Loop Fusion in Array Assignments

compiler can optimize the loop sequence by fusion
- for that, all assignments (loops) should be adjacent
- preserving data dependencies, this can fused:
- for this optimization to work automatically, no code should be placed between the array assignments, such that the assignments are adjacent

F90 code sequence:

A(I:N) = B(I:N)+1

C(I:N) = A(1:N)/2

D(1:N) = 1/C(2:N+1)

Allocate T(1:N)

DO I=1,N

T(I)=B(I)+1

ENDDO

DO I=1,N

A(I) = T(I)

ENDDO

DO I=1,N

T(I)= A(I)/2

ENDDO

DO I=1,N

C(I) = T(I)

ENDDO

DO I=1,N

T(I)=1/C(I+1)

ENDDO

DO I=1,N

D(I) = T(I)

ENDDO

Compiler will typically

generate the following

instruction sequence

Fused loops:

DO I=1,N

A(I) = B(I)+1

C(I) = A(I)/2

ENDDO

DO I=1,N

D(I) = 1/C(I+1)

ENDDO

Further peeling to break data dependencies

will merge the two remaining loops

Loop Fusion is instrumental in generating good F90 code