llvm-g++ produces non-working output

[Quuxplusone]

Bugzilla Link	PR689
Status	RESOLVED FIXED
Importance	P normal
Reported by	Mycroft Holmes (hp48gx@gmail.com)
Reported on	2006-01-12 03:39:31 -0800
Last modified on	2010-03-06 13:59:19 -0800
Version	1.6
Hardware	Macintosh MacOS X
CC	llvm-bugs@lists.llvm.org, martin@mpa-garching.mpg.de
Fixed by commit(s)
Attachments
Blocks
Blocked by
See also

It looks like there's a bug in llvm-g++: a very simple c++ source code compiled
by llvm-gcc produces
a working bytecode, but the same file compiled by llvm-g++ generates a smaller
and non working
output:

PowerBookG4:/llvm/test mycroft$ cat hello.cpp
#include <iostream>

int main()
{
     std::cout << "hello world\n";
     return 0;
}

PowerBookG4:/llvm/test mycroft$ llvm-gcc hello.cpp
PowerBookG4:/llvm/test mycroft$ ls -l
total 440
-rwxr-xr-x   1 mycroft  admin      64 Jan 12 00:03 a.out
-rwxr-xr-x   1 mycroft  admin  216923 Jan 12 00:03 a.out.bc
-rw-r--r--   1 mycroft  admin      95 Jan 12 00:00 hello.cpp
PowerBookG4:/llvm/test mycroft$ ./a.out
hello world

PowerBookG4:/llvm/test mycroft$ rm a.*

PowerBookG4:/llvm/test mycroft$ llvm-g++ hello.cpp
PowerBookG4:/llvm/test mycroft$ ls -l
total 112
-rwxr-xr-x   1 mycroft  admin     64 Jan 12 00:04 a.out
-rwxr-xr-x   1 mycroft  admin  47320 Jan 12 00:04 a.out.bc
-rw-r--r--   1 mycroft  admin     95 Jan 12 00:00 hello.cpp
PowerBookG4:/llvm/test mycroft$ ./a.out
Segmentation fault

[Quuxplusone]

This implies that llvm-g++'s libstdc++ is corrupt.  Where did you get your
llvmgcc?  Is it llvm-gcc 1.6 or
something earlier?  If it is 1.6, please try updating it to the latest CVS
version: it works fine for me on OS X
10.4.4.

-Chris

[Quuxplusone]

yes, it's the released 1.6 binary package downloaded from website.
I run fixheader with MacOS 10.4.3 but now it's running in 10.4.4

I'll have some more check and let you know, but why gcc works when g++ does not?

[Quuxplusone]

when you link with 'gcc', it uses the system libstdc++ library.  When you link
with g++, it uses the LLVM
version.

BTW, note that compiling a program like that invokes the JIT when you run a.out.

-Chris

[Quuxplusone]

Were you able to reproduce this problem with a CVS llvm-gcc?  If you can't
check, I'm going to have to
close this as "works for me".

Thanks,

-Chris

[Quuxplusone]

thanks, I've only a 56k dial-up connection here, so I won't try to checkout
llvm-gcc...
If you prefer, close this bug. I'll open a new one if necessary (I hope not to!
:)

[Quuxplusone]

ok, sounds fair.

-Chris

[Quuxplusone]

I'm observing the same behaviour on i386 with LLVM 1.7.
Whenever I compile a C++ code that tries to write to cout I get a segmentation
fault as soon as I run it.
I tried to compile the frontend from SVN sources but failed so far.

Anyway, since it appears to work fine for most other users, I expect that I must
have made a simple mistake during my LLVM setup. Any idea what this could be?

[Quuxplusone]

Hi,

the bug reappeared with llvm 1.7, and it's much worse:

I'm working on a powerbook g4, with 10.4.6
I use the binary version of llvm-gcc v3.4 from the download section on the
website.
I configure llvm with "--enable-optimized --enable-jit"

llvm compiles fine, but now BOTH llvm-gcc and llvm-g++ produce an identical
bytecode (and
segmentation fault...)

if I use llvm-gcc4 (powerpc) binaries I get a bizarre result: first, in the
archive, the executables don't
have the llvm- prefix. If I rename them and compile llvm as above, when I
compile something I don't
get bytecode, but a native executable, which however does not run (it says:
invalid CPU...)

please, tell me what's happening (hint: please, provide a BINARY setup...)

