COORDINATED PREFETCHING IN HIERARCHICALLY CACHED PROCESSORS

• Main Authors: SADOUGHI YARANDI MAHNAZ, KANNAN HARI S, SUBRAMONIAM PERUMAL R, KANAPATHIPILLAI PRADEEP, LILLY BRIAN P, WILLIAMS III GERARD R
• Format: Patent
• Language: English; Korean
• Published: 30.09.2013
• Subjects: CALCULATING; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

PURPOSE: A pre-fetching method is provided to adjust effectively processors which are caching hierarchically. CONSTITUTION: A pre-fetching method for adjusting effectively at processors which are caching hierarchically comprises the following steps: a system generates pre-fetch demands; a training mechanism combined to a first cash level is trained about a given stream (130); training information is stored in a first table (132); the training information is delivered to a second cash level (134); the second cash level receives the training information (136); pre-fetch demands are generated at the second cache level based on the received training information (138); at the second cash level, the training information is stored in a second table (140); the training information is delivered to a third cash level (142); the third cash level uses the received information (144); the pre-fetch demands are published (145); the training information is stored in a third table (148); and a pre-fetching of processors which are caching is progressed while a similar task is repeated. [Reference numerals] (130) Train a stream given at a first cash level; (132) Store training information in a first table at the first cash level; (134) Deliver the training information from the first cache level to a second cash level; (136) Receive training information at the second cache level; (138) Generate pre-fetch demands at the second cache level based on the received training information; (140) Store training information in a second table at the second cash level; (142) Training information is delivered to a third cache level; (144) Training information is received at the third cash level; (146) Generate pre-fetch demands at the third cache level based on the received training information; (148) Store the training information in the third table at a third cache level; (AA) Start-generate pre-fetch demands