[cherry-pick] polish kunlun error message for 2.0 rc (#28048)

* polish error message,test=kunlun

* polish error,test=kunlun

* polish error,test=kunlun

* polish error,test=kunlun

paddle/fluid/operators/conv_op_xpu.cc

@@ -50,11 +50,17 @@ class GemmConvXPUKernel : public framework::OpKernel<T> {
     PADDLE_ENFORCE_EQ(
         xpu::findmax(dev_ctx.x_context(), input->data<T>(), input->numel(),
                      max_input->data<T>()) == xpu::Error_t::SUCCESS,
-        true, platform::errors::InvalidArgument("XPU kernel error!"));
+        true, platform::errors::InvalidArgument(
+                  "XPU conv kernel error,can not finde max_input,please "
+                  "check whether Baidu Kunlun "
+                  "Card is properly installed."));
     PADDLE_ENFORCE_EQ(
         xpu::findmax(dev_ctx.x_context(), filter.data<T>(), filter.numel(),
                      max_filter->data<T>()) == xpu::Error_t::SUCCESS,
-        true, platform::errors::InvalidArgument("XPU kernel error!"));
+        true, platform::errors::InvalidArgument(
+                  "XPU conv kernel error,can not find max_filter,please "
+                  "check whether Baidu Kunlun "
+                  "Card is properly installed."));
     if (groups == 1) {
       int r = xpu::conv2d_forward_int16<float, float, float, float>(
           dev_ctx.x_context(), batch_size, img_c, img_h, img_w, f, win_h, win_w,
@@ -63,8 +69,12 @@ class GemmConvXPUKernel : public framework::OpKernel<T> {
           output->data<float>(), nullptr, nullptr, xpu::Activation_t::LINEAR,
           // nullptr, nullptr);
           max_input->data<float>(), max_filter->data<float>());
-      PADDLE_ENFORCE_EQ(r == xpu::Error_t::SUCCESS, true,
-                        platform::errors::InvalidArgument("XPU kernel error!"));
+      PADDLE_ENFORCE_EQ(
+          r, XPU_SUCCESS,
+          platform::errors::External("XPU conv kernel return wrong value[%d], "
+                                     "please check whether Baidu Kunlun Card "
+                                     "is properly installed.",
+                                     r));
     } else {
       int r = xpu::conv2d_int16_with_group<float, float, float>(
           dev_ctx.x_context(), input->data<float>(), filter.data<float>(),
@@ -72,8 +82,12 @@ class GemmConvXPUKernel : public framework::OpKernel<T> {
           win_w, groups, strides[0], strides[1], paddings[0], paddings[1],
           // nullptr, nullptr);
           max_input->data<float>(), max_filter->data<float>());
-      PADDLE_ENFORCE_EQ(r == xpu::Error_t::SUCCESS, true,
-                        platform::errors::InvalidArgument("XPU kernel error!"));
+      PADDLE_ENFORCE_EQ(
+          r, XPU_SUCCESS,
+          platform::errors::External("XPU conv kernel return wrong value[%d], "
+                                     "please check whether Baidu Kunlun Card "
+                                     "is properly installed.",
+                                     r));
     }
   }
 };
@@ -125,7 +139,11 @@ class GemmConvGradXPUKernel : public framework::OpKernel<T> {
         xpu::findmax(dev_ctx.x_context(), output_grad->data<T>(),
                      output_grad->numel(),
                      max_output_grad->data<T>()) == xpu::Error_t::SUCCESS,
-        true, platform::errors::InvalidArgument("XPU kernel error!"));
+        true,
+        platform::errors::External(
+            "XPU conv kernel error, can not find max_output_grad, please check "
+            "whether Baidu Kunlun Card is "
+            "properly installed."));
     if (input_grad) {
       int r = xpu::conv2d_backward_int16(
           dev_ctx.x_context(), batch_size, img_c, img_h, img_w, f, win_h, win_w,
@@ -134,8 +152,12 @@ class GemmConvGradXPUKernel : public framework::OpKernel<T> {
           filter.data<float>(), input_grad->data<float>(),
           // nullptr, nullptr,
           max_output_grad->data<float>(), max_filter->data<float>());
-      PADDLE_ENFORCE_EQ(r == xpu::Error_t::SUCCESS, true,
-                        platform::errors::InvalidArgument("XPU kernel error!"));
+      PADDLE_ENFORCE_EQ(
+          r, XPU_SUCCESS,
+          platform::errors::External("XPU conv kernel return wrong value[%d], "
+                                     "please check whether Baidu Kunlun Card "
+                                     "is properly installed.",
+                                     r));
     }
     if (filter_grad) {
       int r = xpu::conv2d_backward_weight_int16(
@@ -145,8 +167,12 @@ class GemmConvGradXPUKernel : public framework::OpKernel<T> {
           input->data<float>(), filter_grad->data<float>(),
           // nullptr, nullptr,
           max_output_grad->data<float>(), max_input->data<float>());
-      PADDLE_ENFORCE_EQ(r == xpu::Error_t::SUCCESS, true,
-                        platform::errors::InvalidArgument("XPU kernel error!"));
+      PADDLE_ENFORCE_EQ(
+          r, XPU_SUCCESS,
+          platform::errors::External("XPU conv kernel return wrong value[%d], "
+                                     "please check whether Baidu Kunlun Card "
+                                     "is properly installed.",
+                                     r));
     }
   }
 };

