Change `PADDLE_ONLY_CPU` to `PADDLE_WITH_GPU`

By shell command

```bash
sed -i 's#ifdef PADDLE_ONLY_CPU#ifndef PADDLE_WITH_GPU#g' `find ./paddle/ -name '*.h' -o -name '*.cc' -o -name '*.cpp' -o -name '*.c' -o -name '*.cu'`
sed -i 's#ifndef PADDLE_ONLY_CPU#ifdef PADDLE_WITH_GPU#g' `find ./paddle/ -name '*.h' -o -name '*.cc' -o -name '*.cpp' -o -name '*.c' -o -name '*.cu'`
```

cmake/configure.cmake

@@ -49,11 +49,11 @@ if(NOT WITH_GOLANG)
 endif(NOT WITH_GOLANG)
 
 if(NOT WITH_GPU)
-    add_definitions(-DPADDLE_ONLY_CPU)
     add_definitions(-DHPPL_STUB_FUNC)
 
     list(APPEND CMAKE_CXX_SOURCE_FILE_EXTENSIONS cu)
 else()
+    add_definitions(-DPADDLE_WITH_GPU)
     FIND_PACKAGE(CUDA REQUIRED)
 
     if(${CUDA_VERSION_MAJOR} VERSION_LESS 7)

paddle/api/Util.cpp

@@ -47,7 +47,7 @@ bool isUsingGpu() { return FLAGS_use_gpu; }
 void setUseGpu(bool useGpu) { FLAGS_use_gpu = useGpu; }
 
 bool isGpuVersion() {
-#ifdef PADDLE_ONLY_CPU
+#ifndef PADDLE_WITH_GPU
   return false;
 #else
   return true;

paddle/capi/Matrix.cpp

@@ -46,7 +46,7 @@ paddle_error paddle_matrix_set_row(paddle_matrix mat,
   if (rowID >= ptr->mat->getHeight()) return kPD_OUT_OF_RANGE;
   paddle::real* buf = ptr->mat->getRowBuf(rowID);
   size_t width = ptr->mat->getWidth();
-#ifndef PADDLE_ONLY_CPU
+#ifdef PADDLE_WITH_GPU
   hl_memcpy(buf, rowArray, sizeof(paddle::real) * width);
 #else
   std::copy(rowArray, rowArray + width, buf);

paddle/framework/lod_tensor.h

@@ -15,7 +15,7 @@
 #pragma once
 
 #include <memory>
-#ifndef PADDLE_ONLY_CPU
+#ifdef PADDLE_WITH_GPU
 #include <thrust/device_vector.h>
 #include <thrust/host_vector.h>
 #include <thrust/system/cuda/experimental/pinned_allocator.h>
@@ -29,7 +29,7 @@
 namespace paddle {
 namespace framework {
 
-#ifdef PADDLE_ONLY_CPU
+#ifndef PADDLE_WITH_GPU
 template <typename T>
 using Vector = std::vector<T>;
 #else

paddle/framework/op_registry.h

@@ -211,7 +211,7 @@ class OpKernelRegistrar : public Registrar {
 // TODO(fengjiayi): The following macros
 // seems ugly, do we have better method?
 
-#ifdef PADDLE_ONLY_CPU
+#ifndef PADDLE_WITH_GPU
 #define USE_OP_KERNEL(op_type) USE_OP_DEVICE_KERNEL(op_type, CPU)
 #else
 #define USE_OP_KERNEL(op_type)        \

paddle/framework/operator.cc

@@ -25,7 +25,7 @@ Eigen::DefaultDevice& ExecutionContext::GetEigenDevice<
   return *device_context_.GetEigenDevice<platform::CPUPlace>();
 }
 
-#ifndef PADDLE_ONLY_CPU
+#ifdef PADDLE_WITH_GPU
 template <>
 Eigen::GpuDevice&
 ExecutionContext::GetEigenDevice<platform::GPUPlace, Eigen::GpuDevice>() const {

