to use run_distributed.py

.github/scripts/ut_result_check.sh

@@ -73,19 +73,13 @@ if [[ "${ut_suite}" == 'torch_xpu' ]]; then
     fi
 fi
 if [[ "${ut_suite}" == 'xpu_distributed' ]]; then
-    grep "^FAILED" c10d_ops_xccl_test.log | awk '{print $2}' > ./"${ut_suite}"_ops_xccl_test_failed.log
-    grep "^FAILED" c10d_xccl_test.log | awk '{print $2}' > ./"${ut_suite}"_c10d_xccl_test_failed.log
-    num_failed_ops_xccl=$(wc -l < "./${ut_suite}_ops_xccl_test_failed.log")
-    num_failed_c10d_xccl=$(wc -l < "./${ut_suite}_c10d_xccl_test_failed.log")
+    grep "^FAILED" xpu_distributed_test.log | awk '{print $2}' > ./"${ut_suite}"_xpu_distributed_test_failed.log
+    num_failed_xpu_distributed=$(wc -l < "./${ut_suite}_xpu_distributed_test_failed.log")
     echo -e "========================================================================="
-    echo -e "Show Failed cases in ${ut_suite} c10d ops xccl"
+    echo -e "Show Failed cases in ${ut_suite} xpu distributed"
     echo -e "========================================================================="
-    cat "./${ut_suite}_ops_xccl_test_failed.log"
-    echo -e "========================================================================="
-    echo -e "Show Failed cases in ${ut_suite} c10d xccl"
-    echo -e "========================================================================="
-    cat "./${ut_suite}_c10d_xccl_test_failed.log"
-    ((num_failed=num_failed_ops_xccl+num_failed_c10d_xccl))
+    cat "./${ut_suite}_xpu_distributed_test_failed.log"
+    ((num_failed=num_failed_xpu_distributed))
     if [[ $num_failed -gt 0 ]]; then
       echo -e "[ERROR] UT ${ut_suite} test Fail"
       exit 1

.github/workflows/_linux_ut.yml

@@ -246,9 +246,7 @@ jobs:
           mkdir -p ut_log/xpu_distributed
           cd ../pytorch/third_party/torch-xpu-ops/test/xpu
           python -c "import torch;print(torch.distributed.is_xccl_available())"
-          # timeout 10000 python run_distributed.py 2>${{ github.workspace }}/ut_log/xpu_distributed/xpu_distributed_test_error.log | tee ${{ github.workspace }}/ut_log/xpu_distributed/xpu_distributed_test.log
-          timeout 10000 python distributed/test_c10d_ops_xccl.py 2>${{ github.workspace }}/ut_log/xpu_distributed/c10d_ops_xccl_test_error.log | tee ${{ github.workspace }}/ut_log/xpu_distributed/c10d_ops_xccl_test.log
-          timeout 10000 python distributed/test_c10d_xccl.py 2>${{ github.workspace }}/ut_log/xpu_distributed/c10d_xccl_test_error.log | tee ${{ github.workspace }}/ut_log/xpu_distributed/c10d_xccl_test.log
+          timeout 10000 python run_distributed.py 2>${{ github.workspace }}/ut_log/xpu_distributed/xpu_distributed_test_error.log | tee ${{ github.workspace }}/ut_log/xpu_distributed/xpu_distributed_test.log
           cd ${{ github.workspace }}
           sudo cp ptrace_scope.bk /proc/sys/kernel/yama/ptrace_scope
       - name: UT Test Results Check
@@ -264,7 +262,7 @@ jobs:
               }
           }
           set -xe
-          for ut_suite in $(echo ${{ inputs.ut }} |sed 's/,/ /g')
+          for ut_suite in $(echo ${{ inputs.ut }} |sed 's/-/ /g')
           do
             contains "op_regression,op_regression_dev1,op_extended,op_ut,torch_xpu,xpu_distributed" $ut_suite
             $contains_status

.github/workflows/pull.yml

@@ -55,7 +55,7 @@ jobs:
     uses: ./.github/workflows/_linux_ut.yml
     with:
       pytorch: ${{ needs.preci-linux-build.outputs.torch_commit_id }}
-      ut: op_regression,op_regression_dev1,op_extended,op_ut
+      ut: op_regression-op_regression_dev1-op_extended-op_ut
       runner: linux.idc.xpu
 
   preci-ut-distributed:

test/xpu/run_distributed.py

@@ -4,7 +4,9 @@
 
 
 def run(test_command):
-    result = subprocess.run(test_command, capture_output=True)
+    result = subprocess.run(test_command, capture_output=True, text=True)
+    print(result.stdout)
+    print(result.stderr)
     if "FAILED" in result.stdout or "FAILED" in result.stderr:
         return 0
     else: