I'm using GMP 4.2 and MPFR 2.2.0, both built with gcc-4.0.2. When I go to build gcc-4.0.3, I get the following error: /afs/ir.stanford.edu/src/pubsw/languages/gcc-build/sun4x_59/gcc/gfortran -B/afs/ir.stanford.edu/src/pubsw/languages/gcc-build/sun4x_59/gcc/ -B/usr/pubsw/sparc-sun-solaris2.9/bin/ -B/usr/pubsw/sparc-sun-solaris2.9/lib/ -isystem /usr/pubsw/sparc-sun-solaris2.9/include -isystem /usr/pubsw/sparc-sun-solaris2.9/sys-include -I . -Wall -fno-repack-arrays -fno-underscoring -c ../../../../gcc-4.0.3/libgfortran/intrinsics/selected_int_kind.f90 -fPIC -DPIC -o .libs/selected_int_kind.o ld.so.1: f951: fatal: /afs/ir.stanford.edu/src/pubsw/languages/gcc-build/sun4x_59/gcc/f951: object requires TLS, but TLS failed to initialize gfortran: Internal error: Killed (program f951) Please submit a full bug report. See <URL:http://gcc.gnu.org/bugs.html> for instructions.
change to fortran, instead of libfortran.
On Solaris 8, the build fails also, except with the following error: /afs/ir.stanford.edu/src/pubsw/languages/gcc-build/sun4x_58/gcc/gfortran -B/afs/ir.stanford.edu/src/pubsw/languages/gcc-build/sun4x_58/gcc/ -B/usr/pubsw/sparc-sun-solaris2.8/bin/ -B/usr/pubsw/sparc-sun-solaris2.8/lib/ -isystem /usr/pubsw/sparc-sun-solaris2.8/include -isystem /usr/pubsw/sparc-sun-solaris2.8/sys-include -I . -Wall -fno-repack-arrays -fno-underscoring -c ../../../../gcc-4.0.3/libgfortran/intrinsics/selected_int_kind.f90 -fPIC -DPIC -o .libs/selected_int_kind.o <built-in>:0: internal compiler error: Segmentation Fault
Very likely a GMP/MFPR problem, try to disable TLS support in the library.
Disable TLS support in *which* library? GMP, MPFR, or libfortran? There are no options in the configure for GMP or MPFR referring to TLS/SSL. --Quanah
> Disable TLS support in *which* library? GMP/MPFR. > There are no options in the configure for GMP or MPFR referring to TLS/SSL. Did you build gcc-4.0.2 with Fortran support? If so, could you try building 4.0.3 with the same version of GMP/MPFR as with 4.0.2? In the meantime, I'll try to reproduce the problem.
I don't see a way to disable "TLS" support in GMP or MPFR, but I'll keep poking. I didn't build gcc 4.0.2 with Fortran support because I needed to build gcc to build GMP and MPFR. ;) Building gcc 4.0.1 also fails for me, but with an entirely different error and location. I'll open a separate bug on that.
> I didn't build gcc 4.0.2 with Fortran support because I needed to build gcc to > build GMP and MPFR. ;) Of course. :-) I just rebuild 4.0.3 with 4.0.2 on SPARC/Solaris 8: gax% CC=gcc40 /home/eric/svn/gcc-4_0-branch/configure --prefix=/opt/build/eric/local/gcc-4_0-branch --with-gmp=/opt/build/eric/local --with-mpfr=/opt/build/eric/local --with-local-prefix=/opt/build/eric/local --enable-languages=c,f95 gax% gcc/xgcc -v Using built-in specs. Target: sparc-sun-solaris2.8 Configured with: /home/eric/svn/gcc-4_0-branch/configure --prefix=/opt/build/eric/local/gcc-4_0-branch --with-gmp=/opt/build/eric/local --with-mpfr=/opt/build/eric/local --with-local-prefix=/opt/build/eric/local --enable-languages=c,f95 Thread model: posix gcc version 4.0.3 > Building gcc 4.0.1 also fails for me, but with an entirely different error and > location. I'll open a separate bug on that. Really weird. GCC 4.0.x is known to build without a hitch on SPARC/Solaris. Please specify the exact configure line and the exact make line.
sorry, I meant 4.1.0, and you answered that well enough. ;) *how* do I build gmp/mpfr without TLS? I don't see any configure options that make sense for this... I'll note that GMP was built with ABI=32 in my case.
