From 84500f9487164f3cf17625c876c15d754b932ced Mon Sep 17 00:00:00 2001 From: Yu Yang Date: Wed, 4 Oct 2017 11:25:46 -0700 Subject: [PATCH] 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 | 2 +- paddle/api/Util.cpp | 2 +- paddle/capi/Matrix.cpp | 2 +- paddle/framework/lod_tensor.h | 4 +-- paddle/framework/op_registry.h | 2 +- paddle/framework/operator.cc | 2 +- paddle/framework/tensor_impl.h | 4 +-- paddle/framework/tensor_test.cc | 8 +++--- paddle/function/BlockExpandOp.cpp | 2 +- paddle/function/ContextProjectionOp.cpp | 2 +- paddle/function/CosSimOp.cpp | 2 +- paddle/function/CropOp.cpp | 2 +- paddle/function/CrossMapNormalOp.cpp | 2 +- paddle/function/DepthwiseConvOp.cpp | 2 +- paddle/function/DepthwiseConvOpTest.cpp | 2 +- paddle/function/GemmConvOp.cpp | 2 +- paddle/function/GemmConvOpTest.cpp | 2 +- paddle/function/Im2ColTest.cpp | 2 +- paddle/function/MulOp.cpp | 2 +- paddle/function/PadOp.cpp | 2 +- paddle/function/RowConvOp.cpp | 2 +- paddle/function/SwitchOp.cpp | 2 +- paddle/gserver/layers/BatchNormBaseLayer.cpp | 2 +- .../layers/BatchNormalizationLayer.cpp | 6 ++--- paddle/gserver/layers/PoolLayer.cpp | 4 +-- paddle/gserver/tests/LayerGradUtil.cpp | 2 +- paddle/gserver/tests/test_BatchNorm.cpp | 2 +- paddle/gserver/tests/test_ConvUnify.cpp | 2 +- paddle/gserver/tests/test_DetectionOutput.cpp | 2 +- paddle/gserver/tests/test_Evaluator.cpp | 2 +- paddle/gserver/tests/test_KmaxSeqScore.cpp | 2 +- paddle/gserver/tests/test_LayerGrad.cpp | 26 +++++++++---------- paddle/gserver/tests/test_NetworkCompare.cpp | 2 +- paddle/gserver/tests/test_PriorBox.cpp | 2 +- .../gserver/tests/test_ProtoDataProvider.cpp | 6 ++--- paddle/gserver/tests/test_PyDataProvider.cpp | 4 +-- .../gserver/tests/test_SelectiveFCLayer.cpp | 8 +++--- .../gserver/tests/test_SeqSliceLayerGrad.cpp | 2 +- paddle/gserver/tests/test_WarpCTCLayer.cpp | 2 +- paddle/math/Matrix.cpp | 6 ++--- paddle/math/SparseMatrix.cpp | 2 +- paddle/math/Vector.cpp | 6 ++--- paddle/math/tests/test_Allocator.cpp | 4 +-- paddle/math/tests/test_BaseMatrix.cpp | 2 +- paddle/math/tests/test_CpuGpuVector.cpp | 2 +- paddle/math/tests/test_ExecViaCpu.cpp | 2 +- paddle/math/tests/test_GpuProfiler.cpp | 2 +- paddle/math/tests/test_Matrix.cpp | 2 +- paddle/math/tests/test_SparseMatrix.cpp | 6 ++--- paddle/math/tests/test_Tensor.cu | 20 +++++++------- paddle/math/tests/test_TrainingAlgorithm.cpp | 2 +- paddle/math/tests/test_batchTranspose.cpp | 2 +- paddle/math/tests/test_lazyAssign.cu | 4 +-- paddle/math/tests/test_matrixCompare.cpp | 2 +- paddle/math/tests/test_perturbation.cpp | 2 +- .../math/tests/test_sparseMatrixCompare.cpp | 2 +- paddle/memory/detail/buddy_allocator.cc | 2 +- paddle/memory/detail/system_allocator.cc | 2 +- paddle/memory/detail/system_allocator.h | 2 +- paddle/memory/detail/system_allocator_test.cc | 2 +- paddle/memory/memcpy.cc | 2 +- paddle/memory/memcpy.h | 2 +- paddle/memory/memory.cc | 2 +- paddle/memory/memory_test.cc | 2 +- paddle/operators/detail/strided_memcpy.h | 2 +- paddle/operators/math/im2col_test.cc | 4 +-- paddle/operators/math/math_function_test.cc | 2 +- paddle/operators/strided_memcpy_test.cc | 2 +- paddle/platform/device_context.cc | 2 +- paddle/platform/device_context.h | 4 +-- paddle/platform/enforce.h | 4 +-- paddle/platform/gpu_info.h | 2 +- paddle/platform/variant.h | 2 +- paddle/pserver/test/SocketTest.cpp | 2 +- paddle/pserver/test/test_ProtoServer.cpp | 2 +- paddle/pybind/pybind.cc | 12 ++++----- paddle/pybind/tensor_py.h | 2 +- paddle/trainer/MergeModel.cpp | 2 +- paddle/trainer/tests/test_Compare.cpp | 2 +- paddle/trainer/tests/test_CompareSparse.cpp | 4 +-- paddle/trainer/tests/test_Trainer.cpp | 4 +-- paddle/trainer/tests/test_TrainerOnePass.cpp | 6 ++--- .../test_recurrent_machine_generation.cpp | 2 +- paddle/utils/Flags.cpp | 2 +- paddle/utils/Util.h | 2 +- paddle/utils/Version.h | 2 +- 86 files changed, 141 insertions(+), 141 deletions(-) diff --git a/cmake/configure.cmake b/cmake/configure.cmake index 51c3b918c..926a7b1d6 100644 --- a/cmake/configure.cmake +++ b/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) diff --git a/paddle/api/Util.cpp b/paddle/api/Util.cpp index d369df5d4..7446d892f 100644 --- a/paddle/api/Util.cpp +++ b/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; diff --git a/paddle/capi/Matrix.cpp b/paddle/capi/Matrix.cpp index d898ebe26..5b3737a75 100644 --- a/paddle/capi/Matrix.cpp +++ b/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); diff --git a/paddle/framework/lod_tensor.h b/paddle/framework/lod_tensor.h index 49786a4a6..b12c95b6b 100644 --- a/paddle/framework/lod_tensor.h +++ b/paddle/framework/lod_tensor.h @@ -15,7 +15,7 @@ #pragma once #include -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU #include #include #include @@ -29,7 +29,7 @@ namespace paddle { namespace framework { -#ifdef PADDLE_ONLY_CPU +#ifndef PADDLE_WITH_GPU template using Vector = std::vector; #else diff --git a/paddle/framework/op_registry.h b/paddle/framework/op_registry.h index 4ee2c7d27..aca6579f3 100644 --- a/paddle/framework/op_registry.h +++ b/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) \ diff --git a/paddle/framework/operator.cc b/paddle/framework/operator.cc index 1012a30b0..21c1c6f9e 100644 --- a/paddle/framework/operator.cc +++ b/paddle/framework/operator.cc @@ -25,7 +25,7 @@ Eigen::DefaultDevice& ExecutionContext::GetEigenDevice< return *device_context_.GetEigenDevice(); } -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU template <> Eigen::GpuDevice& ExecutionContext::GetEigenDevice() const { diff --git a/paddle/framework/tensor_impl.h b/paddle/framework/tensor_impl.h index a5405f9c3..1cde1f74b 100644 --- a/paddle/framework/tensor_impl.h +++ b/paddle/framework/tensor_impl.h @@ -65,7 +65,7 @@ inline T* Tensor::mutable_data(platform::Place place) { holder_.reset(new PlaceholderImpl( boost::get(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(dst_place), dst_ptr, boost::get(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(dst_place), dst_ptr, diff --git a/paddle/framework/tensor_test.cc b/paddle/framework/tensor_test.cc index e2ec738de..86c6945ab 100644 --- a/paddle/framework/tensor_test.cc +++ b/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(), dst_tensor.data()); } -#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(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; diff --git a/paddle/function/BlockExpandOp.cpp b/paddle/function/BlockExpandOp.cpp