Re: For testing: full __float128 patch

Re: For testing: full __float128 patch

[Uros Bizjak]

[-- Attachment #1: Type: text/plain, Size: 1382 bytes --]

Hello!

> On Sun, Aug 29, 2010 at 09:01:49AM +0200, FX wrote:
> > GCC >= 4.5 is needed. I'll add a check to libquad's configure.
> > 
> > FX
> > 
> 
> Further testing on i386-*-freebsd runs into undefined symbols.
> 
> libquad.so: undefined reference to `__getf2'
> libquad.so: undefined reference to `__eqtf2'
> libquad.so: undefined reference to `__addtf3'
> libquad.so: undefined reference to `__floatsitf'
> libquad.so: undefined reference to `__trunctfxf2'
> libquad.so: undefined reference to `__divtf3'
> libquad.so: undefined reference to `__letf2'
> libquad.so: undefined reference to `__unordtf2'
> libquad.so: undefined reference to `__fixtfdi'
> libquad.so: undefined reference to `__fixtfsi'
> libquad.so: undefined reference to `__lttf2'
> libquad.so: undefined reference to `__netf2'
> libquad.so: undefined reference to `__extendxftf2'
> libquad.so: undefined reference to `__extenddftf2'
> libquad.so: undefined reference to `__multf3'
> libquad.so: undefined reference to `__gttf2'
> libquad.so: undefined reference to `__subtf3'
> libgfortran.so: undefined reference to `__floatunditf'
> libquad.so: undefined reference to `__trunctfdf2'
> 
> 
> AFAICT, these should come from libgcc_s.so.1, but for whatever
> reason these are missing.  I have been unable to find how
> to induce gcc to build the required files. 

Try to build gcc with attached patch...

Uros.

[-- Attachment #2: c.diff.txt --]
[-- Type: text/plain, Size: 523 bytes --]

Index: config.gcc
===================================================================
--- config.gcc	(revision 163630)
+++ config.gcc	(working copy)
@@ -3485,6 +3485,9 @@
 	i[34567]86-*-cygwin* | i[34567]86-*-mingw* | x86_64-*-mingw*)
 		tmake_file="${tmake_file} i386/t-fprules-softfp soft-fp/t-softfp"
 		;;
+	i[34567]86-*-freebsd* | x86_64-*-freebsd*)
+		tmake_file="${tmake_file} i386/t-fprules-softfp soft-fp/t-softfp"
+		;;
 	ia64*-*-linux*)
 		tmake_file="${tmake_file} ia64/t-fprules-softfp soft-fp/t-softfp"
 		;;

Re: For testing: full __float128 patch

[Steve Kargl]

On Sun, Aug 29, 2010 at 10:48:46PM +0200, Uros Bizjak wrote:
> Hello!
> 
> > On Sun, Aug 29, 2010 at 09:01:49AM +0200, FX wrote:
> > > GCC >= 4.5 is needed. I'll add a check to libquad's configure.
> > > 
> > > FX
> > > 
> > 
> > Further testing on i386-*-freebsd runs into undefined symbols.
> > 
> > libquad.so: undefined reference to `__getf2'
> > libquad.so: undefined reference to `__eqtf2'
> > libquad.so: undefined reference to `__addtf3'
> > libquad.so: undefined reference to `__floatsitf'
> > libquad.so: undefined reference to `__trunctfxf2'
> > libquad.so: undefined reference to `__divtf3'
> > libquad.so: undefined reference to `__letf2'
> > libquad.so: undefined reference to `__unordtf2'
> > libquad.so: undefined reference to `__fixtfdi'
> > libquad.so: undefined reference to `__fixtfsi'
> > libquad.so: undefined reference to `__lttf2'
> > libquad.so: undefined reference to `__netf2'
> > libquad.so: undefined reference to `__extendxftf2'
> > libquad.so: undefined reference to `__extenddftf2'
> > libquad.so: undefined reference to `__multf3'
> > libquad.so: undefined reference to `__gttf2'
> > libquad.so: undefined reference to `__subtf3'
> > libgfortran.so: undefined reference to `__floatunditf'
> > libquad.so: undefined reference to `__trunctfdf2'
> > 
> > 
> > AFAICT, these should come from libgcc_s.so.1, but for whatever
> > reason these are missing.  I have been unable to find how
> > to induce gcc to build the required files. 
> 
> Try to build gcc with attached patch...
> 

Thanks for the patch.  It is clearly a step in the right
direction.  Unfortunately, my build dies with fixtfti.c,
fixunstfti.c, floattitf.c, and floatuntitf.c, which I 
assume are used with conversions involving TI and TF modes.
If I use #if 0 ... #endif to block out the code in 
those files, my build proceeds to libgomp, which dies with
only two missing symbols,

/usr/home/kargl/gcc/obj4x/./gcc/libgcc_s.so: undefined reference to `__fabstf2'
/usr/home/kargl/gcc/obj4x/./gcc/libgcc_s.so: undefined reference to `__copysigntf3'

I suppose this means that on FreeBSD, I need to enable TI mode
as well as TF.

-- 
Steve

Re: For testing: full __float128 patch

[Uros Bizjak]

[-- Attachment #1: Type: text/plain, Size: 2372 bytes --]

On Mon, Aug 30, 2010 at 1:47 AM, Steve Kargl
<sgk@troutmask.apl.washington.edu> wrote:

>> > Further testing on i386-*-freebsd runs into undefined symbols.
>> >
>> > libquad.so: undefined reference to `__getf2'
>> > libquad.so: undefined reference to `__eqtf2'
>> > libquad.so: undefined reference to `__addtf3'
>> > libquad.so: undefined reference to `__floatsitf'
>> > libquad.so: undefined reference to `__trunctfxf2'
>> > libquad.so: undefined reference to `__divtf3'
>> > libquad.so: undefined reference to `__letf2'
>> > libquad.so: undefined reference to `__unordtf2'
>> > libquad.so: undefined reference to `__fixtfdi'
>> > libquad.so: undefined reference to `__fixtfsi'
>> > libquad.so: undefined reference to `__lttf2'
>> > libquad.so: undefined reference to `__netf2'
>> > libquad.so: undefined reference to `__extendxftf2'
>> > libquad.so: undefined reference to `__extenddftf2'
>> > libquad.so: undefined reference to `__multf3'
>> > libquad.so: undefined reference to `__gttf2'
>> > libquad.so: undefined reference to `__subtf3'
>> > libgfortran.so: undefined reference to `__floatunditf'
>> > libquad.so: undefined reference to `__trunctfdf2'
>> >
>> >
>> > AFAICT, these should come from libgcc_s.so.1, but for whatever
>> > reason these are missing.  I have been unable to find how
>> > to induce gcc to build the required files.
>>
>> Try to build gcc with attached patch...
>>
>
> Thanks for the patch.  It is clearly a step in the right
> direction.  Unfortunately, my build dies with fixtfti.c,
> fixunstfti.c, floattitf.c, and floatuntitf.c, which I
> assume are used with conversions involving TI and TF modes.
> If I use #if 0 ... #endif to block out the code in
> those files, my build proceeds to libgomp, which dies with
> only two missing symbols,
>
> /usr/home/kargl/gcc/obj4x/./gcc/libgcc_s.so: undefined reference to `__fabstf2'
> /usr/home/kargl/gcc/obj4x/./gcc/libgcc_s.so: undefined reference to `__copysigntf3'
>
> I suppose this means that on FreeBSD, I need to enable TI mode
> as well as TF.

Oh, additional patch for libgcc is also needed. Attached patch will
enable filtering out TImode on 32bits and will also enable fallbacks
for the two missing functions, so no need for any other changes.

Can you please regression test these two patches on freebsd (32 and 64 bit)?

Uros.

[-- Attachment #2: d.diff.txt --]
[-- Type: text/plain, Size: 572 bytes --]

Index: config.host
===================================================================
--- config.host	(revision 163630)
+++ config.host	(working copy)
@@ -608,7 +608,8 @@
   i[34567]86-*-linux* | x86_64-*-linux* | \
   i[34567]86-*-gnu* | \
   i[34567]86-*-solaris2* | \
-  i[34567]86-*-cygwin* | i[34567]86-*-mingw* | x86_64-*-mingw*)
+  i[34567]86-*-cygwin* | i[34567]86-*-mingw* | x86_64-*-mingw* | \
+  i[34567]86-*-freebsd* | x86_64-*-freebsd*)
 	if test "${host_address}" = 32; then
 		tmake_file="${tmake_file} t-softfp i386/${host_address}/t-fprules-softfp"
 	fi

Re: For testing: full __float128 patch

[Gerald Pfeifer]

On Mon, 30 Aug 2010, Uros Bizjak wrote:
> Oh, additional patch for libgcc is also needed. Attached patch will
> enable filtering out TImode on 32bits and will also enable fallbacks
> for the two missing functions, so no need for any other changes.
> 
> Can you please regression test these two patches on freebsd (32 and 64 
> bit)?

Tests running...

Steve, I assume you run a number of bootstrap+test cycles over time? Would 
you mind submitting your results to gcc-testresults as well? So far I'm 
the only one doing that for FreeBSD it seems, and comparisons etc. would
be useful.

Gerald

Re: For testing: full __float128 patch

[Gerald Pfeifer]

On Mon, 30 Aug 2010, Gerald Pfeifer wrote:
>> Can you please regression test these two patches on freebsd (32 and 64 
>> bit)?
> Tests running...

Looks like a nice improvement, actually.  On both this fixes previously
existing test failures and adds some passes (no changes on the failure
side on i386).

From all I can tell, please go ahead, Uros. :-)  I wonder whether these
might also be suitable for GCC 4.5?



First x86_64-unknown-freebsd8.1 (cf. 
http://gcc.gnu.org/ml/gcc-testresults/2010-08/msg02964.html without
and http://gcc.gnu.org/ml/gcc-testresults/2010-08/msg02959.html with):

 		=== gcc tests ===
 
-FAIL: gcc.dg/torture/fp-int-convert-float128-timode.c  -O0  (test for excess errors)
-WARNING: gcc.dg/torture/fp-int-convert-float128-timode.c  -O0  compilation failed to produce executable
-FAIL: gcc.dg/torture/fp-int-convert-float128-timode.c  -O1  (test for excess errors)
-WARNING: gcc.dg/torture/fp-int-convert-float128-timode.c  -O1  compilation failed to produce executable
-FAIL: gcc.dg/torture/fp-int-convert-float128-timode.c  -O2  (test for excess errors)
-WARNING: gcc.dg/torture/fp-int-convert-float128-timode.c  -O2  compilation failed to produce executable
-FAIL: gcc.dg/torture/fp-int-convert-float128-timode.c  -O3 -fomit-frame-pointer  (test for excess errors)
-WARNING: gcc.dg/torture/fp-int-convert-float128-timode.c  -O3 -fomit-frame-pointer  compilation failed to produce executable
-FAIL: gcc.dg/torture/fp-int-convert-float128-timode.c  -O3 -g  (test for excess errors)
-WARNING: gcc.dg/torture/fp-int-convert-float128-timode.c  -O3 -g  compilation failed to produce executable
-FAIL: gcc.dg/torture/fp-int-convert-float128-timode.c  -Os  (test for excess errors)
-WARNING: gcc.dg/torture/fp-int-convert-float128-timode.c  -Os  compilation failed to produce executable
-FAIL: gcc.dg/torture/fp-int-convert-float128.c  -O0  (test for excess errors)
-WARNING: gcc.dg/torture/fp-int-convert-float128.c  -O0  compilation failed to produce executable
-FAIL: gcc.dg/torture/fp-int-convert-float128.c  -O1  (test for excess errors)
-WARNING: gcc.dg/torture/fp-int-convert-float128.c  -O1  compilation failed to produce executable
-FAIL: gcc.dg/torture/fp-int-convert-float128.c  -O2  (test for excess errors)
-WARNING: gcc.dg/torture/fp-int-convert-float128.c  -O2  compilation failed to produce executable
-FAIL: gcc.dg/torture/fp-int-convert-float128.c  -O3 -fomit-frame-pointer  (test for excess errors)
-WARNING: gcc.dg/torture/fp-int-convert-float128.c  -O3 -fomit-frame-pointer  compilation failed to produce executable
-FAIL: gcc.dg/torture/fp-int-convert-float128.c  -O3 -g  (test for excess errors)
-WARNING: gcc.dg/torture/fp-int-convert-float128.c  -O3 -g  compilation failed to produce executable
-FAIL: gcc.dg/torture/fp-int-convert-float128.c  -Os  (test for excess errors)
-WARNING: gcc.dg/torture/fp-int-convert-float128.c  -Os  compilation failed to produce executable
 
