[Bug other/39573] New: linking fails when optimizations are enabled

[Bug other/39573] New: linking fails when optimizations are enabled

[fpbeekhof at gmail dot com]

Linking in debugging mode is fine:
$ make CNF=gcc MODE=debug
scons -j 2 CNF=gcc MODE=debug BACKEND=
scons: Reading SConscript files ...
scons: done reading SConscript files.
scons: Building targets ...
g++ -o main.o -c -pipe -I. -Wall -g -fopenmp main.cpp
g++ -o tinystr.o -c -pipe -I. -Wall -g -fopenmp tinystr.cpp
g++ -o tinyxml.o -c -pipe -I. -Wall -g -fopenmp tinyxml.cpp
g++ -o tinyxmlerror.o -c -pipe -I. -Wall -g -fopenmp tinyxmlerror.cpp
g++ -o tinyxmlparser.o -c -pipe -I. -Wall -g -fopenmp tinyxmlparser.cpp
g++ -o shapes -g -fopenmp tinystr.o tinyxml.o tinyxmlerror.o tinyxmlparser.o
main.o -lcvmlcpp
scons: done building targets.

But in release mode, it's not:
$ make CNF=gcc
scons -j 2 CNF=gcc MODE= BACKEND=
scons: Reading SConscript files ...
scons: done reading SConscript files.
scons: Building targets ...
g++ -o main.o -c -pipe -I. -Wall -O3 -funroll-loops -fopenmp -march=native
-ftree-vectorize -DNDEBUG main.cpp
g++ -o tinystr.o -c -pipe -I. -Wall -O3 -funroll-loops -fopenmp -march=native
-ftree-vectorize -DNDEBUG tinystr.cpp
g++ -o tinyxml.o -c -pipe -I. -Wall -O3 -funroll-loops -fopenmp -march=native
-ftree-vectorize -DNDEBUG tinyxml.cpp
g++ -o tinyxmlerror.o -c -pipe -I. -Wall -O3 -funroll-loops -fopenmp
-march=native -ftree-vectorize -DNDEBUG tinyxmlerror.cpp
g++ -o tinyxmlparser.o -c -pipe -I. -Wall -O3 -funroll-loops -fopenmp
-march=native -ftree-vectorize -DNDEBUG tinyxmlparser.cpp
g++ -o shapes -fopenmp tinystr.o tinyxml.o tinyxmlerror.o tinyxmlparser.o
main.o -lcvmlcpp
main.o: In function
`_ZN7cvmlcppL25extractSurfaceFromAdapterIN6shapes20ShapeSurfaceAdaptor_IdEEdEEvRKT_RNS_8GeometryIT0_EEd.omp_fn.1':
main.cpp:(.text+0x5838): undefined reference to `void
cvmlcpp::extractSurfaceFromAdapter<shapes::ShapeSurfaceAdaptor_<double>,
double>(shapes::ShapeSurfaceAdaptor_<double> const&,
cvmlcpp::Geometry<double>&, double)::C.724'
collect2: ld returned 1 exit status
scons: *** [shapes] Error 1
scons: building terminated because of errors.
make: *** [compile] Error 2

$ g++ -v
Using built-in specs.
Target: x86_64-linux-gnu
Configured with: ../src/configure -v --with-pkgversion='Ubuntu 4.3.2-1ubuntu12'
--with-bugurl=file:///usr/share/doc/gcc-4.3/README.Bugs
--enable-languages=c,c++,fortran,objc,obj-c++ --prefix=/usr --enable-shared
--with-system-zlib --libexecdir=/usr/lib --without-included-gettext
--enable-threads=posix --enable-nls --with-gxx-include-dir=/usr/include/c++/4.3
--program-suffix=-4.3 --enable-clocale=gnu --enable-libstdcxx-debug
--enable-objc-gc --enable-mpfr --enable-checking=release
--build=x86_64-linux-gnu --host=x86_64-linux-gnu --target=x86_64-linux-gnu
Thread model: posix
gcc version 4.3.2 (Ubuntu 4.3.2-1ubuntu12) 

Unfortunately, I'm not exactly sure how to complete this report. I have no idea
what component does this, so I just guessed. And should I simply attach all
source code ?


-- 
           Summary: linking fails when optimizations are enabled
           Product: gcc
           Version: 4.3.2
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: other
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: fpbeekhof at gmail dot com


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39573

[Bug other/39573] linking fails when optimizations are enabled

[fpbeekhof at gmail dot com]

------- Comment #1 from fpbeekhof at gmail dot com  2009-04-03 12:01 -------
I've tried compiling the same code on several machines. Outcome: intel machines
are fine, AMD machines exhibit this behaviour.

