Take a leap the next time you use SPI and don’t poll for the busy flag. “What, are you crazy? That’s the whole point of the busy flag! It’s a quick check to make sure you don’t kill a byte waiting to be shifted out!” Sure, we thought the same thing, but the other side of the coin is that it takes time to check the busy flag, and that’s time he could be transmitting data. [bigjosh2] calculates that his technique saves 20% of those wasted cycles in this particular case. And he’s “using the force” only because he’s
a Jedi master able to rely on the cycle count of a chunk of assembly code.
He’s working with an AVR processor, and pumping out bits to drive the vintage LED display pictured above. The ancient chips don’t have buffered SPI so he has to blank the display while shifting new data in to prevent it from glitching. Because the display blank during the SPI transmission, the slower it goes, the dimmer the lights.
He attacks the problem with synchronous code. It takes 2 cycles for the hardware SPI to send each bit, so he twiddles his thumbs (that’s exactly what he wrote in his code comments) for 16 cycles before reloading the SPI register with his next value. This leaves it up to faith in the silicon that the shifting will always take the same number of cycles, but the nice thing about hardware is that it’s deterministic. He ends up killing a few cycles in order to save time by not polling the busy flag.
Still need a crash course in what SPI actually does? [Bil Herd] has you covered with this SPI communication demo.