源码编译（ACL）
简介
ARM计算库(ACL)是一套计算机视觉和机器学习功能，使用SIMD技术为ARM cpu和gpu优化。 Tengine支持与ACL的OpenCL库集成，通过ARM-Mail GPU对CNN进行推理。
support check:
sudo apt install clinfo
clinfo
结果:
Number of platforms 1
.....
Build
ACL GPU Library
下载 ACL
$ git clone https://github.com/ARM-software/ComputeLibrary.git
$ git checkout v20.05
构建 ACL
$ scons Werror=1 -j4 debug=0 asserts=1 neon=0 opencl=1 embed_kernels=1 os=linux arch=arm64-v8a
下载 Tengine
$ git clone https://github.com/OAID/Tengine.git Tengine
$ cd Tengine
创建依赖文件
$ cd <tengine-lite-root-dir>
$ mkdir -p ./3rdparty/acl/lib
$ mkdir -p ./3rdparty/acl/include
$ cp -rf ComputeLibrary/include/* Tengine/3rdparty/acl/include
$ cp -rf ComputeLibrary/arm_compute Tengine/3rdparty/acl/include
$ cp -rf ComputeLibrary/support Tengine/3rdparty/acl/include
$ cp -rf ComputeLibrary/build/libarm_compute*.so Tengine/3rdparty/acl/lib/
构建选项
$ mkdir build-acl-arm64 && cd build-acl-arm64
$ cmake -DCMAKE_TOOLCHAIN_FILE=../toolchains/aarch64-linux-gnu.toolchain.cmake \
-DTENGINE_ENABLE_ACL=ON ..
$ make -j4
$ make install
示例
依赖库
3rdparty/acl/lib/
├── libarm_compute.so
├── libarm_compute_core.so
└── libarm_compute_graph.so
build-acl-arm64/install/lib/
└── libtengine-lite.so
Set FP16 Inference mode
Enable GPU FP16 mode
/* set runtime options */
struct options opt;
opt.num_thread = num_thread;
opt.cluster = TENGINE_CLUSTER_ALL;
opt.precision = TENGINE_MODE_FP16;
opt.affinity = 0;
结果
[root@localhost tengine-lite]# ./tm_mssd_acl -m mssd.tmfile -i ssd_dog.jpg -t 1 -r 10
start to run register cpu allocator
start to run register acl allocator
tengine-lite library version: 1.0-dev
run into gpu by acl
Repeat 10 times, thread 2, avg time 82.32 ms, max_time 135.70 ms, min_time 74.10 ms
--------------------------------------
detect result num: 3
dog :99.8%
BOX:( 138 , 209 ),( 324 , 541 )
car :99.7%
BOX:( 467 , 72 ),( 687 , 171 )
bicycle :99.6%
BOX:( 106 , 141 ),( 574 , 415 )
======================================
[DETECTED IMAGE SAVED]:
======================================
支持硬件列表
| 芯片厂家 | 型号 |
|---|---|
| arm-mali | T-860、G31、G52 |