S. Chatterjee, J.R. Gilbert, R. Schreiber, and T.J. Sheffler. Modeling dataparallel programs with the alignment-distribution graph. Journal of Programming Languages, Special issue on compiling and run-time issues for distributed address space machines, (2), 1994.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....in such cases either by manual analysis of the design or using experiments on a simulator. Compiler Optimizations Most compiler optimizations are in essence choices between several ways of performing a set of operations. the optimizer tries to choose the fastest way. A great deal of research [27, 67, 75] has been done for example in the area of optimizing data distributions, alignment, and redistributions in data parallel languages such as High Performance Fortran [68, 59] While some of that research uses fairly crude cost models, some of the research on performance modeling for high performance ....

S. Chatterjee, J. R. Gilbert, R. Schreiber, and T. J. Sheffler. Modeling data-parallel programs with the alignment distribution graph. Journal of Programming Languages, 2(3):227--258, 1994.

....patterns within computational intensive phases and parallelism exploitation out of them. There has been a significant amount of work concerning static mappings, where the mapping of each array remains fixed along the execution of the whole program ( LC90] KLS90] LC91] Gup92] Who92] CGSS94b] AGG 94] Our work focuses on dynamic mappings in which the mapping of an array may change over its lifetime. Data remapping is one of the topics in this area subject of current research ( CP93] BKK94] CGSS94a] PB95] The main objective of this work has been to devise an algorithm ....

....which remapping options are worth considering at each transition. From this information, a tree showing all the different alternatives of remapping is built. The aim is to determine the path in the tree with the lowest cost. The full remapping tree can easily grow to intractable proportions. CGSS94b] represent the problem as an alignment distribution graph and use a divide and conquer approach to the dynamic mapping problem [CGSS94a] It initially assigns a static mapping to all the nodes and then recursively divides it into regions which are assigned different mappings. Two regions are ....

S. Chatterjee, J.R. Gilbert, R. Schreiber, and T.J. Sheffler. Modeling dataparallel programs with the alignment-distribution graph. Journal of Programming Languages, Special issue on compiling and run-time issues for distributed address space machines, (2), 1994.

....groups of communication is necessary: the first statement copies B to the owner of column 5 of A, and the loop copies B on each column of A. If the second communication is moved above the first assignment, the first communication may be removed. A data flow approach is well suited to this problem [1, 42, 10, 20]. Through advanced program analyses, the source of a value in a program is identified (last write tree, array data flow graph) and the communication is issued accordingly. 3.7 Affine framework Let us introduce the general mapping example in Figure 11. Array A is mapped onto a processor ....

Siddharta Chatterjee, John R. Gilbert, Robert Schreiber, and Thomas J. Sheffler. Modeling data parallel programs with the alignment-distribution graph. Journal of Programming Languages, (2):227--258, 1994.

No context found.

S. Chatterjee, J. R. Gilbert, R. Schreiber, and T. J. Sheffler. Modeling data-parallel programs with the alignmentdistribution graph. Journal of Programming Languages, 1994. Special issue on compiling and run-time issues for distributed address space machines. To appear.

....data locality is important when compiling array parallel languages for distributed memory parallel computers. The languages mentioned above require the user to provide data placement directives in the source code. There has also been considerable interest in automating the task of data placement [2, 5, 6, 16, 17, 18, 22]. This compiler optimization is important for insuring the portability of new scientific codes and for supporting old codes developed without a distributed memory model in mind. Completion time has two components: computation and communication. Communication can be separated into intrinsic and ....

....Our algorithm for axis and stride alignment amplifies their claims of the importance of data layout optimization, and improves upon their methods in several ways. First, we use a more comprehensive cost model, inherited from our alignment distribution graph representation of data parallel programs [6]. We also defer offset concerns to a later phase of alignment, because the shift communication needed to change offset is typically much less expensive than the general communication needed to change axis or stride. 7 Second, we develop a heuristic optimization framework that is more flexible ....

[Article contains additional citation context not shown here]

S. Chatterjee, J. R. Gilbert, R. Schreiber, and T. J. Sheffler. Modeling data-parallel programs with the alignment-distribution graph. Journal of Programming Languages, 2:227--258, Sept. 1994. 30

....important when compiling array parallel languages for distributed memory parallel computers. Languages such as High Performance Fortran [19] require the user to provide data placement directives in the source code. There has also been considerable interest in automating the task of data placement [3, 8, 11, 22, 23, 24, 30]. This compiler optimization is important for insuring the portability of new scientific codes and for supporting old codes developed without a distributed memory model in mind. Data placement optimization is a two step process. First, the alignment phase examines relationships between array ....

....axis and stride alignment amplifies their claims of the importance of data layout optimization, and improves upon their methods in several ways. First, we use a more comprehensive cost model. This is inherited directly from our alignment distribution graph representation of data parallel programs [11]. We also defer offset concerns to a later phase of alignment, because the shift communication needed to change offset is typically much less expensive than the general communication needed to change axis or stride. Second, we develop a heuristic optimization framework that is more flexible than ....

[Article contains additional citation context not shown here]

S. Chatterjee, J. R. Gilbert, R. Schreiber, and T. J. Sheffler. Modeling data-parallel programs with the alignment-distribution graph. Journal of Programming Languages, 2:227--258, Sept. 1994. Special issue on compiling and run-time issues for distributed address space machines.

....edge, giving the alignment of an array at its definition and use. Distribution must also be given for each node, specifying the distribution that is applied to each of the arrays involved in the node computation. The graph, along with these labels, is called an Alignment Distribution Graph (ADG) [3]. We allow the same mappings as HPF: an array is aligned to a template, which is distributed over the available processors. A template is an abstract array used as a target in alignment directives. An alignment is specified by four separate components. These are axis, stride, offset, and ....

....developed earlier. The process of locating subsets and verifying that they are O.S. is heuristic; such a study is therefore mandated, and we view the data below as preliminary, pending better tools and a larger base of experimental programs. Using program analysis tools we have developed earlier [3], we constructed the distribution graphs for seven test programs and applied various combinations of the contraction operations. The contraction operations are sensitive to the adjacency structure of the graph as well as values of the cost entries. For this reason, it is important to understand ....

S. Chatterjee, J. R. Gilbert, R. Schreiber, and T. J. Sheffler. Modeling data-parallel programs with the alignment-distribution graph. Journal of Programming Languages, 2:227--258, 1994. Special issue on compiling and run-time issues for distributed address space machines.

No context found.

S. Chatterjee, J.R. Gilbert, R. Schreiber, and T.J. Sheffler. Modeling dataparallel programs with the alignment-distribution graph. Journal of Programming Languages, Special issue on compiling and run-time issues for distributed address space machines, (2), 1994.

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