[SOLVED] CPU speed seems different with SDK or mbed

[Nathan O.]

Hi,

I own a Redbearlab BLE Nano with a nrf51822.

I want to use it to drive some WS2812 rgb LED (neopixels), using precised timings for bit-banging.

Before driving the LED, I wanted to know what the execution speed of a single I/O toggle instruction. I used this code, and measured the output through a logic analyzer :

while(1) {
	 NRF_GPIO->OUTSET = (1UL << 8);
	 NRF_GPIO->OUTCLR = (1UL << 8);
}

Compiling from a barebone Eclipse project using the nrf5 SDK v11.0.0 (and SoftDevice 130 although I am not using it in my program), I get a square signal with a high duration of about 500ns.

When I run the same code in the mbed online compiler, I obtain a high duration of about 125ns.

Looking at the system files, both seem to use a SYSTEM_CLOCK of 16000000UL (16MHz). Is there any PLL or anything that I have missed that would allow me to get that 4 times speed increase without using mbed ?

Regards

[Nathan O.]

I found the solution.

In case anyone gets the same issue trying to do bit-banging on the nrf51 : the problem was simply a lack of compiler optimization. I was compiling with -O0. When moved to -Os, the toggle is now done every 125ms.

From there I am able to drive the LED using the RGB_Show function from https://developer.mbed.org/questions/54210/RedBear-nRF51822-driving-NeoPixels/

[Roger Clark]

Which example project are you using ?

I checked all the makefiles in SDK 12.1.0 and only the blinky uses -O0 most the rest use -O3

[Nathan O.]

I'm not exactly sure as I've started off from a previous project of mine but it might have been based on the blinky one so that makes sense.

[Roger Clark]

No worries.

I'd try -O3 instead of -Os and see if it makes much difference, as it will reduce you code side, as -Os will favour speed over size