Re: PATCH for java.nio FileChannel and MappedByteBuffer

[Andrew Pinski <pinskia at physics dot uc dot edu>]

On Mar 18, 2004, at 16:56, Mike Stump wrote:
> I am seeing a bootstrap error on darwin:
>
>  /Volumes/mrs3/net/gcc-darwin/gcc/xgcc -shared-libgcc  
> -B/Volumes/mrs3/net/gcc-darwin/gcc/ -nostdinc++  
> -L/Volumes/mrs3/net/gcc-darwin/powerpc-apple-darwin7.2.0/libstdc++-v3/ 
> src  
> -L/Volumes/mrs3/net/gcc-darwin/powerpc-apple-darwin7.2.0/libstdc++-v3/ 
> src/.libs  
> -B/Volumes/mrs3/Packages/gcc-20040315/powerpc-apple-darwin7.2.0/bin/  
> -B/Volumes/mrs3/Packages/gcc-20040315/powerpc-apple-darwin7.2.0/lib/  
> -isystem  
> /Volumes/mrs3/Packages/gcc-20040315/powerpc-apple-darwin7.2.0/include  
> -isystem  
> /Volumes/mrs3/Packages/gcc-20040315/powerpc-apple-darwin7.2.0/sys- 
> include -dynamiclib -flat_namespace -undefined suppress -o  
> .libs/libgcj.6.0.0.dylib .libs/libgcj.6.0.0.dylib-master.o    
> -L/Volumes/mrs3/net/gcc-darwin/powerpc-apple-darwin7.2.0/libstdc++-v3/ 
> src  
> -L/Volumes/mrs3/net/gcc-darwin/powerpc-apple-darwin7.2.0/libstdc++-v3/ 
> src/.libs  
> -L/Volumes/mrs3/net/gcc-darwin/powerpc-apple-darwin7.2.0/libjava  
> ../boehm-gc/.libs/libgcjgc_convenience.a  
> ../libffi/.libs/libffi_convenience.a  
> ../zlib/.libs/libzgcj_convenience.a -lpthread  
> ./libltdl/.libs/libltdlc.a -ldl     -lc  -install_name   
> /Volumes/mrs3/Packages/gcc-20040315/lib/libgcj.6.dylib  
> -compatibility_version 7 -current_version 7.0
> ld: warning multiple definitions of symbol _sqrt
> .libs/libgcj.6.0.0.dylib-master.o definition of _sqrt in section  
> (__TEXT,__text)
> /usr/lib/libpthread.dylib(sqrt970.o) definition of _sqrt
> ld: .libs/libgcj.6.0.0.dylib-master.o has external relocation entries  
> in non-writable section (__TEXT,__text) for symbols:
> _libiconv
> _msync
> _munmap
> /usr/bin/libtool: internal link edit command failed
> make[2]: *** [libgcj.la] Error 1
> make[2]: Leaving directory  
> `/Volumes/mrs3/net/gcc-darwin/powerpc-apple-darwin7.2.0/libjava'
> make[1]: *** [all-recursive] Error 1
> make[1]: Leaving directory  
> `/Volumes/mrs3/net/gcc-darwin/powerpc-apple-darwin7.2.0/libjava'
> make: *** [all-target-libjava] Error 2
>
> I think Geoff's tester is offline, as otherwise I think it would have  
> complained about it.  Could the person that broke it fix it, thanks.   
> I don't know exactly who that is.


I know who caused it by really it is a latent bug in Darwin's back end.

See PR 14580 (if you can):

Here is the latest from Roger (who caused the failure), it is the bug  
already:
Well, it looks like combine is doing "the right thing" as far as the  
RTL-level
transformation is concerned.  It's turning

(set (reg119) (...)) (REG_EQUAL symbol_ref (!t__unmap)))
(set (reg3) (call (mem (reg119))))

into

(set (reg3) (call (mem (symbol_ref (!t__unmap)))))

i.e. it's optimized an indirect jump into a direct jump.  I suspect that
the problem with the rs6000.md backend may be that the constraints for  
the
*call_value_nonlocal_sysv pattern shouldn't be accepting this  
replacement,
recognizing "mem (symbol_ref (!t__unmap))" as a invalid operand on  
darwin.

I think rs6000's use of REG_EQUAL notes is reasonable in this case, as  
they
can be used in other optimizations.  So I believe that the correct fix  
is to
inform the backend that library functions must be called indirectly via  
their
thunks.  I'm also surprised that this hasn't bitten folks in CSE and/or  
GCSE
before now.

Of course, I'm assuming that the RTL is accurately describing the  
intended
semantics here.


Thanks,
Andrew Pinski