unit-at-a-time fixes

unit-at-a-time fixes

[Andreas Jaeger]

[-- Attachment #1: Type: text/plain, Size: 4099 bytes --]


With GCC 3.4 and also the hammer-branch of GCC 3.3, unit-at-a-time
compilation is enabled and rearranges the output order of functions.
This is not desirable in some cases.

Here's the patch that we're currently using at SuSE.  Ok to commit?

Andreas

2003-11-30  Andreas Jaeger  <aj@suse.de>

	* csu/Makefile (CFLAGS-initfini.s): Add -fno-unit-at-a-time.
	* linuxthreads/sysdeps/unix/sysv/linux/x86_64/Makefile
	(CFLAGS-pt-initfini.s): Likewise.
	* locale/Makefile (CFLAGS-loadlocale.c): Likewise.

For linuxthreads:
	* Makefile (CFLAGS-tst-cancel.c): Add -fno-unit-at-a-time.
	(CFLAGS-pt-initfini.s): Likewise.

For nptl:
	* Makefile (CFLAGS-pt-initfini.s): Add -fno-unit-at-a-time.

============================================================
Index: csu/Makefile
--- csu/Makefile	26 Jun 2003 07:27:39 -0000	1.70
+++ csu/Makefile	18 Jul 2003 05:27:14 -0000
@@ -104,7 +104,7 @@ omit-deps += $(crtstuff)
 $(crtstuff:%=$(objpfx)%.o): %.o: %.S $(objpfx)defs.h
 	$(compile.S) -g0 $(ASFLAGS-.os) -o $@
 
-CFLAGS-initfini.s = -g0 -fPIC -fno-inline-functions
+CFLAGS-initfini.s = -g0 -fPIC -fno-inline-functions -fno-unit-at-a-time
 
 vpath initfini.c $(full_config_sysdirs)
 
============================================================
Index: linuxthreads/sysdeps/unix/sysv/linux/x86_64/Makefile
--- linuxthreads/sysdeps/unix/sysv/linux/x86_64/Makefile	11 Apr 2003 23:34:02 -0000	1.5
+++ linuxthreads/sysdeps/unix/sysv/linux/x86_64/Makefile	18 Jul 2003 05:27:14 -0000
@@ -1,3 +1,3 @@
 ifeq ($(subdir),linuxthreads)
-CFLAGS-pt-initfini.s = -g0 -fPIC -fno-inline-functions -fno-asynchronous-unwind-tables
+CFLAGS-pt-initfini.s = -g0 -fPIC -fno-inline-functions -fno-asynchronous-unwind-tables  -fno-unit-at-a-time
 endif
============================================================
Index: linuxthreads/Makefile
--- linuxthreads/Makefile	10 Jun 2003 02:19:43 -0000	1.77
+++ linuxthreads/Makefile	18 Jul 2003 05:27:14 -0000
@@ -66,7 +66,7 @@ LDFLAGS-pthread.so = $(nodelete-$(have-z
 vpath %.c Examples
 
 tst-cancel-ARGS = "$(objpfx)"
-CFLAGS-tst-cancel.c = -fno-inline -fno-inline-functions
+CFLAGS-tst-cancel.c = -fno-inline -fno-inline-functions -fno-unit-at-a-time
 
 include ../Makeconfig
 
@@ -99,7 +99,7 @@ endif
 extra-objs += $(crti-objs) $(crtn-objs)
 omit-deps += crti crtn
 
-CFLAGS-pt-initfini.s = -g0 -fPIC -fno-inline-functions
+CFLAGS-pt-initfini.s = -g0 -fPIC -fno-inline-functions -fno-unit-at-a-time
 endif
 
 librt-tests = ex10 ex11


============================================================
Index: locale/Makefile
--- locale/Makefile	13 Jun 2003 20:42:48 -0000	1.71
+++ locale/Makefile	28 Jul 2003 10:45:07 -0000
@@ -100,6 +100,7 @@ locale-CPPFLAGS := -DLOCALE_PATH='$(loca
 CFLAGS-charmap.c = -Wno-write-strings -Wno-char-subscripts
 CFLAGS-locfile.c = -Wno-write-strings -Wno-char-subscripts
 CFLAGS-charmap-dir.c = -Wno-write-strings
+CFLAGS-loadlocale.c = -fno-unit-at-a-time
 
 # This makes sure -DNOT_IN_libc is passed for all these modules.
 cpp-srcs-left := $(addsuffix .c,$(localedef-modules) $(localedef-aux) \

