R. Triolet, F. Irigoin, P. Feautrier, Direct Parallelization of CALL Statements, Proceedings of ACM SIGPLAN '86 Symposium on Compiler Construction, Palo Alto, CA, July 1986, pp. 176-185  Home/Search   Document Not in Database   Summary   Related Articles   Check

....by a section of code has proceeded in three general directions: linear constraint based forms, reference list forms, and triplet notation based forms. 3. 1 Linear Constraint based Techniques Using linear constraint based techniques to represent array accesses was rst proposed by Triolet, et al. [35]. Their representation was called a region and the overall parallelization technique was called direct parallelization. The array regions accessed in a subroutine were attached to a call statement for the subroutine in the form of a set of linear constraints constructed from the subscript ....

R. Triolet, F. Irigoin, and P. Feautrier. Direct Parallelization of Call Statements. Proceedings of the SIGPLAN Symposium on Compiler Construction, pages 176-185, 1986.

.... improvements on shared memory multiprocessors by manually applying interprocedural analysis and transformation techniques to enhance parallelism or memory hierarchy utilization [5, 29, 49] Techniques that have proven useful for this purpose include scalar and array side effect analysis [31, 32, 39, 51], interprocedural constant propagation [23, 44] array KILL analysis [5, 26, 49] and transformation to expose loop nests to parallelization [29] In addition to automatic parallelization, many problem domains in high performance computing can greatly benefit from exploiting interprocedural ....

....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, ....

R. Triolet, F. Irigoin, and P. Feautrier. Direct parallelization of CALL statements. In Proceedings of the SIGPLAN '86 Symposium on Compiler Construction, Palo Alto, CA, June 1986. 21

....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, ....

R. Triolet, F. Irigion, and P. Feautrier. Direct parallelization of call statements. In Proc. SIGPLAN '86 Symposium on Compiler Construction, pages 176--185, June 1986.

....Representing arbitrary iteration spaces and expressions accurately and analyzing them is not practical in a compiler. Thus, we restrict our analysis to loop bounds and expressions that are affine functions of induction variables and loop constants. In this domain, many representations are possible [6, 14, 20, 21, 24, 25]. We choose to view the iteration spaces as multi dimensional convex regions in an integer space [2, 3, 4] We use systems of inequalities to represent these multi dimensional convex regions and program expressions. The analysis and strength reduction optimizations are then performed by ....

R. Triolet, F. Irigoin, and P. Feautrier. Direct parallelization of CALL statements. In Proceedings of the SIGPLAN '86 Symposium on Compiler Construction, Palo Alto, CA, June 1986.

....Representing arbitrary iteration spaces and expressions accurately and analyzing them is not practical in a compiler. Thus, we restrict our analysis to loop bounds and expressions that are affine functions of induction variables and loop constants. In this domain, many representations are possible [6, 14, 20, 21, 24, 25]. We choose to view the iteration spaces as multi dimensional convex regions in an integer space [2, 3, 4] We use systems of inequalities to represent these multi dimensional convex regions and program expressions. The analysis and strength reduction optimizations are then performed by ....

R. Triolet, F. Irigoin, and P. Feautrier. Direct parallelization of CALL statements. In Proceedings of the SIGPLAN '86 Symposium on Compiler Construction, Palo Alto, CA, June 1986.

....to each loop nest) provide information (to the compiler) about the sections of the arrays accessed in the nest, thereby enabling the compiler to detect the exact set of modules used by the nest. A similar effect can also be obtained using sophisticated compiler based region analysis techniques [20, 42, 45]. Another potential problem is that loop splitting might incur a slight increase in datapath (processor core) and instruction cache energy consumptions due to more complex code. However, previous research [43] shows that main memory energy dominates cache and datapath energies; consequently, this ....

R. Triolet, F. Irigoin, and P. Feautrier. Direct parallelization of CALL statements. In Proc. ACM SIGPLAN'86 Symposium on Compiler Construction, Palo Alto, CA, pp. 175--185, June 1986.

....However, all these previous techniques based on the RSD were not accurate enough to handle the remaining types. Thus, as an alternative to the RSD, others have proposed a different type of descriptor based on convex regions, which we collectively call the convex region descriptor (CRD) [45, 48]. The CRD can summarize the CS, Multiple Subscript (MS) and Triangular Affine (TA) types in addition to the SA type. In this sense, it has more representational power than the RSD. However, it also suffers several critical drawbacks, such as (1) it cannot summarize Non Affine (NA) type accesses, ....

