Build: unified usage of macros USE_GPU and CPU_ONLY as USE_GPU

modules/perception/base/blob_test.cc

@@ -116,7 +116,7 @@ TEST(BlobTest, source_test) {
   blob_4d.Reshape({1, 3, 4, 5});
   blob_4d.set_cpu_data(cpu_data);
 
-#ifndef CPU_ONLY
+#if USE_GPU == 1
   blob_4d.Reshape({2, 3, 4, 5});
   float* gpu_data = nullptr;
   BASE_CUDA_CHECK(cudaMalloc(&gpu_data, blob_4d.count() * sizeof(float)));

modules/perception/base/common.h

@@ -16,7 +16,9 @@
 
 #pragma once
 
-#ifndef CPU_ONLY
+#include <cassert>
+
+#if USE_GPU == 1
 
 #include <cublas_v2.h>
 #include <cuda_runtime.h>
@@ -31,7 +33,7 @@ namespace base {
 #define NO_GPU assert(false)
 #endif
 
-#ifndef CPU_ONLY
+#if USE_GPU == 1
 
 #define BASE_CUDA_CHECK(condition) \
   { apollo::perception::base::GPUAssert((condition), __FILE__, __LINE__); }

modules/perception/base/common_test.cc

@@ -22,7 +22,7 @@ namespace perception {
 namespace base {
 
 TEST(CommonTest, GPUAssertTest) {
-#ifndef CPU_ONLY
+#if USE_GPU == 1
   GPUAssert(cudaSetDevice(-1), __FILE__, __LINE__, false);
 #endif
 }

modules/perception/base/options_define.h

-/******************************************************************************
- * Copyright 2018 The Apollo Authors. All Rights Reserved.
- *
- * Licensed under the Apache License, Version 2.0 (the "License");
- * you may not use this file except in compliance with the License.
- * You may obtain a copy of the License at
- *
- * http://www.apache.org/licenses/LICENSE-2.0
- *
- * Unless required by applicable law or agreed to in writing, software
- * distributed under the License is distributed on an "AS IS" BASIS,
- * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
- * See the License for the specific language governing permissions and
- * limitations under the License.
- *****************************************************************************/
-#pragma once
-
-// build perception with cpu or gpu
-/* #undef CPU_ONLY */
-
-// build perception with using cyber log lib or glog lib
-/* #undef PERCEPTION_USE_CYBER */

modules/perception/base/syncedmem.cc

@@ -75,7 +75,7 @@ SyncedMemory::SyncedMemory(bool use_cuda)
       cpu_malloc_use_cuda_(use_cuda),
       own_gpu_data_(false),
       device_(-1) {
-#ifndef CPU_ONLY
+#if USE_GPU == 1
 #ifdef PERCEPTION_DEBUG
   BASE_CUDA_CHECK(cudaGetDevice(&device_));
 #endif
@@ -91,7 +91,7 @@ SyncedMemory::SyncedMemory(size_t size, bool use_cuda)
       cpu_malloc_use_cuda_(use_cuda),
       own_gpu_data_(false),
       device_(-1) {
-#ifndef CPU_ONLY
+#if USE_GPU == 1
 #ifdef PERCEPTION_DEBUG
   BASE_CUDA_CHECK(cudaGetDevice(&device_));
 #endif
@@ -104,11 +104,11 @@ SyncedMemory::~SyncedMemory() {
     PerceptionFreeHost(cpu_ptr_, cpu_malloc_use_cuda_);
   }
 
-#ifndef CPU_ONLY
+#if USE_GPU == 1
   if (gpu_ptr_ && own_gpu_data_) {
     BASE_CUDA_CHECK(cudaFree(gpu_ptr_));
   }
-#endif  // CPU_ONLY
+#endif  // USE_GPU
 }
 
 inline void SyncedMemory::to_cpu() {
@@ -125,7 +125,7 @@ inline void SyncedMemory::to_cpu() {
       own_cpu_data_ = true;
       break;
     case HEAD_AT_GPU:
-#ifndef CPU_ONLY
+#if USE_GPU == 1
       if (cpu_ptr_ == nullptr) {
         PerceptionMallocHost(&cpu_ptr_, size_, cpu_malloc_use_cuda_);
         own_cpu_data_ = true;
@@ -144,7 +144,7 @@ inline void SyncedMemory::to_cpu() {
 
