docs/html/ndk/guides/abis.jd

page.title=ABI Management
@jd:body

<div id="qv-wrapper">
    <div id="qv">
      <h2>On this page</h2>

      <ol>
        <li><a href="#sa">Supported ABIs</a></li>
        <li><a href="#gc">Generating Code for a Specific ABI</a></li>
        <li><a href="#am">ABI Management on the Android Platform</a></li>
      </ol>
    </div>
  </div>

<p>Different Android handsets use different CPUs, which in turn support different instruction sets.
Each combination of CPU and instruction sets has its own Application Binary Interface, or
<i>ABI</i>. The ABI defines, with great precision, how an application's machine code is supposed to
interact with the system at runtime. You must specify an ABI for each CPU architecture you want
your app to work with.</p>

<p>A typical ABI includes the following information:</p>

<ul>
<li>The CPU instruction set(s) that the machine code should use.</li>
<li>The endianness of memory stores and loads at runtime.</li>
<li>The format of executable binaries, such as programs and shared libraries, and
the types of content they support.</li>
<li>Various conventions for passing data between your code and the system.
These conventions include alignment constraints, as well as how the system uses the stack and
registers when it calls functions.</li>
<li>The list of function symbols available to your machine code at runtime,
generally from very specific sets of libraries.</li>
</ul>

<p>This page enumerates the ABIs that the NDK supports, and provides information about how each ABI
works.</p>

<h2 id="sa">Supported ABIs</h2>

<p>Each ABI supports one or more instruction sets. Table 1 provides an at-a-glance overview of
the instruction sets each ABI supports.</p>

<p class="table-caption" id="abi-table">
  <strong>Table 1.</strong> ABIs and supported instruction sets.</p>

<table>
<tr>
<th>ABI</th>
<th>Supported Instruction Set(s)</th>
<th>Notes</th>
</tr>

<tr>
<td><a href="#armeabi">{@code armeabi}</a> </td>
<td><li>ARMV5TE and later</li>
<li>Thumb-1</li></td>
<td>No hard float.</td>
</tr>

<tr>
<td><a href="#v7a">{@code armeabi-v7a} ({@code armeabi-v7a-hard)}</a></td>
<td>
<li>armeabi</li>
<li>Thumb-2</li>
<li>VFPv3-D16</li>
<li>Other, optional</li></td>
<td>Hard float when specified as {@code armeabi-v7a-hard}.
Incompatible with ARMv5, v6 devices.</td>
</tr>

<tr>
<td><a href="#arm64-v8a">{@code arm64-v8a}</a></td>
<td><li>AArch-64</li></td>
</tr>

<tr>
<td>
<a href="#x86">{@code x86}</a></td>
<td><li>x86 (IA-32)</li>
<li>MMX</li>
<li>SSE/2/3</li>
<li>SSSE3</li></td>
<td>No support for MOVBE or SSE4.</td>
</tr>

<tr>
<td><a href="#86-64">{@code x86_64}</a> </td>
<td>
<li>x86-64</li>
<li>MMX</li>
<li>SSE/2/3</li>
<li>SSSE3</li>
<li>SSE4.1, 4.2</li>
<li>POPCNT</li></td>
</tr>

<tr>
<td><a href="#mips">{@code mips}</a></td>
<td><li>MIPS32r1 and later</li></td>
<td>Uses hard-float, and assumes a CPU:FPU clock ratio of 2:1 for maximum
compatibility. Provides neither micromips nor MIPS16.</td>
</tr>

<tr>
<td><a href="#mips64">{@code mips64}</a></td>
<td><li>MIPS64r6</li></td><td>
</td>
</tr>
</table>

<p>More detailed information about each ABI appears below.</p>

<h3 id="armeabi">armeabi</h3>
<p>This ABI is for ARM-based CPUs that support at least
the ARMv5TE instruction set. Please refer to the following documentation for
more details:</p>

