+CMD ["/usr/sbin/sshd", "-D"]

docker/standalone/cpu-ubuntu/Dockerfile

@@ -4,8 +4,9 @@ MAINTAINER [email protected]
 #ENV http_proxy proxy:port
 #ENV https_proxy proxy:port
 
-RUN apt-get update && apt-get install -y --no-install-recommends \
-	cpio \
+RUN apt-get update && \
+    apt-get install -y --no-install-recommends \
+	    cpio \
         build-essential \
         cmake \
         git \
@@ -38,31 +39,20 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
         python-scipy && \
     rm -rf /var/lib/apt/lists/*
 
-ENV CAFFE_ROOT=/opt/caffe
-WORKDIR $CAFFE_ROOT
+# Install conda and Intel Caffe conda package
+WORKDIR /root/
+RUN wget -c https://repo.continuum.io/miniconda/Miniconda2-latest-Linux-x86_64.sh && \
+    bash Miniconda2-latest-Linux-x86_64.sh -b && \
+    ./miniconda2/bin/conda config --add channels intel && \
+    ./miniconda2/bin/conda install -c intel caffe && \
+    rm -rf /root/miniconda2/pkgs/* && \
+    rm ~/Miniconda2-latest-Linux-x86_64.sh -f && \
+    echo "export PATH=/root/miniconda2/bin:$PATH" >> /root/.bashrc
+WORKDIR /root/miniconda2/caffe
 
-# FIXME: clone a specific git tag and use ARG instead of ENV once DockerHub supports this.
-ENV CLONE_TAG=master
-
-RUN pip install --upgrade pip
-
-RUN git clone -b ${CLONE_TAG} --depth 1 https://github.com/intel/caffe.git . && \
-    for req in $(cat python/requirements.txt) pydot; do pip install $req; done && \
-    mkdir build && cd build && \
-    cmake -DCPU_ONLY=1 -DUSE_MLSL=1 -DCMAKE_BUILD_TYPE=Release .. && \
-    make all -j"$(nproc)"
-
-ENV PYCAFFE_ROOT $CAFFE_ROOT/python
-ENV PYTHONPATH $PYCAFFE_ROOT:$PYTHONPATH
-ENV PATH $CAFFE_ROOT/build/tools:$PYCAFFE_ROOT:$PATH
-RUN echo "$CAFFE_ROOT/build/lib" >> /etc/ld.so.conf.d/caffe.conf && ldconfig
-
-WORKDIR /opt/caffe
-
-RUN mkdir /var/run/sshd
-RUN echo 'root:123456' | chpasswd
-RUN sed -i 's/PermitRootLogin prohibit-password/PermitRootLogin yes/' /etc/ssh/sshd_config
-RUN sed -i 's/Port 22/Port 10010/' /etc/ssh/sshd_config
+RUN mkdir /var/run/sshd && \
+    echo 'root:intelcaffe@123' | chpasswd
+RUN sed -i 's/PermitRootLogin prohibit-password/PermitRootLogin yes/;s/Port 22/Port 10010/' /etc/ssh/sshd_config
 
 # SSH login fix. Otherwise user is kicked off after login
 RUN sed 's@session\s*required\s*pam_loginuid.so@session optional pam_loginuid.so@g' -i /etc/pam.d/sshd
@@ -71,5 +61,10 @@ ENV NOTVISIBLE "in users profile"
 RUN echo "export VISIBLE=now" >> /etc/profile
 
 EXPOSE 10010
-
+RUN ssh-keygen -t rsa -A && \
+    mkdir ~/.ssh && \
+    touch ~/.ssh/config && \
+    echo "Host *" > ~/.ssh/config && \
+    echo "Port 10010" >> ~/.ssh/config && \
+    echo "StrictHostKeyChecking no" >> ~/.ssh/config
 CMD ["/usr/sbin/sshd", "-D"]