 inline void SyncedMemory::to_gpu() {
   check_device();
-#ifndef CPU_ONLY
+#if USE_GPU == 1
   switch (head_) {
     case UNINITIALIZED:
       BASE_CUDA_CHECK(cudaMalloc(&gpu_ptr_, size_));
@@ -188,7 +188,7 @@ void SyncedMemory::set_cpu_data(void* data) {
 
 const void* SyncedMemory::gpu_data() {
   check_device();
-#ifndef CPU_ONLY
+#if USE_GPU == 1
   to_gpu();
   return (const void*)gpu_ptr_;
 #else
@@ -199,7 +199,7 @@ const void* SyncedMemory::gpu_data() {
 
 void SyncedMemory::set_gpu_data(void* data) {
   check_device();
-#ifndef CPU_ONLY
+#if USE_GPU == 1
   ACHECK(data);
   if (own_gpu_data_) {
     BASE_CUDA_CHECK(cudaFree(gpu_ptr_));
@@ -221,7 +221,7 @@ void* SyncedMemory::mutable_cpu_data() {
 
 void* SyncedMemory::mutable_gpu_data() {
   check_device();
-#ifndef CPU_ONLY
+#if USE_GPU == 1
   to_gpu();
   head_ = HEAD_AT_GPU;
   return gpu_ptr_;
@@ -231,7 +231,7 @@ void* SyncedMemory::mutable_gpu_data() {
 #endif
 }
 
-#ifndef CPU_ONLY
+#if USE_GPU == 1
 void SyncedMemory::async_gpu_push(const cudaStream_t& stream) {
   check_device();
   CHECK_EQ(head_, HEAD_AT_CPU);
@@ -247,7 +247,7 @@ void SyncedMemory::async_gpu_push(const cudaStream_t& stream) {
 #endif
 
 void SyncedMemory::check_device() {
-#ifndef CPU_ONLY
+#if USE_GPU == 1
 #ifdef PERCEPTION_DEBUG
   int device;
   cudaGetDevice(&device);

modules/perception/base/syncedmem.h

@@ -70,7 +70,7 @@ namespace perception {
 namespace base {
 
 inline void PerceptionMallocHost(void** ptr, size_t size, bool use_cuda) {
-#ifndef CPU_ONLY
+#if USE_GPU == 1
   if (use_cuda) {
     BASE_CUDA_CHECK(cudaMallocHost(ptr, size));
     return;
@@ -81,7 +81,7 @@ inline void PerceptionMallocHost(void** ptr, size_t size, bool use_cuda) {
 }
 
 inline void PerceptionFreeHost(void* ptr, bool use_cuda) {
-#ifndef CPU_ONLY
+#if USE_GPU == 1
   if (use_cuda) {
     BASE_CUDA_CHECK(cudaFreeHost(ptr));
     return;
@@ -119,7 +119,7 @@ class SyncedMemory {
   void set_head_cpu() { set_head(HEAD_AT_CPU); }
   size_t size() { return size_; }
 
-#ifndef CPU_ONLY
+#if USE_GPU == 1
   void async_gpu_push(const cudaStream_t& stream);
 #endif
 

modules/perception/base/syncedmem_test.cc

@@ -100,7 +100,7 @@ TEST_F(SyncedMemoryTest, TestInitialization) {
   delete p_mem;
 }
 
-#ifndef CPU_ONLY  // GPU test
+#if USE_GPU == 1  // GPU test
 
 TEST_F(SyncedMemoryTest, TestAllocationCPUGPU) {
   SyncedMemory mem(10, true);
@@ -118,7 +118,7 @@ TEST_F(SyncedMemoryTest, TestAllocationCPU) {
   EXPECT_TRUE(mem.mutable_cpu_data());
 }
 
-#ifndef CPU_ONLY  // GPU test
+#if USE_GPU == 1  // GPU test
 
 TEST_F(SyncedMemoryTest, TestAllocationGPU) {
   SyncedMemory mem(10, true);
@@ -145,7 +145,7 @@ TEST_F(SyncedMemoryTest, TestCPUWrite) {
   }
 }
 
-#ifndef CPU_ONLY  // GPU test
+#if USE_GPU == 1  // GPU test
 
 TEST_F(SyncedMemoryTest, TestGPURead) {
   SyncedMemory mem(10, true);

modules/perception/base/test/test_helper.h

@@ -46,7 +46,7 @@ struct CPUDevice {
 template <typename Dtype>
 class CPUDeviceTest : public MultiDeviceTest<CPUDevice<Dtype>> {};
 
