From b36205e296bb0f2ec67bdd5df48b9c33d1903167 Mon Sep 17 00:00:00 2001
From: liaogang <liaogang@baidu.com>
Date: Wed, 2 Aug 2017 19:26:07 +0800
Subject: [PATCH] Refine compute code in operators

---
 paddle/framework/eigen.h      |  1 +
 paddle/operators/add_op.h     | 11 +++++++----
 paddle/operators/mean_op.h    |  7 +++++--
 paddle/operators/mul_op.h     | 13 +++++++++----
 paddle/operators/sgd_op.h     |  8 ++++++--
 paddle/operators/sigmoid_op.h |  8 +++++---
 6 files changed, 33 insertions(+), 15 deletions(-)

diff --git a/paddle/framework/eigen.h b/paddle/framework/eigen.h
index a4667cc51f..85006bb16e 100644
--- a/paddle/framework/eigen.h
+++ b/paddle/framework/eigen.h
@@ -28,6 +28,7 @@ struct EigenDim {
   static Type From(const DDim& dims) {
     PADDLE_ENFORCE(arity(dims) == D, "D must match arity(DDim)");
     Type ret;
+#pragma unroll
     for (int d = 0; d < arity(dims); d++) {
       ret[d] = dims[d];
     }
diff --git a/paddle/operators/add_op.h b/paddle/operators/add_op.h
index d2b649fcbd..2e17334860 100644
--- a/paddle/operators/add_op.h
+++ b/paddle/operators/add_op.h
@@ -28,10 +28,13 @@ public:
 
     output->mutable_data<T>(context.GetPlace());
 
-    EigenVector<T>::Flatten(*output).device(
-        *(context.GetEigenDevice<Place>())) =
-        framework::EigenVector<T>::Flatten(*input0) +
-        framework::EigenVector<T>::Flatten(*input1);
+    auto X = EigenVector<T>::Flatten(*input0);
+    auto Y = EigenVector<T>::Flatten(*input1);
+    auto Z = EigenVector<T>::Flatten(*output);
+
+    auto place = *context.GetEigenDevice<Place>();
+
+    Z.device(place) = X + Y;
   }
 };
 
diff --git a/paddle/operators/mean_op.h b/paddle/operators/mean_op.h
index 5f7d443751..658686c108 100644
--- a/paddle/operators/mean_op.h
+++ b/paddle/operators/mean_op.h
@@ -27,8 +27,11 @@ public:
 
     output->mutable_data<T>(context.GetPlace());
 
-    EigenScalar<T>::From(*output).device(*(context.GetEigenDevice<Place>())) =
-        EigenVector<T>::Flatten(*input).mean();
+    auto X = EigenVector<T>::Flatten(*input);
+    auto y = EigenScalar<T>::From(*output);
+    auto place = *context.GetEigenDevice<Place>();
+
+    y.device(place) = X.mean();
   }
 };
 
diff --git a/paddle/operators/mul_op.h b/paddle/operators/mul_op.h
index eef72ab293..60fa6bdc4a 100644
--- a/paddle/operators/mul_op.h
+++ b/paddle/operators/mul_op.h
@@ -26,13 +26,18 @@ public:
     Eigen::array<Eigen::IndexPair<Eigen::DenseIndex>, 1> dim_pair = {
         {Eigen::IndexPair<Eigen::DenseIndex>(1, 0)}};
 
+    auto input0 = context.Input<Tensor>("X");
+    auto input1 = context.Input<Tensor>("Y");
     auto output = context.Output<Tensor>(0);
+
     output->mutable_data<T>(context.GetPlace());
 
-    EigenMatrix<T>::From(*output).device(*(context.GetEigenDevice<Place>())) =
-        EigenMatrix<T>::From(*context.Input<Tensor>("X"))
-            .contract(EigenMatrix<T>::From(*context.Input<Tensor>("Y")),
-                      dim_pair);
+    auto X = EigenMatrix<T>::From(*input0);
+    auto Y = EigenMatrix<T>::From(*input1);
+    auto Z = EigenMatrix<T>::From(*output);
+    auto place = *context.GetEigenDevice<Place>();
+
+    Z.device(place) = X.contract(Y, dim_pair);
   }
 };
 }  // namespace operators
diff --git a/paddle/operators/sgd_op.h b/paddle/operators/sgd_op.h
index af1dfdd756..43681ab82a 100644
--- a/paddle/operators/sgd_op.h
+++ b/paddle/operators/sgd_op.h
@@ -29,8 +29,12 @@ public:
 
     param_out->mutable_data<T>(ctx.GetPlace());
 
-    EigenVector<T>::Flatten(*param_out).device(*(ctx.GetEigenDevice<Place>())) =
-        EigenVector<T>::Flatten(*param) - lr * EigenVector<T>::Flatten(*grad);
+    auto p = EigenVector<T>::Flatten(*param);
+    auto g = EigenVector<T>::Flatten(*grad);
+    auto o = EigenVector<T>::Flatten(*param_out);
+    auto place = *ctx.GetEigenDevice<Place>();
+
+    o.device(place) = p - lr * g;
   }
 };
 
diff --git a/paddle/operators/sigmoid_op.h b/paddle/operators/sigmoid_op.h
index 3dd23a9ebc..16272da789 100644
--- a/paddle/operators/sigmoid_op.h
+++ b/paddle/operators/sigmoid_op.h
@@ -27,9 +27,11 @@ public:
     auto output = context.Output<Tensor>(0);
     output->mutable_data<T>(context.GetPlace());
 
-    EigenVector<T>::Flatten(*output).device(
-        *(context.GetEigenDevice<Place>())) =
-        1.0 / (1.0 + (-1.0 * EigenVector<T>::Flatten(*input)).exp());
+    auto X = EigenVector<T>::Flatten(*input);
+    auto Y = EigenVector<T>::Flatten(*output);
+    auto place = *context.GetEigenDevice<Place>();
+
+    Y.device(place) = 1.0 / (1.0 + (-1.0 * X).exp());
   }
 };
 }  // namespace operators
-- 
GitLab