From f94d053705180635539aa07f03c1300bedb2bb92 Mon Sep 17 00:00:00 2001
From: Guanghua Yu <742925032@qq.com>
Date: Fri, 16 Oct 2020 14:19:11 +0800
Subject: [PATCH] error message optimization in
 mean_xpu,softmax_with_cross_entropy_op_xpu,test=kunlun (#27967)

---
 paddle/fluid/operators/mean_op_xpu.cc         | 14 +++++++---
 .../softmax_with_cross_entropy_op_xpu.cc      | 27 ++++++++++++-------
 2 files changed, 28 insertions(+), 13 deletions(-)

diff --git a/paddle/fluid/operators/mean_op_xpu.cc b/paddle/fluid/operators/mean_op_xpu.cc
index 3d95609b72..71bcc4be15 100644
--- a/paddle/fluid/operators/mean_op_xpu.cc
+++ b/paddle/fluid/operators/mean_op_xpu.cc
@@ -32,8 +32,11 @@ class MeanXPUKernel : public framework::OpKernel<T> {
     const float* x_data = input->data<float>();
     float* y_data = output->data<float>();
     int r = xpu::mean(dev_ctx.x_context(), x_data, y_data, input->numel());
-    PADDLE_ENFORCE_EQ(r, xpu::Error_t::SUCCESS,
-                      platform::errors::InvalidArgument("XPU kernel error!"));
+    PADDLE_ENFORCE_EQ(
+        r, xpu::Error_t::SUCCESS,
+        platform::errors::External(
+            "XPU kernel error, Mean op execution not succeed, error code=%d",
+            r));
   }
 };
 template <typename DeviceContext, typename T>
@@ -49,8 +52,11 @@ class MeanGradXPUKernel : public framework::OpKernel<T> {
     float* dx = IG->data<float>();
     const float* dy = OG->data<float>();
     int r = xpu::mean_grad(dev_ctx.x_context(), dx, dy, IG->numel());
-    PADDLE_ENFORCE_EQ(r, xpu::Error_t::SUCCESS,
-                      platform::errors::InvalidArgument("XPU kernel error!"));
+    PADDLE_ENFORCE_EQ(
+        r, xpu::Error_t::SUCCESS,
+        platform::errors::External(
+            "XPU kernel error. Mean_grad execution not succeed, error code=%d",
+            r));
   }
 };
 
diff --git a/paddle/fluid/operators/softmax_with_cross_entropy_op_xpu.cc b/paddle/fluid/operators/softmax_with_cross_entropy_op_xpu.cc
index 727b9ef812..f4f6eb9cdc 100644
--- a/paddle/fluid/operators/softmax_with_cross_entropy_op_xpu.cc
+++ b/paddle/fluid/operators/softmax_with_cross_entropy_op_xpu.cc
@@ -28,7 +28,7 @@ class SoftmaxWithCrossEntropyXPUKernel : public framework::OpKernel<T> {
   void Compute(const framework::ExecutionContext& context) const override {
     PADDLE_ENFORCE_EQ(
         platform::is_xpu_place(context.GetPlace()), true,
-        platform::errors::InvalidArgument("This kernel only runs on XPU."));
+        platform::errors::PreconditionNotMet("This kernel only runs on XPU."));
     const Tensor* logits = context.Input<Tensor>("Logits");
     const Tensor* labels = context.Input<Tensor>("Label");
     Tensor* softmax = context.Output<Tensor>("Softmax");
@@ -46,8 +46,11 @@ class SoftmaxWithCrossEntropyXPUKernel : public framework::OpKernel<T> {
         context.template device_context<platform::XPUDeviceContext>();
     int r = xpu::softmax2d_forward(dev_ctx.x_context(), logits->data<float>(),
                                    softmax->data<float>(), n, d);
-    PADDLE_ENFORCE_EQ(r, xpu::Error_t::SUCCESS,
-                      platform::errors::InvalidArgument("XPU kernel error!"));
+    PADDLE_ENFORCE_EQ(
+        r, xpu::Error_t::SUCCESS,
+        platform::errors::External("XPU kernel error. Softmax2d_forward "
+                                   "execution not succeed, error code=%d",
+                                   r));
     // cross_entropy
     auto ignore_index = context.Attr<int>("ignore_index");
     const bool soft_label = context.Attr<bool>("soft_label");
@@ -61,10 +64,13 @@ class SoftmaxWithCrossEntropyXPUKernel : public framework::OpKernel<T> {
       int* labels_int32_host =
           reinterpret_cast<int*>(std::malloc(n * sizeof(int)));
       int* labels_int32_device = NULL;
-      PADDLE_ENFORCE_EQ(
-          xpu_malloc(reinterpret_cast<void**>(&labels_int32_device),
-                     n * sizeof(int)),
-          XPU_SUCCESS, platform::errors::InvalidArgument("XPU kernel error!"));
+      int ret = xpu_malloc(reinterpret_cast<void**>(&labels_int32_device),
+                           n * sizeof(int));
+      PADDLE_ENFORCE_EQ(ret, XPU_SUCCESS,
+                        platform::errors::External(
+                            "XPU API return wrong value[%d], please check "
+                            "where Baidu Kunlun Card is properly installed.",
+                            ret));
       dev_ctx.Wait();
       memory::Copy(platform::CPUPlace(), labels_int64_host,
                    BOOST_GET_CONST(platform::XPUPlace, context.GetPlace()),
@@ -78,8 +84,11 @@ class SoftmaxWithCrossEntropyXPUKernel : public framework::OpKernel<T> {
       int r = xpu::cross_entropy_forward(
           dev_ctx.x_context(), n, d, softmax->data<float>(),
           labels_int32_device, loss->data<float>(), nullptr, ignore_index);
-      PADDLE_ENFORCE_EQ(r, xpu::Error_t::SUCCESS,
-                        platform::errors::InvalidArgument("XPU kernel error!"));
+      PADDLE_ENFORCE_EQ(
+          r, xpu::Error_t::SUCCESS,
+          platform::errors::External("XPU kernel error. Cross_entropy_forward "
+                                     "execution not succeed, error code=%d",
+                                     r));
       dev_ctx.Wait();
       std::free(labels_int32_host);
       std::free(labels_int64_host);
-- 
GitLab