BVLC · shelhamer · Jul 17, 2014 · Jun 30, 2014 · Jun 30, 2014 · Jun 30, 2014
diff --git a/.travis.yml b/.travis.yml
@@ -26,12 +26,18 @@ install:
 
 before_script:
   - mv Makefile.config.example Makefile.config
-  - export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda/lib64:/usr/local/lib
+  - export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/lib
   - export NUM_THREADS=4
 
 script:
+  # CPU-GPU: build only.
+  - export CPU_ONLY=0
+  - make --keep-going --jobs=$NUM_THREADS all
+  - make clean
+  # CPU-only: comprehensive.
+  - export CPU_ONLY=1
   - make --keep-going --jobs=$NUM_THREADS all test warn lint
-  - make runtestnogpu
+  - make runtest
   - make --jobs=$NUM_THREADS all
   - make --jobs=$NUM_THREADS test
   - make --jobs=$NUM_THREADS warn

diff --git a/Makefile b/Makefile
@@ -149,11 +149,13 @@ NONEMPTY_WARN_REPORT := $(BUILD_DIR)/$(WARNS_EXT)
 CUDA_INCLUDE_DIR := $(CUDA_DIR)/include
 CUDA_LIB_DIR := $(CUDA_DIR)/lib64 $(CUDA_DIR)/lib
 
-INCLUDE_DIRS += $(BUILD_INCLUDE_DIR)
-INCLUDE_DIRS += ./src ./include $(CUDA_INCLUDE_DIR)
-LIBRARY_DIRS += $(CUDA_LIB_DIR)
-LIBRARIES := cudart cublas curand \
-	pthread \
+INCLUDE_DIRS += $(BUILD_INCLUDE_DIR) ./src ./include
+ifneq ($(CPU_ONLY), 1)
+	INCLUDE_DIRS += $(CUDA_INCLUDE_DIR)
+	LIBRARY_DIRS += $(CUDA_LIB_DIR)
+	LIBRARIES := cudart cublas curand
+endif
+LIBRARIES += pthread \
 	glog protobuf leveldb snappy \
 	lmdb \
 	boost_system \
@@ -202,6 +204,16 @@ ifeq ($(LINUX), 1)
 	endif
 endif
 
+# CPU-only configuration
+ifeq ($(CPU_ONLY), 1)
+	OBJS := $(PROTO_OBJS) $(CXX_OBJS)
+	TEST_OBJS := $(TEST_CXX_OBJS)
+	TEST_BINS := $(TEST_CXX_BINS)
+	ALL_WARNS := $(ALL_CXX_WARNS)
+	TEST_FILTER := --gtest_filter="-*GPU*"
+	COMMON_FLAGS += -DCPU_ONLY
+endif
+
 # OS X:
 # clang++ instead of g++
 # libstdc++ instead of libc++ for CUDA compatibility on 10.9
@@ -222,9 +234,9 @@ endif
 
 # Debugging
 ifeq ($(DEBUG), 1)
-	COMMON_FLAGS := -DDEBUG -g -O0
+	COMMON_FLAGS += -DDEBUG -g -O0
 else
-	COMMON_FLAGS := -DNDEBUG -O2
+	COMMON_FLAGS += -DNDEBUG -O2
 endif
 
 # BLAS configuration (default = ATLAS)
@@ -282,7 +294,7 @@ SUPERCLEAN_EXTS := .so .a .o .bin .testbin .pb.cc .pb.h _pb2.py .cuo
 # Define build targets
 ##############################
 .PHONY: all test clean linecount lint tools examples $(DIST_ALIASES) \
-	py mat py$(PROJECT) mat$(PROJECT) proto runtest runtestnogpu \
+	py mat py$(PROJECT) mat$(PROJECT) proto runtest \
 	superclean supercleanlist supercleanfiles warn
 
 all: $(NAME) $(STATIC_NAME) tools examples
@@ -342,10 +354,7 @@ $(MAT$(PROJECT)_SO): $(MAT$(PROJECT)_SRC) $(STATIC_NAME)
 	@ echo
 
