One really important portion of the argument is that some compilers include the entire math or floating-point library even though only one floating point arithmetic function is used, while others are "smart" enough to include just what's used.

In the latter case, one could, if necessary, extract the required code and incorporate it in an ASM program as a MACRO. I've even considered using a demo version to do that, just to see if it's realistic, but many compiler demos don't allow you to use floating point, presumably because including the library makes the code too large to allow use of the demo version. If ever there were to be an algorithm that simply begged for HLL rather than ASM, I'd attempt to use HLL to generate the module associated with that algorithm and then hand-optimize it for inclusion in an ASM program as a macro. Unfortunately, I haven't yet figured out how to manage that. I've even considered using the sample compiler that SiLabs provides, just to effect the setup of their rather complex chips, since they provide tools to do it in 'C', and then produce the rest of the code in ASM. I haven't gotten around to that experiment yet ...

I'm persuaded that ASM can be written and organized in such a way that downstream maintainers can readily comprehend what's supposed to be going on. I'm further persuaded that ANYONE can learn to do it. I'm also persuaded that most people simply don't want to do it that way. When I was in grad-school, we had FORTRAN, and not the later versions, and ASM. We all learned to do it right, else we'd still be there.

RE