dbcppp

A C/C++ DBC file parser based on boost.spirit. This library is designed for decoding performance.

Features

very fast decoding
verbose parser output in error case
DBC is editable through C/C++ interface exported from the library
read/write DBC file
decode functionality for frames of arbitrarily byte length
cantools like decoding
KCD file format support

DBC data types

Supported

version
new_symbols
bit_timing
nodes
value_tables
messages
message_transmitters
environment_variables
environment_variables_data
signal_types
comments
attribute_definitions
attribute_defaults
attribute_values
value_descriptions
signal_extended_value_type_list

Not supported yet

sigtype_attr_list
signal_type_refs
signal_groups

Getting started

Dependencies

boost
libxml2

Build & Install

git clone https://github.com/xR3b0rn/dbcppp.git
cd dbcppp
mkdir build
cd build
cmake ..
make -j
make RunTests
make install

Usage example

Command line tool

dbc2

# generate C source from DBC/KCD
dbcppp dbc2 --dbc=file.dbc --format=C
# beauty or merge DBC/KCD
dbcppp dbc2 --dbc=file1.dbc --dbc=file2.kcd --format=DBC
# print DBC/KCD in human readable format
dbcppp dbc2 --dbc=file1.dbc --dbc=file2.kcd --format=human

decode

cantools like decoding:

candump any | dbcppp decode --bus=vcan0,file1.dbc --bus=vcan1,file2.dbc

Library

Examples
C++

#include <fstream>
#include <dbcppp/Network.h>
int main()
{
    std::ifstream dbc_file{"your_dbc.dbc"};
    std::unique_ptr<dbcppp::Network> net = dbcppp::Network::fromDBC(dbc_file);
    if (net)
    {
        can_frame frame;
        while (1)
        {
            receive_can_frame_from_somewhere(&frame);
            const Message* msg = net->getMessageById(frame.id);
            if (msg)
            {
                std::cout << "Received message: " << msg->getName() << std::endl;
                msg->forEachSignal(
                    [&](const Signal& signal)
                    {
                        uint64_t raw = signal.decode(frame.data);
                        std::cout << "\t" << signal.getName() << "=" << signal.rawToPhys(raw) << std::endl;
                    });
            }
        }
    }
}

#include <stdio.h>
#include <dbcppp/CApi.h>
int main()
{
    const dbcppp_Nework* net = dbcppp_NetworkLoadDBCFromFile("your_dbc.dbc");
    if (net)
    {
        can_frame frame;
        while (1)
        {
            receive_can_frame_from_somewhere(&frame);
            const dbcppp_Message* msg = dbcppp_NetworkGetMessageById(net, frame.id);
            if (msg)
            {
                printf("Received message: %s\n", dbcppp_MessageGetName(msg));
                void print_signal_data(const dbcppp_Signal* sig, void* data)
                {
                    can_frame* frame = (can_frame*)data;
                    uint64_t raw = dbcppp_SignalDecode(sig, frame->data);
                    double phys = dbcppp_SignalRawToPhys(sig, raw);
                    printf("\t%s=%f\n", dbcppp_SignalGetName(sig), phys);
                }
                dbcppp_MessageForEachSignal(msg, print_signal_data, &frame);
            }
        }
    }
}

Decode-function

The signals decode function is using prestored masks and fixed offsets to speed up calculation, therefore the decoding-function should be almost as fast as a code generated decode function would be. The assembly of the decode-function on its critical path (signed and byte swap must happen) looks like this (VS19 10.0.18362.0 compiler):

template <Alignment aAlignment, Signal::ByteOrder aByteOrder, Signal::ValueType aValueType, Signal::ExtendedValueType aExtendedValueType>
double template_decode(const Signal* sig, const void* nbytes) noexcept
00007FF8025BCA73  mov         rax,rcx  
00007FF8025BCA76  mov         rcx,qword ptr [rcx+140h]  
00007FF8025BCA7D  xorps       xmm0,xmm0  
00007FF8025BCA80  bswap       r8  
00007FF8025BCA83  shr         r8,cl  
00007FF8025BCA86  and         r8,qword ptr [rax+130h]  
00007FF8025BCA8D  mov         rcx,qword ptr [rax+138h]  
00007FF8025BCA94  mov         rax,rcx  
00007FF8025BCA97  or          rcx,r8  
00007FF8025BCA9A  and         rax,r8  
00007FF8025BCA9D  cmove       rcx,r8  
00007FF8025BCAA1  cvtsi2sd    xmm0,rcx  
00007FF8025BCAA6  ret

On the best path (no byteswap must take place and ExtendedValueType == Double) the decode function only has 5 instructions:

template <Alignment aAlignment, Signal::ByteOrder aByteOrder, Signal::ValueType aValueType, Signal::ExtendedValueType aExtendedValueType>
double template_decode(const Signal* sig, const void* nbytes) noexcept
00007FF8025BCAF0  mov         rax,qword ptr [rdx]  
00007FF8025BCAF3  mov         qword ptr [rsp+8],rcx  
00007FF8025BCAF8  mov         qword ptr [sig],rax  
00007FF8025BCAFD  movsd       xmm0,mmword ptr [data]  
00007FF8025BCB03  ret

Known issues

tests for decoding function for float/double is failing on some maschines (currently only confirmed for System/s390x)

Similar projects

Vector_DBC Does basically the same, the biggest difference is that it uses bison instead of boost::spirit for grammar parsing
CAN BUS tools in Python 3 (cantools)

ByunghunLee/dbcppp