 runtest: $(TEST_ALL_BIN)
-	$(TEST_ALL_BIN) $(TEST_GPUID) --gtest_shuffle
-
-runtestnogpu: $(TEST_ALL_BIN)
-	$(TEST_ALL_BIN) --gtest_shuffle --gtest_filter="-*GPU*:*/2.*:*/3.*"
+	$(TEST_ALL_BIN) $(TEST_GPUID) --gtest_shuffle $(TEST_FILTER)
 
 warn: $(EMPTY_WARN_REPORT)
 

diff --git a/Makefile.config.example b/Makefile.config.example
@@ -1,6 +1,9 @@
 ## Refer to http://caffe.berkeleyvision.org/installation.html
 # Contributions simplifying and improving our build system are welcome!
 
+# CPU-only switch (uncomment to build without GPU support).
+# CPU_ONLY := 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++

diff --git a/include/caffe/common.hpp b/include/caffe/common.hpp
@@ -4,12 +4,10 @@
 #define CAFFE_COMMON_HPP_
 
 #include <boost/shared_ptr.hpp>
-#include <cublas_v2.h>
-#include <cuda.h>
-#include <curand.h>
-#include <driver_types.h>  // cuda driver types
 #include <glog/logging.h>
 
+#include "caffe/util/device_alternate.hpp"
+
 // Disable the copy and assignment operator for a class.
 #define DISABLE_COPY_AND_ASSIGN(classname) \
 private:\
@@ -25,6 +23,8 @@ private:\
 // is executed we will see a fatal log.
 #define NOT_IMPLEMENTED LOG(FATAL) << "Not Implemented Yet"
 
+#ifndef CPU_ONLY
+
 // CUDA: various checks for different function calls.
 #define CUDA_CHECK(condition) \
   /* Code block avoids redefinition of cudaError_t error */ \
@@ -56,6 +56,8 @@ private:\
 // CUDA: check for error after kernel execution and exit loudly if there is one.
 #define CUDA_POST_KERNEL_CHECK CUDA_CHECK(cudaPeekAtLastError())
 
+#endif  // CPU_ONLY
+
 namespace caffe {
 
 // We will use the boost shared_ptr instead of the new C++11 one mainly
@@ -99,10 +101,12 @@ class Caffe {
     }
     return *(Get().random_generator_);
   }
+#ifndef CPU_ONLY
   inline static cublasHandle_t cublas_handle() { return Get().cublas_handle_; }
   inline static curandGenerator_t curand_generator() {
     return Get().curand_generator_;
   }
+#endif
 
   // Returns the mode: running on CPU or GPU.
   inline static Brew mode() { return Get().mode_; }
@@ -125,8 +129,10 @@ class Caffe {
   static void DeviceQuery();
 
  protected:
+#ifndef CPU_ONLY
   cublasHandle_t cublas_handle_;
   curandGenerator_t curand_generator_;
+#endif
   shared_ptr<RNG> random_generator_;
 
   Brew mode_;
@@ -140,6 +146,8 @@ class Caffe {
   DISABLE_COPY_AND_ASSIGN(Caffe);
 };
 
+#ifndef CPU_ONLY
+
 // NVIDIA_CUDA-5.5_Samples/common/inc/helper_cuda.h
 const char* cublasGetErrorString(cublasStatus_t error);
 const char* curandGetErrorString(curandStatus_t error);
@@ -158,6 +166,7 @@ inline int CAFFE_GET_BLOCKS(const int N) {
   return (N + CAFFE_CUDA_NUM_THREADS - 1) / CAFFE_CUDA_NUM_THREADS;
 }
 
+#endif  // CPU_ONLY
 
 }  // namespace caffe
 

diff --git a/include/caffe/layer.hpp b/include/caffe/layer.hpp
@@ -9,6 +9,7 @@
 #include "caffe/blob.hpp"
 #include "caffe/common.hpp"
 #include "caffe/proto/caffe.pb.h"
+#include "caffe/util/device_alternate.hpp"
 
 using std::string;
 using std::vector;

diff --git a/include/caffe/test/test_caffe_main.hpp b/include/caffe/test/test_caffe_main.hpp
@@ -5,7 +5,6 @@
 #ifndef CAFFE_TEST_TEST_CAFFE_MAIN_HPP_
 #define CAFFE_TEST_TEST_CAFFE_MAIN_HPP_
 
