Texas Instruments Corporation

Oluleye Olorode

Ramakrishnan Venkatasubramanian

Mehrdad Nourani

Hung Ong

Bipin Prasad Heremagalur Ramaprasad

This invention involves a cache system in a digital data processing apparatus including: a central processing unit core; a level one instruction cache; and a level two cache. The cache lines in the second level cache are twice the size of the cache lines in the first level instruction cache. The central processing unit core requests additional program instructions when needed via a request address. Upon a miss in the level one instruction cache that causes a hit in the upper half of a level two cache line, the level two cache supplies the upper half level cache line to the level one instruction cache. On a following level two cache memory cycle, the level two cache supplies the lower half of the cache line to the level one instruction cache. This cache technique thus prefetches the lower half level two cache line employing fewer resources than an ordinary prefetch.

Zero latency prefetching in caches

Disclosed embodiments relate to a dNap architecture that accurately transitions cache lines to full power state before an access to them. This ensures that there are no additional delays due to waking up drowsy lines. Only cache lines that are determined by the DMC to be accessed in the immediate future are fully powered while others are put in drowsy mode. As a result, we are able to significantly reduce leakage power with no cache performance degradation and minimal hardware overhead, especially at higher associativities. Up to 92% static/leakage power savings are accomplished with minimal hardware overhead and no performance tradeoff.

Static power reduction in caches using deterministic naps

Disclosed embodiments provide a technique in which a memory controller determines whether a fetch address is a miss in an L1 cache and, when a miss occurs, allocates a way of the L1 cache, determines whether the allocated way matches a scoreboard entry of pending service requests, and, when such a match is found, determine whether a request address of the matching scoreboard entry matches the fetch address. When the matching scoreboard entry also has a request address matching the fetch address, the scoreboard entry is modified to a demand request.

Servicing CPU demand requests with inflight prefetches

Disclosed embodiments relate to a dNap architecture that accurately transitions cache lines to full power state before an access to them. This ensures that there are no additional delays due to waking up drowsy lines. Only cache lines that are determined by the DMC to be accessed in the immediate future are fully powered while others are put in drowsy mode. As a result, we are able to significantly reduce leakage power with no cache performance degradation and minimal hardware overhead, especially at higher associativities. Up to 92% static/Leakage power savings are accomplished with minimal hardware overhead and no performance tradeoff.

Static power reduction in caches using deterministic Naps

This invention involves a particular cache hazard. It is possible that an instruction request that is a miss in the cache occurs while the cache system is servicing a pending prefetch for the same instructions. In the prior art, this hazard is detected by comparing request addresses for all entries in a scoreboard. The program memory controller stores the allocated way in the scoreboard. The program memory controller compares the allocated way of the demand request to the allocated way of all the scoreboard entries. The cache hazard only occurs when the allocated ways match. Following way compare, the demand request address is compared to the request addresses of only those scoreboard entries having matching ways. Other address comparators are not powered during this time. This serves to reduce the electrical power required in detecting this cache hazard.

Servicing CPU demand requests with inflight prefetchs

The dNap architecture is able to accurately transition cache lines to full power state before an access to them. This ensures that there are no additional delays due to waking up drowsy lines. Only cache lines that are determined by the DMC to be accessed in the immediate future are fully powered while others are put in drowsy mode. As a result, we are able to significantly reduce leakage power with no cache performance degradation and minimal hardware overhead, especially at higher associativities. Up to 92% static/Leakage power savings are accomplished with minimal hardware overhead and no performance tradeoff.

With the increasing demand for improved processor performance, memory systems have been growing increasingly larger to keep up with this performance demand. Caches, which dictate the performance of memory systems are often the focus of improved performance in memory systems, and the most common techniques used to increase cache performance are increased size and associativity. Unfortunately, these methods yield increased static and dynamic power consumption. In this invention, a technique is shown that reduces the power consumption in associative caches with some improvement in cache performance. The architecture shown achieves these power savings by reducing the number of ways queried on each cache access, using a simple hash function and no additional storage, while skipping some pipe stages for improved performance. Up to 90% reduction in power consumption with a 4.6% performance improvement was observed.

Dynamic power reduction and performance improvement in caches using fast access

This invention hides the page miss translation latency for program fetches. In this invention whenever an access is requested by CPU that crosses a memory page boundary, the L1I cache controller request a next page translation along with the current page. This pipelines requests to the &#x3bc;TLB without waiting for L1I cache controller to begin processing the next page requests. This becomes a deterministic prefetch of the second page translation request. The translation information for the second page is stored locally in L1I cache controller and used when the access crosses the next page boundary.

Growing community of inventors

Garland, TX, United States of America

Oluleye Olorode