1*4882a593Smuzhiyun# 说明 2*4882a593SmuzhiyunAndroid平台有两种方式来调用RKNN API 3*4882a593Smuzhiyun1）应用直接链接librknnrt.so 4*4882a593Smuzhiyun2）应用链接Android平台HIDL实现的librknn_api_android.so 5*4882a593Smuzhiyun对于需要通过CTS/VTS测试的Android设备可以使用基于Android平台HIDL实现的RKNN API。如果不需要通过CTS/VTS测试的设备建议直接链接使用librknnrt.so，对各个接口调用流程的链路更短，可以提供更好的性能。 6*4882a593Smuzhiyun 7*4882a593Smuzhiyun对于使用Android HIDL实现的RKNN API的代码位于RK3566_RK3568/RK3588 Android系统SDK的vendor/rockchip/hardware/interfaces/neuralnetworks目录下。当完成Android系统编译后，将会生成一些NPU相关的库（对于应用只需要链接使用librknn_api_android.so即可） 8*4882a593Smuzhiyun 9*4882a593Smuzhiyun**本示例适用于librknn_api_android.so。** 10*4882a593Smuzhiyun 11*4882a593Smuzhiyun 12*4882a593Smuzhiyun 13*4882a593Smuzhiyun# 编译 14*4882a593Smuzhiyun 15*4882a593Smuzhiyun- 编译librknn_api_android.so 16*4882a593Smuzhiyun 17*4882a593Smuzhiyun 需要先下载RK3566_RK3568/RK3588 Android SDK，在Android SDK根目录执行 18*4882a593Smuzhiyun 19*4882a593Smuzhiyun ``` 20*4882a593Smuzhiyun source build/envsetup.sh 21*4882a593Smuzhiyun lunch your target ##需要根据自己的实际情况进行选择 22*4882a593Smuzhiyun mmm vendor/rockchip/hardware/interfaces/neuralnetworks/ -j16 23*4882a593Smuzhiyun 24*4882a593Smuzhiyun ``` 25*4882a593Smuzhiyun 26*4882a593Smuzhiyun 将生成 27*4882a593Smuzhiyun 28*4882a593Smuzhiyun ``` 29*4882a593Smuzhiyun /vendor/lib/librknn_api_android.so 30*4882a593Smuzhiyun /vendor/lib/librknnhal_bridge.rockchip.so 31*4882a593Smuzhiyun /vendor/lib64/librknn_api_android.so 32*4882a593Smuzhiyun /vendor/lib64/librknnhal_bridge.rockchip.so 33*4882a593Smuzhiyun /vendor/lib64/rockchip.hardware.neuralnetworks@1.0.so 34*4882a593Smuzhiyun /vendor/lib64/rockchip.hardware.neuralnetworks@1.0-adapter-helper.so 35*4882a593Smuzhiyun /vendor/lib64/hw/rockchip.hardware.neuralnetworks@1.0-impl.so 36*4882a593Smuzhiyun /vendor/bin/hw/rockchip.hardware.neuralnetworks@1.0-service 37*4882a593Smuzhiyun ``` 38*4882a593Smuzhiyun 39*4882a593Smuzhiyun 40*4882a593Smuzhiyun- 编译本demo 41*4882a593Smuzhiyun 42*4882a593Smuzhiyun 将$RKNPU2_SDK拷贝到Android SDK根目录，并执行： 43*4882a593Smuzhiyun 44*4882a593Smuzhiyun ``` 45*4882a593Smuzhiyun mmm rknpu2/examples/librknn_api_android_demo 46*4882a593Smuzhiyun ``` 47*4882a593Smuzhiyun 48*4882a593Smuzhiyun 将生成的vendor/bin/rknn_create_mem_demo 49*4882a593Smuzhiyun 50*4882a593Smuzhiyun# 运行 51*4882a593Smuzhiyun 52*4882a593Smuzhiyun 53*4882a593Smuzhiyun- 将rknn_create_mem_demo推到板子/vendor/bin/目录 54*4882a593Smuzhiyun 55*4882a593Smuzhiyun- 将model推到板子/data/目录 56*4882a593Smuzhiyun 57*4882a593Smuzhiyun- 确保板子的rockchip.hardware.neuralnetworks@1.0-service已经运行 58*4882a593Smuzhiyun 59*4882a593Smuzhiyun``` 60*4882a593Smuzhiyun rknn_create_mem_demo /data/model/RK3566_RK3568/mobilenet_v1.rknn /data/model/dog_224x224.jpg 61*4882a593Smuzhiyun``` 62*4882a593Smuzhiyun 63*4882a593Smuzhiyun 64*4882a593Smuzhiyun 65*4882a593Smuzhiyun# FAQ 66*4882a593Smuzhiyun 67*4882a593Smuzhiyun- rockchip.hardware.neuralnetworks@1.0-service服务没有运行怎么办 68*4882a593Smuzhiyun 69*4882a593Smuzhiyun 如果该服务没有运行，从Android SDK确保vendor/rockchip/hardware/interfaces/neuralnetworks/目录存在，并且重新编译系统固件，并重新烧写到板子上，具体步骤请参考SDK编译固件的说明。 70*4882a593Smuzhiyun 71*4882a593Smuzhiyun- 遇到sizeof(rknn_tensor_attr) != sizeof(::rockchip::hardware::neuralnetworks::V1_0::RKNNTensorAttr)的错误 72*4882a593Smuzhiyun 73*4882a593Smuzhiyun 需要更新vendor/rockchip/hardware/interfaces/neuralnetworks到最新代码 74
