improve performance of instance_norm, test=develop (#25005)

paddle/fluid/operators/instance_norm_op.cc

@@ -311,8 +311,8 @@ class InstanceNormGradKernel<platform::CPUDeviceContext, T>
     auto dy_arr = dy_e.reshape(shape);
     auto x_arr = x_e.reshape(shape);
 
-    auto tmp =
-        (x_arr - mean_arr.broadcast(bcast)) * inv_var_arr.broadcast(bcast);
+    auto tmp = (x_arr - mean_arr.eval().broadcast(bcast)) *
+               inv_var_arr.eval().broadcast(bcast);
 
     math::SetConstant<platform::CPUDeviceContext, T> set_constant;
     // math: d_bias = np.sum(d_y, axis=(n,h,w))
@@ -333,7 +333,8 @@ class InstanceNormGradKernel<platform::CPUDeviceContext, T>
           (tmp * dy_arr).sum(mean_rdims).reshape(param_shape).sum(rdims);
     }
 
-    auto dy_mean = dy_arr.mean(mean_rdims).reshape(NxC_shape).broadcast(bcast);
+    auto dy_mean =
+        dy_arr.mean(mean_rdims).reshape(NxC_shape).eval().broadcast(bcast);
 
     Eigen::DSizes<int, 2> bcast_param(N, sample_size);
     set_constant(dev_ctx, d_x, static_cast<T>(0));
@@ -351,6 +352,7 @@ class InstanceNormGradKernel<platform::CPUDeviceContext, T>
                                  (dy_arr * tmp)
                                      .mean(mean_rdims)
                                      .reshape(NxC_shape)
+                                     .eval()
                                      .broadcast(bcast));
   }
 };