#!/bin/bash
#SBATCH --partition=gre
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=1
# NuScenes conversion is CPU-bound. If this partition requires GPU allocation,
# uncomment the next line.
##SBATCH --gres=gpu:1
#SBATCH --time=24:00:00
#SBATCH --job-name=lzh_gpu
#SBATCH --output=%x-%j.out
#SBATCH --mem=128G
#SBATCH --cpus-per-task=16

set -euo pipefail

PROJECT_ROOT="${PROJECT_ROOT:-/mingli01/project/zihanliu/gpudrive-main}"
NUSCENES_ROOT="${NUSCENES_ROOT:-/mingli01/data/nuscenes}"
OUTPUT_DIR="${OUTPUT_DIR:-${PROJECT_ROOT}/data/processed/nuscenes}"
CONDA_ENV_NAME="${CONDA_ENV_NAME:-gpudrive}"

NUSCENES_VERSION="${NUSCENES_VERSION:-v1.0-trainval}"
NUSCENES_SPLIT="${NUSCENES_SPLIT-train}"
SCENE_NAMES="${SCENE_NAMES:-}"
MAX_SCENES="${MAX_SCENES:-}"
TARGET_HZ="${TARGET_HZ:-10.0}"
NUM_STEPS="${NUM_STEPS:-91}"
MAX_ANNOTATION_GAP_SEC="${MAX_ANNOTATION_GAP_SEC:-1.1}"
WINDOW_STRIDE_SEC="${WINDOW_STRIDE_SEC:-}"
FILE_PREFIX="${FILE_PREFIX:-tfrecord-nuscenes}"
MAP_RADIUS_M="${MAP_RADIUS_M:-120.0}"
MAP_DISCRETIZATION_M="${MAP_DISCRETIZATION_M:-1.0}"
MAX_ROAD_ELEMENTS="${MAX_ROAD_ELEMENTS:-900}"
MAX_GEOMETRY_POINTS="${MAX_GEOMETRY_POINTS:-512}"
ALLOW_PARTIAL_WINDOW="${ALLOW_PARTIAL_WINDOW:-false}"
NO_EGO="${NO_EGO:-false}"
NO_MAP="${NO_MAP:-false}"
FAIL_ON_MISSING_MAP="${FAIL_ON_MISSING_MAP:-false}"
QUIET="${QUIET:-true}"

module load conda/3
source "$(conda info --base)/etc/profile.d/conda.sh"
conda activate "${CONDA_ENV_NAME}"

export PYTHONNOUSERSITE=1
export OMP_NUM_THREADS="${OMP_NUM_THREADS:-${SLURM_CPUS_PER_TASK:-1}}"
export MKL_NUM_THREADS="${MKL_NUM_THREADS:-${SLURM_CPUS_PER_TASK:-1}}"

cd "${PROJECT_ROOT}"
export PYTHONPATH="${PROJECT_ROOT}:${PYTHONPATH:-}"

if [[ ! -d "${PROJECT_ROOT}" ]]; then
  echo "ERROR: PROJECT_ROOT does not exist: ${PROJECT_ROOT}"
  exit 1
fi

if [[ ! -f "${PROJECT_ROOT}/data_utils/process_nuscenes_files.py" ]]; then
  echo "ERROR: NuScenes converter not found under PROJECT_ROOT=${PROJECT_ROOT}"
  exit 1
fi

if [[ ! -d "${NUSCENES_ROOT}" ]]; then
  echo "ERROR: NUSCENES_ROOT does not exist: ${NUSCENES_ROOT}"
  exit 1
fi

mkdir -p "${OUTPUT_DIR}"

echo "SLURM_JOB_ID=${SLURM_JOB_ID:-unknown}"
echo "PROJECT_ROOT=${PROJECT_ROOT}"
echo "NUSCENES_ROOT=${NUSCENES_ROOT}"
echo "OUTPUT_DIR=${OUTPUT_DIR}"
echo "CONDA_ENV_NAME=${CONDA_ENV_NAME}"
echo "NUSCENES_VERSION=${NUSCENES_VERSION}"
echo "NUSCENES_SPLIT=${NUSCENES_SPLIT:-<all>}"
echo "SCENE_NAMES=${SCENE_NAMES:-<empty>}"
echo "MAX_SCENES=${MAX_SCENES:-<empty>}"
echo "TARGET_HZ=${TARGET_HZ}"
echo "NUM_STEPS=${NUM_STEPS}"
echo "WINDOW_STRIDE_SEC=${WINDOW_STRIDE_SEC:-<single-window>}"
echo "NO_MAP=${NO_MAP}"
echo "HOSTNAME=$(hostname)"
echo "PWD=$(pwd)"
echo "SLURM_CPUS_PER_TASK=${SLURM_CPUS_PER_TASK:-unset}"
echo "OMP_NUM_THREADS=${OMP_NUM_THREADS}"
echo "MKL_NUM_THREADS=${MKL_NUM_THREADS}"

CMD=(
  python data_utils/process_nuscenes_files.py
  --dataroot "${NUSCENES_ROOT}"
  --version "${NUSCENES_VERSION}"
  --output-dir "${OUTPUT_DIR}"
  --file-prefix "${FILE_PREFIX}"
  --target-hz "${TARGET_HZ}"
  --num-steps "${NUM_STEPS}"
  --max-annotation-gap-sec "${MAX_ANNOTATION_GAP_SEC}"
  --map-radius-m "${MAP_RADIUS_M}"
  --map-discretization-m "${MAP_DISCRETIZATION_M}"
  --max-road-elements "${MAX_ROAD_ELEMENTS}"
  --max-geometry-points "${MAX_GEOMETRY_POINTS}"
)

if [[ -n "${NUSCENES_SPLIT}" ]]; then
  CMD+=(--split "${NUSCENES_SPLIT}")
fi

if [[ -n "${SCENE_NAMES}" ]]; then
  CMD+=(--scene-names "${SCENE_NAMES}")
fi

if [[ -n "${MAX_SCENES}" ]]; then
  CMD+=(--max-scenes "${MAX_SCENES}")
fi

if [[ -n "${WINDOW_STRIDE_SEC}" ]]; then
  CMD+=(--window-stride-sec "${WINDOW_STRIDE_SEC}")
fi

if [[ "${ALLOW_PARTIAL_WINDOW}" == "true" ]]; then
  CMD+=(--allow-partial-window)
fi

if [[ "${NO_EGO}" == "true" ]]; then
  CMD+=(--no-ego)
fi

if [[ "${NO_MAP}" == "true" ]]; then
  CMD+=(--no-map)
fi

if [[ "${FAIL_ON_MISSING_MAP}" == "true" ]]; then
  CMD+=(--fail-on-missing-map)
fi

if [[ "${QUIET}" == "true" ]]; then
  CMD+=(--quiet)
fi

echo "COMMAND_BEGIN"
printf '%q ' "${CMD[@]}"
echo
echo "COMMAND_END"

"${CMD[@]}"