B. Blount and S. Chatterjee. An evaluation of Java for numerical computing. In Proceedings of ISCOPE'98, volume 1505 of Lecture Notes in Computer Science, pages 35--46. Springer Verlag, 1998.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....elements. However, given that (a) Java does not currently support generic classes, that (b) the ocial plans [JSR] to incorporate generic classes do not support primitive types (float, double, int, etc. and (c) that emulating generic classes with inheritance delivers poor performance [BK99, BC99] OoLaLa is implemented by developing a version for each numerical data type. OoLaLa represents two dimensional arrays by mapping them to one dimensional arrays in a column wise form (as in the Array Package [MMG99] and JLAPACK [BC99] In this way, a two dimensional array is stored continuously ....

....classes with inheritance delivers poor performance [BK99, BC99] OoLaLa is implemented by developing a version for each numerical data type. OoLaLa represents two dimensional arrays by mapping them to one dimensional arrays in a column wise form (as in the Array Package [MMG99] and JLAPACK [BC99] In this way, a two dimensional array is stored continuously by columns in memory (as in Fortran) and the number of exception tests (array index out of bounds and null object) is halved. The operation norm1 (jjAjj 1 ) shown in Figure 2, is used to illustrate and understand the 1 UML class ....

[Article contains additional citation context not shown here]

Brian Blount and Siddhartha Chatterjee. An evaluation of Java for numerical computing. Scientic Programming, 7(2):97-110, 1999.

....substantial for MATMUL and BSOM, less so for TOMCATV. For MICRO DC, no fmas were generated by the compilers (not even by the Fortran compiler) 5 Related work The importance of having high performance libraries for numerical computing in Java has been recognized by many authors. In particular, [4, 5, 29] describe projects to develop such libraries entirely in Java. In addition, there are approaches in which access to existing numerical libraries (typically coded in C and Fortran) is provided to Java applications [3, 9, 16] All of these approaches shift the burden of delivering high performance ....

B. Blount and S. Chatterjee. An evaluation of Java for numerical computing. In Proceedings of ISCOPE'98, volume 1505 of Lecture Notes in Computer Science, pages 35--46. Springer Verlag, 1998.

....of the matrix elements. But, given that Java does not currently support generic classes, that the ocial plans 2 to incorporate generic classes do not support primitive types (float, double, int, etc. and that emulating generic classes with inheritance ( 33] App. B. 4) delivers poor performance [12, 9], OoLaLa is implemented by developing a version for each numerical data type. OoLaLa represents two dimensional arrays by mapping them to one dimensional arrays in a column wise form (as in the Array Package [34] and JLAPACK [9] In this way, a two dimensional array is stored continuously by ....

....classes with inheritance ( 33] App. B.4) delivers poor performance [12, 9] OoLaLa is implemented by developing a version for each numerical data type. OoLaLa represents two dimensional arrays by mapping them to one dimensional arrays in a column wise form (as in the Array Package [34] and JLAPACK [9]) In this way, a two dimensional array is stored continuously by columns in memory (as in Fortran) and the number of exception tests (array index out of bounds and null object) is halved. The operation norm1 (jjAjj 1 ) is used to illustrate and understand the differences among implementations at ....

[Article contains additional citation context not shown here]

Brian Blount and Siddhartha Chatterjee. An evaluation of Java for numerical computing. Scientic Programming, 7(2):97-110, 1999.

....omitted: the class of the elements of matrices; constructors, and methods that query the attributes. The syntax for declaring attributes and methods is a relaxed Java like syntax. Library References LAPACK [37, 38, 32, 60] SparseLib and IML [30, 78, 31, 85, 53] Paladin [48, 49] JLAPACK [14, 15, 26] OwlPack [20, 19, 18, 24] MTL and ITL [80, 81, 83, 79, 82, 59, 58] PMLP [11, 12, 51] Di pack [17, 75] ISIS [2, 27] Sparspak or Sparspak90 [45] Oblio and Spindle [29, 28, 57] JAMA [68] Jampack [88, 25] BPKIT [21, 22, 23] Table 2: Object oriented linear algebra libraries. 3.1 Initial ....

