Hello,
@Lourens
I wonder if someone has some assembly code for a PI Controller, I have some C code but it is a bit slow and I would like to try a faster approach.
I will be using a PIC32MZ with MIPS32 instructions and for now I do not have assembly coding skills but I start working on it right now.

Julien

 

Instead, I would determine what is taking all the time and recode those parts in assembly.

 

This is quite fast for C code almost 600ns but I think there could be some optimisations done in assembly. The current code is a bit complex to be directly converted to assembly without assembly experience. I am using an integrator anti windup and doing many overflow checks but without the if branches those are all simple additions multiplications and variables shift (not using floating point). Would love to see an existing implementation of it if someone already did this

 

This is quite fast for C code almost 600ns but I think there could be some optimisations done in assembly. The current code is a bit complex to be directly converted to assembly without assembly experience. I am using an integrator anti windup and doing many overflow checks but without the if branches those are all simple additions multiplications and variables shift (not using floating point). Would love to see an existing implementation of it if someone already did this

Converting to asm while possible, would be my last choice. Do you need the optimization or would you like it is the primary question. With xc32 you can request a trial full optimization license for testing.

I've used the pic32mz for several motor or inverter projects with PI routines with anti windup and overflow. Most versions of that chip have hardware floating point that can be faster than the equivalent integer operations. And you have lots of code/space to make memory/speed tradeoffs.

https://www.google.com/url?sa=t&sou...kQFnoECB4QAQ&usg=AOvVaw0QIVqaPsCJ1IlktxGRCBAB

https://forum.allaboutcircuits.com/threads/pic32mk-mc-qei-example.150351/post-1560143

 

@nsaspook thank you for the hint I will give a try to FPU ! I did not think that could be faster

 

@nsaspook thank you for the hint I will give a try to FPU ! I did not think that could be faster

YMMV
Easy to instrument and run some test cases. Memory access is often a prime bottleneck on fast controllers. If you have I/D cache try to optimizations that reduce misses.