[LLVMdev] How to get the original function name in C++?

Nick Lewycky nicholas at mxc.ca
Thu Dec 11 11:35:30 PST 2014

zy jj wrote:
> Hi, everyone!
>      I'm new here trapped by a simple problem for days.
>      When LLVM translates C++ source code to IR, it will add a prefix to
> the function name. For example:
> source code:
> int foo(){
> return 1;
> }
> IR form:
> define i32 @_Z3foov() #0 {
> entry:
>    ret i32 1, !dbg !20
> }
>      The getName() method returns _Z3foov, then how can I get foo? I
> know the debugging information is contained in metadata, but I've no
> idea on using it.
>      Thanks a lot for any help.

You got a few answers about name mangling, but you also asked about 
debug info. If you want to find the original name through debug info 
instead of demangling the mangled name, the relevant subset of debug 
info works like this.

!llvm.dbg.cu = !{!0}

"CU" is short for "compilation unit", the DWARF[1] term for what C and 
C++ call a translation unit, a single .c/.cc file with all its includes. 
The only CU in this .ll file is !0.

!0 = metadata !{metadata !"0x11\004\00clang version 3.6.0 (trunk 
223918)\000\00\000\00\001", metadata !1, metadata !2, metadata !2, 
metadata !3, metadata !2, metadata !2} ; [ DW_TAG_compile_unit ] 
[/home/nicholas/zyjj.cc] [DW_LANG_C_plus_plus]

On the one hand our debug info format is free to change, but on the 
other hand it closely follows the DWARF spec. You can construct an 
llvm::DICompileUnit and pass the the MDNode* for !0, then query its 
getSubprograms() method to find what DWARF calls "subprograms" or what 
C/C++ call "functions". It returns a DIArray. For reference, the 
subprograms field is field 3 (zero-indexed) of the compile_unit, so it's 
"metadata !3".

!3 = metadata !{metadata !4}

Our array of subprograms, with one subprogram. Note that this need not 
include all subprograms; methods may only be listed in the 
DW_TAG_class_type for example.

!4 = metadata !{metadata 
metadata !1, metadata !5, metadata !6, null, i32 ()* @_Z3foov, null, 
null, metadata !2} ; [ DW_TAG_subprogram ] [line 1] [def] [foo]

You can construct a DISubprogram with this, and a DISubprogram has a few 
kinds of names. It has a "linkage name" which is the name that you need 
to know in order to link against it, aka. the mangled name. The other 
two names are called "name" and "display name". DWARF does define what a 
name should be, "a string representing the name as it appears in the 
source program" and goes on to clarify that it should be the unmangled 
form. Get it by calling DISubprogram::getName().

However, nowhere does DWARF define what a "display name" is. Taking a 
quick look through clang, I can't see where we ever set it differently 
from "name". If anyone else knows this, please let me know.


[1] - http://dwarfstd.org/Dwarf4Std.php

More information about the llvm-dev mailing list