STM32H745 input fast read

[EsKi]

Hi, I have problem. I have idea: my STM32H745 must read GPIO state as quickly as possible. FCPU = ~400MHz

I wrote the program in assembler for M7

LED_ON()

ldr r2, [r3, #IDR]

ldr r2, [r3, #IDR]

... ~650x

LED_OFF()

...

only read, IDR, no write (no record yet)

And when I measure with an oscilloscope the led output is on for 25us -> one LDR 15 cycles clock CPU. When I read from RAM it takes about 12us, but GPIO 25us.

This is too slow for me. I tried all the tricks and combinations of LDR; STR ... LDR LDR STR STR ... LDR AND LDR STR AND ... Minimum is 25us.

LDR gpio and STR ram are without comment at all. Some astronomical values ​​come out that can be obtained from AVR or DSPIC.

I tried FMC to read and it was even worse.

What is the fastest way to read the state of a ~40MHz pin?

[gbm · Accepted solution]

Two possibilities:

- For single pin - use SPI MISO line and SPI RX DMA. This guarantees fixed sampling frequency.

- For multiple pins - use timer-triggered DMA from port to memory. Some frequency variations and jitter will appear.

[EsKi]

Ok works. It's a pity that it can't be done differently. Creating a 30-40MHz real-time system with non-standard pin analysis is practically impossible with ARM :(. Too large a range of possible problems. My system worked somehow, but it was supposed to be very easy, but it was very confusing and I probably won't come up with such an idea again :)

In my life I've written a lot in various assemblers x86, PIC, DSPIC, AVR, Z80, even 6502 and a bit MC68000. I thought that ARM, popular in the 21st century, and a large 400MHz clock would solve the problem, but here I had to figure it out :). Fortunately, it worked, but never again :).

[AScha.3]

If you talk about the direct r/w to a port/pin, its not "ARM" making it fast or slow - its how the SOC is built and connected to the core.

So on H7 you more on a "PC" , fast core and calculations, but gpio is "far away" from the core, going over some bus connections to reach a port.

If you want direct r/w to a port/pin fast, look at the M33-core series, maybe stm32H563 .

Here ports are on direct connection to core, so you get : (in my tests)

(speed test, option  -O2 , at 250MHz core )  bsrr 4ns ; so port access here 1 cyc, 4 ns.