【Milk-V Duo S 开发板免费体验】RISC-V核心NCNN基准测试

ncnn是腾讯优图实验室推出的一个为移动端极致优化的高性能神经网络前向计算框架，是目前同样也比较罕见的为 RISC-V 架构做过适配和优化的神经网络框架。

本文是一份教程，步骤 (step by step) 地展示了如何在一块全新的Milk-V Duo S 开发板开发板上，跑个 ncnn 神经网络推理框架的 demo。

首先从网络上下载NCNN的最新版源码。

git clone https://github.com/Tencent/ncnn.git

然后配置编译环境，配置好RISC-V的gcc和g++编译器地址，最好在ncnn 的cmakefile文件中set好。

set(CMAKE_C_COMPILER /home/gcc/riscv64-linux-musl-x86_64/bin/riscv64-unknown-linux-musl-gcc)
set(CMAKE_CXX_COMPILER /home/gcc/riscv64-linux-musl-x86_64/bin/riscv64-unknown-linux-musl-g++)

之后，新建一个duos的文件夹，并进入，执行cmake -DCMAKE_TOOLCHAIN_FILE=../toolchains/riscv64-linux-gnu.toolchain.cmake ..配置makefile文件。

root@DeepThink:~/ncnn/duo$ cmake -DCMAKE_TOOLCHAIN_FILE=../toolchains/riscv64-linux-gnu.toolchain.cmake ..
-- CMAKE_TOOLCHAIN_FILE = /home/ncnn/toolchains/riscv64-linux-gnu.toolchain.cmake
-- CMAKE_INSTALL_PREFIX = /home/ncnn/duo/install
-- NCNN_VERSION_STRING = 1.0.20250824
-- The C compiler identification is GNU 10.2.0
-- The CXX compiler identification is GNU 10.2.0
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Check for working C compiler: /home/gcc/riscv64-linux-musl-x86_64/bin/riscv64-unknown-linux-musl-gcc - skipped
-- Detecting C compile features
-- Detecting C compile features - done
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Check for working CXX compiler: /home/gcc/riscv64-linux-musl-x86_64/bin/riscv64-unknown-linux-musl-g++ - skipped
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- Performing Test NCNN_COMPILER_SUPPORT_GNU_INLINE_ASM
-- Performing Test NCNN_COMPILER_SUPPORT_GNU_INLINE_ASM - Success
-- Performing Test NCNN_COMPILER_SUPPORT_RISCV_V
-- Performing Test NCNN_COMPILER_SUPPORT_RISCV_V - Failed
-- Performing Test NCNN_COMPILER_SUPPORT_RISCV_ZFH
-- Performing Test NCNN_COMPILER_SUPPORT_RISCV_ZFH - Failed
-- Performing Test NCNN_COMPILER_SUPPORT_RISCV_ZVFH
-- Performing Test NCNN_COMPILER_SUPPORT_RISCV_ZVFH - Failed
-- Performing Test NCNN_COMPILER_SUPPORT_RISCV_XTHEADVECTOR
-- Performing Test NCNN_COMPILER_SUPPORT_RISCV_XTHEADVECTOR - Failed
CMake Warning at CMakeLists.txt:446 (message):
  The compiler does not support risc-v v or xtheadvector extension.  NCNN_RVV
  will be OFF.


CMake Warning at CMakeLists.txt:452 (message):
  The compiler does not support risc-v xtheadvector extension.
  NCNN_XTHEADVECTOR will be OFF.


CMake Warning at CMakeLists.txt:465 (message):
  The compiler does not support risc-v zfh extension.  NCNN_ZFH will be OFF.


-- Target arch: riscv 64bit
-- Performing Test COMPILER_HAS_HIDDEN_VISIBILITY
-- Performing Test COMPILER_HAS_HIDDEN_VISIBILITY - Success
-- Performing Test COMPILER_HAS_HIDDEN_INLINE_VISIBILITY
-- Performing Test COMPILER_HAS_HIDDEN_INLINE_VISIBILITY - Success
-- Performing Test COMPILER_HAS_DEPRECATED_ATTR
-- Performing Test COMPILER_HAS_DEPRECATED_ATTR - Success
-- Found OpenMP_C: -fopenmp (found version "4.5")
-- Found OpenMP_CXX: -fopenmp (found version "4.5")
-- Found OpenMP: TRUE (found version "4.5")
-- Performing Test CMAKE_HAVE_LIBC_PTHREAD
-- Performing Test CMAKE_HAVE_LIBC_PTHREAD - Success
-- Found Threads: TRUE
-- Configuring done (17.6s)
-- Generating done (0.0s)
-- Build files have been written to: /home/ncnn/duos

然后执行make命令编译文件，直至完成。

将生成的benchncnn和主目录下的模型文件全部拷贝到Milk-V Duo S 开发板。

执行NCNN基准测试，结果如下：