gmp-5.0.1 passes all tests on baseline Sol 8

[Torbjorn Granlund]

Dennis Clarke <dclarke at blastwave.org> writes:

  Using GCC 4.3.4 and binutils and the Sun ( vendor ) provided linker I get
  perfect test results for both pentiumpro and i486 processor types both
  with and without mmx options.
  
  However, Studio 11 C Compiler fails to compile at this point :
  
   /opt/studio/SOS11/SUNWspro/bin/cc -c -DHAVE_CONFIG_H -I. -I..
  -D__GMP_WITHIN_GMP -I.. -DOPERATION_divrem_2 -I/opt/csw/include -erroff
  -xstrconst -xildoff -xarch=386 -xnolibmil -Xa -Kpic -xregs=no%frameptr
  -xlibmieee -g -Qy -xdebugformat=dwarf -xs -D_TS_ERRNO tmp-divrem_2.s
  -KPIC -DPIC -o .libs/divrem_2.o
  Assembler:
          "tmp-divrem_2.s", line 115 : Syntax error
          Near line: "    or      %dl, %al"
          "tmp-divrem_2.s", line 210 : Syntax error
          Near line: "    or      %dl, %al"
  cc: assembler failed for tmp-divrem_2.s
  
This looks like a bug in the Sun assembler.

  Also, GCC 4.3.4 ( from Blastwave ) builds everything on Sparc for both
  32-bit Sparc v8 and v9 with perfect results.
  
We're making some progress. :-)

  Fails rather neatly with Sun Studio 12 on an AMD Opteron based ( x86_64 )
  Solaris 10 ( update 8, latest release ) thus :

  m4:addlsh_n.asm:44 bad macro name
  define(cnt      %r8)
  gmake[2]: *** [addlsh_n.lo] Error 1
  gmake[2]: Leaving directory
  `/export/medusa/dclarke/build/libgmp/i386/Studio12_amd64/gmp-5.0.1/mpn'
  gmake[1]: *** [all-recursive] Error 1
  gmake[1]: Leaving directory
  `/export/medusa/dclarke/build/libgmp/i386/Studio12_amd64/gmp-5.0.1'
  gmake: *** [all] Error 2
  $

This is a bug in GMP that is harmless unless m4 gets miffed.  It is
harmless since the defined macro cnt is actually not used.

Thanks for your testing!

-- 
Torbjörn