Method for spatial memory stream prefetching engine, apparatus, manufacture method and test method

Abstract

The invention describes a method for a spatial memory stream prefetching engine, an apparatus, a manufacture method and a test method. In one aspect, the spatial memory stream prefetching engine usestrigger-to-trigger stride detection to make training table entries to be pattern history table entries and to drive spatially related prefetching in regions of greater distance. In the other aspect, the spatial memory stream prefetching engine maintains a blacklist of program counter values to not be used as trigger values. In another aspect, the spatial memory stream prefetching engine uses, forexample, a filter table, a training table, and a hash value of a particular field of the trigger program counter in an entry in the pattern history table as an index value for a table.

Description

[0001] priority [0002] This application claims priority to U.S. Application Serial No. 15 / 690,931, filed August 30, 2017, and to U.S. Provisional Patent Application Serial No. 62 / 505,318, filed May 12, 2017 and 7, 2017 Priority to U.S. Provisional Patent Application Serial No. 62 / 535,460, filed May 21, all applications are incorporated herein by reference in their entirety. technical field [0003] The present disclosure relates generally to processor memory management, and more specifically, to spatial memory stream confidence mechanisms. Background technique [0004] A computer processor performs cache prefetching to store instructions or data by fetching instructions or data from its initial storage in slower memory (i.e., with slower read / write times) to a faster location before the instruction or data actually needs to be fetched. Faster local memory (ie, has faster read / write times and is often located closer to the instruction / data pipeline) to enhance execution pe...

AI Technical Summary

Problems solved by technology

[0008] However, the spatial memory streaming scheme has several disadvantages

Spatial memory streaming cannot handle shift alignment of patterns with respect to line boundaries between caches

In addition, spatial bit vectors are often used to access spatial patterns at a greater granularity per access and cannot track temporal order

Spatial memory streaming also lacks a robust confidence mechanism and is not dynamically adaptable, i.e. spatial memory streaming cannot adapt to program phase changes when dynamic branching behavior changes offset patterns, for example

These disadvantages lead to reduced coverage and accuracy, loss of timeliness, thereby reducing performance and increasing power consumption

Images