Improving Randomization of Random Test Generator

Abstract

Random stimulus generation or Random Instruction Sequence (RIS) is widely recognized as an effective approach for verifying corner cases that are hard to anticipate. While most of design bugs are flushed out by the deterministic approach, RIS tools are also highly effective in hitting obscure cases. The work presented here uses the RIS tool to solve the problem. The tool uses template library that contains test-cases. A test file contains registers and memory values. Random test generator never generates a test which is not a valid test. The work presented here can be subdivided into two parts, one is to generate template (test-case) for specific conditions like instruction optimization, early forward and crypto. These templates are developed to increase the hit-rate of corner cases to meet required targets, which are required for verifying any core. So it helps to improve the randomization of RIS tools. These templates hit different conditions of early forward with various instruction sets like Arch32 and Arch64, and also with different subsets of instructions like branch, logical, arithmetic and load-store. The second part is to analyse the cache and branch behaviour with some random test suits. It is very tedious and time consuming task to find out behaviour for different types of templates. So, automation is required. These automation scripts make analyses easy. The thesis shows statistical analysis of cache and branch behaviour. One set of graphs show cache hit-miss rate for each core while the other set show how cache coherency policy will work. Thesis also contains a set of graphs which show branch prediction/missprediction rate for random templates.