STM32H7xx尽可能快地切换IO

Question

我正在尝试尽可能快地切换STM32H743上的IO。我使用的是外部10 the时钟，供电电压为3.3V，我确信我的主时钟运行在400 the，与GPIO (AHB4)对话的总线时钟运行在200 the。下面是我用来配置芯片和切换IO的一些示例代码：

RCC_ClkInitTypeDef clock;
RCC_OscInitTypeDef osc;

MODIFY_REG(PWR->CR3, PWR_CR3_SCUEN, 0);
__HAL_PWR_VOLTAGESCALING_CONFIG(PWR_REGULATOR_VOLTAGE_SCALE1);
while (!__HAL_PWR_GET_FLAG(PWR_FLAG_VOSRDY)) {}
// External 10.000MHz oscillator
osc.OscillatorType = RCC_OSCILLATORTYPE_HSE;
osc.HSEState = RCC_HSE_ON;
osc.HSIState = RCC_HSI_OFF;
osc.CSIState = RCC_CSI_OFF;
osc.PLL.PLLState = RCC_PLL_ON;
osc.PLL.PLLSource = RCC_PLLSOURCE_HSE;
// ((10 / 5) * 400 ) / 2 = 400
osc.PLL.PLLM = 5;
osc.PLL.PLLN = 400;
osc.PLL.PLLP = 2;
osc.PLL.PLLVCOSEL = RCC_PLL1VCOWIDE;
osc.PLL.PLLRGE    = RCC_PLL1VCIRANGE_3;
HAL_RCC_OscConfig(&osc);

// Sysclk source is PLL
clock.ClockType = (RCC_CLOCKTYPE_SYSCLK  | RCC_CLOCKTYPE_HCLK  |
                   RCC_CLOCKTYPE_D1PCLK1 | RCC_CLOCKTYPE_PCLK1 |
                   RCC_CLOCKTYPE_PCLK2   | RCC_CLOCKTYPE_D3PCLK1);

clock.SYSCLKSource   = RCC_SYSCLKSOURCE_PLLCLK;
clock.SYSCLKDivider  = RCC_SYSCLK_DIV1;
// HCLK = SYSCLK / 2 = 200
clock.AHBCLKDivider  = RCC_HCLK_DIV2;
clock.APB3CLKDivider = RCC_APB3_DIV2;  
clock.APB1CLKDivider = RCC_APB1_DIV2; 
clock.APB2CLKDivider = RCC_APB2_DIV2; 
clock.APB4CLKDivider = RCC_APB4_DIV2; 
HAL_RCC_ClockConfig(&clock, FLASH_LATENCY_4);
__HAL_RCC_CSI_ENABLE();
__HAL_RCC_SYSCFG_CLK_ENABLE();
HAL_EnableCompensationCell();

// GPIOC0 is pin I'm toggling
__HAL_RCC_GPIOC_CLK_ENABLE();
GPIO_InitTypeDef gpio;
gpio.Mode  = GPIO_MODE_OUTPUT_PP;
gpio.Pull  = GPIO_NOPULL;
gpio.Speed = GPIO_SPEED_FREQ_VERY_HIGH;
gpio.Pin   = GPIO_PIN_0;
HAL_GPIO_Init(GPIOC, &gpio);

// Configure GPIOC9 as SYSCLK/2 (200MHz)
gpio.Mode      = GPIO_MODE_AF_PP;
gpio.Alternate = GPIO_AF0_MCO;
gpio.Pull      = GPIO_NOPULL;
gpio.Speed     = GPIO_SPEED_FREQ_VERY_HIGH;
gpio.Pin       = GPIO_PIN_9;
HAL_GPIO_Init(GPIOC, &gpio);    
HAL_RCC_MCOConfig(RCC_MCO2, RCC_MCO2SOURCE_SYSCLK, RCC_MCODIV_2);

while (1)
{
    GPIOC->ODR = 0;
    GPIOC->ODR = 1;
    GPIOC->ODR = 0;
    GPIOC->ODR = 1;
    GPIOC->ODR = 0;
    GPIOC->ODR = 1;
    GPIOC->ODR = 0;
    GPIOC->ODR = 1;
    GPIOC->ODR = 0;
    GPIOC->ODR = 1;
    GPIOC->ODR = 0;
    GPIOC->ODR = 1;
    GPIOC->ODR = 0;
    GPIOC->ODR = 1;
    GPIOC->ODR = 0;
    GPIOC->ODR = 1;
    GPIOC->ODR = 0;
    GPIOC->ODR = 1;
    GPIOC->ODR = 0;
    GPIOC->ODR = 1;
}

