1 Android NDK x86 (a.k.a. IA-32) instruction set support 2 === 3 4 Introduction: 5 ------------- 6 7 Android NDK r6 added support for the '`x86`' ABI, that allows native code to 8 run on Android-based devices running on CPUs supporting the IA-32 instruction 9 set. 10 11 The Android x86 ABI itself is fully specified in docs/CPU-ARCH-ABIS.html. 12 13 Overview: 14 --------- 15 16 Generating x86 machine code is simple: just add 'x86' to your APP_ABI 17 definition in your Application.mk file, for example: 18 19 APP_ABI := armeabi armeabi-v7a x86 20 21 Alternatively, since NDK r7, you can use: 22 23 APP_ABI := all 24 25 will generate machine code for all supported ABIs with this NDK. Doing so 26 will ensure that your application package contains libraries for all target 27 ABIs. Note that this has an impact on package size, since each ABI will 28 correspond to its own set of native libraries built from the same sources. 29 30 The default ABI is still '`armeabi`', if unspecified in your project. 31 32 As you would expect, generated libraries will go into `$PROJECT/libs/x86/`, and 33 will be embedded into your .apk under `/lib/x86/`. 34 35 And just like other ABIs, the Android package manager will extract these 36 libraries on a *compatible* x86-based device automatically at install time, 37 to put them under <dataPath>/lib, where <dataPath> is the 38 application's private data directory. 39 40 Similarly, the Google Play server is capable of filtering applications 41 based on the native libraries they embed and your device's target CPU. 42 43 Debugging with ndk-gdb should work exactly as described under docs/NDK-GDB.html. 44 45 ARM NEON intrinsics support: 46 ---------------------------- 47 48 The solution is shaped as C/C++ language header with the same name as standard 49 arm neon intrinsics header "arm_neon.h" which is also available in all NDK x86 50 toolchains. It translates neon intrinsics to native x86 SSE ones. 51 52 By default SSE up to SSE3 is used for porting ARM NEON to Intel SSE. 53 54 Current solution covers by default ~41% NEON functions (889 of total 1884) and 55 47% when -mssse3 is enabled. It is highly recommended to use the -mssse3 compiler 56 flag for more coverage and performance. 57 58 If currently provided coverage is not enough to port application please look 59 into next version preview (up to 98% NEON instrinsics covered) at 60 61 > http://software.intel.com/en-us/blogs/2012/12/12/from-arm-neon-to-intel-mmxsse-automatic-porting-solution-tips-and-tricks 62 63 The solution 64 65 - Redefines ARM NEON 128 bit vectors as the corresponding x86 SIMD data. 66 67 - Redefines some functions from ARM NEON to Intel SSE if 1:1 correspondence 68 exists. 69 70 - Implements some ARM NEON functions using Intel SIMD if the performance 71 effective implementation is possible. 72 73 - Implements some of the remaining NEON functions using the serial solution 74 and issuing the corresponding "low performance" compiler warning. 75 76 ### Performance: 77 78 For the major number of cases it is expected to obtain the similar to ARM NEON 79 native perfomance gain for vectorized vs. serial code. 80 81 ### Porting considerations and best known methods are: 82 83 - Use 16-byte data alignment for faster load and store 84 85 - Avoid NEON functions working with constants. It produces performance 86 penalty for constants load\propagation. 87 If constants usage is necessary try to move constants initialization out of 88 hotspot loops and if applicable replace it with logical and compare 89 operations. 90 91 - Try to avoid functions marked as "serialy implemented" because they need to 92 store data from registers to memory, process them serialy and load them again. 93 Probably you could change the data type or algorithm used to make the whole 94 port vectorized not a serial one. 95 96 To learn more about it, see 97 98 > http://software.intel.com/en-us/blogs/2012/12/12/from-arm-neon-to-intel-mmxsse-automatic-porting-solution-tips-and-tricks 99 100 ### Sample code: 101 102 In your project add 'x86' to APP_ABI definition and make sure "arm_neon.h" 103 header is included. 104 Your code will be ported to x86 without any other changes necessary. 105 106 107 Standalone-toolchain: 108 --------------------- 109 110 It is possible to use the x86 toolchain with NDK r6 in stand-alone mode. 111 See docs/STANDALONE-TOOLCHAIN.html for more details. Briefly speaking, 112 it is now possible to run: 113 114 $NDK/build/tools/make-standalone-toolchain.sh --arch=x86 --install-dir=<path> 115 116 The toolchain binaries have the `i686-linux-android- prefix`. 117 118 119 Compatibility: 120 -------------- 121 122 The minimal native API level provided by official Android x86 platform builds 123 is 9, which corresponds to all the native APIs provided by Android 2.3, i.e. 124 Gingerbread (note also that no new native APIs were introduced by Honeycomb). 125 126 You won't have to change anything to your project files if you target an older 127 API level: the NDK build script will automatically select the right set of 128 native platform headers/libraries for you. 129
