Using PRU Arithmetics with C++

Max_

Hello,
I'd like to try to utilize one of the PRUs to perform arithmetic tasks.
My goal is to perform float multiplication on one of the PRUs using Assembly. (I will focus on how to perform float multiplication with the limited PRU assembly instruction set later ;-) )

Previously I have successfully played around using CPU instruction in Assembly and calling them in C++ (as thought in online lecture 21).

Apart from that, I'm a complete beginner in Assembly and I'm wondering how to communicate with the PRU in C++ (which registers to read from, how to initialize the PRU, ...)

Any reference to suitable reading is very welcome!

giuliomoro

Max_ I'd like to try to utilize one of the PRUs to perform arithmetic tasks.

Why? It is not the ideal processor for performing arithmetics beyond simple integer additions and subtractions. According to its documentation, it doesn't even have a multiply instruction for integers, however I heard a few times over the years that it can perform integer multiplication in some quirky way, and this example seems to show that https://github.com/beagleboard/am335x_pru_package/tree/master/pru_sw/old_example/PRU_multiply .

My goal is to perform float multiplication on one of the PRUs using Assembly. (I will focus on how to perform float multiplication with the limited PRU assembly instruction set later ;-) )

How is the part before the parenthesis different from the one within parentheses?

Max_ Previously I have successfully played around using CPU instruction in Assembly and calling them in C++ (as thought in online lecture 21).

Keep in mind that PRU assembly is a different language from the ARM assembly used there.

Max_ Any reference to suitable reading is very welcome!

Here's the reference manual for the PRU instruction set. As mentioned, it is at least partially incomplete because MPY is not included.
https://www.ti.com/lit/ug/spruij2/spruij2.pdf?ts=1673347804758&ref_url=https%253A%252F%252Fwww.google.com%252F

To use PRU assembly in a Bela project and communicate back and forth with ARM, see Bela examples such as Extras/second-pru and Extras/pru-pwm. These are admittedly dated in that the communication between PRU and ARM takes place via shared memory and there is none of the remoteproc messaging and interrupt stuff that is possible with more modern toolchains, but which is also not particularly real time friendly.

Most importantly, you'll want to read about the general principles of how floating point multiplications are achieved on non-floating-point hardware, but I don't have a good recommended reading for that.

Max_

Hey Giuliomoro!
Thanks for your fast reply! :-D

giuliomoro Why? It is not the ideal processor for performing arithmetics beyond simple integer additions and subtractions.

I'm trying evaluate every possible way of a complex multiplication (both, the imaginary and the real value are stored as float) and compare the runtime of each implementation method as a part of my masters thesis. I am aware that the PRU has a limited instruction set and that it's not designed for this purpose. Anyhow, if possible, I want to give it a try, at least just for the comparison purpose. (Other implementations: softfp (in assembly), hardfp (in assembly), and neon (not implamented yet))

giuliomoro How is the part before the parenthesis different from the one within parentheses?

I meant that I'm trying to initialize and communicate with the PRU first and afterwards figure out a way to implement floating point multiplication with the limited instruction set.

giuliomoro Keep in mind that PRU assembly is a different language from the ARM assembly used there.

I'm aware of the different instruction set. I'm struggling with the directives. Are the directives (at least syntax-wise) similar to ARM? Sorry, I'm a complete assembly noob not knowing where to start :-/

giuliomoro To use PRU assembly in a Bela project and communicate back and forth with ARM, see Bela examples such as Extras/second-pru and Extras/pru-pwm. These are admittedly dated in that the communication between PRU and ARM takes place via shared memory and there is none of the remoteproc messaging and interrupt stuff that is possible with more modern toolchains, but which is also not particularly real time friendly.

Great, thanks! I already had a look at those but was not aware that part of it was designed in Assembly. Is the Assembly part available online? Couldn't find it..
I guess that the shared values are loaded using a load operator and the pointer address?

Again, thanks for your fast & informative answer! :-D

giuliomoro

Max_ Is the Assembly part available online?

it's in the .p file in each example's folder.

Max_ I guess that the shared values are loaded using a load operator and the pointer address?

communication there is done via accessing the same shared memory from ARM and PRU. This is PRU ARM so it's really fast and deterministic to access from PRU, but a bit slower from ARM.

Max_ Are the directives (at least syntax-wise) similar to ARM?

Yes and no. Many concepts are the same (load, store, branch), but the actual mnemonics are different. PRU assembly is much simpler (doesn't have any of the modifier suffix that come with ARM) and limited (it has fewer instructions, I think), but it's Turing-complete, so you can build a multiplier out of it :-) .