NVDLA runtime vp 搭建

简介

NVDLA虚拟平台提供了一个可精确注册的系统，可以在该系统上快速开发和调试软件。该虚拟平台基于GreenSocs QBOX，它是与QEMU和SystemC共同仿真的解决方案。下面的图89显示了NVDLA虚拟模拟器的顶层示意图。包含ARMv8“虚拟” SoC板的QEMU仿真器，以提供高性能CPU仿真和通用设备。该仿真器包装在带有一组TLM-2.0接口的标准SystemC模块内，可以与其他SystemC模块进行交互。

1、下载 Virtual Simulator

$ git clone https://github.com/nvdla/vp.git
$ cd vp $ git submodule update --init --recursive

2、安装依赖

2.1、安装包

$ sudo apt-get install g++ cmake libboost-dev python-dev libglib2.0-dev libpixman-1-dev liblua5.2-dev swig libcap-dev libattr1-dev

2.2. 下载安装 SystemC 2.3.0

$ wget -O systemc-2.3.0a.tar.gz
http://www.accellera.org/images/downloads/standards/systemc/systemc-2.3.0a.tar.gz
$ tar -xzvf systemc-2.3.0a.tar.gz
$ cd systemc-2.3.0a
$ sudo mkdir -p
/usr/local/systemc-2.3.0/
$ mkdir objdir
$ cd objdir
$ …/configure --prefix=/usr/local/systemc-2.3.0
$ make
$ sudo make install
如果遇到make install没有安装到制定目录的问题，查看Makefile中prefix变量赋值 prefix = /usr/local/systemc-2.3.0

3、下载安装perl、YAML等包

$ wget -O YAML-1.24.tar.gz
http://search.cpan.org/CPAN/authors/id/T/TI/TINITA/YAML-1.24.tar.gz
$ tar -xzvf YAML-1.24.tar.gz $ cd YAML-1.24
$ perl Makefile.PL
$ make $
sudo make install
$ wget -O IO-Tee-0.65.tar.gz
http://search.cpan.org/CPAN/authors/id/N/NE/NEILB/IO-Tee-0.65.tar.gz
$ tar -xzvf IO-Tee-0.65.tar.gz
$ cd IO-Tee-0.65
$ perl Makefile.PL
$ make
$ sudo make install

4. 下载编译NVDLA CMOD

ps:确定有相关环境：jdk1.8等

$ git clone https://github.com/nvdla/hw.git $ cd hw
$ make 如果遇到make install没有安装到制定目录的问题，查看Makefile中prefix变量赋值 prefix =
/usr/local/systemc-2.3.0
$ cd hw
$ make
这里提示确认工具链位置，按照实际填写或者修改makefile
$ tools/bin/tmake -build cmod_top

5、编译安装Virtual Simulator

5.1 cmake

$ cmake -DCMAKE_INSTALL_PREFIX=[install dir] -DSYSTEMC_PREFIX=[systemc
prefix] -DNVDLA_HW_PREFIX=[nvdla_hw prefix]
-DNVDLA_HW_PROJECT=[nvdla_hw project name]

CMAKE_INSTALL_PREFIX是设置安装位置，SYSTEMC_PREFIX是systemC的安装位置，NVDLA_HW_PREFIX是hw的安装位置，NVDLA_HW_PROJECT是项目名称

例如： cmake -DCMAKE_INSTALL_PREFIX=build
-DSYSTEMC_PREFIX=/usr/local/systemc-2.3.0/ -DNVDLA_HW_PREFIX=/root/nvdla/vp/vp/hw -DNVDLA_HW_PROJECT=nv_full

5.2、编译安装

$ make

$ make install

如果是用的gcc python和文档中设定的版本不同make会报错
1
/usr/local/systemc-2.3.0/include/sysc/packages/boost/get_pointer.hpp:21:40: error: ‘template class std::auto_ptr’ is deprecated [-Werror=deprecated-declarations]
template T * get_pointer(std::auto_ptr const& p)

需要把CMakeList.txt中的-Werror去除去掉警告当错误，修改如下
修改vp/CMakeLists.txt

set(CMAKE_CXX_FLAGS “${CMAKE_CXX_FLAGS} -DSC_INCLUDE_DYNAMIC_PROCESSES -Wall ${AWS_FPGA_CFLAGS} $TRAC{E}_{C}FLAGS")set(CMAK{E}_{C}X{X}_{F}LAG{S}_{D}EBUG"${CMAKE_CXX_FLAGS_DEBUG} -DSC_INCLUDE_DYNAMIC_PROCESSES -Wall -DDEBUG_LOG=1 ${AWS_FPGA_CFLAGS} $TRAC{E}_{C}FLAGS")set(CMAK{E}_{C}X{X}_{F}LAG{S}_{R}ELEASE"${CMAKE_CXX_FLAGS_RELEASE} -DSC_INCLUDE_DYNAMIC_PROCESSES -Wall ${AWS_FPGA_CFLAGS} ${TRACE_CFLAGS}”)

将子目录下CMakeList.txt中的 CMAKE_CXX_FLAGS语句中的-Werror删除

./models/nvdla/CMakeLists.txt +87
./models/nvdla/CMakeLists.txt +88
./models/nvdla/CMakeLists.txt +89

