Adding nccl patch fixes (#47)

* adding patch file for torch 1.13

* renaming patch file

patches/torch1.12.nccl.cpp.patch

@@ -7,7 +7,7 @@ index 5817449c1a..edc4f7781a 100644
    auto* recvbuff = reinterpret_cast<char *>(output.data_ptr());
    auto comm = to_nccl_comm(_comm);
 -#if defined(USE_ROCM) && ROCM_VERSION >= 50000
-+#if 1 
++#if 1
    NCCL_CHECK(ncclAllToAll(sendbuff , recvbuff , count,  type, comm, stream));
  #else
    NCCL_CHECK(ncclCommCount(comm, &numranks));

patches/torch1.13.nccl.cpp.patch

@@ -0,0 +1,13 @@
+diff --git a/torch/csrc/cuda/nccl.cpp b/torch/csrc/cuda/nccl.cpp
+index 83729084ae..934bf24ea4 100644
+--- a/torch/csrc/cuda/nccl.cpp
++++ b/torch/csrc/cuda/nccl.cpp
+@@ -655,7 +655,7 @@ void all2all_single_equal_split(
+   const auto* sendbuff = reinterpret_cast<char*>(input.data_ptr());
+   auto* recvbuff = reinterpret_cast<char*>(output.data_ptr());
+   auto comm = to_nccl_comm(_comm);
+-#if defined(USE_ROCM) && ROCM_VERSION >= 50000
++#if 1
+   NCCL_CHECK(ncclAllToAll(sendbuff, recvbuff, count, type, comm, stream));
+ #else
+   NCCL_CHECK(ncclCommCount(comm, &numranks));