index a89b6bba4..ad78f5f58 100644 --- a/paddle/function/BlockExpandOp.cpp +++ b/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 diff --git a/paddle/function/ContextProjectionOp.cpp b/paddle/function/ContextProjectionOp.cpp index b87750b74..ab18c39df 100644 --- a/paddle/function/ContextProjectionOp.cpp +++ b/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); diff --git a/paddle/function/CosSimOp.cpp b/paddle/function/CosSimOp.cpp index 7ece7b2df..4418f144d 100644 --- a/paddle/function/CosSimOp.cpp +++ b/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 diff --git a/paddle/function/CropOp.cpp b/paddle/function/CropOp.cpp index f12ee43e3..39504cc2c 100644 --- a/paddle/function/CropOp.cpp +++ b/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 diff --git a/paddle/function/CrossMapNormalOp.cpp b/paddle/function/CrossMapNormalOp.cpp index ef878bfbb..1cf0918be 100644 --- a/paddle/function/CrossMapNormalOp.cpp +++ b/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 diff --git a/paddle/function/DepthwiseConvOp.cpp b/paddle/function/DepthwiseConvOp.cpp index 2f3112fe6..7656ab3d0 100644 --- a/paddle/function/DepthwiseConvOp.cpp +++ b/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, diff --git a/paddle/function/DepthwiseConvOpTest.cpp b/paddle/function/DepthwiseConvOpTest.cpp index d8e8c889d..39033ecb2 100644 --- a/paddle/function/DepthwiseConvOpTest.cpp +++ b/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( "GemmConv-CPU", "DepthwiseConv-GPU", forward); diff --git a/paddle/function/GemmConvOp.cpp b/paddle/function/GemmConvOp.cpp index f8cf4ebea..68e08c148 100644 --- a/paddle/function/GemmConvOp.cpp +++ b/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); diff --git a/paddle/function/GemmConvOpTest.cpp b/paddle/function/GemmConvOpTest.cpp index 5283d79a5..bd1cf3c6a 100644 --- a/paddle/function/GemmConvOpTest.cpp +++ b/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( "GemmConv-CPU", "GemmConv-GPU", forward); diff --git a/paddle/function/Im2ColTest.cpp b/paddle/function/Im2ColTest.cpp index acc88a553..55325e94b 100644 --- a/paddle/function/Im2ColTest.cpp +++ b/paddle/function/Im2ColTest.cpp @@ -116,7 +116,7 @@ void TestIm2ColFunctor() { TEST(Im2ColFunctor, CPU) { TestIm2ColFunctor(); } -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU TEST(Im2ColFunctor, GPU) { TestIm2ColFunctor(); } diff --git a/paddle/function/MulOp.cpp b/paddle/function/MulOp.cpp index 25e41edad..655026320 100644 --- a/paddle/function/MulOp.cpp +++ b/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 diff --git a/paddle/function/PadOp.cpp b/paddle/function/PadOp.cpp index adba7c92e..24c9bf4e7 100644 --- a/paddle/function/PadOp.cpp +++ b/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 diff --git a/paddle/function/RowConvOp.cpp b/paddle/function/RowConvOp.cpp index b6501e8f4..09e702f71 100644 --- a/paddle/function/RowConvOp.cpp +++ b/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 diff --git a/paddle/function/SwitchOp.cpp b/paddle/function/SwitchOp.cpp index 01e252a8d..db839b5b7 100644 --- a/paddle/function/SwitchOp.cpp +++ b/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 diff --git a/paddle/gserver/layers/BatchNormBaseLayer.cpp b/paddle/gserver/layers/BatchNormBaseLayer.cpp index f7a80e23e..55f52816a 100644 --- a/paddle/gserver/layers/BatchNormBaseLayer.cpp +++ b/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 diff --git a/paddle/gserver/layers/BatchNormalizationLayer.cpp b/paddle/gserver/layers/BatchNormalizationLayer.cpp index 412762d38..33cf24431 100644 --- a/paddle/gserver/layers/BatchNormalizationLayer.cpp +++ b/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(batchSize * imgPixels_)); if (useGpu_) { -#ifdef PADDLE_ONLY_CPU +#ifndef PADDLE_WITH_GPU LOG(FATAL) << "paddle is compiled only for cpu"; #else batchTranspose( diff --git a/paddle/gserver/layers/PoolLayer.cpp b/paddle/gserver/layers/PoolLayer.cpp index 96d5c54ac..43ab4e4d4 100644 --- a/paddle/gserver/layers/PoolLayer.cpp +++ b/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 diff --git a/paddle/gserver/tests/LayerGradUtil.cpp b/paddle/gserver/tests/LayerGradUtil.cpp index a38880e14..59df057a8 100644 --- a/paddle/gserver/tests/LayerGradUtil.cpp +++ b/paddle/gserver/tests/LayerGradUtil.cpp @@ -674,7 +674,7 @@ void testLayerGradKernel(TestConfig testConf, bool useGpu, bool useWeight, float epsilon) { -#ifdef PADDLE_ONLY_CPU +#ifndef PADDLE_WITH_GPU if (useGpu) return; #endif FLAGS_use_gpu = useGpu; diff --git a/paddle/gserver/tests/test_BatchNorm.cpp b/paddle/gserver/tests/test_BatchNorm.cpp index 659eefa31..c1c85f8fa 100644 --- a/paddle/gserver/tests/test_BatchNorm.cpp +++ b/paddle/gserver/tests/test_BatchNorm.cpp @@ -119,7 +119,7 @@ TEST(Layer, batchNorm) { CHECK_EQ(static_cast(convLayer->getOutputValue()->getWidth()), 576); } -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU void batchNormInference(int n, int c, int h, int w) { MatrixPtr input = std::make_shared(n, c * h * w); MatrixPtr cudnnOut = std::make_shared(n, c * h * w); diff --git a/paddle/gserver/tests/test_ConvUnify.cpp b/paddle/gserver/tests/test_ConvUnify.cpp index e7325e0cc..16556469c 100644 --- a/paddle/gserver/tests/test_ConvUnify.cpp +++ b/paddle/gserver/tests/test_ConvUnify.cpp @@ -117,7 +117,7 @@ MatrixPtr doOneConvTest(size_t imgSize, } TEST(Layer, convParaUnified) { -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU MatrixPtr input, resultCpu, resultGpu; /// TEST1 for conv /// diff --git a/paddle/gserver/tests/test_DetectionOutput.cpp b/paddle/gserver/tests/test_DetectionOutput.cpp index af43dc51f..1a83f48fa 100644 --- a/paddle/gserver/tests/test_DetectionOutput.cpp +++ b/paddle/gserver/tests/test_DetectionOutput.cpp @@ -150,7 +150,7 @@ TEST(Layer, detectionOutputLayerFwd) { useGpu, result2); -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU // GPU case 1. useGpu = true; inputLoc = Matrix::create(1, 16, false, useGpu); diff --git a/paddle/gserver/tests/test_Evaluator.cpp b/paddle/gserver/tests/test_Evaluator.cpp index 93996392d..42bb57057 100644 --- a/paddle/gserver/tests/test_Evaluator.cpp +++ b/paddle/gserver/tests/test_Evaluator.cpp @@ -51,7 +51,7 @@ void testEvaluator(TestConfig testConf, string testEvaluatorName, size_t batchSize, bool useGpu) { -#ifdef PADDLE_ONLY_CPU +#ifndef PADDLE_WITH_GPU if (useGpu) return; #endif FLAGS_use_gpu = useGpu; diff --git a/paddle/gserver/tests/test_KmaxSeqScore.cpp