Some retard once said on this board, and I'm paraphrasing here; "Its possible to run Super Mario 64 on a SNES, it just doesn't because nobody wants to put in the effort to optimize"
And I had to actually think about this for a second, because I know such a feat is impossible, but you have to think about why its impossible. The question as to wither or not a given program can or can not theoretically run on a given piece of hardware, when taken down to its fundamental levels, is memory bandwidth and throughput. Nothing else actually matters, it all comes down to memory bandwidth and throughput. That's why its fundamentally impossible to run Mario 64 on a SNES, even if you imagine the perfect scenario such as if Mario 64 was made completely in machine code somehow, you'd still be bottlenecked by memory bandwidth and throughput. A SNES, no matter how hard you try, is never going to be able to push each frame of the game at a reasonable framerate.
The reason specialized hardware exists, GPUs, SIMD instruction sets, and various rendering pipelines, is to reduce the overall memory bandwidth and throughput requirements of a given piece of software, keyword is reduce, not eliminate.
That all being said, to properly optimize a piece of software, in an ideal scenario, you would develop that given piece of software to run smoothly without specialized rendering and computational pipelines first, then you would move on to utilizing such pipelines to get more headroom out of it