Burke, M., and Cytron, R. Interprocedural dependence analysis and parallelization. In Proceedings of Sigplan '86 Symposium on Compiler Construction (July 1986), pp. 162--175.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....The typical set operations (intersection, union, etc) are de ned on DADs. 3. 2 Reference List based Techniques Li and Yew proposed Atom Images [26] and Burke and Cytron proposed the conversion of all arrays to a single declared dimension and all subscript expressions to a linearized form [11]. The reference list techniques rely on making a list of each individual array reference in the code section. These methods retain all the precision of the original program because they keep all relevant information about each access, but they were not designed to summarize this 18 information. ....

M. Burke and R. Cytron. Interprocedural Dependence Analysis and Parallelization. Proceedings of the SIGPLAN Symposium on Compiler Construction, pages 162-175, July 1986.

....of the call. Dependence analysis, which provides the compiler s fundamental understanding of a program s inherent parallelism, must be extended to incorporate, in addition to scalar MOD and Rrr analysis, 5 more precise side effect information about the subportions of arrays affected by a call [51, 40, 7, 12]. Regular section analysis, which derives rectangular descriptions of array accesses due to a call, will serve this purpose [31] Finally, CONSTANTS and interprocedural symbolic analysis refine dependence information by deriving information about loop bounds and subscript expressions. Of these, ....

M. Burke and R. Cytron. Interprocedural dependence analysis and parallelization. In Proceedings of the SIGPLAN '86 Symposium on Compiler Construction, SIGPLAN Notices 21(7), pages 163-275. ACM, July 1986.

....calls, or restricted ourselves to individual procedures while analyzing real application codes with Paradigm. Clearly, there is a need to develop techniques for interprocedural analysis. A number of researchers have worked on this problem for improving the effectiveness of parallelizing compilers [9, 10, 76, 49, 1]. Those ideas need to be extended to allow determination of constraints and their quality measures across procedure boundaries, and to summarize such information for data accessed in various procedures. ffl Redistribution of data : Currently, Paradigm assigns a fixed distribution to each array, ....

M. Burke and R. Cytron. Interprocedural dependence analysis and parallelization. In Proc. SIGPLAN '86 Symposium on Compiler Construction, pages 162--175, June 1986.

....of [AK87] this field is equivalent to specifying that the dependence is loop independent rather than loop carried with respect to an outer loop. The application of this external field also corresponds to the different cases of dependence checking for different data direction vectors of [BC86] and [Wo189] In summary, the above test can be done only if sequential loop index variables are identical and the source subarray is external with respect to the relevant loop. doall (k=l,100) a[j l,k 1] Sl In order to detect dependences between two linear induction variables ....

Michael Burke and Ron Cytron. Interprocedural dependence analysis and par- allelization. ACM Transactions on Programming Languages and Systems, 8(3):162- 175, June 1986.

....form as A 0 where is the combined vector (i, i, This is exactly the same formulation as for the affine memory disambiguation problem. We can solve for the data flow vectors in the same way that dependence analysis solves for direction vectors using a variation of Burke and Cytron s method[4]. 14 The properties of our LWTs can be exploited in optimizing this procedure. Burke and Cytron s method works by enumerating and verifying possible direction vectors. If the dependence level of a leaf is k, any data flow vector with a direction other than 0 in any of its first k 1 dimensions ....

M. Burke and R. Cytron. Interprocedural dependence analysis and parallelization. In Proceedings of the SIGPLAN 1986.

....the three direction speci ers to appear in the triplet related vector positions. Thus for scalarization dependences, it is no longer the case that a true dependence with a as the rst non = direction is equivalent to an antidependence with the direction reversed, as has been previously noted [4, 6]. By de nition, scalarization dependences are loop independent with regard to surrounding loops. This has several implications. First, any such dependence of a statement on itself will always be an antidependence (ignoring input dependences) whereas such a dependence from one statement to a ....

M. Burke and R. Cytron. Interprocedural dependence analysis and parallelization. In Proceedings of the SIGPLAN '86 Symposium on Compiler Construction, Palo Alto, CA, June 1986.

....aspect of the propagation phase is renaming from formal parameters to actual parameters across a call site. Just as with inlining, this can be difficult if the programmer has reshaped the array at the call so that dimension sizes are different. We could linearize the subscript expressions [BC86] reducing all arrays to a single dimension. However, this yields very complicated subscript expressions that may greatly hamper the ability of the dependence analyzer in disproving dependences, especially since dependence analysis techniques are more successful on simple subscript expressions. ....

....to determine if there exists a dependence between them. ffl Two techniques avoid summarizing accesses to the same array by building a list of all accesses. The first of these, by Burke and Cytron, linearizes all subscript expressions, resulting in a list of accesses to 1 dimensional objects [BC86] The effect of linearization can be a significant loss of information for dependence testing. Li and Yew developed a more precise representation of array accesses, called atom images [LY88a] LY88b] These precisely represent linear subscript 147 expressions in rectangular and triangular ....

