[sve] A redundant csel instruction for overflow check

[vfdff]

• test: https://gcc.godbolt.org/z/jde11xv53

void mset (int *a, long long num) {
   for (long long i=0; i< num; i++)
     a[i] = 2;
}

• gcc:

mset:
        cmp     x1, 0
        ble     .L1
        mov     x2, 0
        cntw    x3
        whilelo p7.s, xzr, x1
        mov     z31.s, #2
.L3:
        st1w    z31.s, p7, [x0, x2, lsl 2]
        add     x2, x2, x3    --  before the whilelo
        whilelo p7.s, x2, x1
        b.any   .L3

• llvm:

mset:                                   // @mset
        cmp     x1, #1
        b.lt    .LBB0_3
        cntw    x8
        mov     z0.s, #2                        // =0x2
        mov     x10, xzr
        subs    x9, x1, x8
        csel    x9, xzr, x9, lo    ---> this is redundant  compare to gcc ??
        whilelo p0.s, xzr, x1
.LBB0_2:                                // =>This Inner Loop Header: Depth=1
        st1w    { z0.s }, p0, [x0, x10, lsl #2]
        whilelo p0.s, x10, x9
        add     x10, x10, x8        -- after the whilelo
        b.mi    .LBB0_2

• the above redundant csel is used to check the overflow of loop iv, but it seems not necessary.

[vfdff]

• in the gcc's body, it add the the iter number before the compare (whilelo), so it don't need the csel to guard the overflow