今天研究下cpufeatures库,它是一个轻量级的展示CPU特性的模块,调用者在运行时候可以根据这个库提供的简单接口获取到目标机的CPU架构和特性。

一、代码路径:

$NDK_ROOT/sources/android/cpufeatures

二、主要功能

1. 获取目标机CPU架构,涵盖了android可能运行所有CPU架构,设置ABI的时候可以参考

extern AndroidCpuFamily android_getCpuFamily(void);
typedef enum {
ANDROID_CPU_FAMILY_UNKNOWN = 0,
ANDROID_CPU_FAMILY_ARM,
ANDROID_CPU_FAMILY_X86,
ANDROID_CPU_FAMILY_MIPS,
ANDROID_CPU_FAMILY_ARM64,
ANDROID_CPU_FAMILY_X86_64,
ANDROID_CPU_FAMILY_MIPS64,
ANDROID_CPU_FAMILY_MAX /* do not remove */
} AndroidCpuFamily;

2. 获取CPU性能,不同的CPU架构有不同的性能,例如arm支持的指令集有VFPV2、ARMv7-A、NEON

extern uint64_t android_getCpuFeatures(void);  // 注意:此方法32bit和64bit arm内核的返回不对等
enum {
ANDROID_CPU_ARM_FEATURE_ARMv7 = (1 << 0),
ANDROID_CPU_ARM_FEATURE_VFPv3 = (1 << 1),
ANDROID_CPU_ARM_FEATURE_NEON = (1 << 2),
ANDROID_CPU_ARM_FEATURE_LDREX_STREX = (1 << 3),
ANDROID_CPU_ARM_FEATURE_VFPv2 = (1 << 4),
ANDROID_CPU_ARM_FEATURE_VFP_D32 = (1 << 5),
ANDROID_CPU_ARM_FEATURE_VFP_FP16 = (1 << 6),
ANDROID_CPU_ARM_FEATURE_VFP_FMA = (1 << 7),
ANDROID_CPU_ARM_FEATURE_NEON_FMA = (1 << 8),
ANDROID_CPU_ARM_FEATURE_IDIV_ARM = (1 << 9),
ANDROID_CPU_ARM_FEATURE_IDIV_THUMB2 = (1 << 10),
ANDROID_CPU_ARM_FEATURE_iWMMXt = (1 << 11),
ANDROID_CPU_ARM_FEATURE_AES = (1 << 12),
ANDROID_CPU_ARM_FEATURE_PMULL = (1 << 13),
ANDROID_CPU_ARM_FEATURE_SHA1 = (1 << 14),
ANDROID_CPU_ARM_FEATURE_SHA2 = (1 << 15),
ANDROID_CPU_ARM_FEATURE_CRC32 = (1 << 16),
};

要支持不同的指令集,要添加对应的编译选项,详细见附录

-mfpu=neon   // 支持neon指令集

3. 获取CPU个数

extern int android_getCpuCount(void);

4. 设置CPU个数,沙盒模拟测试

extern int android_setCpu(int cpu_count, uint64_t cpu_features);
extern uint32_t android_getCpuIdArm(void);
extern int android_setCpuArm(int cpu_count, uint64_t cpu_features, uint32_t cpu_id);

三、 使用举例

1. 以下展示了一个筛选armv7并支持neon指令集的代码

#include <cpu-features.h>

AndroidCUPFamily family = android_getCpuFamily();
if (family != ANDROID_CPU_FAMILY_ARM)
{
printf("Not an ARM CPU !\n");
return;
} uint64_t features = android_getCpuFeatures();
if ((features & ANDROID_CPU_ARM_FEATURE_ARMv7) == 0)
{
printf("Not an ARMv7 CPU !\n")
return;
}

if ((features & ANDROID_CPU_ARM_FEATURE_NEON) == 0)
{
printf("CPU doesn't support NEON !\n");
return;
}

2. 工程引入

# build cpufeatures as a static lib
add_library(cpufeatures STATIC ${ANDROID_NDK}/sources/android/cpufeatures/cpu-features.c)
# include directories
target_include_directories(hello-neon PRIVATE ${ANDROID_NDK}/sources/android/cpufeatures)
# link
target_link_libraries(hello-neon cpufeatures)

附录:

