droidcon2013 ndk cpu_architecture_optimization_weggerle_intel

Optimizing NDK projects for multiple CPU architectures

Alexander Weggerle Technical Consultant Engineer

Intel Software and Services Group

Copyright© 2012, Intel Corporation. All rights reserved. *Other brands and names are the property of their respective owners.

Agenda

• Compatibility

• Compiler options

• Code paths

• Differences between ARM and x86

• Publishing


Introduction

• Performance critical apps are popular

• Games

• Real time multimedia

• Augmented reality

• Users are sensitive

• Don’t accept lags

• Fluid animations

• Minimal load time

Optimization is important


Compatibility

• You already done for Java & HTML

• NDK based apps usually just needs a recompilation

“I want my app to run on all architectures”

APP_ABI := all

Application.mk


Agenda

• Compatibility


• Code paths


• Publishing


Target compiler options

• NDK-build system Android.mk file evaluated for each architecture

• Variable TARGET_ARCH_ABI describes actual architecture

TARGET_ARCH_ABI

x86

armeabi

armeabi-v7a

mips

ifeq ($(TARGET_ARCH_ABI),x86)

LOCAL_CFLAGS := -mtune=atom -mssse3

endif

ifeq ($(TARGET_ARCH_ABI),armeabi-v7a)

LOCAL_CFLAGS := -march=armv7-a

Endif

Android.mk


Recommended Compiler options (x86)

• -mtune=atom

• Out of Order scheduling

• -march=atom

• movbe instruction

• Code is only guaranteed to run on Atom.

– Might not run on emulator or other CPUs

• -ansi-alias / -restrict / -no-prec-div


Agenda

• Compatibility


• Code paths


• Publishing


Multiple Code Paths

• Different reasons for multiple code paths

• Optimizing for multiple architectures

• Single Instruction Multiple Data (SIMD)

• Assembly kernels

• Memory alignment

• Instruction set support


Multiple Code Paths Compiler macros

• Directly inside source code

• No runtime overhead

• Works with all build configurations

#ifdef __i386__

strlcat(buf, "__i386__", sizeof(buf));

#endif

#ifdef __arm__

strlcat(buf, "__arm__", sizeof(buf));

#endif

#ifdef _MIPS_ARCH

strlcat(buf, "_MIPS_ARCH", sizeof(buf));

#endif

source.c


Multiple Code Paths Make system

• Make system will select source file depending on architecture

• No runtime overhead

arch_x86.cpp

const char *getTarget ();

arch_arm.cpp


arch_default.cpp


arch.h


cpufeaturestest.cpp

printf(“%s”, getTarget());


Multiple Code Paths Cpufeatures API

• Checks for architecture and special CPU features

• Uses compiler macros + runtime detection internally

• Small runtime overhead


Agenda

• Compatibility


• Code paths


• Publishing


Porting Native (C/C++) Android* Apps to x86

Native Apps

• Optimized NDK for Intel Atom based devices available on http://developer.android.com/sdk/ndk/index.html since July’11.

• For most apps, changing the make file and a recompile should be sufficient to port to Intel Atom devices.

• If ARM-specific features are used, Intel SSE equivalents should be added (build flag)

• Developer recompiles, re-packages and publishes.

http://software.intel.com/en-us/articles/ndk-android-application-porting-methodologies/

http://developer.android.com/sdk/ndk/index.html

http://developer.android.com/sdk/ndk/index.html


ARM* NEON to Intel® SSE

• Single Instruction Multiple Data (SIMD)

• Most ARM NEON functions have a 1:1 equivalent in Intel® SSE

• Intel provides C++ header file to do the mapping

// VADD.I8 d0,d0,d0

int8x8_t vadd_s8(int8x8_t a, int8x8_t b);

#ifdef USE_MMX

#define vadd_s8 _mm_add_pi8 //MMX

#else

#define vadd_s8 _mm_add_epi8

#endif

neonvssse.h


Differences between ARM and x86 Alignment

• ARM uses aligned, x86 uses packed memory

• Compiler parameter helps to workaround

struct TestStruct

{

int mVar1;

long long mVar2;

int mVar3;

};

ARM

x86

-malign-double


Agenda

• Compatibility


• Code paths


• Publishing


Market Fat binary

• Include all binaries into one apk

• Device removes incompatible libs at installation

Source

libs/armeabi-v7a

libs/x86

libs/armeabi

ndk-build apkbuilder

…


Market Multiple APKs

• Support for different …

• Texture formats

• Screen sizes and densities

• Platform versions

• CPU architectures

• Saves bandwidth and space while installation


Conclusion & Call to action

• NDK provides plenty of mechanisms to differentiate between architectures

• Recompile your NDK based app with APP_ABI := all


INFORMATION IN THIS DOCUMENT IS PROVIDED “AS IS”. NO LICENSE, EXPRESS OR IMPLIED, BY ESTOPPEL OR OTHERWISE, TO ANY INTELLECTUAL PROPERTY RIGHTS IS GRANTED BY THIS DOCUMENT. INTEL ASSUMES NO LIABILITY WHATSOEVER AND INTEL DISCLAIMS ANY EXPRESS OR IMPLIED WARRANTY, RELATING TO THIS INFORMATION INCLUDING LIABILITY OR WARRANTIES RELATING TO FITNESS FOR A PARTICULAR PURPOSE, MERCHANTABILITY, OR INFRINGEMENT OF ANY PATENT, COPYRIGHT OR OTHER INTELLECTUAL PROPERTY RIGHT. Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSmark and MobileMark, are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. Copyright © , Intel Corporation. All rights reserved. Intel, the Intel logo, Xeon, Core, VTune, and Cilk are trademarks of Intel Corporation in the U.S. and other countries.

Optimization Notice

Intel’s compilers may or may not optimize to the same degree for non-Intel microprocessors for optimizations that are not unique to Intel microprocessors. These optimizations include SSE2, SSE3, and SSSE3 instruction sets and other optimizations. Intel does not guarantee the availability, functionality, or effectiveness of any optimization on microprocessors not manufactured by Intel. Microprocessor-dependent optimizations in this product are intended for use with Intel microprocessors. Certain optimizations not specific to Intel microarchitecture are reserved for Intel microprocessors. Please refer to the applicable product User and Reference Guides for more information regarding the specific instruction sets covered by this notice.

Notice revision #20110804

Legal Disclaimer & Optimization Notice


22

17.04.2013

Intel Confidential

droidcon2013 ndk cpu_architecture_optimization_weggerle_intel

Technology