saymrwulf/pytorch
1
0
Fork 0
mirror of https://github.com/saymrwulf/pytorch.git synced 2026-05-14 20:57:59 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Enable ProcessGroupGlooTest in CI (take 2) (#42086)

Browse source 
Summary:
Fixes https://github.com/pytorch/pytorch/issues/42073

Pull Request resolved: https://github.com/pytorch/pytorch/pull/42086

Reviewed By: ngimel

Differential Revision: D22765777

Pulled By: malfet

fbshipit-source-id: ebbcd44f448a1e7f9a3d18fa9967461129dd1dcd
This commit is contained in:
Nikita Shulga 2020-07-27 10:20:07 -07:00 committed by Facebook GitHub Bot
parent 96aaa311c0
commit fbdaa555a2
3 changed files with 13 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

10
.jenkins/pytorch/test.sh
View file

@ -228,6 +228,15 @@ test_libtorch() {
fi
}
test_distributed() {
if [[ "$BUILD_ENVIRONMENT" == *cuda* ]]; then
echo "Testing distributed C++ tests"
mkdir -p test/test-reports/cpp-distributed
build/bin/ProcessGroupGlooTest --gtest_output=xml:test/test-reports/cpp-distributed/ProcessGroupGlooTest.xml
build/bin/ProcessGroupNCCLErrorsTest --gtest_output=xml:test/test-reports/cpp-distributed/ProcessGroupNCCLErrorsTest.xml
fi
}
test_custom_backend() {
if [[ "$BUILD_ENVIRONMENT" != *rocm* ]] && [[ "$BUILD_ENVIRONMENT" != *asan* ]] ; then
echo "Testing custom backends"
@ -371,4 +380,5 @@ else
test_custom_script_ops
test_custom_backend
test_torch_function_benchmark
test_distributed
fi

2
torch/lib/c10d/test/CUDATest.cu
View file

@ -21,7 +21,7 @@ void cudaSleep(at::cuda::CUDAStream& stream, uint64_t clocks) {
int cudaNumDevices() {
int n = 0;
AT_CUDA_CHECK(cudaGetDeviceCount(&n));
C10_CUDA_CHECK_WARN(cudaGetDeviceCount(&n));
return n;
}

4
torch/lib/c10d/test/ProcessGroupGlooTest.cpp
View file

@ -224,7 +224,7 @@ void testBroadcast(const std::string& path, const at::DeviceType b) {
std::vector<std::vector<at::Tensor>> inputs(size);
// Try every permutation of root rank and root tensoro
// Try every permutation of root rank and root tensor
for (auto i = 0; i < size; i++) {
for (auto j = 0; j < stride; j++) {
// Initialize inputs
@ -548,7 +548,7 @@ TEST(ProcessGroupGlooTest, testAllReduceCUDA) {
TEST(ProcessGroupGlooTest, testBroadcastCUDA) {
{
if (torch::cuda::is_available()) {
if (torch::cuda::device_count() > 1) {
TemporaryFile file;
testBroadcast(file.path, at::DeviceType::CUDA);
}