mooncake-track
/
Mooncake
0
0
Fork 12
Code Issues Pull Requests 13 Actions Packages Projects Releases Wiki Activity

[CI] Add hixl roce samples on ASCEND platforms. (#1697) 

---------

Co-authored-by: ZhaoBaiwei <zhaobaiwei@huawei.com>
This commit is contained in:
1180300720 2026-03-21 00:39:41 +08:00 committed by GitHub
parent d2bff9b9a8
commit fdb7040787
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 70 additions and 21 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

91
.github/workflows/ci_ascend.yml vendored
View File

@ -130,37 +130,86 @@ jobs:
"batch_put_get_multi_buffers_sample.py"
)
# List of test scenarios (HCCL_INTRA_ROCE_ENABLE settings)
TEST_SCENARIOS=(
"HCCL_INTRA_ROCE_ENABLE=1"
"HCCL_INTRA_ROCE_ENABLE_UNSET"
)
# Track test results
FAILED_TESTS=()
PASSED_TESTS=()
# Run each test case
export ASCEND_BUFFER_POOL=4:8
for test_case in "${TEST_CASES[@]}"; do
# Run each test scenario
for scenario in "${TEST_SCENARIOS[@]}"; do
echo ""
echo "========================================="
echo "Running test: $test_case"
echo "Running scenario: $scenario"
echo "========================================="
if [ ! -f "$test_case" ]; then
echo "Warning: Test file $test_case not found, skipping..."
continue
fi
# Run the test with run.sh
python3 $test_case \
--device_id=0 \
--rank=0 \
2>&1 | tee "/tmp/hixl_test_${test_case%.py}.log"
TEST_RESULT=${PIPESTATUS[0]}
if [ $TEST_RESULT -eq 0 ]; then
echo "✓ $test_case PASSED"
PASSED_TESTS+=("$test_case")
# Configure environment variables for the current scenario
if [ "$scenario" = "HCCL_INTRA_ROCE_ENABLE=1" ]; then
export HCCL_INTRA_ROCE_ENABLE=1
unset ASCEND_BUFFER_POOL
echo "HCCL_INTRA_ROCE_ENABLE is set to 1, ASCEND_BUFFER_POOL is unset"
else
echo "✗ $test_case FAILED with exit code: $TEST_RESULT"
FAILED_TESTS+=("$test_case")
unset HCCL_INTRA_ROCE_ENABLE
export ASCEND_BUFFER_POOL=4:8
echo "HCCL_INTRA_ROCE_ENABLE is not set, ASCEND_BUFFER_POOL is set to 4:8"
fi
# Run each test case in the current scenario
for test_case in "${TEST_CASES[@]}"; do
echo ""
echo "-----------------------------------------"
echo "Test: $test_case"
echo "-----------------------------------------"
if [ ! -f "$test_case" ]; then
echo "Warning: Test file $test_case not found, skipping..."
continue
fi
# Run the test with 2 devices in distributed mode
# Run rank 0 on device 0
python3 $test_case \
--device_id=0 \
--rank=0 \
--world_size=2 \
--distributed \
2>&1 | tee "/tmp/hixl_test_${scenario//=/}_${test_case%.py}_rank0.log" &
PID0=$!
# Run rank 1 on device 1
python3 $test_case \
--device_id=2 \
--rank=1 \
--world_size=2 \
--distributed \
2>&1 | tee "/tmp/hixl_test_${scenario//=/}_${test_case%.py}_rank1.log" &
PID1=$!
# Wait for both processes to complete
wait $PID0
TEST_RESULT0=$?
wait $PID1
TEST_RESULT1=$?
# Check test results
if [ $TEST_RESULT0 -eq 0 ] && [ $TEST_RESULT1 -eq 0 ]; then
echo "✓ $test_case PASSED (scenario: $scenario)"
PASSED_TESTS+=("$scenario:$test_case")
else
echo "✗ $test_case FAILED (scenario: $scenario)"
if [ $TEST_RESULT0 -ne 0 ]; then
echo " Rank 0 failed with code: $TEST_RESULT0"
fi
if [ $TEST_RESULT1 -ne 0 ]; then
echo " Rank 1 failed with code: $TEST_RESULT1"
fi
FAILED_TESTS+=("$scenario:$test_case")
fi
done
done
echo ""