Report error position.

dmlc · Oct 19, 2020 · b12174d · b12174d
1 parent 3708601
commit b12174d
Show file tree

Hide file tree

Showing 2 changed files with 20 additions and 12 deletions.
diff --git a/rabit/src/allreduce_base.cc b/rabit/src/allreduce_base.cc
@@ -217,7 +217,7 @@ void AllreduceBase::SetParam(const char *name, const char *val) {
       rabit_enable_tcp_no_delay = true;
     } else {
       rabit_enable_tcp_no_delay = false;
-}
+    }
   }
 }
 /*!

diff --git a/rabit/src/allreduce_base.h b/rabit/src/allreduce_base.h
@@ -109,9 +109,11 @@ class AllreduceBase : public IEngine {
     if (world_size == 1 || world_size == -1) {
       return;
     }
-    utils::Assert(TryAllgatherRing(sendrecvbuf_, total_size, slice_begin,
-                                   slice_end, size_prev_slice) == kSuccess,
-                  "AllgatherRing failed");
+    auto ret = TryAllgatherRing(sendrecvbuf_, total_size, slice_begin,
+                                slice_end, size_prev_slice);
+    if (ret != kSuccess) {
+      utils::Error("AllgatherRing failed: %d\n", ret.line);
+    }
   }
   /*!
    * \brief perform in-place allreduce, on sendrecvbuf
@@ -135,9 +137,10 @@ class AllreduceBase : public IEngine {
                  const char *_caller = _CALLER) override {
     if (prepare_fun != nullptr) prepare_fun(prepare_arg);
     if (world_size == 1 || world_size == -1) return;
-    utils::Assert(TryAllreduce(sendrecvbuf_, type_nbytes, count, reducer) ==
-                      kSuccess,
-                  "Allreduce failed");
+    auto ret = TryAllreduce(sendrecvbuf_, type_nbytes, count, reducer);
+    if (ret != kSuccess) {
+      utils::Error("Allreduce failed: %d\n", ret.line);
+    }
   }
   /*!
    * \brief broadcast data from root to all nodes
@@ -152,8 +155,10 @@ class AllreduceBase : public IEngine {
                  const char *_file = _FILE, const int _line = _LINE,
                  const char *_caller = _CALLER) override {
     if (world_size == 1 || world_size == -1) return;
-    utils::Assert(TryBroadcast(sendrecvbuf_, total_size, root) == kSuccess,
-                  "Broadcast failed");
+    auto ret = TryBroadcast(sendrecvbuf_, total_size, root);
+    if (ret != kSuccess) {
+      utils::Error("Broadcast failed: %d\n", ret.line);
+    }
   }
   /*!
    * \brief load latest check point
@@ -272,9 +277,11 @@ class AllreduceBase : public IEngine {
   struct ReturnType {
     /*! \brief internal return type */
     ReturnTypeEnum value;
+    int32_t line { -1 };
     // constructor
-    ReturnType() = default;
-    ReturnType(ReturnTypeEnum value) : value(value) {}  // NOLINT(*)
+    explicit ReturnType(int l  = __builtin_LINE()) : line{l} {}
+    ReturnType(ReturnTypeEnum value, int32_t l = __builtin_LINE()) : value(value), line{l} {}  // NOLINT(*)
+
     inline bool operator==(const ReturnTypeEnum &v) const {
       return value == v;
     }
@@ -518,7 +525,8 @@ class AllreduceBase : public IEngine {
    * \param err the error type
    */
   inline ReturnType ReportError(LinkRecord *link, ReturnType err) {
-    err_link = link; return err;
+    err_link = link;
+    return err;
   }
   //---- data structure related to model ----
   // call sequence counter, records how many calls we made so far