[XPU] improve error print format for some kernels (#72787)

cqulilujia · web-flow · commit 9fd40f9926be · 2025-05-20T12:12:30.000+08:00
diff --git a/paddle/phi/backends/gpu/cuda/gpu_event_timer.cc b/paddle/phi/backends/gpu/cuda/gpu_event_timer.cc
@@ -37,7 +37,7 @@ void EventPair::Start(cudaStream_t stream) {
   PADDLE_ENFORCE_EQ(
       started_,
       false,
-      phi::errors::InvalidArgument("GPUEventTimer has been started."));
+      common::errors::InvalidArgument("GPUEventTimer has been started."));
   PADDLE_ENFORCE_GPU_SUCCESS(cudaEventRecord(start_, stream));
   started_ = true;
 }
@@ -46,7 +46,7 @@ void EventPair::Stop(cudaStream_t stream) {
   PADDLE_ENFORCE_EQ(
       started_,
       true,
-      phi::errors::InvalidArgument("GPUEventTimer has not been started."));
+      common::errors::InvalidArgument("GPUEventTimer has not been started."));
   PADDLE_ENFORCE_GPU_SUCCESS(cudaEventRecord(end_, stream));
   started_ = false;
 }
@@ -57,7 +57,7 @@ double EventPair::Elapsed() {
   PADDLE_ENFORCE_EQ(
       started_,
       false,
-      phi::errors::InvalidArgument("GPUEventTimer has not been stopped."));
+      common::errors::InvalidArgument("GPUEventTimer has not been stopped."));
   PADDLE_ENFORCE_GPU_SUCCESS(cudaEventSynchronize(start_));
   PADDLE_ENFORCE_GPU_SUCCESS(cudaEventSynchronize(end_));
   float ms;
@@ -74,7 +74,7 @@ EventPair *GPUEventTimer::GetLatest() {
   PADDLE_ENFORCE_GT(
       length_,
       0,
-      phi::errors::InvalidArgument("GPUEventTimer has not been started."));
+      common::errors::InvalidArgument("GPUEventTimer has not been started."));
   auto &back = events_[length_ - 1];
   if (back == nullptr) {
     back.reset(new EventPair());
diff --git a/paddle/phi/infermeta/multiary.cc b/paddle/phi/infermeta/multiary.cc
@@ -486,14 +486,14 @@ void ApVariadicInferMeta(const std::vector<const MetaTensor*>& xs,
   PADDLE_ENFORCE_EQ(
       ret.HasError(),
       false,
-      phi::errors::Fatal(
+      common::errors::Fatal(
           "ApVariadicInferMeta failed. \nTraceback (most recent call "
           "last):\n%s\n%s: %s. ",
           ret.GetError().CallStackToString(),
           ret.GetError().class_name(),
           ret.GetError().msg()));
 #else
-  PADDLE_THROW(phi::errors::Unimplemented(
+  PADDLE_THROW(common::errors::Unimplemented(
       "ap_variadic is not implemented when cinn is not enabled."));
 #endif
 }
@@ -512,14 +512,14 @@ void ApFacadeInferMeta(
   const auto& ret = helper.InferMetaByAxprHook(
       xs, infer_meta_func_name, serialized_attributes, outs);
   PADDLE_ENFORCE(!ret.HasError(),
-                 phi::errors::Fatal(
+                 common::errors::Fatal(
                      "ApFacadeInferMeta failed. \nTraceback (most recent call "
                      "last):\n%s\n%s: %s. ",
                      ret.GetError().CallStackToString(),
                      ret.GetError().class_name(),
                      ret.GetError().msg()));
 #else
-  PADDLE_THROW(phi::errors::Unimplemented(
+  PADDLE_THROW(common::errors::Unimplemented(
       "ap_facade is not implemented when cinn is not enabled."));
 #endif
 }
diff --git a/paddle/phi/infermeta/spmd_rules/argsort.cc b/paddle/phi/infermeta/spmd_rules/argsort.cc
@@ -45,7 +45,7 @@ SpmdInfo ArgSortInferSpmd(const DistMetaTensor& x,
   PADDLE_ENFORCE_EQ(
       0 <= axis && axis < x_ndim,
       true,
-      phi::errors::InvalidArgument(
+      common::errors::InvalidArgument(
           "The axis of argsort should be in range [0, %d), but got %d.",
           x_ndim,
           axis));
@@ -139,7 +139,7 @@ SpmdInfo ArgSortGradInferSpmd(const DistMetaTensor& indices,
   PADDLE_ENFORCE_EQ(
       0 <= axis && axis < x_ndim,
       true,
-      phi::errors::InvalidArgument(
+      common::errors::InvalidArgument(
           "The axis of argsort should be in range [0, %d), but got %d.",
           x_ndim,
           axis));
diff --git a/paddle/phi/infermeta/spmd_rules/topk.cc b/paddle/phi/infermeta/spmd_rules/topk.cc
@@ -28,7 +28,7 @@ SpmdInfo TopkInferSpmd(
   PADDLE_ENFORCE_EQ(
       0 <= axis && axis < x_ndim,
       true,
-      phi::errors::InvalidArgument(
+      common::errors::InvalidArgument(