M. Burke and R. Cytron. Interprocedural dependence analysis and parallelization. In Proceedings of the SIGPLAN 86 Symposium on Compiler Construction. ACM, June 1986.

....direction speci ers to appear in the triplet related vector positions. Thus for scalarization dependences, it is no longer the case that a true dependence with a as the rst non = direction is equivalent to an antidependence with the direction reversed, as has been previously noted by others [2, 4]. By de nition, scalarization dependences are loop independent with regard to surrounding loops. This has several implications. First, any such dependence of a statement on itself is always an antidependence (ignoring input dependences) Secondly, all inter statement scalarization dependences ow ....

M. Burke and R. Cytron. Interprocedural dependence analysis and parallelization. In Proceedings of the SIGPLAN '86 Symposium on Compiler Construction, Palo Alto, CA, June 1986.

....It must use the results of this analysis to transform the program into a form where it is more amenable to efficient execution on the target system. Interprocedural analysis. Algorithms for performing data flow analysis across whole programs are both readily available and well understood [5, 7, 8, 10]. These techniques are beginning to appear in commercial systems [25] Perhaps less well appreciated are the limitations of this kind of analysis, the role that it can play in code optimization, and the rational consequences of that role. Interprocedural data flow analysis is a technique for ....

M. Burke and R. Cytron. Interprocedural dependence analysis and parallelization. In Proceedings of the SIGPLAN 86 Symposium on Compiler Construction. ACM, June 1986.

....c Gamman Figure 1: The constant propagation lattice. by a particular loop is an important factor in determining both the amount of work it represents and the number of processors that it can profitably employ. Several techniques for discovering interprocedural constants have been proposed [3, 4, 13, 16] (See Section 5, Other Work ) The study presented in this paper is based on the work of Callahan et al. 4] They suggest several jump functions for their framework. Those jump functions vary in complexity and in the class of interprocedural constants that they are expected to propagate. This ....

....Zadeck. This technique has been implemented in several compilation systems and has proved to be efficient in practice. Burke and Cytron propose a similar approach for computing interprocedural constants within a unified framework that integrates dependence analysis with interprocedural analysis [3]. Intraprocedural constant propagation is employed to determine which constants are available for propagation at call sites. Interprocedural constants are propagated from call sites to the called procedure and used as initial information in the intraprocedural constant propagation. The method ....

Michael Burke and Ron Cytron. Interprocedural dependence analysis and parallelization. SIGPLAN Notices, 21(7):162--175, July 1986. Proceedings of the ACM SIGPLAN '86 Symposium on Compiler Construction.

....of their execution time performing computations on arrays in loops; in order to characterize the memory usage patterns for these programs, the compiler must analyze how arrays are accessed within loops. In our framework, we capture information on array access patterns using array section analysis [3, 4, 13, 18]. When applied to a portion of the program (typically a basic block, loop, or loop nest) array section analysis produces a summary representation of the region accessed within each array. Each of the summaries (referred to hereafter as sections ) contains a component that describes the geometric ....

M. Burke and R. Cytron. Interprocedural dependence analysis and parallelization. In Proceedings of the SIGPLAN '86 Symposium on Compiler Construction, Palo Alto, CA, June 1986.

....four variables, each with different coefficients, is too complicated for these techniques. We found the following techniques inadequate to disprove this dependence: Single Variable Per Constraint Test [MHL91, Ban88] Acyclic Test [MHL91] Simple Loop Residue Test [MHL91] Shostak algorithm [Sho81, BC86] GCD test [AK87, Ban88] Our approach. Our approach is to break dependence equation (1) into two equations: ae 10j 1 Gamma 10j 2 = 0 j 1 ; j 2 2 [0; 9] and ae i 1 Gamma i 2 Gamma 5 = 0 i 1 ; i 2 2 [0; 4] which then can be solved independently of each other by existing techniques. Let ....

....not performed by most C compilers and this requirement is unknown to majority of users. So we have to linearize all the references in order to analyze correctly working programs which may be not standard conforming. 2 Background In this section we give formal definition of dependence taken from [BC86, AK87, WB87] Let us consider two FORTRAN 77 loop nests with n 0 0 common outermost loops (Figure 2) DO x 1 = 0, X 1 . DO xn0 = 0, Xn0 DO xn0 1 = 0, Xn0 1 . DO xn1 = 0, Xn1 S 1 A(f 1 (x 1 ; xn1 ) f l (x 1 ; xn1 ) ENDDO . ENDDO DO yn0 1 = 0, ....

[Article contains additional citation context not shown here]

M. Burke and R. Cytron. Interprocedural dependence analysis and parallelization. In Proc. of the SIGPLAN '86 Symposium on Compiler Construction, July 1986.

