<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:"MS Mincho";
panose-1:2 2 6 9 4 2 5 8 3 4;}
@font-face
{font-family:"MS Mincho";
panose-1:2 2 6 9 4 2 5 8 3 4;}
@font-face
{font-family:Consolas;
panose-1:2 11 6 9 2 2 4 3 2 4;}
@font-face
{font-family:"\@MS Mincho";
panose-1:2 2 6 9 4 2 5 8 3 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;}
span.EmailStyle17
{mso-style-type:personal-reply;
font-family:"Calibri","sans-serif";
color:#1F497D;}
.MsoChpDefault
{mso-style-type:export-only;
mso-fareast-language:JA;}
@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" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">Hi Eli,<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">Thanks for the feedback. Please see below.<br>
- Gao.<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA"><o:p> </o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">From: Eli Friedman [mailto:eli.friedman@gmail.com]
<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">Sent: Thursday, September 19, 2013 12:31 PM<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">To: Gao, Yunzhong<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">Cc: llvmdev@cs.uiuc.edu<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">Subject: Re: [LLVMdev] Proposal to improve vzeroupper optimization strategy<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA"><o:p> </o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">> This is essentially equivalent to "don't insert vzeroupper anywhere", as<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">> far as I can tell. (The case of SSE instructions without a v- prefixed<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">> equivalent is rare enough we can separate it from this discussion.)<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA"><o:p> </o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">So will you be interested in a patch that disables vzeroupper by default?<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA"><o:p> </o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">I implemented this possibly over-engineering solution in our local tree to work<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">around some bad instruction selection issues in LLVM backend. When benchmarking<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">on our game codes, I noticed that sometimes legacy SSE instructions were<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">selected despite existence of AVX equivalent, in which case the vzeroupper<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">instruction was needed. And it is much easier to detect existence of vzeroupper<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">instruction than to detect each single legacy SSE instructions.<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA"><o:p> </o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">The instruction selection issues were later fixed in our tree (patches to be<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">submitted later), at least for the handful of games I tested on. So a simple<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">change to just disable vzeroupper by default will be acceptable to us as well.<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA"><o:p> </o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">> The reason we need vzeroupper in the first place is because we can't assume<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">> other functions won't use legacy SSE instructions; for example, on most<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">> systems, calling sin() will use legacy SSE instructions. I mean, if you can<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">> make some unusual guarantee about your platform, it might make sense to<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">> disable vzeroupper generation in general, but it simply doesn't make sense<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">> on most platforms.<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA"><o:p> </o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">I am confused by this point. By "most systems," do you have in mind a platform<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">where the sin() function was compiled by gcc but the application codes were<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">compiled by clang?<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA"><o:p> </o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">If the sin() function was compiled by clang for a platform that supports AVX<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">instructions, I do not expect it to contain legacy SSE instructions. Is it not<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">the case for your platform?<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA"><o:p> </o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">I just looked at the library code for our sin() function and I do not see any<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">legacy SSE instructions (but for license restrictions I cannot share our<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">library codes; sorry).<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA"><o:p> </o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">> If you want a mechanism to disable vzeroupper generation for particular<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">> function calls, that might make sense...<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA">> -Eli<o:p></o:p></span></p>
<p class="MsoNormal" style="text-autospace:none"><span style="font-size:9.5pt;font-family:Consolas;mso-fareast-language:JA"><o:p> </o:p></span></p>
</div>
</body>
</html>