b/paddle/gserver/tests/test_KmaxSeqScore.cpp index 308abe681..1594de850 100644 --- a/paddle/gserver/tests/test_KmaxSeqScore.cpp +++ b/paddle/gserver/tests/test_KmaxSeqScore.cpp @@ -97,7 +97,7 @@ TEST(Layer, kmaxSeqScoreLayer) { Matrix::create(subSeqStartPosition.back(), 1, false, false); std::vector mode = {false}; -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU mode.push_back(true); #endif diff --git a/paddle/gserver/tests/test_LayerGrad.cpp b/paddle/gserver/tests/test_LayerGrad.cpp index 090bde7b2..e887dee5f 100644 --- a/paddle/gserver/tests/test_LayerGrad.cpp +++ b/paddle/gserver/tests/test_LayerGrad.cpp @@ -12,7 +12,7 @@ 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. */ -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU #include #endif #include @@ -258,7 +258,7 @@ void testProjectionConv(size_t groups, bool isDeconv) { true); } -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU TEST(Projection, conv) { /// test ConvProjection testProjectionConv(1, false); @@ -422,7 +422,7 @@ TEST(Layer, depthwiseConvLayer) { // 'depthwise_conv' is a sepecial case of 'exconv' whose // groups size equals to the input channels size. testDepthwiseConvLayer("exconv", /* useGpu= */ false); -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU testDepthwiseConvLayer("exconv", /* useGpu= */ true); #endif } @@ -480,7 +480,7 @@ void testConvLayer(const string& type, bool trans, bool useGpu) { TEST(Layer, convLayer) { testConvLayer("exconv", /* trans= */ false, /* useGpu= */ false); -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU testConvLayer("exconv", /* trans= */ false, /* useGpu= */ true); testConvLayer("cudnn_conv", /* trans= */ false, /* useGpu= */ true); #endif @@ -525,7 +525,7 @@ TEST(Layer, convTransLayer) { for (auto useGpu : {false, true}) { testConvTransLayer("exconvt", /* trans= */ false, /* useGpu= */ useGpu); } -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU testConvTransLayer("cudnn_convt", /* trans= */ false, /* useGpu= */ true); #endif } @@ -638,7 +638,7 @@ TEST(Layer, SelectiveFullyConnectedLayer) { /* trans= */ false, /* useGup= */ false, false); -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU testLayerGrad(config, "selective_fc", 100, @@ -1210,7 +1210,7 @@ void testPoolLayer(const string& poolType, bool trans, bool useGpu) { testLayerGrad(config, "pool", 100, trans, useGpu); } -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU void testPoolLayer2(const string& poolType, bool trans, bool useGpu) { TestConfig config; config.inputDefs.push_back({INPUT_DATA, "layer_0", 3200, 0}); @@ -1236,7 +1236,7 @@ TEST(Layer, PoolLayer) { testPoolLayer("avg-projection", /* trans= */ false, /* useGpu= */ false); testPoolLayer("max-projection", /* trans= */ false, /* useGpu= */ false); -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU testPoolLayer("avg-projection", /* trans= */ false, /* useGpu= */ true); testPoolLayer("max-projection", /* trans= */ false, /* useGpu= */ true); testPoolLayer("cudnn-max-pool", /* trans= */ false, /* useGpu= */ true); @@ -1309,7 +1309,7 @@ void testPool3DLayer(const string& poolType, bool trans, bool useGpu) { TEST(Layer, Pool3DLayer) { testPool3DLayer("avg", /* trans= */ false, /* useGpu= */ false); testPool3DLayer("max", /* trans= */ false, /* useGpu= */ false); -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU testPool3DLayer("avg", /* trans= */ false, /* useGpu= */ true); testPool3DLayer("max", /* trans= */ false, /* useGpu= */ true); #endif @@ -1695,7 +1695,7 @@ void testBatchNormLayer(const string& type, bool trans, bool useGpu) { TEST(Layer, BatchNormalizationLayer) { testBatchNormLayer("batch_norm", false, false); -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU testBatchNormLayer("batch_norm", false, true); if (hl_get_cudnn_lib_version() >= int(4000)) { testBatchNormLayer("cudnn_batch_norm", false, true); @@ -1744,7 +1744,7 @@ void testBatchNorm3DLayer(const string& type, bool trans, bool useGpu) { TEST(Layer, testBatchNorm3DLayer) { testBatchNorm3DLayer("batch_norm", false, false); -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU testBatchNorm3DLayer("batch_norm", false, true); if (hl_get_cudnn_lib_version() >= int(4000)) { testBatchNorm3DLayer("cudnn_batch_norm", false, true); @@ -2262,7 +2262,7 @@ void test3DConvLayer(const string& type, bool trans, bool useGpu) { TEST(Layer, test3DConvLayer) { test3DConvLayer("conv3d", /* trans= */ false, /* useGpu= */ false); -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU test3DConvLayer("conv3d", /* trans= */ false, /* useGpu= */ true); #endif } @@ -2339,7 +2339,7 @@ void test3DDeConvLayer(const string& type, bool trans, bool useGpu) { TEST(Layer, test3DDeConvLayer) { test3DDeConvLayer("deconv3d", /* trans= */ false, /* useGpu= */ false); -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU test3DDeConvLayer("deconv3d", /* trans= */ false, /* useGpu= */ true); #endif } diff --git a/paddle/gserver/tests/test_NetworkCompare.cpp b/paddle/gserver/tests/test_NetworkCompare.cpp index d36f72360..e322fef9a 100644 --- a/paddle/gserver/tests/test_NetworkCompare.cpp +++ b/paddle/gserver/tests/test_NetworkCompare.cpp @@ -243,7 +243,7 @@ TEST(Compare, concat_slice) { compareNetwork(config_file_a, config_file_b); } -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU TEST(Compare, img_pool) { std::string config_file_a = "./gserver/tests/img_pool_a.conf"; std::string config_file_b = "./gserver/tests/img_pool_b.conf"; diff --git a/paddle/gserver/tests/test_PriorBox.cpp b/paddle/gserver/tests/test_PriorBox.cpp index ae0e3bc3d..cbc0fff7b 100644 --- a/paddle/gserver/tests/test_PriorBox.cpp +++ b/paddle/gserver/tests/test_PriorBox.cpp @@ -151,7 +151,7 @@ TEST(Layer, priorBoxLayerFwd) { useGpu, result); -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU // reset the input parameters variance[1] = 0.1; variance[3] = 0.2; diff --git a/paddle/gserver/tests/test_ProtoDataProvider.cpp b/paddle/gserver/tests/test_ProtoDataProvider.cpp index e11bf402c..988dbc251 100644 --- a/paddle/gserver/tests/test_ProtoDataProvider.cpp +++ b/paddle/gserver/tests/test_ProtoDataProvider.cpp @@ -485,7 +485,7 @@ TEST(ProtoDataProvider, test) { // Currently in async mode, useGpu is not supported continue; } -#ifdef PADDLE_ONLY_CPU +#ifndef PADDLE_WITH_GPU if (useGpu) { continue; } @@ -525,7 +525,7 @@ TEST(ProtoDataProvider, constant_slots) { for (int numConstantSlots : {1, 2}) { for (int useGpu : numTwoArray) { for (int dataCompression : numTwoArray) { -#ifdef PADDLE_ONLY_CPU +#ifndef PADDLE_WITH_GPU if (useGpu) { continue; } @@ -708,7 +708,7 @@ TEST(ProtoSequenceDataProvider, test) { // Currently in async mode, useGpu is not supported continue; } -#ifdef PADDLE_ONLY_CPU +#ifndef PADDLE_WITH_GPU if (useGpu) { continue; } diff --git a/paddle/gserver/tests/test_PyDataProvider.cpp b/paddle/gserver/tests/test_PyDataProvider.cpp index db883543c..f6522febf 100644 --- a/paddle/gserver/tests/test_PyDataProvider.cpp +++ b/paddle/gserver/tests/test_PyDataProvider.cpp @@ -37,7 +37,7 @@ TEST(PyDataProvider, py_fill_slots) { config.clear_files(); std::string dataFile = "gserver/tests/pyDataProvider/pyDataProviderList"; config.set_files(dataFile); -#ifdef PADDLE_ONLY_CPU +#ifndef PADDLE_WITH_GPU bool useGpu = false; #else bool useGpu = true; @@ -71,7 +71,7 @@ TEST(PyDataProvider, py_fill_nest_slots) { std::string dataFile = "gserver/tests/pyDataProvider/pyDataProviderList"; config.set_files(dataFile); EXPECT_EQ(config.IsInitialized(), true); -#ifdef PADDLE_ONLY_CPU +#ifndef PADDLE_WITH_GPU bool useGpu = false; #else bool useGpu = true; diff --git a/paddle/gserver/tests/test_SelectiveFCLayer.cpp b/paddle/gserver/tests/test_SelectiveFCLayer.cpp index ab23d00a2..b25d32fb2 100644 --- a/paddle/gserver/tests/test_SelectiveFCLayer.cpp +++ b/paddle/gserver/tests/test_SelectiveFCLayer.cpp @@ -321,7 +321,7 @@ TEST(Layer, SelectiveFcLayer_train_dense_mul) { "filelist=gserver/tests/SelectiveFcTest/dense_mul_list"; for (auto useGpu : {false, true}) { -#ifdef PADDLE_ONLY_CPU +#ifndef PADDLE_WITH_GPU if (useGpu) { break; } @@ -388,7 +388,7 @@ void testSelectiveFcLayerTrainSparseMul(const LayerConfig& config, outMatSelfc->getWidth(), outMatSelfc->getElementCnt())); cpuOutMatSelfc->copyFrom(*outMatSelfc, HPPL_STREAM_DEFAULT); -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU if (useGpu) { hl_stream_synchronize(HPPL_STREAM_DEFAULT); } @@ -418,7 +418,7 @@ void testSelectiveFcLayerTrainSparseMul(const LayerConfig& config, MatrixPtr cpuOutMatFc( new CpuMatrix(outMatFc->getHeight(), outMatFc->getWidth())); cpuOutMatFc->copyFrom(*outMatFc, HPPL_STREAM_DEFAULT); -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU if (useGpu) { hl_stream_synchronize(HPPL_STREAM_DEFAULT); } @@ -443,7 +443,7 @@ TEST(Layer, SelectiveFcLayer_train_sparse_mul) { selLayerConfig.set_size(fcLayerWidth); testSelectiveFcLayerTrainSparseMul(selLayerConfig, false); -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU testSelectiveFcLayerTrainSparseMul(selLayerConfig, true); #endif } diff --git a/paddle/gserver/tests/test_SeqSliceLayerGrad.cpp b/paddle/gserver/tests/test_SeqSliceLayerGrad.cpp index e1d4ae161..f28149081 100644 --- a/paddle/gserver/tests/test_SeqSliceLayerGrad.cpp +++ b/paddle/gserver/tests/test_SeqSliceLayerGrad.cpp @@ -195,7 +195,7 @@ TEST(Layer, SeqSliceLayer) { vector> ends; std::vector mode = {false}; -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU mode.push_back(true); #endif genSeqInfo(seqStartPos, subSeqStartPos); diff --git a/paddle/gserver/tests/test_WarpCTCLayer.cpp b/paddle/gserver/tests/test_WarpCTCLayer.cpp index 55427e2f1..ae5b64257 100644 --- a/paddle/gserver/tests/test_WarpCTCLayer.cpp +++ b/paddle/gserver/tests/test_WarpCTCLayer.cpp @@ -199,7 +199,7 @@ TEST(Layer, WarpCTCLayer) { for (auto batchSize : {1, 10, 32}) { for (auto normByTimes : {false, true}) { for (auto useGpu : {false, true}) { -#ifdef PADDLE_ONLY_CPU +#ifndef PADDLE_WITH_GPU if (useGpu) continue; #endif LOG(INFO) << "layerSize=" << layerSize << " batchSize=" << batchSize diff --git a/paddle/math/Matrix.cpp b/paddle/math/Matrix.cpp index 0023b4d0f..de02f9c0d 100644 --- a/paddle/math/Matrix.cpp +++ b/paddle/math/Matrix.cpp @@ -670,7 +670,7 @@ void GpuMatrix::leftMul(Matrix& a, real scaleAB, real scaleT) { } void GpuMatrix::selectRows(Matrix& table, IVector& ids) { -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU CHECK(dynamic_cast(&table)); CHECK(table.useGpu()); CHECK(ids.useGpu()); @@ -694,7 +694,7 @@ void GpuMatrix::selectRows(Matrix& table, IVector& ids) { } void GpuMatrix::addToRows(Matrix& table, IVector& ids) { -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU CHECK(dynamic_cast(&table)); CHECK(table.useGpu()); CHECK(ids.useGpu()); @@ -741,7 +741,7 @@ void GpuMatrix::rowMax(Matrix& max) { } void GpuMatrix::rowMax(IVector& maxIds, Matrix& maxVal) { -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU CHECK(maxIds.useGpu() && maxVal.useGpu()) << "Matrix type are not equal"; size_t numSamples = getHeight(); size_t beam = maxVal.getWidth(); diff --git a/paddle/math/SparseMatrix.cpp b/paddle/math/SparseMatrix.cpp index 6370c7738..1f31082ae 100644 --- a/paddle/math/SparseMatrix.cpp +++ b/paddle/math/SparseMatrix.cpp @@ -836,7 +836,7 @@ void GpuSparseMatrix::zeroMem() { } void GpuSparseMatrix::rowMax(IVector& maxIds, Matrix& maxVal) { -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU CHECK(maxIds.useGpu() && maxVal.useGpu()) << "Matrix type are not equal"; size_t numSamples = getHeight(); size_t beam = maxVal.getWidth(); diff --git a/paddle/math/Vector.cpp b/paddle/math/Vector.cpp index eb87ee9bb..54e57b255 100644 --- a/paddle/math/Vector.cpp +++ b/paddle/math/Vector.cpp @@ -172,7 +172,7 @@ void GpuVectorT::isEqualTo(const VectorT& b, const T& value) { template void GpuVectorT::selectFrom(const VectorT& src, const VectorT& ids) { -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU hl_vector_select_from(this->getData(), this->getSize(), src.getData(), @@ -850,7 +850,7 @@ CpuGpuVectorT::CpuGpuVectorT(CpuGpuVectorT& src, size_t size) : sync_(nullptr) { CHECK_LE(offset + size, static_cast(src.getSize())); -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU SyncedFlag* flag = src.getSync(); if (*flag == DATA_AT_CPU) { src.copyToGpu(); // will set synchronous data between CPU and GPU @@ -861,7 +861,7 @@ CpuGpuVectorT::CpuGpuVectorT(CpuGpuVectorT& src, auto cMemHandle = (src.getVector(false))->getMemoryHandle(); cpuVectorT_ = std::make_shared>( size, std::dynamic_pointer_cast(cMemHandle), offset); -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU auto gMemHandle = (src.getVector(true))->getMemoryHandle(); gpuVectorT_ = std::make_shared>( size, std::dynamic_pointer_cast(gMemHandle), offset); diff --git a/paddle/math/tests/test_Allocator.cpp b/paddle/math/tests/test_Allocator.cpp index 1ca70ea84..cf2f66aea 100644 --- a/paddle/math/tests/test_Allocator.cpp +++ b/paddle/math/tests/test_Allocator.cpp @@ -68,7 +68,7 @@ void testPoolAllocator() { TEST(Allocator, Pool) { testPoolAllocator(); -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU testPoolAllocator(); #endif } @@ -92,7 +92,7 @@ TEST(MemoryHandle, Cpu) { EXPECT_EQ(ptr1, ptr2); } -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU TEST(MemoryHandle, Gpu) { int numGpu = hl_get_device_count(); diff --git a/paddle/math/tests/test_BaseMatrix.cpp b/paddle/math/tests/test_BaseMatrix.cpp index 22ce39701..730759f3d 100644 --- a/paddle/math/tests/test_BaseMatrix.cpp +++ b/paddle/math/tests/test_BaseMatrix.cpp @@ -12,7 +12,7 @@ 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. */ -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU /** * This test file use autotest::AutoCompare and cmpWithoutArg to compares the * implementation of CPU and GPU member function in diff --git a/paddle/math/tests/test_CpuGpuVector.cpp b/paddle/math/tests/test_CpuGpuVector.cpp index 58bc43a38..ccb4a902b 100644 --- a/paddle/math/tests/test_CpuGpuVector.cpp +++ b/paddle/math/tests/test_CpuGpuVector.cpp @@ -12,7 +12,7 @@ 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. */ -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU #include #include "paddle/math/Vector.h" diff --git a/paddle/math/tests/test_ExecViaCpu.cpp b/paddle/math/tests/test_ExecViaCpu.cpp index 04c856453..2d439cd06 100644 --- a/paddle/math/tests/test_ExecViaCpu.cpp +++ b/paddle/math/tests/test_ExecViaCpu.cpp @@ -94,7 +94,7 @@ void testWrapper(F&& f) { } } -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU TEST(ExecViaCpu, test1) { testWrapper(f); testWrapper(&f); diff --git a/paddle/math/tests/test_GpuProfiler.cpp b/paddle/math/tests/test_GpuProfiler.cpp index e6b5dba44..6dab187e3 100644 --- a/paddle/math/tests/test_GpuProfiler.cpp +++ b/paddle/math/tests/test_GpuProfiler.cpp @@ -12,7 +12,7 @@ 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. */ -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU #include #include "paddle/math/Matrix.h" diff --git a/paddle/math/tests/test_Matrix.cpp b/paddle/math/tests/test_Matrix.cpp index 1c21da5b7..7a145eae6 100644 --- a/paddle/math/tests/test_Matrix.cpp +++ b/paddle/math/tests/test_Matrix.cpp @@ -12,7 +12,7 @@ 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. */ -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU /** * This test file use autotest::AutoCompare and cmpWithArg to compares the * implementation of CPU and GPU member function in Matrix.cpp. diff --git a/paddle/math/tests/test_SparseMatrix.cpp b/paddle/math/tests/test_SparseMatrix.cpp index c0572dfdb..8151dde10 100644 --- a/paddle/math/tests/test_SparseMatrix.cpp +++ b/paddle/math/tests/test_SparseMatrix.cpp @@ -47,7 +47,7 @@ struct MatrixPara { SparseFormat format; }; -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU void test_sparse_matrix_mul(MatrixPara paraA, MatrixPara paraB, MatrixPara paraC) { @@ -452,7 +452,7 @@ TEST(Matrix, SparseMatrixCSRFormatTrimFrom) { matB->trimFrom(*mat); checkSMatrixEqual2(matA, matB); -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU GpuSparseMatrixPtr matC = std::make_shared( height, trimedWidth, height, FLOAT_VALUE, SPARSE_CSR, true); matC->trimFrom(*mat); @@ -546,7 +546,7 @@ TEST(Matrix, SparseMatrixCSCFormatTrimFrom) { matB->trimFrom(*mat); checkSMatrixEqual2(matA, matB); -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU GpuSparseMatrixPtr matC = std::make_shared( height, trimedWidth, height, FLOAT_VALUE, SPARSE_CSC, true); matC->trimFrom(*mat); diff --git a/paddle/math/tests/test_Tensor.cu b/paddle/math/tests/test_Tensor.cu index 31b693afa..d03698dee 100644 --- a/paddle/math/tests/test_Tensor.cu +++ b/paddle/math/tests/test_Tensor.cu @@ -270,7 +270,7 @@ TEST(Unary, BaseOp) { TestUnaryVectorT testCpuIVector( testUnaryBaseOpInt); -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU TestUnaryMatrix testGpuMatrix(testUnaryBaseOp); TestUnaryVectorT testGpuVector(testUnaryBaseOp); TestUnaryVectorT testGpuIVector( @@ -317,7 +317,7 @@ void testUnayrMathOp(Tensor& A1, Tensor& A2) { TEST(Unary, MathOp) { TestUnaryMatrix testCpu(testUnayrMathOp); -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU TestUnaryMatrix testGpu(testUnayrMathOp); #endif } @@ -374,7 +374,7 @@ void testUnayrCompareOp(Tensor& A1, Tensor& A2) { TEST(Unary, CompareOp) { TestUnaryMatrix testCpu(testUnayrCompareOp); -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU TestUnaryMatrix testGpu(testUnayrCompareOp); #endif } @@ -536,7 +536,7 @@ void testBinaryBaseOp(Tensor& A1, Tensor& A2, Tensor& B) { TEST(Binary, BaseOp) { TestBinaryMatrix testCpu(testBinaryBaseOp); -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU TestBinaryMatrix testGpu(testBinaryBaseOp); #endif } @@ -710,7 +710,7 @@ void testBinaryMathOp(Tensor& A1, Tensor& A2, Tensor& B) { TEST(Binary, MathOp) { TestBinaryMatrix testCpu(testBinaryMathOp); -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU TestBinaryMatrix testGpu(testBinaryMathOp); #endif } @@ -810,7 +810,7 @@ void testBinaryCompareOp(Tensor& A1, Tensor& A2, Tensor& B) { TEST(Binary, CompareOp) { TestBinaryMatrix testCpu(testBinaryCompareOp); -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU TestBinaryMatrix testGpu(testBinaryCompareOp); #endif } @@ -955,7 +955,7 @@ void testTernaryBaseOp(Tensor& A1, Tensor& A2, Tensor& B, Tensor& C) { TEST(Ternary, BaseOp) { TestTernaryMatrix testCpu(testTernaryBaseOp); -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU TestTernaryMatrix testGpu(testTernaryBaseOp); #endif } @@ -1058,7 +1058,7 @@ void testTernaryCompareOp(Tensor& A1, Tensor& A2, Tensor& B, Tensor& C) { TEST(Ternary, CompareOp) { TestTernaryMatrix testCpu(testTernaryCompareOp); -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU TestTernaryMatrix testGpu(testTernaryCompareOp); #endif } @@ -1086,7 +1086,7 @@ void testQuaternaryAdd( TEST(Quaternary, BaseOp) { TestQuaternaryMatrix testCpu(testQuaternaryAdd); -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU TestQuaternaryMatrix testGpu(testQuaternaryAdd); #endif } @@ -1156,7 +1156,7 @@ void testQuaternaryCompareOp( TEST(Quaternary, CompareOp) { TestQuaternaryMatrix testCpu(testQuaternaryCompareOp); -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU TestQuaternaryMatrix testGpu(testQuaternaryCompareOp); #endif } diff --git a/paddle/math/tests/test_TrainingAlgorithm.cpp b/paddle/math/tests/test_TrainingAlgorithm.cpp