Fix(critical): Introduce SMI_DISABLED define to forcibly disable SMI usage in headers

projects/rccl/CMakeLists.txt

@@ -284,6 +284,7 @@ if(ROCM_VERSION VERSION_GREATER_EQUAL "71100" AND ENABLE_AMDSMI)
 else()
   # Force unset if disabled
   unset(USE_AMDSMI CACHE)
+  add_definitions("-DSMI_DISABLED")
 endif()
 
 if(NOT USE_AMDSMI AND ENABLE_AMDSMI)

projects/rccl/src/include/amdsmi_wrap.h

@@ -7,7 +7,7 @@
 #include "amd_smi/amdsmi.h"
 #include "nccl.h"
 
-#ifdef USE_AMDSMI
+#if defined(USE_AMDSMI) && !defined(SMI_DISABLED)
 ncclResult_t amd_smi_init();
 ncclResult_t amd_smi_shutdown();
 ncclResult_t amd_smi_getNumDevice(uint32_t* num_devs);

projects/rccl/src/include/rocm_smi_wrap.h

@@ -29,7 +29,7 @@ THE SOFTWARE.
 #endif
 #include "nccl.h"
 
-#ifdef USE_ROCMSMI
+#if defined(USE_ROCMSMI) && !defined(SMI_DISABLED)
 ncclResult_t rocm_smi_init();
 ncclResult_t rocm_smi_getNumDevice(uint32_t* num_devs);
 ncclResult_t rocm_smi_getDevicePciBusIdString(uint32_t deviceIndex, char* pciBusId, size_t len);