From ca5489532606e8405284120604720e74143d743d Mon Sep 17 00:00:00 2001
From: dsashidh <dsashidh@redhat.com>
Date: Mon, 21 Jul 2025 19:45:57 +0000
Subject: [PATCH 1/3] Add header check for rocm_smi.h to avoid build errors

---
 CMakeLists.txt                                           | 9 +++++++++
 torch/csrc/distributed/c10d/symm_mem/intra_node_comm.cpp | 2 +-
 2 files changed, 10 insertions(+), 1 deletion(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 99c0b9e0ea0c..7156fcff3a29 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -1396,3 +1396,12 @@ if(BUILD_BUNDLE_PTXAS AND USE_CUDA)
   install(PROGRAMS "${PROJECT_BINARY_DIR}/ptxas"
           DESTINATION "${CMAKE_INSTALL_BINDIR}")
 endif()
+
+if(USE_ROCM)
+  include(CheckIncludeFile)
+  check_include_file("rocm_smi/rocm_smi.h" HAS_ROCM_SMI_H)
+  message(STATUS "HAS_ROCM_SMI_H = ${HAS_ROCM_SMI_H}")
+  if(HAS_ROCM_SMI_H)
+    add_definitions(-DHAS_ROCM_SMI)
+  endif()
+endif()
diff --git a/torch/csrc/distributed/c10d/symm_mem/intra_node_comm.cpp b/torch/csrc/distributed/c10d/symm_mem/intra_node_comm.cpp
index 0d53d100cee7..72415909ab00 100644
--- a/torch/csrc/distributed/c10d/symm_mem/intra_node_comm.cpp
+++ b/torch/csrc/distributed/c10d/symm_mem/intra_node_comm.cpp
@@ -2,7 +2,7 @@
 #include <torch/csrc/distributed/c10d/symm_mem/DMAConnectivity.hpp>
 #include <torch/csrc/distributed/c10d/symm_mem/intra_node_comm.hpp>
 
-#if defined(USE_ROCM)
+#if defined(USE_ROCM) && defined(HAS_ROCM_SMI)
 #include <rocm_smi/rocm_smi.h>
 #endif
 

From 3e76a0f43e02d6bdc29a3fcaa6e9af5c79ea3920 Mon Sep 17 00:00:00 2001
From: dsashidh <dsashidh@redhat.com>
Date: Tue, 22 Jul 2025 15:29:34 +0000
Subject: [PATCH 2/3] Guard rsmi ussage with HAS_ROCM_SMI to avoid build errors

---
 torch/csrc/distributed/c10d/symm_mem/intra_node_comm.cpp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/torch/csrc/distributed/c10d/symm_mem/intra_node_comm.cpp b/torch/csrc/distributed/c10d/symm_mem/intra_node_comm.cpp
index 72415909ab00..9600a156486c 100644
--- a/torch/csrc/distributed/c10d/symm_mem/intra_node_comm.cpp
+++ b/torch/csrc/distributed/c10d/symm_mem/intra_node_comm.cpp
@@ -171,7 +171,7 @@ bool IntraNodeComm::rendezvous() {
   gethostname(devInfo.hostname, sizeof(devInfo.hostname));
   devInfo.deviceIdx = deviceIdx_;
 
-#if defined(USE_ROCM)
+#if defined(USE_ROCM) && defined(HAS_ROCM_SMI)
   auto ret = rsmi_init(0);
   if (ret != RSMI_STATUS_SUCCESS) {
     LOG(ERROR) << "IntraNodeComm:: rendezvous failed in rsmi_init, ret=" << ret;

From 07b5dbe2202933abc43f494b63bd146be3ae27b1 Mon Sep 17 00:00:00 2001
From: Dev Sashidhar <dsashidh@redhat.com>
Date: Tue, 5 Aug 2025 17:29:15 +0000
Subject: [PATCH 3/3] Move rocm_smi.h check to top-level CMakeLists.txt and
 fail early if missing

---
 CMakeLists.txt                                  | 17 ++++++++---------
 .../c10d/symm_mem/intra_node_comm.cpp           |  2 +-
 2 files changed, 9 insertions(+), 10 deletions(-)

diff --git a/CMakeLists.txt b/CMakeLists.txt
index 7156fcff3a29..35bbfb563899 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -28,6 +28,14 @@ endif()
 # ---[ Project and semantic versioning.
 project(Torch CXX C)
 
+if(DEFINED ENV{USE_ROCM})
+  include(CheckIncludeFile)
+  check_include_file("rocm_smi/rocm_smi.h" HAS_ROCM_SMI_H)
+  if(NOT HAS_ROCM_SMI_H)
+    message(FATAL_ERROR "rocm_smi.h not found. Please install the rocm-smi-lib package.")
+  endif()
+endif()
+
 if(${CMAKE_SYSTEM_NAME} STREQUAL "Linux")
   set(LINUX TRUE)
 else()
@@ -1396,12 +1404,3 @@ if(BUILD_BUNDLE_PTXAS AND USE_CUDA)
   install(PROGRAMS "${PROJECT_BINARY_DIR}/ptxas"
           DESTINATION "${CMAKE_INSTALL_BINDIR}")
 endif()
-
-if(USE_ROCM)
-  include(CheckIncludeFile)
-  check_include_file("rocm_smi/rocm_smi.h" HAS_ROCM_SMI_H)
-  message(STATUS "HAS_ROCM_SMI_H = ${HAS_ROCM_SMI_H}")
-  if(HAS_ROCM_SMI_H)
-    add_definitions(-DHAS_ROCM_SMI)
-  endif()
-endif()
diff --git a/torch/csrc/distributed/c10d/symm_mem/intra_node_comm.cpp b/torch/csrc/distributed/c10d/symm_mem/intra_node_comm.cpp
index 9600a156486c..f4c1c3fbbf57 100644
--- a/torch/csrc/distributed/c10d/symm_mem/intra_node_comm.cpp
+++ b/torch/csrc/distributed/c10d/symm_mem/intra_node_comm.cpp
@@ -2,7 +2,7 @@
 #include <torch/csrc/distributed/c10d/symm_mem/DMAConnectivity.hpp>
 #include <torch/csrc/distributed/c10d/symm_mem/intra_node_comm.hpp>
 
-#if defined(USE_ROCM) && defined(HAS_ROCM_SMI)
+#if defined(USE_ROCM)
 #include <rocm_smi/rocm_smi.h>
 #endif