AT32F435的qspi和spi的测试对比，发现速度相差无几？

只看该作者 · 2023-5-30 00:18

本帖最后由 li460135301 于 2023-5-30 00:37 编辑

最近打算用AT32F435外挂一篇spi flash做存储，因为看到AT32F435本身支持qspi，想着qspi支持四线制读写，速度和效率肯定高很多。但是测试下来差别基本不大？

我用AT32F435的spi和qspi各挂载了一片w25q64（8MB）。也就是一片用spi驱动，一片用qspi驱动。

测试环境用rt-thread studio，基于SFUD，这里需要说明一下：SFUD的QSPI写入模式目前好像不支持4线写入，所以写入数据也是采用单线写入的。另外，由于AT32F435的rt-thread的BSP中，qspi没有支持dma模式（如果能增加就最好不过了，用户我们自己修改的话太麻烦了），而spi可以开启dma模式。所以这里测试了一下，spi在不开启dma和开启dma模式于qspi的对比。所以该测试并不全面？另外，有不合理的地方，可以讨论，或者雅特力的技术人员有没有更好的提升qspi的速度的方式。谢谢。
另外，qspi（qspi1）的速度采用默认设置50MHZ，实际速度应该是48Mhz，spi（spi2）的速度位AT32F435的spi最大速度36MHZ。

首先，spi默认不开启DMA，进行sf bench测试，结果如下：

复制

msh />sf probe qspi10

[I/SFUD] Find a Winbond flash chip. Size is 8388608 bytes.

[I/SFUD] sf_cmd flash device is initialize success.

[I/SFUD] Probe SPI flash sf_cmd by SPI device qspi10 success.

8 MB sf_cmd is current selected device.

msh />sf bench yes

Erasing the sf_cmd 8388608 bytes data, waiting...

Erase benchmark success, total time: 17.340S.

Writing the sf_cmd 8388608 bytes data, waiting...

Write benchmark success, total time: 32.768S.

Reading the sf_cmd 8388608 bytes data, waiting...

Read benchmark success, total time: 2.060S.

msh />sf probe spi20

[I/SFUD] Find a Winbond flash chip. Size is 8388608 bytes.

[I/SFUD] sf_cmd flash device is initialize success.

[I/SFUD] Probe SPI flash sf_cmd by SPI device spi20 success.

8 MB sf_cmd is current selected device.

msh />sf bench yes

Erasing the sf_cmd 8388608 bytes data, waiting...

Erase benchmark success, total time: 16.881S.

Writing the sf_cmd 8388608 bytes data, waiting...

Write benchmark success, total time: 32.769S.

Reading the sf_cmd 8388608 bytes data, waiting...

Read benchmark success, total time: 8.165S.

可以看到擦除和写入速度几乎一样，而读取速度，qspi的速度位spi的4倍左右，也就是qspi读取一片8MB的SPI flash 要2秒钟。

然后我们开启spi的dma，结果如下：

复制

msh />sf probe qspi10

[I/SFUD] Find a Winbond flash chip. Size is 8388608 bytes.

[I/SFUD] sf_cmd flash device is initialize success.

[I/SFUD] Probe SPI flash sf_cmd by SPI device qspi10 success.

8 MB sf_cmd is current selected device.

msh />sf bench yes

Erasing the sf_cmd 8388608 bytes data, waiting...

Erase benchmark success, total time: 16.439S.

Writing the sf_cmd 8388608 bytes data, waiting...

Write benchmark success, total time: 32.768S.

Reading the sf_cmd 8388608 bytes data, waiting...

Read benchmark success, total time: 2.060S.

msh />sf probe spi20

[I/SFUD] Find a Winbond flash chip. Size is 8388608 bytes.

[I/SFUD] sf_cmd flash device is initialize success.

[I/SFUD] Probe SPI flash sf_cmd by SPI device spi20 success.

8 MB sf_cmd is current selected device.

msh />sf bench yes

Erasing the sf_cmd 8388608 bytes data, waiting...

Erase benchmark success, total time: 16.642S.

Writing the sf_cmd 8388608 bytes data, waiting...

Write benchmark success, total time: 32.768S.

Reading the sf_cmd 8388608 bytes data, waiting...

Read benchmark success, total time: 2.920S.

我们发现，再开启的DMA模式之后，spi读取一片8MB的flash需要2.92s，仅仅比qspi慢了0.9s不到，这个还是再qspi速度比spi高12M的情况下。所以我感觉不合理啊？
不知道是不是我测试的方法问题。还请请教？否则用qspi的意义就不大了，qspi还比spi多占用了两个io。

附件是我测试的工程代码，基于rt-thread studio。
另外，AT32的rt-thread studio的pack最新版本0.1.4的串口有bug，中断函数都采用UART1_IRQHandler(实际应该是USART1_IRQHandler,串口1，2，3，6都是这样)，否则串口只能打印不能输入，这个疏忽就很尬尴，每次创建工程，需要自己修改drv_usart.c文件。