[Quuxplusone]

recompiling the gcc4 frontend partially fixes: the new gcc4 does compile the
"hello world" c++ sample, it
will NOT produce bytecode, but at least the binary file works.

[Quuxplusone]

run llvm-gcc4 -v. Does it print the string "LLVM"? If not, you have a native GCC, not an LLVM GCC.

[Quuxplusone]

I'm not that stupid... ;)
It's the first thing I checked.

powerbookg4:/llvm/llvm-gcc4/bin mycroft$ ./llvm-gcc -v
Using built-in specs.
Target: powerpc-apple-darwin8.6.0
Configured with: ../gccsrc/configure --prefix=/llvm/llvm-gcc -enable-
llvm=/llvm/llvmobj --enable-
languages=c,c++,objc,obj-c++
Thread model: posix
gcc version 4.0.1 LLVM (Apple Computer, Inc. build 5400)

[Quuxplusone]

I forgot to mention: this is the version I recompiled, however the binaries you
distribute have a similar
version dump.
(the filename is originally gcc and not llvm-gcc, but I renamed it)

[Quuxplusone]

> I'm not that stupid... ;)
> It's the first thing I checked.

No insult intended.  Often the most obvious cause is the right one :)

If I had to guess, I would guess it's because you renamed the executables.
Please try naming them
"gcc" and using a symlink from llvm-gcc to gcc.  GCC is funny about how it
finds parts like "cc1".

If that doesn't help, please try a simple example, e.g.:

echo "int foo(){}" > test.c
llvm-gcc test.c -S -o - -emit-llvm -v

and include the output.

Thanks,

-Chris

[Quuxplusone]

still the same: no bytecode.
here's the gcc4 directory  (it's in the path)

powerbookg4:/llvm/llvm-gcc/bin mycroft$ ls -l
total 97864
-rwxr-xr-x   4 mycroft  admin  5586284 Jun 22 07:25 c++
-rwxr-xr-x   1 mycroft  admin  5585780 Jun 22 07:25 cpp
-rwxr-xr-x   4 mycroft  admin  5586284 Jun 22 07:25 g++
-rwxr-xr-x   3 mycroft  admin  5584860 Jun 22 07:25 gcc
-rwxr-xr-x   1 mycroft  admin    15728 Jun 22 07:25 gccbug
-rwxr-xr-x   1 mycroft  admin  5367872 Jun 22 07:25 gcov
lrwxr-xr-x   1 mycroft  admin        3 Jun 27 09:58 llvm-c++ -> c++
lrwxr-xr-x   1 mycroft  admin        3 Jun 27 09:58 llvm-cpp -> cpp
lrwxr-xr-x   1 mycroft  admin        3 Jun 27 09:58 llvm-g++ -> g++
lrwxr-xr-x   1 mycroft  admin        3 Jun 27 09:58 llvm-gcc -> gcc
lrwxr-xr-x   1 mycroft  admin        6 Jun 27 09:58 llvm-gccbug -> gccbug
lrwxr-xr-x   1 mycroft  admin        4 Jun 27 09:58 llvm-gcov -> gcov
-rwxr-xr-x   4 mycroft  admin  5586284 Jun 22 07:25 powerpc-apple-darwin8.6.0-
c++
-rwxr-xr-x   4 mycroft  admin  5586284 Jun 22 07:25 powerpc-apple-darwin8.6.0-
g++
-rwxr-xr-x   3 mycroft  admin  5584860 Jun 22 07:25 powerpc-apple-darwin8.6.0-
gcc
-rwxr-xr-x   3 mycroft  admin  5584860 Jun 22 07:25 powerpc-apple-darwin8.6.0-
gcc-4.0.1

powerbookg4:/llvm/test mycroft$ ls
hello.cpp       test.c

powerbookg4:/llvm/test mycroft$ llvm-gcc hello.cpp
/usr/bin/ld: Undefined symbols:
std::ios_base::Init::Init()
std::ios_base::Init::~Init()
std::cout
std::basic_ostream<char, std::char_traits<char> >& std::operator<<
<std::char_traits<char> >
(std::basic_ostream<char, std::char_traits<char> >&, char const*)
collect2: ld returned 1 exit status

powerbookg4:/llvm/test mycroft$ llvm-g++ hello.cpp
powerbookg4:/llvm/test mycroft$ ls
a.out           hello.cpp       test.c
powerbookg4:/llvm/test mycroft$ ./a.out
hello world