....case. Ideally, generic classes would be used to develop only one version of ######, independent of the data type of the matrix elements, but, given that Java does not currently support generic classes and that emulating generic classes with inheritance ( 71] App. B. 4) delivers poor performance [18, 15], ###### is implemented by developing aversion for eachnumerical data type. ###### represents two dimensional arrays by mapping them to one dimensional language arrays in a column wise form (as in ArrayPackage [73] and JLAPACK [15] In this way, atwo dimensional array is stored contiguously by ....

[Article contains additional citation context not shown here]

B. Blount and S. Chatterjee. An evaluation of Java for numerical computing. Scientic Programming, 7(2):97-110, 1999.

....omitted: the class of the elements of matrices; constructors, and methods that query the attributes. The syntax for declaring attributes and methods is a relaxed Java like syntax. Library References LAPACK [37, 38, 32, 60] SparseLib and IML [30, 78, 31, 85, 53] Paladin [48, 49] JLAPACK [14, 15, 26] OwlPack [20, 19, 18, 24] MTL and ITL [80, 81, 83, 79, 82, 59, 58] PMLP [11, 12, 51] Di pack [17, 75] ISIS [2, 27] Sparspak or Sparspak90 [45] Oblio and Spindle [29, 28, 57] JAMA [68] Jampack [88, 25] BPKIT [21, 22, 23] Table 2: Object oriented linear algebra libraries. 3.1 Initial ....

B. Blount and S. Chatterjee. An evaluation of Java for numerical computing. In D. Caromel, R. R. Oldehoeft, and M. Tholburn, editors, Computing in Object-Oriented Parallel Environments, Second International Symposium ISCOPE 98,volume 1505 of Lecture Notes in Computer Science, pages 35-46. Springer-Verlag, 1998.

....by a dense vector # j , of the same length as J j , which has only the nonzero entries for row j of S. This representation of S is shown in Figure 2(b) Matrix N does not have to be represented explicitly. S = # # # # # #N = # # # 1 1 1 1 1 1 1 1 1 1 1 1 1 # # # #1 = J1 = [ 1 3 5 ] #2 = J2 = 1 2 5 ] #3 = J3 = 1 2 3 4 ] #4 = J4 = 2 3 4 ] a) b) Figure 2: Structure of the spending matrix S and normalizing matrix N (a) representation of matrix S (b) Let # i be the product dependent scaling factor for product i as discussed in the preceding ....

...., which has only the nonzero entries for row j of S. This representation of S is shown in Figure 2(b) Matrix N does not have to be represented explicitly. S = # # # # # #N = # # # 1 1 1 1 1 1 1 1 1 1 1 1 1 # # # #1 = J1 = 1 3 5 ] #2 = J2 = 1 2 5 ] #3 = J3 = [ 1 2 3 4 ] #4 = J4 = 2 3 4 ] a) b) Figure 2: Structure of the spending matrix S and normalizing matrix N (a) representation of matrix S (b) Let # i be the product dependent scaling factor for product i as discussed in the preceding section. The score # ij of customer j against product i is ....

[Article contains additional citation context not shown here]

B. Blount and S. Chatterjee. An evaluation of Java for numerical computing. In Proceedings of ISCOPE'98, volume 1505 of Lecture Notes in Computer Science, pages 35--46. Springer Verlag, 1998.

....applies the versioning transformation for a different purpose, to create safe and alias free regions, taking into account the Java semantics regarding exceptions and pointers. One approach to high performance numerical computing in Java is the use of high performance libraries. In particular, [5, 6, 23] describe projects to develop such libraries entirely in Java. In addition, there are approaches in which access to existing numerical libraries (typically coded in C and Fortran) is provided to Java applications [4, 8, 13] All of these approaches shift the burden of delivering high performance ....

B. Blount and S. Chatterjee. An evaluation of Java for numerical computing. In Proceedings of ISCOPE'98, volume 1505 of Lecture Notes in Computer Science, pages 35--46. Springer Verlag, 1998.

....the applet and the remote source machine in which we can measure the current network performance. In addition, as research and implementation continue to facilitate high performance computing in Java, performanceoriented Java applications will benefit from resource measurement and prediction tools [18, 15, 2, 12]. This project, however, is not purely an engineering endeavor; this study lends insight into the overhead associated with using Java TCP sockets. We discuss the differences between the Java the C language measurements and provide a quantitative comparison between TCP socket implementation in the ....

B. Blount and S. Chatterjee. An evaluation of java for numerical computing. In Scientific Programing, volume 7 (2), pages 97--119, 1999. Special issue on high performance Java compilation and runtime issues.