/*
* Copyright (C) 2010 The Android Open Source Project
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* * Redistributions of source code must retain the above copyright
* notice, this list of conditions and the following disclaimer.
* * Redistributions in binary form must reproduce the above copyright
* notice, this list of conditions and the following disclaimer in
* the documentation and/or other materials provided with the
* distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
* LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
* FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
* COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
* INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
* BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS
* OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
* AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
* OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
* OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*/
#ifndef CPU_FEATURES_H
#define CPU_FEATURES_H #include <sys/cdefs.h>
#include <stdint.h> __BEGIN_DECLS /* A list of valid values returned by android_getCpuFamily().
* They describe the CPU Architecture of the current process.
*/
typedef enum {
ANDROID_CPU_FAMILY_UNKNOWN = ,
ANDROID_CPU_FAMILY_ARM,
ANDROID_CPU_FAMILY_X86,
ANDROID_CPU_FAMILY_MIPS,
ANDROID_CPU_FAMILY_ARM64,
ANDROID_CPU_FAMILY_X86_64,
ANDROID_CPU_FAMILY_MIPS64, ANDROID_CPU_FAMILY_MAX /* do not remove */ } AndroidCpuFamily; /* Return the CPU family of the current process.
*
* Note that this matches the bitness of the current process. I.e. when
* running a 32-bit binary on a 64-bit capable CPU, this will return the
* 32-bit CPU family value.
*/
extern AndroidCpuFamily android_getCpuFamily(void); /* Return a bitmap describing a set of optional CPU features that are
* supported by the current device's CPU. The exact bit-flags returned
* depend on the value returned by android_getCpuFamily(). See the
* documentation for the ANDROID_CPU_*_FEATURE_* flags below for details.
*/
extern uint64_t android_getCpuFeatures(void); /* The list of feature flags for ANDROID_CPU_FAMILY_ARM that can be
* recognized by the library (see note below for 64-bit ARM). Value details
* are:
*
* VFPv2:
* CPU supports the VFPv2 instruction set. Many, but not all, ARMv6 CPUs
* support these instructions. VFPv2 is a subset of VFPv3 so this will
* be set whenever VFPv3 is set too.
*
* ARMv7:
* CPU supports the ARMv7-A basic instruction set.
* This feature is mandated by the 'armeabi-v7a' ABI.
*
* VFPv3:
* CPU supports the VFPv3-D16 instruction set, providing hardware FPU
* support for single and double precision floating point registers.
* Note that only 16 FPU registers are available by default, unless
* the D32 bit is set too. This feature is also mandated by the
* 'armeabi-v7a' ABI.
*
* VFP_D32:
* CPU VFP optional extension that provides 32 FPU registers,
* instead of 16. Note that ARM mandates this feature is the 'NEON'
* feature is implemented by the CPU.
*
* NEON:
* CPU FPU supports "ARM Advanced SIMD" instructions, also known as
* NEON. Note that this mandates the VFP_D32 feature as well, per the
* ARM Architecture specification.
*
* VFP_FP16:
* Half-width floating precision VFP extension. If set, the CPU
* supports instructions to perform floating-point operations on
* 16-bit registers. This is part of the VFPv4 specification, but