> I'll note that GMP was built with ABI=32 in my case. I'd recommend to configure it as specified in the instructions: http://gcc.gnu.org/install/specific.html#sparc-sun-solaris2 And you also need to make sure the resulting library is visible from the dynamic linker, i.e. 'ldd f951' should not have unresolved references.
I rebuilt gmp with: --build=sparc-sun-solaris2.8 and then rebuilt MPFR, and then rebuilt gcc 4.0.3. On Solaris 8, it is still dying, with: /afs/ir.stanford.edu/src/pubsw/languages/gcc-build/sun4x_58/gcc/gfortran -B/afs/ir.stanford.edu/src/pubsw/languages/gcc-build/sun4x_58/gcc/ -B/usr/pubsw/sparc-sun-solaris2.8/bin/ -B/usr/pubsw/sparc-sun-solaris2.8/lib/ -isystem /usr/pubsw/sparc-sun-solaris2.8/include -isystem /usr/pubsw/sparc-sun-solaris2.8/sys-include -I . -Wall -fno-repack-arrays -fno-underscoring -c ../../../../gcc-4.0.3/libgfortran/intrinsics/selected_int_kind.f90 -fPIC -DPIC -o .libs/selected_int_kind.o <built-in>:0: internal compiler error: Segmentation Fault Please submit a full bug report, with preprocessed source if appropriate. See <URL:http://gcc.gnu.org/bugs.html> for instructions. make[2]: *** [selected_int_kind.lo] Error 1 make[2]: Leaving directory `/afs/ir.stanford.edu/src/pubsw/languages/gcc-build/sun4x_58/sparc-sun-solaris2.8/libgfortran' make[1]: *** [all] Error 2 make[1]: Leaving directory `/afs/ir.stanford.edu/src/pubsw/languages/gcc-build/sun4x_58/sparc-sun-solaris2.8/libgfortran' make: *** [all-target-libgfortran] Error 2 ldd shows no issues with f951 or gfortran: solaris8-build:/afs/ir/src/pubsw/languages/gcc-4.0.3> ldd ../gcc-build/sun4x_58/gcc/gfortran libintl.so.3 => /usr/pubsw/lib/libintl.so.3 libiconv.so.2 => /usr/pubsw/lib/libiconv.so.2 libc.so.1 => /usr/lib/libc.so.1 libgcc_s.so.1 => /usr/pubsw/lib/libgcc_s.so.1 libdl.so.1 => /usr/lib/libdl.so.1 /usr/platform/SUNW,UltraAX-i2/lib/libc_psr.so.1 solaris8-build:/afs/ir/src/pubsw/languages/gcc-4.0.3> ldd ../gcc-build/sun4x_58/gcc/f951 libgmp.so.3 => /usr/pubsw/lib/libgmp.so.3 libintl.so.3 => /usr/pubsw/lib/libintl.so.3 libiconv.so.2 => /usr/pubsw/lib/libiconv.so.2 libc.so.1 => /usr/lib/libc.so.1 libgcc_s.so.1 => /usr/pubsw/lib/libgcc_s.so.1 libdl.so.1 => /usr/lib/libdl.so.1 /usr/platform/SUNW,UltraAX-i2/lib/libc_psr.so.1 I'm guessing Solaris 9 will still fail, too. It just hasn't gotten as far as Solaris 8 yet.
> I rebuilt gmp with: > > --build=sparc-sun-solaris2.8 > > and then rebuilt MPFR, and then rebuilt gcc 4.0.3. On Solaris 8, it is still > dying, with: Please post the exact configure line and command line.
