blob: 1cc590813cbcdafe609fd601763d6b5256d1a479 [file] [log] [blame]
<?xml version="1.0" encoding="utf-8" ?>
<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Transitional//EN" "">
<html xmlns="" xml:lang="en" lang="en">
<meta http-equiv="Content-Type" content="text/html; charset=utf-8" />
<meta name="generator" content="Docutils 0.6:" />
<title>libbcc: A Versatile Bitcode Execution Engine for Mobile Devices</title>
<style type="text/css">
:Author: David Goodger (
: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 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% } {
list-style-type: none }
<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">
<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)
<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 their
own runtime bitcode library (lib*.bc) that differentiates their
system. Specialization becomes ecosystem-friendly.</li>
<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
<li>load from the cache file upon cache-hit.</li>
<p>Highlights of libbcc are:</p>
<li><p class="first">libbcc supports bitcode from various language frontends, such as
Renderscript, GLSL (pixelflinger2).</p>
<li><p class="first">libbcc strives to balance between library size, launch time and
steady-state performance:</p>
<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><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
MagicSmoke 182ms 3ms
Halo 127ms 3ms
Balls 149ms 3ms
SceneGraph 146ms 90ms
Model 104ms 4ms
Fountain 57ms 3ms
<p>AOT also masks the launching time overhead of on-device linking
and helps it become reality.</p>
<li><p class="first">For steady-state performance, we enable VFP3 and aggressive
<li><p class="first">Currently we disable Lazy JITting.</p>
<div class="section" id="api">
<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 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 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>
<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>
<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 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,
<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><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>