* not mandated by any Android ABI.
*
* VFP_FMA:
* Fused multiply-accumulate VFP instructions extension. Also part of
* the VFPv4 specification, but not mandated by any Android ABI.
*
* NEON_FMA:
* Fused multiply-accumulate NEON instructions extension. Optional
* extension from the VFPv4 specification, but not mandated by any
* Android ABI.
*
* IDIV_ARM:
* Integer division available in ARM mode. Only available
* on recent CPUs (e.g. Cortex-A15).
*
* IDIV_THUMB2:
* Integer division available in Thumb-2 mode. Only available
* on recent CPUs (e.g. Cortex-A15).
*
* iWMMXt:
* Optional extension that adds MMX registers and operations to an
* ARM CPU. This is only available on a few XScale-based CPU designs
* sold by Marvell. Pretty rare in practice.
*
* AES:
* CPU supports AES instructions. These instructions are only
* available for 32-bit applications running on ARMv8 CPU.
*
* CRC32:
* CPU supports CRC32 instructions. These instructions are only
* available for 32-bit applications running on ARMv8 CPU.
*
* SHA2:
* CPU supports SHA2 instructions. These instructions are only
* available for 32-bit applications running on ARMv8 CPU.
*
* SHA1:
* CPU supports SHA1 instructions. These instructions are only
* available for 32-bit applications running on ARMv8 CPU.
*
* PMULL:
* CPU supports 64-bit PMULL and PMULL2 instructions. These
* instructions are only available for 32-bit applications
* running on ARMv8 CPU.
*
* If you want to tell the compiler to generate code that targets one of
* the feature set above, you should probably use one of the following
* flags (for more details, see technical note at the end of this file):
*
* -mfpu=vfp
* -mfpu=vfpv2
* These are equivalent and tell GCC to use VFPv2 instructions for
* floating-point operations. Use this if you want your code to
* run on *some* ARMv6 devices, and any ARMv7-A device supported
* by Android.
*
* Generated code requires VFPv2 feature.
*
* -mfpu=vfpv3-d16
* Tell GCC to use VFPv3 instructions (using only 16 FPU registers).
* This should be generic code that runs on any CPU that supports the
* 'armeabi-v7a' Android ABI. Note that no ARMv6 CPU supports this.
*
* Generated code requires VFPv3 feature.
*
* -mfpu=vfpv3
* Tell GCC to use VFPv3 instructions with 32 FPU registers.
* Generated code requires VFPv3|VFP_D32 features.
*
* -mfpu=neon
* Tell GCC to use VFPv3 instructions with 32 FPU registers, and
* also support NEON intrinsics (see <arm_neon.h>).
* Generated code requires VFPv3|VFP_D32|NEON features.
*
* -mfpu=vfpv4-d16
* Generated code requires VFPv3|VFP_FP16|VFP_FMA features.
*
* -mfpu=vfpv4
* Generated code requires VFPv3|VFP_FP16|VFP_FMA|VFP_D32 features.
*
* -mfpu=neon-vfpv4
* Generated code requires VFPv3|VFP_FP16|VFP_FMA|VFP_D32|NEON|NEON_FMA
* features.
*
* -mcpu=cortex-a7
* -mcpu=cortex-a15
* Generated code requires VFPv3|VFP_FP16|VFP_FMA|VFP_D32|
* NEON|NEON_FMA|IDIV_ARM|IDIV_THUMB2
* This flag implies -mfpu=neon-vfpv4.
*
* -mcpu=iwmmxt
* Allows the use of iWMMXt instrinsics with GCC.
*
* IMPORTANT NOTE: These flags should only be tested when
* android_getCpuFamily() returns ANDROID_CPU_FAMILY_ARM, i.e. this is a
* 32-bit process.
*
* When running a 64-bit ARM process on an ARMv8 CPU,
* android_getCpuFeatures() will return a different set of bitflags
*/
enum {
ANDROID_CPU_ARM_FEATURE_ARMv7 = ( << ),
ANDROID_CPU_ARM_FEATURE_VFPv3 = ( << ),
ANDROID_CPU_ARM_FEATURE_NEON = ( << ),
ANDROID_CPU_ARM_FEATURE_LDREX_STREX = ( << ),
