current egcs in CVS

current egcs in CVS

[Ulrich Drepper]

Hi,

Does nobody else has problems with the current egcs in the CVS (i.e.,
not the versions on the branch).  On ix86 I get a completely unusable
result since the floating-point handling is horribly broken.  This is
what enquire prints:

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
PROPERTIES OF FLOAT
Base = -2147483648

*** WARNING: Function return or parameter passing faulty? (This is a guess.)

PROPERTIES OF DOUBLE
Base = -2147483648

*** WARNING: Function return or parameter passing faulty? (This is a guess.)

PROPERTIES OF LONG DOUBLE
Base = -2147483648

*** WARNING: Function return or parameter passing faulty? (This is a guess.)

Float expressions are evaluated in a higher precision than double,
   using 1 base digits
Double expressions are evaluated in a higher precision than double,
   using 1 base digits
Long double expressions are evaluated in a higher precision than double,
   using 1 base digits
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Another problem is that `make' went on building the compiler.
Shouldn't enquire report this as a severe problem and stop the
compilation?

-- Uli
---------------.      drepper at gnu.org  ,-.   Rubensstrasse 5
Ulrich Drepper  \    ,-------------------'   \  76149 Karlsruhe/Germany
Cygnus Solutions `--' drepper at cygnus.com   `------------------------

Re: current egcs in CVS

[Jeffrey A Law]

  In message < x73ek7k8bv.fsf@myware.rz.uni-karlsruhe.de >you write:
  > Hi,
  > 
  > Does nobody else has problems with the current egcs in the CVS (i.e.,
  > not the versions on the branch).  On ix86 I get a completely unusable
  > result since the floating-point handling is horribly broken.  This is
  > what enquire prints:
I bootstrapped on my x86 yesterday without any problems at all.
i686-pc-linux-gnulibc1.

I'm in the process of doing a merge from gcc2 -> egcs, once that's done
I'll be making a fresh snapshot (from the development tree) so other
folks can look into this problem.

  > Another problem is that `make' went on building the compiler.
  > Shouldn't enquire report this as a severe problem and stop the
  > compilation?
Nope -- at least in the past enquire generated warnings/errors on
many systems due to bugs in their C-libraries.  It was decreed that
this should not stop the build process.

jeff

Re: current egcs in CVS

[Ulrich Drepper]

Jeffrey A Law <law@hurl.cygnus.com> writes:

> I bootstrapped on my x86 yesterday without any problems at all.
> i686-pc-linux-gnulibc1.

The compiler builds fine.  What does the generated gcc/include/float.h
contain?  Mine is empty.

I'll try to update the gcc source tree of egcs.  Maybe some unwanted
change crept in.

-- Uli
---------------.      drepper at gnu.org  ,-.   Rubensstrasse 5
Ulrich Drepper  \    ,-------------------'   \  76149 Karlsruhe/Germany
Cygnus Solutions `--' drepper at cygnus.com   `------------------------

Re: current egcs in CVS

[Jeffrey A Law]

  In message < x7btyue6m7.fsf@myware.rz.uni-karlsruhe.de >you write:
  > Jeffrey A Law <law@hurl.cygnus.com> writes:
  > 
  > > I bootstrapped on my x86 yesterday without any problems at all.
  > > i686-pc-linux-gnulibc1.
  > 
  > The compiler builds fine.  What does the generated gcc/include/float.h
  > contain?  Mine is empty.
Mine looks normal.

jeff

Re: current egcs in CVS

[Alexandre Oliva]

Ulrich Drepper writes:

> Does nobody else has problems with the current egcs in the CVS (i.e.,
> not the versions on the branch).  On ix86 I get a completely unusable
> result since the floating-point handling is horribly broken.  This is
> what enquire prints:

I used to get this problem on ix86 on older FSF gcc snapshots, if
BOOT_CFLAGS contained -O2 or higher.  This is fixed in the current
snapshot; it is possible that Jeff hasn't merged whatever fixed it
yet.

-- 
Alexandre Oliva
mailto:oliva@dcc.unicamp.br mailto:aoliva@acm.org
http://www.dcc.unicamp.br/~oliva
Universidade Estadual de Campinas, SP, Brasil

Re: current egcs in CVS

[Ulrich Drepper]

Alexandre Oliva <oliva@dcc.unicamp.br> writes:

> I used to get this problem on ix86 on older FSF gcc snapshots, if
> BOOT_CFLAGS contained -O2 or higher.  This is fixed in the current
> snapshot; it is possible that Jeff hasn't merged whatever fixed it
> yet.

We found out that it's no problem with -O2 but with -O3.  I.e., don't run

	make CFLAGS=-O3

in the current version.

-- Uli
---------------.      drepper at gnu.org  ,-.   Rubensstrasse 5
Ulrich Drepper  \    ,-------------------'   \  76149 Karlsruhe/Germany
Cygnus Solutions `--' drepper at cygnus.com   `------------------------

Re: current egcs in CVS

[Jeffrey A Law]

  In message < x7pvn9dbwq.fsf@myware.rz.uni-karlsruhe.de >you write:
  > Alexandre Oliva <oliva@dcc.unicamp.br> writes:
  > 
  > > I used to get this problem on ix86 on older FSF gcc snapshots, if
  > > BOOT_CFLAGS contained -O2 or higher.  This is fixed in the current
  > > snapshot; it is possible that Jeff hasn't merged whatever fixed it
  > > yet.
  > 
  > We found out that it's no problem with -O2 but with -O3.  I.e., don't run
  > 
  > 	make CFLAGS=-O3
  > 
  > in the current version.
I just got a clean -O3 build; so it was probably something that got
fixed when I did the merge from gcc-2.8.

jeff

Re: current egcs in CVS

[Jeffrey A Law]

  In message < oryb1xzv30.fsf@grupiara.dcc.unicamp.br >you write:
  > Ulrich Drepper writes:
  > 
  > > Does nobody else has problems with the current egcs in the CVS (i.e.,
  > > not the versions on the branch).  On ix86 I get a completely unusable
  > > result since the floating-point handling is horribly broken.  This is
  > > what enquire prints:
  > 
  > I used to get this problem on ix86 on older FSF gcc snapshots, if
  > BOOT_CFLAGS contained -O2 or higher.  This is fixed in the current
  > snapshot; it is possible that Jeff hasn't merged whatever fixed it
  > yet.
I just did a merge with the gcc-2.8 tarball, so if it was fixed at
the FSF sources, it should be OK in the mainline tree now...

I'll be making a snapshot of the mainline tree later today.

Jeff