 		=== gcc Summary ===
 
-# of expected passes		60905
-# of unexpected failures	21
+# of expected passes		60929
+# of unexpected failures	9



Second, i386-unknown-freebsd7.1 (cf. 
http://gcc.gnu.org/ml/gcc-testresults/2010-08/msg02948.html without, 
and http://gcc.gnu.org/ml/gcc-testresults/2010-08/msg02975.html with):

 		=== gcc tests ===
 
@@ -127,30 +127,6 @@
 FAIL: gcc.dg/torture/fp-int-convert-float.c  -O3 -fomit-frame-pointer  execution test
 FAIL: gcc.dg/torture/fp-int-convert-float.c  -O3 -g  execution test
 FAIL: gcc.dg/torture/fp-int-convert-float.c  -Os  execution test
-FAIL: gcc.dg/torture/fp-int-convert-float128-timode.c  -O0  (test for excess errors)
-WARNING: gcc.dg/torture/fp-int-convert-float128-timode.c  -O0  compilation failed to produce executable
-FAIL: gcc.dg/torture/fp-int-convert-float128-timode.c  -O1  (test for excess errors)
-WARNING: gcc.dg/torture/fp-int-convert-float128-timode.c  -O1  compilation failed to produce executable
-FAIL: gcc.dg/torture/fp-int-convert-float128-timode.c  -O2  (test for excess errors)
-WARNING: gcc.dg/torture/fp-int-convert-float128-timode.c  -O2  compilation failed to produce executable
-FAIL: gcc.dg/torture/fp-int-convert-float128-timode.c  -O3 -fomit-frame-pointer  (test for excess errors)
-WARNING: gcc.dg/torture/fp-int-convert-float128-timode.c  -O3 -fomit-frame-pointer  compilation failed to produce executable
-FAIL: gcc.dg/torture/fp-int-convert-float128-timode.c  -O3 -g  (test for excess errors)
-WARNING: gcc.dg/torture/fp-int-convert-float128-timode.c  -O3 -g  compilation failed to produce executable
-FAIL: gcc.dg/torture/fp-int-convert-float128-timode.c  -Os  (test for excess errors)
-WARNING: gcc.dg/torture/fp-int-convert-float128-timode.c  -Os  compilation failed to produce executable
-FAIL: gcc.dg/torture/fp-int-convert-float128.c  -O0  (test for excess errors)
-WARNING: gcc.dg/torture/fp-int-convert-float128.c  -O0  compilation failed to produce executable
-FAIL: gcc.dg/torture/fp-int-convert-float128.c  -O1  (test for excess errors)
-WARNING: gcc.dg/torture/fp-int-convert-float128.c  -O1  compilation failed to produce executable
-FAIL: gcc.dg/torture/fp-int-convert-float128.c  -O2  (test for excess errors)
-WARNING: gcc.dg/torture/fp-int-convert-float128.c  -O2  compilation failed to produce executable
-FAIL: gcc.dg/torture/fp-int-convert-float128.c  -O3 -fomit-frame-pointer  (test for excess errors)
-WARNING: gcc.dg/torture/fp-int-convert-float128.c  -O3 -fomit-frame-pointer  compilation failed to produce executable
-FAIL: gcc.dg/torture/fp-int-convert-float128.c  -O3 -g  (test for excess errors)
-WARNING: gcc.dg/torture/fp-int-convert-float128.c  -O3 -g  compilation failed to produce executable
-FAIL: gcc.dg/torture/fp-int-convert-float128.c  -Os  (test for excess errors)
-WARNING: gcc.dg/torture/fp-int-convert-float128.c  -Os  compilation failed to produce executable
 FAIL: gcc.dg/torture/fp-int-convert-float80.c  -O0  execution test
 FAIL: gcc.dg/torture/fp-int-convert-float80.c  -O1  execution test
 FAIL: gcc.dg/torture/fp-int-convert-float80.c  -O2  execution test
@@ -176,12 +152,12 @@
 
 		=== gcc Summary ===
 
-# of expected passes		60614
-# of unexpected failures	78
+# of expected passes		60638
+# of unexpected failures	66
 # of unexpected successes	3
 # of expected failures		204
 # of unsupported tests		1251
 
 		=== gfortran Summary ===
 
-# of expected passes		35511
+# of expected passes		35527
 # of unexpected failures	12
 # of expected failures		39
 # of unsupported tests		275



Gerald

Re: For testing: full __float128 patch

[Joseph S. Myers]

On Mon, 30 Aug 2010, Gerald Pfeifer wrote:

> On Mon, 30 Aug 2010, Gerald Pfeifer wrote:
> >> Can you please regression test these two patches on freebsd (32 and 64 
> >> bit)?
> > Tests running...
> 
> Looks like a nice improvement, actually.  On both this fixes previously
> existing test failures and adds some passes (no changes on the failure
> side on i386).
> 
> From all I can tell, please go ahead, Uros. :-)  I wonder whether these
> might also be suitable for GCC 4.5?

Assuming that symbol versioning is enabled on FreeBSD, you need to set 
appropriate versions for all the new symbols in libgcc_s - which would be 
GCC_4.6.0 for something going on trunk only but GCC_4.5.2 for something 
going on 4.5 as well (and with the right %inherit settings in any case).

-- 
Joseph S. Myers
joseph@codesourcery.com

Re: For testing: full __float128 patch

[Steve Kargl]

On Mon, Aug 30, 2010 at 08:28:58AM +0200, Uros Bizjak wrote:
> 
> Oh, additional patch for libgcc is also needed. Attached patch will
> enable filtering out TImode on 32bits and will also enable fallbacks
> for the two missing functions, so no need for any other changes.
> 
> Can you please regression test these two patches on freebsd (32 and 64 bit)?
> 
> Uros.

I won't be able to test the new patch until
later tonight (about 10 hours from now 10:30 am PST).
Gerald's testing appears to show good results.

-- 
Steve

Re: For testing: full __float128 patch

[Uros Bizjak]

[-- Attachment #1: Type: text/plain, Size: 1514 bytes --]

On Mon, Aug 30, 2010 at 6:48 PM, Joseph S. Myers
<joseph@codesourcery.com> wrote:

>> >> Can you please regression test these two patches on freebsd (32 and 64
>> >> bit)?
>> > Tests running...
>>
>> Looks like a nice improvement, actually.  On both this fixes previously
>> existing test failures and adds some passes (no changes on the failure
>> side on i386).
>>
>> From all I can tell, please go ahead, Uros. :-)  I wonder whether these
>> might also be suitable for GCC 4.5?
>
> Assuming that symbol versioning is enabled on FreeBSD, you need to set
> appropriate versions for all the new symbols in libgcc_s - which would be
> GCC_4.6.0 for something going on trunk only but GCC_4.5.2 for something
> going on 4.5 as well (and with the right %inherit settings in any case).

I'm not familiar in versioning stuff, but since soft-fp symbols are
part of generic libgcc-std.ver for a long time (before 4.5.0), I
believe, it is enough to patch only configure files (see i.e. commits
that enabled soft-fp on mingw/cygwin/-gnu*).

gcc/ChangeLog:

2010-08-30  Uros Bizjak  <ubizjak@gmail.com>

	* config.gcc: Add i386/t-fprules-softfp and soft-fp/t-softfp
	to tmake_file for i[34567]86-*-freebsd* and x86_64-*-freebsd*.

libgcc/ChangeLog:

2010-08-30  Uros Bizjak  <ubizjak@gmail.com>

	* config.host: Add i386/${host_address}/t-fprules-softfp and
	t-softfp to tmake_file for i[34567]86-*-freebsd* and x86_64-*-freebsd*.

Tested by Gerald. OK for 4.5.2 and mainline?

Uros.

[-- Attachment #2: c.diff.txt --]
[-- Type: text/plain, Size: 1128 bytes --]

Index: gcc/config.gcc
===================================================================
--- gcc/config.gcc	(revision 163647)
+++ gcc/config.gcc	(working copy)
@@ -3485,6 +3485,9 @@
 	i[34567]86-*-cygwin* | i[34567]86-*-mingw* | x86_64-*-mingw*)
 		tmake_file="${tmake_file} i386/t-fprules-softfp soft-fp/t-softfp"
 		;;
+	i[34567]86-*-freebsd* | x86_64-*-freebsd*)
+		tmake_file="${tmake_file} i386/t-fprules-softfp soft-fp/t-softfp"
+		;;
 	ia64*-*-linux*)
 		tmake_file="${tmake_file} ia64/t-fprules-softfp soft-fp/t-softfp"
 		;;
Index: libgcc/config.host
===================================================================
--- libgcc/config.host	(revision 163647)
+++ libgcc/config.host	(working copy)
@@ -608,7 +608,8 @@
   i[34567]86-*-linux* | x86_64-*-linux* | \
   i[34567]86-*-gnu* | \
   i[34567]86-*-solaris2* | \
-  i[34567]86-*-cygwin* | i[34567]86-*-mingw* | x86_64-*-mingw*)
+  i[34567]86-*-cygwin* | i[34567]86-*-mingw* | x86_64-*-mingw* | \
+  i[34567]86-*-freebsd* | x86_64-*-freebsd*)
 	if test "${host_address}" = 32; then
 		tmake_file="${tmake_file} t-softfp i386/${host_address}/t-fprules-softfp"
 	fi

Re: For testing: full __float128 patch

[Richard Henderson]

On 08/30/2010 10:37 AM, Uros Bizjak wrote:
> I'm not familiar in versioning stuff, but since soft-fp symbols are
> part of generic libgcc-std.ver for a long time (before 4.5.0), I
> believe, it is enough to patch only configure files (see i.e. commits
> that enabled soft-fp on mingw/cygwin/-gnu*).

It will "work", but not properly.  One should never have symbols
magically appear in old versions.  If you add the symbols to the
proper current version number in a config/gcc/i386/libgcc-bsd.ver,
they'll appear at the right place.


r~

Re: For testing: full __float128 patch

[Uros Bizjak]

On Mon, Aug 30, 2010 at 7:48 PM, Richard Henderson <rth@redhat.com> wrote:
> On 08/30/2010 10:37 AM, Uros Bizjak wrote:
>> I'm not familiar in versioning stuff, but since soft-fp symbols are
>> part of generic libgcc-std.ver for a long time (before 4.5.0), I
>> believe, it is enough to patch only configure files (see i.e. commits
>> that enabled soft-fp on mingw/cygwin/-gnu*).
>
> It will "work", but not properly.  One should never have symbols
> magically appear in old versions.  If you add the symbols to the
> proper current version number in a config/gcc/i386/libgcc-bsd.ver,
> they'll appear at the right place.

Can someone please help here?

Thanks,
Uros.

Re: For testing: full __float128 patch

[Uros Bizjak]

[-- Attachment #1: Type: text/plain, Size: 866 bytes --]

On Mon, Aug 30, 2010 at 7:53 PM, Uros Bizjak <ubizjak@gmail.com> wrote:

>>> I'm not familiar in versioning stuff, but since soft-fp symbols are
>>> part of generic libgcc-std.ver for a long time (before 4.5.0), I
>>> believe, it is enough to patch only configure files (see i.e. commits
>>> that enabled soft-fp on mingw/cygwin/-gnu*).
>>
>> It will "work", but not properly.  One should never have symbols
>> magically appear in old versions.  If you add the symbols to the
>> proper current version number in a config/gcc/i386/libgcc-bsd.ver,
>> they'll appear at the right place.
>
> Can someone please help here?

Attached is my best (untested) shot at libgcc-bsd.ver. Someone has to
write i386/t-freebsd (similar to i386/t-linux) and connect it to the
build system. The versioning assumes that this will be committed to
4.5.x first.

Uros.

[-- Attachment #2: libgcc-bsd.ver --]
[-- Type: application/octet-stream, Size: 1793 bytes --]

# Copyright (C) 2010 Free Software Foundation, Inc.
#
# This file is part of GCC.
#
# GCC is free software; you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation; either version 3, or (at your option)
# any later version.
#
# GCC is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with GCC; see the file COPYING3.  If not see
# <http://www.gnu.org/licenses/>.