           "The axis of topk should be in range [0, %d), but got %d.",
           x_ndim,
           axis));
@@ -89,7 +89,7 @@ SpmdInfo TopkGradInferSpmd(const DistMetaTensor& x,
   PADDLE_ENFORCE_EQ(
       0 <= axis && axis < x_ndim,
       true,
-      phi::errors::InvalidArgument(
+      common::errors::InvalidArgument(
           "The axis of topk_grad should be in range [0, %d), but got %d.",
           x_ndim,
           axis));
diff --git a/paddle/phi/kernels/cpu/lu_solve_kernel.cc b/paddle/phi/kernels/cpu/lu_solve_kernel.cc
@@ -70,7 +70,7 @@ void LuSolveKernel(const Context& dev_ctx,
     PADDLE_ENFORCE_EQ(
         info,
         0,
-        phi::errors::PreconditionNotMet(
+        common::errors::PreconditionNotMet(
             "LU solve failed with error code %d. Check if matrix is singular.",
             info));
   }
diff --git a/paddle/phi/kernels/cpu/svdvals_kernel.cc b/paddle/phi/kernels/cpu/svdvals_kernel.cc
@@ -47,7 +47,7 @@ void LapackSvdvals(const T* X, T* S, int rows, int cols) {
                                                 nullptr,  // iwork is not needed
                                                 &info);
   if (info != 0) {
-    PADDLE_THROW(phi::errors::InvalidArgument(
+    PADDLE_THROW(common::errors::InvalidArgument(
         "Error during LAPACK lwork query. Invalid matrix or arguments."));
   }
   lwork = static_cast<int>(work[0]);
@@ -68,11 +68,11 @@ void LapackSvdvals(const T* X, T* S, int rows, int cols) {
                                                 nullptr,  // iwork is not needed
                                                 &info);
   if (info < 0) {
-    PADDLE_THROW(phi::errors::InvalidArgument(
+    PADDLE_THROW(common::errors::InvalidArgument(
         "This %s-th argument has an illegal value.", info));
   }
   if (info > 0) {
-    PADDLE_THROW(phi::errors::InvalidArgument(
+    PADDLE_THROW(common::errors::InvalidArgument(
         "SVD computation did not converge. Input matrix may be invalid."));
   }
 }
@@ -97,17 +97,17 @@ void SvdvalsKernel(const Context& dev_ctx,
   PADDLE_ENFORCE_GT(
       rows,
       0,
-      phi::errors::InvalidArgument("The row of Input(X) must be > 0."));
+      common::errors::InvalidArgument("The row of Input(X) must be > 0."));
   PADDLE_ENFORCE_GT(
       cols,
       0,
-      phi::errors::InvalidArgument("The column of Input(X) must be > 0."));
+      common::errors::InvalidArgument("The column of Input(X) must be > 0."));
   int k = std::min(rows, cols);
   int batches = static_cast<int>(X.numel() / (rows * cols));
-  PADDLE_ENFORCE_GT(
-      batches,
-      0,
-      phi::errors::InvalidArgument("The batch size of Input(X) must be > 0."));
+  PADDLE_ENFORCE_GT(batches,
+                    0,
+                    common::errors::InvalidArgument(
+                        "The batch size of Input(X) must be > 0."));
   DDim s_dims;
   if (x_dims.size() <= 2) {
     s_dims = {k};
diff --git a/paddle/phi/kernels/funcs/index_put_utils.h b/paddle/phi/kernels/funcs/index_put_utils.h
@@ -188,7 +188,7 @@ T** GetDevicePointerArray(const Context& ctx,
                           phi::Allocator::AllocationPtr* holder_ptr) {
   PADDLE_ENFORCE_NOT_NULL(
       holder_ptr,
-      phi::errors::InvalidArgument(
+      common::errors::InvalidArgument(
           "hold_ptr should be provided when calling GetDevicePointerArray."));
   std::vector<const T*> h_indices_v(indices_v.size());
   for (size_t i = 0; i < indices_v.size(); ++i) {
diff --git a/paddle/phi/kernels/gpu/ap_variadic_kernel.cu b/paddle/phi/kernels/gpu/ap_variadic_kernel.cu
@@ -28,27 +28,27 @@ void AllocateOutTensors(const Context& dev_ctx,
                         const std::vector<DenseTensor*>& outs) {
   for (auto* out : outs) {
     auto out_dtype = ap::axpr::GetDataTypeFromPhiDataType(out->dtype());
-    PADDLE_ENFORCE_EQ(
-        out_dtype.HasOkValue(),
-        true,
-        phi::errors::InvalidArgument("GetDataTypeFromPhiDataType() failed !"));
+    PADDLE_ENFORCE_EQ(out_dtype.HasOkValue(),
+                      true,
+                      common::errors::InvalidArgument(
+                          "GetDataTypeFromPhiDataType() failed !"));
 
