| Message ID | 20110216135154.GV30899@tyan-ft48-01.lab.bos.redhat.com |
|---|---|
| State | New |
| Headers | show |
On 02/16/2011 02:51 PM, Jakub Jelinek wrote: > As the testcases below show, some exports from libgfortran.so > were forgotten on targets with symbol versioning. > This patch fixes it, bootstrapped/regtested on x86_64-linux and i686-linux, > ok for trunk? OK. Thanks for the patch. I think we should do something similar before every new release. Tobias > 2011-02-16 Jakub Jelinek<jakub@redhat.com> > > PR libfortran/47757 > * libgfortran/gfortran.map (GFORTRAN_1.4): Export > _gfortran_{m,s}i{all,any,parity}_i{1,2,4,8,16} and > _gfortran_{cshift0,eoshift{0,2}}_16_char4. > > * gfortran.dg/pr47757-1.f90: New test. > * gfortran.dg/pr47757-2.f90: New test. > * gfortran.dg/pr47757-3.f90: New test.
On Wed, Feb 16, 2011 at 06:10:12PM +0100, Tobias Burnus wrote: > On 02/16/2011 02:51 PM, Jakub Jelinek wrote: > >As the testcases below show, some exports from libgfortran.so > >were forgotten on targets with symbol versioning. > >This patch fixes it, bootstrapped/regtested on x86_64-linux and i686-linux, > >ok for trunk? > > OK. Thanks for the patch. I think we should do something similar > before every new release. Agreed. Anyway, there are other issues: 1) for __float128 using targets, lots of new symbols appeared even in older GFORTRAN_1.0 etc. versions, ideally they would go to GFORTRAN_1.4 if they have been added on those targets for 4.6 (you'd probably need to preprocess gfortran.map for this) 2) I see quite many PLT slots for _gfortran_* functions in libgfortran.so.3, while some of them may be from Fortran sources and harder to avoid, perhaps a bunch of new iexport_proto's are needed Jakub
On Wed, Feb 16, 2011 at 19:23, Jakub Jelinek <jakub@redhat.com> wrote: > On Wed, Feb 16, 2011 at 06:10:12PM +0100, Tobias Burnus wrote: >> On 02/16/2011 02:51 PM, Jakub Jelinek wrote: >> >As the testcases below show, some exports from libgfortran.so >> >were forgotten on targets with symbol versioning. >> >This patch fixes it, bootstrapped/regtested on x86_64-linux and i686-linux, >> >ok for trunk? >> >> OK. Thanks for the patch. I think we should do something similar >> before every new release. > > Agreed. Anyway, there are other issues: > 1) for __float128 using targets, lots of new symbols appeared even in older > GFORTRAN_1.0 etc. versions, ideally they would go to GFORTRAN_1.4 > if they have been added on those targets for 4.6 (you'd probably need > to preprocess gfortran.map for this) > 2) I see quite many PLT slots for _gfortran_* functions in libgfortran.so.3, > while some of them may be from Fortran sources and harder to avoid, > perhaps a bunch of new iexport_proto's are needed Would it make sense to compile libgfortran with -fvisibility=hidden?
On Wed, Feb 16, 2011 at 08:18:39PM +0200, Janne Blomqvist wrote: > On Wed, Feb 16, 2011 at 19:23, Jakub Jelinek <jakub@redhat.com> wrote: > > On Wed, Feb 16, 2011 at 06:10:12PM +0100, Tobias Burnus wrote: > >> On 02/16/2011 02:51 PM, Jakub Jelinek wrote: > >> >As the testcases below show, some exports from libgfortran.so > >> >were forgotten on targets with symbol versioning. > >> >This patch fixes it, bootstrapped/regtested on x86_64-linux and i686-linux, > >> >ok for trunk? > >> > >> OK. Thanks for the patch. I think we should do something similar > >> before every new release. > > > > Agreed. Anyway, there are other issues: > > 1) for __float128 using targets, lots of new symbols appeared even in older > > GFORTRAN_1.0 etc. versions, ideally they would go to GFORTRAN_1.4 > > if they have been added on those targets for 4.6 (you'd probably need > > to preprocess gfortran.map for this) > > 2) I see quite many PLT slots for _gfortran_* functions in libgfortran.so.3, > > while some of them may be from Fortran sources and harder to avoid, > > perhaps a bunch of new iexport_proto's are needed > > Would it make sense to compile libgfortran with -fvisibility=hidden? It wouldn't, it would just completely break it. Jakub