The second AMD machine has ubuntu 8.04 on it, and a different compiler version:
$ gcc -v
Using built-in specs.
Target: x86_64-linux-gnu
Configured with: ../src/configure -v
--enable-languages=c,c++,fortran,objc,obj-c++,treelang --prefix=/usr
--enable-shared --with-system-zlib --libexecdir=/usr/lib
--without-included-gettext --enable-threads=posix --enable-nls
--with-gxx-include-dir=/usr/include/c++/4.2 --program-suffix=-4.2
--enable-clocale=gnu --enable-libstdcxx-debug --enable-objc-gc --enable-mpfr
--enable-checking=release --build=x86_64-linux-gnu --host=x86_64-linux-gnu
--target=x86_64-linux-gnu
Thread model: posix
gcc version 4.2.4 (Ubuntu 4.2.4-1ubuntu3)

Still, the result is the same:
$ make CNF=gcc
scons -j 2 CNF=gcc MODE= BACKEND= CVMLCPP_PREFIX=/user/l1/beekhof/
scons: Reading SConscript files ...
scons: done reading SConscript files.
scons: Building targets ...
g++ -o main.o -c -pipe -I. -Wall -O3 -funroll-loops -fopenmp -march=native
-ftree-vectorize -DNDEBUG -I/user/l1/beekhof//include/ main.cpp
g++ -o tinystr.o -c -pipe -I. -Wall -O3 -funroll-loops -fopenmp -march=native
-ftree-vectorize -DNDEBUG -I/user/l1/beekhof//include/ tinystr.cpp
g++ -o tinyxml.o -c -pipe -I. -Wall -O3 -funroll-loops -fopenmp -march=native
-ftree-vectorize -DNDEBUG -I/user/l1/beekhof//include/ tinyxml.cpp
g++ -o tinyxmlerror.o -c -pipe -I. -Wall -O3 -funroll-loops -fopenmp
-march=native -ftree-vectorize -DNDEBUG -I/user/l1/beekhof//include/
tinyxmlerror.cpp
g++ -o tinyxmlparser.o -c -pipe -I. -Wall -O3 -funroll-loops -fopenmp
-march=native -ftree-vectorize -DNDEBUG -I/user/l1/beekhof//include/
tinyxmlparser.cpp
g++ -o shapes -fopenmp -L/user/l1/beekhof//lib/ tinystr.o tinyxml.o
tinyxmlerror.o tinyxmlparser.o main.o -lcvmlcpp -lz -lboost_iostreams
main.o: In function
`_ZN7cvmlcpp25extractSurfaceFromAdapterIN6shapes20ShapeSurfaceAdaptor_IdEEdEEvRKT_RNS_8GeometryIT0_EEd.omp_fn.9':
main.cpp:(.text+0x2af2): undefined reference to
`_ZN7cvmlcpp25extractSurfaceFromAdapterIN6shapes20ShapeSurfaceAdaptor_IdEEdEEvRKT_RNS_8GeometryIT0_EEd.omp_fn.9::C.497'
collect2: ld returned 1 exit status
scons: *** [shapes] Error 1
scons: building terminated because of errors.
make: *** [compile] Error 2

I'll see if I can make a reduced test-case...


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39573

[Bug other/39573] linking fails when optimizations are enabled

[fpbeekhof at gmail dot com]

------- Comment #2 from fpbeekhof at gmail dot com  2009-04-04 16:57 -------
The problem is the "-march=native" option. If I remove that, all is fine even
on AMD machines.

$ make CNF=gcc BACKEND=gsl
scons -j 2 CNF=gcc MODE= BACKEND=gsl CVMLCPP_PREFIX=/user/l1/beekhof/
scons: Reading SConscript files ...
scons: done reading SConscript files.
scons: Building targets ...
g++ -o main.o -c -pipe -I. -Wall -O3 -funroll-loops -fopenmp -ftree-vectorize
-DNDEBUG -DUSE_GSL -I/user/l1/beekhof//include/ main.cpp
g++ -o tinystr.o -c -pipe -I. -Wall -O3 -funroll-loops -fopenmp
-ftree-vectorize -DNDEBUG -DUSE_GSL -I/user/l1/beekhof//include/ tinystr.cpp
g++ -o tinyxml.o -c -pipe -I. -Wall -O3 -funroll-loops -fopenmp
-ftree-vectorize -DNDEBUG -DUSE_GSL -I/user/l1/beekhof//include/ tinyxml.cpp
g++ -o tinyxmlerror.o -c -pipe -I. -Wall -O3 -funroll-loops -fopenmp
-ftree-vectorize -DNDEBUG -DUSE_GSL -I/user/l1/beekhof//include/
tinyxmlerror.cpp
g++ -o tinyxmlparser.o -c -pipe -I. -Wall -O3 -funroll-loops -fopenmp
-ftree-vectorize -DNDEBUG -DUSE_GSL -I/user/l1/beekhof//include/
tinyxmlparser.cpp
g++ -o shapes -fopenmp -DUSE_GSL -L/user/l1/beekhof//lib/ tinystr.o tinyxml.o
tinyxmlerror.o tinyxmlparser.o main.o -lcvmlcpp -lz -lboost_iostreams -lgsl
-lgslcblas
scons: done building targets.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39573

