PaddlePaddle
diff --git a/‎paddle/phi/api/yaml/legacy_ops.yaml
Copy file name to clipboard
-1Lines changed: 0 additions & 1 deletion b/‎paddle/phi/api/yaml/legacy_ops.yaml
Copy file name to clipboard
-1Lines changed: 0 additions & 1 deletion
diff --git a/‎paddle/phi/api/yaml/ops.yaml
Copy file name to clipboard
+30-15Lines changed: 30 additions & 15 deletions b/‎paddle/phi/api/yaml/ops.yaml
Copy file name to clipboard
+30-15Lines changed: 30 additions & 15 deletions
diff --git a/‎paddle/phi/kernels/funcs/activation_functor.h
Copy file name to clipboard
+17-11Lines changed: 17 additions & 11 deletions b/‎paddle/phi/kernels/funcs/activation_functor.h
Copy file name to clipboard
+17-11Lines changed: 17 additions & 11 deletions
@@ -237,7 +237,6 @@
     func : ElementwiseInferMeta
   kernel :
     func : elementwise_pow
-  inplace: (x -> out)
   backward : elementwise_pow_grad
 
 - op : embedding
 
@@ -13,6 +13,7 @@
   kernel :
     func : abs
     data_type : x
+  inplace: (x -> out)
   backward : abs_grad
 
 - op : accuracy
@@ -26,20 +27,22 @@
 
 - op : acos
   args : (Tensor x)
-  output : Tensor
+  output : Tensor(out)
   infer_meta :
     func : UnchangedInferMeta
   kernel :
     func : acos
+  inplace: (x -> out)
   backward : acos_grad
 
 - op : acosh
   args : (Tensor x)
-  output : Tensor
+  output : Tensor(out)
   infer_meta :
     func : UnchangedInferMeta
   kernel :
     func : acosh
+  inplace: (x -> out)
   backward : acosh_grad
 
 - op : adagrad_
@@ -90,12 +93,13 @@
 
 - op : addmm
   args : (Tensor input, Tensor x, Tensor y, float beta=1.0, float alpha=1.0)
-  output : Tensor
+  output : Tensor(out)
   infer_meta :
     func : AddmmInferMeta
   kernel :
     func : addmm
     data_type : x
+  inplace: (input -> out)
   backward : addmm_grad
 
 - op : affine_grid
@@ -176,34 +180,37 @@
 
 - op : asin
   args : (Tensor x)
-  output : Tensor
+  output : Tensor(out)
   infer_meta :
     func : UnchangedInferMeta
   kernel :
     func : asin
+  inplace: (x -> out)
   backward : asin_grad
 
 - op : asinh
   args : (Tensor x)
-  output : Tensor
+  output : Tensor(out)
   infer_meta :
     func : UnchangedInferMeta
   kernel :
     func : asinh
+  inplace: (x -> out)
   backward : asinh_grad
 
 - op : atan
   args : (Tensor x)
-  output : Tensor
+  output : Tensor(out)
   infer_meta :
     func : UnchangedInferMeta
   kernel :
     func : atan
+  inplace: (x -> out)
   backward : atan_grad
 
 - op : atan2
   args : (Tensor x, Tensor y)
-  output : Tensor
+  output : Tensor(out)
   infer_meta :
     func : Atan2InferMeta
   kernel :
@@ -212,11 +219,12 @@
 
 - op : atanh
   args : (Tensor x)
-  output : Tensor
+  output : Tensor(out)
   infer_meta :
     func : UnchangedInferMeta
   kernel :
     func : atanh
+  inplace: (x -> out)
   backward : atanh_grad
 
 - op : auc
@@ -524,20 +532,22 @@
 
 - op : cos
   args : (Tensor x)
-  output : Tensor
+  output : Tensor(out)
   infer_meta :
     func : UnchangedInferMeta
   kernel :
     func : cos
+  inplace: (x -> out)
   backward : cos_grad
 
 - op : cosh
   args : (Tensor x)
-  output : Tensor
+  output : Tensor(out)
   infer_meta :
     func : UnchangedInferMeta
   kernel :
     func : cosh
+  inplace: (x -> out)
   backward : cosh_grad
 
 - op : crop
@@ -756,11 +766,12 @@
 
 - op : erf
   args : (Tensor x)
-  output : Tensor
+  output : Tensor(out)
   infer_meta :
     func : UnchangedInferMeta
   kernel :
     func : erf
+  inplace : (x -> out)
   backward : erf_grad
 
 - op : erfinv
@@ -806,12 +817,13 @@
 
 - op : expm1
   args : (Tensor x)
-  output : Tensor
+  output : Tensor(out)
   infer_meta :
     func : UnchangedInferMeta
     param : [x]
   kernel :
     func : expm1
+  inplace: (x -> out)
   backward : expm1_grad
 
 - op : fft_c2c
@@ -2250,20 +2262,22 @@
 
 - op : sin
   args : (Tensor x)
-  output : Tensor
+  output : Tensor(out)
   infer_meta :
     func : UnchangedInferMeta
   kernel :
     func : sin
+  inplace: (x -> out)
   backward : sin_grad
 
