<table border="1" cellspacing="0" cellpadding="8">
    <tr>
        <th>Issue</th>
        <td>
            <a href=https://github.com/llvm/llvm-project/issues/62701>62701</a>
        </td>
    </tr>

    <tr>
        <th>Summary</th>
        <td>
            missed optimisation combining >= while > is ok?
        </td>
    </tr>

    <tr>
      <th>Labels</th>
      <td>
            new issue
      </td>
    </tr>

    <tr>
      <th>Assignees</th>
      <td>
      </td>
    </tr>

    <tr>
      <th>Reporter</th>
      <td>
          adamse
      </td>
    </tr>
</table>

<pre>
    Exhibited in rust when doing some parsing: https://rust.godbolt.org/z/xv1WjY55o 
- top is good, the cmp's are combined, I use >
- bottom is bad, cmp's are not combined, I use >=

Someone had this to say about the issue (referencing https://godbolt.org/z/YW6b4fMeP which puts it in llvm ir terms):

> I think I see what's happening. The >= code is hitting InstCombinerImpl::foldICmpAddConstant, where it runs this transformation:
> ```
>   // X+C >u C2 -> (X & ~C2) != C
>   // iff C & C2 == 0
>   //       C2+1 is a power of 2
> ```
> 
> which turns
> ```
>   %A = add i64 %val, -4
>   %B = icmp ugt i64 %A, 3
> 
> =>
>   %0 = and i64 %val, -4
>   %B = icmp ne i64 %0, 4
> ```
> 
> But this later impacts InstCombinerImpl::foldAndOrOfICmpsUsingRanges(), which can fold icmps that involve adds, but not ands. The > version of the code doesn't hit the first fold, so the add/icmp doesn't get turned into an and/icmp, meaning the second fold runs, which is what actually combines the two icmps into one 
> 
</pre>
<img width="1px" height="1px" alt="" src="http://email.email.llvm.org/o/eJyUVE2P2zYQ_TX0ZbCGRFmyfdDBq10DPhQp-oEkR0ocSUwkUuBQdraH_vZiKDtOt9u0MQxboN7MvDczfIrIdBaxFPmjyJ9Wag6986XSaiRc1U6_lM9felObgBqMBT9TgEuPFrQztgNyI8KkPBnbiewAfQgTiewg5FHII6PXndO1G8La-U7I4x9CHr-c0_efPua5A5E8ieTwAMFNYAg657SQFYQeoRknIbcEyiM0bqyNxfjuBDMhiOz5Flu7ENzI4bWKiG8jrQv_Ep09LQmW31_diM4i9EpD6A1BcEDqBVTt5hD5GKIZQcidxxY92obl_13uP5V-fF_Um_Yn_BkuvWl6mOZAYAJ3chjOIxgPAf1IQu45yzeERPYMJ6ZiP8MJCBEuvQpRV6-mCa2x3Rp-629ioHGaSUJvQmBqJ0uhWpT70zgNnD87tG7Qp2qcDlpXzlJQNnBbLj16ZGJ-tnRtgFeWWudHFYyzd3bZM4giuX6_HgEsLYAPQj5WzGmGSsJDhMvdBxCygD8rKeQehEyZcPVGtGlbqCK2ksAjyp7grSrLh9M9pixaweQu6MG1IL_L8_60DCTM3tJ_KssPTAaU1mCKDR-c1cB9e9i8Aj5GoGnGCeYu3NAHxmZvsYgan18lSZZq9keqWbyBE4Zu_mcXHuN2G4JBBfRgxkk1gb6zPAer3_l3Le8Q_c63_hdlOyQhd7zDcZW4r42ywPBIjjdK8dKf3XBGbiMxsp5DvJ_Kavq6ynBGT8ZZHmW0AV5r7ZCskNvA2x2PW-MpxAqciVw8VFoLeYztuEd0GOKUo38FB8pywSuOg0dUfJliBsLGWb0w56twF2QoXkBQTZjVMLzcXIViXLi4q9JYg53k3uOVLjO9z_ZqhWVa7DZJXqTbbNWXmSp2smjSrc6aWhdtu2uKNsftbp_nTa7TlSllIrMkTzfJXso8W29x18idzvdp2upsn4tNgqMyw5rdhH1nFW2qLOQ2SVeDqnGgaO1SWrwsHiakZKf3Jcc81HNHYpMMhgLdswQTBixHQ4Qa3BTMaCj6wFU29-vqO5feDMvkDIH7LLLjavZD-coZTejnet24UcgjV7n-PUzefcIm8DiYGwl5jNz_CgAA__8UePia">