[RAS] Add support for RAS client (#1748)

Enable RAS client binary `rcclras`

src/ras/client.cc

@@ -84,7 +84,7 @@ static void parseArgs(int argc, char** argv) {
         printUsage(argv[0]);
         exit(0);
       case 'r':
-        fprintf(stderr, "NCCL RAS client version " STR(NCCL_MAJOR) "." STR(NCCL_MINOR) "."
+        fprintf(stderr, "RCCL RAS client version " STR(NCCL_MAJOR) "." STR(NCCL_MINOR) "."
                 STR(NCCL_PATCH) NCCL_SUFFIX "\n");
         exit(0);
       default:

src/ras/client_support.cc

@@ -487,18 +487,20 @@ static ncclResult_t rasClientRunInit(struct rasClient* client) {
   int firstIdx, nPeers;
   struct rasValCount valCounts[NCCL_MAX_LOCAL_RANKS];
   int nValCounts;
-  static int cudaDriver = -1, cudaRuntime = -1;
+  static int hipRuntime = -1, amdgpuDriver = -1;
 
   TRACE(NCCL_RAS, "RAS: rasClientRunInit: starting");
 
   rasOutReset();
-  rasOutAppend("NCCL version " STR(NCCL_MAJOR) "." STR(NCCL_MINOR) "." STR(NCCL_PATCH) NCCL_SUFFIX
-               " compiled with CUDA " STR(CUDA_MAJOR) "." STR(CUDA_MINOR) "\n");
-  if (cudaRuntime == -1)
-    cudaRuntimeGetVersion(&cudaRuntime);
-  if (cudaDriver == -1)
-    cudaDriverGetVersion(&cudaDriver);
-  rasOutAppend("CUDA runtime version %d, driver version %d\n\n", cudaRuntime, cudaDriver);
+  rasOutAppend("RCCL version " STR(NCCL_MAJOR) "." STR(NCCL_MINOR) "." STR(NCCL_PATCH) NCCL_SUFFIX
+               " compiled with ROCm " STR(ROCM_BUILD_INFO) "\n");
+  if (hipRuntime == -1)
+    hipRuntimeGetVersion(&hipRuntime);
+  if (amdgpuDriver == -1)
+    hipDriverGetVersion(&amdgpuDriver);
+    //Find a better way to query amdgpu driver version, as hipDriverGetVersion() reports the same as hipRuntimeGetVersion()
+    //Else, cudaRuntimeGetVersion() and cudaDriverGetVersion() are anyways hipified, so no need of this mod
+  rasOutAppend("HIP runtime version %d, amdgpu driver version %d\n\n", hipRuntime, amdgpuDriver);
   msgLen = rasOutLength();
   NCCLCHECKGOTO(rasClientAllocMsg(&msg, msgLen), ret, fail);
   rasOutExtract(msg);

src/ras/ras_internal.h

@@ -22,6 +22,7 @@
 #include "ras.h"
 #include "socket.h"
 #include "utils.h"
+#include "hip_rocm_version_info.h"
 
 // Type of a RAS network or client message.
 typedef enum {