.... 4 Theta Theta Theta Theta Theta Theta Theta Theta Theta Theta Theta Theta Theta 3 7 5 N = 2 6 4 1 1 1 1 1 1 1 1 1 1 1 1 1 3 7 5 oe 1 = Theta Theta Theta ] J1 = 1 3 5 ] oe 2 = Theta Theta Theta ] J2 = 1 2 5 ] oe 3 = Theta Theta Theta Theta ] J3 = [ 1 2 3 4 ] oe 4 = Theta Theta Theta ] J4 = 2 3 4 ] Figure 2: Structures of the spending matrix S and normalizing matrix N , and representation of matrix S. 3 product group i can be computed in the following way. First, two vectors and j of size n are initialized to 0: 1 : n] 0; j[1 : n] ....

.... Theta Theta Theta Theta Theta Theta 3 7 5 N = 2 6 4 1 1 1 1 1 1 1 1 1 1 1 1 1 3 7 5 oe 1 = Theta Theta Theta ] J1 = 1 3 5 ] oe 2 = Theta Theta Theta ] J2 = 1 2 5 ] oe 3 = Theta Theta Theta Theta ] J3 = 1 2 3 4 ] oe 4 = Theta Theta Theta ] J4 = [ 2 3 4 ] Figure 2: Structures of the spending matrix S and normalizing matrix N , and representation of matrix S. 3 product group i can be computed in the following way. First, two vectors and j of size n are initialized to 0: 1 : n] 0; j[1 : n] 0: 4) Then, the compressed vector oe j is ....

[Article contains additional citation context not shown here]

B. Blount and S. Chatterjee. An evaluation of Java for numerical computing. In Proceedings of ISCOPE'98, volume 1505 of Lecture Notes in Computer Science, pages 35--46. Springer Verlag, 1998.

....substantial for MATMUL and BSOM, less so for TOMCATV. For MICRO DC, no fmas were generated by the compilers (not even by the Fortran compiler) 5 Related work The importance of having high performance libraries for numerical computing in Java has been recognized by many authors. In particular, [4, 5, 28] describe projects to develop such libraries entirely in Java. In addition, there are approaches in which access to existing numerical libraries (typically coded in C and 13 Fortran) is provided to Java applications [3, 9, 15] All of these approaches shift the burden of delivering high ....

B. Blount and S. Chatterjee. An evaluation of Java for numerical computing. In Proceedings of ISCOPE'98, volume 1505 of Lecture Notes in Computer Science, pages 35--46. Springer Verlag, 1998.

....is an extra level of indirection to get to any complex number in the array. Second, at least 50 of the storage is used for object descriptors. That is, this organization uses at least twice as much memory as a Fortran array of complex numbers. c[n 1] descriptor re 0:0 im 0:0 . c[1] descriptor re 0:0 im 0:0 c[0] descriptor re 0:0 im 0:0 Figure 2: Structure of a Java array of Complex objects. We quantify the performance impact of supporting complex numbers in Java through a working example. The MICROSTRIP benchmark [15] computes the value of the potential ....

....do not need global analysis and in our integration of complex and true multidimensional array optimizations. 7 Conclusions and future work We have demonstrated in this paper that high performance numerical codes using complex numbers can be developed in Java. Earlier work by our group and others [1, 2, 3, 13] has demonstrated that high performance can be obtained for codes written using the Java primitive floating point types. Our present results, combined with those earlier results, show that the answer to the question of whether Java is a suitable language for developing high performance numerical ....

B. Blount and S. Chatterjee. An evaluation of Java for numerical computing. In Proceedings of ISCOPE'98, volume 1505 of Lecture Notes in Computer Science, pages 35--46. Springer Verlag, 1998.

....is best Fortran performance. Numbers at the top of the bars indicate actual Mflops. The fma instruction was disabled for both Java and Fortran. 7 Related work The importance of having high performance libraries for numerical computing in Java has been recognized by many authors. In particular, [7, 9, 31] describe projects to develop such libraries entirely in Java. In addition, there are approaches in which access to existing numerical libraries (typically coded in C and Fortran) is provided to Java applications [6, 10, 14] The Array package includes a BLAS class, so that efficient linear ....