当我在示波器上看到这一点时，我在200 my上正确地看到了我的SYSCLK/2，但切换引脚每60 is才切换一次。

​

​

当我查看这段代码的反汇编(编译后的"-O3")时，切换被编译为单条STR指令(r1 = 0，r2 = 1，r3 = &GPIOC->ODR)：

...
str r1, [r3, #20] 
str r2, [r3, #20] 
str r1, [r3, #20] 
str r2, [r3, #20] 
str r1, [r3, #20] 
str r2, [r3, #20] 
str r1, [r3, #20] 
str r2, [r3, #20] 
str r1, [r3, #20] 
str r2, [r3, #20] 
str r1, [r3, #20]
...

我找不到Cortex-M7处理器的周期信息，但是当我查看Cortex-M4处理器(http://infocenter.arm.com/help/topic/com.arm.doc.ddi0439b/DDI0439B_cortex_m4_r0p0_trm.pdf)表3-1的周期时间时，我发现执行一个STR应该需要两个时钟周期。我希望看到我的IO大约每10纳秒(或者200 the AHB4总线上每两个时钟周期)切换一次。

我尝试过从闪存和SRAM运行代码，但IO速度没有差别。

为什么我的IO不是每两个时钟周期切换一次？

编辑：

在参考手册(st.com/resource/en/reference_manual/dm00314099.pdf)第10.2节中，它说：“快速切换能够每两个时钟周期改变一次。”

编辑：

因此，一些评论提到，处理器并不是真的打算直接与GPIO对话。因此，我重写了代码以使用DMA (我必须使用BDMA，并专门在RAM_D3中分配数据，以使其运行速度与软件循环一样快)。最终，我希望将计算出的数据从RAM输出到整个GPIO端口。如何按照参考手册的建议，每两个时钟周期输出一次数据？

在NUCLEO-H743ZI上运行的DMA代码：

#include <stm32h7xx_hal.h>
#include <stdint.h>

DMA_HandleTypeDef dma;

void SysTick_Handler(void)
{
    HAL_IncTick();
}

static void ClockConfig(void)
{
    RCC_ClkInitTypeDef clock;
    RCC_OscInitTypeDef osc;  
    MODIFY_REG(PWR->CR3, PWR_CR3_SCUEN, 0);
    __HAL_PWR_VOLTAGESCALING_CONFIG(PWR_REGULATOR_VOLTAGE_SCALE1);
    while (!__HAL_PWR_GET_FLAG(PWR_FLAG_VOSRDY)) {}

    osc.OscillatorType = RCC_OSCILLATORTYPE_CSI;
    osc.CSIState = RCC_CSI_ON;
    osc.CSICalibrationValue = 16;
    osc.PLL.PLLState = RCC_PLL_ON;
    osc.PLL.PLLSource = RCC_PLLSOURCE_CSI;
    osc.PLL.PLLM = 1;
    osc.PLL.PLLN = 200;
    osc.PLL.PLLP = 2;
    osc.PLL.PLLQ = 2;
    osc.PLL.PLLR = 2;
    osc.PLL.PLLRGE = RCC_PLL1VCIRANGE_2;
    osc.PLL.PLLVCOSEL = RCC_PLL1VCOWIDE;
    osc.PLL.PLLFRACN = 0;
    HAL_RCC_OscConfig(&osc);

    clock.ClockType      = (RCC_CLOCKTYPE_SYSCLK | RCC_CLOCKTYPE_HCLK | RCC_CLOCKTYPE_D1PCLK1 | RCC_CLOCKTYPE_PCLK1 | RCC_CLOCKTYPE_PCLK2  | RCC_CLOCKTYPE_D3PCLK1);
    clock.SYSCLKSource   = RCC_SYSCLKSOURCE_PLLCLK;
    clock.SYSCLKDivider  = RCC_SYSCLK_DIV1;
    clock.AHBCLKDivider  = RCC_HCLK_DIV2;
    clock.APB3CLKDivider = RCC_APB3_DIV2;  
    clock.APB1CLKDivider = RCC_APB1_DIV2; 
    clock.APB2CLKDivider = RCC_APB2_DIV2; 
    clock.APB4CLKDivider = RCC_APB4_DIV2; 
    HAL_RCC_ClockConfig(&clock, FLASH_LATENCY_4);
    __HAL_RCC_CSI_ENABLE();
    __HAL_RCC_SYSCFG_CLK_ENABLE();
    HAL_EnableCompensationCell();

