<html>
<head>
<meta http-equiv="Content-Type" content="text/html;
charset=windows-1252">
</head>
<body text="#000000" bgcolor="#FFFFFF">
<div class="moz-cite-prefix">On 7/20/2018 11:02 AM, via cfe-dev
wrote:<br>
</div>
<blockquote type="cite"
cite="mid:004701d42053$c9109ab0$5b31d010$@gmail.com">
<meta http-equiv="Content-Type" content="text/html;
charset=windows-1252">
<meta name="Generator" content="Microsoft Word 15 (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: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:#0563C1;
text-decoration:underline;}
a:visited, span.MsoHyperlinkFollowed
{mso-style-priority:99;
color:#954F72;
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: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]-->
<div class="WordSection1">
<p class="MsoNormal">Hello all!<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">I’ve been mucking around in an old codebase
at work looking for easy performance wins. One avenue involves
replacing a switch-based variable assignment with something
derived from the parity of an input variable. I was pretty
surprised when I saw the generated assembly, and I’m wondering
about the reasoning behind it.<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">In short, it boils down to the assembly <span
style="font-family:Consolas">__builtin_parity()</span>
produces. Clang 6.0.1 (and trunk on Godbolt) produces:<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal"><span style="font-family:Consolas">parity(int):
# @parity(int)<o:p></o:p></span></p>
<p class="MsoNormal"><span style="font-family:Consolas">
mov eax, edi<o:p></o:p></span></p>
<p class="MsoNormal"><span style="font-family:Consolas">
shr eax<o:p></o:p></span></p>
<p class="MsoNormal"><span style="font-family:Consolas">
and eax, 1431655765<o:p></o:p></span></p>
<p class="MsoNormal"><span style="font-family:Consolas">
sub edi, eax<o:p></o:p></span></p>
<p class="MsoNormal"><span style="font-family:Consolas">
mov eax, edi<o:p></o:p></span></p>
<p class="MsoNormal"><span style="font-family:Consolas">
and eax, 858993459<o:p></o:p></span></p>
<p class="MsoNormal"><span style="font-family:Consolas">
shr edi, 2<o:p></o:p></span></p>
<p class="MsoNormal"><span style="font-family:Consolas">
and edi, 858993459<o:p></o:p></span></p>
<p class="MsoNormal"><span style="font-family:Consolas">
add edi, eax<o:p></o:p></span></p>
<p class="MsoNormal"><span style="font-family:Consolas">
mov eax, edi<o:p></o:p></span></p>
<p class="MsoNormal"><span style="font-family:Consolas">
shr eax, 4<o:p></o:p></span></p>
<p class="MsoNormal"><span style="font-family:Consolas">
add eax, edi<o:p></o:p></span></p>
<p class="MsoNormal"><span style="font-family:Consolas">
and eax, 17764111<o:p></o:p></span></p>
<p class="MsoNormal"><span style="font-family:Consolas">
imul eax, eax, 16843009<o:p></o:p></span></p>
<p class="MsoNormal"><span style="font-family:Consolas">
shr eax, 24<o:p></o:p></span></p>
<p class="MsoNormal"><span style="font-family:Consolas">
and eax, 1<o:p></o:p></span></p>
<p class="MsoNormal"><span style="font-family:Consolas">
ret<o:p></o:p></span></p>
<p class="MsoNormal"><span style="font-family:Consolas"><o:p> </o:p></span></p>
<p class="MsoNormal">While GCC 8.1.0 (and trunk on Godbolt)
produces<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal"><span style="font-family:Consolas">parity(int):<o:p></o:p></span></p>
<p class="MsoNormal"><span style="font-family:Consolas">
mov eax, edi<o:p></o:p></span></p>
<p class="MsoNormal"><span style="font-family:Consolas">
shr edi, 16<o:p></o:p></span></p>
<p class="MsoNormal"><span style="font-family:Consolas">
xor eax, edi<o:p></o:p></span></p>
<p class="MsoNormal"><span style="font-family:Consolas">
xor al, ah<o:p></o:p></span></p>
<p class="MsoNormal"><span style="font-family:Consolas">
setnp al<o:p></o:p></span></p>
<p class="MsoNormal"><span style="font-family:Consolas">
movzx eax, al<o:p></o:p></span></p>
<p class="MsoNormal"><span style="font-family:Consolas">
ret<o:p></o:p></span></p>
<p class="MsoNormal"><span style="font-family:Consolas"><o:p> </o:p></span></p>
<p class="MsoNormal">I know a popcnt followed by an and would be
better, but unfortunately some of my users don’t have
computers that support the popcnt instruction, so I can’t use
a newer -march flag.<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
<p class="MsoNormal">Could someone explain why the difference
between Clang and GCC here, and whether it should make a
difference? The code in question is in a hot loop in my code,
so I’d imagine the size difference could impact unrolling (and
result in icache differences too), but I haven’t finished
poking around with benchmarks.<o:p></o:p></p>
<p class="MsoNormal"><o:p> </o:p></p>
</div>
</blockquote>
<br>
LLVM doesn't have any special support for computing parity, so it's
just getting lowered to "popcount(x)&1"; if your target doesn't
have a popcount instruction, it uses the generic expansion
(something like
<a class="moz-txt-link-freetext" href="https://graphics.stanford.edu/~seander/bithacks.html#CountBitsSetParallel">https://graphics.stanford.edu/~seander/bithacks.html#CountBitsSetParallel</a>
). This is obviously not the fastest lowering, but computing parity
is not a common operation, so nobody has spent any time optimizing
it.<br>
<br>
-Eli<br>
<pre class="moz-signature" cols="72">--
Employee of Qualcomm Innovation Center, Inc.
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, a Linux Foundation Collaborative Project</pre>
</body>
</html>