Hello,
I’ve been trying to build some normal linux applications with clang and I keep running into a snag in configure scripts. Perhaps these scripts are not written very well, but the current clang behavior seems strange. When I run clang with the -emit-llvm option, it still uses /usr/bin/ld as the linker which fails to compile llvm bitcode on linux (though it works on mac). Here’s a simple example:
$ cat > test.c
int main(void) { return 0; }
$ clang -emit-llvm test.c
/tmp/cc-pDfs56.o: file not recognized: File format not recognized
clang: error: linker command failed with exit code 1 (use -v to see invocation)
$ clang -emit-llvm test.c -c
$
It seems to me that clang should call llvm-ld when it generates llvm code and ld otherwise, but maybe there’s something that I have configured incorrectly. What is the appropriate way to address this?
Thanks.
Hello,
I've been trying to build some normal linux applications with clang and I keep running into a snag in configure scripts. Perhaps these scripts are not written very well, but the current clang behavior seems strange. When I run clang with the -emit-llvm option, it still uses /usr/bin/ld as the linker which fails to compile llvm bitcode on linux (though it works on mac). Here's a simple example:
$ cat > test.c
int main(void) { return 0; }
$ clang -emit-llvm test.c
/tmp/cc-pDfs56.o: file not recognized: File format not recognized
clang: error: linker command failed with exit code 1 (use -v to see invocation)
$ clang -emit-llvm test.c -c
$
It seems to me that clang should call llvm-ld when it generates llvm code and ld otherwise, but maybe there's something that I have configured incorrectly. What is the appropriate way to address this?
llvm-ld is not a full replacement for a system ld. If you want to do LTO (which is why I assume you
are using -emit-llvm), you should use a linker that support it. Gold support plugins and can
use the LLVM one. Recently gnu ld implemented the same interface, but I haven't tested it.
http://llvm.org/docs/GoldPlugin.html
Thanks.
Cheers,
Rafael
Hi Rafael,
I’ve tried the LTO options with dragonegg, but I haven’t had any luck getting it to work, I’m not sure if it is related to my installation of gold or my gcc configuration.
I’m really interested in getting the resulting program as bitcode though. Will LTO do that for me? What I’m trying to do is build a bitcode version of libc that I can combine with a program and analyze/optimize the resulting program.
2011/6/28 Rafael Ávila de Espíndola <rafael.espindola@gmail.com>
Hi Gregory,
I've tried the LTO options with dragonegg, but I haven't had any luck getting it
to work,
details?
Ciao, Duncan.
I'm not sure if it is related to my installation of gold or my gcc
Hi Duncan,
Hi Gregory,
I’ve tried the LTO options with dragonegg, but I haven’t had any luck getting it
to work,
details?
I think this is a misunderstanding on my part about what the dragonegg plugin is supposed to do. I am looking for something that will essentially be a replacement for gcc except that it generates llvm bitcode instead of normal object files. If I run something like:
gcc-4.5 -fplugin=/homes/malecha/tmp/libc-scratch/dragonegg/dragonegg.so a.c -c -o a.o -flto
/tmp/ccRKW0GD.s: Assembler messages:
/tmp/ccRKW0GD.s:1: Error: junk at end of line, first unrecognized character is .' /tmp/ccRKW0GD.s:2: Error: no such instruction: target triple=“x86_64–linux-gnu”’
/tmp/ccRKW0GD.s:4: Error: no such instruction: `module asm “\09.ident\09\22GCC: (GNU) 4.5.2 LLVM: 133848\22”’
…
I assume because my system assembler doesn’t understand llvm bitcode. So I ended up wrapping it up in a script that calls gcc with -emit-llvm -S and then runs llvm-as on the resulting llvm assembly code. I think that I just need to manually (i.e. with a script) expand the gcc calls to incorporate the llvm toolchain.
Hi Gregory,
I think this is a misunderstanding on my part about what the dragonegg plugin is
supposed to do. I am looking for something that will essentially be a
replacement for gcc except that it generates llvm bitcode instead of normal
object files. If I run something like:
you need to use -S otherwise gcc will pass the output to the system assembler,
which will barf (as you saw) since it doesn't understand LLVM IR.
gcc-4.5 -fplugin=/homes/malecha/tmp/libc-scratch/dragonegg/dragonegg.so a.c -c
-o a.o -flto
/tmp/ccRKW0GD.s: Assembler messages:
/tmp/ccRKW0GD.s:1: Error: junk at end of line, first unrecognized character is `.'
/tmp/ccRKW0GD.s:2: Error: no such instruction: `target triple="x86_64--linux-gnu"'
/tmp/ccRKW0GD.s:4: Error: no such instruction: `module asm "\09.ident\09\22GCC:
(GNU) 4.5.2 LLVM: 133848\22"'
...
I assume because my system assembler doesn't understand llvm bitcode. So I ended
up wrapping it up in a script that calls gcc with -emit-llvm -S and then runs
llvm-as on the resulting llvm assembly code. I think that I just need to
manually (i.e. with a script) expand the gcc calls to incorporate the llvm
toolchain.
Right, you need to run llvm-as manually. This is a pain, but I didn't find I
good solution for this yet.
Ciao, Duncan.