powerbookg4:/llvm/test mycroft$ llvm-gcc test.c -S -o - -emit-llvm -v
Using built-in specs.
Target: powerpc-apple-darwin8.6.0
Configured with: ../gccsrc/configure --prefix=/llvm/llvm-gcc -enable-
llvm=/llvm/llvmobj --enable-
languages=c,c++,objc,obj-c++
Thread model: posix
gcc version 4.0.1 LLVM (Apple Computer, Inc. build 5400)
 /llvm/llvm-gcc/libexec/gcc/powerpc-apple-darwin8.6.0/4.0.1/cc1 -quiet -v -D__DYNAMIC__ test.c -
emit-llvm -fPIC -quiet -dumpbase test.c -auxbase-strip - -version -o -
ignoring nonexistent directory "/llvm/llvm-gcc/lib/gcc/powerpc-apple-
darwin8.6.0/4.0.1/../../../../
powerpc-apple-darwin8.6.0/include"
#include "..." search starts here:
#include <...> search starts here:
 /usr/local/include
 /llvm/llvm-gcc/include
 /llvm/llvm-gcc/lib/gcc/powerpc-apple-darwin8.6.0/4.0.1/include
 /usr/include
 /System/Library/Frameworks
 /Library/Frameworks
End of search list.
GNU C version 4.0.1 LLVM (Apple Computer, Inc. build 5400) (powerpc-apple-
darwin8.6.0)
        compiled by GNU C version 4.0.1 (Apple Computer, Inc. build 5341).
GGC heuristics: --param ggc-min-expand=100 --param ggc-min-heapsize=131072
Compiler executable checksum: 552bb8881767f70d3bc8f89513fbe650
; ModuleID = 'test.c'
target endian = big
target pointersize = 32
target triple = "powerpc-apple-darwin8.6.0"

implementation   ; Functions:

int %foo() {
entry:
        %retval = alloca int, align 4           ; <int*> [#uses=1]
        "alloca point" = cast int 0 to int              ; <int> [#uses=0]
        br label %return

return:         ; preds = %entry
        %retval = load int* %retval             ; <int> [#uses=1]
        ret int %retval
}

if you are interested:

