digraph G {
  graph [rankdir=TB, bgcolor="transparent", pad=0.15, nodesep=0.28, ranksep=0.38,
         fontname="Noto Sans CJK SC"];
  node [shape=box, style="rounded,filled", fontname="Noto Sans CJK SC", fontsize=10,
        color="#44546A", fillcolor="#F7F8FA", margin="0.12,0.08"];
  edge [fontname="Noto Sans CJK SC", fontsize=9, color="#596777", arrowsize=0.7];

  cfg [label="CPU 一次配置\n基址 / 组数 / stride", fillcolor="#EAF2F8"];
  read [label="DMA 读取 128 B\nA、B → 空闲 Bank", fillcolor="#E8F3E8"];
  calc [label="4 路 32x32 LUT 乘法\n每拍形成一个输出元素的 4 项乘积", fillcolor="#FFF1D6"];
  write [label="DMA 写回 192 B\nC → ExtRAM", fillcolor="#E8F3E8"];
  crc [label="硬件流水 CRC32\n不写回时仍可判定整批结果", fillcolor="#F8E6EA"];
  next [label="Bank 翻转 / group+1", shape=diamond, fillcolor="#F2F2F2"];
  done [label="done + dma_cycles", fillcolor="#D9EAF7"];

  { rank=same; read; calc; write; }
  cfg -> read;
  read -> calc -> write;
  write -> crc -> next;
  next -> read [label="未完成", constraint=false];
  next -> done [label="全部完成"];
}