Yes, it is unfortunate that in the pursuit of both compatibility and performance we now live in a world where even the CPU is so maddeningly complex that it is basically impossible to reason about how it will perform on any given piece of code. Even writing assembly leaves you a few layers of abstraction above where it would have 30 years ago.
I like to think that it is possible we'll one day develop a CPU architecture that is both simple enough to reason about and also highly performant.
Yes, and it's also the my reason of mentioning microcontroller projects. For example, some find the AVR instruction set is clear and powerful, optimized for both ASM programming and C compilers, and can be used as a good introduction to both hardware and assembly.
> I like to think that it is possible we'll one day develop a CPU architecture that is both simple enough to reason about and also highly performant.
RISC-V is getting there, with open-hardware cores such as Rocket (in-order) and BOOM (out-of-order). Too bad that many and perhaps most of the peripheral components even on a general-purpose SiFive SoC are still closed hardware blocks. But people are working on opening these up as well.
I like to think that it is possible we'll one day develop a CPU architecture that is both simple enough to reason about and also highly performant.