[Bug target/39573] Linking fails on AMD with -march=native, works with generic x86_64

[steven at gcc dot gnu dot org]

------- Comment #3 from steven at gcc dot gnu dot org  2009-04-04 17:30 -------
I'm not sure how the -march=native option changes codegen behavior, but the
most likely suspect is the vectorizer.  Do things pass with -march=native if
you don't vectorize (needs explicit -ftree-vectorize)?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39573

[Bug target/39573] Linking fails on AMD with -march=native, works with generic x86_64

[rguenth at gcc dot gnu dot org]

------- Comment #4 from rguenth at gcc dot gnu dot org  2009-04-04 17:40 -------
Testcase?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39573

[Bug target/39573] Linking fails on AMD with -march=native, works with generic x86_64

[fpbeekhof at gmail dot com]

------- Comment #5 from fpbeekhof at gmail dot com  2009-04-04 17:45 -------
Created an attachment (id=17586)
 --> (http://gcc.gnu.org/bugzilla/attachment.cgi?id=17586&action=view)
preprocessed source; with a Makefile!

Everything fails perfectly:
$ make clean; make
rm -f *.o *~ shapes
gcc -pipe -I. -Wall -O3 -funroll-loops -fopenmp -march=native -ftree-vectorize 
 -c -o mainpp.o mainpp.cc
gcc -pipe -I. -Wall -O3 -funroll-loops -fopenmp -march=native -ftree-vectorize 
 -c -o tinystrpp.o tinystrpp.cc
gcc -pipe -I. -Wall -O3 -funroll-loops -fopenmp -march=native -ftree-vectorize 
 -c -o tinyxmlpp.o tinyxmlpp.cc
gcc -pipe -I. -Wall -O3 -funroll-loops -fopenmp -march=native -ftree-vectorize 
 -c -o tinyxmlerrorpp.o tinyxmlerrorpp.cc
gcc -pipe -I. -Wall -O3 -funroll-loops -fopenmp -march=native -ftree-vectorize 
 -c -o tinyxmlparserpp.o tinyxmlparserpp.cc
gcc -pipe -I. -Wall -O3 -funroll-loops -fopenmp -march=native -ftree-vectorize 
 -c -o SurfaceExtractorpp.o SurfaceExtractorpp.cc
gcc -o shapes -pipe -I. -Wall -O3 -funroll-loops -fopenmp -march=native
-ftree-vectorize -lz -lboost_iostreams -lgsl -lgslcblas mainpp.o tinystrpp.o
tinyxmlpp.o tinyxmlerrorpp.o tinyxmlparserpp.o SurfaceExtractorpp.o
mainpp.o: In function
`_ZN7cvmlcpp25extractSurfaceFromAdapterIN6shapes20ShapeSurfaceAdaptor_IdEEdEEvRKT_RNS_8GeometryIT0_EEd.omp_fn.8':
mainpp.cc:(.text+0x5641): undefined reference to `void
cvmlcpp::extractSurfaceFromAdapter<shapes::ShapeSurfaceAdaptor_<double>,
double>(shapes::ShapeSurfaceAdaptor_<double> const&,
cvmlcpp::Geometry<double>&, double)::C.907'
collect2: ld returned 1 exit status
make: *** [shapes] Error 1

Oh, and this is the CPU:
$ cat /proc/cpuinfo 
processor       : 0
vendor_id       : AuthenticAMD
cpu family      : 15
model           : 47
model name      : AMD Athlon(tm) 64 Processor 3200+
stepping        : 0
cpu MHz         : 1000.000
cache size      : 512 KB
fpu             : yes
fpu_exception   : yes
cpuid level     : 1
wp              : yes
flags           : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov
pat pse36 clflush mmx fxsr sse sse2 syscall nx mmxext fxsr_opt lm 3dnowext
3dnow up rep_good nopl pni lahf_lm
bogomips        : 2002.51
TLB size        : 1024 4K pages
clflush size    : 64
cache_alignment : 64
address sizes   : 40 bits physical, 48 bits virtual
power management: ts fid vid ttp tm stc


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39573

[Bug target/39573] Linking fails on AMD with -march=native, works with generic x86_64

[rguenth at gcc dot gnu dot org]

------- Comment #6 from rguenth at gcc dot gnu dot org  2009-04-04 17:47 -------
So, does it reproduce with -march=athlon64 then?  (or any specific -march?)