ANDROID_CPU_ARM_FEATURE_VFPv2 = ( << ),
ANDROID_CPU_ARM_FEATURE_VFP_D32 = ( << ),
ANDROID_CPU_ARM_FEATURE_VFP_FP16 = ( << ),
ANDROID_CPU_ARM_FEATURE_VFP_FMA = ( << ),
ANDROID_CPU_ARM_FEATURE_NEON_FMA = ( << ),
ANDROID_CPU_ARM_FEATURE_IDIV_ARM = ( << ),
ANDROID_CPU_ARM_FEATURE_IDIV_THUMB2 = ( << ),
ANDROID_CPU_ARM_FEATURE_iWMMXt = ( << ),
ANDROID_CPU_ARM_FEATURE_AES = ( << ),
ANDROID_CPU_ARM_FEATURE_PMULL = ( << ),
ANDROID_CPU_ARM_FEATURE_SHA1 = ( << ),
ANDROID_CPU_ARM_FEATURE_SHA2 = ( << ),
ANDROID_CPU_ARM_FEATURE_CRC32 = ( << ),
}; /* The bit flags corresponding to the output of android_getCpuFeatures()
* when android_getCpuFamily() returns ANDROID_CPU_FAMILY_ARM64. Value details
* are:
*
* FP:
* CPU has Floating-point unit.
*
* ASIMD:
* CPU has Advanced SIMD unit.
*
* AES:
* CPU supports AES instructions.
*
* CRC32:
* CPU supports CRC32 instructions.
*
* SHA2:
* CPU supports SHA2 instructions.
*
* SHA1:
* CPU supports SHA1 instructions.
*
* PMULL:
* CPU supports 64-bit PMULL and PMULL2 instructions.
*/
enum {
ANDROID_CPU_ARM64_FEATURE_FP = ( << ),
ANDROID_CPU_ARM64_FEATURE_ASIMD = ( << ),
ANDROID_CPU_ARM64_FEATURE_AES = ( << ),
ANDROID_CPU_ARM64_FEATURE_PMULL = ( << ),
ANDROID_CPU_ARM64_FEATURE_SHA1 = ( << ),
ANDROID_CPU_ARM64_FEATURE_SHA2 = ( << ),
ANDROID_CPU_ARM64_FEATURE_CRC32 = ( << ),
}; /* The bit flags corresponding to the output of android_getCpuFeatures()
* when android_getCpuFamily() returns ANDROID_CPU_FAMILY_X86 or
* ANDROID_CPU_FAMILY_X86_64.
*/
enum {
ANDROID_CPU_X86_FEATURE_SSSE3 = ( << ),
ANDROID_CPU_X86_FEATURE_POPCNT = ( << ),
ANDROID_CPU_X86_FEATURE_MOVBE = ( << ),
ANDROID_CPU_X86_FEATURE_SSE4_1 = ( << ),
ANDROID_CPU_X86_FEATURE_SSE4_2 = ( << ),
ANDROID_CPU_X86_FEATURE_AES_NI = ( << ),
ANDROID_CPU_X86_FEATURE_AVX = ( << ),
ANDROID_CPU_X86_FEATURE_RDRAND = ( << ),
ANDROID_CPU_X86_FEATURE_AVX2 = ( << ),
ANDROID_CPU_X86_FEATURE_SHA_NI = ( << ),
}; /* The bit flags corresponding to the output of android_getCpuFeatures()
* when android_getCpuFamily() returns ANDROID_CPU_FAMILY_MIPS
* or ANDROID_CPU_FAMILY_MIPS64. Values are:
*
* R6:
* CPU executes MIPS Release 6 instructions natively, and
* supports obsoleted R1..R5 instructions only via kernel traps.
*
* MSA:
* CPU supports Mips SIMD Architecture instructions.
*/
enum {
ANDROID_CPU_MIPS_FEATURE_R6 = ( << ),
ANDROID_CPU_MIPS_FEATURE_MSA = ( << ),
}; /* Return the number of CPU cores detected on this device. */
extern int android_getCpuCount(void); /* The following is used to force the CPU count and features
* mask in sandboxed processes. Under 4.1 and higher, these processes
* cannot access /proc, which is the only way to get information from
* the kernel about the current hardware (at least on ARM).
*
* It _must_ be called only once, and before any android_getCpuXXX
* function, any other case will fail.
*
* This function return 1 on success, and 0 on failure.
*/
extern int android_setCpu(int cpu_count,
uint64_t cpu_features); #ifdef __arm__
/* Retrieve the ARM 32-bit CPUID value from the kernel.
* Note that this cannot work on sandboxed processes under 4.1 and
* higher, unless you called android_setCpuArm() before.
*/
extern uint32_t android_getCpuIdArm(void); /* An ARM-specific variant of android_setCpu() that also allows you
* to set the ARM CPUID field.
*/
extern int android_setCpuArm(int cpu_count,
uint64_t cpu_features,
uint32_t cpu_id);
#endif __END_DECLS #endif /* CPU_FEATURES_H */