<ul>
<li><a href="https://www.scss.tcd.ie/~waldroj/3d1/arm_arm.pdf">ARM Architecture
Reference Manual</a></li>
<li><a
href="http://infocenter.arm.com/help/topic/com.arm.doc.ihi0042e/IHI0042E_aapcs.pdf">
Procedure Call Standard for the ARM Architecture</a></li>
<li><a
href="http://infocenter.arm.com/help/topic/com.arm.doc.dui0101a/DUI0101A_Elf.pdf">
ARM ELF File Format</a></li>
<li><a
href="http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.subset.swdev.abi/index.html">Application Binary Interface (ABI) for the ARM Architecture</a></li>
<li><a
href="http://infocenter.arm.com/help/topic/com.arm.doc.ihi0037c/IHI0037C_bpabi.pdf">
Base Platform ABI for the ARM Architecture</a></li>
<li><a
href="http://infocenter.arm.com/help/topic/com.arm.doc.ihi0039c/IHI0039C_clibabi.pdf">
C Library ABI for the ARM Architecture</a></li>
<li><a
href="http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.ihi0041d/index.html">
C++ ABI for the ARM Architecture</a></li>
<li><a
href="http://infocenter.arm.com/help/topic/com.arm.doc.ihi0043d/IHI0043D_rtabi.pdf">
Run-time ABI for the ARM Architecture</a></li>
<li><a href="http://www.sco.com/developers/gabi/2001-04-24/contents.html">ELF
System V Application Binary Interface</a></li>
<li><a href="http://mentorembedded.github.com/cxx-abi/abi.html">Generic/Itanium C++
ABI</a></li>
</ul>

<p>The AAPCS standard defines EABI as a family of similar
but distinct ABIs. Also, Android follows the little-endian
<a href="http://sourcery.mentor.com/sgpp/lite/arm/portal/kbattach142/arm_gnu_linux_ abi.pdf">
ARM GNU/Linux ABI</a>.</p>

<p>This ABI does not support hardware-assisted floating point
computations. Instead, all floating-point operations use software helper
functions from the compiler's {@code libgcc.a} static library.</p>

<p>The armeabi ABI supports ARM’s
<a href="http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.ddi0210c/CACBCAAE.html">
Thumb (a.k.a. Thumb-1) instruction set</a>. The NDK generates Thumb
code by default unless you specify different behavior using the
<code>LOCAL_ARM_MODE</code> variable in your
<a href="{@docRoot}ndk/guides/android_mk.html">{@code Android.mk}</a>
file.</p>

<h3 id="v7a">armeabi-v7a (armeabi-v7a-hard)</h3>
<p>This ABI extends armeabi to include several
<a href="http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.ddi0406c/index.html">
CPU instruction set extensions</a>. The instruction extensions that this Android-specific
ABI supports are:</p>

<ul>
<li>The Thumb-2 instruction set extension, which provides performance comparable to 32-bit ARM
instructions with similar compactness to Thumb-1.</li>
<li>The VFP hardware-FPU instructions. More specifically, VFPv3-D16, which
includes 16 dedicated 64-bit floating point registers, in addition to another
16 32-bit registers from the ARM core.</li>
</ul>

<p>Other extensions that the v7-a ARM spec describes, including
<a href="http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.ddi0388f/Beijfcja.html">
Advanced SIMD</a> (a.k.a. NEON), VFPv3-D32, and ThumbEE, are optional
to this ABI. Since their presence is not guaranteed, the system should check at runtime
whether the extensions are available. If they are not, you must use alternative code paths. This
check is similar to the one that the system typically performs to check or use
<a href="http://en.wikipedia.org/wiki/MMX_%28instruction_set%29">MMX</a>,
<a href="http://en.wikipedia.org/wiki/SSE2">SSE2</a>, and other specialized
instruction sets on x86 CPUs.</p>

<p>For information about how to perform these runtime checks, refer to
<a href="{@docRoot}ndk/guides/cpu-features.html">The {@code cpufeatures} Library</a>.
Also, for information about the NDK's support for building
machine code for NEON, see
<a href="{@docRoot}ndk/guides/cpu-arm-neon.html">NEON Support</a>.</p>

<p>The {@code armeabi-v7a} ABI uses the {@code -mfloat-abi=softfp} switch to
enforce the rule that the compiler must pass all double values in core register pairs during
function calls, instead of dedicated floating-point ones. The system can perform all internal
computations using the FP registers. Doing so speeds up the computations greatly.</p>

<p>Although the requirement to use core register pairs produces a modest performance hit, it ensures
compatibility with all existing armeabi binaries. If you need the additional
performance, you can specify your ABI as {@code armeabi-v7a-hard} instead. Doing so
allows you to use hard floats, while still linking with Android native APIs
that use {@code softfp}. For more information, refer to the comments in
{@code $NDK/tests/device/hard-float/jni/android.mk}.</p>

<p class="note"><strong>Note:</strong> You cannot specify {@code APP_ABI} as both
{@code armeabi-v7a} and {@code armeabi-v7a-hard}. In either case, the build system places the
shared libraries in the {@code armeabi-v7a/} directory.</p>

