add.f32: from registration to PTX

An instruction registered in TIRx and called from TIRx-lite

dst, a, b : f32

Read a, b; write dst

PTX template

add.f32 dst, a, b;

Call from a kernel

txl.ptx.add.f32(x, x, y)

dst → x, a → x, b → y

Inspect the PTX

add.f32 %f0, %f0, %f1;

x → %f0, y → %f1

Symbolic register names

Registration and PTX are schematic.

PTX → ptxas → Machine code