> I would submit that the answer is actually "no". Accuracy would plummet due to the > complexity of writing a Naomi emulation entirely in assembly, and if the same exact > algorithms were used (e.g. an interpreter SH-4 core and a PowerVR drawing routine > that piped each pixel through MAME's memory system) you would get identically absymal > performance.
The answer is no. I disagree on your reasoning though.
It would certainly be harder to write in assembler. It would be possible to match or exceed MAME's current accuracy with a version written in assembler. However in this very unlikely situation it wouldn't be assembler that helped. You could improve the accuracy much easier by improving the current code (even if you threw it all out and started again).
You probably wouldn't have time to work or sleep as Intel & AMD are likely to have released hardware that dwarfs your optimisations before you finish.
Previous 
Index 
Next 
Flat 
Re: Should I learn C or C++?
Reply
