add rocm test (#64639)

risemeup1 · web-flow · commit 86d347b67c9d · 2024-06-03T09:39:48.000+08:00
* add rocm test

* add rocm test

* fix

* fix

* fix

* fix,test=ducoment_fix
diff --git a/paddle/scripts/paddle_build.sh b/paddle/scripts/paddle_build.sh
@@ -1474,7 +1474,7 @@ function card_test() {
     if [ "${WITH_XPU}" == "ON" ];then
         CUDA_DEVICE_COUNT=1
     elif [ "${WITH_ROCM}" == "ON" ];then
-        CUDA_DEVICE_COUNT=$(rocm-smi -i | grep GPU | wc -l)
+        CUDA_DEVICE_COUNT=$(rocm-smi -i | grep DCU | wc -l)
     elif [ "${WITH_IPU}" == "ON" ];then
         CUDA_DEVICE_COUNT=1
     else
@@ -1517,13 +1517,22 @@ function card_test() {
             if [[ $cardnumber == $CUDA_DEVICE_COUNT ]]; then
                 (ctest -I $i,,$NUM_PROC -R "($testcases)" -E "($disable_ut_quickly)" ${run_label_mode} -V --timeout 120 -j $parallel_job | tee $tmpfile; test ${PIPESTATUS[0]} -eq 0) &
             else
-                (env CUDA_VISIBLE_DEVICES=$cuda_list ctest -I $i,,$NUM_PROC -R "($testcases)" -E "($disable_ut_quickly)" ${run_label_mode} --timeout 120 -V -j $parallel_job | tee $tmpfile; test ${PIPESTATUS[0]} -eq 0) &
+                if [ "$WITH_ROCM" == "ON" ];then
+                    (env HIP_VISIBLE_DEVICES=$cuda_list ctest -I $i,,$NUM_PROC -R "($testcases)" -E "($disable_ut_quickly)" ${run_label_mode} --timeout 120 -V -j $parallel_job | tee $tmpfile; test ${PIPESTATUS[0]} -eq 0) &
+                else
+                    (env CUDA_VISIBLE_DEVICES=$cuda_list ctest -I $i,,$NUM_PROC -R "($testcases)" -E "($disable_ut_quickly)" ${run_label_mode} --timeout 120 -V -j $parallel_job | tee $tmpfile; test ${PIPESTATUS[0]} -eq 0) &
+                fi
             fi
         else
             if [[ $cardnumber == $CUDA_DEVICE_COUNT ]]; then
                 (ctest -I $i,,$NUM_PROC -R "($testcases)" -E "($disable_ut_quickly)" ${run_label_mode} --timeout 120 --output-on-failure  -j $parallel_job | tee $tmpfile; test ${PIPESTATUS[0]} -eq 0) &
             else
-                (env CUDA_VISIBLE_DEVICES=$cuda_list ctest -I $i,,$NUM_PROC -R "($testcases)" -E "($disable_ut_quickly)" ${run_label_mode} --timeout 120 --output-on-failure  -j $parallel_job | tee $tmpfile; test ${PIPESTATUS[0]} -eq 0) &
+                if [ "$WITH_ROCM" == "ON" ];then
+                    (env HIP_VISIBLE_DEVICES=$cuda_list ctest -I $i,,$NUM_PROC -R "($testcases)" -E "($disable_ut_quickly)" ${run_label_mode} --timeout 120 --output-on-failure  -j $parallel_job | tee $tmpfile; test ${PIPESTATUS[0]} -eq 0) &
+                else
+                    (env CUDA_VISIBLE_DEVICES=$cuda_list ctest -I $i,,$NUM_PROC -R "($testcases)" -E "($disable_ut_quickly)" ${run_label_mode} --timeout 120 --output-on-failure  -j $parallel_job | tee $tmpfile; test ${PIPESTATUS[0]} -eq 0) &
+                fi
+                
             fi
         fi
     done
@@ -2652,7 +2661,11 @@ set -x
         fi
         if [ -a "$PADDLE_ROOT/added_ut" ];then
             added_uts=^$(awk BEGIN{RS=EOF}'{gsub(/\n/,"$|^");print}' $PADDLE_ROOT/added_ut)$
-            env CUDA_VISIBLE_DEVICES=0 ctest -R "(${added_uts})" -LE "RUN_TYPE=DIST|RUN_TYPE=EXCLUSIVE|RUN_TYPE=HYBRID" --output-on-failure --repeat-until-fail 3 --timeout 15;added_ut_error=$?
+            if [ "$WITH_ROCM" == "ON" ];then
+                env HIP_VISIBLE_DEVICES=0 ctest -R "(${added_uts})" -LE "RUN_TYPE=DIST|RUN_TYPE=EXCLUSIVE|RUN_TYPE=HYBRID" --output-on-failure --repeat-until-fail 3 --timeout 15;added_ut_error=$?
+            else
+                env CUDA_VISIBLE_DEVICES=0 ctest -R "(${added_uts})" -LE "RUN_TYPE=DIST|RUN_TYPE=EXCLUSIVE|RUN_TYPE=HYBRID" --output-on-failure --repeat-until-fail 3 --timeout 15;added_ut_error=$?
+            fi
             ctest -R "(${added_uts})" -L "RUN_TYPE=DIST|RUN_TYPE=EXCLUSIVE" --output-on-failure --repeat-until-fail 3 --timeout 15;added_ut_error_1=$?
             if [ "$added_ut_error" != 0 ] && [ "$added_ut_error_1" != 0 ];then
                 echo "========================================"
@@ -2826,7 +2839,9 @@ set +x
         rerun_ut_endTime_s=`date +%s`
         echo "ipipe_log_param_Rerun_TestCases_Total_Time: $[ $rerun_ut_endTime_s - $rerun_ut_startTime_s ]s"
         echo "ipipe_log_param_Rerun_TestCases_Total_Time: $[ $rerun_ut_endTime_s - $rerun_ut_startTime_s ]s" >> ${PADDLE_ROOT}/build/build_summary.txt
-        cp $PADDLE_ROOT/build/Testing/Temporary/CTestCostData.txt ${cfs_dir}/coverage/${AGILE_PULL_ID}/${AGILE_REVISION}/
+        if [ "$WITH_ROCM" != "ON" ];then
+            cp $PADDLE_ROOT/build/Testing/Temporary/CTestCostData.txt ${cfs_dir}/coverage/${AGILE_PULL_ID}/${AGILE_REVISION}/
+        fi
         if [[ "$EXIT_CODE" != "0" ]]; then
             show_ut_retry_result
         fi
@@ -4475,6 +4490,9 @@ function main() {
         export FLAGS_PIR_OPTEST=True
         parallel_test true
         ;;
+      hyg_dcu_test)
+        parallel_test
+        ;;
       nv_cicheck_coverage)
         parallel_test
         nv_test
