blob: c4610a4873985e6445b40eb57dfa0dc5fb7dafc7 [file] [log] [blame]
<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "xhtml1-transitional.dtd">
<!-- saved from url=(0013)about:internet -->
<?xml-stylesheet type="text/xsl" href="mathml.xsl"?><html xmlns="http://www.w3.org/1999/xhtml" xmlns:pref="http://www.w3.org/2002/Math/preference" xmlns:xlink="http://www.w3.org/1999/xlink" pref:renderer="mathplayer-dl">
<head>
<meta http-equiv="Content-Type" content="text/html; charset=UTF-8" />
<style xmlns="" type="text/css">
/* This style sets a margin around the entire page */
html, body {
margin: 10px;
}
p {
font: normal 16px verdana, sans-serif;
margin: 0;
padding-bottom:12px;
}
h1 {
font: bold 25px verdana, sans-serif;
margin-top: 0;
margin-bottom: 3px;
padding-top: 0;
padding-bottom: 0;
}
h2 {
font: bold 19px verdana, sans-serif;
margin-top: 28px;
margin-bottom: 3px;
padding-top: 0;
padding-bottom: 0;
}
h3 {
font: bold 19px verdana, sans-serif !important;
margin-top: 28px;
margin-bottom: 3px;
padding-top: 0;
padding-bottom: 0;
}
li {
font: normal 16px verdana, sans-serif;
margin-top: 0;
margin-bottom: 18px;
padding-top: 0;
padding-bottom: 0;
}
.pdparam {
font: italic 16px verdana, sans-serif;
}
.term {
font: italic 16px verdana, sans-serif;
font-weight: normal;
}
.type {
font: normal 16px verdana, sans-serif !important;
}
.parameter {
font-style: italic;
}
a:link, a:visited {
color: blue;
text-decoration: none;
font: normal 16px;
}
a:hover {
background-color: #FFFF99;
font: normal 16px;
}
div.funcsynopsis {
text-align: left;
background-color: #e6e6e6;
font: normal 16px verdana, sans-serif;
padding-top: 10px;
padding-bottom: 10px;
}
div.funcsynopsis table {
border-collapse: separate;
font: normal 16px verdana, sans-serif;
}
div.funcsynopsis td {
background-color: #e6e6e6;
border: 0 solid #000;
padding: 1px;
font: normal 16px verdana, sans-serif;
}
div.refsect1 {
font-family: verdana, sans-serif;
font-size: 16px;
}
code.constant {
font: normal 16px courier new, monospace !important;
}
span.errorname {
font: normal 16px verdana, sans-serif !important;
}
code.function {
font: bold 16px verdana, sans-serif !important;
}
b.fsfunc {
font: bold 16px verdana, sans-serif !important;
}
code.varname {
font: italic 16px verdana, sans-serif;
}
code.replaceable {
font: italic 16px courier new, monospace;
}
code.funcdef {
font: normal 16px verdana, sans-serif !important;
}
.citerefentry {
font: normal 16px verdana, sans-serif !important;
}
.parameter {
font-style: italic;
}
code.fsfunc {
font: normal 16px verdana, sans-serif !important;
}
/* PARAMETER: This style controls spacing between the terms in Parameter section */
dt {
margin-top: 15px;
}
/* TABLES: These styles apply to all tables OTHER than the Synopsis and Example tables */
div.refsect1 table {
width: 100%;
margin-top: 10px;
background-color: #FFF;
border-collapse: collapse;
border-color: #000;
border-width: 1px;
font: normal 16px verdana, sans-serif;
}
div.refsect1 th {
border-collapse: collapse;
border-color: #000;
border-width: 1px;
font: bold 16px verdana, sans-serif;
}
div.refsect1 td {
background-color: #FFF;
padding: 5px;
vertical-align: text-top;
border-collapse: collapse;
border-color: #000;
border-width: 1px;
font: normal 16px verdana, sans-serif;
}
div.refsect1 p{
font: normal 16px verdana, sans-serif;
margin-top: 8px;
margin-bottom: 8px;
padding-top: 0;
padding-bottom: 0;
}
/* EXAMPLE: These styles apply only to the Example section */
div.refsect2 {
font: normal 16px courier new, monospace !important;
}
div.refsect2 table {
margin-top: 0;
background-color: #e6e6e6;
width: 100%;
border: 0 solid #000;
padding: 2px;
font: normal 16px courier new, monospace !important;
}
div.refsect2 td {
background-color: #e6e6e6;
font: normal 16px courier new, monospace !important;
white-space:pre;
}
/* COPYRIGHT: This style formats the text of the copyright statement at the bottom of the page */
div.refsect3 {
font: normal 11px verdana, sans-serif;
margin-top: 50px;
margin-bottom: 20px;
padding-top: 0;
padding-bottom: 0;
}
</style>
<title>Function Qualifiers</title>
<meta name="generator" content="DocBook XSL Stylesheets V1.79.1" />
<meta name="keywords" content=" &#10; Function Qualifiers&#10; " />
</head>
<body>
<div class="refentry">
<a id="idm1"></a>
<div class="titlepage"></div>
<div xmlns="" class="refnamediv">
<a xmlns="http://www.w3.org/1999/xhtml" id="Function Qualifiers"></a>
<h1>Function Qualifiers</h1>
<p>
Qualifiers for kernel functions.
