<html xmlns:v="urn:schemas-microsoft-com:vml" xmlns:o="urn:schemas-microsoft-com:office:office" xmlns:w="urn:schemas-microsoft-com:office:word" xmlns:m="http://schemas.microsoft.com/office/2004/12/omml" xmlns="http://www.w3.org/TR/REC-html40">
<head>
<meta http-equiv="Content-Type" content="text/html; charset=us-ascii">
<meta name="Generator" content="Microsoft Word 14 (filtered medium)">
<style><!--
/* Font Definitions */
@font-face
{font-family:Calibri;
panose-1:2 15 5 2 2 2 4 3 2 4;}
/* Style Definitions */
p.MsoNormal, li.MsoNormal, div.MsoNormal
{margin:0cm;
margin-bottom:.0001pt;
font-size:11.0pt;
font-family:"Calibri","sans-serif";}
a:link, span.MsoHyperlink
{mso-style-priority:99;
color:blue;
text-decoration:underline;}
a:visited, span.MsoHyperlinkFollowed
{mso-style-priority:99;
color:purple;
text-decoration:underline;}
span.EmailStyle17
{mso-style-type:personal-compose;
font-family:"Calibri","sans-serif";
color:windowtext;}
.MsoChpDefault
{mso-style-type:export-only;
font-family:"Calibri","sans-serif";}
@page WordSection1
{size:612.0pt 792.0pt;
margin:72.0pt 72.0pt 72.0pt 72.0pt;}
div.WordSection1
{page:WordSection1;}
--></style><!--[if gte mso 9]><xml>
<o:shapedefaults v:ext="edit" spidmax="1026" />
</xml><![endif]--><!--[if gte mso 9]><xml>
<o:shapelayout v:ext="edit">
<o:idmap v:ext="edit" data="1" />
</o:shapelayout></xml><![endif]-->
</head>
<body lang="EN-US" link="blue" vlink="purple">
<div class="WordSection1">
<p class="MsoNormal">Hi,<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">I have a test case with an unrolled loop body, consisting of a long sequence of non-overlapping stores. I found that with alias analysis it became very compile time expensive (53 seconds, without AA it was 0.02 sec). The program spent a
lot of time in WalkChainUsers() (isel), and it was due to all the token factor nodes introduced by the combiner.<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">I spent some time looking into it, and found that nearly all of the compile time disappeared if the combiner added a new token factor node to the worklist. The case was that there was a mess of token factors at the bottom of the DAG, that
disappeared after reiteration over those nodes.<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">Test case supplied, to run: llc -mtriple=arm -combiner-alias-analysis unrolledloop.opt.ll<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">I supply a patch for this. Let me know if this seems reasonable to use and if I can commit.<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">/Jonas Paulsson<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">Patch:<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">From 0123157edefc2aea3137b702b1fa24ebd41e7709 Mon Sep 17 00:00:00 2001<o:p></o:p></p>
<p class="MsoNormal">From: Jonas Paulsson <jonas.paulsson@ericsson.com><o:p></o:p></p>
<p class="MsoNormal">Date: Mon, 9 Feb 2015 16:19:53 +0100<o:p></o:p></p>
<p class="MsoNormal">Subject: [PATCH] Fix SelectionDAG compile time issue with alias analysis.<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">Add new token factor node to worklist if alias analysis is turned on,<o:p></o:p></p>
<p class="MsoNormal">in DAGCombiner::visitTokenFactor().<o:p></o:p></p>
<p class="MsoNormal">---<o:p></o:p></p>
<p class="MsoNormal">lib/CodeGen/SelectionDAG/DAGCombiner.cpp | 6 ++++--<o:p></o:p></p>
<p class="MsoNormal">1 file changed, 4 insertions(+), 2 deletions(-)<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">diff --git a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp<o:p></o:p></p>
<p class="MsoNormal">index 8a4b602..dbf6ef7 100644<o:p></o:p></p>
<p class="MsoNormal">--- a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp<o:p></o:p></p>
<p class="MsoNormal">+++ b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp<o:p></o:p></p>
<p class="MsoNormal">@@ -1539,8 +1539,10 @@ SDValue DAGCombiner::visitTokenFactor(SDNode *N) {<o:p></o:p></p>
<p class="MsoNormal"> Result = DAG.getNode(ISD::TokenFactor, SDLoc(N), MVT::Other, Ops);<o:p></o:p></p>
<p class="MsoNormal"> }<o:p></o:p></p>
<p class="MsoNormal"><o:p></o:p></p>
<p class="MsoNormal">- // Don't add users to work list.<o:p></o:p></p>
<p class="MsoNormal">- return CombineTo(N, Result, false);<o:p></o:p></p>
<p class="MsoNormal">+ // Don't add users to work list, unless alias analysis is used.<o:p></o:p></p>
<p class="MsoNormal">+ bool UseAA = CombinerAA.getNumOccurrences() > 0 ? CombinerAA<o:p></o:p></p>
<p class="MsoNormal">+ : DAG.getSubtarget().useAA();<o:p></o:p></p>
<p class="MsoNormal">+ return CombineTo(N, Result, UseAA /*add to worklist*/);<o:p></o:p></p>
<p class="MsoNormal"> }<o:p></o:p></p>
<p class="MsoNormal"><o:p></o:p></p>
<p class="MsoNormal"> return Result;<o:p></o:p></p>
<p class="MsoNormal">-- <o:p></o:p></p>
<p class="MsoNormal">1.8.4.2<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
</div>
</body>
</html>