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39573

[Bug target/39573] Linking fails on AMD with -march=native, works with generic x86_64

[fpbeekhof at gmail dot com]

------- Comment #7 from fpbeekhof at gmail dot com  2009-04-04 18:02 -------
I tried with "-march=native" but without "-ftree-vectorize", it still fails.
I tried with "-march=athlon64" but without "-ftree-vectorize", it still fails.
On an Intel machine, with -march=native and -ftree-vectorize, it is fine.
cpu family      : 6
model           : 15
model name      : Intel(R) Core(TM)2 CPU          6400  @ 2.13GHz
stepping        : 2


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39573

[Bug target/39573] Linking fails on AMD with -march=native, works with generic x86_64

[rguenth at gcc dot gnu dot org]

------- Comment #8 from rguenth at gcc dot gnu dot org  2009-04-04 18:03 -------
Btw, with -O3 the vectorizer is on by default, so you need -fno-tree-vectorize
to turn it off.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39573

[Bug target/39573] Linking fails on AMD with -march=native, works with generic x86_64

[fpbeekhof at gmail dot com]

------- Comment #9 from fpbeekhof at gmail dot com  2009-04-04 18:13 -------
Subject: Re:  Linking fails on AMD with -march=native, works
 with generic x86_64

rguenth at gcc dot gnu dot org wrote:
> ------- Comment #8 from rguenth at gcc dot gnu dot org  2009-04-04 18:03 -------
> Btw, with -O3 the vectorizer is on by default, so you need -fno-tree-vectorize
> to turn it off.

(Oh...)
I tried with "-march=native" and "-fno-tree-vectorize", it still fails:

gccbug39573$ make clean; make
rm -f *.o *~ shapes
gcc -pipe -I. -Wall -O3 -funroll-loops -fopenmp -march=native 
-fno-tree-vectorize   -c -o mainpp.o mainpp.cc
gcc -pipe -I. -Wall -O3 -funroll-loops -fopenmp -march=native 
-fno-tree-vectorize   -c -o tinystrpp.o tinystrpp.cc
gcc -pipe -I. -Wall -O3 -funroll-loops -fopenmp -march=native 
-fno-tree-vectorize   -c -o tinyxmlpp.o tinyxmlpp.cc
gcc -pipe -I. -Wall -O3 -funroll-loops -fopenmp -march=native 
-fno-tree-vectorize   -c -o tinyxmlerrorpp.o tinyxmlerrorpp.cc
gcc -pipe -I. -Wall -O3 -funroll-loops -fopenmp -march=native 
-fno-tree-vectorize   -c -o tinyxmlparserpp.o tinyxmlparserpp.cc
gcc -pipe -I. -Wall -O3 -funroll-loops -fopenmp -march=native 
-fno-tree-vectorize   -c -o SurfaceExtractorpp.o SurfaceExtractorpp.cc
gcc -o shapes -pipe -I. -Wall -O3 -funroll-loops -fopenmp -march=native 
-fno-tree-vectorize -lz -lboost_iostreams -lgsl -lgslcblas mainpp.o 
tinystrpp.o tinyxmlpp.o tinyxmlerrorpp.o tinyxmlparserpp.o 
SurfaceExtractorpp.o
mainpp.o: In function 
`_ZN7cvmlcpp25extractSurfaceFromAdapterIN6shapes20ShapeSurfaceAdaptor_IdEEdEEvRKT_RNS_8GeometryIT0_EEd.omp_fn.8':
mainpp.cc:(.text+0x5688): undefined reference to `void 
cvmlcpp::extractSurfaceFromAdapter<shapes::ShapeSurfaceAdaptor_<double>, 
double>(shapes::ShapeSurfaceAdaptor_<double> const&, 
cvmlcpp::Geometry<double>&, double)::C.907'
collect2: ld returned 1 exit status
make: *** [shapes] Error 1

On an
vendor_id       : AuthenticAMD
cpu family      : 15
model           : 47
model name      : AMD Athlon(tm) 64 Processor 3200+
stepping        : 0

As well as on an
processor       : 31
vendor_id       : AuthenticAMD
cpu family      : 16
model           : 2
model name      : Quad-Core AMD Opteron(tm) Processor 8356
stepping        : 3


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39573

[Bug target/39573] Linking fails on AMD with -march=native, works with generic x86_64

[steven at gcc dot gnu dot org]

------- Comment #10 from steven at gcc dot gnu dot org  2009-04-04 18:41 -------
*sigh* I wanted to say in comment #3 that you need explitit
-f*no*-tree-vectorize and got it wrong :-)


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39573

[Bug target/39573] Linking fails on AMD with -march=native, works with generic x86_64

[fpbeekhof at gmail dot com]

------- Comment #11 from fpbeekhof at gmail dot com  2009-04-04 19:50 -------
No problem. I did a few more tests...