-#ifdef CPU_ONLY
+#if USE_GPU == 0
 
 typedef ::testing::Types<CPUDevice<float>, CPUDevice<double>>
     TestDtypesAndDevices;

modules/perception/camera/common/data_provider.cc

@@ -15,7 +15,9 @@
  *****************************************************************************/
 #include "modules/perception/camera/common/data_provider.h"
 
+#if USE_GPU == 1
 #include <nppi.h>
+#endif
 
 #include "cyber/common/log.h"
 
@@ -124,7 +126,7 @@ bool DataProvider::FillImageData(int rows, int cols, const uint8_t *data,
 
   bool success = false;
 
-#ifdef CPU_ONLY  // copy to host memory
+#if USE_GPU == 0  // copy to host memory
   AINFO << "Fill in CPU mode ...";
   if (handler_ != nullptr) {
     AERROR << "Undistortion DO NOT support CPU mode!";

modules/perception/camera/common/undistortion_handler.cc

@@ -17,7 +17,10 @@
 
 #include <vector>
 
-#include "cuda/include/npp.h"
+#if USE_GPU == 1
+#include <npp.h>
+#endif
+
 #include "Eigen/Dense"
 
 #include "cyber/common/log.h"

modules/perception/camera/test/camera_lib_traffic_light_detector_detection_test.cc

@@ -116,7 +116,7 @@ TEST(DetectionTest, all) {
   gt_lights.push_back(light2);
   frame.traffic_lights = lights;
 
-#ifndef CPU_ONLY
+#if USE_GPU == 1
   EXPECT_TRUE(detector->Detect(detetor_options, &frame));
 
   EXPECT_EQ(lights.size(), gt_lights.size());
@@ -185,7 +185,7 @@ TEST(DetectionTest, no_light) {
   lights.clear();
   frame.traffic_lights = lights;
 
-#ifndef CPU_ONLY
+#if USE_GPU == 1
   EXPECT_TRUE(detector->Detect(detetor_options, &frame));
 #endif
 }
@@ -255,7 +255,7 @@ TEST(DetectionTest, out_of_img_light) {
   gt_lights.push_back(light2);
   frame.traffic_lights = lights;
 
-#ifndef CPU_ONLY
+#if USE_GPU == 1
   EXPECT_TRUE(detector->Detect(detetor_options, &frame));
 
   EXPECT_FALSE(lights[0]->region.is_detected);

modules/perception/camera/test/camera_lib_traffic_light_detector_recognition_test.cc

@@ -69,7 +69,7 @@ TEST(RecognizeTest, yellow) {
 
   frame.traffic_lights = lights;
 
-#ifndef CPU_ONLY
+#if USE_GPU == 1
   ASSERT_TRUE(recognition->Detect(recognition_options, &frame));
   AINFO << "COLOR " << static_cast<int>(lights[0]->status.color);
   ASSERT_TRUE(base::TLColor::TL_YELLOW == lights[0]->status.color);
@@ -125,7 +125,7 @@ TEST(RecognizeTest, red) {
 
   frame.traffic_lights = lights;
 
-#ifndef CPU_ONLY
+#if USE_GPU == 1
   ASSERT_TRUE(recognition->Detect(recognition_options, &frame));
   AINFO << "COLOR " << static_cast<int>(lights[0]->status.color);
   ASSERT_TRUE(base::TLColor::TL_RED == lights[0]->status.color);
@@ -233,7 +233,7 @@ TEST(RecognizeTest, black) {
 
   frame.traffic_lights = lights;
 
-#ifndef CPU_ONLY
+#if USE_GPU == 1
   ASSERT_TRUE(recognition->Detect(recognition_options, &frame));
   AINFO << "COLOR " << static_cast<int>(lights[0]->status.color);
   ASSERT_TRUE(base::TLColor::TL_BLACK == lights[0]->status.color);
@@ -285,7 +285,7 @@ TEST(RecognizeTest, no_detection) {
   lights[0]->region.is_detected = false;
   frame.traffic_lights = lights;
 
-#ifndef CPU_ONLY
+#if USE_GPU == 1
   ASSERT_TRUE(recognition->Detect(recognition_options, &frame));
   AINFO << "COLOR " << static_cast<int>(lights[0]->status.color);
   ASSERT_TRUE(base::TLColor::TL_UNKNOWN_COLOR == lights[0]->status.color);
@@ -341,7 +341,7 @@ TEST(RecognizeTest, unknown_class) {
 
   frame.traffic_lights = lights;
 