# 128 bit long double support was introduced with GCC 4.5.0.
# These lines make the symbols to get a @@GCC_4.5.0 attached.

%exclude {
  __addtf3
  __divtc3
  __divtf3
  __eqtf2
  __extenddftf2
  __extendsftf2
  __extendxftf2
  __fixtfdi
  __fixtfsi
  __fixtfti
  __fixunstfdi
  __fixunstfsi
  __fixunstfti
  __floatditf
  __floatsitf
  __floattitf
  __floatunditf
  __floatunsitf
  __floatuntitf
  __getf2
  __gttf2
  __letf2
  __lttf2
  __multc3
  __multf3
  __negtf2
  __netf2
  __powitf2
  __subtf3
  __trunctfdf2
  __trunctfsf2
  __trunctfxf2
  __unordtf2
}

GCC_4.5.0 {
  __addtf3
  __divtc3
  __divtf3
  __eqtf2
  __extenddftf2
  __extendsftf2
  __extendxftf2
  __fixtfdi
  __fixtfsi
  __fixunstfdi
  __fixunstfsi
  __floatditf
  __floatsitf
  __floattitf
  __floatunditf
  __floatunsitf
  __getf2
  __gttf2
  __letf2
  __lttf2
  __multc3
  __multf3
  __negtf2
  __netf2
  __powitf2
  __subtf3
  __trunctfdf2
  __trunctfsf2
  __trunctfxf2
  __unordtf2
}

%ifdef __x86_64__
GCC_4.5.0 {
  __fixtfti
  __fixunstfti
  __floatuntitf
}
%else
GCC_4.5.0 {
  __copysigntf3
  __fabstf2
}
%endif

Re: For testing: full __float128 patch

[Richard Henderson]

On 08/30/2010 12:25 PM, Uros Bizjak wrote:
> Attached is my best (untested) shot at libgcc-bsd.ver. Someone has
> to write i386/t-freebsd (similar to i386/t-linux) and connect it to
> the build system. The versioning assumes that this will be committed
> to 4.5.x first.

That file looks plausible, though of course the actual
version will be 4.5.2.


r~

Re: For testing: full __float128 patch

[Jakub Jelinek]

On Mon, Aug 30, 2010 at 09:25:42PM +0200, Uros Bizjak wrote:
> On Mon, Aug 30, 2010 at 7:53 PM, Uros Bizjak <ubizjak@gmail.com> wrote:
> 
> >>> I'm not familiar in versioning stuff, but since soft-fp symbols are
> >>> part of generic libgcc-std.ver for a long time (before 4.5.0), I
> >>> believe, it is enough to patch only configure files (see i.e. commits
> >>> that enabled soft-fp on mingw/cygwin/-gnu*).
> >>
> >> It will "work", but not properly.  One should never have symbols
> >> magically appear in old versions.  If you add the symbols to the
> >> proper current version number in a config/gcc/i386/libgcc-bsd.ver,
> >> they'll appear at the right place.
> >
> > Can someone please help here?
> 
> Attached is my best (untested) shot at libgcc-bsd.ver. Someone has to
> write i386/t-freebsd (similar to i386/t-linux) and connect it to the
> build system. The versioning assumes that this will be committed to
> 4.5.x first.

It is wrong, GCC 4.5.0 has been released already, so no new symbols should
be added to GCC_4.5.0 symver.
IMNSHO this isn't a regression fix, so you should just apply it to the
trunk, and introduce them as GCC_4.6.0 version.  Also, there is no need
to have two chunks for the same symver in the *.ver file, just put the
3 resp. 2 symbols before the preceeding closing brace, together with the
%ifdef/%else/%endif.

	Jakub

Re: For testing: full __float128 patch

[Uros Bizjak]

[-- Attachment #1: Type: text/plain, Size: 1381 bytes --]

On Mon, Aug 30, 2010 at 9:38 PM, Jakub Jelinek <jakub@redhat.com> wrote:

>> >>> I'm not familiar in versioning stuff, but since soft-fp symbols are
>> >>> part of generic libgcc-std.ver for a long time (before 4.5.0), I
>> >>> believe, it is enough to patch only configure files (see i.e. commits
>> >>> that enabled soft-fp on mingw/cygwin/-gnu*).
>> >>
>> >> It will "work", but not properly.  One should never have symbols
>> >> magically appear in old versions.  If you add the symbols to the
>> >> proper current version number in a config/gcc/i386/libgcc-bsd.ver,
>> >> they'll appear at the right place.
>> >
>> > Can someone please help here?
>>
>> Attached is my best (untested) shot at libgcc-bsd.ver. Someone has to
>> write i386/t-freebsd (similar to i386/t-linux) and connect it to the
>> build system. The versioning assumes that this will be committed to
>> 4.5.x first.
>
> It is wrong, GCC 4.5.0 has been released already, so no new symbols should
> be added to GCC_4.5.0 symver.
> IMNSHO this isn't a regression fix, so you should just apply it to the
> trunk, and introduce them as GCC_4.6.0 version.  Also, there is no need
> to have two chunks for the same symver in the *.ver file, just put the
> 3 resp. 2 symbols before the preceeding closing brace, together with the
> %ifdef/%else/%endif.

So, attached and untested.

Uros.

[-- Attachment #2: c.diff.txt --]
[-- Type: text/plain, Size: 4074 bytes --]

Index: libgcc/config.host
===================================================================
--- libgcc/config.host	(revision 163647)
+++ libgcc/config.host	(working copy)
@@ -608,7 +608,8 @@
   i[34567]86-*-linux* | x86_64-*-linux* | \
   i[34567]86-*-gnu* | \
   i[34567]86-*-solaris2* | \
-  i[34567]86-*-cygwin* | i[34567]86-*-mingw* | x86_64-*-mingw*)
+  i[34567]86-*-cygwin* | i[34567]86-*-mingw* | x86_64-*-mingw* | \
+  i[34567]86-*-freebsd* | x86_64-*-freebsd*)
 	if test "${host_address}" = 32; then
 		tmake_file="${tmake_file} t-softfp i386/${host_address}/t-fprules-softfp"
 	fi
Index: gcc/config.gcc
===================================================================
--- gcc/config.gcc	(revision 163647)
+++ gcc/config.gcc	(working copy)
@@ -1153,10 +1153,11 @@
 	;;
 i[34567]86-*-freebsd*)
 	tm_file="${tm_file} i386/unix.h i386/att.h dbxelf.h elfos.h ${fbsd_tm_file} i386/freebsd.h"
+	tmake_file="${tmake_file} i386/t-freebsd"
 	;;
 x86_64-*-freebsd*)
 	tm_file="${tm_file} i386/unix.h i386/att.h dbxelf.h elfos.h ${fbsd_tm_file} i386/x86-64.h i386/freebsd.h i386/freebsd64.h"
-	tmake_file="${tmake_file} i386/t-crtstuff"
+	tmake_file="${tmake_file} i386/t-crtstuff i386/t-freebsd"
 	;;
 i[34567]86-*-netbsdelf*)
 	tm_file="${tm_file} i386/unix.h i386/att.h dbxelf.h elfos.h netbsd.h netbsd-elf.h i386/netbsd-elf.h"
@@ -3485,6 +3486,9 @@
 	i[34567]86-*-cygwin* | i[34567]86-*-mingw* | x86_64-*-mingw*)
 		tmake_file="${tmake_file} i386/t-fprules-softfp soft-fp/t-softfp"
 		;;
+	i[34567]86-*-freebsd* | x86_64-*-freebsd*)
+		tmake_file="${tmake_file} i386/t-fprules-softfp soft-fp/t-softfp"
+		;;
 	ia64*-*-linux*)
 		tmake_file="${tmake_file} ia64/t-fprules-softfp soft-fp/t-softfp"
 		;;
Index: gcc/config/i386/t-freebsd
===================================================================
--- gcc/config/i386/t-freebsd	(revision 0)
+++ gcc/config/i386/t-freebsd	(revision 0)
@@ -0,0 +1 @@
+SHLIB_MAPFILES = $(srcdir)/config/i386/libgcc-bsd.ver
Index: gcc/config/i386/libgcc-bsd.ver
===================================================================
--- gcc/config/i386/libgcc-bsd.ver	(revision 0)
+++ gcc/config/i386/libgcc-bsd.ver	(revision 0)
@@ -0,0 +1,98 @@
+# Copyright (C) 2010 Free Software Foundation, Inc.
+#
+# This file is part of GCC.
+#
+# GCC is free software; you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation; either version 3, or (at your option)
+# any later version.
+#
+# GCC is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with GCC; see the file COPYING3.  If not see
+# <http://www.gnu.org/licenses/>.
+
+# 128 bit long double support was introduced with GCC 4.6.0.
+# These lines make the symbols to get a @@GCC_4.6.0 attached.
+
+%exclude {
+  __addtf3
+  __divtc3
+  __divtf3
+  __eqtf2
+  __extenddftf2
+  __extendsftf2
+  __extendxftf2
+  __fixtfdi
+  __fixtfsi
+  __fixtfti
+  __fixunstfdi
+  __fixunstfsi
+  __fixunstfti
+  __floatditf
+  __floatsitf
+  __floattitf
+  __floatunditf
+  __floatunsitf
+  __floatuntitf
+  __getf2
+  __gttf2
+  __letf2
+  __lttf2
+  __multc3
+  __multf3
+  __negtf2
+  __netf2
+  __powitf2
+  __subtf3
+  __trunctfdf2
+  __trunctfsf2
+  __trunctfxf2
+  __unordtf2
+}
+
+GCC_4.6.0 {
+  __addtf3
+  __divtc3
+  __divtf3
+  __eqtf2
+  __extenddftf2
+  __extendsftf2
+  __extendxftf2
+  __fixtfdi
+  __fixtfsi
+  __fixunstfdi
+  __fixunstfsi
+  __floatditf
+  __floatsitf
+  __floattitf
+  __floatunditf
+  __floatunsitf
+  __getf2
+  __gttf2
+  __letf2
+  __lttf2
+  __multc3
+  __multf3
+  __negtf2
+  __netf2
+  __powitf2
+  __subtf3
+  __trunctfdf2
+  __trunctfsf2
+  __trunctfxf2
+  __unordtf2
+
+%ifdef __x86_64__
+  __fixtfti
+  __fixunstfti
+  __floatuntitf
+%else
+  __copysigntf3
+  __fabstf2
+%endif
+}
\ No newline at end of file

Re: For testing: full __float128 patch

[Joseph S. Myers]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: TEXT/PLAIN; charset=X-UNKNOWN, Size: 1760 bytes --]

On Mon, 30 Aug 2010, Uros Bizjak wrote:

> On Mon, Aug 30, 2010 at 7:53 PM, Uros Bizjak <ubizjak@gmail.com> wrote:
> 
> >>> I'm not familiar in versioning stuff, but since soft-fp symbols are
> >>> part of generic libgcc-std.ver for a long time (before 4.5.0), I
> >>> believe, it is enough to patch only configure files (see i.e. commits
> >>> that enabled soft-fp on mingw/cygwin/-gnu*).
> >>
> >> It will "work", but not properly.  One should never have symbols
> >> magically appear in old versions.  If you add the symbols to the
> >> proper current version number in a config/gcc/i386/libgcc-bsd.ver,
> >> they'll appear at the right place.
> >
> > Can someone please help here?
> 
> Attached is my best (untested) shot at libgcc-bsd.ver. Someone has to
> write i386/t-freebsd (similar to i386/t-linux) and connect it to the
> build system. The versioning assumes that this will be committed to
> 4.5.x first.

I advise copying libgcc-sol2.ver, with GCC_4.5.0 changed to GCC_4.6.0, 
unless you have some reason for actually needing different sets of 
symbols, since the Solaris version has already been reviewed as including 
all relevant symbols.  (Plus, as previously discussed, the appropriate 
%inherit and empty GCC_4.6.0 version in libgcc-std.ver - like those for 
GCC_4.1.0 which also only has symbols for some targets, for example.)

The Windows versions didn't need to allow for symbol versioning since that 
target doesn't support it.  The GNU/Linux versions got some symbol 
versions wrong initially so are not a good example to work from for other 
OSes.  Thus I recommend the Solaris version as an example from after the 
bugs in this area had been worked out.