<h3 id="hard">armeabi-v7a-hard</h3>
<p>This variant of the {@code armeabi-v7a} ABI is unique to the NDK. The NDK build
system adds the following flags in addition to those that it uses for the
{@code armeabi-v7a} ABI:</p>

<pre class="no-pretty-print">
TARGET_CFLAGS += -mhard-float -D_NDK_MATH_NO_SOFTFP=1
TARGET_LDFLAGS += -Wl,--no-warn-mismatch -lm_hard
</pre>

<p>The compiler compiles all code with hard-float, and links it with {@code libm_hard.a}.
This math library is the same one as {@code libm.a}, except that it follows hard-float ABI
conventions. In the APK, the generated shared libraries reside in {@code /lib/armeabi-v7a/}.</p>

<h3 id="arm64-v8a">arm64-v8a</h3>
<p>This ABI is for ARMv8-based CPUs that support AArch64. It also includes the NEON and
VFPv4 instruction sets.</p>

<p>For more information, see the
<a href="http://www.arm.com/files/downloads/ARMv8_Architecture.pdf">ARMv8
Technology Preview</a>, and contact ARM for further details.</p>

<h3 id="x86">x86</h3>
<p>This ABI is for CPUs supporting the instruction set commonly
referred to as "x86" or "IA-32". Characteristics of this ABI include:</p>

<ul>
<li>Instructions normally generated by GCC with compiler flags such as the following:

<pre class="no-pretty-print">
-march=i686 -mtune=intel -mssse3 -mfpmath=sse -m32
</pre>

<p>These flags target the the Pentium Pro instruction set, along with the
the <a href="http://en.wikipedia.org/wiki/MMX_%28instruction_set%29">MMX</a>,
<a href="http://en.wikipedia.org/wiki/Streaming_SIMD_Extensions">SSE</a>,
<a href="http://en.wikipedia.org/wiki/SSE2">SSE2</a>,
<a href="http://en.wikipedia.org/wiki/SSE3">SSE3</a>, and
<a href="http://en.wikipedia.org/wiki/SSSE3">SSSE3</a> instruction set extensions.
The generated code is an optimization balanced across the top Intel 32-bit
CPUs.</p>
<p> For more information on compiler flags, particularly related to performance optimization,
refer to <a href="http://software.intel.com/blogs/2012/09/26/gcc-x86-performance-hints">GCC
x86 performance hints</a>.</p>
</li>
<li>Use of the standard Linux x86 32-bit calling convention, as opposed to the one for SVR. For
more information, see section 6, "Register Usage", of
<a href="http://www.agner.org/optimize/calling_conventions.pdf">Calling conventions for different
C++ compilers and operating systems</a>.</li>
</ul>

<p>The ABI does not include any other optional IA-32 instruction set
extensions, such as:</p>
<ul>
<li>MOVBE</li>
<li>Any variant of SSE4.</li>
</ul>
<p>You can still use these extensions, as long as you use runtime feature-probing to
enable them, and provide fallbacks for devices that do not support them.</p>
<p>The NDK toolchain assumes 16-byte stack alignment before a function call. The default tools and
options enforce this rule. If you are writing assembly code, you must make sure to maintain stack
alignment, and ensure that other compilers also obey this rule.</p>

<p>Refer to the following documents for more details:</p>
<ul>
<li>
<a href="https://gcc.gnu.org/onlinedocs/gcc-4.9.2/gcc/i386-and-x86-64-Options.html">
GCC online documentation: Intel 386 and AMD x86-64 Options</a></li>
<li><a href="http://www.agner.org/optimize/calling_conventions.pdf">Calling
conventions for different C++ compilers and operating systems</a></li>
<li><a
href="http://www.intel.com/content/dam/www/public/us/en/documents/manuals/64-ia-32-architectures-software-developer-instruction-set-reference-manual-325383.pdf"
>Intel IA-32 Intel Architecture Software Developer's Manual, Volume 2:
Instruction Set Reference</a></li>
<li><a
href="http://www.intel.com/content/dam/www/public/us/en/documents/manuals/64-ia-32-architectures-software-developer-system-programming-manual-325384.pdf">Intel
IA-32 Intel Architecture Software Developer's Manual, Volume 3: System
Programming Guide</a></li>
<li><a href="http://www.sco.com/developers/devspecs/abi386-4.pdf">System V Application Binary
Interface: Intel386 Processor Architecture Supplement</a></li>
</ul>

