ESP32驱动Apple Lisa硬盘模拟器性能优化求助

问题背景 #

我正在开发一款基于ESP32的Apple Lisa古董电脑硬盘模拟器，当前遇到代码性能瓶颈：模拟器在一款Lisa机型上运行正常，但在另一款硬盘操作速度更快的Lisa机型上无法跟上节奏。问题集中在emulatorRead函数的数据传输部分，高速机型下会丢失1-2个strobe脉冲，导致数据错误。已尝试编译优化从-Os改为-Ofast、改用中断检测strobe（反而更慢）、将关键变量放入IRAM等方法，但均无明显效果。

当前核心代码：

#define busOffset 12 // The parallel bus starts on ESP32 pin 12.
#define CMDPin 21 // CMD is on ESP32 pin 21.
#define STRBPin 25 // STRB is on ESP32 pin 25.

uint32_t IRAM_ATTR prevState = 1; // Variables used for falling-edge detection on the strobe signal.
uint32_t IRAM_ATTR currentState = 1;
uint8_t blockData[532]; // The 532-byte hard disk block that we're currently working with.
volatile uint32_t IRAM_ATTR continueLoop = true;
byte *pointer;

void IRAM_ATTR exitLoopISR(){ // An ISR that clears the continueLoop flag when it's time to stop the data transfer.
  continueLoop = false;
}

void emulatorRead(){
  pointer = blockData; // Make our pointer point to the start of the blockData array.

  REG_WRITE(GPIO_OUT_W1TS_REG, *pointer << busOffset); // Put the first data byte on the bus and increment the value of the pointer.
  REG_WRITE(GPIO_OUT_W1TC_REG, ((byte)~*pointer++ << busOffset)); 

  attachInterrupt(CMDPin, exitLoopISR, FALLING); // Attach a falling-edge interrupt to CMD so that whenever the host computer lowers CMD, exitLoopISR will be called and will make us exit the data transfer loop.

  while(continueLoop){ // Keep transferring data until the host computer lowers CMD.
    currentState = bitRead(REG_READ(GPIO_IN_REG), STRBPin); // Check the current state of the data strobe.
    if(currentState == 0 and prevState == 1){ // If we're on the falling edge of the strobe, it's time to put the next data byte on the bus.
      REG_WRITE(GPIO_OUT_W1TS_REG, *pointer << busOffset); // So go ahead and put it on the bus. And then increment our blockData pointer to the next byte.
      REG_WRITE(GPIO_OUT_W1TC_REG, ((byte)~*pointer++ << busOffset));
    }
    prevState = currentState; // Set the previous strobe state to the current strobe state to prepare for the next iteration of the loop.
  }

  detachInterrupt(CMDPin); // Detach the CMD pin interrupt now that we're done.
}

优化方案 #

1. 消除bitRead的额外开销 #

bitRead函数会引入不必要的运算，直接用位掩码操作读取STRB引脚状态，速度更快：

// 预计算STRB引脚的掩码（全局常量）
const uint32_t STRB_MASK = 1ULL << STRBPin;

// 循环内替换bitRead
currentState = (REG_READ(GPIO_IN_REG) & STRB_MASK) ? 1 : 0;

2. 合并GPIO输出操作，减少寄存器写入次数 #

当前代码分两次调用W1TS/W1TC设置总线，可合并为一次写入GPIO_OUT_REG，大幅降低总线操作延迟：

// 预计算总线引脚的掩码（全局常量）
const uint32_t BUS_MASK = 0xFF << busOffset;

// 替换原有的两次REG_WRITE
uint32_t outVal = REG_READ(GPIO_OUT_REG);
outVal = (outVal & ~BUS_MASK) | ((*pointer << busOffset) & BUS_MASK);
REG_WRITE(GPIO_OUT_REG, outVal);
pointer++;

3. 提前计算不变值，避免重复运算 #

将BUS_MASK、STRB_MASK这类固定值提前定义为全局常量，避免每次循环/传输时重复移位计算。

4. 优化循环结构，减少冗余操作 #

将GPIO_IN_REG的读取放在循环开头，避免重复读取寄存器；同时简化边沿检测逻辑：

while(continueLoop) {
    uint32_t inReg = REG_READ(GPIO_IN_REG);
    uint32_t currState = (inReg & STRB_MASK);
    if (!currState && prevState) {
        // 合并后的GPIO输出操作
        uint32_t outVal = (REG_READ(GPIO_OUT_REG) & ~BUS_MASK) | ((*pointer << busOffset) & BUS_MASK);
        REG_WRITE(GPIO_OUT_REG, outVal);
        pointer++;
    }
    prevState = currState;
}

5. 配置高速IO模式 #

确保STRB引脚和总线引脚都设置为高速驱动模式，降低引脚切换延迟：

// 初始化阶段配置引脚
pinMode(STRBPin, INPUT_PULLUP); // 根据实际电平需求调整
for (int i = busOffset; i < busOffset + 8; i++) {
    pinMode(i, OUTPUT);
    gpio_set_drive_capability((gpio_num_t)i, GPIO_DRIVE_CAP_3); // 启用最大驱动能力
}

6. 优化变量存储位置 #

将pointer变量放入IRAM，避免Flash读取延迟：

byte *IRAM_ATTR pointer;

内容的提问来源于stack exchange，提问作者Alex Anderson-Mcleod