paddle/fluid/operators/dropout_op_xpu.cc

@@ -64,10 +64,16 @@ class DropoutXPUKernel : public framework::OpKernel<T> {
             }
           }
         }
-        PADDLE_ENFORCE(
+        PADDLE_ENFORCE_EQ(
             xpu_malloc(reinterpret_cast<void**>(&mask_data_table),
-                       max_data_size * sizeof(float)) == xpu::Error_t::SUCCESS,
-            "XPU no enough memory");
+                       max_data_size * sizeof(float)),
+            XPU_SUCCESS,
+            platform::errors::ResourceExhausted(
+                "\n\nOut of memory error on XPU, Cannot"
+                "allocate %s memory on XPU. \n\nPlease "
+                "check whether there is any other process "
+                "using XPU.\n",
+                string::HumanReadableSize(max_data_size * sizeof(void*))));
         memory::Copy(BOOST_GET_CONST(platform::XPUPlace, context.GetPlace()),
                      mask_data_table, platform::CPUPlace(), mask_data_host,
                      max_data_size * sizeof(float));
@@ -84,8 +90,12 @@ class DropoutXPUKernel : public framework::OpKernel<T> {
       auto& dev_ctx = context.template device_context<DeviceContext>();
       int r = xpu::dropout(dev_ctx.x_context(), mask_data_table, x_data,
                            mask_data, y_data, max_data_size, size);
-      PADDLE_ENFORCE_EQ(r == xpu::Error_t::SUCCESS, true,
-                        platform::errors::InvalidArgument("XPU kernel error!"));
+      PADDLE_ENFORCE_EQ(
+          r, xpu::Error_t::SUCCESS,
+          platform::errors::External(
+              "XPU dropout return wrong value[%d], please check whether "
+              "Baidu Kunlun Card is properly installed.",
+              r));
     } else {  // Infer
       float scale = 0.0f;
       if (dropout_implementation == "upscale_in_train") {
@@ -96,8 +106,12 @@ class DropoutXPUKernel : public framework::OpKernel<T> {
       auto& dev_ctx = context.template device_context<DeviceContext>();
       int r = xpu::scale(dev_ctx.x_context(), x->numel(), scale, 0.0f, 0,
                          x_data, y_data);
-      PADDLE_ENFORCE_EQ(r == xpu::Error_t::SUCCESS, true,
-                        platform::errors::InvalidArgument("XPU kernel error!"));
+      PADDLE_ENFORCE_EQ(
+          r, xpu::Error_t::SUCCESS,
+          platform::errors::External(
+              "XPU dropout return wrong value[%d], please check whether "
+              "Baidu Kunlun Card is properly installed.",
+              r));
     }
   }
 };
@@ -105,8 +119,9 @@ template <typename DeviceContext, typename T>
 class DropoutGradXPUKernel : public framework::OpKernel<T> {
  public:
   void Compute(const framework::ExecutionContext& context) const override {
-    PADDLE_ENFORCE(!context.Attr<bool>("is_test"),
-                   "GradOp is only callable when is_test is false");
+    PADDLE_ENFORCE_EQ(!context.Attr<bool>("is_test"), true,
+                      platform::errors::InvalidArgument(
+                          "GradOp is only callable when is_test is false"));
     auto* grad_x = context.Output<Tensor>(framework::GradVarName("X"));
     auto* grad_y = context.Input<Tensor>(framework::GradVarName("Out"));
     auto* mask = context.Input<Tensor>("Mask");
@@ -115,8 +130,12 @@ class DropoutGradXPUKernel : public framework::OpKernel<T> {
     int r = xpu::elementwise_mul(dev_ctx.x_context(), grad_y->data<T>(),
                                  mask->data<T>(), grad_x->data<T>(),
                                  grad_y->numel());
-    PADDLE_ENFORCE_EQ(r == xpu::Error_t::SUCCESS, true,
-                      platform::errors::InvalidArgument("XPU kernel error!"));
+    PADDLE_ENFORCE_EQ(
+        r, xpu::Error_t::SUCCESS,
+        platform::errors::External(
+            "XPU dropout return wrong value[%d], please check whether "
+            "Baidu Kunlun Card is properly installed.",
+            r));
   }
 };
 }  // namespace operators

paddle/fluid/operators/log_loss_op_xpu.cc

@@ -29,8 +29,12 @@ class LogLossXPUKernel : public framework::OpKernel<T> {
     int r =
         xpu::log_loss_fwd(dev_ctx.x_context(), n, epsilon, predict->data<T>(),
                           labels->data<T>(), loss->data<T>());
-    PADDLE_ENFORCE_EQ(r == xpu::Error_t::SUCCESS, true,
-                      platform::errors::InvalidArgument("XPU kernel error!"));
+    PADDLE_ENFORCE_EQ(
+        r, xpu::Error_t::SUCCESS,
+        platform::errors::External(
+            "XPU log_loss kernel return wrong value[%d], please check whether "
+            "Baidu Kunlun Card is properly installed.",
+            r));
   }
 };
 template <typename DeviceContext, typename T, typename AttrType = T>
@@ -51,8 +55,12 @@ class LogLossGradXPUKernel : public framework::OpKernel<T> {
     int r = xpu::log_loss_bwd(dev_ctx.x_context(), n, epsilon,
                               predict->data<T>(), labels->data<T>(),
                               dloss->data<T>(), dpred->data<T>());
-    PADDLE_ENFORCE_EQ(r == xpu::Error_t::SUCCESS, true,
-                      platform::errors::InvalidArgument("XPU kernel error!"));
+    PADDLE_ENFORCE_EQ(
+        r, xpu::Error_t::SUCCESS,
+        platform::errors::External(
+            "XPU log_loss kernel return wrong value[%d], please check whether "
+            "Baidu Kunlun Card is properly installed.",
+            r));
   }
 };