More robust msccl shared directory location discovery (#869)

ROCm · Aug 29, 2023 · 6dca2cf · 6dca2cf
1 parent 26490f4
commit 6dca2cf
Showing 1 changed file with 4 additions and 3 deletions.
diff --git a/src/misc/msccl/msccl_lifecycle.cc b/src/misc/msccl/msccl_lifecycle.cc
@@ -71,8 +71,8 @@ static const char* mscclAlgoDirEnv = "MSCCL_ALGO_DIR";
 static const char* mscclAlgoDefaultDir = "msccl-algorithms";
 extern "C" bool mscclUnitTestMode() __attribute__((__weak__));
 static const char* mscclUnitTestAlgoDefaultDir = "msccl-unit-test-algorithms";
-static const char* mscclAlgoShareDirPath = "share/rccl/msccl-algorithms";
-static const char* mscclUnitTestAlgoShareDirPath = "share/rccl/msccl-unit-test-algorithms";
+static const char* mscclAlgoShareDirPath = "../share/rccl/msccl-algorithms";
+static const char* mscclUnitTestAlgoShareDirPath = "../share/rccl/msccl-unit-test-algorithms";
 
 static ncclResult_t mscclInternalSchedulerInit() {
  mscclStatus& status = mscclGetStatus();
@@ -94,7 +94,7 @@ static ncclResult_t mscclInternalSchedulerInit() {
  mscclAlgoDirStr += (mscclUnitTestMode && mscclUnitTestMode()) ? mscclUnitTestAlgoDefaultDir : mscclAlgoDefaultDir;
  mscclAlgoDir = mscclAlgoDirStr.c_str();
  // Get share Directory Paths
- mscclAlgoShareDirStr = selfLibPath.substr(0, selfLibPath.rfind("lib"));
+ mscclAlgoShareDirStr = selfLibPath.substr(0, selfLibPath.find_last_of("/\\") + 1);
  mscclAlgoShareDirStr += (mscclUnitTestMode && mscclUnitTestMode()) ? mscclUnitTestAlgoShareDirPath : mscclAlgoShareDirPath;
  mscclAlgoShareDir = mscclAlgoShareDirStr.c_str();
  }
@@ -112,6 +112,7 @@ static ncclResult_t mscclInternalSchedulerInit() {
  } else {
  fullDirPath = mscclAlgoDir;
  }
+ INFO(NCCL_INIT, "Using MSCCL files from %s", fullDirPath);
  while ((entry = readdir(dp))) {
  if (entry->d_type != DT_LNK && entry->d_type != DT_REG) {
  continue;