For GMP (4.1.4 and 4.2, I've tried both versions): This file contains any messages produced by compilers while running configure, to aid debugging if configure makes a mistake. It was created by configure, which was generated by GNU Autoconf 2.52. Invocation command line was $ ../../configure --datadir=${prefix}/lib --libexecdir=${prefix}/lib --sharedstatedir=${prefix}/lib --prefix=/usr/pubsw --build=sparc-sun-solaris2.8 For MPFR: This file contains any messages produced by compilers while running configure, to aid debugging if configure makes a mistake. It was created by MPFR configure 2.2.0, which was generated by GNU Autoconf 2.59. Invocation command line was $ ../../configure --datadir=${prefix}/lib --libexecdir=${prefix}/lib --sharedstatedir=${prefix}/lib --prefix=/usr/pubsw --enable-thread-safe --with-gnu-ld --with-gmp=/usr/pubsw For GCC: # The gcc driver likes to know the arguments it was configured with. TOPLEVEL_CONFIGURE_ARGUMENTS=../../gcc-4.0.3/configure --datadir=$${prefix}/lib --libexecdir=$${prefix}/lib --sharedstatedir=$${prefix}/lib --prefix=/usr/pub sw --enable-threads --with-gnu-as --with-as=/usr/pubsw/bin/as --with-gnu-ld --with-ld=/usr/pubsw/bin/ld --with-libintl-prefix=/usr/pubsw --with-libiconv-prefix=/usr/pubsw --enable-libada --enable-languages=c,c++,f95,java,objc,ada --Quanah
And the command line for the compiler?
For building GCC, I execute: cd ../gcc-build/@sys && make CFLAGS='-O' \ LIBCFLAGS='-g -O2' LIBCXXFLAGS='-g -O2 -fno-implicit-templates' 2>&1 | tee -a makeLOG (@sys is an AFS thing, it translates to sun4x_58 or sun4x_59 in this case, depending on the OS).
> For building GCC, I execute: > > cd ../gcc-build/@sys && make CFLAGS='-O' \ > LIBCFLAGS='-g -O2' LIBCXXFLAGS='-g -O2 -fno-implicit-templates' 2>&1 | tee > -a makeLOG Use 'make bootstrap' instead of 'make'.
I will try "make bootstrap", but I note: "make" worked fine with gcc-3.4.x, gcc-3.3.x, gcc-2.9x etc. And for i386 and amd64 it worked just fine, too, for gcc 4.0.3 with these same versions of GMP and MPFR to build fortran. Is this a solaris specific issue?
> "make" worked fine with gcc-3.4.x, gcc-3.3.x, gcc-2.9x etc. Yes, but it's wrong up to (and including) 4.1.x. We only guarantee that the compiler correctly compiles itself ("bootstraps"), not that it is correctly compiled by other compilers at random optimization levels. > Is this a solaris specific issue? No, this has always been documented: @section Building a native compiler For a native build issue the command @samp{make bootstrap}. This will build the entire GCC system, which includes the following steps: @itemize @bullet @item Build host tools necessary to build the compiler such as texinfo, bison, gperf. @item Build target tools for use by the compiler such as binutils (bfd, binutils, gas, gprof, ld, and opcodes) if they have been individually linked or moved into the top level GCC source tree before configuring. @item Perform a 3-stage bootstrap of the compiler. @item Perform a comparison test of the stage2 and stage3 compilers. @item Build runtime libraries using the stage3 compiler from the previous step. @end itemize
