[AArch64] Unnecessary `addv` for `popcount(uint8_t)`

[Kmeakin]

https://godbolt.org/z/e9qPT9z81

C++ code

#include <cstdint>

auto popcount_u8(std::uint8_t x) { return __builtin_popcountg(x); }

Assembly (LLVM)

popcount_u8(unsigned char):
        and     w8, w0, #0xff
        fmov    s0, w8
        cnt     v0.8b, v0.8b
        addv    b0, v0.8b
        fmov    w0, s0
        ret

Assembly (GCC)

popcount_u8(unsigned char):
        and     w0, w0, 255
        fmov    d31, x0
        cnt     v31.8b, v31.8b
        smov    w0, v31.b[0]
        ret

As you can see from GCC's assembly, the addv is unecessary, because all the lanes except for the first will be zero.

[llvmbot]

@llvm/issue-subscribers-backend-aarch64

Author: Karl Meakin (Kmeakin)

https://godbolt.org/z/e9qPT9z81

C++ code

#include <cstdint>

auto popcount_u8(std::uint8_t x) { return __builtin_popcountg(x); }

Assembly (LLVM)

popcount_u8(unsigned char):
        and     w8, w0, #<!-- -->0xff
        fmov    s0, w8
        cnt     v0.8b, v0.8b
        addv    b0, v0.8b
        fmov    w0, s0
        ret

Assembly (GCC)

popcount_u8(unsigned char):
        and     w0, w0, 255
        fmov    d31, x0
        cnt     v31.8b, v31.8b
        smov    w0, v31.b[0]
        ret

As you can see from GCC's assembly, the addv is unecessary, because all the lanes except for the first will be zero.

[XChy]

We may be able to simplify it by known bits.