fix: disable AMD SMI for gfx1151 targets in CMake and remove a debug error from the SMI wrapper header.

projects/rccl/CMakeLists.txt

@@ -46,12 +46,7 @@ option(QUIET_WARNINGS                          "Supress compiler warnings"
 option(ENABLE_ROCSHMEM                         "Enable rocSHMEM support in RCCL"               OFF)
 option(ENABLE_AMDSMI                           "Enable AMD/ROCm SMI support"                   ON)
 
-if(ENABLE_AMDSMI)
-  message(STATUS "SMI Support: ENABLED")
-  add_compile_definitions(RCCL_SMI_ENABLED)
-else()
-  message(STATUS "SMI Support: DISABLED via ENABLE_AMDSMI=OFF")
-endif()
+
 
 # Default GPU architectures to build
 #==================================================================================================
@@ -125,9 +120,23 @@ else()
   set(SUPPORTED_GPUS ${DEFAULT_GPUS})
 endif()
 
+set(GPU_TARGETS "${SUPPORTED_GPUS}")
 set(GPU_TARGETS "${SUPPORTED_GPUS}")
 message(STATUS "Compiling for ${GPU_TARGETS}")
 
+# Auto-disable SMI for gfx1151 as it is not supported
+if("${GPU_TARGETS}" MATCHES "gfx1151")
+  message(STATUS "Detected gfx1151 target: Forcing ENABLE_AMDSMI=OFF")
+  set(ENABLE_AMDSMI OFF CACHE BOOL "Force disable SMI for gfx1151" FORCE)
+endif()
+
+if(ENABLE_AMDSMI)
+  message(STATUS "SMI Support: ENABLED")
+  add_compile_definitions(RCCL_SMI_ENABLED)
+else()
+  message(STATUS "SMI Support: DISABLED")
+endif()
+
 ## NOTE: Reload rocm-cmake in order to update GPU_TARGETS
 include(cmake/Dependencies.cmake) # Reloading to use desired GPU_TARGETS instead of defaults
 

projects/rccl/src/include/rocm_smi_wrap.h

@@ -36,7 +36,6 @@ ncclResult_t rocm_smi_getDevicePciBusIdString(uint32_t deviceIndex, char* pciBus
 ncclResult_t rocm_smi_getDeviceIndexByPciBusId(const char* pciBusId, uint32_t* deviceIndex);
 ncclResult_t rocm_smi_getLinkInfo(int srcDev, int dstDev, RSMI_IO_LINK_TYPE* rsmi_type, int *hops, int *count);
 #else
-#error "DEBUG_TRACE: SMI is indeed disabled, using inline functions."
 inline ncclResult_t rocm_smi_init() { return ncclSuccess; }
 inline ncclResult_t rocm_smi_getNumDevice(uint32_t* num_devs) { *num_devs = 0; return ncclSuccess; }
 inline ncclResult_t rocm_smi_getDevicePciBusIdString(uint32_t deviceIndex, char* pciBusId, size_t len) { if (len > 0) pciBusId[0] = '\0'; return ncclSuccess; }