<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 - Performance regression: SROA apepars to add unnecessary code to critical path function"
href="https://bugs.llvm.org/show_bug.cgi?id=33438">33438</a>
</td>
</tr>
<tr>
<th>Summary</th>
<td>Performance regression: SROA apepars to add unnecessary code to critical path function
</td>
</tr>
<tr>
<th>Product</th>
<td>new-bugs
</td>
</tr>
<tr>
<th>Version</th>
<td>4.0
</td>
</tr>
<tr>
<th>Hardware</th>
<td>Other
</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>new bugs
</td>
</tr>
<tr>
<th>Assignee</th>
<td>unassignedbugs@nondot.org
</td>
</tr>
<tr>
<th>Reporter</th>
<td>eric.schweitz@pgroup.com
</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=18630" name="attach_18630" title="SROA adding many phi-nodes">attachment 18630</a> <a href="attachment.cgi?id=18630&action=edit" title="SROA adding many phi-nodes">[details]</a></span>
SROA adding many phi-nodes
A performance regression in a large code has been tracked down to a difference
between the way LLVM 3.9 and LLVM 4.0 produce code for a particular function.
The reproducer is attached.
In LLVM 4.0, one can see that the function performs saves on all registers
# BB#0: # %L.entry
std 20, -96(1) # 8-byte Folded Spill
std 21, -88(1) # 8-byte Folded Spill
std 22, -80(1) # 8-byte Folded Spill
std 23, -72(1) # 8-byte Folded Spill
std 24, -64(1) # 8-byte Folded Spill
std 25, -56(1) # 8-byte Folded Spill
std 26, -48(1) # 8-byte Folded Spill
std 27, -40(1) # 8-byte Folded Spill
lots of register-to-register moves
mr 27, 0
mr 30, 0
mr 29, 0
mr 28, 0
mr 26, 0
mr 8, 0
mr 7, 0
etc.
The code produced by LLVM 3.9 does not contain these instructions.
It appears that SROA may play some role in this regression.</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>