<h3 id="86-64">x86_64</h3>
<p>This ABI is for CPUs supporting the instruction set commonly referred to as
"x86-64." It supports instructions that GCC typically generates with the following
compiler flags:</p>
<pre class="no-pretty-print">
-march=x86-64 -msse4.2 -mpopcnt -m64 -mtune=intel
</pre>

<p>These flags target the x86-64 instruction set, according to the GCC
documentation. along with the
<a href="http://en.wikipedia.org/wiki/MMX_%28instruction_set%29">MMX</a>,
<a href="http://en.wikipedia.org/wiki/Streaming_SIMD_Extensions">SSE</a>,
<a href="http://en.wikipedia.org/wiki/SSE2">SSE2</a>,
<a href="http://en.wikipedia.org/wiki/SSE3">SSE3</a>,
<a href="http://en.wikipedia.org/wiki/SSSE3">SSSE3</a>,
<a href="http://en.wikipedia.org/wiki/SSE4#SSE4.1">SSE4.1</a>,
<a href="http://en.wikipedia.org/wiki/SSE4#SSE4.2">SSE4.2</a>, and
<a href="https://software.intel.com/en-us/node/512035">POPCNT</a>
instruction-set extensions. The generated code is an optimization balanced
across the top Intel 64-bit CPUs.</p>

<p> For more information on compiler flags, particularly related to performance optimization,
refer to <a href="http://software.intel.com/blogs/2012/09/26/gcc-x86-performance-hints">GCC
x86 Performance</a>.</p>

<p>This ABI does not include any other optional x86-64 instruction set
extensions, such as:</p>

<ul>
<li>MOVBE</li>
<li>SHA</li>
<li>AVX</li>
<li>AVX2</li>
</ul>

<p>You can still use these extensions, as long as you use runtime feature probing to
enable them, and provide fallbacks for devices that do not support them.</p>
<p>Refer to the following documents for more details:</p>

<ul>
<li><a href="http://www.agner.org/optimize/calling_conventions.pdf">Calling conventions for
different C++ compilers and operating systems</a></li>
<li>
<a href="http://www.intel.com/content/www/us/en/processors/architectures-software-developer-manuals.html?iid=tech_vt_tech+64-32_manuals">
Intel64 and IA-32 Architectures Software Developer's Manual, Volume 2: Instruction Set
Reference</a></li>
<li>
<a href="http://www.intel.com/content/www/us/en/processors/architectures-software-developer-manuals.html?iid=tech_vt_tech+64-32_manuals">
Intel64 and IA-32 Intel Architecture Software Developer's Manual Volume 3: System Programming</a>
</li>
</ul>

<h3 id="mips">mips</h3>
<p>This ABI is for MIPS-based CPUs that support at least the MIPS32r1 instruction set. It includes
the following features:</p>

<ul>
<li>MIPS32 revision 1 ISA</li>
<li>Little-endian</li>
<li>O32</li>
<li>Hard-float</li>
<li>No DSP application-specific extensions</li>
</ul>

<p>For more information, please refer to the following documentation:</p>

<ul>
<li>Architecture for Programmers ("MIPSARCH")</li>
<li><a href="https://refspecs.linuxbase.org/elf/gabi4+/contents.html">ELF
System V Application Binary Interface</a></li>
<li><a href="http://sourcery.mentor.com/public/cxx-abi/abi.html">Itanium/Generic C++
ABI</a></li>
</ul>

<p>The MIPS-specific documentation is available
<a href="http://www.imgtec.com/mips/architectures/mips32.asp">here</a>, with
further information
<a href="https://sourcery.mentor.com/sgpp/lite/mips/portal/target_arch?@action=faq&amp;target_arch=MIPS">here</a>.</p>
</li>
</ul>

<h3 id="mips64">mips64</h3>
<p>This ABI is for MIPS64 R6. For more information, see
<a href="http://www.imgtec.com/mips/architectures/mips64.asp">MIPS64 Architecture</a>.</p>

<h2 id="gc">Generating Code for a Specific ABI</h2>
<p>By default, the NDK generates machine code for the armeabi ABI. You can
generate ARMv7-a-compatible machine code, instead, by adding the following line
to your <a href="{@docRoot}ndk/guides/application_mk.html">{@code Application.mk}</a> file.</p>
<pre class="no-pretty-print">
APP_ABI := armeabi-v7a
</pre>

<p>To build machine code for two or more distinct ABIs, using spaces as delimiters. For
example:</p>

