RE: ppc-eabi with gcc-3.0.4

[David Korn <dkorn at pixelpower dot com>]

>-----Original Message-----
>From: Vermeulen Jan [mailto:Jan.Vermeulen@siemens.atea.be]
>Sent: 26 February 2002 10:47

>Ah, thanks for the pointer... I checked it out and I see:
>
><quoting>
>MULTILIB_OPTIONS = msoft-float \
>                   mlittle/mbig \
>                   mcall-sysv/mcall-aix \
>                   fleading-underscore
>
>MULTILIB_DIRNAMES = nof \
>                    le be \
>                    cs ca \
>                    und
></quoting>
>
>So, i guess that some libraries will not have been made.
>But which ones? The ones containing leading underscores for 
>the functions or those without them?

  No; the way multilibbing works is that not only do all these variant
libraries get made, but unless you take steps to prevent it, each and every
*combination* of them will be made; then, according to the combination of
flags that you use for *your* code at compile time, gcc can choose the
correct variant of multilib, by simply looking in the right combination
of subdirectory names under $prefix/lib/gcc-lib/$target/$version.

  In the MULTILIB_OPTIONS, when you see two options separated by /, that
means they are mutually exclusive:

>                   mlittle/mbig \

means only use -mbig or -mlittle, not both.  Let me simplify that example:

>MULTILIB_OPTIONS = msoft-float \
>                   mlittle/mbig
>
>MULTILIB_DIRNAMES = nof \
>                    le be

  Now, in this case, it would try and build all 8 combinations (think
binary: each option can be on or off, and there are three).  In the table
here, the subdirectories are given considering . (the current dir) to be
the appropriate libs dir.  The switches listed are found by combining the
entries in MULTILIB_OPTIONS, and the subdirectory is chosen by appending
the path elements from MULTILIB_DIRNAMES.  The switches are used when
compiling libgcc (and the other target libs, such as libstdc++ and
libiberty), and the directory hierarchy exists under $prefix/$target/lib
(libstdc++, libiberty get installed here) and also as the same tree under
$prefix/lib/gcc-lib/$target/$version, where libgcc variants get installed:

#    combination       into subdir    using switches
=    ===========       ===========    ==============
0    -none-            .        
1    be                ./be           -mbig
2    le                ./le           -mlittle
3    be + le           ./be/le        -mbig -mlittle
4    nof               ./nof          -msoft-float
5    nof + be          ./nof/be       -msoft-float -mbig
6    nof + le          ./nof/le       -msoft-float -mlittle
7    nof + le + be     ./nof/le/be    -msoft-float -mlittle -mbig

  But mlittle/mbig are marked as mutually exclusive, so it doesn't do that:

#    combination       into subdir    using switches
=    ===========       ===========    ==============
0    -none-            .        
1    be                ./be           -mbig
2    le                ./le           -mlittle
4    nof               ./nof          -msoft-float
5    nof + be          ./nof/be       -msoft-float -mbig
6    nof + le          ./nof/le       -msoft-float -mlittle

  OK, so going back to your original example:

><quoting>
>MULTILIB_OPTIONS = msoft-float \
>                   mlittle/mbig \
>                   mcall-sysv/mcall-aix \
>                   fleading-underscore
>
>MULTILIB_DIRNAMES = nof \
>                    le be \
>                    cs ca \
>                    und
></quoting>

  We should expect to see every combination of libraries, except that
they must all be either big or little endian, but not both; and that they
should all use sysv or aix calling conventions, but not both.

  There's one last bit:  the MULTILIB_EXCEPTIONS, which provides a way to
discard a few more specific combinations.  In 2.95.x it says:

MULTILIB_EXCEPTIONS	= *mbig/*mcall-linux* \
			  *mlittle/*mcall-linux* \
			  *msoft-float/*mcall-linux*

  Now, this means that no library will be built for any of the flag
combinations "-mbig -mcall-linux", "-mlittle -mcall-linux", and
"-msoft-float -mcall-linux".  Why is this?  Well, just to save disk space
and time spent building the compiler.  Linux is always big-endian on ppc,
and since big-endian is the default when *no* endian flag is supplied, there
is no need to make a "-mbig -mcall-linux" variant when you already have
plain "-mcall-linux".  There is no need for "-mlittle -mcall-linux" because
PPC linux does not run in little endian mode, so it would never work, and
likewise, ppc linux *demands* hardware floating point, so there is no need
to make a soft-float variant either.

  If you want more info than this, read the relevant chapters of the
"Using and Porting the gcc compiler" doc:

" Fragments::       Writing the `t-TARGET' and `x-HOST' files."


    cheers,
     DaveK
-- 
Burn your ID card!  http://www.optional-identity.org.uk/
Help support the campaign, copy this into your .sig!


**********************************************************************
This email and any files transmitted with it are confidential and
intended solely for the use of the individual or entity to whom they
are addressed. If you have received this email in error please notify
the system manager.

This footnote also confirms that this email message has been swept by
MIMEsweeper for the presence of computer viruses.

www.mimesweeper.com
**********************************************************************

------
Want more information?  See the CrossGCC FAQ, http://www.objsw.com/CrossGCC/
Want to unsubscribe? Send a note to crossgcc-unsubscribe@sources.redhat.com