In short, it's the combination of -fopenmp and -march=native that makes fail on
AMD64.

Note: To test without openMP you can't use the preprocessed source.

# openmp and march=native --> FAIL
$ make clean; make CNF=gcc BACKEND=gsl
....
g++ -o tinystr.o -c -pipe -I. -Wall -fopenmp -march=native -DNDEBUG -DUSE_GSL
-I/user/l1/beekhof//include/ tinystr.cpp
g++ -o main.o -c -pipe -I. -Wall -fopenmp -march=native -DNDEBUG -DUSE_GSL
-I/user/l1/beekhof//include/ main.cpp
g++ -o tinyxml.o -c -pipe -I. -Wall -fopenmp -march=native -DNDEBUG -DUSE_GSL
-I/user/l1/beekhof//include/ tinyxml.cpp
g++ -o tinyxmlerror.o -c -pipe -I. -Wall -fopenmp -march=native -DNDEBUG
-DUSE_GSL -I/user/l1/beekhof//include/ tinyxmlerror.cpp
g++ -o tinyxmlparser.o -c -pipe -I. -Wall -fopenmp -march=native -DNDEBUG
-DUSE_GSL -I/user/l1/beekhof//include/ tinyxmlparser.cpp
g++ -o shapes -fopenmp -DUSE_GSL -L/user/l1/beekhof//lib/ tinystr.o tinyxml.o
tinyxmlerror.o tinyxmlparser.o main.o -lcvmlcpp -lz -lboost_iostreams -lgsl
-lgslcblas
main.o: In function
`_ZN7cvmlcpp25extractSurfaceFromAdapterIN6shapes20ShapeSurfaceAdaptor_IdEEdEEvRKT_RNS_8GeometryIT0_EEd.omp_fn.9':
main.cpp:(.text+0x11cc): undefined reference to
`_ZN7cvmlcpp25extractSurfaceFromAdapterIN6shapes20ShapeSurfaceAdaptor_IdEEdEEvRKT_RNS_8GeometryIT0_EEd.omp_fn.9::C.467'
collect2: ld returned 1 exit status
scons: *** [shapes] Error 1
scons: building terminated because of errors.
make: *** [compile] Error 2

# No march=native --> FINE
$ make clean; make CNF=gcc BACKEND=gsl
....
g++ -o tinystr.o -c -pipe -I. -Wall -fopenmp -DNDEBUG -DUSE_GSL
-I/user/l1/beekhof//include/ tinystr.cpp
g++ -o main.o -c -pipe -I. -Wall -fopenmp -DNDEBUG -DUSE_GSL
-I/user/l1/beekhof//include/ main.cpp
g++ -o tinyxml.o -c -pipe -I. -Wall -fopenmp -DNDEBUG -DUSE_GSL
-I/user/l1/beekhof//include/ tinyxml.cpp
g++ -o tinyxmlerror.o -c -pipe -I. -Wall -fopenmp -DNDEBUG -DUSE_GSL
-I/user/l1/beekhof//include/ tinyxmlerror.cpp
g++ -o tinyxmlparser.o -c -pipe -I. -Wall -fopenmp -DNDEBUG -DUSE_GSL
-I/user/l1/beekhof//include/ tinyxmlparser.cpp
g++ -o shapes -fopenmp -DUSE_GSL -L/user/l1/beekhof//lib/ tinystr.o tinyxml.o
tinyxmlerror.o tinyxmlparser.o main.o -lcvmlcpp -lz -lboost_iostreams -lgsl
-lgslcblas
scons: done building targets.


# No openmp --> FINE
$ make clean; make CNF=gcc BACKEND=gsl
....
g++ -o main.o -c -pipe -I. -Wall -march=native -DNDEBUG -DUSE_GSL
-I/user/l1/beekhof//include/ main.cpp
g++ -o tinystr.o -c -pipe -I. -Wall -march=native -DNDEBUG -DUSE_GSL
-I/user/l1/beekhof//include/ tinystr.cpp
g++ -o tinyxml.o -c -pipe -I. -Wall -march=native -DNDEBUG -DUSE_GSL
-I/user/l1/beekhof//include/ tinyxml.cpp
g++ -o tinyxmlerror.o -c -pipe -I. -Wall -march=native -DNDEBUG -DUSE_GSL
-I/user/l1/beekhof//include/ tinyxmlerror.cpp
g++ -o tinyxmlparser.o -c -pipe -I. -Wall -march=native -DNDEBUG -DUSE_GSL
-I/user/l1/beekhof//include/ tinyxmlparser.cpp
g++ -o shapes -DUSE_GSL -L/user/l1/beekhof//lib/ tinystr.o tinyxml.o
tinyxmlerror.o tinyxmlparser.o main.o -lcvmlcpp -lz -lboost_iostreams -lgsl
-lgslcblas
scons: done building targets.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39573

