|
| 1 | +#!/usr/bin/env bash |
| 2 | + |
| 3 | +# Hygon DCU-specific unit-test environment setup. |
| 4 | + |
| 5 | +export CUDA_PATH="${CUDA_PATH:-/opt/dtk/cuda/cuda-12}" |
| 6 | +export CUDA_HOME="${CUDA_HOME:-$CUDA_PATH}" |
| 7 | + |
| 8 | +if [[ -f /opt/dtk/env.sh ]]; then |
| 9 | + # shellcheck source=/dev/null |
| 10 | + source /opt/dtk/env.sh |
| 11 | +fi |
| 12 | + |
| 13 | +FLAGCX_CI_MPI_BASE_HOME=${MPI_HOME:-/opt/mpi} |
| 14 | + |
| 15 | +# Use the real OpenMPI launcher if the image provides a wrapper. |
| 16 | +if [[ -x "$FLAGCX_CI_MPI_BASE_HOME/bin/mpirun.real" ]]; then |
| 17 | + FLAGCX_CI_MPI_HOME=$(mktemp -d) |
| 18 | + mkdir -p "$FLAGCX_CI_MPI_HOME/bin" |
| 19 | + ln -s "$FLAGCX_CI_MPI_BASE_HOME/bin/mpirun.real" \ |
| 20 | + "$FLAGCX_CI_MPI_HOME/bin/mpirun" |
| 21 | + ln -s "$FLAGCX_CI_MPI_BASE_HOME/include" "$FLAGCX_CI_MPI_HOME/include" |
| 22 | + ln -s "$FLAGCX_CI_MPI_BASE_HOME/lib" "$FLAGCX_CI_MPI_HOME/lib" |
| 23 | + export MPI_HOME=$FLAGCX_CI_MPI_HOME |
| 24 | +else |
| 25 | + export MPI_HOME=$FLAGCX_CI_MPI_BASE_HOME |
| 26 | +fi |
| 27 | + |
| 28 | +export FLAGCX_ADAPTOR=du |
| 29 | +export USE_DU=1 |
| 30 | + |
| 31 | +FLAGCX_CI_COMMON_MAKE_ARGS=( |
| 32 | + USE_DU=1 |
| 33 | + DEVICE_HOME="$CUDA_PATH" |
| 34 | + CCL_HOME="$CUDA_PATH" |
| 35 | +) |
| 36 | + |
| 37 | +# The DU makefile does not currently pull the default device-api backend into |
| 38 | +# libflagcx.so by itself, so CI passes it through as an extra source. |
| 39 | +FLAGCX_CI_PROJECT_MAKE_ARGS=( |
| 40 | + "${FLAGCX_CI_COMMON_MAKE_ARGS[@]}" |
| 41 | + PLATFORM_EXTRA_SRCS=flagcx/adaptor/device_api/default_dev_api_backend.cc |
| 42 | +) |
| 43 | +FLAGCX_CI_TEST_MAKE_ARGS=("${FLAGCX_CI_COMMON_MAKE_ARGS[@]}") |
| 44 | + |
| 45 | +FLAGCX_CI_INTRA_NP=8 |
| 46 | +FLAGCX_CI_NODE_NP=4 |
| 47 | +FLAGCX_CI_RUNNER_NP=8 |
| 48 | +export NP=8 |
| 49 | + |
| 50 | +flagcx_ci_configure_suite() { |
| 51 | + local suite=$1 |
| 52 | + |
| 53 | + case "$suite" in |
| 54 | + device_api) |
| 55 | + FLAGCX_CI_PROJECT_MAKE_ARGS+=(COMPILE_KERNEL=1) |
| 56 | + FLAGCX_CI_TEST_MAKE_ARGS+=(COMPILE_KERNEL=1) |
| 57 | + ;; |
| 58 | + rma) |
| 59 | + FLAGCX_CI_TEST_MAKE_ARGS+=( |
| 60 | + "HETERO_ENV=-x FLAGCX_USE_HETERO_COMM=1 -x FLAGCX_MEM_ENABLE=1 -x FLAGCX_VMM_ENABLE=0" |
| 61 | + ) |
| 62 | + ;; |
| 63 | + esac |
| 64 | +} |
| 65 | + |
| 66 | +flagcx_ci_prepare() { |
| 67 | + local suite=$1 |
| 68 | + echo "Preparing Hygon DCU environment for unit-test suite: $suite" |
| 69 | + command -v mpirun |
| 70 | + command -v nvcc |
| 71 | + mpirun --version |
| 72 | + nvcc --version |
| 73 | + hy-smi --showproductname || true |
| 74 | +} |
| 75 | + |
| 76 | +flagcx_ci_build_suite_override() { |
| 77 | + local suite=$1 |
| 78 | + local suite_dir=${2:-} |
| 79 | + shift 2 || true |
| 80 | + local -a args=("$@") |
| 81 | + |
| 82 | + if [[ "$suite" == "device_api" ]]; then |
| 83 | + FLAGCX_CI_BUILD_SUITE_OVERRIDE_HANDLED=1 |
| 84 | + echo "Skipping Hygon device_api build: DU test kernels do not provide all launchers required by the current device_api tests." |
| 85 | + return |
| 86 | + fi |
| 87 | + |
| 88 | + if [[ "$suite" == "symmem" ]]; then |
| 89 | + FLAGCX_CI_BUILD_SUITE_OVERRIDE_HANDLED=1 |
| 90 | + cmake -S "$PROJECT_ROOT/third-party/googletest" \ |
| 91 | + -B "$PROJECT_ROOT/third-party/googletest/build" |
| 92 | + cmake --build "$PROJECT_ROOT/third-party/googletest/build" --parallel "$(nproc)" |
| 93 | + make -C "$suite_dir" --jobs="$(nproc)" "${args[@]}" |
| 94 | + return |
| 95 | + fi |
| 96 | + |
| 97 | + FLAGCX_CI_BUILD_SUITE_OVERRIDE_HANDLED=0 |
| 98 | +} |
| 99 | + |
| 100 | +flagcx_ci_run_suite_override() { |
| 101 | + local suite=$1 |
| 102 | + local suite_dir=$2 |
| 103 | + shift 2 |
| 104 | + local -a args=("$@") |
| 105 | + |
| 106 | + if [[ "$suite" == "device_api" ]]; then |
| 107 | + FLAGCX_CI_RUN_SUITE_OVERRIDE_HANDLED=1 |
| 108 | + echo "Skipping Hygon device_api tests: DU launcher coverage is incomplete in the current test kernels." |
| 109 | + return |
| 110 | + fi |
| 111 | + |
| 112 | + if [[ "$suite" == "runner" ]]; then |
| 113 | + FLAGCX_CI_RUN_SUITE_OVERRIDE_HANDLED=1 |
| 114 | + make -C "$suite_dir" run-unit "${args[@]}" |
| 115 | + cd "$suite_dir" |
| 116 | + mpirun -np "$FLAGCX_CI_RUNNER_NP" --allow-run-as-root \ |
| 117 | + ./build/bin/runner_mpi_tests |
| 118 | + echo "Skipping Hygon runner hetero-mode MPI variants: current DU path stalls in the single-node hetero simulation." |
| 119 | + return |
| 120 | + fi |
| 121 | + |
| 122 | + FLAGCX_CI_RUN_SUITE_OVERRIDE_HANDLED=0 |
| 123 | +} |
0 commit comments