     out_dtype.GetOkValue().Match(
         [&](const ap::axpr::CppDataType<ap::adt::Undefined>&) {
-          PADDLE_THROW(
-              phi::errors::InvalidArgument("allocate not support undefined !"));
+          PADDLE_THROW(common::errors::InvalidArgument(
+              "allocate not support undefined !"));
         },
         [&](const ap::axpr::CppDataType<uint64_t>&) {
-          PADDLE_THROW(
-              phi::errors::InvalidArgument("allocate not support uint64_t !"));
+          PADDLE_THROW(common::errors::InvalidArgument(
+              "allocate not support uint64_t !"));
         },
         [&](const ap::axpr::CppDataType<uint32_t>&) {
-          PADDLE_THROW(
-              phi::errors::InvalidArgument("allocate not support uint32_t !"));
+          PADDLE_THROW(common::errors::InvalidArgument(
+              "allocate not support uint32_t !"));
         },
         [&](const ap::axpr::CppDataType<uint16_t>&) {
-          PADDLE_THROW(
-              phi::errors::InvalidArgument("allocate not support uint16_t !"));
+          PADDLE_THROW(common::errors::InvalidArgument(
+              "allocate not support uint16_t !"));
         },
         [&](const auto& impl) {
           using tensor_type = typename std::decay_t<decltype(impl)>::type;
@@ -71,13 +71,13 @@ void ApVariadicKernel(const Context& dev_ctx,
   PADDLE_ENFORCE_GT(
       xs.size(),
       0,
-      phi::errors::InvalidArgument(
+      common::errors::InvalidArgument(
           "At least 1 input is required. current number out uts: // %d",
           xs.size()));
   PADDLE_ENFORCE_GT(
       outs.size(),
       0,
-      phi::errors::InvalidArgument(
+      common::errors::InvalidArgument(
           "num_outputs must be greater than 1. current _outputs: // %d",
           outs.size()));
 
@@ -94,13 +94,14 @@ void ApVariadicKernel(const Context& dev_ctx,
                                             kernel_dispatch_lambda,
                                             kernel_dispatch_const_data_lambda,
                                             outs);
-  PADDLE_ENFORCE_EQ(ret.HasError(),
-                    false,
-                    phi::errors::Fatal("ap_variadic failed. \nTraceback (most "
-                                       "recent call last):\n%s\n%s: %s. ",
-                                       ret.GetError().CallStackToString(),
-                                       ret.GetError().class_name(),
-                                       ret.GetError().msg()));
+  PADDLE_ENFORCE_EQ(
+      ret.HasError(),
+      false,
+      common::errors::Fatal("ap_variadic failed. \nTraceback (most "
+                            "recent call last):\n%s\n%s: %s. ",
+                            ret.GetError().CallStackToString(),
+                            ret.GetError().class_name(),
+                            ret.GetError().msg()));
 }
 
