How hard would it be to get DMD to generate SIMD code for all x86 targets?

[solidstate1991]

It would be also great if 32 bit and 64 bit long vector loading 
could be supported, so I don't have to write separate functions 
for that. Had to delete a lot of code I wrote in Assembly after I 
made a big mistake thanks to its much poorer readability compared 
to much more modern languages.

If it's not a hard task, I might do it by myself.