NDK开发之<cpu-features.h模块功能>的更多相关文章

  1. [Abp vNext 入坑分享] - 3.简单的用户模块功能开发

    一.简要说明 本篇文章开始进行业务模块的开发模拟,借助user模块来进行业务开发,主要是用户相关的基础操作.主要是先使用Users来体验整个开发的流程.主要是先把一个基础流程跑顺利,在这里我并不会过于 ...

  2. NDK开发之一

    2015.07.22 Wiki_Tree: --NDK开发: --NDK特征: --MK文件编写规则: NDK开发: Ndk-build编译时会生成的两个同名的so库,位于不同的目录/project ...

  3. Android之NDK开发(转)

    Android之NDK开发 一.NDK产生的背景 Android平台从诞生起,就已经支持C.C++开发.众所周知,Android的SDK基于Java实现,这意味着基于Android SDK进行开发的第 ...

  4. 5.7 NDK开发

    JNI开发流程主要分为以下6步: 编写Java源代码 将Java源代码编译成class字节码文件 用javah -jni命令生成.h头文件(-jni参数表示将class中用native声明的函数生成j ...

  5. Android之NDK开发

    转自:http://www.cnblogs.com/devinzhang/archive/2012/02/29/2373729.html 一.NDK产生的背景 Android平台从诞生起,就已经支持C ...

  6. ndk开发教程以及问题解决方案

    一.NDK产生的背景 Android平台从诞生起,就已经支持C.C++开发.众所周知,Android的SDK基于Java实现,这意味着基于Android SDK进行开发的第三方应用都必须使用Java语 ...

  7. 【转】Android之NDK开发

    原文网址:http://www.cnblogs.com/devinzhang/archive/2012/02/29/2373729.html 一.NDK产生的背景 Android平台从诞生起,就已经支 ...

  8. Android NDK开发指南---Application.mk文件和android.mk文件

    https://android.googlesource.com/platform/development/+/donut-release/ndk/docs/OVERVIEW.TXT https:// ...

  9. 【转】Android NDK开发入门实例

    写这个,目的就是记录一下我自己的NDK是怎么入门的.便于以后查看,而不会忘了又用搜索引擎一顿乱搜.然后希望能够帮助刚学的人入门. 先转一段别人说的话:“NDK全称:Native Development ...

随机推荐

  1. 对Java面向对象中多态的理解

    理解的要点:多态意味着父亲的变量可以指向子类对象 面向对象程序设计的三大支柱是封装.继承和多态 封装对外把相应的属性和方法实现的细节进行了隐藏.继承关系使一个子类继承父亲的特征,并且加上了一些新的特征 ...

  2. Spring Cloud Zuul 网关服务的fallback

    当我们的zuul进行路由分发时,如果后端服务没有启动,或者调用超时,这时候我们希望Zuul提供一种降级功能,而不是将异常暴露出来. Spring cloud zuul提供这种降级功能,操作步骤如下: ...

  3. [腾讯 TMQ] 接口测试用例设计

    接口测试 [腾讯 TMQ] 接口测试用例设计 腾讯移动品质中心 · 2018年01月17日 · 最后由 于静 回复于 20 天前 · 21794 次阅读 本帖已被设为精华帖! 目录 作者:刘燕 团队: ...

  4. 史上最难PHPer笔试题,40分就能月薪过万!附答案

    请批判性的学习,欢迎大牛指正错误 1.有关PHP字符串的说法,不对的是:A.如果一个脚本的编码是 ISO-8859-1,则其中的字符串也会被编码为 ISO-8859-1.B.PHP的字符串在内部是字节 ...

  5. Juniper srx新增接口IP,使PC直连srx(转)

    转自:https://www.jianshu.com/p/bc27134bde3d Juniper srx新增接口IP,使PC直连srx 2018.11.19 14:24:15字数 424 概述 需求 ...

  6. spark shc hbase 超时问题 hbase.client.scanner.timeout.period 配置

    异常信息 20/02/27 19:36:21 INFO TaskSetManager: Starting task 17.1 in stage 3.0 (TID 56, 725.slave.adh, ...

  7. 面向对象 part3 构造函数 原型函数

    6.2创建对象 方法:对象字面量  object构造函数 缺点:都是创建单个对象.同一个接口创建多个对象,会产生大量重复代码 6.2.1工厂模式 用函数封装以特定的接口创建对象 function cr ...

  8. ZJNU 2201 - 挖矿谷物语

    在dfs过程中加上栈记录当次dfs走过的路径 如果当次dfs到了一个之前的dfs已经经过的点 又因为只对没有访问过的点开始dfs 所以这种情况就说明接下来不可能返回到当次dfs开始的点 将栈内元素取出 ...

  9. altium designer 画板 电子元器件的名称不能集体修改

    今天在画板子的时候,遇到一个问题,就是发现电子元器件的名字太大了,想把他们集体都改小一点,方便布局,结果修改的时候却发现高度可以改,宽度改不了,(集体改的话,改不了,但是可以单独改,只是单独改,工程量 ...

  10. Vue 项目中应用

    Vue使用 一.vue生命周期 # main.js import Vue from 'vue' import App from './App.vue' import router from './ro ...