Vectorizing HIRLAM 1: This dependence is determinable.

[Toon Moene <toon at moene dot indiv dot nluug dot nl>]

L.S.,

This code:

      SUBROUTINE S(N, M)
      DIMENSION A(N, M), B(N, M)
      READ*,A,B
      DO J = 1, M
         DO I = 1, N
            A(I, J) = A(I, J) + B(I, J)
         ENDDO
      ENDDO
      PRINT*,A
      END

when compiled thusly:

$ gfortran -g -S -O3 -ftree-vectorize -ftree-vectorizer-verbose=2 -msse2 vect1.f

draws the following "not vectorized" message:

vect1.f:5: note: not vectorized: can't determine dependence between (*a_23)[D.951_86] and (*a_23)[D.951_86]
vect1.f:5: note: vectorized 0 loops in function.

That's pretty silly, of course :-)
Source and destination A(I,J) overlap exactly, so vectorization of this loop
is certainly possible.  The equivalent rank-1 example is vectorized without
problems.

Exempting this one special case from "can't determine dependence" will mean
about 1000 more vectorized loops in HIRLAM.

Kind regards,

-- 
Toon Moene - e-mail: toon@moene.indiv.nluug.nl - phone: +31 346 214290
Saturnushof 14, 3738 XG  Maartensdijk, The Netherlands
A maintainer of GNU Fortran 95: http://gcc.gnu.org/fortran/