Post #2766066

TIL that #llvm does jump table compression (that is, stores scaled code offsets in u8, u16, or u32 code offsets rather than full 64-bit addresses) on Aarch64! That’s awesome and makes a huge difference on some application (huge switches/matches with many shared branches)! Sadly it appears to not go below u32 on RISC-V. I wonder if similar optimizations can be relied upon for #WASM jits.