Replace wide integer multiplication

Hello!

I'm using fixed-point data in a Simulink Embedded Coder project. The multipliers henceforth generate code that requires wide integer operations, because two 32-bit fixedpoint types get multiplied into a 64bit type and then shifted. For 32x32 bit multiplication, the code generator uses the utility-function

mul_wide_u32
mul_wide_s32

This function is complex and for my usecase unecessary as the Cortex M3 architecture has the MULL (long multiply) instruction. Is there a way to automatically replace mul_wide_u32 and mul_wide_s32 with custom functions? Since I use fixed point, I'd need to define a multiplication-function for every n of bitshift, so I'd like to replace just the multiplication and let matlab handle the shifting (mul_u32_loSR and so on).

Tl;Dr;: Can I replace generated code strings or functions with custom ones (not replace blocks, but replace generated code)

Thank you

Leon

Answer 1

[Leon Loeser]

I have now found the code replacement library for Cortex M3, but few functions still use the utility functions generated by matlab despite the data-types being the same as in the multiplications replaced with Cortex functions. Any ideas?

Answer 2

[Andy Bartlett]

Hi Leon,

Solving the problem may be a simple as turning on support long long for the model.

Please see this article.

https://www.linkedin.com/pulse/how-dramatically-increase-efficiency-simulink-code-andy-bartlett-phd/?trackingId=Y7%2FJ7xObPtD5N5zzzghrjQ%3D%3D

and/or this video

https://www.youtube.com/watch?v=_rfZMVQG04M

Please let us know if that solves the problem

Andy