Synthetic Floating-Point Workloader
Original Publication Date: 1986-May-01
Included in the Prior Art Database: 2005-Mar-09
This article describes a standard method by which floating-point performance of different systems can be tested. In the development of high performance floating-point processors, the need has arisen for a standard by which the floating-point performance of the different systems can be compared. Historically, the need to analyze floating-point performance is not new. One previously developed solution to the dilemma was to determine the type of high level operation (sine, cosine, matrix manipulation, integer to floating-point conversion, etc.) which normally is performed and creates a set of test kernels, each kernel representing adifferent type of numerical operation. Next, the ratio of frequency of occurrence between the different types of operations was determined.