<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 12 (filtered medium)">
<style><!--
/* Font Definitions */
@font-face
{font-family:"Cambria Math";
panose-1:2 4 5 3 5 4 6 3 2 4;}
@font-face
{font-family:Calibri;
panose-1:2 15 5 2 2 2 4 3 2 4;}
@font-face
{font-family:Tahoma;
panose-1:2 11 6 4 3 5 4 4 2 4;}
@font-face
{font-family:Consolas;
panose-1:2 11 6 9 2 2 4 3 2 4;}
/* Style Definitions */
p.MsoNormal, li.MsoNormal, div.MsoNormal
{margin:0in;
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;
font-family:"Calibri","sans-serif";
color:windowtext;}
span.EmailStyle18
{mso-style-type:personal-reply;
font-family:"Calibri","sans-serif";
color:windowtext;}
.MsoChpDefault
{mso-style-type:export-only;
font-size:10.0pt;}
@page WordSection1
{size:8.5in 11.0in;
margin:1.0in 1.0in 1.0in 1.0in;}
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">Ok, I think I have a handle on the problem. The changes that cause this were introduced in R141177.<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">Take this basic block:<o:p></o:p></p>
<p class="MsoNormal">f.end: ; preds = %if.else, %if.then62, %if.then<o:p></o:p></p>
<p class="MsoNormal"> %x.0 = phi i32 [ %conv, %if.then ], [ %tmp81, %if.then62 ], [ %tmp55, %if.else ]<o:p></o:p></p>
<p class="MsoNormal"> %y.0 = phi i32 [ %conv37, %if.then ], [ %tmp71, %if.then62 ], [ %conv45, %if.else ]<o:p></o:p></p>
<p class="MsoNormal"> ...<o:p></o:p></p>
<p class="MsoNormal"> %cmp106 = icmp eq i32 %x.0, %y.0<o:p></o:p></p>
<p class="MsoNormal"> br i1 %cmp106, label %if.then108, label %if.else109<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">When processing the branch instruction, GVN somehow thinks that it is safe to switch %y.0 for %x.0<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">Here is if.then108:<o:p></o:p></p>
<p class="MsoNormal">if.then108: ; preds = %if.end<o:p></o:p></p>
<p class="MsoNormal"> br i1 %cmp83, label %if.then113, label %if.end112<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">Here is if.end112:<o:p></o:p></p>
<p class="MsoNormal">if.end112: ; preds = %if.then113, %if.then108<o:p></o:p></p>
<p class="MsoNormal"> tail call void @barrier(i32 0, i32 1) nounwind<o:p></o:p></p>
<p class="MsoNormal"> %cmp151 = icmp ult i32 %tmp95, 216<o:p></o:p></p>
<p class="MsoNormal"> %tmp153 = shl i32 %x.0, 5<o:p></o:p></p>
<p class="MsoNormal"> %tmp155 = or i32 %., %tmp153<o:p></o:p></p>
<p class="MsoNormal"> %tmp209 = extractelement <4 x float> %tmp99, i32 0<o:p></o:p></p>
<p class="MsoNormal"> %tmp210 = insertelement <3 x float> undef, float %tmp209, i32 0<o:p></o:p></p>
<p class="MsoNormal"> %tmp211 = extractelement <4 x float> %tmp99, i32 1<o:p></o:p></p>
<p class="MsoNormal"> %tmp212 = insertelement <3 x float> %tmp210, float %tmp211, i32 1<o:p></o:p></p>
<p class="MsoNormal"> %tmp213 = extractelement <4 x float> %tmp99, i32 2<o:p></o:p></p>
<p class="MsoNormal"> %tmp214 = insertelement <3 x float> %tmp212, float %tmp213, i32 2<o:p></o:p></p>
<p class="MsoNormal"> %tmp279 = extractelement <4 x float> %tmp99, i32 3<o:p></o:p></p>
<p class="MsoNormal"> %tmp280 = fmul float %tmp279, 0xC061252760000000<o:p></o:p></p>
<p class="MsoNormal"> br label %for.body<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">Somehow, this optimization on the br instruction is turning<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal"> %tmp153 = shl i32 %y.0, 5<o:p></o:p></p>
<p class="MsoNormal">into<o:p></o:p></p>
<p class="MsoNormal"> %tmp153 = shl i32 %x.0, 5<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">breaking our applications.<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">My proposed solution is to add this to propagateEquality:<o:p></o:p></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas">
<span style="color:green">// Don't try to propogate equalities between phi nodes.</span><o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas">
<span style="color:blue">if</span> (isa<PHINode>(LHS) || isa<PHINode>(RHS)) <span style="color:blue">
continue</span>;<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas"><o:p> </o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas">The only other thing I can think of is that isOnlyReachableViaThisEdge has a bug in its logic in this case.<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas"><o:p> </o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas">Micah<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas"><o:p> </o:p></span></p>
<p class="MsoNormal"><o:p> </o:p></p>
<div style="border:none;border-left:solid blue 1.5pt;padding:0in 0in 0in 4.0pt">
<div>
<div style="border:none;border-top:solid #B5C4DF 1.0pt;padding:3.0pt 0in 0in 0in">
<p class="MsoNormal"><b><span style="font-size:10.0pt;font-family:"Tahoma","sans-serif"">From:</span></b><span style="font-size:10.0pt;font-family:"Tahoma","sans-serif""> llvmdev-bounces@cs.uiuc.edu [mailto:llvmdev-bounces@cs.uiuc.edu]
<b>On Behalf Of </b>Villmow, Micah<br>
<b>Sent:</b> Friday, August 10, 2012 11:34 AM<br>
<b>To:</b> Developers Mailing List<br>
<b>Subject:</b> [LLVMdev] GVN miscompile debugging help<o:p></o:p></span></p>
</div>
</div>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">I found a case where GVN seems to miscompile an OpenCL program. What I am trying to figure out is given a bitcode file, how can I reduce it to a simpler case with bugpoint when I don't have a valid reference compiler available.<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">Thanks for any tips,<o:p></o:p></p>
<p class="MsoNormal">Micah <o:p></o:p></p>
</div>
</div>
</body>
</html>