Graphics Automated Triage Software

Abstract

In Intel, typically a CPU design process takes around 2-3 years and critical part of it is the post silicon validation. Around 600K tests run every year under graphics environment for every project which results in hundreds of failures. A significant amount of time is spent by a validation engineer in triaging and debugging these failures. A validation engineer’s task is to understand whether it is a new failure or an existing debugged failure. At present, debugger has to manually go through all the log files that are generated by the tests and triage the failures. These failures can occur due to many different types of errors. Debugger manually creates buckets with an error signature title and bucketizes the same types of error under the same bucket. This helps in further debugging and root causing the failure. A post silicon validation debugging tool is essential to eradicate this manual effort and automate the bucketing process which helps in saving significant time of the debugger. Graphics Automated Triage Software is developed which automates the bucketing process, saving significant time of the debugger and hence helps in bringing the efficiency in the system. It is a failure analysis and an auto-bucketing tool. It parses the log files, fetches the required error data and matches with the existing data from the repository. The matched error signature title is the bucket title. All the error signatures with same title falls under the same bucket. GATS also allows the debugger to define their own error signatures in the user defined repository. This feature makes the tool more maintainable as debugger can add up the new signatures coming up.