powerbookg4:/llvm/test mycroft$ llvm-gcc hello.cpp  -S -o - -emit-llvm -v
Using built-in specs.
Target: powerpc-apple-darwin8.6.0
Configured with: ../gccsrc/configure --prefix=/llvm/llvm-gcc -enable-
llvm=/llvm/llvmobj --enable-
languages=c,c++,objc,obj-c++
Thread model: posix
gcc version 4.0.1 LLVM (Apple Computer, Inc. build 5400)
 /llvm/llvm-gcc/libexec/gcc/powerpc-apple-darwin8.6.0/4.0.1/cc1plus -quiet -v -D__DYNAMIC__
hello.cpp -emit-llvm -fPIC -quiet -dumpbase hello.cpp -auxbase-strip - -version
-o - -
D__private_extern__=extern
ignoring nonexistent directory "/llvm/llvm-gcc/lib/gcc/powerpc-apple-
darwin8.6.0/4.0.1/../../../../
powerpc-apple-darwin8.6.0/include"
#include "..." search starts here:
#include <...> search starts here:
 /llvm/llvm-gcc/lib/gcc/powerpc-apple-darwin8.6.0/4.0.1/../../../../include/c++/4.0.1
 /llvm/llvm-gcc/lib/gcc/powerpc-apple-darwin8.6.0/4.0.1/../../../../include/c++/4.0.1/powerpc-
apple-darwin8.6.0
 /llvm/llvm-gcc/lib/gcc/powerpc-apple-darwin8.6.0/4.0.1/../../../../include/c++/4.0.1/backward
 /usr/local/include
 /llvm/llvm-gcc/include
 /llvm/llvm-gcc/lib/gcc/powerpc-apple-darwin8.6.0/4.0.1/include
 /usr/include
 /System/Library/Frameworks
 /Library/Frameworks
End of search list.
GNU C++ version 4.0.1 LLVM (Apple Computer, Inc. build 5400) (powerpc-apple-
darwin8.6.0)
        compiled by GNU C version 4.0.1 (Apple Computer, Inc. build 5341).
GGC heuristics: --param ggc-min-expand=100 --param ggc-min-heapsize=131072
Compiler executable checksum: 099f891e5e9b8ccebf1bf8512c58370d
; ModuleID = 'hello.cpp'
target endian = big
target pointersize = 32
target triple = "powerpc-apple-darwin8.6.0"
        "struct.std::basic_ios<char,std::char_traits<char> >" = type { "struct.std::ios_base",
"struct.std::basic_ostream<char,std::char_traits<char> >"*, sbyte, uint,
"struct.std::basic_streambuf<char,std::char_traits<char> >"*,
"struct.std::ctype<char>"*,
"struct.std::num_get<char,std::istreambuf_iterator<char, std::char_traits<char>
> >"*,
"struct.std::num_get<char,std::istreambuf_iterator<char, std::char_traits<char>
> >"* }
        "struct.std::basic_ostream<char,std::char_traits<char> >" = type { int (...)**,
"struct.std::basic_ios<char,std::char_traits<char> >" }
        "struct.std::basic_streambuf<char,std::char_traits<char> >" = type { int (...)**, sbyte*, sbyte*,
sbyte*, sbyte*, sbyte*, sbyte*, "struct.std::locale" }
        "struct.std::ctype<char>" = type { "struct.std::locale::facet", int*, uint, int*, int*, uint*, sbyte, [256
x sbyte], [256 x sbyte], sbyte }
        "struct.std::ctype_base" = type { ubyte }
        "struct.std::ios_base" = type { int (...)**, int, int, uint, uint, uint,
"struct.std::ios_base::_Callback_list"*, "struct.std::ios_base::_Words", [8 x
"struct.std::ios_base::_Words"],
int, "struct.std::ios_base::_Words"*, "struct.std::locale" }
        "struct.std::ios_base::Init" = type { ubyte }
        "struct.std::ios_base::_Callback_list" = type { "struct.std::ios_base::_Callback_list"*, void (uint,
"struct.std::ios_base"*, int)*, int, int }
        "struct.std::ios_base::_Words" = type { sbyte*, int }
        "struct.std::locale" = type { "struct.std::locale::_Impl"* }
        "struct.std::locale::_Impl" = type { int, "struct.std::locale::facet"**, uint, "struct.std::locale::facet"**,
sbyte** }
        "struct.std::locale::facet" = type { int (...)**, int }
        "struct.std::num_get<char,std::istreambuf_iterator<char, std::char_traits<char> > >" = type
{ "struct.std::locale::facet" }
        "struct.std::num_put<char,std::ostreambuf_iterator<char, std::char_traits<char> > >" = type
{ "struct.std::locale::facet" }
%_ZSt8__ioinit = internal global "struct.std::ctype_base" zeroinitializer
;
<"struct.std::ctype_base"*> [#uses=2]
%_ZSt4cout = external global
"struct.std::basic_ostream<char,std::char_traits<char> >"          ;
<"struct.std::basic_ostream<char,std::char_traits<char> >"*> [#uses=1]
%str = internal constant [13 x sbyte] c"hello world\0A\00"              ; <[13
x sbyte]*> [#uses=1]
%llvm.global_ctors = appending global [1 x { int, void ()* }] [ { int, void ()*
} { int 65535, void ()* %
_GLOBAL__I_main } ]            ; <[1 x { int, void ()* }]*> [#uses=0]
%llvm.global_dtors = appending global [1 x { int, void ()* }] [ { int, void ()*
} { int 65535, void ()* %
_GLOBAL__D_main } ]            ; <[1 x { int, void ()* }]*> [#uses=0]

implementation   ; Functions:

int %main() {
entry:
        %retval = alloca int, align 4           ; <int*> [#uses=3]
        %tmp = alloca int, align 4              ; <int*> [#uses=4]
        "alloca point" = cast int 0 to int              ; <int> [#uses=0]
        %tmp = getelementptr [13 x sbyte]* %str, int 0, uint 0          ; <sbyte*> [#uses=1]
        %tmp = call "struct.std::basic_ostream<char,std::char_traits<char> >"* %
_ZStlsISt11char_traitsIcEERSt13basic_ostreamIcT_ES5_PKc
( "struct.std::basic_ostream<char,std::char_traits<char> >"* %_ZSt4cout, sbyte*
%tmp )               ;
<"struct.std::basic_ostream<char,std::char_traits<char> >"*> [#uses=0]
        store int 0, int* %tmp
        %tmp = load int* %tmp           ; <int> [#uses=1]
        store int %tmp, int* %retval
        br label %return
                ; No predecessors!
        store int 0, int* %tmp
        %tmp1 = load int* %tmp          ; <int> [#uses=1]
        store int %tmp1, int* %retval
        br label %return