-- 
Joseph S. Myers
joseph@codesourcery.com

Re: For testing: full __float128 patch

[Steve Kargl]

On Mon, Aug 30, 2010 at 09:51:09PM +0200, Uros Bizjak wrote:
> 
> So, attached and untested.
> 
> Uros.

I tested your patch on x86_64-*-freebsd.  My ppl and cloog
libraries were built with FreeBSD's system gcc (4.2.1).  All
tests failed due to missing symbols that are marked GCC_3.0.

> Index: gcc/config/i386/t-freebsd
> ===================================================================
> --- gcc/config/i386/t-freebsd	(revision 0)
> +++ gcc/config/i386/t-freebsd	(revision 0)
> @@ -0,0 +1 @@
> +SHLIB_MAPFILES = $(srcdir)/config/i386/libgcc-bsd.ver

Changing the = to += seems to fix the problem.  My
regression testing is still running, so I can't 
report stats.

-- 
Steve

Re: For testing: full __float128 patch

[Steve Kargl]

On Mon, Aug 30, 2010 at 03:29:20PM -0700, Steve Kargl wrote:
> On Mon, Aug 30, 2010 at 09:51:09PM +0200, Uros Bizjak wrote:
> > 
> > So, attached and untested.
> > 
> > Uros.
> 
> I tested your patch on x86_64-*-freebsd.  My ppl and cloog
> libraries were built with FreeBSD's system gcc (4.2.1).  All
> tests failed due to missing symbols that are marked GCC_3.0.
> 
> > Index: gcc/config/i386/t-freebsd
> > ===================================================================
> > --- gcc/config/i386/t-freebsd	(revision 0)
> > +++ gcc/config/i386/t-freebsd	(revision 0)
> > @@ -0,0 +1 @@
> > +SHLIB_MAPFILES = $(srcdir)/config/i386/libgcc-bsd.ver
> 
> Changing the = to += seems to fix the problem.  My
> regression testing is still running, so I can't 
> report stats.
> 

With my change to +=, on x86_64-*-freebsd I get

		=== gcc Summary ===

# of expected passes		72950
# of unexpected failures	74
# of unexpected successes	3
# of expected failures		215
# of unresolved testcases	3
# of unsupported tests		1081
/usr/home/sgk/gcc/obj4x/gcc/xgcc  version 4.6.0 20100830 (experimental) (GCC) 


		=== gfortran Summary ===

# of expected passes		36155
# of unexpected failures	4
# of unexpected successes	7
# of expected failures		59
# of unsupported tests		53
/usr/home/sgk/gcc/obj4x/gcc/testsuite/gfortran/../../gfortran  version 4.6.0 20100830 (experimental) (GCC) 


I quickly looked through the gcc.log file and noticed
a couple of ICE's but I believe those are unrelated to
the TFmode patch.

Note, I believe that I do not have the authority to
approve this patch.
-- 
Steve

Re: For testing: full __float128 patch

[Steve Kargl]

On Sun, Aug 29, 2010 at 09:01:49AM +0200, FX wrote:
> GCC >= 4.5 is needed. I'll add a check to libquad's configure.
> 
> FX
> 

Further testing on i386-*-freebsd runs into undefined symbols.

libquad.so: undefined reference to `__getf2'
libquad.so: undefined reference to `__eqtf2'
libquad.so: undefined reference to `__addtf3'
libquad.so: undefined reference to `__floatsitf'
libquad.so: undefined reference to `__trunctfxf2'
libquad.so: undefined reference to `__divtf3'
libquad.so: undefined reference to `__letf2'
libquad.so: undefined reference to `__unordtf2'
libquad.so: undefined reference to `__fixtfdi'
libquad.so: undefined reference to `__fixtfsi'
libquad.so: undefined reference to `__lttf2'
libquad.so: undefined reference to `__netf2'
libquad.so: undefined reference to `__extendxftf2'
libquad.so: undefined reference to `__extenddftf2'
libquad.so: undefined reference to `__multf3'
libquad.so: undefined reference to `__gttf2'
libquad.so: undefined reference to `__subtf3'
libgfortran.so: undefined reference to `__floatunditf'
libquad.so: undefined reference to `__trunctfdf2'


AFAICT, these should come from libgcc_s.so.1, but for whatever
reason these are missing.  I have been unable to find how
to induce gcc to build the required files. 

-- 
Steve

Re: For testing: full __float128 patch

[FX]

GCC >= 4.5 is needed. I'll add a check to libquad's configure.

FX

Le 29 août 2010 à 02:56, Steve Kargl <sgk@troutmask.apl.washington.edu> a écrit :

> On Sat, Aug 28, 2010 at 05:47:28PM -0700, Steve Kargl wrote:
>> On Sun, Aug 29, 2010 at 01:24:47AM +0200, FX wrote:
>>> Before I submit it officially for review, I want this patch to get some exposure while I clean up a few remaining dusty corners. To test the patch on x86_64-linux, proceed as follows:
>>> 
>>>  -- Get libquad from http://quatramaran.ens.fr/~coudert/tmp/libquad.tar.bz2 , then ./configure --prefix=/foo && make && make install 
>>>  -- Build GCC with the extra configure argument: --with-quad=/foo (to be given to the toplevel configure command, like --with-gmp, nothing fancy)
>>> 
>>> Things to test in particular:
>>> 
>>>  -- I've tested this on linux and Mac OS; although it's pretty much OS-insensitive, does it work on things like FreeBSD and Windows?
>>>  -- check carefully I/O support; in particular, the "output formatting" (i.e. writing out the values) involves a wrapper of my own around the gdtoa code, which was a bit painful to write, so bugs might have crawled in there
>>> 
>>> 
>> 
>> On i386-*-freebsd,
>> 
>> ./configure --prefix=$HOME/work
>> gmake
>> ...
>> 
>> -O2 -MT libgfortran_io.lo -MD -MP -MF .deps/libgfortran_io.Tpo -c -o libgfortran_io.lo libgfortran_io.c
>> libtool: compile:  gcc -DHAVE_CONFIG_H -I. -I.. -g -O2 -MT libgfortran_io.lo -MD -MP -MF .deps/libgfortran_io.Tpo -c libgfortran_io.c  -fPIC -DPIC -o .libs/libgfortran_io.o
>> In file included from libgfortran_io.c:1:
>> quad.h:8: error: unable to emulate 'TC'
> 
> gcc, here, is the FreeBSD system compiler.
> gcc (GCC) 4.2.1 20070719  [FreeBSD]
> 
> If I set my path to include ~/work/bin (where the 4.6.0 gcc lives)
> as the first directory to search, libquad builds fine.  So, what
> is the minimum required gcc version needs to build libquad?
> 
> -- 
> Steve

Re: For testing: full __float128 patch

[Steve Kargl]

On Sat, Aug 28, 2010 at 05:56:57PM -0700, Steve Kargl wrote:
> On Sat, Aug 28, 2010 at 05:47:28PM -0700, Steve Kargl wrote:
> > On Sun, Aug 29, 2010 at 01:24:47AM +0200, FX wrote:
> > > Before I submit it officially for review, I want this patch to get some exposure while I clean up a few remaining dusty corners. To test the patch on x86_64-linux, proceed as follows:
> > > 
> > >   -- Get libquad from http://quatramaran.ens.fr/~coudert/tmp/libquad.tar.bz2 , then ./configure --prefix=/foo && make && make install 
> > >   -- Build GCC with the extra configure argument: --with-quad=/foo (to be given to the toplevel configure command, like --with-gmp, nothing fancy)
> > > 
> > > Things to test in particular:
> > > 
> > >   -- I've tested this on linux and Mac OS; although it's pretty much OS-insensitive, does it work on things like FreeBSD and Windows?
> > >   -- check carefully I/O support; in particular, the "output formatting" (i.e. writing out the values) involves a wrapper of my own around the gdtoa code, which was a bit painful to write, so bugs might have crawled in there
> > > 
> > > 
> > 
> > On i386-*-freebsd,
> > 
> > ./configure --prefix=$HOME/work
> > gmake
> > ...
> > 
> >  -O2 -MT libgfortran_io.lo -MD -MP -MF .deps/libgfortran_io.Tpo -c -o libgfortran_io.lo libgfortran_io.c
> > libtool: compile:  gcc -DHAVE_CONFIG_H -I. -I.. -g -O2 -MT libgfortran_io.lo -MD -MP -MF .deps/libgfortran_io.Tpo -c libgfortran_io.c  -fPIC -DPIC -o .libs/libgfortran_io.o
> > In file included from libgfortran_io.c:1:
> > quad.h:8: error: unable to emulate 'TC'
> 
> gcc, here, is the FreeBSD system compiler.
> gcc (GCC) 4.2.1 20070719  [FreeBSD]
> 
> If I set my path to include ~/work/bin (where the 4.6.0 gcc lives)
> as the first directory to search, libquad builds fine.  So, what
> is the minimum required gcc version needs to build libquad?
> 
laptop:kargl[246] setenv | grep LD
LD_LIBRARY_PATH=/usr/home/kargl/work/lib
LD_RUN_PATH=/usr/home/kargl/work/lib

laptop:kargl[246] gmake check-gfortran
...
100s of FAILs
^C

From testsuite/gfortran/gfortran.log

FAIL: gfortran.dg/PR19754_2.f90  -O0  (test for excess errors)
Excess errors:
/usr/home/kargl/gcc/obj4x/i386-unknown-freebsd9.0/./libgfortran/.libs/libgfortran.so: undefined reference to `scalbnq'
/usr/home/kargl/gcc/obj4x/i386-unknown-freebsd9.0/./libgfortran/.libs/libgfortran.so: undefined reference to `__getf2'
/usr/home/kargl/gcc/obj4x/i386-unknown-freebsd9.0/./libgfortran/.libs/libgfortran.so: undefined reference to `__eqtf2'
/usr/home/kargl/gcc/obj4x/i386-unknown-freebsd9.0/./libgfortran/.libs/libgfortran.so: undefined reference to `__addtf3'
/usr/home/kargl/gcc/obj4x/i386-unknown-freebsd9.0/./libgfortran/.libs/libgfortran.so: undefined reference to `ldexpq'
/usr/home/kargl/gcc/obj4x/i386-unknown-freebsd9.0/./libgfortran/.libs/libgfortran.so: undefined reference to `acosq'
/usr/home/kargl/gcc/obj4x/i386-unknown-freebsd9.0/./libgfortran/.libs/libgfortran.so: undefined reference to `cabsq'
/usr/home/kargl/gcc/obj4x/i386-unknown-freebsd9.0/./libgfortran/.libs/libgfortran.so: undefined reference to `csqrtq'
/usr/home/kargl/gcc/obj4x/i386-unknown-freebsd9.0/./libgfortran/.libs/libgfortran.so: undefined reference to `__floatsitf'
/usr/home/kargl/gcc/obj4x/i386-unknown-freebsd9.0/./libgfortran/.libs/libgfortran.so: undefined reference to `__trunctfxf2'
/usr/home/kargl/gcc/obj4x/i386-unknown-freebsd9.0/./libgfortran/.libs/libgfortran.so: undefined reference to `clogq'
/usr/home/kargl/gcc/obj4x/i386-unknown-freebsd9.0/./libgfortran/.libs/libgfortran.so: undefined reference to `atan2q'
/usr/home/kargl/gcc/obj4x/i386-unknown-freebsd9.0/./libgfortran/.libs/libgfortran.so: undefined reference to `lroundq'
/usr/home/kargl/gcc/obj4x/i386-unknown-freebsd9.0/./libgfortran/.libs/libgfortran.so: undefined reference to `expq'
/usr/home/kargl/gcc/obj4x/i386-unknown-freebsd9.0/./libgfortran/.libs/libgfortran.so: undefined reference to `sinq'
/usr/home/kargl/gcc/obj4x/i386-unknown-freebsd9.0/./libgfortran/.libs/libgfortran.so: undefined reference to `libgfortran_dto q'
/usr/home/kargl/gcc/obj4x/i386-unknown-freebsd9.0/./libgfortran/.libs/libgfortran.so: undefined reference to `__divtf3'
/usr/home/kargl/gcc/obj4x/i386-unknown-freebsd9.0/./libgfortran/.libs/libgfortran.so: undefined reference to `frexpq'
/usr/home/kargl/gcc/obj4x/i386-unknown-freebsd9.0/./libgfortran/.libs/libgfortran.so: undefined reference to `__letf2'
/usr/home/kargl/gcc/obj4x/i386-unknown-freebsd9.0/./libgfortran/.libs/libgfortran.so: undefined reference to `sqrtq'
/usr/home/kargl/gcc/obj4x/i386-unknown-freebsd9.0/./libgfortran/.libs/libgfortran.so: undefined reference to `cosq'
/usr/home/kargl/gcc/obj4x/i386-unknown-freebsd9.0/./libgfortran/.libs/libgfortran.so: undefined reference to `nanq'
/usr/home/kargl/gcc/obj4x/i386-unknown-freebsd9.0/./libgfortran/.libs/libgfortran.so: undefined reference to `roundq'
/usr/home/kargl/gcc/obj4x/i386-unknown-freebsd9.0/./libgfortran/.libs/libgfortran.so: undefined reference to `libgfortran_str

