Optimizer mmcpy if _rand_len=16 and remove data copy in GradKernel (#21099)

paddle/fluid/operators/pyramid_hash_op.cc

@@ -163,10 +163,14 @@ class CPUPyramidHashOPKernel : public framework::OpKernel<T> {
                          int _space_len) const {
     for (unsigned int j = 0; j != _num_emb; j += _rand_len) {
       unsigned int pos = XXH32(hash_id, len * sizeof(T), j) % _space_len;
+      if (_rand_len == 16) {
+        memcpy(top_pos + j, const_cast<float*>(weights + pos), 16 * sizeof(T));
+      } else {
         memcpy(top_pos + j, const_cast<float*>(weights + pos),
                _rand_len * sizeof(T));
       }
     }
+  }
 
   void Compute(const framework::ExecutionContext& ctx) const override {
     auto* bottom = ctx.Input<LoDTensor>("X");
@@ -322,6 +326,8 @@ class PyramidHashOpGrad : public framework::OperatorWithKernel {
     PADDLE_ENFORCE_EQ(ctx->HasInput("W"), true, "Input(W) should not be null.");
     PADDLE_ENFORCE_EQ(ctx->HasInput("DropPos"), true,
                       "Input(DropPos) should not be null.");
+    PADDLE_ENFORCE_EQ(ctx->HasInput("X_Temp_Out"), true,
+                      "Input(X_Temp_Out) should not be null.");
     PADDLE_ENFORCE_EQ(
         ctx->HasInput(framework::GradVarName("Out")), true,
         "Input(Out@GRAD) of PyramidHashGradOp should not be null.");
@@ -347,6 +353,7 @@ class PyramidHashGradOpMaker : public framework::SingleGradOpMaker<T> {
     op_desc_ptr->SetInput("X", this->Input("X"));
     op_desc_ptr->SetInput("W", this->Input("W"));
     op_desc_ptr->SetInput("DropPos", this->Output("DropPos"));
+    op_desc_ptr->SetInput("X_Temp_Out", this->Output("X_Temp_Out"));
 
     op_desc_ptr->SetInput(framework::GradVarName("Out"),
                           this->OutputGrad("Out"));
@@ -380,13 +387,8 @@ class CPUPyramidHashOPGradKernel : public framework::OpKernel<T> {
     int _space_len = ctx.Attr<int>("space_len");
     int _pyramid_layer = ctx.Attr<int>("pyramid_layer");
 
-    const auto* bottom_data_ori = bottom->data<int32_t>();
-    Tensor buff;
-    buff.Resize(framework::make_ddim({bottom->dims()[0], bottom->dims()[1]}));
-    T* bottom_data = buff.mutable_data<T>(ctx.GetPlace());
-    for (size_t i = 0; i < bottom->dims()[0]; i++) {
-      bottom_data[i] = bottom_data_ori[i];
-    }
+    auto* buff = ctx.Input<LoDTensor>("X_Temp_Out");
+    auto* bottom_data = buff->data<T>();
 
     int _slot_len = bottom->dims()[0];
     if (_slot_len == bottom->lod()[0].size() - 1 &&