J. Janssen and H. Corporaal. Controlled Node Splitting. In Proceedings of the 6th International Conference on Compiler Construction, pages 44 -- 58, Linkoping, Sweden, Apr. 1996. volume 1060 of Springer Lecture Notes in Computer Science.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....1 gives an example. As has been observed in the literature that for various node splitting algorithms, there is an exponential blowup in the size of the program or flow graph. However, there are di#erent ways to choose which nodes to split, and to choose an order of splitting these nodes [14, 16]. This leaves open two questions, which we have been unable to find answered in the published literature. Is it possible that some (perhaps as yet undiscovered) node splitting algorithm avoids exponential blowup . Is it possible that for any flow graph, there is an equivalent, reducible ....

....to finite automata and regular languages the di#erence is that flow graphs have labeled nodes, whereas finite automata have labeled edges. In fact, the language of a flow graph G is regular, but flow graphs produce more There are several equivalent definitions of reducible flow graph; see [13, 1, 14]. We overload the symbol L to mean, in di#erent contexts, the label of a node, the language of a flow graph, or other ways of associating a set of strings to various objects (such as leftist trees or annotated flow graphs) that will come later. If the label of each node is put on each ....

J. Janssen and H. Corporaal. Controlled Node Splitting. In Proceedings of the 6th International Conference on Compiler Construction, pages 44 -- 58, Linkoping, Sweden, Apr. 1996. volume 1060 of Springer Lecture Notes in Computer Science.

....that require single entry loops. However, many algorithms for scheduling (which should ideally follow PRE) rely on reducibility. 104 After PRE(MR) a reducible graph can be obtained with additional code duplication. An effective algorithm for normalizing irreducible programs is given in [JC96] To suppress an unnecessary invocation of the algorithm, we can employ a simple test of whether irreducibility may be created after a region duplication. The test is based upon examining only the CMP entry and exit edges, rather than the entire program. Assuming we start from a reducible graph, ....

....target nodes of region exit edges may become the (multiple) loop entry nodes. Consider the loop in Figure 6.19(a) Two of the three exits of CMP#a # b# fall into the loop. After restructuring, they will become loop entries, as shown in Figure 6. 2(c) Rather than applying a global algorithm like [JC96] a straightforward approach to make the affected loop reducible is to peel off a part of its body. The goal is to extend the replication scope so that the region exits sink onto a single loop node, which will then become the new loop entry. Such a node is the closest common postdominator (within ....

Johan Janssen and Henk Corporaal. Controlled node splitting. In Compiler Construction, 6th International Conference, volume 1060 of Springer Lecture Notes in Computer Science, pages 44--58, Sweden, April 1996.

....that presuppose single entry loops prior to PRE. However, many algorithms for scheduling (which should follow PRE) rely on reducibility. After ComPRE, a reducible graph can be obtained with additional code duplication. An effective algorithm for normalizing irreducible programs is given in [23]. To suppress an unnecessary invocation of the algorithm, we can employ a simple test of whether irreducibility may be created after a region duplication. The test is based upon examining only the CMP entry and exit edges, rather than the entire program. Assuming we start from a reducible graph, ....

....target nodes of region exit edges may become the (multiple) loop entry nodes. Consider the loop in Figure 4(a) Two of the three exits of CMP [a b] fall into the loop. After restructuring, they will become loop entries, as shown in Figure 1(c) Rather than applying a global algorithm like [23], a straightforward approach to make the affected loop reducible is to peel off a part of its body. The goal is to extend the replication scope so that the region exits sink onto a single loop node, which will then become the new loop entry. Such a node is the closest common postdominator (within ....

Johan Janssen and Henk Corporaal. Controlled node splitting. In Compiler Construction, 6th International Conference, volume 1060 of Springer Lecture Notes in Computer Science, pages 44--58, Sweden, April 1996.

No context found.

J. Janssen and H. Corporaal. Controlled Node Splitting. In Proceedings of the 6th International Conference on Compiler Construction, pages 44 -- 58, Linkoping, Sweden, Apr. 1996. volume 1060 of Springer Lecture Notes in Computer Science.

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