R. Triolet, F. Irigoin, and P. Feautrier. Direct Parallelization of Call Statements. Proceedings of the SIGPLAN Symposium on Compiler Construction, pages 176--185, 1986.

....a procedure based on interprocedural information is only limited in the feasibility of the analysis required. Experimental evidence has demonstrated that interprocedural analysis designed for parallelizing compilers can significantly reduce dependences assumed in the presence of procedure calls [TIF86] LY88b] LY88a] HK91] resulting in significant execution time performance improvements [MS91] For scalar compilation, research on the effectiveness of interprocedural information has produced mixed results, ranging from moderate [Con83] to marginal improvement [RG89b] This dissertation does ....

....iteration spaces. In both cases, testing whether two lists have a non empty intersection requires a traversal of the lists of accesses. Thus, the cost of these techniques can be significant. ffl Triolet summarizes array accesses using a more precise representation of the summary than rsds[TIF86] This technique locates the convex hull surrounding two regions representing accesses. Dependence testing requires an assymptotically exponential linear inequality solver. ffl Balasundaram proposed an approach to summarizing array accesses designed to locate opportunities for task level ....

[Article contains additional citation context not shown here]

R. Triolet, F. Irigoin, and P. Feautrier. Direct parallelization of call statements. In Proceedings of the SIGPLAN 86 Symposium on Compiler Construction. ACM, June 1986.

....is in our future agenda. 4.4 Evaluation of Array Renaming Figure 14 presents the normalized energy consumption due to array renaming. We found that only two benchmarks (btrix and phods) take advantage of this optimization. However, we believe that a more sophisticated (inter procedural [19]) analysis can find more opportunities for this optimization in large codes that use many temporary arrays. Figure 15 summarizes the energy impact of our optimizations in each code. Apart from individual optimizations, it also gives the combined memory energy effect of the optimizations. From ....

R. Triolet, F. Irigoin, and P. Feautrier. Direct parallelization of CALL statements. In Proc. ACM SIGPLAN'86 Symposium on Compiler Construction, Palo Alto, CA, pp. 175--185, June 1986.

....later in this thesis. 2. 4 Related Work This thesis is largely based on combining the work that has been done on compiler optimizations for object oriented languages [1, 28, 27, 28, 36, 37, 55, 64] functional languages [42, 54, 56, 66, 68] and compilers for vector and parallel machines [40, 48, 69], as 16 well as some techniques from classical scalar compilers [14, 30] The unique paradigm compiling for the Internet that Java has imposed on compiler research has allowed us to explore these techniques, modify, extend and apply them in this new environment. We will now address some of the ....

....The performance improvements due to these techniques cannot be expected to be as high as in the classical compilers for procedural languages, but are still significant enough to justify their implementation and the development of the exception hiding. Havlak and Kennedy [48] and Triolet et al. [69] have developed regular section analysis techniques to efficiently describe parts of arrays that are affected by particular loops in the program. While these techniques were mainly targeting high performance compilation for parallel machines, their simplified versions can be used to help isolate ....

Remi Triolet, Francois Irigoin, and Paul Feautrier. Direct parallelization of CALL statements. In Proceedings of the SIGPLAN '86 Symposium on Compiler Construction, pages 176--185, Palo Alto, California, July 1986.

.... the analysis to generate conditions that guard conditionally optimized code, and to generate conditions that use run time information to identify parallel loops [16] There has been a significant amount of research on extracting array sections in scientific programs that manipulate dense matrices [20, 12, 11]. These techniques are all designed to work for programs with loop nests that access matrices using affine access functions. The techniques presented in this paper, on the other hand, are designed to work for pointer references in recursive procedures with general control flow. 10 Conclusion ....

R. Triolet, F. Irigoin, and P. Feautrier. Direct parallelization of CALL statements. In Proceedings of the SIGPLAN '86 Symposium on Compiler Construction, Palo Alto, CA, June 1986.

....Analysis Several researchers have developed systems that automatically characterize the array regions that procedures access. The first systems were designed to analyze Design Driven Compilation 13 scientific programs with loop nests that manipulate dense matrices using affine access functions [17, 12, 11]. These systems use the loop bounds and the array index expressions to derive the array regions that each procedure accesses. They then propagate accessed array regions from callees to callers to derive the regions accessed by the complete execution of each procedure. Researchers have recently ....

