From 9f37923767300e50b24d82c76501de4e747ebd3d Mon Sep 17 00:00:00 2001
From: Divya Shikre <DivyaUday.Shikre@amd.com>
Date: Thu, 3 Dec 2020 13:01:15 -0500
Subject: [PATCH] Fix for inconsistent GPU indexing between rocm-smi and
 rbt/hip.

Signed-off-by: Divya Shikre <DivyaUday.Shikre@amd.com>
Change-Id: I0d966c91bfe0f0d51859ff098d15011a3e4e8b29


[ROCm/rocm_smi_lib commit: 47ca37aef75ac57ecd1214a8ab6dac7be65302fb]
---
 projects/rocm-smi-lib/src/rocm_smi_main.cc | 25 +++++++++++++++-------
 1 file changed, 17 insertions(+), 8 deletions(-)

diff --git a/projects/rocm-smi-lib/src/rocm_smi_main.cc b/projects/rocm-smi-lib/src/rocm_smi_main.cc
index 9aa4ced9f5..3822a65f4b 100755
--- a/projects/rocm-smi-lib/src/rocm_smi_main.cc
+++ b/projects/rocm-smi-lib/src/rocm_smi_main.cc
@@ -461,6 +461,7 @@ static bool isAMDGPU(std::string dev_path) {
 uint32_t RocmSMI::DiscoverAmdgpuDevices(void) {
   uint32_t ret = 0;
   std::string err_msg;
+  uint32_t count = 0;
 
   // If this gets called more than once, clear previous findings.
   devices_.clear();
@@ -486,18 +487,26 @@ uint32_t RocmSMI::DiscoverAmdgpuDevices(void) {
   while (dentry != nullptr) {
     if (memcmp(dentry->d_name, kDeviceNamePrefix, strlen(kDeviceNamePrefix))
                                                                        == 0) {
-      std::string vend_str_path = kPathDRMRoot;
-      vend_str_path += "/";
-      vend_str_path += dentry->d_name;
-
-      if (isAMDGPU(vend_str_path) ||
-          (init_options_ & RSMI_INIT_FLAG_ALL_GPUS)) {
-        AddToDeviceList(dentry->d_name);
-      }
+        if ((strcmp(dentry->d_name, ".") == 0) ||
+           (strcmp(dentry->d_name, "..") == 0))
+           continue;
+        count++;
     }
     dentry = readdir(drm_dir);
   }
 
+  for (uint32_t node_id = 0; node_id < count; node_id++) {
+    std::string path = kPathDRMRoot;
+    path += "/card";
+    path += std::to_string(node_id);
+    if (isAMDGPU(path) ||
+       (init_options_ & RSMI_INIT_FLAG_ALL_GPUS)) {
+          std::string d_name = "card";
+          d_name += std::to_string(node_id);
+          AddToDeviceList(d_name);
+      }
+  }
+
   if (closedir(drm_dir)) {
     err_msg = "Failed to close drm root directory ";
     err_msg += kPathDRMRoot;