#
✕ nick:pac85
#asahi · 2026-09-29 14:34
pac85The point of that is to take scalar computations and run them in just one thread. Without it the computations would run per thread yielding the same result and using vector registers that are larher. It's common on mobile GPU because it can save power. More commonly desktop GPUs have a scalar alu and scalar regs that can run in parallel with the vector alu.
#asahi · 2026-09-29 14:33
pac85Also yeah important to distinguish from the concept of a pre shader (or preamble) that existed in previous gens as well
#asahi · 2026-09-29 14:32
pac85opticron: the entry shader replaces the USC, so it's emitted basically every time you bind a pipeline. You need it because the HW requires that you allocate a bunch of things via instructions and has no other way of doing it.