Unsuprisingly,the build still fails after changing it to use "make boostrap" in the same exact way: /afs/ir.stanford.edu/src/pubsw/languages/gcc-build/sun4x_58/gcc/gfortran -B/afs/ir.stanford.edu/src/pubsw/languages/gcc-build/sun4x_58/gcc/ -B/usr/pubsw/sparc-sun-solaris2.8/bin/ -B/usr/pubsw/sparc-sun-solaris2.8/lib/ -isystem /usr/pubsw/sparc-sun-solaris2.8/include -isystem /usr/pubsw/sparc-sun-solaris2.8/sys-include -I . -Wall -fno-repack-arrays -fno-underscoring -c ../../../../gcc-4.0.3/libgfortran/intrinsics/selected_int_kind.f90 -fPIC -DPIC -o .libs/selected_int_kind.o <built-in>:0: internal compiler error: Segmentation Fault Please submit a full bug report, with preprocessed source if appropriate. See <URL:http://gcc.gnu.org/bugs.html> for instructions. make[3]: *** [selected_int_kind.lo] Error 1 make[3]: Leaving directory `/afs/ir.stanford.edu/src/pubsw/languages/gcc-build/sun4x_58/sparc-sun-solaris2.8/libgfortran' make[2]: *** [all] Error 2 make[2]: Leaving directory `/afs/ir.stanford.edu/src/pubsw/languages/gcc-build/sun4x_58/sparc-sun-solaris2.8/libgfortran' make[1]: *** [all-target-libgfortran] Error 2 make[1]: Leaving directory `/afs/ir.stanford.edu/src/pubsw/languages/gcc-build/sun4x_58' make: *** [bootstrap] Error 2
(In reply to comment #18) > Unsuprisingly,the build still fails after changing it to > use "make boostrap" in the same exact way: > > ../../../../gcc-4.0.3/libgfortran/intrinsics/selected_int_kind.f90 -fPIC I've scanned this thread and I have a few comments. 1) GMP/MPFR are used only by the gfortran compiler (ie not the runtime library). To avoid the TLS problems, you need only build and install the static libraries for GMP/MPFR. 2) You really, really, really want to use at least the 4.1 tree for gfortran. There are literally 100s of bugs fixed in 4.1 and beyond that will never be backported to 4.0.3.
> Unsuprisingly,the build still fails after changing it to use "make boostrap" > in the same exact way: Maybe "unsurprisingly", but this eliminates a whole category of problems. Could you remove --with-libintl-prefix=/usr/pubsw --with-libiconv-prefix=/usr/pubsw from the configure line and add --disable-nls instead?
> /afs/ir.stanford.edu/src/pubsw/languages/gcc-build/sun4x_58/gcc/gfortran > -B/afs/ir.stanford.edu/src/pubsw/languages/gcc-build/sun4x_58/gcc/ > -B/usr/pubsw/sparc-sun-solaris2.8/bin/ -B/usr/pubsw/sparc-sun-solaris2.8/lib/ > -isystem /usr/pubsw/sparc-sun-solaris2.8/include -isystem > /usr/pubsw/sparc-sun-solaris2.8/sys-include -I . -Wall -fno-repack-arrays > -fno-underscoring -c > ../../../../gcc-4.0.3/libgfortran/intrinsics/selected_int_kind.f90 -fPIC -DPIC Could you attach this selected_int_kind.f90 file to the Bugzilla PR?
I can look at only building GMP static, rather than static & shared. I have a build of 4.1 happening on Solaris 9 at the moment after fixing the ksh issue. --Quanah
Created attachment 11144 [details] selected_int_kind.f90 selected_int_kind.f90 (it is just what is shipped with the gcc-4.0.3 source...)
I will look at removing the libintl/libiconv bits, but in the past gcc failed to find them without adding those flags.
> selected_int_kind.f90 (it is just what is shipped with the gcc-4.0.3 source...) Except that it is generated on the fly by mk-sic-ink.sh...
> Except that it is generated on the fly by mk-sic-ink.sh... Sorry, I was talking about selected_int_kind.inc... Could you attach this one?
I'll try the flags and get the attachment to you tomorrow. Today is my birthday, and my wife doesn't really want me on the computer much. ;) So far, gcc 4.1.0 looks to be building without issue.
Created attachment 11160 [details] selected_int_kind.inc Here's the selected_int_kind.inc file generated on Solaris 8 with gcc 4.0.3.