index 4a88844b4..36ac02400 100644 --- a/paddle/math/tests/test_TrainingAlgorithm.cpp +++ b/paddle/math/tests/test_TrainingAlgorithm.cpp @@ -91,7 +91,7 @@ int VectorCheckErr(const VectorPtr& vector1, const VectorPtr& vector2) { typedef std::function testMatrixFunc; void testCase(testMatrixFunc matrixFunc) { -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU for (auto useGpu : {false, true}) { #else for (auto useGpu : {false}) { diff --git a/paddle/math/tests/test_batchTranspose.cpp b/paddle/math/tests/test_batchTranspose.cpp index 4eb983790..0189e534e 100644 --- a/paddle/math/tests/test_batchTranspose.cpp +++ b/paddle/math/tests/test_batchTranspose.cpp @@ -17,7 +17,7 @@ limitations under the License. */ using namespace paddle; // NOLINT -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU TEST(MatrixBatchTransTest, test_batch_matrix_transpose) { const int nx = 100; const int ny = 50; diff --git a/paddle/math/tests/test_lazyAssign.cu b/paddle/math/tests/test_lazyAssign.cu index 92afab4ff..04f23cff5 100644 --- a/paddle/math/tests/test_lazyAssign.cu +++ b/paddle/math/tests/test_lazyAssign.cu @@ -72,7 +72,7 @@ void testLazyAssign(int height, int width) { TEST(lazyAssign, CPU) { testMatrixCase(testLazyAssign); } -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU TEST(lazyAssign, GPU) { testMatrixCase(testLazyAssign); } #endif @@ -142,6 +142,6 @@ void testSgdUpdate(int height, int width) { TEST(sgdUpdate, CPU) { testMatrixCase(testSgdUpdate); } -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU TEST(sgdUpdate, GPU) { testMatrixCase(testSgdUpdate); } #endif diff --git a/paddle/math/tests/test_matrixCompare.cpp b/paddle/math/tests/test_matrixCompare.cpp index 061fb22e3..7735877ac 100644 --- a/paddle/math/tests/test_matrixCompare.cpp +++ b/paddle/math/tests/test_matrixCompare.cpp @@ -12,7 +12,7 @@ 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. */ -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU /// This unittest checks GpuMatrix/CpuMatrix get same result, so disable when /// only cpu version. diff --git a/paddle/math/tests/test_perturbation.cpp b/paddle/math/tests/test_perturbation.cpp index 60ebae015..dff18136a 100644 --- a/paddle/math/tests/test_perturbation.cpp +++ b/paddle/math/tests/test_perturbation.cpp @@ -12,7 +12,7 @@ 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. */ -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU #include #include diff --git a/paddle/math/tests/test_sparseMatrixCompare.cpp b/paddle/math/tests/test_sparseMatrixCompare.cpp index a9185a4b2..e39cc0a2f 100644 --- a/paddle/math/tests/test_sparseMatrixCompare.cpp +++ b/paddle/math/tests/test_sparseMatrixCompare.cpp @@ -12,7 +12,7 @@ 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. */ -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU /// This unittest checks GpuSparseMatrix/CpuSparseMatrix get same result, // so disable when /// only cpu version. diff --git a/paddle/memory/detail/buddy_allocator.cc b/paddle/memory/detail/buddy_allocator.cc index bb4497010..ed0c3374f 100644 --- a/paddle/memory/detail/buddy_allocator.cc +++ b/paddle/memory/detail/buddy_allocator.cc @@ -175,7 +175,7 @@ void* BuddyAllocator::SystemAlloc(size_t size) { } BuddyAllocator::PoolSet::iterator BuddyAllocator::RefillPool() { -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU if (system_allocator_->UseGpu()) { if ((total_used_ + total_free_) == 0) { // Compute the maximum allocation size for the first allocation. diff --git a/paddle/memory/detail/system_allocator.cc b/paddle/memory/detail/system_allocator.cc index a270bd595..64f8182b5 100644 --- a/paddle/memory/detail/system_allocator.cc +++ b/paddle/memory/detail/system_allocator.cc @@ -62,7 +62,7 @@ void CPUAllocator::Free(void* p, size_t size, size_t index) { bool CPUAllocator::UseGpu() const { return false; } -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU void* GPUAllocator::Alloc(size_t& index, size_t size) { // CUDA documentation doesn't explain if cudaMalloc returns nullptr diff --git a/paddle/memory/detail/system_allocator.h b/paddle/memory/detail/system_allocator.h index 82ba322e0..6b1f40347 100644 --- a/paddle/memory/detail/system_allocator.h +++ b/paddle/memory/detail/system_allocator.h @@ -40,7 +40,7 @@ class CPUAllocator : public SystemAllocator { virtual bool UseGpu() const; }; -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU class GPUAllocator : public SystemAllocator { public: virtual void* Alloc(size_t& index, size_t size); diff --git a/paddle/memory/detail/system_allocator_test.cc b/paddle/memory/detail/system_allocator_test.cc index ba44e06dd..57d5443d5 100644 --- a/paddle/memory/detail/system_allocator_test.cc +++ b/paddle/memory/detail/system_allocator_test.cc @@ -56,7 +56,7 @@ TEST(CPUAllocator, LockMem) { TestAllocator(a, 0); } -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU TEST(GPUAllocator, Alloc) { paddle::memory::detail::GPUAllocator a; TestAllocator(a, 2048); diff --git a/paddle/memory/memcpy.cc b/paddle/memory/memcpy.cc index c96a697a7..184d0f8fa 100644 --- a/paddle/memory/memcpy.cc +++ b/paddle/memory/memcpy.cc @@ -26,7 +26,7 @@ void Copy(platform::CPUPlace, void* dst, std::memcpy(dst, src, num); } -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU template <> void Copy(platform::CPUPlace dst_place, void* dst, diff --git a/paddle/memory/memcpy.h b/paddle/memory/memcpy.h index 2b9c0eada..7142831d4 100644 --- a/paddle/memory/memcpy.h +++ b/paddle/memory/memcpy.h @@ -33,7 +33,7 @@ namespace memory { template void Copy(DstPlace, void* dst, SrcPlace, const void* src, size_t num); -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU /** * \brief Copy memory from one place to another place. diff --git a/paddle/memory/memory.cc b/paddle/memory/memory.cc index 29bc26f9d..6d5a74daf 100644 --- a/paddle/memory/memory.cc +++ b/paddle/memory/memory.cc @@ -62,7 +62,7 @@ size_t Used(platform::CPUPlace place) { return GetCPUBuddyAllocator()->Used(); } -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU BuddyAllocator* GetGPUBuddyAllocator(int gpu_id) { using BuddyAllocVec = std::vector; diff --git a/paddle/memory/memory_test.cc b/paddle/memory/memory_test.cc index 53cc63a09..7a617f04d 100644 --- a/paddle/memory/memory_test.cc +++ b/paddle/memory/memory_test.cc @@ -80,7 +80,7 @@ TEST(BuddyAllocator, CPUMultAlloc) { } } -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU size_t align(size_t size, paddle::platform::GPUPlace place) { size += sizeof(paddle::memory::detail::Metadata); diff --git a/paddle/operators/detail/strided_memcpy.h b/paddle/operators/detail/strided_memcpy.h index b165224b3..9f05a2632 100644 --- a/paddle/operators/detail/strided_memcpy.h +++ b/paddle/operators/detail/strided_memcpy.h @@ -34,7 +34,7 @@ struct StridedMemcpyFunctor { auto& cpu_place = boost::get(place); memory::Copy(cpu_place, dst, cpu_place, src, sizeof(T) * dst_dim.head); } else { -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU auto& gpu_place = boost::get(place); auto& cuda_ctx = reinterpret_cast(dev_ctx); diff --git a/paddle/operators/math/im2col_test.cc b/paddle/operators/math/im2col_test.cc index f0b8c8859..3d040ca2b 100644 --- a/paddle/operators/math/im2col_test.cc +++ b/paddle/operators/math/im2col_test.cc @@ -71,7 +71,7 @@ void testIm2col() { context = new paddle::platform::CPUDeviceContext(paddle::platform::CPUPlace()); } else { -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU context = new paddle::platform::CUDADeviceContext(paddle::platform::GPUPlace()); #else @@ -116,7 +116,7 @@ void testIm2col() { TEST(math, im2col) { testIm2col(); -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU testIm2col(); #endif } diff --git a/paddle/operators/math/math_function_test.cc b/paddle/operators/math/math_function_test.cc index 22468a0c4..225226862 100644 --- a/paddle/operators/math/math_function_test.cc +++ b/paddle/operators/math/math_function_test.cc @@ -1,7 +1,7 @@ #include "paddle/operators/math/math_function.h" #include "gtest/gtest.h" -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU TEST(math_function, notrans_mul_trans) { paddle::framework::Tensor input1; paddle::framework::Tensor input1_gpu; diff --git a/paddle/operators/strided_memcpy_test.cc b/paddle/operators/strided_memcpy_test.cc index 05882a887..e0dd7b19f 100644 --- a/paddle/operators/strided_memcpy_test.cc +++ b/paddle/operators/strided_memcpy_test.cc @@ -72,7 +72,7 @@ TEST(StridedMemcpy, CPUConcat) { } } -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU TEST(StridedMemcpy, GPUCrop) { // clang-format off int src[] = { diff --git a/paddle/platform/device_context.cc b/paddle/platform/device_context.cc index 36af1ac67..8dcc357a1 100644 --- a/paddle/platform/device_context.cc +++ b/paddle/platform/device_context.cc @@ -35,7 +35,7 @@ Eigen::DefaultDevice* CPUDeviceContext::eigen_device() const { Place CPUDeviceContext::GetPlace() const { return CPUPlace(); } -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU template <> Eigen::GpuDevice* diff --git a/paddle/platform/device_context.h b/paddle/platform/device_context.h index d805d2ab0..c1c4c7f76 100644 --- a/paddle/platform/device_context.h +++ b/paddle/platform/device_context.h @@ -14,7 +14,7 @@ limitations under the License. */ #include "paddle/platform/enforce.h" #include "paddle/platform/place.h" -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU #include "paddle/platform/dynload/cublas.h" #include "paddle/platform/dynload/cudnn.h" #include "paddle/platform/gpu_info.h" @@ -61,7 +61,7 @@ class CPUDeviceContext : public DeviceContext { std::unique_ptr eigen_device_; }; -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU template <> struct EigenDeviceConverter { using EigenDeviceType = Eigen::GpuDevice; diff --git a/paddle/platform/enforce.h b/paddle/platform/enforce.h index 52bd23039..f9fe521d5 100644 --- a/paddle/platform/enforce.h +++ b/paddle/platform/enforce.h @@ -29,7 +29,7 @@ limitations under the License. */ #include // for __cxa_demangle #endif -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU #include "paddle/platform/dynload/cublas.h" #include "paddle/platform/dynload/cudnn.h" @@ -113,7 +113,7 @@ inline typename std::enable_if::type throw_on_error( } } -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU template inline typename std::enable_if::type throw_on_error( diff --git a/paddle/platform/gpu_info.h b/paddle/platform/gpu_info.h index f0c825bd9..ac884386d 100644 --- a/paddle/platform/gpu_info.h +++ b/paddle/platform/gpu_info.h @@ -14,7 +14,7 @@ limitations under the License. */ #pragma once -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU #include #include diff --git a/paddle/platform/variant.h b/paddle/platform/variant.h index 16ee00efe..8145799df 100644 --- a/paddle/platform/variant.h +++ b/paddle/platform/variant.h @@ -16,7 +16,7 @@ #include -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU // Because boost's variadic templates has bug on nvcc, boost will disable // variadic template support when GPU enabled on nvcc. diff --git a/paddle/pserver/test/SocketTest.cpp b/paddle/pserver/test/SocketTest.cpp index 6f6c9e596..96724530f 100644 --- a/paddle/pserver/test/SocketTest.cpp +++ b/paddle/pserver/test/SocketTest.cpp @@ -215,7 +215,7 @@ int main(int argc, char** argv) { uint64_t dataSize = FLAGS_dim * sizeof(real); -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU GpuVector gpuParam(FLAGS_dim); GpuVector gpuGrad(FLAGS_dim); #else diff --git a/paddle/pserver/test/test_ProtoServer.cpp b/paddle/pserver/test/test_ProtoServer.cpp index 04236fda2..74ab1f2f7 100644 --- a/paddle/pserver/test/test_ProtoServer.cpp +++ b/paddle/pserver/test/test_ProtoServer.cpp @@ -99,7 +99,7 @@ TEST(ProtoServer, regular) { } TEST(ProtoServer, extended) { -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU ProtoClient* client; if (FLAGS_rdma_tcp == "rdma") client = new ProtoClient(FLAGS_server_addr, FLAGS_port, F_RDMA); diff --git a/paddle/pybind/pybind.cc b/paddle/pybind/pybind.cc index d480427f5..761d82fc4 100644 --- a/paddle/pybind/pybind.cc +++ b/paddle/pybind/pybind.cc @@ -34,7 +34,7 @@ static size_t UniqueIntegerGenerator() { } bool IsCompileGPU() { -#ifdef PADDLE_ONLY_CPU +#ifndef PADDLE_WITH_GPU return false; #else return true; @@ -78,7 +78,7 @@ PYBIND11_PLUGIN(core) { .def("set", PyCPUTensorSetFromArray) .def("set", PyCPUTensorSetFromArray) .def("set", PyCPUTensorSetFromArray) -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU .def("set", PyCUDATensorSetFromArray) .def("set", PyCUDATensorSetFromArray) .def("set", PyCUDATensorSetFromArray) @@ -96,7 +96,7 @@ PYBIND11_PLUGIN(core) { .def( "__init__", [](LoDTensor &instance, const std::vector> &lod) { -#ifdef PADDLE_ONLY_CPU +#ifndef PADDLE_WITH_GPU new (&instance) LoDTensor(lod); #else LoD new_lod; @@ -107,7 +107,7 @@ PYBIND11_PLUGIN(core) { }) .def("set_lod", [](LoDTensor &self, const