C++ parsers are tough to build.
I can't speak with experience about using ANTLR's C++ grammars. Here I discuss what I learned by reading the notes attached to the the one I did see at the ANTLR site; in essence, the author produced an incomplete grammar. And that was for just C++98. It has been awhile since I looked; there may be others.
Our DMS Software Reengineering Toolkit has a robust C++ front end.
The lexer handles all the cruft for ANSI, GCC3, MS Visual Studio 2008, including large-precision floating point numbers, etc.
[EDIT: 12/2011. Now handles C++11 and OpenMP directives]
[EDIT: 3/2015: Now handles C++14 in both GCC and MS variants.
See some parse trees here on SO]
Having "just" a parser is actually not very useful. Above and beyond "just parsing", our front end will build ASTs, build accurate symbol tables (for C++, this is extremely hard to do), perform function-local flow analysis, and allow you to carry out program transformations, etc. See Life After Parsing.
[EDIT: 5/2019: Now handles C++17 in ANSI, GCC and MS variants. Does complete name and type resolution across compilation units. Used to automate large scale God-class refactoring/splitting across systems of 3000 compilation units.]