paddle/framework/tensor_impl.h

@@ -65,7 +65,7 @@ inline T* Tensor::mutable_data(platform::Place place) {
       holder_.reset(new PlaceholderImpl<T, platform::CPUPlace>(
           boost::get<platform::CPUPlace>(place), size));
     } else if (platform::is_gpu_place(place)) {
-#ifdef PADDLE_ONLY_CPU
+#ifndef PADDLE_WITH_GPU
       PADDLE_THROW("'GPUPlace' is not supported in CPU only device.");
     }
 #else
@@ -103,7 +103,7 @@ inline void Tensor::CopyFrom(const Tensor& src,
     memory::Copy(boost::get<platform::CPUPlace>(dst_place), dst_ptr,
                  boost::get<platform::CPUPlace>(src_place), src_ptr, size);
   }
-#ifndef PADDLE_ONLY_CPU
+#ifdef PADDLE_WITH_GPU
   else if (platform::is_gpu_place(src_place) &&
            platform::is_cpu_place(dst_place)) {
     memory::Copy(boost::get<platform::CPUPlace>(dst_place), dst_ptr,

paddle/framework/tensor_test.cc

@@ -74,7 +74,7 @@ TEST(Tensor, MutableData) {
     EXPECT_EQ(p1, p2);
   }
 
-#ifndef PADDLE_ONLY_CPU
+#ifdef PADDLE_WITH_GPU
   {
     Tensor src_tensor;
     float* p1 = nullptr;
@@ -126,7 +126,7 @@ TEST(Tensor, ShareDataWith) {
     ASSERT_EQ(src_tensor.data<int>(), dst_tensor.data<int>());
   }
 
-#ifndef PADDLE_ONLY_CPU
+#ifdef PADDLE_WITH_GPU
   {
     Tensor src_tensor;
     Tensor dst_tensor;
@@ -163,7 +163,7 @@ TEST(Tensor, Slice) {
     EXPECT_EQ(src_data_address + 3 * 4 * 1 * sizeof(int), slice_data_address);
   }
 
-#ifndef PADDLE_ONLY_CPU
+#ifdef PADDLE_WITH_GPU
   {
     Tensor src_tensor;
     src_tensor.mutable_data<double>(make_ddim({6, 9}), GPUPlace());
@@ -218,7 +218,7 @@ TEST(Tensor, CopyFrom) {
       EXPECT_EQ(dst_ptr[i], slice_ptr[i]);
     }
   }
-#ifndef PADDLE_ONLY_CPU
+#ifdef PADDLE_WITH_GPU
   {
     Tensor src_tensor;
     Tensor gpu_tensor;

paddle/function/BlockExpandOp.cpp

@@ -194,7 +194,7 @@ public:
 
 REGISTER_TYPED_FUNC(BlockExpand, CPU, BlockExpandForward);
 REGISTER_TYPED_FUNC(BlockExpandGrad, CPU, BlockExpandBackward);
-#ifndef PADDLE_ONLY_CPU
+#ifdef PADDLE_WITH_GPU
 REGISTER_TYPED_FUNC(BlockExpand, GPU, BlockExpandForward);
 REGISTER_TYPED_FUNC(BlockExpandGrad, GPU, BlockExpandBackward);
 #endif

paddle/function/ContextProjectionOp.cpp

@@ -395,7 +395,7 @@ REGISTER_TYPED_FUNC(ContextProjectionForward,
 REGISTER_TYPED_FUNC(ContextProjectionBackward,
                     CPU,
                     ContextProjectionBackwardFunc);
-#ifndef PADDLE_ONLY_CPU
+#ifdef PADDLE_WITH_GPU
 REGISTER_TYPED_FUNC(ContextProjectionForward,
                     GPU,
                     ContextProjectionForwardFunc);

paddle/function/CosSimOp.cpp

@@ -233,7 +233,7 @@ private:
 
 REGISTER_TYPED_FUNC(CosSimForward, CPU, CosSimForwardFunc);
 REGISTER_TYPED_FUNC(CosSimBackward, CPU, CosSimBackwardFunc);
-#ifndef PADDLE_ONLY_CPU
+#ifdef PADDLE_WITH_GPU
 REGISTER_TYPED_FUNC(CosSimForward, GPU, CosSimForwardFunc);
 REGISTER_TYPED_FUNC(CosSimBackward, GPU, CosSimBackwardFunc);
 #endif

paddle/function/CropOp.cpp

@@ -169,7 +169,7 @@ private:
 
 REGISTER_TYPED_FUNC(Crop, CPU, CropFunc);
 REGISTER_TYPED_FUNC(CropGrad, CPU, CropGradFunc);
-#ifndef PADDLE_ONLY_CPU
+#ifdef PADDLE_WITH_GPU
 REGISTER_TYPED_FUNC(Crop, GPU, CropFunc);
 REGISTER_TYPED_FUNC(CropGrad, GPU, CropGradFunc);
 #endif

paddle/function/CrossMapNormalOp.cpp

@@ -336,7 +336,7 @@ private:
 
