<table border="1" cellspacing="0" cellpadding="8">
    <tr>
        <th>Issue</th>
        <td>
            <a href=https://github.com/llvm/llvm-project/issues/139866>139866</a>
        </td>
    </tr>

    <tr>
        <th>Summary</th>
        <td>
            AArch64 ISel(?) miscompilation from weird Rust MIR
        </td>
    </tr>

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

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

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

<pre>
    The reproducer for this is a bit large, I've (mostly) reduced an output from https://github.com/cbeuw/rustlantis, so what I have is Rust MIR, but I'm linking a godbolt that has the LLVM IR: https://godbolt.org/z/sP7qG363b

The structure of this program is that it hashes some locals, and prints a hash at the end. At some point in the optimization pipeline, the hash changes.

We (me and @nikic, who walked me through some opt-bisect use) are pretty confident this is an Aarch64 ISel issue because this produces the good hash:
```
rustc +nightly main.rs -Copt-level=2 -Cllvm-args=-opt-bisect-limit=3642 && ./main
```
And this produces the bad hash:
```
rustc +nightly main.rs -Copt-level=2 -Cllvm-args=-opt-bisect-limit=3647 && ./main
```
And the passes in that bisect range are:
```
BISECT: running pass (3643) AArch64 Instruction Selection on function (_ZN4main4main17he1f737bca831c768E)
BISECT: running pass (3644) SME Peephole Optimization pass on function (_ZN4main4main17he1f737bca831c768E)
BISECT: running pass (3645) Early Tail Duplication on function (_ZN4main4main17he1f737bca831c768E)
BISECT: running pass (3646) Optimize machine instruction PHIs on function (_ZN4main4main17he1f737bca831c768E)
BISECT: running pass (3647) Merge disjoint stack slots on function (_ZN4main4main17he1f737bca831c768E)
```

We've already checked that
```
rustc +nightly main.rs -Copt-level=2 -Cllvm-args=-no-stack-coloring=1 && ./main
```
produces the bad hash, so it doesn't seem like this is related to stack coloring.
</pre>
<img width="1" height="1" alt="" src="http://email.email.llvm.org/o/eJy8VdFu4zYQ_Br6ZWFDomTJftCDEsetgUsbJEEL9KWgqLXIC0Wq5MqG7-sLSm5yvR56RdEUMCwBpDizM8NdEYLuLGLF1jdsvVuIkZTzVRBIymi7aFx7qZ4VgsfBu3aU6OHoPJDSAXQAAY0mMMJ3yPgtHBgvTwiMb3oXyFwY34LH-FkLwoIbaRgJjt71oIiGwLKa8T3j-06TGpuVdD3je9ngeGZ878dARljSIZ4dHJyVIDiAEieM4I9jILg_PMbVZqQJvQej7Yu2HQjoXNs4Q0DxMyUCkEL48OGnezg8sqz-ksK8e-V8x_j-E-P78FD-9l1WZA1LapbUUYZAfpQ0egR3nEUYvOu86COfCUdPUAoDBNcjGCeFmfgL28LgtaWoWtwCgiZGaNsV1DTvH5y2BNpOK24g3etPgrSzMOgBjbaTznFxOkIqYTsMq5nhz7P0OGGxPLH6Rcu4_6wcnIV5wRZ6BFLejZ2aAd1Ay0YHlARjwGiY8AiDR6ILSGePukVLb4ZbqIWXqsjh8IQGdAgjQoNSjAFfFYmOz3J3zrUT1ahzUrMiuf6SOtorgfEbqztF5gK90HblAyxvIymDJzQs23FY3hpz6pfCd4Flu-Ub46XRvSaW7bIi58B4wXgBK8b38aQv0GrbfoVeI_4fduU_YocwiBAwzP4LgqsxPpocffkrzZvD093tc0yzH62NuY9HxBhkRZ5FO-v6apedwxuz9IQG5zdn4Tja-Z3xza-__JBHetNfWipMj2VWNlJsslSWxeaO8e23UPOI-nR_Bw-Ig3IG4cc_5TjufAfYdYS9E95c4FloA7txMFqK96qyiHDXwhB6IZW2CPozkR--P7xLoWVEvkffIbQ6fJwaRiAhXyAYR_8W8vNQTa1kbuXCeBTtBaRCGdtHjOV_dFOsW060l9IZ57XtWLZLv3FPvn555-mgCVqHwTJeEgTEOApe8LVzeTSCYgXuqtYfsKtFW2XtNtuKBVZpmZc8XSfJdqGqFnkjOU-Pm6xZ4zFPt8cS02y9SfOyyDlf6IonfJ2s0zzZZkmyXTVCrhPRbPOGFzzNSpYn2AttVrHyOFoWU8Os0my7KYqFEQ2aMA1fzi2e53bKOI-z2FeTXM3YBZYnRgcKb8eQJoPV691-QsP4hmX7mI1eB-n6QZs5_NO4PaP27evEXIzeVH8zgSPM9bEcvPuIkhjfT-QC4_sr-1PFfw8AAP__L_ijVQ">