Merge branch 'release/1.0.0' of https://github.com/PaddlePaddle/Paddle into release/1.0.0

Author: shippingwang

cmake/flags.cmake

@@ -138,13 +138,17 @@ if (APPLE)
         # On Mac OS X build fat binaries with x86_64 architectures by default.
         set (CMAKE_OSX_ARCHITECTURES "x86_64" CACHE STRING "Build architectures for OSX" FORCE)
     endif()
-else()
+    # On Mac OS X register class specifier is deprecated and will cause warning error on latest clang 10.0
+    set (COMMON_FLAGS -Wno-deprecated-register)
+endif(APPLE)
+
+if(LINUX)
     set(GPU_COMMON_FLAGS
         -Wall
         -Wextra
         -Werror
         ${GPU_COMMON_FLAGS})
-endif()
+endif(LINUX)
 
 if(UNIX AND NOT APPLE)
   # except apple from nix*Os family

paddle/fluid/framework/details/reference_count_pass.cc

@@ -80,15 +80,15 @@ std::unique_ptr<ir::Graph> ReferenceCountPass::ApplyImpl(
       // This is weird but there is really some variables without var_desc
       // in computation_op
       if (var_desc == nullptr) {
-        if (compute_op->Node()->Op()->Block()->FindVar(var_name) == nullptr)
-          continue;
-      } else {
-        if (var_desc->Persistable()) continue;
-        auto var_type = var_desc->Proto()->type().type();
-        if (var_type != proto::VarType::LOD_TENSOR &&
-            var_type != proto::VarType::SELECTED_ROWS) {
-          continue;
-        }
+        var_desc = compute_op->Node()->Op()->Block()->FindVar(var_name);
+        if (var_desc == nullptr) continue;
+      }
+
+      if (var_desc->Persistable()) continue;
+      auto var_type = var_desc->Proto()->type().type();
+      if (var_type != proto::VarType::LOD_TENSOR &&
+          var_type != proto::VarType::SELECTED_ROWS) {
+        continue;
       }
 
       // compute op only runs in one device

paddle/fluid/framework/parallel_executor.cc

@@ -319,6 +319,13 @@ void ParallelExecutor::Run(const std::vector<std::string> &fetch_tensors,
 #ifdef PADDLE_WITH_CUDA
   if (!gcs_.empty()) {
     ResetReferenceCount();
+    for (auto &pair : cur_ref_cnts_) {
+      auto &name_map = *(pair.second);
+      for (auto &fetch_name : fetch_tensors) {
+        name_map.erase(fetch_name);
+      }
+      name_map.erase(fetched_var_name);
+    }
   }
 #endif
   auto fetch_data = member_->executor_->Run(fetch_tensors);

paddle/fluid/operators/CMakeLists.txt

@@ -301,7 +301,6 @@ op_library(fusion_lstm_op DEPS cpu_lstm_compute)
 if (WITH_GPU)
     op_library(conv_op DEPS vol2col depthwise_conv im2col)
     op_library(layer_norm_op DEPS cub)
-    op_library(reduce_mean_op DEPS cub)
 else()
     op_library(conv_op DEPS vol2col im2col)
 endif()

paddle/fluid/operators/conv_op.h

@@ -380,8 +380,7 @@ class DepthwiseConvKernel : public framework::OpKernel<T> {
     math::DepthwiseConvFunctor<DeviceContext, T> depthwiseConv;
 
     auto& dev_ctx = context.template device_context<DeviceContext>();
-    depthwiseConv(dev_ctx, *input, filter, strides, paddings, dilations,
-                  output);
+    depthwiseConv(dev_ctx, *input, filter, strides, paddings, output);
   }
 };
 
@@ -416,14 +415,14 @@ class DepthwiseConvGradKernel : public framework::OpKernel<T> {
       input_grad->mutable_data<T>(context.GetPlace());
       set_zero(dev_ctx, input_grad, static_cast<T>(0));
       depthwiseConvInputGrad(dev_ctx, *input, filter, *output_grad, strides,
-                             paddings, dilations, input_grad);
+                             paddings, input_grad);
     }
 
     if (filter_grad) {
       filter_grad->mutable_data<T>(context.GetPlace());
       set_zero(dev_ctx, filter_grad, static_cast<T>(0));
       depthwiseConvFilterGrad(dev_ctx, *input, *output_grad, strides, paddings,
-                              dilations, filter_grad);
+                              filter_grad);
     }
   }
 };

paddle/fluid/operators/conv_transpose_op.h

@@ -345,7 +345,7 @@ class DepthwiseConvTransposeKernel : public framework::OpKernel<T> {
     math::DepthwiseConvInputGradFunctor<DeviceContext, T>
         depthwiseConvInputGrad;
     depthwiseConvInputGrad(dev_ctx, *output, filter, *input, strides, paddings,
-                           dilations, output);
+                           output);
   }
 };
 
@@ -367,11 +367,10 @@ class DepthwiseConvTransposeGradKernel : public framework::OpKernel<T> {
     auto& dev_ctx = context.template device_context<DeviceContext>();
     std::vector<int> strides = context.Attr<std::vector<int>>("strides");
     std::vector<int> paddings = context.Attr<std::vector<int>>("paddings");
-    std::vector<int> dilations = context.Attr<std::vector<int>>("dilations");
 
     if (input_grad) {
       math::DepthwiseConvFunctor<DeviceContext, T> depthwiseConv;
-      depthwiseConv(dev_ctx, *output_grad, filter, strides, paddings, dilations,
+      depthwiseConv(dev_ctx, *output_grad, filter, strides, paddings,
                     input_grad);
     }
 
@@ -383,7 +382,7 @@ class DepthwiseConvTransposeGradKernel : public framework::OpKernel<T> {
       math::DepthwiseConvFilterGradFunctor<DeviceContext, T>
           depthwiseConvFilterGrad;
       depthwiseConvFilterGrad(dev_ctx, *output_grad, *input, strides, paddings,
-                              dilations, filter_grad);
+                              filter_grad);
     }
   }
 };