return:         ; preds = %0, %entry
        %retval = load int* %retval             ; <int> [#uses=1]
        ret int %retval
}

internal void %_GLOBAL__I_main() section
"__TEXT,__StaticInit,regular,pure_instructions" {
entry:
        call void %_Z41__static_initialization_and_destruction_0ii( int 1, int 65535 )
        br label %return

return:         ; preds = %entry
        ret void
}

internal void %_GLOBAL__D_main() section
"__TEXT,__StaticInit,regular,pure_instructions" {
entry:
        call void %_Z41__static_initialization_and_destruction_0ii( int 0, int 65535 )
        br label %return

return:         ; preds = %entry
        ret void
}

declare "struct.std::basic_ostream<char,std::char_traits<char> >"* %
_ZStlsISt11char_traitsIcEERSt13basic_ostreamIcT_ES5_PKc
("struct.std::basic_ostream<char,std::char_traits<char> >"*, sbyte*)

internal void %_Z41__static_initialization_and_destruction_0ii(int
%__initialize_p, int %__priority) section
"__TEXT,__StaticInit,regular,pure_instructions" {
entry:
        %__initialize_p_addr = alloca int               ; <int*> [#uses=3]
        %__priority_addr = alloca int           ; <int*> [#uses=3]
        "alloca point" = cast int 0 to int              ; <int> [#uses=0]
        store int %__initialize_p, int* %__initialize_p_addr
        store int %__priority, int* %__priority_addr
        %tmp = load int* %__priority_addr               ; <int> [#uses=1]
        %tmp = seteq int %tmp, 65535            ; <bool> [#uses=2]
        %tmp = cast bool %tmp to uint           ; <uint> [#uses=0]
        br bool %tmp, label %cond_true, label %cond_false4

cond_true:              ; preds = %entry
        %tmp1 = load int* %__initialize_p_addr          ; <int> [#uses=1]
        %tmp2 = seteq int %tmp1, 1              ; <bool> [#uses=2]
        %tmp2 = cast bool %tmp2 to uint         ; <uint> [#uses=0]
        br bool %tmp2, label %cond_true3, label %cond_false

cond_true3:             ; preds = %cond_true
        call void %_ZNSt8ios_base4InitC1Ev( "struct.std::ctype_base"* %_ZSt8__ioinit )
        br label %cond_next

cond_false:             ; preds = %cond_true
        br label %cond_next

cond_next:              ; preds = %cond_false, %cond_true3
        br label %cond_next5

cond_false4:            ; preds = %entry
        br label %cond_next5

cond_next5:             ; preds = %cond_false4, %cond_next
        %tmp6 = load int* %__priority_addr              ; <int> [#uses=1]
        %tmp7 = seteq int %tmp6, 65535          ; <bool> [#uses=2]
        %tmp7 = cast bool %tmp7 to uint         ; <uint> [#uses=0]
        br bool %tmp7, label %cond_true8, label %cond_false14

cond_true8:             ; preds = %cond_next5
        %tmp9 = load int* %__initialize_p_addr          ; <int> [#uses=1]
        %tmp10 = seteq int %tmp9, 0             ; <bool> [#uses=2]
        %tmp10 = cast bool %tmp10 to uint               ; <uint> [#uses=0]
        br bool %tmp10, label %cond_true11, label %cond_false12

cond_true11:            ; preds = %cond_true8
        call void %_ZNSt8ios_base4InitD1Ev( "struct.std::ctype_base"* %_ZSt8__ioinit )
        br label %cond_next13

cond_false12:           ; preds = %cond_true8
        br label %cond_next13

cond_next13:            ; preds = %cond_false12, %cond_true11
        br label %cond_next15

cond_false14:           ; preds = %cond_next5
        br label %cond_next15

cond_next15:            ; preds = %cond_false14, %cond_next13
        br label %return

return:         ; preds = %cond_next15
        ret void
}

declare void %_ZNSt8ios_base4InitC1Ev("struct.std::ctype_base"*)

declare void %_ZNSt8ios_base4InitD1Ev("struct.std::ctype_base"*)

[Quuxplusone]

From your transcript, it looks like everything is working perfectly.  To build
.bc files with llvm-gcc4 use
either:

llvm-gcc file.c -c -emit-llvm -o file.bc -O3

or:
llvm-gcc file.c -S -emit-llvm -o file.ll -O3
llvm-as file.ll -o file.bc

-Chris

[Quuxplusone]

This is fixed in recent drops of llvm-gcc4.

-Chris