MATMUL bug

This is a simplified version of a bug reported by AntonioGJerez in http://forums.silverfrost.com/viewtopic.php?p=18250#18250 . FTN95 7.10 gives wrong results when MATMUL is invoked with a single-precision complex argument and a double-precision real argument. Here is an example in which MATMUL is used to evaluate the outer product of vectors B and C, where B is represented as a 2 X 1 double-precision matrix and C as a 1 X 2 single-precision complex matrix. The bug does not occur if B is made single-precision.

program MatmulBug
implicit none
double precision B(2,1) 
complex C(1,2) 
!
B=dble(reshape((/1.,2./),(/2,1/))) 
C=cmplx(reshape((/1.,3./),(/1,2/))) 
!
write(*,*)'B = ',B
write(*,*)'C = ',C
write(*,*)'B X C = ',matmul(B,C) 

end program

FTN95 gives the product as (in usual Fortran 2-D array order) [(0,1.875), (0,3.75), (0,2.125), (0,4.25)], instead of the correct result, [(1,0), (2,0), (3,0), (6,0)].

[P.S., 15 June 2015: This bug is also present in FTN95-7.20]

Thanks for the feedback. I have logged this for investigation.

Hello,

This is another simple example of conversion from real*8 to complex that doesn´t work (and no warning or error is displayed)

regards

  program bug3

  COMPLEX A(4,4)
  REAL*8 B(4,4)
  
  PRINT*,'This runs ok'
  B=RESHAPE((/1.d0,0.d0,0.d0,0.d0, 0.d0,1.d0,0.d0,0.d0,0.d0,0.d0,1.d0,0.d0, 0.d0,0.d0,0.d0,1.d0/),(/4,4/))               
  print*,B
  print*,CMPLX(B)
  PRINT*
  PRINT*,'This doesn´t run ok'      
  print*,CMPLX(RESHAPE((/1.d0,0.d0,0.d0,0.d0, 0.d0,1.d0,0.d0,0.d0,0.d0,0.d0,1.d0,0.d0, 0.d0,0.d0,0.d0,1.d0/),(/4,4/)))     
  PRINT*
  PRINT*,'This doesn´t run ok'
  A=CMPLX(RESHAPE((/1.d0,0.d0,0.d0,0.d0, 0.d0,1.d0,0.d0,0.d0,0.d0,0.d0,1.d0,0.d0, 0.d0,0.d0,0.d0,1.d0/),(/4,4/))) 
  print*,A      
  
  end program

Thanks. I have logged this also.

The MATMUL bug has now been fixed for the next release.

Using CMPLX on a RESHAPEd REAL array does not work at the moment. Pending a fix you could write...

COMPLEX A(4,4) 
A=RESHAPE((/1.0,..../),(/4,4/))

That is, remove the call to CMPLX and allow the type conversion to take place across the assignment.

The two bugs illustrated in AntonioGJerez's code have been fixed for the next release of FTN95.