std::vector> &lod) { -#ifdef PADDLE_ONLY_CPU +#ifndef PADDLE_WITH_GPU self.set_lod(lod); #else LoD new_lod; @@ -117,7 +117,7 @@ PYBIND11_PLUGIN(core) { #endif }) .def("lod", [](LoDTensor &self) -> std::vector> { -#ifdef PADDLE_ONLY_CPU +#ifndef PADDLE_WITH_GPU return self.lod(); #else auto lod = self.lod(); @@ -203,7 +203,7 @@ All parameter, weight, gradient are variables in Paddle. .def_static("create", [](paddle::platform::GPUPlace& place) -> paddle::platform::DeviceContext* { -#ifdef PADDLE_ONLY_CPU +#ifndef PADDLE_WITH_GPU PADDLE_THROW("GPUPlace is not supported in CPU device."); #else return new paddle::platform::CUDADeviceContext(place); diff --git a/paddle/pybind/tensor_py.h b/paddle/pybind/tensor_py.h index 3e3e6bc03..62e85fa54 100644 --- a/paddle/pybind/tensor_py.h +++ b/paddle/pybind/tensor_py.h @@ -106,7 +106,7 @@ void PyCPUTensorSetFromArray( std::memcpy(dst, array.data(), sizeof(T) * array.size()); } -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU template void PyCUDATensorSetFromArray( framework::Tensor &self, diff --git a/paddle/trainer/MergeModel.cpp b/paddle/trainer/MergeModel.cpp index 91d89b61a..a37d53bc7 100644 --- a/paddle/trainer/MergeModel.cpp +++ b/paddle/trainer/MergeModel.cpp @@ -29,7 +29,7 @@ int main(int argc, char** argv) { initMain(argc, argv); initPython(argc, argv); string confFile = TrainerConfigHelper::getConfigNameFromPath(FLAGS_model_dir); -#ifdef PADDLE_ONLY_CPU +#ifndef PADDLE_WITH_GPU FLAGS_use_gpu = false; #endif auto config = std::make_shared(confFile); diff --git a/paddle/trainer/tests/test_Compare.cpp b/paddle/trainer/tests/test_Compare.cpp index e855a8fe2..b5d29da45 100644 --- a/paddle/trainer/tests/test_Compare.cpp +++ b/paddle/trainer/tests/test_Compare.cpp @@ -146,7 +146,7 @@ void compareGradient(comData& comDataCpu, comData& comDataGpu) { } int main(int argc, char** argv) { -#ifdef PADDLE_ONLY_CPU +#ifndef PADDLE_WITH_GPU exit(0); #endif paddle::initMain(argc, argv); diff --git a/paddle/trainer/tests/test_CompareSparse.cpp b/paddle/trainer/tests/test_CompareSparse.cpp index 813275518..4da9ce20f 100644 --- a/paddle/trainer/tests/test_CompareSparse.cpp +++ b/paddle/trainer/tests/test_CompareSparse.cpp @@ -174,7 +174,7 @@ TEST(compareSparse, multiGradientMachine) { FLAGS_local = local; FLAGS_ports_num_for_sparse = 5; for (bool useGpu : {false, true}) { -#ifdef PADDLE_ONLY_CPU +#ifndef PADDLE_WITH_GPU if (useGpu) continue; #endif FLAGS_parallel_nn = useGpu; @@ -198,7 +198,7 @@ TEST(compareSparse, NeuralNetwork) { FLAGS_local = local; FLAGS_ports_num_for_sparse = 5; for (bool useGpu : {false, true}) { -#ifdef PADDLE_ONLY_CPU +#ifndef PADDLE_WITH_GPU if (useGpu) continue; #endif FLAGS_parallel_nn = useGpu; diff --git a/paddle/trainer/tests/test_Trainer.cpp b/paddle/trainer/tests/test_Trainer.cpp index 264bc46eb..f69e1aafe 100644 --- a/paddle/trainer/tests/test_Trainer.cpp +++ b/paddle/trainer/tests/test_Trainer.cpp @@ -51,7 +51,7 @@ void checkGradientTest(const string& configFile, TEST(checkGradient, cpu) { checkGradientTest(configFile1, false, false); } -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU TEST(checkGradient, gpu) { checkGradientTest(configFile1, true, false); } TEST(checkGradient, multiGpu) { @@ -97,7 +97,7 @@ TEST(checkGradient, hsigmoid) { checkGradientTest(configFile2, false, false); } TEST(checkGradient, chunk) { checkGradientTest(configFile3, false, false); -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU checkGradientTest(configFile3, true, true); #endif } diff --git a/paddle/trainer/tests/test_TrainerOnePass.cpp b/paddle/trainer/tests/test_TrainerOnePass.cpp index 00ba61377..4c4d124fa 100644 --- a/paddle/trainer/tests/test_TrainerOnePass.cpp +++ b/paddle/trainer/tests/test_TrainerOnePass.cpp @@ -79,7 +79,7 @@ void trainerOnePassTest(const string& configFile, // 1. test trainer (cpu, gpu). TEST(trainerOnePass, cpu) { trainerOnePassTest(configFile1, false, false); } -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU TEST(trainerOnePass, gpu) { trainerOnePassTest(configFile1, true, false); } TEST(trainerOnePass, gpu2) { trainerOnePassTest(configFile1, true, false, 2); } @@ -94,7 +94,7 @@ TEST(trainerOnePass, parallel) { #endif // 2. test average_window. -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU TEST(average_window, gpu) { trainerOnePassTest(configFile1, true, false, 4, 0.01); } @@ -266,7 +266,7 @@ TEST(checkRemoteUpdater, cpuTrainerOldUpdater) { checkRemoteParameterUpdaterTest(configFile1, false, false, 1, true); } -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU TEST(checkRemoteUpdater, gpuTrainer) { checkRemoteParameterUpdaterTest(configFile1, true, false); } diff --git a/paddle/trainer/tests/test_recurrent_machine_generation.cpp b/paddle/trainer/tests/test_recurrent_machine_generation.cpp index 1322e7717..74b4fed7e 100644 --- a/paddle/trainer/tests/test_recurrent_machine_generation.cpp +++ b/paddle/trainer/tests/test_recurrent_machine_generation.cpp @@ -113,7 +113,7 @@ void testGeneration(const string& configFile, #ifndef PADDLE_TYPE_DOUBLE TEST(RecurrentGradientMachine, test_generation) { -#ifdef PADDLE_ONLY_CPU +#ifndef PADDLE_WITH_GPU const auto useGpuConfs = {false}; #else const auto useGpuConfs = {true, false}; diff --git a/paddle/utils/Flags.cpp b/paddle/utils/Flags.cpp index ab1c181c6..32155ded3 100644 --- a/paddle/utils/Flags.cpp +++ b/paddle/utils/Flags.cpp @@ -14,7 +14,7 @@ limitations under the License. */ #include "Flags.h" -#ifdef PADDLE_ONLY_CPU +#ifndef PADDLE_WITH_GPU DEFINE_bool(use_gpu, false, "Only support CPU training"); #else DEFINE_bool(use_gpu, true, "Whether to use GPU for training"); diff --git a/paddle/utils/Util.h b/paddle/utils/Util.h index 22ce2534d..904d0f506 100644 --- a/paddle/utils/Util.h +++ b/paddle/utils/Util.h @@ -218,7 +218,7 @@ protected: * *d2* is peer device to enable direct access to by the d1 device. */ inline void enablePeerAccess(int d1, int d2) { -#ifndef PADDLE_ONLY_CPU +#ifdef PADDLE_WITH_GPU if (hl_device_can_access_peer(d1, d2)) { SetDevice dev(d1); hl_device_enable_peer_access(d2); diff --git a/paddle/utils/Version.h b/paddle/utils/Version.h index f53d6420b..611fda83d 100644 --- a/paddle/utils/Version.h +++ b/paddle/utils/Version.h @@ -48,7 +48,7 @@ void printVersion(std::ostream& os); * @return return true if paddle compiled with GPU */ constexpr bool isWithGpu() { -#ifdef PADDLE_ONLY_CPU +#ifndef PADDLE_WITH_GPU return false; #else return true; -- GitLab