-#ifndef CPU_ONLY
+#if USE_GPU == 1
   ASSERT_FALSE(recognition->Detect(recognition_options, &frame));
 #endif
 }
@@ -395,7 +395,7 @@ TEST(RecognizeTest, quadrate) {
 
   frame.traffic_lights = lights;
 
-#ifndef CPU_ONLY
+#if USE_GPU == 1
   ASSERT_TRUE(recognition->Detect(recognition_options, &frame));
   AINFO << "COLOR " << static_cast<int>(lights[0]->status.color);
   ASSERT_TRUE(base::TLColor::TL_GREEN == lights[0]->status.color);
@@ -457,7 +457,7 @@ TEST(RecognizeTest, horizontal) {
 
   frame.traffic_lights = lights;
 
-#ifndef CPU_ONLY
+#if USE_GPU == 1
   ASSERT_TRUE(recognition->Detect(recognition_options, &frame));
   AINFO << "COLOR " << static_cast<int>(lights[0]->status.color);
   ASSERT_TRUE(base::TLColor::TL_GREEN == lights[0]->status.color);
@@ -507,7 +507,7 @@ TEST(RecognizeTest, no_light) {
   lights.clear();
   frame.traffic_lights = lights;
 
-#ifndef CPU_ONLY
+#if USE_GPU == 1
   ASSERT_TRUE(recognition->Detect(recognition_options, &frame));
   EXPECT_EQ(lights.size(), 0);
 #endif

modules/perception/lidar/lib/segmentation/cnnseg/feature_generator.cc

@@ -46,7 +46,7 @@ bool FeatureGenerator::Init(const FeatureParam& feature_param,
 
   // set output feature blob data
   float* out_blob_data = nullptr;
-#ifndef CPU_ONLY
+#if USE_GPU == 1
   log_blob_.reset(
       new base::Blob<float>(1, 1, 1, static_cast<int>(log_table_.size())));
   float* log_table = log_blob_->mutable_gpu_data();
@@ -95,7 +95,7 @@ bool FeatureGenerator::Init(const FeatureParam& feature_param,
     }
 
 // memory copy direction and distance features
-#ifndef CPU_ONLY
+#if USE_GPU == 1
     cudaMemcpy(direction_data_, direction_data.data(),
                direction_data.size() * sizeof(float), cudaMemcpyHostToDevice);
     cudaMemcpy(distance_data_, distance_data.data(),

modules/perception/lidar/lib/segmentation/cnnseg/feature_generator.h

@@ -33,7 +33,7 @@ namespace lidar {
 class FeatureGenerator {
  public:
   FeatureGenerator() {
-#ifndef CPU_ONLY
+#if USE_GPU == 1
     pc_gpu_size_ = kMaxPointCloudGPUSize;
     BASE_CUDA_CHECK(cudaMalloc(reinterpret_cast<void**>(&pc_gpu_),
                                pc_gpu_size_ * sizeof(base::PointF)));
@@ -43,7 +43,7 @@ class FeatureGenerator {
   }
 
   virtual ~FeatureGenerator() {
-#ifndef CPU_ONLY
+#if USE_GPU == 1
     ReleaseGPUMemory();
 #endif
   }
@@ -52,7 +52,7 @@ class FeatureGenerator {
 
   void Generate(const base::PointFCloudPtr& pc_ptr,
                 const std::vector<int>& point2grid) {
-#ifndef CPU_ONLY
+#if USE_GPU == 1
     GenerateGPU(pc_ptr, point2grid);
 #else
     GenerateCPU(pc_ptr, point2grid);
@@ -62,7 +62,7 @@ class FeatureGenerator {
   inline std::string Name() const { return "FeatureGenerator"; }
 
  private:
-#ifndef CPU_ONLY
+#if USE_GPU == 1
   void GenerateGPU(const base::PointFCloudPtr& pc_ptr,
                    const std::vector<int>& point2grid);
   void ReleaseGPUMemory();

modules/planning/open_space/trajectory_smoother/planning_block.h

@@ -16,7 +16,6 @@
 
 #pragma once
 
-#include <cuda.h>
 #include <cuda_runtime.h>
 
 namespace apollo {