./fpga/aws-fpga/fpga_sc_wrapper/CMakeLists.txt +91
./fpga/aws-fpga/fpga_sc_wrapper/CMakeLists.txt +92
./fpga/aws-fpga/fpga_sc_wrapper/CMakeLists.txt +93

./fpga/aws-fpga/cosim_sc_wrapper/CMakeLists.txt +82
./fpga/aws-fpga/cosim_sc_wrapper/CMakeLists.txt +83
./fpga/aws-fpga/cosim_sc_wrapper/CMakeLists.txt +84

2
/nvdla/vp/libs/greenlib/greenscript/src/greenscript.cpp: In member function ‘void gs::script::GreenScriptModule::load(const char*)’:
/home/bg193/works/nvdla/vp/libs/greenlib/greenscript/src/greenscript.cpp:221:46: error: ‘PyString_AsString’ was not declared in this scope
std::string s(PyString_AsString(path_py));
修改vp/libs/greenlib/CMakeLists.txt
/ # set(EXTERNAL_INCLUDE_DIRS ${EXTERNAL_INCLUDE_DIRS} ${PYTHON_INCLUDE_DIRS})
set(EXTERNAL_INCLUDE_DIRS ${EXTERNAL_INCLUDE_DIRS} “/usr/include/python2.7”)

原因为python不是2.x，需要手动指定
修改vp/libs/greenlib/CMakeLists.txt

/# set(EXTERNAL_INCLUDE_DIRS ${EXTERNAL_INCLUDE_DIRS}
${PYTHON_INCLUDE_DIRS}) set(EXTERNAL_INCLUDE_DIRS
${EXTERNAL_INCLUDE_DIRS} “/usr/include/python2.7”)

3
ac_fixed.h:410:108 nvdla error:invalid operands of types ‘’ and ‘int’ to

修改 cmod/hls/include/ac_fixed.h

— ac_fixed_orig.h 2018-08-29 14:47:59.977317789 -0700
+++ ac_fixed.h 2018-08-29 14:35:51.514589435 -0700 @@ -407,7 +407,7 @@ #endif
// Explicit conversion functions to ac_int that captures all integer bits (bits are truncated)

• inline ac_int<AC_MAX(I,1),S> to_ac_int() const { return ((ac_fixed<AC_MAX(I,1),AC_MAX(I,1),S>) *this).slc<AC_MAX(I,1)>(0); }

• inline ac_int<AC_MAX(I,1),S> to_ac_int() const { return ((ac_fixed<AC_MAX(I,1),AC_MAX(I,1),S>) *this).template
  slc<AC_MAX(I,1)>(0); }
  // Explicit conversion functions to C built-in types ------------- inline int to_int() const { return ((I-W) >= 32) ? 0 : (signed int) to_ac_int(); }

4
/home/me/vp/libs/greenlib/greensocket/include/greensocket/generic/green_socket_extension_support_base.tpp:127:29: error: expected primary-expression before ‘>’ token T* tmp=txn.get_extension(); //get the pointer ^ /home/me/vp/libs/greenlib/greensocket/include/greensocket/generic/green_socket_extension_support_base.tpp:127:31: error: expected primary-expression before ‘)’ token T* tmp=txn.get_extension(); //get the pointer

这错误是因为ubuntu gcc版本过高，可以继续修改代码支持该特性，我这边不想修改源码，直接将gcc回退到了4.8.5解决

5
/NVDLA/vp/libs/qbox/util/memfd.c:40:12: error: static declaration of ‘memfd_create’ follows non-static declaration static int memfd_create(const char *name, unsigned int flags)

新版本glibc库导致，修改代码
libs/qbox/util/memfd.c
//static int memfd_create(const char *name, unsigned int flags)
int memfd_create(const char *name, unsigned int flags)

6、运行 Virtual Simulator

如果自己不编译linux kernel和rootfs，sw/prebuild中提供了已经编好的
如果需要自己编译，可以参考这里

6.1 拷贝prebuild中的Image和rootfs.ext4

修改conf/aarch64_nvdla.lua，手动指定image和rootfs到实际位置

-kernel /home/works/nvdla/sw/prebuilt/linux/Image
-drive file=/home/works/nvdla/sw/prebuilt/linux/rootfs.ext4

$ export SC_SIGNAL_WRITE_CHECK=DISABLE
$ ./build/bin/aarch64_toplevel -c conf/aarch64_nvdla.lua
Login the kernel. The demo image uses account ‘root’ and password ‘nvdla’.

vp/tests中已经提供了几个例子能跑

:# mount -t 9p -o trans=virtio r /mnt
:# cd /mnt/tests/hello :#
./aarch64_hello
如果看到打印hello world则成功

如果要跑runtime，那就把prebuilt\arm64-linux中drm.ko、opendla1.ko、opendla2.ko nvdla_runtime拷贝过来

$:insmod drm.ko $:insmod opendla1.ko #根据实际架构选择1还是2

说明：
opendla1.ko :nv_full
opendla2.ko :nv_small、nv_large

./nvdla_runtime --loadable fast-math.nvdla --image 0000.jpg