added mscclppGetErrorString

src/include/core.h

@@ -35,35 +35,4 @@
     ret func(args)
 #endif // end PROFAPI
 
-/*
-static __inline__ int mscclppTypeSize(mscclppDataType_t type) {
-  switch (type) {
-    case mscclppInt8:
-    case mscclppUint8:
-      return 1;
-    case mscclppFloat16:
-#if defined(__CUDA_BF16_TYPES_EXIST__)
-    case mscclppBfloat16:
-#endif
-      return 2;
-    case mscclppInt32:
-    case mscclppUint32:
-    case mscclppFloat32:
-      return 4;
-    case mscclppInt64:
-    case mscclppUint64:
-    case mscclppFloat64:
-      return 8;
-    default:
-      return -1;
-  }
-}
-
-#include "debug.h"
-#include "checks.h"
-#include "cudawrap.h"
-#include "utils.h"
-#include "nvtx.h"
-*/
-
 #endif // end include guard

src/include/mscclpp.h

@@ -198,6 +198,16 @@ mscclppResult_t mscclppBootstrapAllGather(mscclppComm_t comm, void* data, int si
  */
 mscclppResult_t mscclppCommDestroy(mscclppComm_t comm);
 
+/* Return the string for the given error code.
+ * 
+ * Ouput:
+ *   returns the string
+ * 
+ * Inputs:
+ *   result: the error code that this function needs to translate
+ */
+const char*  mscclppGetErrorString(mscclppResult_t result);
+
 /* Connect to a remote rank. This function only prepares metadata for connection. The actual connection
  * is made by a following call of mscclppConnectionSetup(). Note that this function is two-way and a connection
  * from rank i to remote rank j needs to have a counterpart from rank j to rank i.

src/init.cc

@@ -255,6 +255,20 @@ mscclppResult_t mscclppCommDestroy(mscclppComm_t comm){
   return mscclppSuccess;
 }
 
+MSCCLPP_API(const char*, mscclppGetErrorString, mscclppResult_t code);
+const char* mscclppGetErrorString(mscclppResult_t code) {
+  switch (code) {
+    case mscclppSuccess                : return "no error";
+    case mscclppUnhandledCudaError     : return "unhandled cuda error";
+    case mscclppSystemError            : return "unhandled system error";
+    case mscclppInternalError          : return "internal error";
+    case mscclppInvalidArgument        : return "invalid argument";
+    case mscclppInvalidUsage           : return "invalid usage";
+    case mscclppRemoteError            : return "remote process exited or there was a network error";
+    case mscclppInProgress             : return "MSCCL++ operation in progress";
+    default                            : return "unknown result code";
+  }
+}
 
 MSCCLPP_API(mscclppResult_t, mscclppGetDeviceConnection, mscclppComm_t comm, int remoteRank, int tag, mscclppDevConn_t** devConn);
 mscclppResult_t mscclppGetDeviceConnection(mscclppComm_t comm, int remoteRank, int tag, mscclppDevConn_t** devConn){

tests/allgather_test.cu

@@ -18,9 +18,9 @@ static int nranksPerNode = 8;
 #define MSCCLPPCHECK(call) do { \
   mscclppResult_t res = call; \
   if (res != mscclppSuccess && res != mscclppInProgress) { \
-     /* Print the back trace*/ \
-   printf("Failure at %s:%d -> %d\n", __FILE__, __LINE__, res);    \
-       return res; \
+    /* Print the back trace*/ \
+    printf("Failure at %s:%d -> %s\n", __FILE__, __LINE__, mscclppGetErrorString(res));    \
+    return res; \
   } \
 } while (0)