Original Post

Keep in mind it takes AMD and academia in Texas to make a new A x B + C chip compute unit. The complexity of a simple FPU FMA is overwhelming not because it's hard, but because of corporate closed-source know-how and really difficult way to reason about GDSII <-> RTL back-and-forth (think $80k/year per a Cadence seat). Linux is muuuuuch easier to open-source than ASIC IP. Chips are for rich large closed source proprietary Dark Black Magic IP Holders. Just admit it.