</p>
</div>
<div class="refsect2">
<a id="synopsis"></a>
<h3>
</h3>
<div class="informaltable">
<table class="informaltable" border="0">
<colgroup>
<col align="left" class="col1" />
</colgroup>
<tbody>
<tr>
<td align="left">
__kernel
kernel
__attribute__((vec_type_hint(&lt;type<span class="emphasis"><em>n</em></span>&gt;)))
__attribute__((work_group_size_hint(<span class="emphasis"><em>X</em></span>, <span class="emphasis"><em>Y</em></span>, <span class="emphasis"><em>Z</em></span>)))
__attribute__((reqd_work_group_size(<span class="emphasis"><em>X</em></span>, <span class="emphasis"><em>Y</em></span>, <span class="emphasis"><em>Z</em></span>)))
</td>
</tr>
</tbody>
</table>
</div>
</div>
<div class="refsect1">
<a id="description"></a>
<h2>Description</h2>
<p>
The <code class="function">__kernel</code> (or <code class="function">kernel</code>) qualifier
declares a function to be a kernel that can be
executed by an application on an OpenCL device(s).
The following rules apply to functions that
are declared with this qualifier:
</p>
<div class="itemizedlist">
<ul class="itemizedlist" style="list-style-type: bullet; ">
<li class="listitem" style="list-style-type: disc">
<p>
It can be executed on the device only
</p>
</li>
<li class="listitem" style="list-style-type: disc">
<p>
It can be called by the host
</p>
</li>
<li class="listitem" style="list-style-type: disc">
<p>
It is just a regular function call if a <code class="function">__kernel</code>
function is called by another kernel function.
</p>
</li>
</ul>
</div>
<p>
The <code class="function">__kernel</code> qualifier can be used with the keyword
<a class="citerefentry" href="attribute.html"><span class="citerefentry"><span class="refentrytitle">__attribute__</span></span></a> to declare additional
information about the kernel function as described below.
</p>
<p>
The optional
<code class="constant">__attribute__((vec_type_hint(&lt;type<span class="emphasis"><em>n</em></span>&gt;)))</code>
is a hint to the
compiler and is intended to be a representation of the computational
<span class="emphasis"><em>width</em></span> of the
<code class="function">__kernel</code>,
and should serve as the basis for calculating processor
bandwidth utilization when the compiler
is looking to autovectorize the code.
<code class="constant">vec_type_hint (&lt;type<span class="emphasis"><em>n</em></span>&gt;)</code>
shall be one of the built-in scalar or vector data type described in
tables 6.1 and 6.2.
If
<code class="constant">vec_type_hint (&lt;type<span class="emphasis"><em>n</em></span>&gt;)</code>
is not specified, the default value is <span class="type">int</span>.
</p>
<p>
The
<code class="constant">__attribute__((vec_type_hint(int)))</code>
is the default type.
</p>
<p>
For example, where the developer specified a width of <span class="type">float4</span>,
the compiler should assume
that the computation usually uses up 4 lanes of a float vector,
and would decide to merge work-items or possibly even separate
one work-item into many threads to better match the hardware
capabilities. A conforming implementation is not required
to autovectorize code, but shall
support the hint. A compiler may autovectorize, even if no
hint is provided. If an
implementation merges <code class="constant">N</code> work-items into one thread,
it is responsible for correctly handling
cases where the number of global or local work-items
in any dimension modulo <code class="constant">N</code> is not zero.
</p>
<p>
If for example, a <code class="function">__kernel</code> is declared with
<code class="constant">__attribute__(( vec_type_hint (float4)))</code>
(meaning that most operations in the <code class="function">__kernel</code>
are explicitly vectorized using
<span class="type">float4</span>) and the kernel is running using
Intel® Advanced Vector Instructions
(Intel® AVX)
which implements a 8-float-wide vector unit,
the autovectorizer might choose to merge two
work-items to one thread, running a second
work-item in the high half of the 256-bit AVX
register.
</p>
<p>
As another example, a Power4 machine has two scalar
double precision floating-point units with
an 6-cycle deep pipe. An autovectorizer for the
Power4 machine might choose to interleave six
<code class="constant">__attribute__(( vec_type_hint (double2))) __kernel</code>s
into one hardware
thread, to ensure that there is always 12-way
parallelism available to saturate the FPUs. It might
also choose to merge 4 or 8 work-items (or some
other number) if it concludes that these are
better choices, due to resource utilization
concerns or some preference for divisibility by 2.