    HAL_SYSTICK_Config(SystemCoreClock / 1000);
    HAL_SYSTICK_CLKSourceConfig(SYSTICK_CLKSOURCE_HCLK);
    HAL_NVIC_SetPriority(SysTick_IRQn, 0, 0);
}

void GpioConfig(void)
{
    __GPIOC_CLK_ENABLE();
    GPIO_InitTypeDef gpio;
    gpio.Pin   = GPIO_PIN_0;
    gpio.Mode  = GPIO_MODE_OUTPUT_PP;
    gpio.Speed = GPIO_SPEED_FREQ_VERY_HIGH;
    gpio.Pull  = GPIO_NOPULL;
    HAL_GPIO_Init(GPIOC, &gpio);
    gpio.Pin   = GPIO_PIN_1;
    gpio.Mode  = GPIO_MODE_OUTPUT_PP;
    gpio.Speed = GPIO_SPEED_FREQ_VERY_HIGH;
    gpio.Pull  = GPIO_NOPULL;
    HAL_GPIO_Init(GPIOC, &gpio);
    gpio.Mode      = GPIO_MODE_AF_PP;
    gpio.Alternate = GPIO_AF0_MCO;
    gpio.Pull      = GPIO_NOPULL;
    gpio.Speed     = GPIO_SPEED_FAST;
    gpio.Pin       = GPIO_PIN_9;
    HAL_GPIO_Init(GPIOC, &gpio);    
    HAL_RCC_MCOConfig(RCC_MCO2, RCC_MCO2SOURCE_SYSCLK, RCC_MCODIV_2);
}

uint8_t complete = 0;
void DmaComplete(DMA_HandleTypeDef *handle)
{
    complete = 1;
}

void DmaError(DMA_HandleTypeDef *handle)
{
    complete = 1;
}

void DmaConfig(void)
{
    __HAL_RCC_BDMA_CLK_ENABLE();
    dma.Instance                 = BDMA_Channel0;
    dma.Init.Request             = DMA_REQUEST_MEM2MEM;
    dma.Init.Direction           = DMA_MEMORY_TO_MEMORY;
    dma.Init.PeriphInc           = DMA_PINC_ENABLE;
    dma.Init.MemInc              = DMA_MINC_DISABLE;
    dma.Init.PeriphDataAlignment = DMA_PDATAALIGN_HALFWORD;
    dma.Init.MemDataAlignment    = DMA_MDATAALIGN_HALFWORD;
    dma.Init.Mode                = DMA_NORMAL;
    dma.Init.Priority            = DMA_PRIORITY_VERY_HIGH;
    HAL_DMA_Init(&dma);
    HAL_DMA_RegisterCallback(&dma, HAL_DMA_XFER_CPLT_CB_ID,  DmaComplete);
    HAL_DMA_RegisterCallback(&dma, HAL_DMA_XFER_ERROR_CB_ID, DmaError);
    HAL_NVIC_SetPriority(BDMA_Channel0_IRQn, 0, 0);
    HAL_NVIC_EnableIRQ(BDMA_Channel0_IRQn);
}

void BDMA_Channel0_IRQHandler(void)
{
    HAL_DMA_IRQHandler(&dma);
}

void HardFault_Handler(void)
{
    __NOP();
}

__attribute__((section(".ram_d3")))
static uint16_t src[10] = { 0, 1, 2, 3, 4, 5, 6, 7, 8, 9 };

int main(void)
{
    HAL_Init();
    ClockConfig();
    GpioConfig();
    DmaConfig();
    while (1)
    {
        complete = 0;
        HAL_DMA_Start_IT(&dma, (uint32_t)&src, (uint32_t)&GPIOC->ODR, 10);
        while (complete == 0) ;
    }
}

​

​

Answer 1

你的假设是错误的。GPIO速度不是由内核指令时序决定的，而是由外围设备的物理特性决定的。请参阅DS中的表，其中包含静态和动态GPIO端口特征。最大速度取决于许多参数。