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

if [[ $# -lt 1 ]]; then
  echo "usage: $0 command [args ...]" >&2
  exit 2
fi

run_id="${RUN_ID:-$(date -u +%Y%m%dT%H%M%SZ)-unverified}"
out="${ARTIFACT_DIR:-artifacts/$run_id}"
mkdir -p "$out"

nvidia-smi -q > "$out/nvidia-smi-q.txt"
nvidia-smi topo -m > "$out/nvidia-smi-topo.txt"
dcgmi discovery -l > "$out/dcgm-discovery.txt" 2>&1 || true

nsys profile \
  --trace=cuda,nvtx,osrt \
  --sample=none \
  --force-overwrite=true \
  --output="$out/timeline" \
  "$@"

printf 'run_id=%s\ncommand=' "$run_id" > "$out/run.txt"
printf '%q ' "$@" >> "$out/run.txt"
printf '\n' >> "$out/run.txt"

cat <<NOTE
Captured $out/timeline.nsys-rep.
Run Nsight Compute only on a selected kernel, for example:
  ncu --set full --kernel-name regex:<name> --export $out/kernel -- <command>
Run correctness separately:
  compute-sanitizer --tool memcheck <kernel-test>
  compute-sanitizer --tool racecheck <kernel-test>
NOTE