Method, Apparatus, and Program to Efficiently Calculate Cache Prefetching Patterns for Loops

A computer program product, in a recordable-type computer readable medium, for determining cache prefetching patterns, the computer program product comprising:

• instructions for identifying an instruction as a candidate for cache prefetching;
• instructions for placing the instruction in data gathering mode by changing an op-code of the instruction to form a new op-code;
• instructions, upon executing new op-code, recording, in an array of address differences, a difference between a current address being referenced and a last address being referenced, wherein the last address being referenced refers to an address from the immediately previously executed op-code;
• and
• instructions, responsive to a number of address differences being recorded in the array of address differences exceeding a threshold and referring to a same cache line, for determining an expected cache stride value based on the array of address differences.

An apparatus for determining cache prefetching patterns, the apparatus comprising:

• means for identifying an instruction as a candidate for cache prefetching;
• means for placing the instruction in data gathering mode by changing an op-code of the instruction to form a new op-code;
• means for recording, upon executing the new op-code, in an array of address differences, a difference between a current address being referenced and a last address being referenced, wherein the last address being referenced refers to an address from the immediately previously executed op-code;
• and
• responsive to a number of address differences being recorded in the array of address differences exceeding a threshold and referring to a same cache line, determining an expected cache stride value based on the array of address differences.