Larus, J. R. 1991. Compiling lisp programs for parallel execution. Lisp and Symbolic Computation 4, 1, 29--99.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....computation. As a result, the state of a future will be either Stolen or Full when a touch is attempted. 1 9 3.2.3 Discussion. The design of futures in Olden was influenced heavily by our intent to use parallelizing compiler technology [Hendren 1990; Hendren and Nicolau 1990; Hendren et al. 1992; Larus 1991] to insert them automatically. Both the restrictions mentioned in the previous paragraph follow from this intention. These two restrictions allow us to allocate future cells on the stack rather than the heap, which helps make futures less expensive. Another important fact to note about the use of ....

Larus, J. R. 1991. Compiling lisp programs for parallel execution. Lisp and Symbolic Computation 4, 1, 29--99.

....In [Harrison III 1991] he describes how the notion of a loop iteration space can be generalized by counting the number of entries into a while loop or a recursively called procedure. Larus developed a number of constructs for annotating objects, including distinct loc and unique obj [Larus 1989; Larus 1991]. He also developed alias graphs, which capture the structure of Lisp objects; like most others, he employed k limiting to limit the size of these objects at compiletime [Larus and Hilfinger 1988b; Larus and Hilfinger 1988a] Early work in this area was done by Jones and Muchnick [1979; 1982] ....

Larus, J. R. 1991. Compiling Lisp programs for parallel execution. LISP and Symbolic Computation, 4, 1 (Jan.), 29--99.

....The pointer analysis problem can be divided into 2 distinct subproblems: 1) disambiguating pointers that point to objects on the stack, and (2) disambiguating pointers that point to objects on the heap. There has been a considerable amount of work in both of these areas [JM81, JM82, LH88, Lar91, Gua88, HPR89, Har89, CWZ90, HN90, LR92, CBC93, Deu92, Deu94, PCK93, EGH94, WL95, Ruf95] although more attention has been paid to actually implementing methods that work well for stack allocated objects [LR92, CBC93, EGH94, Ruf95, WL95] A complete discussion and comparison of these methods can ....

J. R. Larus. Compiling Lisp programs for parallel execution. Lisp and Symbolic Computation, 4:29-99, 1991.

....The pointer analysis problem can be divided into 2 distinct subproblems: 1) disambiguating pointers that point to objects on the stack, and (2) disambiguating pointers that point to objects on the heap. There has been a considerable amount of work in both of these areas [JM81, JM82, LH88, Lar91, Gua88, HPR89, Har89, CWZ90, HN90, LR92, CBC93, Deu92, Deu94, PCK93, EGH94, WL95, Ruf95] although more attention has been paid to actually implementing methods that work well for stack allocated objects [LR92, CBC93, EGH94, Ruf95, WL95] A complete discussion and comparison of these methods can ....

J. R. Larus. Compiling Lisp programs for parallel execution. Lisp and Symbolic Computation, 4:29--99, 1991.

....input data is already distributed) work on all recursion levels can be initiated simultaneously. Now, let us look outside the logic programmingworld. A parallelization technique for recursion in Lisp, which gives essentially the same degree of parallelism as with AND parallelism, is described by Larus (1991). Parcel (Harrison, 1989) is a compiler, for the Lisp dialect Scheme, that parallelizes recursion for execution on shared memory multiprocessors. This work addresses the same problem as we do: compiling recursion to parallel iteration. Let us point out some notable differences in the solutions. ....

Larus, J. R. (1991) Compiling Lisp programs for parallel execution. Lisp and Symbolic Computation 4, No. 1, 29--99.

....such as C and FORTRAN90. The pointer analysis problem can be divided into 2 distinct subproblems: 1) disambiguating pointers that point to objects on the stack, and (2) disambiguating pointers that point to objects on the heap. There has been a considerable amount of work in both of these areas [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17], although more attention has been paid to actually implementing methods that work well for stack allocated objects [10, 11, 15, 16, 17] A complete discussion and comparison of these methods can be found in [18] Stack directed pointers exhibit the important property that their targets always ....

J. R. Larus, "Compiling Lisp programs for parallel execution," Lisp and Symbolic Computation, vol. 4, pp. 29--99, 1991.

....analyses respectively as stack analysis and heap analysis. A considerable amount of work has been done in both of these areas. Initially, the focus was on heap analysis alone, for languages like Lisp and Scheme or for toy imperative languages that did not include all of the complexities of C [3, 4, 5, 12, 17, 18, 19, 20, 22, 23, 27, 30]. A recent trend has been to actually implement pointer analyses in real C and FORTRAN90 compilers, and to examine if practical and useful solutions can be obtained. The most recently proposed (and implemented) approaches [1, 6, 26, 29, 32, 33] mostly focus on the stack problem and only give ....

....structures built by the program, and provide conservative estimates otherwise. Shape analysis is designed for programs that primarily use recursive data structures, or a combination of arrays and recursive data structures. Shape information can be gainfully exploited to parallelize such programs [12, 17, 22, 24], or to apply optimizing transformations like loop unrolling [15] and software pipelining [16] on them. Much of the previous work on heap analysis also primarily focused on some variation of the problem of shape estimation [3, 5, 17, 19, 21, 23, 28, 30] In general, all of these approaches use a ....

J. R. Larus. Compiling Lisp programs for parallel execution. Lisp and Symbolic Computation, 4:29--99, 1991.

.... for static analyses since an individual procedure call site can invoke many procedures and the task of determining the set of procedures invoked from a call site is difficult [16, 18] Static parallelization systems, therefore, do not precisely analyze programs involving first class procedures [6, 10]. The ML procedure map 2 (Figure 1a) illustrates the problem and serves as an example for dynamic parallelization with tags. 2 map is a canonical example of potential parallel evaluation obscured by unknown side effects of higher order procedure parameters. Other common examples are ....

....this problem is ideally suited to solution by dynamic parallelization. Existing static analyses provide crude, yet expensive, approximations to dynamic structures. These compiletime parallelization techniques are often forced to assume structure sharing due to imprecise alias information [7, 6, 10]. Heap resolution is based on the observation that heap reference counts identify sharing in a heap structure. The destructive quicksort qs of Figure 3 serves as the example for heap resolution. qs sorts the elements of list l according to a comparison predicate cmp. This version of quicksort ....

[Article contains additional citation context not shown here]

J. R. Larus. Compiling Lisp programs for parallel execution. Lisp and Symbolic Computation, 4:29--99, 1991.

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