S-algol

Abstract: The ALGOL eigenvalue procedure Eig 3 in Section 14 of Parlett [1] was transcribed into Extended Algol [2] for the Burroughs B 5000 and used to obtain the eigenvalues of several matrices. The procedures were altered as follows: 1. Formal changes were made to conform to the Burroughs character set for ALGOL. For example, a. All lower case letters were changed to capitals. b. As a result, the label L in procedure Triangle was changed to LL, to avoid conflict with the integer 1. c. V was changed to OR. 2. Two minor changes were made to conform to syntactic differences between B 5000 ALGOL and true ALGOL: a. Each label was declared in the head of its inniermost block. b. The heading of procedure Evaluate was deleted. The body was labeled Hy, and this block was substituted for Parlett's procedure statement labeled Hy. c. The procedure mod was deleted, and the expression mod (j 1, 3) was replaced by (J 1) l/IOD 3, valid in B 5000 ALGOL. 3. To complete the program ALGOL bodies were written for the procedures comsqrt and scale. 4. In the absence of an overflow procedure, overflow was changed to a local Boolean variable in the body of procedure Laguerre, but set permanently to false. 5. Two irrelevant types of change were made in certain expressions: a. In some places a factor like y T 2 was changed to y X y. b. Some inequalities of form a > b were reversed to read b < a. 6. The texts of some comments were changed. In the absence of overflow none of these changes should affect the program at all. Separate run-time indications assured us that we did not have overflow. The B 5000 procedure EIG 3 was used to obtain the eigenvalues of matrices A and B in Section 14 of Parlett [1]. All computed eigenvalues were correct to within 2 X i0-7. The compiling time from a card-deck input was approximately 43 seconds, including a test program. The running time was approximately 8 seconds for each matrix, excluding output time. James Varah of Stanford University used EIG 3 to obtain the eigenvalues of 100 matrices of order 10 generated in a certain random manner. For each matrix,

Abstract: A straightforward ·set 01 short and simple benchmarks has been used to examine the relative performance 01 a wide range 01 mainframe, mini, and micro systems ~Comparable sets ot BASIC, Pascal FORTRAN and Structured Algol codes are given In a series ot Tables and Figuros Special attention has been paid to the role of additional processors to assist In the execution of the benchmark codes These Include a 6809 used to acceJerate APPLE-Pascal 11 p-code and an AMD 9511 to ac:c:elerate BASIC-E and Structured Algol semi-complied cod", Two AMD 9511 versions of MicroSoft ooao FORTRAN FORLIB are included The overlap In performance of mainframe, mini and micro systems is illustrated The major conclusion Is that special attention should be paid to high speed Interpreters for semi-compiled commercially-available software products as a major thrust towards a transferable set of user ellvironments at different levels of execution speed