-- 
Steve

Re: For testing: full __float128 patch

[Steve Kargl]

On Sat, Aug 28, 2010 at 05:47:28PM -0700, Steve Kargl wrote:
> On Sun, Aug 29, 2010 at 01:24:47AM +0200, FX wrote:
> > Before I submit it officially for review, I want this patch to get some exposure while I clean up a few remaining dusty corners. To test the patch on x86_64-linux, proceed as follows:
> > 
> >   -- Get libquad from http://quatramaran.ens.fr/~coudert/tmp/libquad.tar.bz2 , then ./configure --prefix=/foo && make && make install 
> >   -- Build GCC with the extra configure argument: --with-quad=/foo (to be given to the toplevel configure command, like --with-gmp, nothing fancy)
> > 
> > Things to test in particular:
> > 
> >   -- I've tested this on linux and Mac OS; although it's pretty much OS-insensitive, does it work on things like FreeBSD and Windows?
> >   -- check carefully I/O support; in particular, the "output formatting" (i.e. writing out the values) involves a wrapper of my own around the gdtoa code, which was a bit painful to write, so bugs might have crawled in there
> > 
> > 
> 
> On i386-*-freebsd,
> 
> ./configure --prefix=$HOME/work
> gmake
> ...
> 
>  -O2 -MT libgfortran_io.lo -MD -MP -MF .deps/libgfortran_io.Tpo -c -o libgfortran_io.lo libgfortran_io.c
> libtool: compile:  gcc -DHAVE_CONFIG_H -I. -I.. -g -O2 -MT libgfortran_io.lo -MD -MP -MF .deps/libgfortran_io.Tpo -c libgfortran_io.c  -fPIC -DPIC -o .libs/libgfortran_io.o
> In file included from libgfortran_io.c:1:
> quad.h:8: error: unable to emulate 'TC'

gcc, here, is the FreeBSD system compiler.
gcc (GCC) 4.2.1 20070719  [FreeBSD]

If I set my path to include ~/work/bin (where the 4.6.0 gcc lives)
as the first directory to search, libquad builds fine.  So, what
is the minimum required gcc version needs to build libquad?

-- 
Steve

Re: For testing: full __float128 patch

[Steve Kargl]

On Sun, Aug 29, 2010 at 01:24:47AM +0200, FX wrote:
> Before I submit it officially for review, I want this patch to get some exposure while I clean up a few remaining dusty corners. To test the patch on x86_64-linux, proceed as follows:
> 
>   -- Get libquad from http://quatramaran.ens.fr/~coudert/tmp/libquad.tar.bz2 , then ./configure --prefix=/foo && make && make install 
>   -- Build GCC with the extra configure argument: --with-quad=/foo (to be given to the toplevel configure command, like --with-gmp, nothing fancy)
> 
> Things to test in particular:
> 
>   -- I've tested this on linux and Mac OS; although it's pretty much OS-insensitive, does it work on things like FreeBSD and Windows?
>   -- check carefully I/O support; in particular, the "output formatting" (i.e. writing out the values) involves a wrapper of my own around the gdtoa code, which was a bit painful to write, so bugs might have crawled in there
> 
> 

On i386-*-freebsd,

./configure --prefix=$HOME/work
gmake
...

 -O2 -MT libgfortran_io.lo -MD -MP -MF .deps/libgfortran_io.Tpo -c -o libgfortran_io.lo libgfortran_io.c
libtool: compile:  gcc -DHAVE_CONFIG_H -I. -I.. -g -O2 -MT libgfortran_io.lo -MD -MP -MF .deps/libgfortran_io.Tpo -c libgfortran_io.c  -fPIC -DPIC -o .libs/libgfortran_io.o
In file included from libgfortran_io.c:1:
quad.h:8: error: unable to emulate 'TC'
quad.h:12: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'acosq'
quad.h:13: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'acoshq'
quad.h:14: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'asinq'
quad.h:15: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'asinhq'
quad.h:16: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'atanq'
quad.h:17: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'atanhq'
quad.h:18: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'atan2q'
quad.h:19: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'cbrtq'
quad.h:20: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'ceilq'
quad.h:21: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'copysignq'
quad.h:22: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'coshq'
quad.h:23: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'cosq'
quad.h:24: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'erfq'
quad.h:25: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'erfcq'
quad.h:26: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'expq'
quad.h:27: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'expm1q'
quad.h:28: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'fabsq'
quad.h:30: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'floorq'
quad.h:31: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'fmodq'
quad.h:32: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'frexpq'
quad.h:33: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'hypotq'
quad.h:34: warning: parameter names (without types) in function declaration
quad.h:36: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'j0q'
quad.h:37: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'j1q'
quad.h:38: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'jnq'
quad.h:39: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'ldexpq'
quad.h:40: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'lgammaq'
quad.h:41: warning: parameter names (without types) in function declaration
quad.h:42: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'logq'
quad.h:43: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'log10q'
quad.h:44: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'log1pq'
quad.h:45: warning: parameter names (without types) in function declaration
quad.h:46: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'modfq'
quad.h:47: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'nanq'
quad.h:48: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'nextafterq'
quad.h:49: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'powq'
quad.h:50: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'remainderq'
quad.h:51: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'rintq'
quad.h:52: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'roundq'
quad.h:53: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'scalblnq'
quad.h:54: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'scalbnq'
quad.h:56: error: expected ')' before '*' token
quad.h:57: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'sinhq'
quad.h:58: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'sinq'
quad.h:59: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'sqrtq'
quad.h:60: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'tanq'
quad.h:61: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'tanhq'
quad.h:62: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'tgammaq'
quad.h:63: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'truncq'
quad.h:64: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'y0q'
quad.h:65: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'y1q'
quad.h:66: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'ynq'
quad.h:71: error: expected '=', ',', ';', 'asm' or '__attribute__' before 'cargq'
quad.h:75: warning: parameter names (without types) in function declaration
quad.h:88: error: expected declaration specifiers or '...' before '__float128'
libgfortran_io.c:10: error: expected ';', ',' or ')' before 'x'
libgfortran_io.c:86: error: expected declaration specifiers or '...' before '__float128'
libgfortran_io.c: In function 'libgfortran_dtoaq':
libgfortran_io.c:90: error: 'x' undeclared (first use in this function)
libgfortran_io.c:90: error: (Each undeclared identifier is reported only once
libgfortran_io.c:90: error: for each function it appears in.)
gmake[2]: *** [libgfortran_io.lo] Error 1
gmake[2]: Leaving directory `/usr/home/kargl/gcc/libquad/src'
gmake[1]: *** [all-recursive] Error 1
gmake[1]: Leaving directory `/usr/home/kargl/gcc/libquad'
gmake: *** [all] Error 2

-- 
Steve

For testing: full __float128 patch

[FX]

[-- Attachment #1: Type: text/plain, Size: 2378 bytes --]

Before I submit it officially for review, I want this patch to get some exposure while I clean up a few remaining dusty corners. To test the patch on x86_64-linux, proceed as follows:

  -- Get libquad from http://quatramaran.ens.fr/~coudert/tmp/libquad.tar.bz2 , then ./configure --prefix=/foo && make && make install 
  -- Build GCC with the extra configure argument: --with-quad=/foo (to be given to the toplevel configure command, like --with-gmp, nothing fancy)

Things to test in particular:

  -- I've tested this on linux and Mac OS; although it's pretty much OS-insensitive, does it work on things like FreeBSD and Windows?
  -- check carefully I/O support; in particular, the "output formatting" (i.e. writing out the values) involves a wrapper of my own around the gdtoa code, which was a bit painful to write, so bugs might have crawled in there


Things that do not yet work, or may change:

  -- this is only for platform that support __float128 (targets without TFmode would break); I'm working on an updated front-end patch as proposed by Andrew and Tobias

  -- I realized that the exponentiation builtins are not properly handled by the front-end (all exponentiations involving __float128 or its complex counterpart give wrong code); I'll work on a fix

  -- Array intrinsics *do not work at all*. This is a limitation of the array descriptor structure, which I cannot overcome by myself. The current patch only changes these files (in_pack_generic.c, in_unpack_generic.c, pack_generic.c, spread_generic.c, unpack_generic.c, cshift0.c) to make them compile, but they will yield wrong results.

  -- the integration of libquad depends on how the discussion with the SC turns out; the current scheme is not nice at all in case of multilibs, see below

Please note: in case of a multilib build, unless you have a system which supports fat libraries (MacOS), only the variant that corresponds to the libquad you've built will have support enabled. Typically, on a x86_64 system where you've built libquad with the default 64-bit compiler, you'll only have libquad support in the -m64 variant of libgfortran, and not the -m32 variant.



All comment are of course welcome. I think the two preliminary commits allowed to keep the size of this final patch to a reasonable value under 1kLOC (less than 600 without regenerated files).

FX


[-- Attachment #2: full_testing.diff --]
[-- Type: application/octet-stream, Size: 28985 bytes --]

Index: gcc/fortran/trans-types.c
===================================================================
--- gcc/fortran/trans-types.c	(revision 163619)
+++ gcc/fortran/trans-types.c	(working copy)
@@ -414,7 +414,8 @@ gfc_init_kinds (void)
 	 support is done.  */
 	if (mode != TYPE_MODE (float_type_node)
 	  && (mode != TYPE_MODE (double_type_node))
-          && (mode != TYPE_MODE (long_double_type_node)))
+          && (mode != TYPE_MODE (long_double_type_node))
+	  && (mode != TFmode))
 	continue;
 
       /* Let the kind equal the precision divided by 8, rounding up.  Again,
Index: libgfortran/configure
===================================================================
--- libgfortran/configure	(revision 163619)
+++ libgfortran/configure	(working copy)
@@ -603,6 +603,8 @@ LTLIBOBJS
 LIBOBJS
 IEEE_FLAGS
 FPU_HOST_HEADER
+GFOR_USE_QUAD_LIB_FALSE
+GFOR_USE_QUAD_LIB_TRUE
 extra_ldflags_libgfortran
 ac_ct_FC
 FCFLAGS
@@ -691,6 +693,8 @@ host_os
 host_vendor
 host_cpu
 host
+QUADINC
+QUADLIBS
 onestep
 onestep_FALSE
 onestep_TRUE
@@ -746,6 +750,9 @@ enable_option_checking
 with_build_libsubdir
 enable_version_specific_runtime_libs
 enable_intermodule
+with_quad
+with_quad_include
+with_quad_lib
 enable_maintainer_mode
 enable_multilib
 enable_dependency_tracking
@@ -1404,6 +1411,11 @@ Optional Packages:
   --with-PACKAGE[=ARG]    use PACKAGE [ARG=yes]
   --without-PACKAGE       do not use PACKAGE (same as --with-PACKAGE=no)
   --with-build-libsubdir=DIR  Directory where to find libraries for build system
+  --with-quad=PATH         specify prefix directory for the installed quad package.
+                          Equivalent to --with-quad-include=PATH/include
+                          plus --with-quad-lib=PATH/lib
+  --with-quad-include=PATH specify directory for installed quad include files
+  --with-quad-lib=PATH     specify directory for the installed quad library
   --with-pic              try to use only PIC/non-PIC objects [default=use
                           both]
   --with-gnu-ld           assume the C compiler uses GNU ld [default=no]
@@ -2733,6 +2745,40 @@ fi
 
