<html>
<head>
<base href="http://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 --- - Performance of unordered_multimap::insert much slower than GCC 4.8.2"
href="http://llvm.org/bugs/show_bug.cgi?id=21275">21275</a>
</td>
</tr>
<tr>
<th>Summary</th>
<td>Performance of unordered_multimap::insert much slower than GCC 4.8.2
</td>
</tr>
<tr>
<th>Product</th>
<td>libc++
</td>
</tr>
<tr>
<th>Version</th>
<td>3.4
</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>All Bugs
</td>
</tr>
<tr>
<th>Assignee</th>
<td>unassignedclangbugs@nondot.org
</td>
</tr>
<tr>
<th>Reporter</th>
<td>zaxxon2011@gmail.com
</td>
</tr>
<tr>
<th>CC</th>
<td>llvmbugs@cs.uiuc.edu, mclow.lists@gmail.com
</td>
</tr>
<tr>
<th>Classification</th>
<td>Unclassified
</td>
</tr></table>
<p>
<div>
<pre>The performance of unordered_multimap::insert is very slow in LLVM libc++
when used on large data sets that have degenerate non-uniform key distribution.
// g++ -std=c++11 -O3 umm.cpp -o umm && time ./umm
// 13.5s for Apple LLVM version 6.0 (clang-600.0.51) (based on LLVM 3.5svn).
// 0.016s for GCC version 4.8.2.
#include <unordered_map>
int main() {
std::unordered_multimap<int,int> m;
for (int x = 0; x < 100000; ++x)
m.insert(std::pair<int,int>(x%4, x));
return 0;
}
I recognize that a another ticket was filed on this issue and marked as
WONTFIX.
But even using max_load_factor() does not help LLVM libc++ get anywhere near
the performance of GCC.
One cannot always know in advance what the key distribution of the data set
will
be - it can be uniform or degenerate. Data sets can be multi-gigabyte and would
not be conducive to multiple passes to determine key distribution. As a
workaround I used #ifdefs with an ordered std::multimap on Mac which is only
twice as slow as GCC 4.8.2 libstdc++'s unordered_multimap for this scenario,
which is not ideal.</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>