Method and apparatus for data mining from core traces

An apparatus for data mining from core traces in a processing system for wireless baseband design, comprising:

• a non-transitory memory storage comprising instructions;
• and
• one or more processors in communication with the memory, wherein the one or more processors execute the instructions to:
• detect a core trace in the processing system, by an accumulator the core trace being a sequence of instructions executed in the processing system;
• map instruction addresses in the core trace to one or more functions of the processing system;
• sort the mapped functions into a hierarchical format;
• identify a gene function in the hierarchical format, by a data analysis engine, the gene function comprising a fundamental function executed by the processing system and defined by a name, number of loads and stores, memory usage, data flow, workload, pre-emption and power usage, and wherein the gene function is executed as one of the sequence of instructions executed by one or more processors in the processing system;
• derive attributes for the gene function from the hierarchical format, the attributes including at least memory usage, data throughput, workload, pre-emption and power usage of the gene function, the power usage of the processing system during instruction execution, the power usage being derived from counts of load/store and Arithmetic Logic Unit (ALU) operations;
• detect a pre-emption event based on the core trace during execution of the gene function by the one or more processors, wherein a percentage of the core trace attributable to the pre-emption event is defined as a sum, for all pre-emption events, of a size of first and second pre-empting tasks plus overhead, divided by the size of the core trace;
• and
• determine a cost attributable to the pre-emption event by the data analysis engine, the cost of the pre-emption event having an impact on processing latency of at least the first pre-emption task, and, based on the cost attributable to the pre-emption event exceeding a threshold, change an order of the sequence of instructions to thereby reduce the pre-emption cost, wherein the cost of the pre-emptive event is determined as a size of the first pre-emption task plus overhead, where the size is defined by a number of instructions in the sequence of instructions and the overhead includes operating system instructions used to start and end the pre-emption event and is derived from the core trace.

A non-transitory computer-readable medium storing computer instructions for data mining from core traces in a processing system for wireless baseband design, that when executed by one or more processors, cause the one or more processors to:

• detect a core trace in the processing system by an accumulator, the core trace being a sequence of instructions executed in the processing system;
• map instruction addresses in the core trace to one or more functions of the processing system;
• sort the mapped functions into a hierarchical format;
• identify a gene function in the hierarchical format, by a data analysis engine, the gene function comprising a fundamental function executed by the processing system and defined by a name, number of loads and stores, memory usage, data flow, workload, pre-emption and power usage, and wherein the gene function is executed as one of the sequence of instructions executed by one or more processors in the processing system;
• derive attributes for the gene function from the hierarchical format, the gene function defined by a name, number of loads and stores, memory usage, data flow, workload, pre-emption and power usage, and the power usage of the processing system during instruction execution, the power usage being derived from counts of load/store and Arithmetic Logic Unit (ALU) operations;
• detect a pre-emption event based on the core trace during execution of the gene function by the one or more processors, wherein a percentage of the core trace attributable to the pre-emption event is defined as a sum, for all pre-emption events, of a size of first and second pre-empting tasks plus overhead, divided by the size of the core trace;
• and
• determine a cost attributable to the pre-emption event by the data analysis engine, the cost of the pre-emption event having an impact on processing latency of at least the first pre-emption task, and, based on the cost attributable to the pre-emption event exceeding a threshold, change an order of the sequence of instructions to thereby reduce the pre-emption cost, wherein the cost of the pre-emption event is determined as a size of first pre-empting task plus overhead, where the size is defined by a number of instructions in the sequence of instructions and the overhead includes operating system instructions used to start and end the pre-emption event and is derived from the core trace.