-#include <cuda_runtime.h>
 #include <glog/logging.h>
 #include <gtest/gtest.h>
 
@@ -44,6 +43,12 @@ struct DoubleCPU {
   static const Caffe::Brew device = Caffe::CPU;
 };
 
+#ifdef CPU_ONLY
+
+typedef ::testing::Types<FloatCPU, DoubleCPU> TestDtypesAndDevices;
+
+#else
+
 struct FloatGPU {
   typedef float Dtype;
   static const Caffe::Brew device = Caffe::GPU;
@@ -57,6 +62,8 @@ struct DoubleGPU {
 typedef ::testing::Types<FloatCPU, DoubleCPU, FloatGPU, DoubleGPU>
     TestDtypesAndDevices;
 
+#endif
+
 }  // namespace caffe
 
 #endif  // CAFFE_TEST_TEST_CAFFE_MAIN_HPP_
diff --git a/include/caffe/util/benchmark.hpp b/include/caffe/util/benchmark.hpp
@@ -4,7 +4,8 @@
 #define CAFFE_UTIL_BENCHMARK_H_
 
 #include <boost/date_time/posix_time/posix_time.hpp>
-#include <cuda_runtime.h>
+
+#include "caffe/util/device_alternate.hpp"
 
 namespace caffe {
 
@@ -27,8 +28,10 @@ class Timer {
   bool initted_;
   bool running_;
   bool has_run_at_least_once_;
+#ifndef CPU_ONLY
   cudaEvent_t start_gpu_;
   cudaEvent_t stop_gpu_;
+#endif
   boost::posix_time::ptime start_cpu_;
   boost::posix_time::ptime stop_cpu_;
   float elapsed_milliseconds_;

diff --git a/include/caffe/util/device_alternate.hpp b/include/caffe/util/device_alternate.hpp
@@ -0,0 +1,44 @@
+// Copyright 2014 BVLC and contributors.
+
+#ifndef CAFFE_UTIL_DEVICE_ALTERNATE_H_
+#define CAFFE_UTIL_DEVICE_ALTERNATE_H_
+
+#ifdef CPU_ONLY  // CPU-only Caffe.
+
+#include <vector>
+
+// Stub out GPU calls as unavailable.
+
+#define NO_GPU LOG(FATAL) << "CPU-only Mode"
+
+#define STUB_GPU(classname) \
+template <typename Dtype> \
+Dtype classname<Dtype>::Forward_gpu(const vector<Blob<Dtype>*>& bottom, \
+    vector<Blob<Dtype>*>* top) { NO_GPU; } \
+template <typename Dtype> \
+void classname<Dtype>::Backward_gpu(const vector<Blob<Dtype>*>& top, \
+    const vector<bool>& propagate_down, \
+    vector<Blob<Dtype>*>* bottom) { NO_GPU; } \
+
+#define STUB_GPU_FORWARD(classname, funcname) \
+template <typename Dtype> \
+Dtype classname<Dtype>::funcname##_##gpu(const vector<Blob<Dtype>*>& bottom, \
+    vector<Blob<Dtype>*>* top) { NO_GPU; } \
+
+#define STUB_GPU_BACKWARD(classname, funcname) \
+template <typename Dtype> \
+void classname<Dtype>::funcname##_##gpu(const vector<Blob<Dtype>*>& top, \
+    const vector<bool>& propagate_down, \
+    vector<Blob<Dtype>*>* bottom) { NO_GPU; } \
+
+#else  // Normal GPU + CPU Caffe.
+
+#include <cublas_v2.h>
+#include <cuda.h>
+#include <cuda_runtime.h>
+#include <curand.h>
+#include <driver_types.h>  // cuda driver types
+
+#endif  // CPU_ONLY
+
+#endif  // CAFFE_UTIL_DEVICE_ALTERNATE_H_