 - op : sinh
   args : (Tensor x)
-  output : Tensor
+  output : Tensor(out)
   infer_meta :
     func : UnchangedInferMeta
   kernel :
     func : sinh
+  inplace: (x -> out)
   backward : sinh_grad
 
 - op : slogdet
@@ -2409,11 +2423,12 @@
 
 - op : tan
   args : (Tensor x)
-  output : Tensor
+  output : Tensor(out)
   infer_meta :
     func : UnchangedInferMeta
   kernel :
     func : tan
+  inplace: (x -> out)
   backward : tan_grad
 
 - op : tanh
 
@@ -116,7 +116,10 @@ template <typename T>
 struct SinFunctor : public BaseActivationFunctor<T> {
   template <typename Device, typename X, typename Out>
   void operator()(Device d, X x, Out out) const {
-    out.device(d) = x.unaryExpr(Sine<T>());
+    // Note(GGBond8488): Since Eigen3.3, Behavior like {A = (B * A).cwiseAbs()}
+    // will give wrong result, details see
+    // https://eigen.tuxfamily.org/dox/group__TopicAliasing.html
+    out.device(d) = x.unaryExpr(Sine<T>()).eval();
   }
 };
 
@@ -448,7 +451,7 @@ template <typename T>
 struct CosFunctor : public BaseActivationFunctor<T> {
   template <typename Device, typename X, typename Out>
   void operator()(Device d, X x, Out out) const {
-    out.device(d) = x.unaryExpr(Cosine<T>());
+    out.device(d) = x.unaryExpr(Cosine<T>()).eval();
   }
 };
 
@@ -762,7 +765,10 @@ template <typename T>
 struct TanFunctor : public BaseActivationFunctor<T> {
   template <typename Device, typename X, typename Out>
   void operator()(Device d, X x, Out out) const {
-    out.device(d) = x.unaryExpr(Tangent<T>());
+    // Note(GGBond8488): Since Eigen3.3, Behavior like {A = (B * A).cwiseAbs()}
+    // will give wrong result, details see
+    // https://eigen.tuxfamily.org/dox/group__TopicAliasing.html
+    out.device(d) = x.unaryExpr(Tangent<T>()).eval();
   }
 };
 
@@ -795,7 +801,7 @@ template <typename T>
 struct SinhFunctor : public BaseActivationFunctor<T> {
   template <typename Device, typename X, typename Out>
   void operator()(Device d, X x, Out out) const {
-    out.device(d) = x.unaryExpr(Sinh<T>());
+    out.device(d) = x.unaryExpr(Sinh<T>()).eval();
   }
 };
 
@@ -804,7 +810,7 @@ template <typename T>
 struct CoshFunctor : public BaseActivationFunctor<T> {
   template <typename Device, typename X, typename Out>
   void operator()(Device d, X x, Out out) const {
-    out.device(d) = x.unaryExpr(Cosh<T>());
+    out.device(d) = x.unaryExpr(Cosh<T>()).eval();
   }
 };
 
@@ -855,7 +861,7 @@ template <typename T>
 struct AcosFunctor : public BaseActivationFunctor<T> {
   template <typename Device, typename X, typename Out>
   void operator()(Device d, X x, Out out) const {
-    out.device(d) = x.unaryExpr(Acos<T>());
+    out.device(d) = x.unaryExpr(Acos<T>()).eval();
   }
 };
 
@@ -892,7 +898,7 @@ template <typename T>
 struct AsinFunctor : public BaseActivationFunctor<T> {
   template <typename Device, typename X, typename Out>
   void operator()(Device d, X x, Out out) const {
-    out.device(d) = x.unaryExpr(Asin<T>());
+    out.device(d) = x.unaryExpr(Asin<T>()).eval();
   }
 };
 
@@ -929,7 +935,7 @@ template <typename T>
 struct AtanFunctor : public BaseActivationFunctor<T> {
   template <typename Device, typename X, typename Out>
   void operator()(Device d, X x, Out out) const {
-    out.device(d) = x.unaryExpr(Atan<T>());
+    out.device(d) = x.unaryExpr(Atan<T>()).eval();
   }
 };
 
@@ -977,7 +983,7 @@ template <typename T>
 struct AcoshFunctor : public BaseActivationFunctor<T> {
   template <typename Device, typename X, typename Out>
   void operator()(Device d, X x, Out out) const {
-    out.device(d) = x.unaryExpr(Acosh<T>());
+    out.device(d) = x.unaryExpr(Acosh<T>()).eval();
   }
 };
 
@@ -1014,7 +1020,7 @@ template <typename T>
 struct AsinhFunctor : public BaseActivationFunctor<T> {
   template <typename Device, typename X, typename Out>
   void operator()(Device d, X x, Out out) const {
-    out.device(d) = x.unaryExpr(Asinh<T>());
+    out.device(d) = x.unaryExpr(Asinh<T>()).eval();
   }
 };
 
@@ -1051,7 +1057,7 @@ template <typename T>
 struct AtanhFunctor : public BaseActivationFunctor<T> {
   template <typename Device, typename X, typename Out>
   void operator()(Device d, X x, Out out) const {
-    out.device(d) = x.unaryExpr(Atanh<T>());
+    out.device(d) = x.unaryExpr(Atanh<T>()).eval();
   }
 };