[Bug target/39573] Linking fails on AMD with -march=native, works with generic x86_64

[rguenth at gcc dot gnu dot org]

------- Comment #12 from rguenth at gcc dot gnu dot org  2009-04-04 20:00 -------
I do not see that undefined symbol required, but it _does_ print strangely

_ZN7cvmlcpp25extractSurfaceFromAdapterIN6shapes20ShapeSurfaceAdaptor_IdEEdEEvRKT_RNS_8GeometryIT0_EEd.omp_fn.9::C.467

it suggests that this was sort-of "demangled"


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39573

[Bug target/39573] Linking fails on AMD with -march=native, works with generic x86_64

[rguenth at gcc dot gnu dot org]

------- Comment #13 from rguenth at gcc dot gnu dot org  2009-04-04 20:03 -------
What does

nm mainpp.o | grep
_ZN7cvmlcpp25extractSurfaceFromAdapterIN6shapes20ShapeSurfaceAdaptor_IdEEdEEvRKT_RNS_8GeometryIT0_EEd

print for you?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39573

[Bug target/39573] Linking fails on AMD with -march=native and -fopenmp, works with generic x86_64

[fpbeekhof at gmail dot com]

------- Comment #14 from fpbeekhof at gmail dot com  2009-04-04 20:14 -------
For simplicty, in the preprocessed source, edit the Makefile, set
CXXFLAGS= -pipe -I. -Wall -fopenmp -march=native

Then,
gccbug39573$ make clean; make
rm -f *.o *~ shapes
gcc -pipe -I. -Wall -fopenmp -march=native   -c -o mainpp.o mainpp.cc
gcc -pipe -I. -Wall -fopenmp -march=native   -c -o tinystrpp.o tinystrpp.cc
gcc -pipe -I. -Wall -fopenmp -march=native   -c -o tinyxmlpp.o tinyxmlpp.cc
gcc -pipe -I. -Wall -fopenmp -march=native   -c -o tinyxmlerrorpp.o
tinyxmlerrorpp.cc
gcc -pipe -I. -Wall -fopenmp -march=native   -c -o tinyxmlparserpp.o
tinyxmlparserpp.cc
gcc -pipe -I. -Wall -fopenmp -march=native   -c -o SurfaceExtractorpp.o
SurfaceExtractorpp.cc
gcc -o shapes -pipe -I. -Wall -fopenmp -march=native -lz -lboost_iostreams
-lgsl -lgslcblas mainpp.o tinystrpp.o tinyxmlpp.o tinyxmlerrorpp.o
tinyxmlparserpp.o SurfaceExtractorpp.o
mainpp.o: In function
`_ZN7cvmlcpp25extractSurfaceFromAdapterIN6shapes20ShapeSurfaceAdaptor_IdEEdEEvRKT_RNS_8GeometryIT0_EEd.omp_fn.8':
mainpp.cc:(.text+0x1f76): undefined reference to `void
cvmlcpp::extractSurfaceFromAdapter<shapes::ShapeSurfaceAdaptor_<double>,
double>(shapes::ShapeSurfaceAdaptor_<double> const&,
cvmlcpp::Geometry<double>&, double)::C.867'
collect2: ld returned 1 exit status
make: *** [shapes] Error 1

gccbug39573$ nm mainpp.o | grep
_ZN7cvmlcpp25extractSurfaceFromAdapterIN6shapes20ShapeSurfaceAdaptor_IdEEdEEvRKT_RNS_8GeometryIT0_EEd
0000000000000000 W
_ZN7cvmlcpp25extractSurfaceFromAdapterIN6shapes20ShapeSurfaceAdaptor_IdEEdEEvRKT_RNS_8GeometryIT0_EEd
00000000000019af t
_ZN7cvmlcpp25extractSurfaceFromAdapterIN6shapes20ShapeSurfaceAdaptor_IdEEdEEvRKT_RNS_8GeometryIT0_EEd.omp_fn.10
0000000000001c7e t
_ZN7cvmlcpp25extractSurfaceFromAdapterIN6shapes20ShapeSurfaceAdaptor_IdEEdEEvRKT_RNS_8GeometryIT0_EEd.omp_fn.8
0000000000001af2 t
_ZN7cvmlcpp25extractSurfaceFromAdapterIN6shapes20ShapeSurfaceAdaptor_IdEEdEEvRKT_RNS_8GeometryIT0_EEd.omp_fn.9


