Does the update of sp redundant in spill/reload code?

Beginners
,
1

the following test code get from llvm/test/CodeGen/RISCV/rvv/allocate-lmul-2-4-8.ll

define void @lmul1_and_2() nounwind {
  %v2 = alloca <vscale x 2 x i64>
  ret void
}

and both riscv64 and aarch64 generate assemble with sub/add sp (see detail Compiler Explorer) , such as aarch64:

        str     x29, [sp, #-16]!                // 8-byte Folded Spill
        addvl   sp, sp, #-1    // sub sp
        addvl   sp, sp, #1     // add sp
        ldr     x29, [sp], #16

combine addvl sp, sp, #-1 and addvl sp, sp, #1, the sp in fact doesn’t change, so them are redundant ?

2

Yes, once you get to CodeGen, any alloca remaining in the function will lead to an allocation, because it’s assumed earlier optimizations have removed any that aren’t actually used. In this case the fact that you need a special instruction for vscale vectors makes the redundancy clearer, but it’s not fundamentally different to any other unused alloca.

In more realistic code, there would either be genuine uses of %v2 between the sub and the add, or earlier optimizations would have discovered it wasn’t used and removed the alloca entirely.

3

Thanks very much