<html>
<head>
<base href="https://bugs.llvm.org/">
</head>
<body><table border="1" cellspacing="0" cellpadding="8">
<tr>
<th>Bug ID</th>
<td><a class="bz_bug_link
bz_status_NEW "
title="NEW - LICM incorrect load/store promotion in presence of poison"
href="https://bugs.llvm.org/show_bug.cgi?id=51471">51471</a>
</td>
</tr>
<tr>
<th>Summary</th>
<td>LICM incorrect load/store promotion in presence of poison
</td>
</tr>
<tr>
<th>Product</th>
<td>libraries
</td>
</tr>
<tr>
<th>Version</th>
<td>trunk
</td>
</tr>
<tr>
<th>Hardware</th>
<td>All
</td>
</tr>
<tr>
<th>OS</th>
<td>All
</td>
</tr>
<tr>
<th>Status</th>
<td>NEW
</td>
</tr>
<tr>
<th>Severity</th>
<td>normal
</td>
</tr>
<tr>
<th>Priority</th>
<td>P
</td>
</tr>
<tr>
<th>Component</th>
<td>Loop Optimizer
</td>
</tr>
<tr>
<th>Assignee</th>
<td>unassignedbugs@nondot.org
</td>
</tr>
<tr>
<th>Reporter</th>
<td>d.maljutin@yandex.ru
</td>
</tr>
<tr>
<th>CC</th>
<td>llvm-bugs@lists.llvm.org
</td>
</tr></table>
<p>
<div>
<pre>Created <span class=""><a href="attachment.cgi?id=25131" name="attach_25131" title="reproducer, run with opt -licm">attachment 25131</a> <a href="attachment.cgi?id=25131&action=edit" title="reproducer, run with opt -licm">[details]</a></span>
reproducer, run with opt -licm
If there are two groups of memory accesses in the loop all to the same memory
but one is done via "poisoned" pointer, basic-aa can just place them into two
non-aliasing sets. LICM would then promote both of these groups leading to a
miscompile: <a href="https://llvm.godbolt.org/z/KKxdb5e9v">https://llvm.godbolt.org/z/KKxdb5e9v</a>
The load/store with a poison pointer is a dead code and would never execute at
runtime.
Similar IR was produced on an actual example after value propagation through a
switch case.</pre>
</div>
</p>
<hr>
<span>You are receiving this mail because:</span>
<ul>
<li>You are on the CC list for the bug.</li>
</ul>
</body>
</html>