-- 

fpbeekhof at gmail dot com changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
            Summary|Linking fails on AMD with - |Linking fails on AMD with -
                   |march=native, works with    |march=native and -fopenmp,
                   |generic x86_64              |works with generic x86_64


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39573

[Bug target/39573] Linking fails on AMD with -march=native and -fopenmp, works with generic x86_64

[rguenth at gcc dot gnu dot org]

------- Comment #15 from rguenth at gcc dot gnu dot org  2009-04-04 20:45 -------
Ok, I can see

        movl  
$_ZZN7cvmlcpp25extractSurfaceFromAdapterIN6shapes20ShapeSurfaceAdaptor_IdEEdEEvRKT_RNS_8GeometryIT0_EEdE5C.907,
%esi
        movl    $12, %ecx
        movq    56(%rsp), %rdi
        rep movsq
        movl    %r14d, %esi

but that's not on an omp thing.

Expanded from

;; offset = C.907
(insn 2176 2175 2177 /usr/local/include/cvmlcpp/volume/SurfaceExtractor:150
(parallel [
            (set (reg:DI 999)
                (plus:DI (reg/f:DI 54 virtual-stack-vars)
                    (const_int -832 [0xfffffffffffffcc0])))
            (clobber (reg:CC 17 flags))
        ]) -1 (nil))

(insn 2177 2176 2178 /usr/local/include/cvmlcpp/volume/SurfaceExtractor:150
(set (reg:DI 1000)
        (symbol_ref:DI
("_ZZN7cvmlcpp25extractSurfaceFromAdapterIN6shapes20ShapeSurfaceAdaptor_IdEEdEEvRKT_RNS_8GeometryIT0_EEdE5C.907")
[flags 0x2] <var_decl 0xb18423f4 C.907>)) -1 (nil))

That DECL_ASSEMBLER_NAME is bogus.

mainpp.cc.004t.gimple:                static int C.907[8][3] = {{0, 0, 0}, {1,
0, 0}, {1, 0, 1}, {0, 0, 1}, {0, 1, 0}, {1, 1, 0}, {1, 1, 1}, {0, 1, 1}};
mainpp.cc.004t.gimple:                                    offset = C.907;

huh.  Gimplified from

    int offset[8][3] = {{0, 0, 0}, {1, 0, 0}, {1, 0, 1}, {0, 0, 1}, {0, 1, 0},
{1, 1, 0}, {1, 1, 1}, {0, 1, 1}};

this sounds like related to PR38615.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39573

[Bug target/39573] Linking fails on AMD with -march=native and -fopenmp, works with generic x86_64

[fpbeekhof at gmail dot com]

------- Comment #16 from fpbeekhof at gmail dot com  2009-04-05 08:16 -------

Ok, this was the original code that fails
// Prepare points of cube relative to point in matrix.
typename iPoint3D::value_type offset [8] [3] = {
        {0, 0, 0}, {1, 0, 0}, {1, 0, 1}, {0, 0, 1},
        {0, 1, 0}, {1, 1, 0}, {1, 1, 1}, {0, 1, 1} };

After adding "const", it is fine!
// Prepare points of cube relative to point in matrix.
const typename iPoint3D::value_type offset [8] [3] = {
        {0, 0, 0}, {1, 0, 0}, {1, 0, 1}, {0, 0, 1},
        {0, 1, 0}, {1, 1, 0}, {1, 1, 1}, {0, 1, 1} };


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39573

[Bug target/39573] Linking fails on AMD with -march=native and -fopenmp, works with generic x86_64

[jakub at gcc dot gnu dot org]

------- Comment #17 from jakub at gcc dot gnu dot org  2009-04-06 08:59 -------
Smaller self-contained testcase:
int z;

void __attribute__((noinline))
bar (int *x)
{
  #pragma omp atomic
    z += x[2];
  x[2] += x[3];
}

