by fvv on 6/11/24, 8:15 AM with 1 comments
by cherryteastain on 6/11/24, 9:46 AM
To do that AMD must adopt an intermediate representation like Nvidia PTX/SPIR-V that compilers can target, as opposed to per-chip targets (gfx1030 and whatnot) like now. Otherwise they will be perennially behind.