Allow default alignment less then 16.

Author: PatriosTheGreat

src/enqueue.cc

@@ -1302,8 +1302,9 @@ static ncclResult_t uploadWork(struct ncclComm* comm, struct ncclKernelPlan* pla
     plan->kernelArgs->workBuf = comm->workFifoBufDev;
     break;
   case ncclDevWorkStorageTypePersistent:
-    static_assert(16 <= alignof(max_align_t), "We rely on 16-byte alignment.");
-    fifoBufHost = malloc(workBytes);
+    const int alignment = 16;
+    size_t workBytesAlignedUp = ((workBytes + alignment - 1) / alignment) * alignment;
+    fifoBufHost = aligned_alloc(16, workBytesAlignedUp);
     fifoCursor = 0;
     fifoMask = ~0u;
     break;