--- nptl/Makefile	6 Nov 2003 09:47:37 -0000	1.124
+++ nptl/Makefile	18 Nov 2003 20:30:23 -0000
@@ -312,7 +312,7 @@
 extra-objs += $(crti-objs) $(crtn-objs)
 omit-deps += crti crtn
 
-CFLAGS-pt-initfini.s = -g0 -fPIC -fno-inline-functions
+CFLAGS-pt-initfini.s = -g0 -fPIC -fno-inline-functions -fno-unit-at-a-time
 endif
 
 CFLAGS-flockfile.c = -D_IO_MTSAFE_IO
--- nptl/sysdeps/unix/sysv/linux/x86_64/Makefile	9 May 2003 07:29:09 -0000	1.3
+++ nptl/sysdeps/unix/sysv/linux/x86_64/Makefile	18 Nov 2003 20:30:23 -0000
@@ -1,3 +1,3 @@
 ifeq ($(subdir),nptl)
-CFLAGS-pt-initfini.s = -g0 -fPIC -fno-inline-functions -fno-asynchronous-unwind-tables
+CFLAGS-pt-initfini.s = -g0 -fPIC -fno-inline-functions -fno-asynchronous-unwind-tables -fno-unit-at-a-time
 endif

-- 
 Andreas Jaeger, aj@suse.de, http://www.suse.de/~aj
  SuSE Linux AG, Deutschherrnstr. 15-19, 90429 Nürnberg, Germany
   GPG fingerprint = 93A3 365E CE47 B889 DF7F  FED1 389A 563C C272 A126

[-- Attachment #2: Type: application/pgp-signature, Size: 188 bytes --]

Re: unit-at-a-time fixes

[Andreas Jaeger]

[-- Attachment #1: Type: text/plain, Size: 1217 bytes --]

Andreas Jaeger <aj@suse.de> writes:

> With GCC 3.4 and also the hammer-branch of GCC 3.3, unit-at-a-time
> compilation is enabled and rearranges the output order of functions.
> This is not desirable in some cases.
>
> Here's the patch that we're currently using at SuSE.  Ok to commit?
>
> Andreas
>
> 2003-11-30  Andreas Jaeger  <aj@suse.de>
>
> 	* csu/Makefile (CFLAGS-initfini.s): Add -fno-unit-at-a-time.
> 	* linuxthreads/sysdeps/unix/sysv/linux/x86_64/Makefile
> 	(CFLAGS-pt-initfini.s): Likewise.
> 	* locale/Makefile (CFLAGS-loadlocale.c): Likewise.

There're two entries missing:

>
> For linuxthreads:
> 	* Makefile (CFLAGS-tst-cancel.c): Add -fno-unit-at-a-time.
> 	(CFLAGS-pt-initfini.s): Likewise.
        * sysdeps/unix/sysv/linux/x86_64/Makefile (CFLAGS-pt-initfini.s): 
        Likewise.

> For nptl:
> 	* Makefile (CFLAGS-pt-initfini.s): Add -fno-unit-at-a-time.
        * sysdeps/unix/sysv/linux/x86_64/Makefile (CFLAGS-pt-initfini.s): 
        Likewise.



Andreas
-- 
 Andreas Jaeger, aj@suse.de, http://www.suse.de/~aj
  SuSE Linux AG, Deutschherrnstr. 15-19, 90429 Nürnberg, Germany
   GPG fingerprint = 93A3 365E CE47 B889 DF7F  FED1 389A 563C C272 A126

[-- Attachment #2: Type: application/pgp-signature, Size: 188 bytes --]

Re: unit-at-a-time fixes

[Jakub Jelinek]

On Sun, Nov 30, 2003 at 08:38:19PM +0100, Andreas Jaeger wrote:
> 
> With GCC 3.4 and also the hammer-branch of GCC 3.3, unit-at-a-time
> compilation is enabled and rearranges the output order of functions.
> This is not desirable in some cases.
> 
> Here's the patch that we're currently using at SuSE.  Ok to commit?

As long as GCC 3.2.x and 3.3.x are supported to build glibc, this patch is
wrong, since these compilers have no -fno-unit-at-a-time switch.
You should add a configure check for that switch and add
$(fno-unit-at-a-time) to the makefiles instead.