 }  // namespace phi
diff --git a/paddle/phi/kernels/gpu/lu_solve_kernle.cu b/paddle/phi/kernels/gpu/lu_solve_kernle.cu
@@ -133,7 +133,7 @@ void LuSolveKernel(const Context& dev_ctx,
   } else if (trans == "C") {
     trans_op = rocblas_operation_conjugate_transpose;
   } else {
-    PADDLE_THROW(phi::errors::InvalidArgument(
+    PADDLE_THROW(common::errors::InvalidArgument(
         "trans must be one of ['N', 'T', 'C'], but got %s", trans));
   }
 #else
@@ -145,7 +145,7 @@ void LuSolveKernel(const Context& dev_ctx,
   } else if (trans == "C") {
     trans_op = CUBLAS_OP_C;
   } else {
-    PADDLE_THROW(phi::errors::InvalidArgument(
+    PADDLE_THROW(common::errors::InvalidArgument(
         "trans must be one of ['N', 'T', 'C'], but got %s", trans));
   }
 #endif
diff --git a/paddle/phi/kernels/selected_rows/xpu/adam_kernel.cc b/paddle/phi/kernels/selected_rows/xpu/adam_kernel.cc
@@ -57,7 +57,7 @@ void AdamDenseParamSparseGradKernel(
   PADDLE_ENFORCE_NE(
       amsgrad,
       true,
-      phi::errors::Unimplemented("Operation amsgrad is not supported yet."));
+      common::errors::Unimplemented("Operation amsgrad is not supported yet."));
 
   using XPUType = typename XPUTypeTrait<T>::Type;
   xpu::ctx_guard RAII_GUARD(dev_ctx.x_context());
diff --git a/paddle/phi/kernels/xpu/adam_kernel.cc b/paddle/phi/kernels/xpu/adam_kernel.cc
@@ -56,7 +56,7 @@ void AdamDenseKernel(
   PADDLE_ENFORCE_NE(
       amsgrad,
       true,
-      phi::errors::Unimplemented("Operation amsgrad is not supported yet."));
+      common::errors::Unimplemented("Operation amsgrad is not supported yet."));
 
   xpu::ctx_guard RAII_GUARD(dev_ctx.x_context());
   float* param_ptr = nullptr;
@@ -291,7 +291,7 @@ void MergedAdamKernel(
   PADDLE_ENFORCE_NE(
       amsgrad,
       true,
-      phi::errors::Unimplemented("Operation amsgrad is not supported yet."));
+      common::errors::Unimplemented("Operation amsgrad is not supported yet."));
 
   VLOG(4) << "use_global_beta_pow:" << use_global_beta_pow;
 
diff --git a/paddle/phi/kernels/xpu/adamw_kernel.cc b/paddle/phi/kernels/xpu/adamw_kernel.cc
@@ -500,7 +500,7 @@ void AdamwDenseKernel(
   PADDLE_ENFORCE_NE(
       amsgrad,
       true,
-      phi::errors::Unimplemented("Operation amsgrad is not supported yet."));
+      common::errors::Unimplemented("Operation amsgrad is not supported yet."));
 
   auto dev_version =
       phi::backends::xpu::get_xpu_version(dev_ctx.GetPlace().GetDeviceId());
diff --git a/paddle/phi/kernels/xpu/gather_grad_kernel.cc b/paddle/phi/kernels/xpu/gather_grad_kernel.cc
@@ -83,8 +83,9 @@ void GatherGradKernel(const Context& dev_ctx,
         axis_v,
         false);
   } else {
-    PADDLE_THROW(common::errors::InvalidArgument("Unsupported index type: %s",
-                                                 DataTypeToString(index_type)));
+    PADDLE_THROW(common::errors::InvalidArgument(
+        "Unsupported index type, expected int32 or int64, but got type %s",
+        DataTypeToString(index_type)));
   }
   PADDLE_ENFORCE_XDNN_SUCCESS(r, "gather_grad");
 }
diff --git a/paddle/phi/kernels/xpu/gather_kernel.cc b/paddle/phi/kernels/xpu/gather_kernel.cc
@@ -77,8 +77,9 @@ void GatherKernel(const Context& dev_ctx,
         index.dims().size() == 0 ? 1 : index.dims()[0],
         axis_v);
   } else {
-    PADDLE_THROW(common::errors::InvalidArgument("Unsupported index type: %s",
-                                                 DataTypeToString(index_type)));
+    PADDLE_THROW(common::errors::InvalidArgument(
+        "Unsupported index type, expected int32 or int64, but got type %s",
+        DataTypeToString(index_type)));
   }
   PADDLE_ENFORCE_XDNN_SUCCESS(r, "paddle_gather");
 }
diff --git a/paddle/phi/kernels/xpu/p_recv_kernel.cc b/paddle/phi/kernels/xpu/p_recv_kernel.cc
@@ -51,8 +51,9 @@ void PRecvKernel(const Context& dev_ctx,
   dev_ctx.Alloc(out, dtype);
   comm_ctx->Recv(out, out->numel(), peer, stream);
 #else
-  PADDLE_THROW(errors::PreconditionNotMet(
-      "PaddlePaddle should compile with DWITH_XPU_BKCL."));
+  PADDLE_THROW(common::errors::PreconditionNotMet(
+      "PaddlePaddle is not compiled with DWITH_XPU_BKCL, please recompile with "
+      "DWITH_XPU_BKCL for using p_recv_kernel."));
 #endif
 }
 