 
 
+# Build with a quad-float support library
+
+# Check whether --with-quad was given.
+if test "${with_quad+set}" = set; then :
+  withval=$with_quad;
+fi
+
+
+# Check whether --with-quad_include was given.
+if test "${with_quad_include+set}" = set; then :
+  withval=$with_quad_include;
+fi
+
+
+# Check whether --with-quad_lib was given.
+if test "${with_quad_lib+set}" = set; then :
+  withval=$with_quad_lib;
+fi
+
+
+if test "x$with_quad" != x; then
+  QUADLIBS="-L$with_quad/lib $QUADLIBS"
+  QUADINC="-I$with_quad/include $QUADINC"
+fi
+if test "x$with_quad_include" != x; then
+  QUADINC="-I$with_quad_include $QUADINC"
+fi
+if test "x$with_quad_lib" != x; then
+  QUADLIBS="-L$with_quad_lib $QUADLIBS"
+fi
+
+
+
+
 # Gets build, host, target, *_vendor, *_cpu, *_os, etc.
 #
 # You will slowly go insane if you do not grok the following fact:  when
@@ -11362,7 +11408,7 @@ else
   lt_dlunknown=0; lt_dlno_uscore=1; lt_dlneed_uscore=2
   lt_status=$lt_dlunknown
   cat > conftest.$ac_ext <<_LT_EOF
-#line 11365 "configure"
+#line 11411 "configure"
 #include "confdefs.h"
 
 #if HAVE_DLFCN_H
@@ -11468,7 +11514,7 @@ else
   lt_dlunknown=0; lt_dlno_uscore=1; lt_dlneed_uscore=2
   lt_status=$lt_dlunknown
   cat > conftest.$ac_ext <<_LT_EOF
-#line 11471 "configure"
+#line 11517 "configure"
 #include "confdefs.h"
 
 #if HAVE_DLFCN_H
@@ -24468,6 +24514,129 @@ $as_echo "#define HAVE_BROKEN_POWF 1" >>
   fi
 
 
+# Check whether we have a __float128 type
+
+  { $as_echo "$as_me:${as_lineno-$LINENO}: checking whether we have a __float128 type" >&5
+$as_echo_n "checking whether we have a __float128 type... " >&6; }
+if test "${libgfor_cv_have_float128+set}" = set; then :
+  $as_echo_n "(cached) " >&6
+else
+
+    if test x$gcc_no_link = xyes; then
+  as_fn_error "Link tests are not allowed after GCC_NO_EXECUTABLES." "$LINENO" 5
+fi
+cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+
+/* no header */
+
+int
+main ()
+{
+
+  typedef _Complex float __attribute__((mode(TC))) __complex128;
+
+  __float128 x;
+  x = __builtin_huge_valq() - 2.e1000Q;
+
+  __complex128 z1, z2;
+  z1 = x;
+  z2 = 2.Q;
+
+  z1 /= z2;
+
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_c_try_link "$LINENO"; then :
+  eval "libgfor_cv_have_float128=yes"
+else
+  eval "libgfor_cv_have_float128=no"
+fi
+rm -f core conftest.err conftest.$ac_objext \
+    conftest$ac_exeext conftest.$ac_ext
+
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $libgfor_cv_have_float128" >&5
+$as_echo "$libgfor_cv_have_float128" >&6; }
+
+  if test "x$libgfor_cv_have_float128" = xyes; then
+
+$as_echo "#define HAVE_FLOAT128 1" >>confdefs.h
+
+  fi
+
+
+# Check whether we have a quad-precision float library
+
+  { $as_echo "$as_me:${as_lineno-$LINENO}: checking whether we have a quad-precision float library" >&5
+$as_echo_n "checking whether we have a quad-precision float library... " >&6; }
+if test "${libgfor_cv_have_quad_lib+set}" = set; then :
+  $as_echo_n "(cached) " >&6
+else
+
+    save_LIBS="$LIBS"
+    LIBS="$LIBS $QUADLIBS -lquad"
+    save_CFLAGS="$CFLAGS"
+    CFLAGS="$CFLAGS $QUADINC"
+    if test x$gcc_no_link = xyes; then
+  as_fn_error "Link tests are not allowed after GCC_NO_EXECUTABLES." "$LINENO" 5
+fi
+cat confdefs.h - <<_ACEOF >conftest.$ac_ext
+/* end confdefs.h.  */
+
+  #include <quad.h>
+
+int
+main ()
+{
+
+  __complex128 z;
+  __float128 x;
+
+  z = 2.Q;
+  x = cargq (x);
+  x = sqrtq (x);
+
+  char buffer[100];
+  libgfortran_dtoaq (buffer, 30, 40, x);
+
+  ;
+  return 0;
+}
+_ACEOF
+if ac_fn_c_try_link "$LINENO"; then :
+  eval "libgfor_cv_have_quad_lib=yes"
+else
+  eval "libgfor_cv_have_quad_lib=no"
+fi
+rm -f core conftest.err conftest.$ac_objext \
+    conftest$ac_exeext conftest.$ac_ext
+
+    LIBS="$save_LIBS"
+    CFLAGS="$save_CFLAGS"
+
+fi
+{ $as_echo "$as_me:${as_lineno-$LINENO}: result: $libgfor_cv_have_quad_lib" >&5
+$as_echo "$libgfor_cv_have_quad_lib" >&6; }
+
+  if test "x$libgfor_cv_have_quad_lib" = xyes; then
+
+$as_echo "#define GFC_WITH_QUAD_LIB 1" >>confdefs.h
+
+  fi
+
+     if test "x$libgfor_cv_have_quad_lib" = xyes; then
+  GFOR_USE_QUAD_LIB_TRUE=
+  GFOR_USE_QUAD_LIB_FALSE='#'
+else
+  GFOR_USE_QUAD_LIB_TRUE='#'
+  GFOR_USE_QUAD_LIB_FALSE=
+fi
+
+
+
 # Check for GNU libc feenableexcept
 { $as_echo "$as_me:${as_lineno-$LINENO}: checking for feenableexcept in -lm" >&5
 $as_echo_n "checking for feenableexcept in -lm... " >&6; }
@@ -25150,6 +25319,10 @@ if test -z "${LIBGFOR_USE_SYMVER_SUN_TRU
   as_fn_error "conditional \"LIBGFOR_USE_SYMVER_SUN\" was never defined.
 Usually this means the macro was only invoked conditionally." "$LINENO" 5
 fi
+if test -z "${GFOR_USE_QUAD_LIB_TRUE}" && test -z "${GFOR_USE_QUAD_LIB_FALSE}"; then
+  as_fn_error "conditional \"GFOR_USE_QUAD_LIB\" was never defined.
+Usually this means the macro was only invoked conditionally." "$LINENO" 5
+fi
 
 : ${CONFIG_STATUS=./config.status}
 ac_write_fail=0
Index: libgfortran/Makefile.in
===================================================================
--- libgfortran/Makefile.in	(revision 163619)
+++ libgfortran/Makefile.in	(working copy)
@@ -381,6 +381,8 @@ PACKAGE_TARNAME = @PACKAGE_TARNAME@
 PACKAGE_URL = @PACKAGE_URL@
 PACKAGE_VERSION = @PACKAGE_VERSION@
 PATH_SEPARATOR = @PATH_SEPARATOR@
+QUADINC = @QUADINC@
+QUADLIBS = @QUADLIBS@
 RANLIB = @RANLIB@
 SECTION_FLAGS = @SECTION_FLAGS@
 SED = @SED@
@@ -463,12 +465,16 @@ gcc_version := $(shell cat $(top_srcdir)
 @LIBGFOR_USE_SYMVER_FALSE@version_dep = 
 @LIBGFOR_USE_SYMVER_GNU_TRUE@@LIBGFOR_USE_SYMVER_TRUE@version_dep = $(srcdir)/gfortran.map
 @LIBGFOR_USE_SYMVER_SUN_TRUE@@LIBGFOR_USE_SYMVER_TRUE@version_dep = gfortran.map-sun
+@GFOR_USE_QUAD_LIB_FALSE@quad_link = 
+
+# Compile and link with quad float library if present
+@GFOR_USE_QUAD_LIB_TRUE@quad_link = -lquad
 LTLDFLAGS = $(shell $(SHELL) $(top_srcdir)/../libtool-ldflags $(LDFLAGS)) \
 	    -no-undefined -bindir "$(bindir)"
 
 toolexeclib_LTLIBRARIES = libgfortran.la
 libgfortran_la_LINK = $(LINK) $(libgfortran_la_LDFLAGS)
-libgfortran_la_LDFLAGS = -version-info `grep -v '^\#' $(srcdir)/libtool-version` $(LTLDFLAGS) -lm $(extra_ldflags_libgfortran) $(version_arg)
+libgfortran_la_LDFLAGS = -version-info `grep -v '^\#' $(srcdir)/libtool-version` $(LTLDFLAGS) $(QUADLIBS) $(quad_link) -lm $(extra_ldflags_libgfortran) $(version_arg)
 libgfortran_la_DEPENDENCIES = $(version_dep)
 myexeclib_LTLIBRARIES = libgfortranbegin.la
 myexeclibdir = $(libdir)/gcc/$(target_alias)/$(gcc_version)$(MULTISUBDIR)
@@ -477,7 +483,8 @@ libgfortranbegin_la_LDFLAGS = -static
 libgfortranbegin_la_LINK = $(LINK) $(libgfortranbegin_la_LDFLAGS)
 AM_CPPFLAGS = -iquote$(srcdir)/io -I$(srcdir)/$(MULTISRCTOP)../gcc \
 	      -I$(srcdir)/$(MULTISRCTOP)../gcc/config \
-	      -I$(MULTIBUILDTOP)../../$(host_subdir)/gcc -D_GNU_SOURCE
+	      -I$(MULTIBUILDTOP)../../$(host_subdir)/gcc -D_GNU_SOURCE \
+	      $(QUADINC)
 
 gfor_io_src = \
 io/close.c \
Index: libgfortran/runtime/in_pack_generic.c
===================================================================
--- libgfortran/runtime/in_pack_generic.c	(revision 163619)
+++ libgfortran/runtime/in_pack_generic.c	(working copy)
@@ -85,10 +85,6 @@ internal_pack (gfc_array_char * source)
       return internal_pack_r10 ((gfc_array_r10 *) source);
 #endif
 
-#if defined (HAVE_GFC_REAL_16)
-    case GFC_DTYPE_REAL_16:
-      return internal_pack_r16 ((gfc_array_r16 *) source);
-#endif
     case GFC_DTYPE_COMPLEX_4:
       return internal_pack_c4 ((gfc_array_c4 *) source);
 	
@@ -100,10 +96,6 @@ internal_pack (gfc_array_char * source)
       return internal_pack_c10 ((gfc_array_c10 *) source);
 #endif
 
-#if defined (HAVE_GFC_COMPLEX_16)
-    case GFC_DTYPE_COMPLEX_16:
-      return internal_pack_c16 ((gfc_array_c16 *) source);
-#endif
 
     case GFC_DTYPE_DERIVED_2:
       if (GFC_UNALIGNED_2(source->data))
Index: libgfortran/runtime/in_unpack_generic.c
===================================================================
--- libgfortran/runtime/in_unpack_generic.c	(revision 163619)
+++ libgfortran/runtime/in_unpack_generic.c	(working copy)
@@ -95,11 +95,6 @@ internal_unpack (gfc_array_char * d, con
       return;
 #endif
 
-#if defined(HAVE_GFC_REAL_16)
-    case GFC_DTYPE_REAL_16:
-      internal_unpack_r16 ((gfc_array_r16 *) d, (const GFC_REAL_16 *) s);
-      return;
-#endif
     case GFC_DTYPE_COMPLEX_4:
       internal_unpack_c4 ((gfc_array_c4 *)d, (const GFC_COMPLEX_4 *)s);
       return;
@@ -114,11 +109,6 @@ internal_unpack (gfc_array_char * d, con
       return;
 #endif
 