int
main ()
{
  int i;
#pragma omp parallel for
  for (i = 0; i < 65536; i++)
    {
      int x[] =
        {
          0, 0, 0, 1, 0, 0, 1, 0, 1, 0, 0, 1, 0, 1, 0, 1, 1, 0, 1, 1, 1, 0, 1,
1,
          0, 0, 0, 1, 0, 0, 1, 0, 1, 0, 0, 1, 0, 1, 0, 1, 1, 0, 1, 1, 1, 0, 1,
1,
          0, 0, 0, 1, 0, 0, 1, 0, 1, 0, 0, 1, 0, 1, 0, 1, 1, 0, 1, 1, 1, 0, 1,
1,
          0, 0, 0, 1, 0, 0, 1, 0, 1, 0, 0, 1, 0, 1, 0, 1, 1, 0, 1, 1, 1, 0, 1,
1,
          0, 0, 0, 1, 0, 0, 1, 0, 1, 0, 0, 1, 0, 1, 0, 1, 1, 0, 1, 1, 1, 0, 1,
1,
          0, 0, 0, 1, 0, 0, 1, 0, 1, 0, 0, 1, 0, 1, 0, 1, 1, 0, 1, 1, 1, 0, 1,
1,
          0, 0, 0, 1, 0, 0, 1, 0, 1, 0, 0, 1, 0, 1, 0, 1, 1, 0, 1, 1, 1, 0, 1,
1,
          0, 0, 0, 1, 0, 0, 1, 0, 1, 0, 0, 1, 0, 1, 0, 1, 1, 0, 1, 1, 1, 0, 1,
1,
          0, 0, 0, 1, 0, 0, 1, 0, 1, 0, 0, 1, 0, 1, 0, 1, 1, 0, 1, 1, 1, 0, 1,
1,
          0, 0, 0, 1, 0, 0, 1, 0, 1, 0, 0, 1, 0, 1, 0, 1, 1, 0, 1, 1, 1, 0, 1,
1,
          0, 0, 0, 1, 0, 0, 1, 0, 1, 0, 0, 1, 0, 1, 0, 1, 1, 0, 1, 1, 1, 0, 1,
1,
          0, 0, 0, 1, 0, 0, 1, 0, 1, 0, 0, 1, 0, 1, 0, 1, 1, 0, 1, 1, 1, 0, 1,
1,
          0, 0, 0, 1, 0, 0, 1, 0, 1, 0, 0, 1, 0, 1, 0, 1, 1, 0, 1, 1, 1, 0, 1,
1,
        };
      bar (x);
    }
}

I've increased the size of the array to make it fail regardless of the -mtune=
setting.


-- 

jakub at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         AssignedTo|unassigned at gcc dot gnu   |jakub at gcc dot gnu dot org
                   |dot org                     |
             Status|UNCONFIRMED                 |ASSIGNED
     Ever Confirmed|0                           |1
           Keywords|                            |openmp
   Last reconfirmed|0000-00-00 00:00:00         |2009-04-06 08:59:25
               date|                            |


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39573

[Bug middle-end/39573] Linking fails on AMD with -march=native and -fopenmp, works with generic x86_64

[jakub at gcc dot gnu dot org]

------- Comment #18 from jakub at gcc dot gnu dot org  2009-04-08 20:05 -------
Subject: Bug 39573

Author: jakub
Date: Wed Apr  8 20:04:45 2009
New Revision: 145772

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=145772
Log:
        PR middle-end/39573
        * omp-low.c (expand_omp_taskreg): Finalize taskreg static local_decls
        variables.

        * libgomp.c++/pr39573.C: New test.

Added:
    trunk/libgomp/testsuite/libgomp.c++/pr39573.C
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/omp-low.c
    trunk/libgomp/ChangeLog


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39573

[Bug middle-end/39573] Linking fails on AMD with -march=native and -fopenmp, works with generic x86_64

[jakub at gcc dot gnu dot org]

------- Comment #19 from jakub at gcc dot gnu dot org  2009-04-08 20:13 -------
Subject: Bug 39573

Author: jakub
Date: Wed Apr  8 20:13:26 2009
New Revision: 145773

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=145773
Log:
        PR middle-end/39573
        * omp-low.c (expand_omp_taskreg): Finalize taskreg static local_decls
        variables.

        * libgomp.c++/pr39573.C: New test.

Added:
    branches/gcc-4_4-branch/libgomp/testsuite/libgomp.c++/pr39573.C
      - copied unchanged from r145772,
trunk/libgomp/testsuite/libgomp.c++/pr39573.C
Modified:
    branches/gcc-4_4-branch/gcc/ChangeLog
    branches/gcc-4_4-branch/gcc/omp-low.c
    branches/gcc-4_4-branch/libgomp/ChangeLog


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39573

[Bug middle-end/39573] Linking fails on AMD with -march=native and -fopenmp, works with generic x86_64

[jakub at gcc dot gnu dot org]

------- Comment #20 from jakub at gcc dot gnu dot org  2009-04-08 20:14 -------
Fixed for 4.4/4.5+ so far.


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39573

[Bug middle-end/39573] Linking fails on AMD with -march=native and -fopenmp, works with generic x86_64

[pinskia at gcc dot gnu dot org]

------- Comment #21 from pinskia at gcc dot gnu dot org  2009-04-28 23:41 -------
Fixed.


-- 

pinskia at gcc dot gnu dot org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|                            |FIXED
   Target Milestone|---                         |4.4.0


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=39573