Check all CUDA API calls for errors in test/ (pytorch#74921) (pytorch…

…#83954) Summary: Pull Request resolved: pytorch#74921 Test Plan: Sandcastle Reviewed By: ezyang, malfet, ngimel Differential Revision: D35194966 Pull Request resolved: pytorch#83954 Approved by: https://github.com/ezyang
dicky98 · Aug 24, 2022 · 67f0940 · 67f0940
1 parent a741927
commit 67f0940
Show file tree

Hide file tree

Showing 2 changed files with 436 additions and 387 deletions.
diff --git a/test/cpp/c10d/ProcessGroupNCCLTest.cpp b/test/cpp/c10d/ProcessGroupNCCLTest.cpp
@@ -106,7 +106,7 @@ class NCCLTest : public NCCLTestBase {
 
     // Copy inputs to outputs
     for (const auto i : c10::irange(numDevices_)) {
-      cudaStreamSynchronize(streams_[i].stream());
+      C10_CUDA_CHECK(cudaStreamSynchronize(streams_[i].stream()));
       outputs[i] = tensors_[i].cpu();
     }
 
@@ -137,7 +137,7 @@ class NCCLTest : public NCCLTestBase {
 
     // Copy inputs to outputs
     for (const auto i : c10::irange(numDevices_)) {
-      cudaStreamSynchronize(streams_[i].stream());
+      C10_CUDA_CHECK(cudaStreamSynchronize(streams_[i].stream()));
       for (auto j = 0; j < worldSize_ * numDevices_; ++j) {
         outputs[i][j] = tensor_lists[i][j].cpu();
       }