<pre class="no-pretty-print">
APP_ABI := armeabi armeabi-v7a
</pre>

<p>This setting tells the NDK to build two versions of your machine code: one
for each ABI listed on this line. For more information on the values you can specify for the
{@code APP_ABI} variable, see <a href="{@docRoot}ndk/guides/android_mk.html">Android.mk</a>.
</p>

<p>When you build multiple machine-code versions, the build system copies the libraries to your
application project path, and ultimately packages them into your APK, so creating
a <a href="http://en.wikipedia.org/wiki/Fat_binary"><i>fat binary</i></a>. A fat binary
is larger than one containing only the machine code for a single system; the tradeoff is
gaining wider compatibility, but at the expense of a larger APK.</p>

<p>At installation time, the package manager unpacks only the most appropriate
machine code for the target device. For details, see <a href="#aen">Automatic
extraction of native code at install time</a>.</p>


<h2 id="am">ABI Management on the Android Platform</h2>
<p>This section provides details about how the Android platform manages native
code in APKs.</p>

<h3>Native code in app packages</h3>
<p>Both the Play Store and Package Manager expect to find NDK-generated
libraries on filepaths inside the APK matching the following pattern:</p>

<pre class="no-pretty-print">
/lib/&lt;abi&gt;/lib&lt;name&gt;.so
</pre>

<p>Here, {@code &lt;abi&gt;} is one of the ABI names listed under <a href="#sa">Supported ABIs</a>,
and {@code &lt;name&gt;} is the name of the library as you defined it for the {@code LOCAL_MODULE}
variable in the <a href="{@docRoot}ndk/guides/android_mk.html">{@code Android.mk}</a> file. Since
APK files are just zip files, it is trivial to open them and confirm that the shared native
libraries are where they belong.</p>

<p>If the system does not find the native shared libraries where it expects them, it cannot use
them. In such a case, the app itself has to copy the libraries over, and then
perform <code>dlopen()</code>.</p>

<p>In a fat binary, each library resides under a directory whose name matches a corresponding ABI.
For example, a fat binary may contain:</p>

<pre class="no-pretty-print">
/lib/armeabi/libfoo.so
/lib/armeabi-v7a/libfoo.so
/lib/arm64-v8a/libfoo.so
/lib/x86/libfoo.so
/lib/x86_64/libfoo.so
/lib/mips/libfoo.so
/lib/mips64/libfoo.so
</pre>

<p class="note"><strong>Note:</strong> ARMv7-based Android devices running 4.0.3 or earlier
install native libraries from the {@code armeabi} directory instead of the {@code armeabi-v7a}
directory if both directories exist. This is because {@code /lib/armeabi/} comes after
{@code /lib/armeabi-v7a/} in the APK. This issue is fixed from 4.0.4.</p>

<h3>Android Platform ABI support</h3>
<p>The Android system knows at runtime which ABI(s) it supports, because build-specific system
properties indicate:</p>

<ul>
<li>The primary ABI for the device, corresponding to the machine code used in
the system image itself.</li>
<li>An optional, secondary ABI, corresponding to another ABI that the system image also supports.
</li>
</ul>

<p>This mechanism ensures that the system extracts the best machine code from
the package at installation time.</p>

<p>For best performance, you should compile directly for the primary ABI. For example, a
typical ARMv5TE-based device would only define the primary ABI: {@code armeabi}. By contrast, a
typical, ARMv7-based device would define the primary ABI as {@code armeabi-v7a} and the secondary
one as {@code armeabi}, since it can run application native binaries generated for each of them.</p>

<p>Many x86-based devices can also run {@code armeabi-v7a} and {@code armeabi} NDK binaries. For
such devices, the primary ABI would be {@code x86}, and the second one, {@code armeabi-v7a}.</p>

<p>A typical MIPS-based device only defines a primary abi: {@code mips}.</p>

<h3 id="aen">Automatic extraction of native code at install time</h3>

<p>When installing an application, the package manager service scans the APK, and looks for any
shared libraries of the form:</p>

<pre class="no-pretty-print">
lib/&lt;primary-abi&gt;/lib&lt;name&gt;.so
</pre>

<p>If none is found, and you have defined a secondary ABI, the service scans for shared libraries of
the form:</p>

<pre class="no-pretty-print">
lib/&lt;secondary-abi&gt;/lib&lt;name&gt;.so
</pre>

<p>When it finds the libraries that it's looking for, the package manager
copies them to <code>/lib/lib&lt;name&gt;.so</code>, under the application's
{@code data} directory ({@code data/data/&lt;package_name&gt;/lib/}).</p>

<p>If there is no shared-object file at all, the application builds and installs, but crashes at
runtime.</p>