Browse Prior Art Database

Method for completion-time memory renaming Disclosure Number: IPCOM000008570D
Publication Date: 2002-Jun-24
Document File: 3 page(s) / 89K

Publishing Venue

The Prior Art Database


Disclosed is a method for completion-time memory renaming. Benefits include improved functionality.

This text was extracted from a Microsoft Word document.
At least one non-text object (such as an image or picture) has been suppressed.
This is the abbreviated version, containing approximately 50% of the total text.

Method for completion-time memory renaming

Disclosed is a method for completion-time memory renaming. Benefits include improved functionality.


              Associativity increases performance by increasing a cache’s hit rate. Conventional microprocessors employ caches that maintain associativity at a value greater than 1. The associative lookup is typically performed at execution time. Associative lookup can increase the latency of the cache. Optimization techniques are typically applied to the cache to reduce this penalty.

General description

              The disclosed method is completion-time memory renaming. Its goal is to increase the associativity of the DL1 cache while not increasing its access latency. By renaming the data memory addresses at completion time, the associative lookup can be moved off the critical processing path. With increased associativity, the DL1 cache has a lower miss rate, which increases overall performance.

              Completion-time memory renaming changes the data memory addresses at completion time (or retirement time). Data loads and store addresses go through a rename stage prior to cache access. However, the load renaming is performed at the completion time of its prior execution. Therefore, a load receives the benefit of associativity without incurring the extra execution latency. The key to this implementation is that the associativity is implemented at the rename stage. Therefore, the associative lookup is off the critical path of execution.


              The disclosed method provides advantages including improved performance due to reduced latency because the complexity of associativity is moved from the execute time to the completion time.

Detailed description

              The disclosed method includes several major functions (see Figure 1):

·        Load prediction table

·        Rename table

·        Load address predictor table

·        General functionality

Load prediction table

              The purpose of the load prediction table is to hold the renamed data address prediction on a per static load basis. Each entry represents a static load and holds a renamed data address for the next execution. For example, a load (load(EX)) always accesses address X (see Figure 1). Our load address predictor predicts that this load will access address X the next time it executes. Assume that data address X is renamed to entry 5 in the data cache. This load’s...