From 937aec99f0ecdc5e5d000d3cef667941cec328de Mon Sep 17 00:00:00 2001
From: Wenkai Du <43822138+wenkaidu@users.noreply.github.com>
Date: Fri, 13 Nov 2020 17:58:04 -0800
Subject: [PATCH] Use device's link width and speed if port doesn't report
 (#304)

(cherry picked from commit 554729079d98a1d95eaf69ded5f8b2f286532dd8)
---
 src/graph/xml.cc | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)

diff --git a/src/graph/xml.cc b/src/graph/xml.cc
index a1f936bab..3bf4fa783 100644
--- a/src/graph/xml.cc
+++ b/src/graph/xml.cc
@@ -492,7 +492,10 @@ ncclResult_t ncclTopoGetXmlFromSys(struct ncclXmlNode* pciNode, struct ncclXml*
     char portSpeedStr[MAX_STR_LEN];
     float portSpeed;
     NCCLCHECK(ncclTopoGetStrFromSys(path, "../max_link_speed", portSpeedStr));
-    sscanf(portSpeedStr, "%f GT/s", &portSpeed);
+    if (portSpeedStr[0])
+      sscanf(portSpeedStr, "%f GT/s", &portSpeed);
+    else
+      portSpeed = deviceSpeed;
     NCCLCHECK(xmlSetAttr(pciNode, "link_speed", portSpeed < deviceSpeed ? portSpeedStr : deviceSpeedStr));
   }
   NCCLCHECK(xmlGetAttrIndex(pciNode, "link_width", &index));
@@ -502,7 +505,11 @@ ncclResult_t ncclTopoGetXmlFromSys(struct ncclXmlNode* pciNode, struct ncclXml*
     NCCLCHECK(ncclTopoGetStrFromSys(path, "max_link_width", strValue));
     int deviceWidth = strtol(strValue, NULL, 0);
     NCCLCHECK(ncclTopoGetStrFromSys(path, "../max_link_width", strValue));
-    int portWidth = strtol(strValue, NULL, 0);
+    int portWidth;
+    if (strValue[0])
+      portWidth = strtol(strValue, NULL, 0);
+    else
+      portWidth = deviceWidth;
     NCCLCHECK(xmlSetAttrInt(pciNode, "link_width", std::min(deviceWidth,portWidth)));
   }
   struct ncclXmlNode* parent = pciNode->parent;