#!/usr/bin/env bash
set -euo pipefail

command -v nvidia-smi >/dev/null && nvidia-smi topo -m || true
command -v nvidia-smi >/dev/null && nvidia-smi nvlink --status || true
command -v nvlink_bandwidth >/dev/null && nvlink_bandwidth || true

if [[ -n "${NCCL_TESTS_DIR:-}" ]]; then
  "$NCCL_TESTS_DIR/build/all_reduce_perf" -b 8 -e 1G -f 2 -g "${GPU_COUNT:-2}"
  if [[ -x "$NCCL_TESTS_DIR/build/alltoall_perf" ]]; then
    "$NCCL_TESTS_DIR/build/alltoall_perf" -b 8 -e 1G -f 2 -g "${GPU_COUNT:-2}"
  fi
else
  echo "NCCL_TESTS_DIR not set; pin NVIDIA/nccl-tests v2.19.6 or an explicit commit"
fi

for tool in ibstat ibv_devinfo ib_write_bw ib_read_bw ib_send_lat ucx_info; do
  if command -v "$tool" >/dev/null; then
    "$tool" --version 2>/dev/null || "$tool" -v 2>/dev/null || "$tool" 2>/dev/null || true
  else
    echo "$tool=missing"
  fi
done

cat <<'NOTE'
NCCL Tests and verbs tools qualify the fabric. They do not prove GLM-5.2
throughput, KV movement, or accepted-task quality. Join their artifacts to the
same host, topology, driver, and time boundary as the workload receipt.
NOTE