Merge Release_1.1.2

Author: daisyden

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}")