@@ -77,8 +78,9 @@ void PRecvArrayKernel(const Context& dev_ctx,
             << common::product(out_dims) << " from " << peer;
   }
 #else
-  PADDLE_THROW(errors::PreconditionNotMet(
-      "PaddlePaddle should compile with DWITH_XPU_BKCL."));
+  PADDLE_THROW(common::errors::PreconditionNotMet(
+      "PaddlePaddle is not compiled with DWITH_XPU_BKCL, please recompile with "
+      "DWITH_XPU_BKCL for using p_recv_kernel."));
 #endif
 }
 
diff --git a/paddle/phi/kernels/xpu/p_send_kernel.cc b/paddle/phi/kernels/xpu/p_send_kernel.cc
@@ -43,8 +43,9 @@ void PSendKernel(const Context& dev_ctx,
   }
   comm_ctx->Send(x, x.numel(), peer, stream);
 #else
-  PADDLE_THROW(errors::PreconditionNotMet(
-      "PaddlePaddle should compile with DWITH_XPU_BKCL."));
+  PADDLE_THROW(common::errors::PreconditionNotMet(
+      "PaddlePaddle is not compiled with DWITH_XPU_BKCL, please recompile with "
+      "DWITH_XPU_BKCL for using p_send_kernel."));
 #endif
 }
 
@@ -65,8 +66,9 @@ void PSendArrayKernel(const Context& dev_ctx,
             << common::product(x.dims()) << " to " << peer;
   }
 #else
-  PADDLE_THROW(errors::PreconditionNotMet(
-      "PaddlePaddle should compile with DWITH_XPU_BKCL."));
+  PADDLE_THROW(common::errors::PreconditionNotMet(
+      "PaddlePaddle is not compiled with DWITH_XPU_BKCL, please recompile with "
+      "DWITH_XPU_BKCL for using p_send_kernel."));
 #endif
 }
 
diff --git a/paddle/phi/kernels/xpu/pool_kernel.cc b/paddle/phi/kernels/xpu/pool_kernel.cc
@@ -316,7 +316,9 @@ void MaxPool2dWithIndexKernel(const Context& ctx,
   PADDLE_ENFORCE_EQ(kernel_size.size(),
                     2,
                     common::errors::InvalidArgument(
-                        "The Pool2d XPU OP only support 2 dimension pooling!"));
+                        "The Pool2d XPU OP only support 2 dimension pooling, "
+                        "but received kernel_size with size %d",
+                        kernel_size.size()));
   PADDLE_ENFORCE_EQ(!adaptive || (kernel_size[0] * kernel_size[1] == 1),
                     true,
                     common::errors::InvalidArgument(
diff --git a/paddle/phi/kernels/xpu/stack_grad_kernel.cc b/paddle/phi/kernels/xpu/stack_grad_kernel.cc
@@ -62,7 +62,7 @@ void StackGradKernel(const Context& dev_ctx,
   PADDLE_ENFORCE_LE(
       needed_slices,
       n_slices,
-      phi::errors::InvalidArgument(
+      common::errors::InvalidArgument(
           "Number of valid slices (%ld) exceeds out_grad's dimension (%ld) "
           "along axis %d in stack_grad kernel. Mismatch between forward and "
           "backward shapes.",

Original file line number	Diff line number	Diff line change
`@@ -70,7 +70,7 @@ void LuSolveKernel(const Context& dev_ctx,`
`70`	`70`	`PADDLE_ENFORCE_EQ(`
`71`	`71`	`info,`
`72`	`72`	`0,`
`73`		`- phi::errors::PreconditionNotMet(`
	`73`	`+ common::errors::PreconditionNotMet(`
`74`	`74`	`"LU solve failed with error code %d. Check if matrix is singular.",`
`75`	`75`	`info));`
`76`	`76`	`}`