"""Validate fused SwiGLU and native NVFP4 production on real H3 FC1 output."""

from __future__ import annotations

import argparse
import json
import time
from pathlib import Path

import torch

from h3_blackwell_runtime.adaln import H3CurveAdaLN
from h3_blackwell_runtime.block import H3DiTBlock
from h3_blackwell_runtime.checkpoint import H3Checkpoint
from h3_blackwell_runtime.nvfp4_quant import (
    nvfp4_activation_scale,
    vortex_native_quantize_swiglu_nvfp4,
)
from h3_blackwell_runtime.packing import H3PromptPacker
from h3_blackwell_runtime.rope import h3_rope_rotation
from h3_blackwell_runtime.sampler import _audio_sigma, _model_sigma, beta_sigmas
from h3_blackwell_runtime.t2v import random_av_latents


def parse_args() -> argparse.Namespace:
    parser = argparse.ArgumentParser(description=__doc__)
    parser.add_argument("--model-path", default="/models/minimax_h3_fl2va_pruned_nvfp4.safetensors")
    parser.add_argument("--output", type=Path, required=True)
    parser.add_argument("--block-index", type=int, required=True)
    parser.add_argument("--width", type=int, default=1344)
    parser.add_argument("--height", type=int, default=768)
    parser.add_argument("--frames", type=int, default=124)
    parser.add_argument("--steps", type=int, default=12)
    parser.add_argument("--sampler-step", type=int, default=1)
    parser.add_argument("--seed", type=int, default=440420)
    parser.add_argument("--text-tokens", type=int, default=100)
    parser.add_argument("--warmup", type=int, default=3)
    parser.add_argument("--iterations", type=int, default=10)
    parser.add_argument("--device", default="cuda")
    return parser.parse_args()


def main() -> None:
    args = parse_args()
    torch.manual_seed(args.seed)
    checkpoint = H3Checkpoint(args.model_path, device=args.device)
    block = H3DiTBlock.from_checkpoint(
        checkpoint, args.block_index, attention_backend="sage2",
    ).eval()
    adaln = H3CurveAdaLN.from_checkpoint(
        checkpoint, f"blocks.{args.block_index}.adaln_proj",
    ).eval()
    packer = H3PromptPacker(checkpoint)
    video, audio, _ = random_av_latents(
        args.width, args.height, args.frames, args.seed, device=args.device,
    )
    sigma = beta_sigmas(args.steps, device=args.device)[args.sampler_step - 1]
    native_audio = audio.to(torch.bfloat16) * (_audio_sigma(sigma) / sigma)
    text = torch.randn(
        1, args.text_tokens, 5376, device=args.device, dtype=torch.bfloat16,
    )
    hidden, timesteps, segments, positions, _, _ = packer(
        text, video, native_audio, _model_sigma(sigma),
    )
    rotation = h3_rope_rotation(
        positions.to(args.device),
        checkpoint.tensor("rope.inv_freq", dtype=torch.float32),
        hidden.dtype,
    )
    modulation = tuple(value.detach() for value in adaln(timesteps))
    captured = []
    block.fused_nvfp4_modulation = False
    hook = block.mlp.fc1.register_forward_hook(
        lambda _module, _inputs, output: captured.append(output.detach()),
    )
    with torch.inference_mode():
        block(hidden, rotation, *modulation, segments)
    hook.remove()
    gate_up = captured[-1]

    with torch.inference_mode():
        gate, up = gate_up.chunk(2, dim=-1)
        materialized = torch.nn.functional.silu(gate).mul_(up)
        reference_scale = nvfp4_activation_scale(materialized).float()
        import comfy_kitchen as ck
        from comfy_kitchen.tensor import TensorCoreNVFP4Layout

        reference_qdata, reference_sfa = ck.quantize_nvfp4(
            materialized,
            reference_scale,
            pad_16x=TensorCoreNVFP4Layout.get_padded_shape(tuple(materialized.shape))
            != tuple(materialized.shape),
        )
        actual_scale, actual_qdata, actual_sfa = vortex_native_quantize_swiglu_nvfp4(
            gate_up,
        )
        torch.cuda.synchronize()
        for _ in range(args.warmup):
            vortex_native_quantize_swiglu_nvfp4(gate_up)
        times = []
        for _ in range(args.iterations):
            torch.cuda.synchronize()
            started = time.perf_counter()
            vortex_native_quantize_swiglu_nvfp4(gate_up)
            torch.cuda.synchronize()
            times.append(time.perf_counter() - started)

    report = {
        "device": torch.cuda.get_device_name(),
        "block_index": args.block_index,
        "input_shape": list(gate_up.shape),
        "output_shape": list(materialized.shape),
        "scale_equal": torch.equal(actual_scale, reference_scale),
        "scale_reference": reference_scale.item(),
        "scale_actual": actual_scale.item(),
        "qdata_differences": torch.count_nonzero(actual_qdata != reference_qdata).item(),
        "sfa_differences": torch.count_nonzero(
            actual_sfa.view(torch.uint8) != reference_sfa.view(torch.uint8)
        ).item(),
        "producer_p50_ms": sorted(times)[len(times) // 2] * 1000.0,
    }
    report["equal"] = (
        report["scale_equal"]
        and report["qdata_differences"] == 0
        and report["sfa_differences"] == 0
    )
    args.output.parent.mkdir(parents=True, exist_ok=True)
    args.output.write_text(json.dumps(report, indent=2) + "\n", encoding="utf-8")
    print(json.dumps(report, indent=2), flush=True)
    if not report["equal"]:
        raise RuntimeError("fused SwiGLU producer is not byte-exact")


if __name__ == "__main__":
    main()