5.0.1 test failures on alphaev56-dec-osf4.0g (Tru64)

[Torbjorn Granlund]

"Daniel Richard G." <skunk at iSKUNK.ORG> writes:

  On Tue, 2010 Sep  7 00:15+0200, Torbjorn Granlund wrote:
  >
  > Build with ".../configure CFLAGS=-g" to disable optimisation and
  > enable debugging.  If the failures remain, gdb will work better.
  
  I used "-O0 -g3 -Wa,-arch,ev56 -mcpu=ev56" for good measure. (The latter
  flags are from a regular GMP build.) I also edited m4-ccas so it
  wouldn't delete the temporary .s files.
  
  (gdb) r
  Starting program: /tmp/gmp-5.0.1.build/tests/mpn/t-hgcd 
  
  Program received signal SIGSEGV, Segmentation fault.
  __gmpn_invert_limb () at tmp-invert_limb.s:73
  73		lda   $1,  $C36
  
  (gdb) r
  Starting program: /tmp/gmp-5.0.1.build/tests/mpn/t-invert 
  
  Program received signal SIGSEGV, Segmentation fault.
  __gmpn_invert_limb () at tmp-invert_limb.s:73
  73		lda   $1,  $C36
  
  (gdb) r
  Starting program: /tmp/gmp-5.0.1.build/tests/mpn/t-div 
  
  Program received signal SIGSEGV, Segmentation fault.
  __gmpn_invert_limb () at tmp-invert_limb.s:73
  73		lda   $1,  $C36
  
  > You might also try with --build=none to disable assembly code.  That
  > will give some indication of cause (2).
  
  All the tests pass (if slowly) with --build=none.
  
Good work!

Let me guess that mpn/alpha/invert_limb.asm is not correctly assembled.
The assembler used by gcc and cc probably doesn't generate the expected
relocs for the two LEA invocations.

This file is not new, and it has been subject to testing on OSF in the
past.

Incompatibilities in how to access memory with relocs are not uncommon.
But since this file is more than 10 years old, such incompatibilities
seem unlikely.

-- 
Torbjörn