 REGISTER_TYPED_FUNC(CrossMapNormal, CPU, CrossMapNormalFunc);
 REGISTER_TYPED_FUNC(CrossMapNormalGrad, CPU, CrossMapNormalGradFunc);
-#ifndef PADDLE_ONLY_CPU
+#ifdef PADDLE_WITH_GPU
 REGISTER_TYPED_FUNC(CrossMapNormal, GPU, CrossMapNormalFunc);
 REGISTER_TYPED_FUNC(CrossMapNormalGrad, GPU, CrossMapNormalGradFunc);
 #endif

paddle/function/DepthwiseConvOp.cpp

@@ -292,7 +292,7 @@ REGISTER_TYPED_FUNC(DepthwiseConvGradInput,
 REGISTER_TYPED_FUNC(DepthwiseConvGradFilter,
                     CPU,
                     DepthwiseConvGradFilterFunction);
-#ifndef PADDLE_ONLY_CPU
+#ifdef PADDLE_WITH_GPU
 REGISTER_TYPED_FUNC(DepthwiseConv, GPU, DepthwiseConvFunction);
 REGISTER_TYPED_FUNC(DepthwiseConvGradInput,
                     GPU,

paddle/function/DepthwiseConvOpTest.cpp

@@ -17,7 +17,7 @@ limitations under the License. */
 
 namespace paddle {
 
-#ifndef PADDLE_ONLY_CPU
+#ifdef PADDLE_WITH_GPU
 TEST(DepthwiseConv, Forward) {
   DepthwiseConvolution<DEVICE_TYPE_CPU, DEVICE_TYPE_GPU>(
       "GemmConv-CPU", "DepthwiseConv-GPU", forward);

paddle/function/GemmConvOp.cpp

@@ -340,7 +340,7 @@ public:
 REGISTER_TYPED_FUNC(GemmConv, CPU, GemmConvFunction);
 REGISTER_TYPED_FUNC(GemmConvGradInput, CPU, GemmConvGradInputFunction);
 REGISTER_TYPED_FUNC(GemmConvGradFilter, CPU, GemmConvGradFilterFunction);
-#ifndef PADDLE_ONLY_CPU
+#ifdef PADDLE_WITH_GPU
 REGISTER_TYPED_FUNC(GemmConv, GPU, GemmConvFunction);
 REGISTER_TYPED_FUNC(GemmConvGradInput, GPU, GemmConvGradInputFunction);
 REGISTER_TYPED_FUNC(GemmConvGradFilter, GPU, GemmConvGradFilterFunction);

paddle/function/GemmConvOpTest.cpp

@@ -24,7 +24,7 @@ TEST(GemmConv, NaiveConv) {
       "NaiveConv-CPU", "GemmConv-CPU", forward);
 }
 
-#ifndef PADDLE_ONLY_CPU
+#ifdef PADDLE_WITH_GPU
 TEST(GemmConv, Forward) {
   Convolution<DEVICE_TYPE_CPU, DEVICE_TYPE_GPU>(
       "GemmConv-CPU", "GemmConv-GPU", forward);

paddle/function/Im2ColTest.cpp

@@ -116,7 +116,7 @@ void TestIm2ColFunctor() {
 
 TEST(Im2ColFunctor, CPU) { TestIm2ColFunctor<DEVICE_TYPE_CPU, float>(); }
 
-#ifndef PADDLE_ONLY_CPU
+#ifdef PADDLE_WITH_GPU
 
 TEST(Im2ColFunctor, GPU) { TestIm2ColFunctor<DEVICE_TYPE_GPU, float>(); }
 

paddle/function/MulOp.cpp

@@ -341,7 +341,7 @@ private:
 };
 
 REGISTER_TYPED_FUNC(MulOp, CPU, MulFunc);
-#ifndef PADDLE_ONLY_CPU
+#ifdef PADDLE_WITH_GPU
 REGISTER_TYPED_FUNC(MulOp, GPU, MulFunc);
 #endif
 }  // namespace paddle

paddle/function/PadOp.cpp

@@ -207,7 +207,7 @@ private:
 