	Jakub

Re: unit-at-a-time fixes

[Andreas Jaeger]

[-- Attachment #1: Type: text/plain, Size: 897 bytes --]

Jakub Jelinek <jakub@redhat.com> writes:

> On Sun, Nov 30, 2003 at 08:38:19PM +0100, Andreas Jaeger wrote:
>> 
>> With GCC 3.4 and also the hammer-branch of GCC 3.3, unit-at-a-time
>> compilation is enabled and rearranges the output order of functions.
>> This is not desirable in some cases.
>> 
>> Here's the patch that we're currently using at SuSE.  Ok to commit?
>
> As long as GCC 3.2.x and 3.3.x are supported to build glibc, this patch is
> wrong, since these compilers have no -fno-unit-at-a-time switch.

Argh.  You're right.

> You should add a configure check for that switch and add
> $(fno-unit-at-a-time) to the makefiles instead.

OK, will do tomorrow,

Andreas
-- 
 Andreas Jaeger, aj@suse.de, http://www.suse.de/~aj
  SuSE Linux AG, Deutschherrnstr. 15-19, 90429 Nürnberg, Germany
   GPG fingerprint = 93A3 365E CE47 B889 DF7F  FED1 389A 563C C272 A126

[-- Attachment #2: Type: application/pgp-signature, Size: 188 bytes --]

Re: unit-at-a-time fixes

[Andreas Schwab]

Andreas Jaeger <aj@suse.de> writes:

> @@ -1371,6 +1372,36 @@ EOF
>  
>    AC_SUBST(libc_cv_fpie)
>  fi
> +
> +if test $elf != yes; then
> +  AC_CACHE_CHECK(for .init and .fini sections, libc_cv_have_initfini,
> +		 [AC_TRY_COMPILE(, [asm (".section .init");
> +				    asm (".section .fini");
> +				    asm ("${libc_cv_dot_text}");],
> +				 libc_cv_have_initfini=yes,
> +				 libc_cv_have_initfini=no)])
> +  AC_SUBST(libc_cv_have_initfini)dnl
> +  if test $libc_cv_have_initfini = yes; then
> +    AC_DEFINE(HAVE_INITFINI)
> +  fi
> +fi

This looks bogus.

Andreas.

-- 
Andreas Schwab, SuSE Labs, schwab@suse.de
SuSE Linux AG, Deutschherrnstr. 15-19, D-90429 Nürnberg
Key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: unit-at-a-time fixes

[Andreas Jaeger]

[-- Attachment #1: Type: text/plain, Size: 940 bytes --]

Andreas Schwab <schwab@suse.de> writes:

> Andreas Jaeger <aj@suse.de> writes:
>
>> @@ -1371,6 +1372,36 @@ EOF
>>  
>>    AC_SUBST(libc_cv_fpie)
>>  fi
>> +
>> +if test $elf != yes; then
>> +  AC_CACHE_CHECK(for .init and .fini sections, libc_cv_have_initfini,
>> +		 [AC_TRY_COMPILE(, [asm (".section .init");
>> +				    asm (".section .fini");
>> +				    asm ("${libc_cv_dot_text}");],
>> +				 libc_cv_have_initfini=yes,
>> +				 libc_cv_have_initfini=no)])
>> +  AC_SUBST(libc_cv_have_initfini)dnl
>> +  if test $libc_cv_have_initfini = yes; then
>> +    AC_DEFINE(HAVE_INITFINI)
>> +  fi
>> +fi
>
> This looks bogus.

You're right - this part should be removed, I seemed to have copied
too much :-(

Andreas
-- 
 Andreas Jaeger, aj@suse.de, http://www.suse.de/~aj
  SuSE Linux AG, Deutschherrnstr. 15-19, 90429 Nürnberg, Germany
   GPG fingerprint = 93A3 365E CE47 B889 DF7F  FED1 389A 563C C272 A126

[-- Attachment #2: Type: application/pgp-signature, Size: 188 bytes --]

Re: unit-at-a-time fixes

[Ulrich Drepper]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Andreas Jaeger wrote:
> Here's an updated patch following Jakub's suggestion.

Remove the bogus configure.in part and you can apply it.

- -- 
➧ Ulrich Drepper ➧ Red Hat, Inc. ➧ 444 Castro St ➧ Mountain View, CA ❖
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.3 (GNU/Linux)

iD8DBQE/zDPe2ijCOnn/RHQRAsxKAJwJ93njDEmHEqUa8kWej57o1ZLo8wCghJGU
0rDFMoI884nyCchnB1+xthI=
=1Ua3
-----END PGP SIGNATURE-----