</p>
<p>
The optional
<code class="constant">__attribute__((work_group_size_hint(<span class="emphasis"><em>X</em></span>, <span class="emphasis"><em>Y</em></span>, <span class="emphasis"><em>Z</em></span>)))</code>
is a hint to the
compiler and is intended to specify the work-group size
that may be used i.e. value most likely to
be specified by the <code class="varname">local_work_size</code> argument to
<a class="citerefentry" href="clEnqueueNDRangeKernel.html"><span class="citerefentry"><span class="refentrytitle">clEnqueueNDRangeKernel</span></span></a>.
For example the
<code class="constant">__attribute__((work_group_size_hint(1, 1, 1)))</code>
is a hint to the compiler
that the kernel will most likely be executed
with a work-group size of 1.
</p>
<p>
The optional
<code class="constant">__attribute__((reqd_work_group_size(<span class="emphasis"><em>X</em></span>, <span class="emphasis"><em>Y</em></span>, <span class="emphasis"><em>Z</em></span>)))</code>
is the work-group size that must be used as the
<code class="varname">local_work_size</code> argument to
<a class="citerefentry" href="clEnqueueNDRangeKernel.html"><span class="citerefentry"><span class="refentrytitle">clEnqueueNDRangeKernel</span></span></a>.
This allows the compiler to optimize the generated
code appropriately for this kernel. The optional
<code class="constant">__attribute__((reqd_work_group_size(<span class="emphasis"><em>X</em></span>, <span class="emphasis"><em>Y</em></span>, <span class="emphasis"><em>Z</em></span>)))</code>,
if specified, must be (1, 1, 1) if the kernel is executed via
<a class="citerefentry" href="clEnqueueTask.html"><span class="citerefentry"><span class="refentrytitle">clEnqueueTask</span></span></a>.
</p>
<p>
If <code class="varname">Z</code> is one, the <code class="varname">work_dim</code> argument to
<a class="citerefentry" href="clEnqueueNDRangeKernel.html"><span class="citerefentry"><span class="refentrytitle">clEnqueueNDRangeKernel</span></span></a>
can be 2 or 3. If <code class="varname">Y</code> and <code class="varname">Z</code> are
one, the <code class="varname">work_dim</code> argument to
<a class="citerefentry" href="clEnqueueNDRangeKernel.html"><span class="citerefentry"><span class="refentrytitle">clEnqueueNDRangeKernel</span></span></a>
can be 1, 2 or 3.
</p>
</div>
<div class="refsect1">
<a id="notes"></a>
<h2>Notes</h2>
<p>
Implicit in autovectorization is the assumption that
any libraries called from the
__kernel must be recompilable at
run time to handle cases where the compiler decides to
merge or separate workitems. This probably means that such
libraries can never be hard coded binaries or that hard
coded binaries must be accompanied either by source or some
retargetable intermediate representation. This may be
a code security question for some.
</p>
</div>
<div class="refsect2">
<a id="example1"></a>
<h3>
Example
</h3>
<div class="informaltable">
<table class="informaltable" border="0">
<colgroup>
<col align="left" class="col1" />
</colgroup>
<tbody>
<tr>
<td align="left">
// autovectorize assuming float4 as the
// basic computation width
__kernel __attribute__((vec_type_hint(float4)))
void foo( __global float4 *p ) { ....
// autovectorize assuming double as the
// basic computation width
__kernel __attribute__((vec_type_hint(double)))
void foo( __global float4 *p ){ ....
// autovectorize assuming int (default)
// as the basic computation width
__kernel
void foo( __global float4 *p ){ ....
</td>
</tr>
</tbody>
</table>
</div>
</div>
<div class="refsect1">
<a id="specification"></a>
<h2>Specification</h2>
<p>
<img src="pdficon_small1.gif" />
<a href="http://www.khronos.org/registry/cl/specs/opencl-1.0.pdf#page=154" target="OpenCL Spec">OpenCL Specification</a>
</p>
</div>
<div class="refsect1">
<a id="seealso"></a>
<h2>Also see</h2>
<p>
<a class="citerefentry" href="clEnqueueNDRangeKernel.html"><span class="citerefentry"><span class="refentrytitle">clEnqueueNDRangeKernel</span></span></a>
<a class="citerefentry" href="clEnqueueTask.html"><span class="citerefentry"><span class="refentrytitle">clEnqueueTask</span></span></a>
</p>
</div>
<div xmlns="" class="refsect3" lang="en" xml:lang="en"><a xmlns="http://www.w3.org/1999/xhtml" id="Copyright"></a><h4 xmlns="http://www.w3.org/1999/xhtml"></h4><img xmlns="http://www.w3.org/1999/xhtml" src="KhronosLogo.jpg" /><p xmlns="http://www.w3.org/1999/xhtml"></p>Copyright © 2007-2009 The Khronos Group Inc.
Permission is hereby granted, free of charge, to any person obtaining a
copy of this software and/or associated documentation files (the
"Materials"), to deal in the Materials without restriction, including
without limitation the rights to use, copy, modify, merge, publish,
distribute, sublicense, and/or sell copies of the Materials, and to
permit persons to whom the Materials are furnished to do so, subject to
the condition that this copyright notice and permission notice shall be included
in all copies or substantial portions of the Materials.
</div>
</div>
</body>
</html>