.... 62 63 5 Array Liveness Analysis and Its Applications It has been shown that many high level interprocedural analyses and optimizations are needed to create code that effectively runs on a multiprocessor[53] Detecting coarse grain parallelism[51] requires interprocedural dependence analysis[24][51] privatization analysis, and reduction recognition for all scalar and array variables. In addition, symbolic analysis to detect linear relationships between scalar variables is useful in improving the precision of the array analyses. Also, various techniques have been developed to minimize ....

M. Burke and R. Cytron. "Interprocedural dependence analysis and parallelization." In Proceedings of the SIGPLAN '86 Symposium on Compiler Construction, Palo Alto, CA, June 1986.

....S in the current iteration is used by statement S 0 in a future iteration. The dependence arises because for at least a pair of di erent iteration vectors, say v 1 and v 2 , both S1 and S2 access the same memory location. Dependence analysis has been studied extensively in the literature [9, 11, 16, 17, 18, 19, 20], and the details are summarized in [10, 11, 12] Lastly, the level of a dependence is de ned as the loop nest level, numbered from outermost to innermost, that carries the dependence [10, 11] Each dependence arc is associated with kind : true, anti or output dependence, direction : lexically ....

M. Burke and R. Cytron. Interprocedural dependence analysis and parallelization. In Proc. of the ACM SIGPLAN '86 Symp. on Compiler Construction, Palo-Alta, CA, July 1986.

....references is called within, then n 1 Theta n 2 additional dependence tests are necessary to analyze the loop. 4. And finally, the translation mechanisms may reduce the efficiency of subsequent intraprocedural analyses, for example by generating nonlinear terms in array subscript expressions [16]. Partial inlining is more suitable to actually optimize programs. User controlled inlining is usually available in compilers. Another solution is to let the compiler automatically inline some functions or some call sites, according to a heuristic. The most advanced compilers seem to be the ....

....programs has been chiefly driven by research on automatic parallelization, parallelizing large loops containing procedure calls being of the utmost importance to achieve good performances. At first, the main purpose was therefore to analyze interprocedural dependences. And alias analysis [16, 29, 72, 62] as well as summary data flow information or interprocedural side effects (sdfi) 28] on scalar variables were among the main analyses. To enhance the results of these analyses, and to enable other useful code transformations 6 Recursivity and formal procedures are not often used in scientific ....

M. Burke and R. Cytron. Interprocedural dependence analysis and parallelization. In ACM SIGPLAN Symposium on Compiler Construction, pages 162--175, July 1986. 26

....EntryP and where the BirthSite and History are computed as in section 7.1. In computing F orwardBind c , aliases that contain access paths not reachable in the called procedure P are not explicitly removed because such aliases may become reachable aliases of a procedure (transitively) called by P [Burke and Cytron 1986]. 20 These aliases can be removed after the interprocedural alias analysis is done. However, the alias relations whose access paths contain only local variables of the called procedure can be ignored upon its termination when such storage is reclaimed unless the calling and the called procedures ....

Burke, M. and Cytron, R. 1986. Interprocedural dependence analysis and parallelization. In SIGPLAN '86 Symposium on Compiler Construction. ACM, 162--175. SIGPLAN Notices, 21(7).

No context found.

Burke, M., and Cytron, R. Interprocedural dependence analysis and parallelization. In Proceedings of Sigplan '86 Symposium on Compiler Construction (July 1986), pp. 162--175.

No context found.

BURKE, M., AND CYTRON, R. Interprocedural dependence analysis and parallelization. In proceedings of the Symposium on Compiler Construction (1986), pp. 162--175.

No context found.

M. Burke and R. Cytron. Interprocedural Dependence Analysis and Parallelization. Proceedings of the SIGPLAN Symposium on Compiler Construction, pages 162--175, July 1986.

No context found.

M. Burke and R. Cytron. Interprocedural Dependence Analysis and Parallelization. In Proc. of the ACM SIGPLAN Symp. on Compiler Construction, pages 162--175, Palo Alto, CA, June 1986.

No context found.

M. Burke and R. Cytron. Interprocedural Dependence Analysis and Parallelization. Proceedings of the SIGPLAN Symposium on Compiler Construction, pages 162--175, July 1986.

No context found.

M. Burke and R. Cytron, "Interprocedural dependence analysis and parallelization," in Proceedings of the ACM SIGPLAN Symposium on Compiler Construction, Palo Alto, CA, June 1986, pp. 162--175.

No context found.

M. Burke and R. Cytron. Interprocedural Dependence Analysis and Parallelization. Interprocedural dependence analysis and parallelization. In Proceedings of the SIGPLAN '86 Symposium on Compiler Construction, July 1986.

No context found.

M. Burke and R. Cytron, `Interprocedural dependence analysis and parallelization', Proc. ACM SIGPLAN '86 Symposium on Compiler Construction, Palo Alto, CA, July 1986, pp. 162--175. Published as ACM SIGPLAN Notices, 21, (7) (1989).

No context found.

Burke M and Cytron R, "Interprocedural dependence analysis and parallelization", in SIGPLAN Symposium on Compiler Construction, July

First 50 documents  Next 50

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC