From f0c61017cc7d4c6465e84fff041fe8ae6afdbf12 Mon Sep 17 00:00:00 2001
From: tangwei12 <tangwei12@baidu.com>
Date: Thu, 14 May 2020 18:44:55 +0800
Subject: [PATCH] fix op error, test=develop (#24451) (#24539)

---
 .../operators/distributed_ops/recv_op.cc      |  4 ++-
 .../distributed_ops/ref_by_trainer_id_op.cc   | 25 +++++++++++++------
 .../distributed_ops/ref_by_trainer_id_op.h    |  5 +++-
 .../distributed_ops/send_barrier_op.cc        |  4 ++-
 .../operators/distributed_ops/send_op.cc      |  4 ++-
 5 files changed, 30 insertions(+), 12 deletions(-)

diff --git a/paddle/fluid/operators/distributed_ops/recv_op.cc b/paddle/fluid/operators/distributed_ops/recv_op.cc
index 30353ef35d7..aad9aefed4e 100644
--- a/paddle/fluid/operators/distributed_ops/recv_op.cc
+++ b/paddle/fluid/operators/distributed_ops/recv_op.cc
@@ -84,7 +84,9 @@ class RecvOp : public framework::OperatorBase {
       }
       for (size_t i = 0; i < rets.size(); i++) {
         VLOG(7) << "before sync_recv " << outs[i] << "from " << epmap[i];
-        PADDLE_ENFORCE_NE(rets[i]->Wait(), 0U, "internal error in RPCClient");
+        PADDLE_ENFORCE_NE(
+            rets[i]->Wait(), 0U,
+            platform::errors::ExecutionTimeout("internal error in RPCClient"));
         VLOG(7) << "after sync_recv " << outs[i] << "from " << epmap[i];
       }
     }
diff --git a/paddle/fluid/operators/distributed_ops/ref_by_trainer_id_op.cc b/paddle/fluid/operators/distributed_ops/ref_by_trainer_id_op.cc
index 6bf70844491..befdf4e9388 100644
--- a/paddle/fluid/operators/distributed_ops/ref_by_trainer_id_op.cc
+++ b/paddle/fluid/operators/distributed_ops/ref_by_trainer_id_op.cc
@@ -27,14 +27,23 @@ class RefByTrainerIdOp : public framework::OperatorWithKernel {
       : OperatorWithKernel(type, inputs, outputs, attrs) {}
 
   void InferShape(framework::InferShapeContext *ctx) const override {
-    PADDLE_ENFORCE(ctx->HasInputs("X"),
-                   "Input(X) of RefByTrainerIdOp should not be null.");
-    PADDLE_ENFORCE(ctx->HasInput("TrainerId"),
-                   "Input(TrainerId) of RefByTrainerIdOp should not be null.");
-    PADDLE_ENFORCE(ctx->HasOutput("Out"),
-                   "Output(Out) of RefByTrainerIdOp should not be null.");
-    PADDLE_ENFORCE_EQ(ctx->GetInputDim("TrainerId").size(), 1,
-                      "TrainerId should be a scalar.");
+    PADDLE_ENFORCE_EQ(ctx->HasInputs("X"), true,
+                      platform::errors::InvalidArgument(
+                          "Input(X) of RefByTrainerIdOp should not be null."));
+
+    PADDLE_ENFORCE_EQ(
+        ctx->HasInput("TrainerId"), true,
+        platform::errors::InvalidArgument(
+            "Input(TrainerId) of RefByTrainerIdOp should not be null."));
+
+    PADDLE_ENFORCE_EQ(
+        ctx->HasOutput("Out"), true,
+        platform::errors::InvalidArgument(
+            "Output(Out) of RefByTrainerIdOp should not be null."));
+
+    PADDLE_ENFORCE_EQ(
+        ctx->GetInputDim("TrainerId").size(), 1,
+        platform::errors::InvalidArgument("TrainerId should be a scalar."));
     // Out's shape is determined at runtime.
   }
 
diff --git a/paddle/fluid/operators/distributed_ops/ref_by_trainer_id_op.h b/paddle/fluid/operators/distributed_ops/ref_by_trainer_id_op.h
index 49dde44d6b6..d372674555a 100644
--- a/paddle/fluid/operators/distributed_ops/ref_by_trainer_id_op.h
+++ b/paddle/fluid/operators/distributed_ops/ref_by_trainer_id_op.h
@@ -38,7 +38,10 @@ class RefByTrainerIdKernel : public framework::OpKernel<T> {
     } else {
       trainer_id = *trainer_id_data;
     }
-    PADDLE_ENFORCE_LT((size_t)trainer_id, in_list.size());
+    PADDLE_ENFORCE_LT((size_t)trainer_id, in_list.size(),
+                      platform::errors::InvalidArgument(
+                          "X' size must >= TrainerId: [%s], but received [%s]",
+                          trainer_id, in_list.size()));
     out->mutable_data<T>(context.GetPlace());
     framework::TensorCopy(*(in_list[trainer_id]), in_list[trainer_id]->place(),
                           out);
diff --git a/paddle/fluid/operators/distributed_ops/send_barrier_op.cc b/paddle/fluid/operators/distributed_ops/send_barrier_op.cc
index 69c0726b20c..a8e9379d214 100644
--- a/paddle/fluid/operators/distributed_ops/send_barrier_op.cc
+++ b/paddle/fluid/operators/distributed_ops/send_barrier_op.cc
@@ -59,7 +59,9 @@ class SendBarrierOp : public framework::OperatorBase {
     }
 
     for (size_t i = 0; i < rets.size(); i++) {
-      PADDLE_ENFORCE_NE(rets[i]->Wait(), 0U, "internal error in RPCClient");
+      PADDLE_ENFORCE_NE(
+          rets[i]->Wait(), 0U,
+          platform::errors::ExecutionTimeout("internal error in RPCClient"));
     }
   }
 };
diff --git a/paddle/fluid/operators/distributed_ops/send_op.cc b/paddle/fluid/operators/distributed_ops/send_op.cc
index 1e9de787320..6d129a2140f 100644
--- a/paddle/fluid/operators/distributed_ops/send_op.cc
+++ b/paddle/fluid/operators/distributed_ops/send_op.cc
@@ -83,7 +83,9 @@ class SendOp : public framework::OperatorBase {
       }
       for (size_t i = 0; i < rets.size(); i++) {
         VLOG(7) << "before sync_send " << ins[i] << "from " << epmap[i];
-        PADDLE_ENFORCE_NE(rets[i]->Wait(), 0U, "internal error in RPCClient");
+        PADDLE_ENFORCE_NE(
+            rets[i]->Wait(), 0U,
+            platform::errors::ExecutionTimeout("internal error in RPCClient"));
         VLOG(7) << "after sync_send " << ins[i] << "from " << epmap[i];
       }
     }
-- 
GitLab