NCCL 2.27.7-1

Prevent initialization failures in certain configurations when attempting
to load fp8-specific symmetric multicast kernels on GPUs older than
Blackwell.


[ROCm/rccl commit: 593de54e52]

projects/rccl/src/enqueue.cc

@@ -38,12 +38,9 @@ ncclResult_t ncclInitKernelsForDevice(int cudaArch, int maxSharedMem, size_t* ma
       if (fn == nullptr) continue;
 
       cudaError_t errcode = cudaFuncGetAttributes(&attr, fn);
-      if (errcode == cudaErrorNoKernelImageForDevice) continue;
-      CUDACHECKGOTO(errcode, result, ignore0);
-
+      if (errcode != cudaSuccess) continue; // Silently ignore failures
       if (maxStackSize) {
         if (attr.localSizeBytes > *maxStackSize) *maxStackSize = attr.localSizeBytes;
-      ignore0:;
       }
       if (carveout) {
         CUDACHECKGOTO(cudaFuncSetAttribute(fn,