--- libgfortran/gfortran.map.jj 2010-11-17 10:13:20.000000000 +0100 +++ libgfortran/gfortran.map 2011-02-16 10:54:46.979897353 +0100 @@ -1122,16 +1122,46 @@ GFORTRAN_1.4 { _gfortran_iall_i4; _gfortran_iall_i8; _gfortran_iall_i16; + _gfortran_miall_i1; + _gfortran_miall_i2; + _gfortran_miall_i4; + _gfortran_miall_i8; + _gfortran_miall_i16; + _gfortran_siall_i1; + _gfortran_siall_i2; + _gfortran_siall_i4; + _gfortran_siall_i8; + _gfortran_siall_i16; _gfortran_iany_i1; _gfortran_iany_i2; _gfortran_iany_i4; _gfortran_iany_i8; _gfortran_iany_i16; + _gfortran_miany_i1; + _gfortran_miany_i2; + _gfortran_miany_i4; + _gfortran_miany_i8; + _gfortran_miany_i16; + _gfortran_siany_i1; + _gfortran_siany_i2; + _gfortran_siany_i4; + _gfortran_siany_i8; + _gfortran_siany_i16; _gfortran_iparity_i1; _gfortran_iparity_i2; _gfortran_iparity_i4; _gfortran_iparity_i8; _gfortran_iparity_i16; + _gfortran_miparity_i1; + _gfortran_miparity_i2; + _gfortran_miparity_i4; + _gfortran_miparity_i8; + _gfortran_miparity_i16; + _gfortran_siparity_i1; + _gfortran_siparity_i2; + _gfortran_siparity_i4; + _gfortran_siparity_i8; + _gfortran_siparity_i16; _gfortran_norm2_r4; _gfortran_norm2_r8; _gfortran_norm2_r10; @@ -1154,6 +1184,9 @@ GFORTRAN_1.4 { _gfortran_transfer_real_write; _gfortran_transfer_real128; _gfortran_transfer_real128_write; + _gfortran_cshift0_16_char4; + _gfortran_eoshift0_16_char4; + _gfortran_eoshift2_16_char4; } GFORTRAN_1.3; F2C_1.0 { --- gcc/testsuite/gfortran.dg/pr47757-1.f90.jj 2011-02-16 10:32:59.273764520 +0100 +++ gcc/testsuite/gfortran.dg/pr47757-1.f90 2011-02-16 10:35:12.370402168 +0100 @@ -0,0 +1,40 @@ +! PR libfortran/47757 +! { dg-do run } + + integer(1) :: a1(2,2) + integer(2) :: a2(2,2) + integer(4) :: a4(2,2) + integer(8) :: a8(2,2) + logical :: mask(2,2) + logical :: mask2 + a1 = 0 + a2 = 0 + a3 = 0 + a4 = 0 + mask2 = .true. + mask = reshape([.true.,.true.,.false.,.true.],[2,2]) + print *, iany(a1, dim=1, mask=mask) + print *, iany(a2, dim=1, mask=mask) + print *, iany(a4, dim=1, mask=mask) + print *, iany(a8, dim=1, mask=mask) + print *, iall(a1, dim=1, mask=mask) + print *, iall(a2, dim=1, mask=mask) + print *, iall(a4, dim=1, mask=mask) + print *, iall(a8, dim=1, mask=mask) + print *, iparity(a1, dim=1, mask=mask) + print *, iparity(a2, dim=1, mask=mask) + print *, iparity(a4, dim=1, mask=mask) + print *, iparity(a8, dim=1, mask=mask) + print *, iany(a1, dim=1, mask=mask2) + print *, iany(a2, dim=1, mask=mask2) + print *, iany(a4, dim=1, mask=mask2) + print *, iany(a8, dim=1, mask=mask2) + print *, iall(a1, dim=1, mask=mask2) + print *, iall(a2, dim=1, mask=mask2) + print *, iall(a4, dim=1, mask=mask2) + print *, iall(a8, dim=1, mask=mask2) + print *, iparity(a1, dim=1, mask=mask2) + print *, iparity(a2, dim=1, mask=mask2) + print *, iparity(a4, dim=1, mask=mask2) + print *, iparity(a8, dim=1, mask=mask2) +end --- gcc/testsuite/gfortran.dg/pr47757-2.f90.jj 2011-02-16 10:33:09.269420078 +0100 +++ gcc/testsuite/gfortran.dg/pr47757-2.f90 2011-02-16 10:50:30.057714582 +0100 @@ -0,0 +1,16 @@ +! PR libfortran/47757 +! { dg-do run { target fortran_large_int } } + + integer(16) :: a16(2,2) + logical :: mask(2,2) + logical :: mask2 + a16 = 0 + mask2 = .true. + mask = reshape([.true.,.true.,.false.,.true.],[2,2]) + print *, iany(a16, dim=1, mask=mask) + print *, iall(a16, dim=1, mask=mask) + print *, iparity(a16, dim=1, mask=mask) + print *, iany(a16, dim=1, mask=mask2) + print *, iall(a16, dim=1, mask=mask2) + print *, iparity(a16, dim=1, mask=mask2) +end --- gcc/testsuite/gfortran.dg/pr47757-3.f90.jj 2011-02-16 10:35:25.864388810 +0100 +++ gcc/testsuite/gfortran.dg/pr47757-3.f90 2011-02-16 10:50:41.550388829 +0100 @@ -0,0 +1,12 @@ +! PR libfortran/47757 +! { dg-do run { target fortran_large_int } } + + character(kind=4):: str(3,3), s(3) + str(1,:) = [4_'A', 4_'b', 4_'C'] + str(2,:) = [4_'A', 4_'b', 4_'C'] + str(3,:) = [4_'A', 4_'b', 4_'C'] + s = 4_'A' + print *, cshift(str, shift=2_16, dim=1_16) + print *, eoshift(str, shift=2_16, dim=1_16) + print *, eoshift(str, shift=2_16, boundary=s, dim=1_16) +end
Hi! As the testcases below show, some exports from libgfortran.so were forgotten on targets with symbol versioning. This patch fixes it, bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk? 2011-02-16 Jakub Jelinek <jakub@redhat.com> PR libfortran/47757 * libgfortran/gfortran.map (GFORTRAN_1.4): Export _gfortran_{m,s}i{all,any,parity}_i{1,2,4,8,16} and _gfortran_{cshift0,eoshift{0,2}}_16_char4. * gfortran.dg/pr47757-1.f90: New test. * gfortran.dg/pr47757-2.f90: New test. * gfortran.dg/pr47757-3.f90: New test. Jakub