The relationship of code coverage metrics on high-level and RTL code

Abstract

Code coverage is a standard tool for obtaining an indication of the quality of testing done for a hardware design. This is generally applied to the RTL code for a design. As hardware design is moving to a higher level, it is desirable to apply such tools to the original design source code. The high-level code is synthesized into RTL, which is the traditional design representation to which code coverage is applied. It would be desirable to know how code coverage at the higher level is correlated to code coverage at the RTL. However, line coverage, which is commonly used for high-level code, is known to have little correlation to coverage metrics obtained from the RTL produced from the high-level code. In this exercise, we obtain coverage metrics from a high-level model, produce RTL from it using a high-level synthesis tool, and compare those results with RTL coverage metrics. The correlation obtained for this example is quite good.