<?xml version="1.0" encoding="utf-8" ?>
<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "http://www.w3.org/TR/xhtml1/DTD/xhtml1-transitional.dtd">
<html xmlns="http://www.w3.org/1999/xhtml" xml:lang="en" lang="en">
<head>
<meta http-equiv="Content-Type" content="text/html; charset=utf-8" />
<meta name="generator" content="Docutils 0.6: http://docutils.sourceforge.net/" />
<title>libbcc: A Versatile Bitcode Execution Engine for Mobile Devices</title>
<style type="text/css">

/*
:Author: David Goodger (goodger@python.org)
:Id: $Id: html4css1.css 5951 2009-05-18 18:03:10Z milde $
:Copyright: This stylesheet has been placed in the public domain.

Default cascading style sheet for the HTML output of Docutils.

See http://docutils.sf.net/docs/howto/html-stylesheets.html for how to
customize this style sheet.
*/

/* used to remove borders from tables and images */
.borderless, table.borderless td, table.borderless th {
  border: 0 }

table.borderless td, table.borderless th {
  /* Override padding for "table.docutils td" with "! important".
     The right padding separates the table cells. */
  padding: 0 0.5em 0 0 ! important }

.first {
  /* Override more specific margin styles with "! important". */
  margin-top: 0 ! important }

.last, .with-subtitle {
  margin-bottom: 0 ! important }

.hidden {
  display: none }

a.toc-backref {
  text-decoration: none ;
  color: black }

blockquote.epigraph {
  margin: 2em 5em ; }

dl.docutils dd {
  margin-bottom: 0.5em }

/* Uncomment (and remove this text!) to get bold-faced definition list terms
dl.docutils dt {
  font-weight: bold }
*/

div.abstract {
  margin: 2em 5em }

div.abstract p.topic-title {
  font-weight: bold ;
  text-align: center }

div.admonition, div.attention, div.caution, div.danger, div.error,
div.hint, div.important, div.note, div.tip, div.warning {
  margin: 2em ;
  border: medium outset ;
  padding: 1em }

div.admonition p.admonition-title, div.hint p.admonition-title,
div.important p.admonition-title, div.note p.admonition-title,
div.tip p.admonition-title {
  font-weight: bold ;
  font-family: sans-serif }

div.attention p.admonition-title, div.caution p.admonition-title,
div.danger p.admonition-title, div.error p.admonition-title,
div.warning p.admonition-title {
  color: red ;
  font-weight: bold ;
  font-family: sans-serif }

/* Uncomment (and remove this text!) to get reduced vertical space in
   compound paragraphs.
div.compound .compound-first, div.compound .compound-middle {
  margin-bottom: 0.5em }

div.compound .compound-last, div.compound .compound-middle {
  margin-top: 0.5em }
*/

div.dedication {
  margin: 2em 5em ;
  text-align: center ;
  font-style: italic }

div.dedication p.topic-title {
  font-weight: bold ;
  font-style: normal }

div.figure {
  margin-left: 2em ;
  margin-right: 2em }

div.footer, div.header {
  clear: both;
  font-size: smaller }

div.line-block {
  display: block ;
  margin-top: 1em ;
  margin-bottom: 1em }

div.line-block div.line-block {
  margin-top: 0 ;
  margin-bottom: 0 ;
  margin-left: 1.5em }

div.sidebar {
  margin: 0 0 0.5em 1em ;
  border: medium outset ;
  padding: 1em ;
  background-color: #ffffee ;
  width: 40% ;
  float: right ;
  clear: right }

div.sidebar p.rubric {
  font-family: sans-serif ;
  font-size: medium }

div.system-messages {
  margin: 5em }

div.system-messages h1 {
  color: red }

div.system-message {
  border: medium outset ;
  padding: 1em }

div.system-message p.system-message-title {
  color: red ;
  font-weight: bold }

div.topic {
  margin: 2em }

h1.section-subtitle, h2.section-subtitle, h3.section-subtitle,
h4.section-subtitle, h5.section-subtitle, h6.section-subtitle {
  margin-top: 0.4em }

h1.title {
  text-align: center }

h2.subtitle {
  text-align: center }

hr.docutils {
  width: 75% }

img.align-left, .figure.align-left{
  clear: left ;
  float: left ;
  margin-right: 1em }

img.align-right, .figure.align-right {
  clear: right ;
  float: right ;
  margin-left: 1em }