R. Triolet, F. Irigoin, and P. Feautrier. Direct parallelization of CALL statements. In Proceedings of the SIGPLAN '86 Symposium on Compiler Construction, Palo Alto, CA, June 1986.

....Interprocedural Array Region Analysis Several researchers have developed systems that automatically characterize the array regions that procedures access. The first systems were designed to analyze scientific programs with loop nests that manipulate dense matrices using affine access functions [17, 12, 11]. These systems use the loop bounds and the array index expressions to derive the array regions that each procedure accesses. They then propagate accessed array regions from callees to callers to derive the regions accessed by the complete execution of each procedure. Researchers have recently ....

R. Triolet, F. Irigoin, and P. Feautrier. Direct parallelization of CALL statements. In Proceedings of the SIGPLAN '86 Symposium on Compiler Construction, Palo Alto, CA, June 1986.

....with the triplet notation. The same notation was used in papers by Tseng [10] and Chatterjee, Gilbert and Long [11] for message generation. Blume and Eigenmann [5] excluded the stride from the triplet notation in their dependence test for simplicity, but at the expense of accuracy. Convex regions [14, 17] express the geometrical shape of array accesses. They can be used with Fourier Motzkin based dependence tests [21, 22] Balasundaram and Kennedy [15] simplified the convex region to detect task parallelism. Such representations are designed to strike a balance between the efficiency of using the ....

R. Triolet, F. Irigoin, P. Feautrier, Direct Parallelization of CALL Statements, Proceedings of ACM SIGPLAN '86 Symposium on Compiler Construction, Palo Alto, CA, July 1986, pp. 176-185

....Y is a distributed array, its local elements must be taken into account as a special reference and be accessed with (p; c; instead of i. The definition of View is thus altered to take into account array regions . These regions are the result of a precise program analysis which is presented in [80, 50, 9, 11, 10, 33, 32, 31]. An array region is a set of array elements described by equalities and inequalities defining a convex polyhedron. This polyhedron may be parameterized by program variables. Each array dimension is described by a variable. The equations due to subscript expression S Y are replaced by the array ....

R'emi Triolet, Paul Feautrier, and Francois Irigoin. Direct parallelization of call statements. In Proceedings of the ACM Symposium on Compiler Construction, 1986.

No context found.

R. Triolet, F. Irigoin, P. Feautrier, Direct Parallelization of CALL Statements, Proceedings of ACM SIGPLAN '86 Symposium on Compiler Construction, Palo Alto, CA, July 1986, pp. 176-185

No context found.

R. Triolet, F. Irigoin, and P.Feautrier. Direct Parallelization of Call Statements. Proceedings of the SIGPLAN Symposium on Compiler Construction, pages 176--185, 1986.

No context found.

R. Triolet, F. Irigoin, and P. Feautrier. Direct parallelization of CALL statements. In Proc. ACM SIGPLAN'86 Symposium on Compiler Construction, Palo Alto, CA, pp. 175--185, June 1986.

No context found.

R. Triolet, F. Irigoin, P.Feautrier, Direct Parallelization of CALL Statements, Proceedings of ACM SIGPLAN '86 Symposium on Compiler Construction,Palo Alto, CA, July 1986, pp. 176-185

No context found.

R. Triolet, F. Irigoin, and P. Feautrier. Direct parallelization of CALL statements. In Proceedings of the SIGPLAN '86 Symposium on Compiler Construction, Palo Alto, CA, June 1986.

No context found.

R. Triolet, F. Irigion, and P. Feautrier. Direct Parallelization of Call Statements. Proc. of the ACM SIGPLAN Symp. on Compiler Construction, 21(7):176--185, July 1986.

No context found.

R. Triolet, F. Irigoin, and P. Feautrier. Direct Parallelization of Call Statements. Proceedings of the SIGPLAN Symposium on Compiler Construction, pages 176--185, 1986.

No context found.

R. Triolet, F. Irigion, and P. Feautrier, "Direct parallelization of call statements," Proceedings of the ACM SIGPLAN Symposium on Compiler Construction, vol. 21, no. 7, pp. 176--185, July 1986.

No context found.

R. Triolet, F. Irigoin, and P. Feautrier. Direct parallelization of call statements. In Proceedings of the SIGPLAN'86 Symposium on Compiler Construction, pages 176-185, Palo Alto, CA, July 1986.

No context found.

R Triolet et al., "Direct parallelization of CALL statements", in SIGPLAN Symposium on Compiler Construction, July 1986.

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