> Here's the selected_int_kind.inc file generated on Solaris 8 with gcc 4.0.3. Thanks. Victory, at last something obviously broken. :-) The file reads: integer, parameter :: c = 0 type (int_info), parameter :: int_infos(c) = (/ & while it should read: integer, parameter :: c = 4 type (int_info), parameter :: int_infos(c) = (/ & int_info (1, range(0_1)), & int_info (2, range(0_2)), & int_info (4, range(0_4)), & int_info (8, range(0_8)) /) so again the Fortran compiler either doesn't work or is not correctly invoked. In the $objdir/sparc-sun-solaris2.8/libgfortran directory, delete the file (selected_int_kind.inc) and type 'gmake selected_int_kind.inc'. You should see something like: /bin/ksh /home/eric/svn/gcc-4_0-branch/libgfortran/mk-sik-inc.sh '/opt/build/eric/gcc-4_0-branch/gcc/gfortran -B/opt/build/eric/gcc-4_0-branch/gcc/ -B/opt/build/eric/local/gcc-4_0-branch/sparc-sun-solaris2.8/bin/ -B/opt/build/eric/local/gcc-4_0-branch/sparc-sun-solaris2.8/lib/ -isystem /opt/build/eric/local/gcc-4_0-branch/sparc-sun-solaris2.8/include -isystem /opt/build/eric/local/gcc-4_0-branch/sparc-sun-solaris2.8/sys-include -I . -Wall -fno-repack-arrays -fno-underscoring-g -O2' > selected_int_kind.inc Could you try to re-play the script mk-sik-inc.sh by hand and find out when things start to go awry?
Here is what happens when I run it by hand: solaris8-build:/afs/ir/src/pubsw/languages/gcc-build/sun4x_58/sparc-sun-solaris2.8/libgfortran> /bin/ksh ../../../../gcc-4.0.3/libgfortran/mk-sik-inc.sh '/afs/ir.stanford.edu/src/pubsw/languages/gcc-build/sun4x_58/gcc/gfortran -B/afs/ir.stanford.edu/src/pubsw/languages/gcc-build/sun4x_58/gcc/ -B/usr/pubsw/sparc-sun-solaris2.8/bin/ -B/usr/pubsw/sparc-sun-solaris2.8/lib/ -isystem /usr/pubsw/sparc-sun-solaris2.8/include -isystem /usr/pubsw/sparc-sun-solaris2.8/sys-include -I . -Wall -fno-repack-arrays -fno-underscoring ' integer, parameter :: c = 0 type (int_info), parameter :: int_infos(c) = (/ & No errors or anything... It just spits out the broken bits.
> No errors or anything... It just spits out the broken bits. OK. Could you then break apart the script and run the various pieces manually?
Okay, I created the following file (as is generated by the script): solaris8-build:/tmp> cat /tmp/q1.f90 integer (kind=1) :: x end Then ran the build command on it as is done by the script: /afs/ir.stanford.edu/src/pubsw/languages/gcc-build/sun4x_58/gcc/gfortran -B/afs/ir.stanford.edu/src/pubsw/languages/gcc-build/sun4x_58/gcc/ -B/usr/pubsw/sparc-sun-solaris2.8/bin/ -B/usr/pubsw/sparc-sun-solaris2.8/lib/ -isystem /usr/pubsw/sparc-sun-solaris2.8/include -isystem /usr/pubsw/sparc-sun-solaris2.8/sys-include -I . -Wall -fno-repack-arrays -fno-underscoring /tmp/q1.f90 and I get: <built-in>:0: internal compiler error: Segmentation Fault Please submit a full bug report, with preprocessed source if appropriate. See <URL:http://gcc.gnu.org/bugs.html> for instructions. so the compiler is completely flailing on this. :(
GDB shows: GNU gdb 6.3 Copyright 2004 Free Software Foundation, Inc. GDB is free software, covered by the GNU General Public License, and you are welcome to change it and/or distribute copies of it under certain conditions. Type "show copying" to see the conditions. There is absolutely no warranty for GDB. Type "show warranty" for details. This GDB was configured as "sparc-sun-solaris2.8"... (gdb) run Starting program: /afs/ir.stanford.edu/src/pubsw/languages/gcc-build/sun4x_58/gcc/f951 Program received signal SIGSEGV, Segmentation fault. 0x0035c398 in mpfr_set_default_prec () (gdb) bt #0 0x0035c398 in mpfr_set_default_prec () #1 0x0001496c in gfc_arith_init_1 () at ../../../gcc-4.0.3/gcc/fortran/arith.c:180 #2 0x0003e730 in gfc_init_1 () at ../../../gcc-4.0.3/gcc/fortran/misc.c:287 #3 0x0005d7d4 in gfc_init () at ../../../gcc-4.0.3/gcc/fortran/f95-lang.c:292 #4 0x00281514 in toplev_main (argc=Variable "argc" is not available. ) at ../../../gcc-4.0.3/gcc/toplev.c:2019 #5 0x000118d4 in .nope () #6 0x000118d4 in .nope ()
> Program received signal SIGSEGV, Segmentation fault. > 0x0035c398 in mpfr_set_default_prec () > (gdb) bt > #0 0x0035c398 in mpfr_set_default_prec () OK. Could you ditch MPFR 2.2.0 entirely and use the MPFR library bundled with GMP 4.2? The configure line for GMP is on the Solaris build instructions page.