.align-left {
  text-align: left }

.align-center {
  clear: both ;
  text-align: center }

.align-right {
  text-align: right }

/* reset inner alignment in figures */
div.align-right {
  text-align: left }

/* div.align-center * { */
/*   text-align: left } */

ol.simple, ul.simple {
  margin-bottom: 1em }

ol.arabic {
  list-style: decimal }

ol.loweralpha {
  list-style: lower-alpha }

ol.upperalpha {
  list-style: upper-alpha }

ol.lowerroman {
  list-style: lower-roman }

ol.upperroman {
  list-style: upper-roman }

p.attribution {
  text-align: right ;
  margin-left: 50% }

p.caption {
  font-style: italic }

p.credits {
  font-style: italic ;
  font-size: smaller }

p.label {
  white-space: nowrap }

p.rubric {
  font-weight: bold ;
  font-size: larger ;
  color: maroon ;
  text-align: center }

p.sidebar-title {
  font-family: sans-serif ;
  font-weight: bold ;
  font-size: larger }

p.sidebar-subtitle {
  font-family: sans-serif ;
  font-weight: bold }

p.topic-title {
  font-weight: bold }

pre.address {
  margin-bottom: 0 ;
  margin-top: 0 ;
  font: inherit }

pre.literal-block, pre.doctest-block {
  margin-left: 2em ;
  margin-right: 2em }

span.classifier {
  font-family: sans-serif ;
  font-style: oblique }

span.classifier-delimiter {
  font-family: sans-serif ;
  font-weight: bold }

span.interpreted {
  font-family: sans-serif }

span.option {
  white-space: nowrap }

span.pre {
  white-space: pre }

span.problematic {
  color: red }

span.section-subtitle {
  /* font-size relative to parent (h1..h6 element) */
  font-size: 80% }

table.citation {
  border-left: solid 1px gray;
  margin-left: 1px }

table.docinfo {
  margin: 2em 4em }

table.docutils {
  margin-top: 0.5em ;
  margin-bottom: 0.5em }

table.footnote {
  border-left: solid 1px black;
  margin-left: 1px }

table.docutils td, table.docutils th,
table.docinfo td, table.docinfo th {
  padding-left: 0.5em ;
  padding-right: 0.5em ;
  vertical-align: top }

table.docutils th.field-name, table.docinfo th.docinfo-name {
  font-weight: bold ;
  text-align: left ;
  white-space: nowrap ;
  padding-left: 0 }

h1 tt.docutils, h2 tt.docutils, h3 tt.docutils,
h4 tt.docutils, h5 tt.docutils, h6 tt.docutils {
  font-size: 100% }

ul.auto-toc {
  list-style-type: none }

</style>
</head>
<body>
<div class="document" id="libbcc-a-versatile-bitcode-execution-engine-for-mobile-devices">
<h1 class="title">libbcc: A Versatile Bitcode Execution Engine for Mobile Devices</h1>

