#!/usr/bin/env bash

set -euo pipefail

OUTPUT_ROOT="${OUTPUT_ROOT:?OUTPUT_ROOT is required}"
FRONTIER_SOURCE="${FRONTIER_SOURCE:-/home/admin/cpfs/wjh/aituner/frontier-q235-v020-5b953f5}"
VENV_ROOT="${VENV_ROOT:-/tmp/wjh/venvs/vllm-0.20.0-cu129-profiler-v1}"
RUNNER_DIR="${RUNNER_DIR:-$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)}"
MODEL=Qwen3-235B-A22B
TOKENS=(1 2 4 8 16 24 32 40 48 56 64 72 80 88 96 104 112 120 128 256 512 1024 2048 4096 8192)
BATCHES=(1 2 4 8 16 24 32 40 48 56 64 72 80 88 96 104 112 120 128)
KV=(128 1024 2048 4096 8192 16384 32768 40960)

mkdir -p "${OUTPUT_ROOT}/logs" "${OUTPUT_ROOT}/provenance"
exec > >(tee -a "${OUTPUT_ROOT}/controller.log") 2>&1

echo "Q235_PROFILE_LAUNCH_ECHO host=dash0 model=${MODEL} vllm=0.20.0 frontier=$(git -C "${FRONTIER_SOURCE}" rev-parse HEAD) device=H20 methods={CUDA_EVENT,KERNEL_ONLY} attention_tp={4,8} moe_paths={TP4/EP1:Triton,TP1/EP8:FlashInfer-CUTLASS} token_points=${#TOKENS[@]} batch_points=${#BATCHES[@]} kv_points=${#KV[@]} parallel_gpus=6 expected_wall=20-75m expected_cost=2-7_H20-GPUh output=${OUTPUT_ROOT}"
date -u +START_UTC=%Y-%m-%dT%H:%M:%SZ
nvidia-smi --query-gpu=index,memory.used --format=csv,noheader,nounits \
  | awk '$2 > 16 {exit 1}'
git -C "${FRONTIER_SOURCE}" rev-parse HEAD > "${OUTPUT_ROOT}/provenance/frontier.commit"
"${VENV_ROOT}/bin/vllm" --version > "${OUTPUT_ROOT}/provenance/vllm.version"

common_profile() {
  local gpu="$1" method="$2" root="$3"
  env CUDA_VISIBLE_DEVICES="${gpu}" PYTHONPATH="${FRONTIER_SOURCE}" \
    "${VENV_ROOT}/bin/python" -m frontier.profiling.linear_op.main \
      --disable_ray --num_gpus 1 --device h20 --output_dir "${root}" --models "${MODEL}" \
      --num_tensor_parallel_workers 1 4 8 --attn_tp 4 8 --ffn_tp 1 4 \
      --num_tokens_list "${TOKENS[@]}" --profile_method "${method}" --yes
  env CUDA_VISIBLE_DEVICES="${gpu}" PYTHONPATH="${FRONTIER_SOURCE}" \
    "${VENV_ROOT}/bin/python" -m frontier.profiling.attention.main \
      --disable_ray --num_gpus 1 --device h20 --output_dir "${root}" --models "${MODEL}" \
      --num_tensor_parallel_workers 4 8 --max_model_len 40960 --max_seq_len 8192 \
      --batch_size_list "${BATCHES[@]}" --decode_kv_cache_size_list "${KV[@]}" \
      --fixed_chunked_prefill_size 8192 --attention_backend FLASHINFER \
      --profile_method "${method}" --yes
}

moe_profile() {
  local gpu="$1" method="$2" tp="$3" ep="$4" root="$5"
  env CUDA_VISIBLE_DEVICES="${gpu}" PYTHONPATH="${FRONTIER_SOURCE}" \
    "${VENV_ROOT}/bin/python" -m frontier.profiling.moe.main \
      --disable_ray --num_gpus 1 --device h20 --output_dir "${root}" --models "${MODEL}" \
      --num_tensor_parallel_workers "${tp}" --expert_parallel_sizes "${ep}" \
      --num_tokens_list "${TOKENS[@]}" --load_distributions uniform \
      --num_samples_per_distribution 1 --profile_method "${method}" --yes
}

declare -a pids=()
common_profile 0 cuda_event "${OUTPUT_ROOT}/cuda-common" > "${OUTPUT_ROOT}/logs/cuda-common.log" 2>&1 & pids+=("$!")
common_profile 1 record_function "${OUTPUT_ROOT}/kernel-common" > "${OUTPUT_ROOT}/logs/kernel-common.log" 2>&1 & pids+=("$!")
moe_profile 2 cuda_event 4 1 "${OUTPUT_ROOT}/cuda-moe-tp4" > "${OUTPUT_ROOT}/logs/cuda-moe-tp4.log" 2>&1 & pids+=("$!")
moe_profile 3 cuda_event 1 8 "${OUTPUT_ROOT}/cuda-moe-ep8" > "${OUTPUT_ROOT}/logs/cuda-moe-ep8.log" 2>&1 & pids+=("$!")
moe_profile 4 record_function 4 1 "${OUTPUT_ROOT}/kernel-moe-tp4" > "${OUTPUT_ROOT}/logs/kernel-moe-tp4.log" 2>&1 & pids+=("$!")
moe_profile 5 record_function 1 8 "${OUTPUT_ROOT}/kernel-moe-ep8" > "${OUTPUT_ROOT}/logs/kernel-moe-ep8.log" 2>&1 & pids+=("$!")
failed=0
for pid in "${pids[@]}"; do wait "${pid}" || failed=1; done
[[ "${failed}" -eq 0 ]] || { tail -80 "${OUTPUT_ROOT}"/logs/*.log; exit 1; }

"${VENV_ROOT}/bin/python" "${RUNNER_DIR}/assemble_qwen235_v020_profiles.py" \
  --cuda-common "${OUTPUT_ROOT}/cuda-common" \
  --cuda-moe-tp4 "${OUTPUT_ROOT}/cuda-moe-tp4" --cuda-moe-ep8 "${OUTPUT_ROOT}/cuda-moe-ep8" \
  --kernel-common "${OUTPUT_ROOT}/kernel-common" \
  --kernel-moe-tp4 "${OUTPUT_ROOT}/kernel-moe-tp4" --kernel-moe-ep8 "${OUTPUT_ROOT}/kernel-moe-ep8" \
  --output-root "${OUTPUT_ROOT}/frozen"
date -u +END_UTC=%Y-%m-%dT%H:%M:%SZ
echo Q235_V020_PROFILES_COMPLETE