Kill c10d/private/CUDAUtils.hpp (#13681)

Summary:
Use AT_CUDA_CHECK instead
Pull Request resolved: https://github.com/pytorch/pytorch/pull/13681

Differential Revision: D12966607

Pulled By: teng-li

fbshipit-source-id: da0431f588969791a19519368edb909b9c3dc5ab

7 files changed, 3 insertions, 35 deletions

diff --git a/torch/lib/c10d/ProcessGroupGloo.cpp b/torch/lib/c10d/ProcessGroupGloo.cpp
index cd226f0fff..6b6a77932d 100644
--- a/torch/lib/c10d/ProcessGroupGloo.cpp
+++ b/torch/lib/c10d/ProcessGroupGloo.cpp
@@ -21,8 +21,6 @@
 #include <gloo/cuda_allreduce_halving_doubling.h>
 #include <gloo/cuda_allreduce_ring_chunked.h>
 #include <gloo/cuda_broadcast_one_to_all.h>
-
-#include <c10d/private/CUDAUtils.hpp>
 #endif
 
 #include <gloo/rendezvous/context.h>
diff --git a/torch/lib/c10d/ProcessGroupNCCL.cpp b/torch/lib/c10d/ProcessGroupNCCL.cpp
index 9e64fe7fae..a3ede7b1cc 100644
--- a/torch/lib/c10d/ProcessGroupNCCL.cpp
+++ b/torch/lib/c10d/ProcessGroupNCCL.cpp
@@ -10,7 +10,6 @@
 #include <ATen/cuda/CUDAGuard.h>
 
 #include <c10d/Utils.hpp>
-#include <c10d/private/CUDAUtils.hpp>
 
 namespace c10d {
 
@@ -103,7 +102,7 @@ bool ProcessGroupNCCL::WorkNCCL::finishedGPUExecution() const {
     // Checking the work's corresponding CUDA events' status
     auto ret = cudaEventQuery(cudaEvents_[i]);
     if (ret != cudaSuccess && ret != cudaErrorNotReady) {
-      C10D_CUDA_CHECK(ret);
+      AT_CUDA_CHECK(ret);
     }
     if (ret == cudaErrorNotReady) {
       return false;
diff --git a/torch/lib/c10d/private/CUDAUtils.hpp b/torch/lib/c10d/private/CUDAUtils.hpp
deleted file mode 100644
index 6b31263add..0000000000
--- a/torch/lib/c10d/private/CUDAUtils.hpp
+++ /dev/null
@@ -1,23 +0,0 @@
-#pragma once
-
-#include <sstream>
-#include <stdexcept>
-
-#include <cuda.h>
-#include <cuda_runtime.h>
-
-// TODO: Use AT_CHECK or similar here
-#define C10D_CUDA_CHECK(condition)        \
-  do {                                    \
-    cudaError_t error = (condition);      \
-    if (error != cudaSuccess) {           \
-      std::stringstream ss;               \
-      ss << "Error at: ";                 \
-      ss << __FILE__;                     \
-      ss << ":";                          \
-      ss << __LINE__;                     \
-      ss << ": ";                         \
-      ss << cudaGetErrorString(error);    \
-      throw std::runtime_error(ss.str()); \
-    }                                     \
-  } while (0)
diff --git a/torch/lib/c10d/test/CUDATest.cu b/torch/lib/c10d/test/CUDATest.cu
index ef8a2780c6..b6efe6fceb 100644
--- a/torch/lib/c10d/test/CUDATest.cu
+++ b/torch/lib/c10d/test/CUDATest.cu
@@ -1,6 +1,5 @@
 #include "CUDATest.hpp"
-
-#include <c10d/private/CUDAUtils.hpp>
+#include <ATen/cuda/Exceptions.h>
 
 namespace c10d {
 namespace test {
@@ -22,7 +21,7 @@ void cudaSleep(at::cuda::CUDAStream& stream, uint64_t clocks) {
 
 int cudaNumDevices() {
   int n = 0;
-  C10D_CUDA_CHECK(cudaGetDeviceCount(&n));
+  AT_CUDA_CHECK(cudaGetDeviceCount(&n));
   return n;
 }
 
diff --git a/torch/lib/c10d/test/CUDATest.hpp b/torch/lib/c10d/test/CUDATest.hpp
index d2275c128e..5e02e317fb 100644
--- a/torch/lib/c10d/test/CUDATest.hpp
+++ b/torch/lib/c10d/test/CUDATest.hpp
@@ -1,8 +1,5 @@
 #pragma once
 
-#include <cuda.h>
-#include <cuda_runtime.h>
-
 #include <ATen/cuda/CUDAStream.h>
 
 namespace c10d {
diff --git a/torch/lib/c10d/test/ProcessGroupGlooAsyncTest.cpp b/torch/lib/c10d/test/ProcessGroupGlooAsyncTest.cpp
index 5c4912706c..10b7ab1c10 100644
--- a/torch/lib/c10d/test/ProcessGroupGlooAsyncTest.cpp
+++ b/torch/lib/c10d/test/ProcessGroupGlooAsyncTest.cpp
@@ -4,7 +4,6 @@
 
 #include <c10d/FileStore.hpp>
 #include <c10d/ProcessGroupGloo.hpp>
-#include <c10d/private/CUDAUtils.hpp>
 #include <c10d/test/CUDATest.hpp>
 #include <c10d/test/TestUtils.hpp>
 
diff --git a/torch/lib/c10d/test/ProcessGroupNCCLTest.cpp b/torch/lib/c10d/test/ProcessGroupNCCLTest.cpp
index 4f0f0c33f7..20ae4dda56 100644
--- a/torch/lib/c10d/test/ProcessGroupNCCLTest.cpp
+++ b/torch/lib/c10d/test/ProcessGroupNCCLTest.cpp
@@ -2,7 +2,6 @@
 
 #include <c10d/FileStore.hpp>
 #include <c10d/ProcessGroupNCCL.hpp>
-#include <c10d/private/CUDAUtils.hpp>
 #include <c10d/test/CUDATest.hpp>
 #include <c10d/test/TestUtils.hpp>