From the GMP 4.2 NEWS file: Mis-features: * mpfr is gone, and will from now on be released only separately. Please see www.mpfr.org.
Just to note, if I haven't already, I was able to build gcc 4.0.3 with GMP 4.2 and MPFR 2.2.0 on i386 and amd64 without a problem (including fortran), so this seems to be a problem specific to Solaris 8/9.
> From the GMP 4.2 NEWS file: > > Mis-features: > * mpfr is gone, and will from now on be released only separately. Please see > www.mpfr.org. Grumpf... Could you downgrade to 4.1.x then?
Yeah, I'll give that a shot next week (i'm now in a freeze period). I've also been poking at MPFR. There are apparently 10 or more patches now for 2.2.0, that may resolve the issues, too. I'll look at that. I've rebuilt it, and ran the "check" area for mpfr, and 115/117 tests fail, so I'm guessing there are some serious issues there with MPFR. I've also subscribed to their list, because I can see that until MPFR is no longer busted, gcc isn't going to be working. ;)
> I've also been poking at MPFR. There are apparently 10 or more patches now for > 2.2.0, that may resolve the issues, too. I'll look at that. I've rebuilt it, > and ran the "check" area for mpfr, and 115/117 tests fail. I cannot reproduce that: MPFR 2.2.0 is clean (117/117 tests OK ) if configured with './configure --build=sparc-sun-solaris2.8 --with-gmp=/opt/build/eric/local' against GMP 4.1.3 and built by GCC 4.0.2 with 'gmake' on the Solaris 8 box. GMP 4.1.3 was configured with './configure --build=sparc-sun-solaris2.8' and built by GCC 4.0.2 with 'gmake'.
This is the build with GMP 4.1.3 and MPFR 2.2.0: gax% gcc/xgcc -v Using built-in specs. Target: sparc-sun-solaris2.8 Configured with: /home/eric/svn/gcc-4_0-branch/configure --prefix=/opt/build/eric/local/gcc-4_0-branch --with-gmp=/opt/build/eric/local2 --with-mpfr=/opt/build/eric/local2 --with-local-prefix=/opt/build/eric/local2 --enable-languages=c,f95 Thread model: posix gcc version 4.0.3
I'm puzzled by this PR and PR 26893: what is the status on those reports? Did you reproduce this on a given platform, with *known good* GMP and MPFR and with a gfortran >= 4.1 ? Otherwise, I supposed we might close both: I do regularly build on gfortran on sparc-sun-solaris2.9, with no problem. Note that it doesn't mean we can't work on resolving the issue, reopening a PR if the reporter has time to investigate further.
I will be working on this after the students leave, starting in June. Please keep this ticket open. I note that the versions of GMP & MPFR I'm using work *fine* on linux, so I consider them "known good". They only don't work with Solaris.
No feedback in 3 months and gfortran is known to build just fine on sparc-solaris (using working GMP).
This entire problem stems from having: --enable-thread-safe when building MPFR on Solaris. I finally got gcc 4.0.3 to build gfortran for me last week after fixing that configure line to MPFR.