Merge branch 'develop' of https://github.com/PaddlePaddle/Paddle into custom_op_zyf

Author: zyfncg

paddle/fluid/operators/bce_loss_op_npu.cc

@@ -34,8 +34,9 @@ class BCELossNPUKernel : public framework::OpKernel<T> {
         ctx.template device_context<paddle::platform::NPUDeviceContext>()
             .stream();
 
-    const auto& runner = NpuOpRunner("BinaryCrossEntropy", {*x, *labels},
-                                     {*out}, {{"reduction", "none"}});
+    const auto& runner =
+        NpuOpRunner("BinaryCrossEntropy", {*x, *labels}, {*out},
+                    {{"reduction", static_cast<std::string>("none")}});
     runner.Run(stream);
   }
 };
@@ -57,7 +58,7 @@ class BCELossGradNPUKernel : public framework::OpKernel<T> {
 
     const auto& runner =
         NpuOpRunner("BinaryCrossEntropyGrad", {*x, *labels, *dout}, {*dx},
-                    {{"reduction", "none"}});
+                    {{"reduction", static_cast<std::string>("none")}});
     runner.Run(stream);
   }
 };

paddle/fluid/pybind/pybind.cc

@@ -2221,6 +2221,7 @@ All parameter, weight, gradient are variables in Paddle.
     auto &pool = platform::DeviceContextPool::Instance();
     auto devices = platform::GetSelectedNPUDevices();
     for (size_t i = 0; i < devices.size(); ++i) {
+      platform::NPUDeviceGuard guard(devices[i]);
       pool.Get(platform::NPUPlace(devices[i]))->Wait();
     }
     platform::AclInstance::Instance().Finalize();

paddle/scripts/paddle_build.bat

@@ -161,13 +161,23 @@ if "%WITH_SCCACHE%"=="ON" (
     del D:\sccache\sccache_log.txt
     cmd /C sccache -V || call :install_sccache
     sccache --stop-server 2> NUL
+
+    :: Localy storage on windows
     if not exist D:\sccache mkdir D:\sccache
     set SCCACHE_DIR=D:\sccache\.cache
-    :: sccache will shut down if a source file takes more than 10 mins to compile
+    
+    :: Sccache will shut down if a source file takes more than 10 mins to compile
     set SCCACHE_IDLE_TIMEOUT=0
-    set SCCACHE_CACHE_SIZE=30G
+    set SCCACHE_CACHE_SIZE=100G
     set SCCACHE_ERROR_LOG=D:\sccache\sccache_log.txt
     set SCCACHE_LOG=quiet
+
+    :: Distributed storage on windows
+    set SCCACHE_ENDPOINT=s3.bj.bcebos.com
+    set SCCACHE_BUCKET=paddle-windows
+    set SCCACHE_S3_KEY_PREFIX=sccache/
+    set SCCACHE_S3_USE_SSL=true
+
     sccache --start-server
     sccache -z
     goto :CASE_%1

python/paddle/fluid/tests/unittests/npu/test_bce_loss_npu.py

@@ -96,7 +96,7 @@ def test_dygraph_layer(place,
                        label_np,
                        reduction='mean',
                        weight_np=None):
-    paddle.disable_static()
+    paddle.disable_static(place)
     if weight_np is not None:
         weight = paddle.to_tensor(weight_np)
         bce_loss = paddle.nn.loss.BCELoss(weight=weight, reduction=reduction)
@@ -113,7 +113,7 @@ def test_dygraph_functional(place,
                             label_np,
                             reduction='mean',
                             weight_np=None):
-    paddle.disable_static()
+    paddle.disable_static(place)
     input = paddle.to_tensor(input_np)
     label = paddle.to_tensor(label_np)