BUG: g77 -O -march=pentiumpro, once again

[Toon Moene]

>  I can provide a much smaller test programm for the bug,
>  Jason found:  (it's from slaein.f) gcc -O
>  -march=pentiumpro -c h.f

      SUBROUTINE SLAEIN(N,VR, VI,EPS3)
      IMPLICIT NONE
      INTEGER N
      REAL VR(*),VI(*),EPS3,ROOTN,NORM,NRMSML,REC
      REAL SNRM2,SLAPY2
      EXTERNAL SNRM2,SLAPY2

      NORM = SLAPY2( SNRM2( N, VR, 1 ), SNRM2( N, VI, 1 ) )
      REC = ( EPS3*ROOTN ) / MAX( NORM, NRMSML )
      CALL SSCAL( N, REC, VR, 1 )
      RETURN
      END

If the error doesn't occur with just the -O flag it might be  
related to the handling of conditional moves (isn't it so that  
-march=pentiumpro enables their use ?)

Note that there is a MAX function in the code (4th line from below)  
that might lead g77/gcc into generating a (floating point)  
conditional move.

However, I don't have access to a Pentium Pro (or Pentium II, they  
are architecturally the same) to test this - Craig, how's your Dell  
?

Cheers,
Toon.