<table border="1" cellspacing="0" cellpadding="8">
    <tr>
        <th>Issue</th>
        <td>
            <a href=https://github.com/llvm/llvm-project/issues/129199>129199</a>
        </td>
    </tr>

    <tr>
        <th>Summary</th>
        <td>
            AMDGPU's getLargestLegalSuperClass should ignore register alignment requirements
        </td>
    </tr>

    <tr>
      <th>Labels</th>
      <td>
            backend:AMDGPU
      </td>
    </tr>

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

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

<pre>
    On gfx90a+ VGPRs and AGPRs have a requirement to be even aligned, but this only applies to operations operating on tuple registers. We can still insert copies to and from the unaligned classes, as long as the ultimate use sees a copy back to an aligned use. In theory this should avoid some spills since there's more freedom in finding a free register in the wider class 
</pre>
<img width="1" height="1" alt="" src="http://email.email.llvm.org/o/eJxMUl1v1DAQ_DXOy6qn2PeVPOTh2uoqpCIqEPC8iTeOwbGD1z64f4-cXgtPu0pmxzOjQWZrPFEn9vdi_1hhTlOIHUYmP1d90Nfukwcz_mlrFOoevj29fGZAr-G0bhNeCBAi_co20kw-QQrQE9CFPKAr5FqoB-hzgjRZhuDdFXBZnCUu2LBQxGSD57fVGwgeUl4cQSRjOVHkDXwnGNADJ-scWM8UEwxhudEUSWMMM6SJIPvbyzA4ZCYuCpDBBW_KXDEu2RkTQWYCJmLAQneFHoefr4xv-gtkAx98uQvx-uqDp5CdBrwEq4HDTMCLdY6BrR-oQCMJdWSYQyQYI5EOM1gPo_W6eMT147vD8qvo-m01xVfdUOluq9ttixV18rirj2p7kLKaur7BujnI_V6OTd-3zUGOfbtvt4NUmmQzVLZTtdrXSjX1QdVSboZRUd3udqOWmsYWxa6mGa3bOHeZNyGayjJn6qRqZdtWDntyvLZCqZIIeS22p9PHx6eXr0Kp0pXYldu7PhsWu9pZTvyPLdnkqHvDHxkMpWeMhjg9k0H3JS8UH1aXtySt8SWp9zzW8NdG_dcurnJ03ZTSwmJ7Euos1NnYNOV-M4RZqHMRcBt3Sww_aEhCnVdvLNT5Zu_Sqb8BAAD__zghBPU">