-#if defined(HAVE_GFC_COMPLEX_16)
-    case GFC_DTYPE_COMPLEX_16:
-      internal_unpack_c16 ((gfc_array_c16 *) d, (const GFC_COMPLEX_16 *) s);
-      return;
-#endif
     case GFC_DTYPE_DERIVED_2:
       if (GFC_UNALIGNED_2(d->data) || GFC_UNALIGNED_2(s))
 	break;
Index: libgfortran/intrinsics/pack_generic.c
===================================================================
--- libgfortran/intrinsics/pack_generic.c	(revision 163619)
+++ libgfortran/intrinsics/pack_generic.c	(working copy)
@@ -310,12 +310,6 @@ pack (gfc_array_char *ret, const gfc_arr
       return;
 #endif
 
-#ifdef HAVE_GFC_REAL_16
-    case GFC_DTYPE_REAL_16:
-      pack_r16 ((gfc_array_r16 *) ret, (gfc_array_r16 *) array,
-		(gfc_array_l1 *) mask, (gfc_array_r16 *) vector);
-      return;
-#endif
     case GFC_DTYPE_COMPLEX_4:
       pack_c4 ((gfc_array_c4 *) ret, (gfc_array_c4 *) array,
 	       (gfc_array_l1 *) mask, (gfc_array_c4 *) vector);
@@ -333,12 +327,6 @@ pack (gfc_array_char *ret, const gfc_arr
       return;
 #endif
 
-#ifdef HAVE_GFC_COMPLEX_16
-    case GFC_DTYPE_COMPLEX_16:
-      pack_c16 ((gfc_array_c16 *) ret, (gfc_array_c16 *) array,
-		(gfc_array_l1 *) mask, (gfc_array_c16 *) vector);
-      return;
-#endif
 
       /* For derived types, let's check the actual alignment of the
 	 data pointers.  If they are aligned, we can safely call
Index: libgfortran/intrinsics/erfc_scaled_inc.c
===================================================================
--- libgfortran/intrinsics/erfc_scaled_inc.c	(revision 163619)
+++ libgfortran/intrinsics/erfc_scaled_inc.c	(working copy)
@@ -39,7 +39,7 @@ see the files COPYING3 and COPYING.RUNTI
 # define EXP(x) exp(x)
 # define TRUNC(x) trunc(x)
 
-#else
+#elif (KIND == 10) || (KIND == 16 && defined(GFC_REAL_16_IS_LONG_DOUBLE))
 
 # ifdef HAVE_EXPL
 #  define EXP(x) expl(x)
@@ -48,6 +48,15 @@ see the files COPYING3 and COPYING.RUNTI
 #  define TRUNC(x) truncl(x)
 # endif
 
+#elif (KIND == 16 && defined(GFC_REAL_16_IS_FLOAT128))
+
+#  define EXP(x) expq(x)
+#  define TRUNC(x) truncq(x)
+
+#else
+
+# error "What exactly is it that you want me to do?"
+
 #endif
 
 #if defined(EXP) && defined(TRUNC)
Index: libgfortran/intrinsics/cshift0.c
===================================================================
--- libgfortran/intrinsics/cshift0.c	(revision 163619)
+++ libgfortran/intrinsics/cshift0.c	(working copy)
@@ -141,13 +141,6 @@ cshift0 (gfc_array_char * ret, const gfc
       return;
 #endif
 
-#ifdef HAVE_GFC_REAL_16
-    case GFC_DTYPE_REAL_16:
-      cshift0_r16 ((gfc_array_r16 *)ret, (gfc_array_r16 *) array, shift,
-		   which);
-      return;
-#endif
-
     case GFC_DTYPE_COMPLEX_4:
       cshift0_c4 ((gfc_array_c4 *)ret, (gfc_array_c4 *) array, shift, which);
       return;
@@ -163,12 +156,6 @@ cshift0 (gfc_array_char * ret, const gfc
       return;
 #endif
 
-#ifdef HAVE_GFC_COMPLEX_16
-    case GFC_DTYPE_COMPLEX_16:
-      cshift0_c16 ((gfc_array_c16 *)ret, (gfc_array_c16 *) array, shift,
-		   which);
-      return;
-#endif
 
     default:
       break;
Index: libgfortran/intrinsics/spread_generic.c
===================================================================
--- libgfortran/intrinsics/spread_generic.c	(revision 163619)
+++ libgfortran/intrinsics/spread_generic.c	(working copy)
@@ -329,12 +329,6 @@ spread (gfc_array_char *ret, const gfc_a
       return;
 #endif
 
-#ifdef GFC_HAVE_REAL_16
-    case GFC_DTYPE_REAL_16:
-      spread_r16 ((gfc_array_r16 *) ret, (gfc_array_r16 *) source,
-		 *along, *pncopies);
-      return;
-#endif
 
     case GFC_DTYPE_COMPLEX_4:
       spread_c4 ((gfc_array_c4 *) ret, (gfc_array_c4 *) source,
@@ -353,13 +347,6 @@ spread (gfc_array_char *ret, const gfc_a
       return;
 #endif
 
-#ifdef GFC_HAVE_COMPLEX_16
-    case GFC_DTYPE_COMPLEX_16:
-      spread_c16 ((gfc_array_c16 *) ret, (gfc_array_c16 *) source,
-		 *along, *pncopies);
-      return;
-#endif
-
     case GFC_DTYPE_DERIVED_2:
       if (GFC_UNALIGNED_2(ret->data) || GFC_UNALIGNED_2(source->data))
 	break;
@@ -508,13 +495,6 @@ spread_scalar (gfc_array_char *ret, cons
       return;
 #endif
 
-#ifdef HAVE_GFC_REAL_16
-    case GFC_DTYPE_REAL_16:
-      spread_scalar_r16 ((gfc_array_r16 *) ret, (GFC_REAL_16 *) source,
-			*along, *pncopies);
-      return;
-#endif
-
     case GFC_DTYPE_COMPLEX_4:
       spread_scalar_c4 ((gfc_array_c4 *) ret, (GFC_COMPLEX_4 *) source,
 			*along, *pncopies);
@@ -532,13 +512,6 @@ spread_scalar (gfc_array_char *ret, cons
       return;
 #endif
 
-#ifdef HAVE_GFC_COMPLEX_16
-    case GFC_DTYPE_COMPLEX_16:
-      spread_scalar_c16 ((gfc_array_c16 *) ret, (GFC_COMPLEX_16 *) source,
-			*along, *pncopies);
-      return;
-#endif
-
     case GFC_DTYPE_DERIVED_2:
       if (GFC_UNALIGNED_2(ret->data) || GFC_UNALIGNED_2(source))
 	break;
Index: libgfortran/intrinsics/unpack_generic.c
===================================================================
--- libgfortran/intrinsics/unpack_generic.c	(revision 163619)
+++ libgfortran/intrinsics/unpack_generic.c	(working copy)
@@ -267,13 +267,6 @@ unpack1 (gfc_array_char *ret, const gfc_
 	  return;
 #endif
 
-#ifdef HAVE_GFC_REAL_16
-    case GFC_DTYPE_REAL_16:
-      unpack1_r16 ((gfc_array_r16 *) ret, (gfc_array_r16 *) vector,
-		   mask, (gfc_array_r16 *) field);
-      return;
-#endif
-
     case GFC_DTYPE_COMPLEX_4:
       unpack1_c4 ((gfc_array_c4 *) ret, (gfc_array_c4 *) vector,
 		  mask, (gfc_array_c4 *) field);
@@ -291,13 +284,6 @@ unpack1 (gfc_array_char *ret, const gfc_
       return;
 #endif
 
-#ifdef HAVE_GFC_COMPLEX_16
-    case GFC_DTYPE_COMPLEX_16:
-      unpack1_c16 ((gfc_array_c16 *) ret, (gfc_array_c16 *) vector,
-		   mask, (gfc_array_c16 *) field);
-      return;
-#endif
-
     case GFC_DTYPE_DERIVED_2:
       if (GFC_UNALIGNED_2(ret->data) || GFC_UNALIGNED_2(vector->data)
 	  || GFC_UNALIGNED_2(field->data))
@@ -460,13 +446,6 @@ unpack0 (gfc_array_char *ret, const gfc_
       return;
 #endif
 
-#ifdef HAVE_GFC_REAL_16
-    case GFC_DTYPE_REAL_16:
-      unpack0_r16 ((gfc_array_r16 *) ret, (gfc_array_r16 *) vector,
-		   mask, (GFC_REAL_16 *) field);
-      return;
-#endif
-
     case GFC_DTYPE_COMPLEX_4:
       unpack0_c4 ((gfc_array_c4 *) ret, (gfc_array_c4 *) vector,
 		  mask, (GFC_COMPLEX_4 *) field);
@@ -484,12 +463,6 @@ unpack0 (gfc_array_char *ret, const gfc_
       return;
 #endif
 
-#ifdef HAVE_GFC_COMPLEX_16
-    case GFC_DTYPE_COMPLEX_16:
-      unpack0_c16 ((gfc_array_c16 *) ret, (gfc_array_c16 *) vector,
-		   mask, (GFC_COMPLEX_16 *) field);
-      return;
-#endif
     case GFC_DTYPE_DERIVED_2:
       if (GFC_UNALIGNED_2(ret->data) || GFC_UNALIGNED_2(vector->data)
 	  || GFC_UNALIGNED_2(field))
Index: libgfortran/configure.ac
===================================================================
--- libgfortran/configure.ac	(revision 163619)
+++ libgfortran/configure.ac	(working copy)
@@ -34,6 +34,27 @@ AC_MSG_RESULT($enable_intermodule)
 AM_CONDITIONAL(onestep,[test x$onestep = x-onestep])
 AC_SUBST(onestep)
 
+# Build with a quad-float support library
+AC_ARG_WITH(quad, [  --with-quad=PATH         specify prefix directory for the installed quad package.
+                          Equivalent to --with-quad-include=PATH/include
+                          plus --with-quad-lib=PATH/lib])
+AC_ARG_WITH(quad_include, [  --with-quad-include=PATH specify directory for installed quad include files])
+AC_ARG_WITH(quad_lib, [  --with-quad-lib=PATH     specify directory for the installed quad library])
+
+if test "x$with_quad" != x; then
+  QUADLIBS="-L$with_quad/lib $QUADLIBS"
+  QUADINC="-I$with_quad/include $QUADINC"
+fi
+if test "x$with_quad_include" != x; then
+  QUADINC="-I$with_quad_include $QUADINC"
+fi
+if test "x$with_quad_lib" != x; then
+  QUADLIBS="-L$with_quad_lib $QUADLIBS"
+fi
+
+AC_SUBST(QUADLIBS)
+AC_SUBST(QUADINC)
+
 # Gets build, host, target, *_vendor, *_cpu, *_os, etc.
 #
 # You will slowly go insane if you do not grok the following fact:  when
@@ -468,6 +489,12 @@ LIBGFOR_CHECK_MINGW_SNPRINTF
 # Check for a broken powf implementation
 LIBGFOR_CHECK_FOR_BROKEN_POWF
 
+# Check whether we have a __float128 type
+LIBGFOR_CHECK_FLOAT128
+
+# Check whether we have a quad-precision float library
+LIBGFOR_CHECK_QUAD_LIB
+
 # Check for GNU libc feenableexcept
 AC_CHECK_LIB([m],[feenableexcept],[have_feenableexcept=yes AC_DEFINE([HAVE_FEENABLEEXCEPT],[1],[libm includes feenableexcept])])
 
Index: libgfortran/libgfortran.h
===================================================================
--- libgfortran/libgfortran.h	(revision 163619)
+++ libgfortran/libgfortran.h	(working copy)
@@ -314,6 +314,30 @@ internal_proto(big_endian);
   (GFC_INTEGER_16)((((GFC_UINTEGER_16)1) << 127) - 1)
 #endif
 
+
+/* What are the C types corresponding to the real(kind=10) and
+   real(kind=16) types? We currently rely on the following assumptions:
+     -- if real(kind=10) exists, i.e. if HAVE_GFC_REAL_10 is defined,
+        then it is necessarily the "long double" type
+     -- if real(kind=16) exists, then:
+         * if HAVE_GFC_REAL_10, real(kind=16) is "__float128"
+	 * otherwise, real(kind=16) is "long double"
+   To allow to change this in the future, we create the
+   GFC_REAL_16_IS_FLOAT128 macro that is used throughout libgfortran.  */
+#if defined(HAVE_GFC_REAL_16)
+# if defined(HAVE_GFC_REAL_10)
+#  define GFC_REAL_16_IS_FLOAT128
+#  if !defined(HAVE_FLOAT128)
+#   error "Where has __float128 gone?"
+#  endif
+# else
+#  define GFC_REAL_16_IS_LONG_DOUBLE
+# endif
+#endif
+
+//#include "quad.h"
+
+
 /* M{IN,AX}{LOC,VAL} need also infinities and NaNs if supported.  */
 
