<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;}
@font-face
        {font-family:Tahoma;
        panose-1:2 11 6 4 3 5 4 4 2 4;}
@font-face
        {font-family:"Arial Narrow";
        panose-1:2 11 6 6 2 2 2 3 2 4;}
/* Style Definitions */
p.MsoNormal, li.MsoNormal, div.MsoNormal
        {margin:0in;
        margin-bottom:.0001pt;
        font-size:12.0pt;
        font-family:"Times New Roman","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;}
p.emailquote, li.emailquote, div.emailquote
        {mso-style-name:emailquote;
        mso-margin-top-alt:auto;
        margin-right:0in;
        mso-margin-bottom-alt:auto;
        margin-left:1.0pt;
        border:none;
        padding:0in;
        font-size:12.0pt;
        font-family:"Times New Roman","serif";}
span.EmailStyle18
        {mso-style-type:personal-reply;
        font-family:"Calibri","sans-serif";
        color:#1F497D;}
.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"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1F497D">Typo.  The first sentence should read as:<o:p></o:p></span></p>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif"">… test case (Part B) where loop vectorization is disabled due to possible store-load forwarding conflict (Part C).</span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1F497D"><o:p></o:p></span></p>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1F497D"><o:p> </o:p></span></p>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif";color:#1F497D"><o:p> </o:p></span></p>
<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>Das, Dibyendu<br>
<b>Sent:</b> Friday, June 12, 2015 10:42 AM<br>
<b>To:</b> llvmdev@cs.uiuc.edu<br>
<b>Subject:</b> [LLVMdev] Loop Vectorization and Store-Load Forwarding issue<o:p></o:p></span></p>
</div>
</div>
<p class="MsoNormal"><o:p> </o:p></p>
<div>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif"">I have been looking into this small test case (Part A) where loop vectorization is disabled due to possible store-load forwarding conflict (Part B). As you can see, due to
 the presence of dependence distance 2 the loop is vectorizable only for a width of 2. However, the presence of dependence distance 15 (due to y[j-15]) results in store-load forwarding issue as store packet of y[16:17] (iteration j=16) partially overlaps with
 load packets of y[15:16] (iteration j=30) and  y[17:18] (iteration j=32). As conflicts introduce additional delays in the store->load forwarding pipes, this fact is modeled in the method
</span><span style="font-size:11.0pt;font-family:"Arial Narrow","sans-serif"">MemoryDepChecker::couldPreventStoreLoadForward()</span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""> in LoopAccessAnalysis.cpp. The function may turn off vectorization
 in the presence of such conflicts. Looking through the code gives me the feeling that it may be more conservative than desired. The reason being, if the dependence distance is high, the conflicting store may flush out of the store pipe by the time the load
 is issued. And vectorization may become beneficial.<o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""> <o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif"">I am seeing some performance improvements when I disable the method above. This is for x86. Hence I am seeking some advice on how to improve the following logic. Can we better
 model </span><span style="font-size:8.0pt;font-family:"Calibri","sans-serif"">NumCyclesForStoreLoadThroughMemory
</span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif"">? This may be way too high ? Or there are other ways to circumvent the basic problem ?<o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""> <o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif"">-TIA<o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif"">Dibyendu<o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""> <o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif"">Part A:<o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:8.0pt;font-family:"Calibri","sans-serif"">  const unsigned NumCyclesForStoreLoadThroughMemory = 8*TypeByteSize;  // 512 for the test case shown</span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""><o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:8.0pt;font-family:"Calibri","sans-serif"">  // Maximum vector factor.</span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""><o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:8.0pt;font-family:"Calibri","sans-serif"">  unsigned MaxVFWithoutSLForwardIssues = VectorizerParams::MaxVectorWidth * TypeByteSize; 
</span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""><o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:8.0pt;font-family:"Calibri","sans-serif"">  if(MaxSafeDepDistBytes < MaxVFWithoutSLForwardIssues)</span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""><o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:8.0pt;font-family:"Calibri","sans-serif"">    MaxVFWithoutSLForwardIssues = MaxSafeDepDistBytes;</span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""><o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:8.0pt;font-family:"Calibri","sans-serif""> </span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""><o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:8.0pt;font-family:"Calibri","sans-serif"">  for (unsigned vf = 2*TypeByteSize; vf <= MaxVFWithoutSLForwardIssues; vf *= 2) {</span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""><o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:8.0pt;font-family:"Calibri","sans-serif"">    if (Distance % vf && Distance / vf < NumCyclesForStoreLoadThroughMemory) {</span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""><o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:8.0pt;font-family:"Calibri","sans-serif"">      MaxVFWithoutSLForwardIssues = (vf >>=1);</span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""><o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:8.0pt;font-family:"Calibri","sans-serif"">      break;</span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""><o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:8.0pt;font-family:"Calibri","sans-serif"">    }</span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""><o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:8.0pt;font-family:"Calibri","sans-serif"">  }</span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""><o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:8.0pt;font-family:"Calibri","sans-serif""> </span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""><o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:8.0pt;font-family:"Calibri","sans-serif"">  if (MaxVFWithoutSLForwardIssues< 2*TypeByteSize) {</span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""><o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:8.0pt;font-family:"Calibri","sans-serif"">    DEBUG(dbgs() << "LAA: Distance " << Distance <<</span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""><o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:8.0pt;font-family:"Calibri","sans-serif"">          " that could cause a store-load forwarding conflict\n");</span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""><o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:8.0pt;font-family:"Calibri","sans-serif"">    return true;</span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""><o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:8.0pt;font-family:"Calibri","sans-serif"">  }</span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""><o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif"">----------------------------<o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:10.0pt;font-family:"Calibri","sans-serif"">Part B:</span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""><o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:10.0pt;font-family:"Calibri","sans-serif"">typedef unsigned long long uint64;</span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""><o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:10.0pt;font-family:"Calibri","sans-serif""> </span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""><o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:10.0pt;font-family:"Calibri","sans-serif"">void foo(const unsigned char *m, unsigned int block, uint64 y[80])</span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""><o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:10.0pt;font-family:"Calibri","sans-serif"">{</span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""><o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:10.0pt;font-family:"Calibri","sans-serif"">    const unsigned char *sblock;</span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""><o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:10.0pt;font-family:"Calibri","sans-serif"">    int i, j;</span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""><o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:10.0pt;font-family:"Calibri","sans-serif""> </span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""><o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:10.0pt;font-family:"Calibri","sans-serif"">    for (i = 0; i < (int) block; i++) {</span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""><o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:10.0pt;font-family:"Calibri","sans-serif"">        sblock = m + (i << 7);</span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""><o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:10.0pt;font-family:"Calibri","sans-serif""> </span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""><o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:10.0pt;font-family:"Calibri","sans-serif"">        for (j = 16; j < 80; j++) {</span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""><o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:10.0pt;font-family:"Calibri","sans-serif"">           y[j] = y[j - 2] + y[j - 15] ;</span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""><o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:10.0pt;font-family:"Calibri","sans-serif"">        }</span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""><o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:10.0pt;font-family:"Calibri","sans-serif"">    }</span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""><o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:10.0pt;font-family:"Calibri","sans-serif"">}</span><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""><o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif"">Part C:<o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""><snip> from the debug dump during the LoopAccessAnalysis phase:<o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""> <o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif"">LAA: Checking memory dependencies<o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif"">LAA: Src Scev: {(8 + %y),+,8}<%for.body3>Sink Scev: {(128 + %y),+,8}<nsw><%for.body3>(Induction step: 1)<o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif"">LAA: Distance for   %3 = load i64, i64* %arrayidx6, align 8 to   store i64 %add, i64* %arrayidx8, align 8: 120<o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif"">LAA: Distance 120 that could cause a store-load forwarding conflict<o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""> <o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""> <o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""> <o:p></o:p></span></p>
</div>
<div>
<p class="MsoNormal"><span style="font-size:11.0pt;font-family:"Calibri","sans-serif""> <o:p></o:p></span></p>
</div>
</div>
</body>
</html>