Intel® Fortran Compiler 17.0 Developer Guide and Reference
General Compiler Directive: Assists the compiler's dependence analysis of iterative DO loops.
!DIR$ IVDEP [: option]
option |
Is LOOP or BACK. This argument is only available on processors using IA-32 architecture. |
The IVDEP directive is an assertion to the compiler's optimizer about the order of memory references inside a DO loop.
IVDEP:LOOP implies no loop-carried dependencies. IVDEP:BACK implies no backward dependencies.
When no option is specified, the following occurs:
The compiler begins dependence analysis by assuming all dependences occur in the same forward direction as their appearance in the normal scalar execution order. This contrasts with normal compiler behavior, which is for the dependence analysis to make no initial assumptions about the direction of a dependence.
!DIR$ IVDEP with no option can also be spelled !DIR$ INIT_DEP_FWD (INITialize DEPendences ForWarD).
The IVDEP directive is applied to a DO loop in which the user knows that dependences are in lexical order. For example, if two memory references in the loop touch the same memory location and one of them modifies the memory location, then the first reference to touch the location has to be the one that appears earlier lexically in the program source code. This assumes that the right-hand side of an assignment statement is "earlier" than the left-hand side.
The IVDEP directive informs the compiler that the program would behave correctly if the statements were executed in certain orders other than the sequential execution order, such as executing the first statement or block to completion for all iterations, then the next statement or block for all iterations, and so forth. The optimizer can use this information, along with whatever else it can prove about the dependences, to choose other execution orders.
In the following example, the IVDEP directive provides more information about the dependences within the loop, which may enable loop transformations to occur:
!DIR$ IVDEP
DO I=1, N
A(INDARR(I)) = A(INDARR(I)) + B(I)
END DO
In this case, the scalar execution order follows:
Retrieve INDARR(I).
Use the result from step 1 to retrieve A(INDARR(I)).
Retrieve B(I).
Add the results from steps 2 and 3.
Store the results from step 4 into the location indicated by A(INDARR(I)) from step 1.
IVDEP directs the compiler to initially assume that when steps 1 and 5 access a common memory location, step 1 always accesses the location first because step 1 occurs earlier in the execution sequence. This approach lets the compiler reorder instructions, as long as it chooses an instruction schedule that maintains the relative order of the array references.