 REGISTER_TYPED_FUNC(Pad, CPU, PadFunc);
 REGISTER_TYPED_FUNC(PadGrad, CPU, PadGradFunc);
-#ifndef PADDLE_ONLY_CPU
+#ifdef PADDLE_WITH_GPU
 REGISTER_TYPED_FUNC(Pad, GPU, PadFunc);
 REGISTER_TYPED_FUNC(PadGrad, GPU, PadGradFunc);
 #endif

paddle/function/RowConvOp.cpp

@@ -217,7 +217,7 @@ public:
 
 REGISTER_TYPED_FUNC(RowConv, CPU, RowConvFunc);
 REGISTER_TYPED_FUNC(RowConvGrad, CPU, RowConvGradFunc);
-#ifndef PADDLE_ONLY_CPU
+#ifdef PADDLE_WITH_GPU
 REGISTER_TYPED_FUNC(RowConv, GPU, RowConvFunc);
 REGISTER_TYPED_FUNC(RowConvGrad, GPU, RowConvGradFunc);
 #endif

paddle/function/SwitchOp.cpp

@@ -132,7 +132,7 @@ public:
 
 REGISTER_TYPED_FUNC(NCHW2NHWC, CPU, NCHW2NHWCFunc);
 REGISTER_TYPED_FUNC(NHWC2NCHW, CPU, NHWC2NCHWFunc);
-#ifndef PADDLE_ONLY_CPU
+#ifdef PADDLE_WITH_GPU
 REGISTER_TYPED_FUNC(NCHW2NHWC, GPU, NCHW2NHWCFunc);
 REGISTER_TYPED_FUNC(NHWC2NCHW, GPU, NHWC2NCHWFunc);
 #endif

paddle/gserver/layers/BatchNormBaseLayer.cpp

@@ -16,7 +16,7 @@ limitations under the License. */
 #include "BatchNormalizationLayer.h"
 #include "Layer.h"
 #include "paddle/utils/Stat.h"
-#ifndef PADDLE_ONLY_CPU
+#ifdef PADDLE_WITH_GPU
 #include "CudnnBatchNormLayer.h"
 #endif
 

paddle/gserver/layers/BatchNormalizationLayer.cpp

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/utils/Stat.h"
-#ifndef PADDLE_ONLY_CPU
+#ifdef PADDLE_WITH_GPU
 #include "hl_batch_transpose.h"
 #endif
 #include "BatchNormalizationLayer.h"
@@ -90,7 +90,7 @@ void BatchNormalizationLayer::expandMat(const MatrixPtr& in, MatrixPtr& out) {
   size_t batchSize = in->getHeight();
   CHECK_EQ(out->getHeight(), batchSize * imgPixels_);
   if (useGpu_) {
-#ifdef PADDLE_ONLY_CPU
+#ifndef PADDLE_WITH_GPU
     LOG(FATAL) << "paddle is compiled only for cpu";
 #else
     batchTranspose(
@@ -127,7 +127,7 @@ void BatchNormalizationLayer::shrinkMat(const MatrixPtr& in, MatrixPtr& out) {
   }
   CHECK_EQ(in->getHeight(), static_cast<size_t>(batchSize * imgPixels_));
   if (useGpu_) {
-#ifdef PADDLE_ONLY_CPU
+#ifndef PADDLE_WITH_GPU
     LOG(FATAL) << "paddle is compiled only for cpu";
 #else
     batchTranspose(

paddle/gserver/layers/PoolLayer.cpp

@@ -15,7 +15,7 @@ limitations under the License. */
 #include "PoolLayer.h"
 #include "PoolProjectionLayer.h"
 #include "paddle/utils/Logging.h"
-#ifndef PADDLE_ONLY_CPU
+#ifdef PADDLE_WITH_GPU
 #include "CudnnPoolLayer.h"
 #endif
 namespace paddle {
@@ -53,7 +53,7 @@ Layer* PoolLayer::create(const LayerConfig& config) {
   const std::string& pool = config.inputs(0).pool_conf().pool_type();
   if (pool == "max-projection" || pool == "avg-projection") {
     return new PoolProjectionLayer(config);
-#ifndef PADDLE_ONLY_CPU
+#ifdef PADDLE_WITH_GPU
   } else if (CudnnPoolLayer::typeCheck(pool)) {
     return new CudnnPoolLayer(config);
 #endif