 #ifdef __FLT_HAS_INFINITY__
@@ -327,7 +351,11 @@ internal_proto(big_endian);
 #  define GFC_REAL_10_INFINITY __builtin_infl ()
 # endif
 # ifdef HAVE_GFC_REAL_16
-#  define GFC_REAL_16_INFINITY __builtin_infl ()
+#  ifdef GFC_REAL_16_IS_LONG_DOUBLE
+#   define GFC_REAL_16_INFINITY __builtin_infl ()
+#  else
+#   define GFC_REAL_16_INFINITY __builtin_infq ()
+#  endif
 # endif
 #endif
 #ifdef __FLT_HAS_QUIET_NAN__
@@ -341,7 +369,11 @@ internal_proto(big_endian);
 #  define GFC_REAL_10_QUIET_NAN __builtin_nanl ("")
 # endif
 # ifdef HAVE_GFC_REAL_16
-#  define GFC_REAL_16_QUIET_NAN __builtin_nanl ("")
+#  ifdef GFC_REAL_16_IS_LONG_DOUBLE
+#   define GFC_REAL_16_QUIET_NAN __builtin_nanl ("")
+#  else
+#   define GFC_REAL_16_QUIET_NAN nanq ("")
+#  endif
 # endif
 #endif
 
Index: libgfortran/config.h.in
===================================================================
--- libgfortran/config.h.in	(revision 163619)
+++ libgfortran/config.h.in	(working copy)
@@ -1,5 +1,8 @@
 /* config.h.in.  Generated from configure.ac by autoheader.  */
 
+/* Define if we have a quad-precision float library. */
+#undef GFC_WITH_QUAD_LIB
+
 /* Define to 0 if the target shouldn't use #pragma weak */
 #undef GTHREAD_USE_WEAK
 
@@ -390,6 +393,9 @@
 /* Define to 1 if you have the <fenv.h> header file. */
 #undef HAVE_FENV_H
 
+/* Define if have a __float128 type. */
+#undef HAVE_FLOAT128
+
 /* Define to 1 if you have the <floatingpoint.h> header file. */
 #undef HAVE_FLOATINGPOINT_H
 
Index: libgfortran/acinclude.m4
===================================================================
--- libgfortran/acinclude.m4	(revision 163619)
+++ libgfortran/acinclude.m4	(working copy)
@@ -388,3 +388,66 @@ esac])
     AC_DEFINE(HAVE_BROKEN_POWF, 1, [Define if powf is broken.])
   fi
 ])
+
+dnl Check whether we have a __float128 type
+AC_DEFUN([LIBGFOR_CHECK_FLOAT128], [
+  AC_CACHE_CHECK([whether we have a __float128 type],
+                 libgfor_cv_have_float128, [
+    AC_TRY_LINK([
+/* no header */
+],[
+  typedef _Complex float __attribute__((mode(TC))) __complex128;
+
+  __float128 x;
+  x = __builtin_huge_valq() - 2.e1000Q;
+
+  __complex128 z1, z2;
+  z1 = x;
+  z2 = 2.Q;
+
+  z1 /= z2;
+],
+    eval "libgfor_cv_have_float128=yes",
+    eval "libgfor_cv_have_float128=no")
+  ])
+
+  if test "x$libgfor_cv_have_float128" = xyes; then
+    AC_DEFINE(HAVE_FLOAT128, 1, [Define if have a __float128 type.])
+  fi
+])
+
+dnl Check whether we have a quad-precision float library
+AC_DEFUN([LIBGFOR_CHECK_QUAD_LIB], [
+  AC_CACHE_CHECK([whether we have a quad-precision float library],
+                 libgfor_cv_have_quad_lib, [
+    save_LIBS="$LIBS"
+    LIBS="$LIBS $QUADLIBS -lquad"
+    save_CFLAGS="$CFLAGS"
+    CFLAGS="$CFLAGS $QUADINC"
+    AC_TRY_LINK([
+  #include <quad.h>
+],[
+  __complex128 z;
+  __float128 x;
+
+  z = 2.Q;
+  x = cargq (x);
+  x = sqrtq (x);
+  
+  char buffer[100];
+  libgfortran_dtoaq (buffer, 30, 40, x);
+],
+      eval "libgfor_cv_have_quad_lib=yes",
+      eval "libgfor_cv_have_quad_lib=no")
+
+    LIBS="$save_LIBS"
+    CFLAGS="$save_CFLAGS"
+  ])
+
+  if test "x$libgfor_cv_have_quad_lib" = xyes; then
+    AC_DEFINE(GFC_WITH_QUAD_LIB, 1, [Define if we have a quad-precision float library.])
+  fi
+
+  dnl We need a conditional for the Makefile
+  AM_CONDITIONAL(GFOR_USE_QUAD_LIB, [test "x$libgfor_cv_have_quad_lib" = xyes])
+])
Index: libgfortran/io/read.c
===================================================================
--- libgfortran/io/read.c	(revision 163619)
+++ libgfortran/io/read.c	(working copy)
@@ -162,10 +162,16 @@ convert_real (st_parameter_dt *dtp, void
       break;
 #endif
 
-#if defined(HAVE_GFC_REAL_16) && defined (HAVE_STRTOLD)
+#if defined(HAVE_GFC_REAL_16)
+# if defined(GFC_REAL_16_IS_FLOAT128)
+    case 16:
+      libgfortran_strtopQ (buffer, NULL, dest);
+      break;
+# elif defined(HAVE_STRTOLD)
     case 16:
       *((GFC_REAL_16*) dest) = gfc_strtold (buffer, NULL);
       break;
+# endif
 #endif
 
     default:
Index: libgfortran/io/write.c
===================================================================
--- libgfortran/io/write.c	(revision 163619)
+++ libgfortran/io/write.c	(working copy)
@@ -1459,6 +1459,7 @@ set_fnode_default (st_parameter_dt *dtp,
 /* Output a real number with default format.
    This is 1PG14.7E2 for REAL(4), 1PG23.15E3 for REAL(8),
    1PG28.19E4 for REAL(10) and 1PG43.34E4 for REAL(16).  */
+// FX -- FIXME: should we change the default format for __float128-real(16)?
 
 void
 write_real (st_parameter_dt *dtp, const char *source, int length)
Index: libgfortran/io/write_float.def
===================================================================
--- libgfortran/io/write_float.def	(revision 163619)
+++ libgfortran/io/write_float.def	(working copy)
@@ -973,6 +973,11 @@ sprintf (buffer, "%+-#" STR(MIN_FIELD_WI
 
 #endif
 
+#if defined(GFC_REAL_16_IS_FLOAT128)
+#define DTOAQ \
+libgfortran_dtoaq (buffer, size, ndigits - 1, tmp);
+#endif
+
 #define WRITE_FLOAT(x,y)\
 {\
 	GFC_REAL_ ## x tmp;\
@@ -1057,7 +1062,11 @@ write_float (st_parameter_dt *dtp, const
 #endif
 #ifdef HAVE_GFC_REAL_16
     case 16:
+# ifdef GFC_REAL_16_IS_FLOAT128
+      WRITE_FLOAT(16,Q)
+# else
       WRITE_FLOAT(16,L)
+# endif
       break;
 #endif
     default:
Index: libgfortran/Makefile.am
===================================================================
--- libgfortran/Makefile.am	(revision 163619)
+++ libgfortran/Makefile.am	(working copy)
@@ -30,12 +30,19 @@ version_arg =
 version_dep =
 endif
 
+# Compile and link with quad float library if present
+if GFOR_USE_QUAD_LIB
+quad_link = -lquad
+else
+quad_link =
+endif
+
 LTLDFLAGS = $(shell $(SHELL) $(top_srcdir)/../libtool-ldflags $(LDFLAGS)) \
 	    -no-undefined -bindir "$(bindir)"
 
 toolexeclib_LTLIBRARIES = libgfortran.la
 libgfortran_la_LINK = $(LINK) $(libgfortran_la_LDFLAGS)
-libgfortran_la_LDFLAGS = -version-info `grep -v '^\#' $(srcdir)/libtool-version` $(LTLDFLAGS) -lm $(extra_ldflags_libgfortran) $(version_arg)
+libgfortran_la_LDFLAGS = -version-info `grep -v '^\#' $(srcdir)/libtool-version` $(LTLDFLAGS) $(QUADLIBS) $(quad_link) -lm $(extra_ldflags_libgfortran) $(version_arg)
 libgfortran_la_DEPENDENCIES = $(version_dep)
 
 myexeclib_LTLIBRARIES = libgfortranbegin.la
@@ -48,7 +55,8 @@ libgfortranbegin_la_LINK = $(LINK) $(lib
 ## use -iquote
 AM_CPPFLAGS = -iquote$(srcdir)/io -I$(srcdir)/$(MULTISRCTOP)../gcc \
 	      -I$(srcdir)/$(MULTISRCTOP)../gcc/config \
-	      -I$(MULTIBUILDTOP)../../$(host_subdir)/gcc -D_GNU_SOURCE
+	      -I$(MULTIBUILDTOP)../../$(host_subdir)/gcc -D_GNU_SOURCE \
+	      $(QUADINC)
 
 # Fortran rules for complex multiplication and division
 AM_CFLAGS += -fcx-fortran-rules
Index: libgfortran/mk-kinds-h.sh
===================================================================
--- libgfortran/mk-kinds-h.sh	(revision 163619)
+++ libgfortran/mk-kinds-h.sh	(working copy)
@@ -44,7 +44,14 @@ echo "#define GFC_UINTEGER_LARGEST GFC_U
 echo "#define GFC_DEFAULT_CHAR ${smallest}"
 echo ""
 
-REAL_10_FOUND=
+
+# Get the kind value for long double, so we may disambiguate it
+# from __float128.
+echo "use iso_c_binding; print *, c_long_double ; end" > tmq$$.f90
+long_double_kind=`$compile -S -fdump-parse-tree tmq$$.f90 | grep TRANSFER \
+			| sed 's/ *TRANSFER *//'`
+rm -f tmq$$.*
+
 
 for k in $possible_real_kinds; do
   echo "  real (kind=$k) :: x" > tmp$$.f90
@@ -52,17 +59,18 @@ for k in $possible_real_kinds; do
   echo "  end" >> tmp$$.f90
   if $compile -S tmp$$.f90 > /dev/null 2>&1; then
     case $k in
-      4) ctype="float" ; suffix="f" ;;
-      8) ctype="double" ; suffix="" ;;
-      10) ctype="long double" ; suffix="l" ; REAL_10_FOUND=1 ;;
-      16) ctype="long double"
-	  suffix="l"
-	  # Disable REAL(16) if it is just __float128
-	  # until the library is fixed
-	  if [ -n "$REAL_10_FOUND" ]; then
-	    continue
-	  fi
-	  ;;
+      4) ctype="float" ; cplxtype="complex float" ; suffix="f" ;;
+      8) ctype="double" ; cplxtype="complex double" ; suffix="" ;;
+      10) ctype="long double" ; cplxtype="complex long double" ; suffix="l" ;;
+      16) if [ $long_double_kind -eq 10 ]; then
+	    ctype="__float128"
+	    cplxtype="_Complex float __attribute__((mode(TC)))"
+	    suffix="q"
+	  else
+	    ctype="long double"
+	    cplxtype="complex long double"
+	    suffix="l"
+	  fi ;;
       *) echo "$0: Unknown type" >&2 ; exit 1 ;;
     esac
 
@@ -86,7 +94,7 @@ for k in $possible_real_kinds; do
 
     # Output the information we've gathered
     echo "typedef ${ctype} GFC_REAL_${k};"
-    echo "typedef complex ${ctype} GFC_COMPLEX_${k};"
+    echo "typedef ${cplxtype} GFC_COMPLEX_${k};"
     echo "#define HAVE_GFC_REAL_${k}"
     echo "#define HAVE_GFC_COMPLEX_${k}"
     echo "#define GFC_REAL_${k}_HUGE ${huge}${suffix}"