B. Blount and S. Chatterjee. An evaluation of Java for numerical computing. In Proceedings of ISCOPE'98, volume 1505 of Lecture Notes in Computer Science, pages 35--46. Springer Verlag, 1998.

....parts are visible. 7 Conclusions and future work We have demonstrated in this paper that high performance numerical codes using complex numbers can be developed in Java. We have achieved with Java 60 to 90 of Fortran performance complex arithmetic codes. Earlier work by our group and others [2, 3, 4, 13] has demonstrated that high performance can be obtained for codes written using the Java primitive floating point types. Our present results, combined with those earlier results, show that the answer to the question of whether Java is a suitable language for developing high performance numerical ....

B. Blount and S. Chatterjee. An evaluation of Java for numerical computing. In Proceedings of ISCOPE'98, volume 1505 of Lecture Notes in Computer Science, pages 35--46. Springer Verlag, 1998.

....an extra level of indirection to get to any of the complex numbers in the array. Second, at least 50 of the storage is used for object descriptors. That is, this organization uses at least twice as much memory as a Fortran array of complex numbers. c[n 1] descriptor re 0:0 im 0:0 . c[1] descriptor re 0:0 im 0:0 c[0] descriptor re 0:0 im 0:0 Figure 2: Structure of a Java array of Complex objects. We quantify the performance impact of supporting complex numbers in Java through a working example. The MICROSTRIP benchmark [15] computes the value of the potential ....

....do not need global analysis and in our integration of complex and true multidimensional array optimizations. 7 Conclusions and future work We have demonstrated in this paper that high performance numerical codes using complex numbers can be developed in Java. Earlier work by our group and others [1, 2, 3, 13] has demonstrated that high performance can be obtained for codes written using the Java primitive floating point types. Our present results, combined with those earlier results, show that the answer to the question of whether Java is a suitable language for developing high performance numerical ....

B. Blount and S. Chatterjee. An evaluation of Java for numerical computing. In Proceedings of ISCOPE'98, volume 1505 of Lecture Notes in Computer Science, pages 35--46. Springer Verlag, 1998.

....community has been reluctant to adopt Java as the language of choice for numerical applications. To demonstrate the viability of high performance computing in Java and to encourage its greater adoption in the computational science community, several authors have ported numerical libraries to Java [1, 5, 19], written Fortran to Java translators [8, 10] developed compilation technology for Java [6, 7, 25] and written class libraries to address deficiencies in the Java language for numerical computing [26] Although these studies demonstrate the potential of Java for high performance computing, ....

B. Blount and S. Chatterjee. An evaluation of Java for numerical computing. Scientific Programming, 7:97--110, 1999.

....threads than originally intended. Our translation solves this problem by modifying the finalBarrier method. If a thread enters this method and cannot proceed, it executes work off of the work pile until it can proceed. The collections within our library resemble collections from previous work [8] and consist of three components: a data object, a shape object, and a map object. The data object is a one dimensional array containing all the elements of a given collection in some arbitrary order. The shape object specifies the shape of the collection by defining a domain on which the ....

B. L. Blount and S. Chatterjee. An evaluation of Java for numerical computing. In Proc. ISCOPE'98, Dec. 1998.

No context found.

B. Blount and S. Chatterjee. An evaluation of Java for numerical computing. In Proceedings of ISCOPE'98, volume 1505 of Lecture Notes in Computer Science, pages 35--46. Springer Verlag, 1998.

No context found.

BLOUNT,B.AND CHATTERJEE, S. 1999. An evaluation of Java for numerical computing. Scientific Programming 7(2), 97--110. Special Issue: High Performance Java Compilation and Runtime Issues.

No context found.

B. L. Blount and S. Chatterjee. An evaluation of java for numerical computing. In Computing in Object-Oriented Parallel Environments: Proceedings, Second International Symposium, ISCOPE 98, pages 35--46, 1998.

No context found.

B. Blount and S. Chatterjee. An evaluation of Java for numerical computing. Scientific Programming, 7(2):97--110, 1999.

No context found.

B. Blount and S. Chatterjee. An evaluation of Java for numerical computing. Scientific Programming, 7(2):97--110, 1999.

No context found.

B. Blount and S. Chatterjee. An evaluation of Java for numerical computing. In Computing in Object-Oriented Parallel Environments, Second International Symposium ISCOPE 98, volume 1505 of Lecture Notes in Computer Science, pages 35--46. Springer-Verlag, 1998.

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