<html>
<head>
<base href="https://llvm.org/bugs/" />
</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 --- - clang-3.7: OpenMP corrupts C++11 lambda captures?"
href="https://llvm.org/bugs/show_bug.cgi?id=24769">24769</a>
</td>
</tr>
<tr>
<th>Summary</th>
<td>clang-3.7: OpenMP corrupts C++11 lambda captures?
</td>
</tr>
<tr>
<th>Product</th>
<td>OpenMP
</td>
</tr>
<tr>
<th>Version</th>
<td>unspecified
</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>normal
</td>
</tr>
<tr>
<th>Priority</th>
<td>P
</td>
</tr>
<tr>
<th>Component</th>
<td>Clang Compiler Support
</td>
</tr>
<tr>
<th>Assignee</th>
<td>unassignedclangbugs@nondot.org
</td>
</tr>
<tr>
<th>Reporter</th>
<td>zerolo@gmail.com
</td>
</tr>
<tr>
<th>CC</th>
<td>llvm-bugs@lists.llvm.org
</td>
</tr>
<tr>
<th>Classification</th>
<td>Unclassified
</td>
</tr></table>
<p>
<div>
<pre>Created <span class=""><a href="attachment.cgi?id=14863" name="attach_14863" title="Repo. Compile with "-fopenmp -std=c++11".">attachment 14863</a> <a href="attachment.cgi?id=14863&action=edit" title="Repo. Compile with "-fopenmp -std=c++11".">[details]</a></span>
Repo. Compile with "-fopenmp -std=c++11".
When using OpenMP directives around a C++11 lambda, the captures of that lambda
suddenly refer to invalid / corrupt data.
The attached example contains a templated function taking a functor / lambda.
That lambda is executed on a single thread while all other threads are blocked
in the function.
With clang + openmp, the variable / reference captures within that lambda refer
to bogus data, at least that's what it looks like.
After compiling the attached example with
<span class="quote">> clang++ -fopenmp -std=c++11 clang_omp_repo.cpp</span >
The output looks like this
<span class="quote">> ./a.out </span >
[before parallel] &shared: 0x7fff9ec106c0
[inside lambda] &shared: 0x7fff9ec10218
^ the shared variable inside the lambda has a different address, this shouldn't
be
[inside lambda] -1631518096: shared: -1631516992 (should be 32767)
^ the captured thread index and numThreads are bogus values as well
[before lambda] 0: shared: 9 (should be <= 8)
^ because the lambda referred to a different place, the "real" shared wasn't
reset
[before lambda] 7: shared: 10 (should be <= 8)
[before lambda] 3: shared: 11 (should be <= 8)
[before lambda] 6: shared: 15 (should be <= 8)
[before lambda] 2: shared: 12 (should be <= 8)
[before lambda] 5: shared: 16 (should be <= 8)
[before lambda] 1: shared: 13 (should be <= 8)
[before lambda] 4: shared: 14 (should be <= 8)
[inside lambda] &shared: 0x7fff9ec10218
[inside lambda] -1631518096: shared: -1631516992 (should be 32767)
[after parallel] shared: 16 (should be 0)
The same code works fine on g++-4.9.2 or Intel Compiler 15.0.2.
Version info:
clang version 3.7.0 (tags/RELEASE_370/final)
Target: x86_64-unknown-linux-gnu
Thread model: posix
Found candidate GCC installation:
/home_as/shared/arbeitsgruppen/digitalx/installs/gcc/4.9.2/install/lib/gcc/x86_64-unknown-linux-gnu/4.9.2
Selected GCC installation:
/home_as/shared/arbeitsgruppen/digitalx/installs/gcc/4.9.2/install/lib/gcc/x86_64-unknown-linux-gnu/4.9.2
Candidate multilib: .;@m64
Candidate multilib: 32;@m32
Selected multilib: .;@m64</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>