<div class="section" id="introduction">
<h1>Introduction</h1>
<p>libbcc is an LLVM bitcode execution engine that compiles the bitcode
to an in-memory executable. libbcc is versatile because:</p>
<ul class="simple">
<li>it implements both AOT (Ahead-of-Time) and JIT (Just-in-Time)
compilation.</li>
<li>Android devices demand fast start-up time, small size, and high
performance <em>at the same time</em>. libbcc attempts to address these
design constraints.</li>
<li>it supports on-device linking. Each device vendor can supply his or
her own runtime bitcode library (lib*.bc) that differentiates his or
her system. Specialization becomes ecosystem-friendly.</li>
</ul>
<p>libbcc provides:</p>
<ul class="simple">
<li>a <em>just-in-time bitcode compiler</em>, which translates the LLVM bitcode
into machine code</li>
<li>a <em>caching mechanism</em>, which can:<ul>
<li>after each compilation, serialize the in-memory executable into a
cache file.  Note that the compilation is triggered by a cache
miss.</li>
<li>load from the cache file upon cache-hit.</li>
</ul>
</li>
</ul>
<p>Highlights of libbcc are:</p>
<ul>
<li><p class="first">libbcc supports bitcode from various language frontends, such as
Renderscript, GLSL (pixelflinger2).</p>
</li>
<li><p class="first">libbcc strives to balance between library size, launch time and
steady-state performance:</p>
<ul>
<li><p class="first">The size of libbcc is aggressively reduced for mobile devices. We
customize and improve upon the default Execution Engine from
upstream. Otherwise, libbcc's execution engine can easily become
at least 2 times bigger.</p>
</li>
<li><p class="first">To reduce launch time, we support caching of
binaries. Just-in-Time compilation are oftentimes Just-too-Late,
if the given apps are performance-sensitive. Thus, we implemented
AOT to get the best of both worlds: Fast launch time and high
steady-state performance.</p>
<p>AOT is also important for projects such as NDK on LLVM with
portability enhancement. Launch time reduction after we
implemented AOT is signficant:</p>
<pre class="literal-block">
Apps          libbcc without AOT       libbcc with AOT
              launch time in libbcc    launch time in libbcc
App_1            1218ms                   9ms
App_2            842ms                    4ms
Wallpaper:
  MagicSmoke     182ms                    3ms
  Halo           127ms                    3ms
Balls            149ms                    3ms
SceneGraph       146ms                    90ms
Model            104ms                    4ms
Fountain         57ms                     3ms
</pre>
<p>AOT also masks the launching time overhead of on-device linking
and helps it become reality.</p>
</li>
<li><p class="first">For steady-state performance, we enable VFP3 and aggressive
optimizations.</p>
</li>
</ul>
</li>
<li><p class="first">Currently we disable Lazy JITting.</p>
</li>
</ul>
</div>
<div class="section" id="api">
<h1>API</h1>
<p><strong>Basic:</strong></p>
<ul class="simple">
<li><strong>bccCreateScript</strong> - Create new bcc script</li>
<li><strong>bccRegisterSymbolCallback</strong> - Register the callback function for external
symbol lookup</li>
<li><strong>bccReadBC</strong> - Set the source bitcode for compilation</li>
<li><strong>bccReadModule</strong> - Set the llvm::Module for compilation</li>
<li><strong>bccLinkBC</strong> - Set the library bitcode for linking</li>
<li><strong>bccPrepareExecutable</strong> - <em>deprecated</em> - Use bccPrepareExecutableEx instead</li>
<li><strong>bccPrepareExecutableEx</strong> - Create the in-memory executable by either
just-in-time compilation or cache loading</li>
<li><strong>bccGetFuncAddr</strong> - Get the entry address of the function</li>
<li><strong>bccDisposeScript</strong> - Destroy bcc script and release the resources</li>
<li><strong>bccGetError</strong> - <em>deprecated</em> - Don't use this</li>
</ul>
<p><strong>Reflection:</strong></p>
<ul class="simple">
<li><strong>bccGetExportVarCount</strong> - Get the count of exported variables</li>
<li><strong>bccGetExportVarList</strong> - Get the addresses of exported variables</li>
<li><strong>bccGetExportFuncCount</strong> - Get the count of exported functions</li>
<li><strong>bccGetExportFuncList</strong> - Get the addresses of exported functions</li>
<li><strong>bccGetPragmaCount</strong> - Get the count of pragmas</li>
<li><strong>bccGetPragmaList</strong> - Get the pragmas</li>
</ul>
<p><strong>Debug:</strong></p>
<ul class="simple">
<li><strong>bccGetFuncCount</strong> - Get the count of functions (including non-exported)</li>
<li><strong>bccGetFuncInfoList</strong> - Get the function information (name, base, size)</li>
</ul>
</div>
<div class="section" id="cache-file-format">
<h1>Cache File Format</h1>
<p>A cache file (denoted as *.oBCC) for libbcc consists of several sections:
header, string pool, dependencies table, relocation table, exported
variable list, exported function list, pragma list, function information
table, and bcc context.  Every section should be aligned to a word size.
Here is the brief description of each sections:</p>
<ul class="simple">
<li><strong>Header</strong> (MCO_Header) - The header of a cache file. It contains the
magic word, version, machine integer type information (the endianness,
the size of off_t, size_t, and ptr_t), and the size
and offset of other sections.  The header section is guaranteed
to be at the beginning of the cache file.</li>
<li><strong>String Pool</strong> (MCO_StringPool) - A collection of serialized variable
length strings.  The strp_index in the other part of the cache file
represents the index of such string in this string pool.</li>
<li><strong>Dependencies Table</strong> (MCO_DependencyTable) - The dependencies table.
This table stores the resource name (or file path), the resource
type (rather in APK or on the file system), and the SHA1 checksum.</li>
<li><strong>Relocation Table</strong> (MCO_RelocationTable) - <em>not enabled</em></li>
<li><strong>Exported Variable List</strong> (MCO_ExportVarList) -
The list of the addresses of exported variables.</li>
<li><strong>Exported Function List</strong> (MCO_ExportFuncList) -
The list of the addresses of exported functions.</li>
<li><strong>Pragma List</strong> (MCO_PragmaList) - The list of pragma key-value pair.</li>
<li><strong>Function Information Table</strong> (MCO_FuncTable) - This is a table of
function information, such as function name, function entry address,
and function binary size.  Besides, the table should be ordered by
function name.</li>
<li><strong>Context</strong> - The context of the in-memory executable, including
the code and the data.  The offset of context should aligned to
a page size, so that we can mmap the context directly into memory.</li>
</ul>
<p>For furthur information, you may read <a class="reference external" href="include/bcc/bcc_cache.h">bcc_cache.h</a>,
<a class="reference external" href="lib/bcc/CacheReader.cpp">CacheReader.cpp</a>, and
<a class="reference external" href="lib/bcc/CacheWriter.cpp">CacheWriter.cpp</a> for details.</p>
</div>
<div class="section" id="jit-ed-code-calling-conventions">
<h1>JIT'ed Code Calling Conventions</h1>
<ol class="arabic">
<li><p class="first">Calls from Execution Environment or from/to within script:</p>
<p>On ARM, the first 4 arguments will go into r0, r1, r2, and r3, in that order.
The remaining (if any) will go through stack.</p>
<p>For ext_vec_types such as float2, a set of registers will be used. In the case
of float2, a register pair will be used. Specifically, if float2 is the first
argument in the function prototype, float2.x will go into r0, and float2.y,
r1.</p>
<p>Note: stack will be aligned to the coarsest-grained argument. In the case of
float2 above as an argument, parameter stack will be aligned to an 8-byte
boundary (if the sizes of other arguments are no greater than 8.)</p>
</li>
<li><p class="first">Calls from/to a separate compilation unit: (E.g., calls to Execution
Environment if those runtime library callees are not compiled using LLVM.)</p>
<p>On ARM, we use hardfp.  Note that double will be placed in a register pair.</p>
</li>
</ol>
</div>
</div>
</body>
</html>