<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 - LLVM miscompiles calls to "operator delete" under -Oz"
href="https://bugs.llvm.org/show_bug.cgi?id=34581">34581</a>
</td>
</tr>
<tr>
<th>Summary</th>
<td>LLVM miscompiles calls to "operator delete" under -Oz
</td>
</tr>
<tr>
<th>Product</th>
<td>libraries
</td>
</tr>
<tr>
<th>Version</th>
<td>trunk
</td>
</tr>
<tr>
<th>Hardware</th>
<td>PC
</td>
</tr>
<tr>
<th>OS</th>
<td>Linux
</td>
</tr>
<tr>
<th>Status</th>
<td>NEW
</td>
</tr>
<tr>
<th>Severity</th>
<td>enhancement
</td>
</tr>
<tr>
<th>Priority</th>
<td>P
</td>
</tr>
<tr>
<th>Component</th>
<td>Scalar Optimizations
</td>
</tr>
<tr>
<th>Assignee</th>
<td>unassignedbugs@nondot.org
</td>
</tr>
<tr>
<th>Reporter</th>
<td>richard-llvm@metafoo.co.uk
</td>
</tr>
<tr>
<th>CC</th>
<td>llvm-bugs@lists.llvm.org
</td>
</tr></table>
<p>
<div>
<pre>Test case:
void f(char *c) {
delete c;
}
void g(char *c) {
if (c)
::operator delete(c);
}
void h(char *c) {
if (c)
delete c;
}
At -Oz, LLVM optimizes away the implied "if (c)" in f. It's highly questionable
for LLVM to be doing this, but a C++ implementation is permitted to elide the
null check in this case.
LLVM also optimizes away the explicit "if (c)" in g, which is a miscompile. A
user replacement operator delete may have observable effects (for instance,
logging) even when called on a null pointer, so LLVM is not allowed to invent
calls to it. Note that in this case the "operator delete" invocation is
"nobuiltin", making it doubly-clear that this is a miscompile.
Finally, it optimizes away both null checks in h, which again is a miscompile.
Despite being a builtin call, this is still an incorrect transformation, again
because operator delete can have observable side-effects.</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>