Aart J.C. Bik. Compiler Support for Sparse Matrix Computations. PhD thesis, Leiden University, Netherland, May 1996.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....for data, and one array which lists all indices of columns for data) The work on discovering parallelism in irregular data parallel applications is limited by the lack of language support to express irregular data structures. Data restructuring techniques are used for sparse array representations [22,79,105]. These techniques are based on the linear array representation as well. Many irregular applications use non standard data representations such as graphs, trees, or general geometry meshes. These applications are hard or even impossible to parallelize with existing compiler and run time support. ....

....approach based on the inspector executor model. However, the approach applies in the context of multi dimensional arrays in Fotran languages and regular data partitioning (block and cyclic) There is an important body of work on irregular applications that addresses sparse array representations [22, 105, 111]. The existing data parallel frameworks work well for numerical applications that use finite difference, or stencil discretization schemes (Figure 8.2) The code excerpt in Figure 8.2 shows an example of a regular code. Transformations and subsequent optimizations of regular computations for ....

[Article contains additional citation context not shown here]

Aart J.C. Bik. Compiler Support for Sparse Matrix Computations. PhD thesis, Leiden University, Netherland, May 1996.

....Sparsity is related to several other projects that automatically tune the performance of algorithmic kernels for specific machines. In the area of sparse matrices, these systems include the sparse compiler that takes a dense matrix program as input and generates code for a sparse implementation [Bik96] As in Sparsity, the matrix is examined during optimization, although the sparse compiler looks for higher level structure, such as bands or symmetry. This type of analysis is orthogonal to ours, and it is likely that the combination would prove useful. The Bernoulli compiler also takes a ....

Aart J. C. Bik. Compiler Support for Sparse Matrix Computations. PhD thesis, Leiden University, 1996.

....size. PHiPAC [BAD 97] approaches this problem in a different way in that blocked code is generated on a target machine using a parameterized code generator and the block size is chosen by measuring the performance of those generated codes with different parameters. For sparse matrices, Bik [Bik96] and Kotlyar s [KPS97] work generates a sparse code from dense code with the compiler, in order to relieve the programmer s effort in writing complex sparse code and to increase the flexibility of the code for various storage formats of sparse matrices. As a result of effort to provide a generic ....

Aart J. C. Bik. Compiler Support for Sparse Matrix Computations. PhD thesis, Leiden University, 1996.

....is discussed in section 3. Finally, we state our conclusions in section 4, where we also explain why the current prototype has more difficulties with LU factorization of sparse matrices. 2. PRELIMINARIES In this section, we briefly discuss the functionality of the sparse compiler (we refer to [4] for a more elaborate discussion) and give some preliminary information related to the experiments reported in this paper. 2.1 A Brief Introduction to the Sparse Compiler The input of the sparse compiler consists of a FORTRAN program in which twodimensional arrays are used as enveloping data ....

....values and row indices, respectively. Arrays LOW A and HGH A are used to locate the entries in each sparse vector. New entries can be inserted by moving a sparse vector to free space at the end of the parallel arrays which, occasionally, requires a left compression [29, p25 33] 35, p16 21] see [4] for more details) 2) The following annotations indicate that A has a sparse strict lower and upper triangular part with preferred column wise access, and a dense main diagonal that should be accessed separately along the direction (1; 1) Delta 5 DN1 A DN2 A DN3 A DN1 A DN1 A LDU ....

[Article contains additional citation context not shown here]

Aart J.C. Bik. Compiler Support for Sparse Matrix Computations. PhD thesis, Department of Computer Science, Leiden University, 1996. ISBN 90-9009442-3.

....level, as is done traditionally, the sparsity should be dealt with at the compilation level by a special kind of restructuring compiler, referred to as a sparse compiler. Elaboration of these ideas have resulted in the development of an automatic data structure selection and transformation method [7] and [12, 15] The method has been incorporated into MT1 to actually test the feasibility of automatically generating sparse codes. Hence, although MT1 can be used as a conventional restructuring compiler, MT1 can also be used as a sparse compiler to automatically convert a dense program into ....

....serves as a reference guide for programmers that want to use MT1 as a sparse compiler. A detailed presentation of the other features of MT1 is given in the MT1 Reference Manual [22] A thorough presentation of the automatic data structure selection and transformation method can be found in [7]. 2 Motivation Although this document is not intended to be a tutorial on sparse matrix computations, a brief motivation for using a sparse compiler to generate sparse codes is given. 2.1 Sparse Matrix Computations If many elements in a matrix are zero, then this matrix is called a sparse ....

[Article contains additional citation context not shown here]

Aart J.C. Bik. Compiler Support for Sparse Matrix Computations. PhD thesis, Department of Computer Science, Leiden University, 1996. ISBN 90-9009442-3.

No context found.

Aart Johannes Casimir Bik. Compiler support for sparse matrix computations. PhD thesis, Leiden University, 1996.

No context found.

Aart Johannes Casimir Bik. Compiler support for sparse matrix computations. PhD thesis, Leiden University, 1996.

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