Cache memory bank access prediction

Details Cache memory bank access prediction Cache memory bank access prediction

1999-12-29

2004-02-17

Peikari, B. James (Department: 2186)

Electrical computers and digital processing systems: memory

Address formation

Address mapping

C711S213000, C711S005000, C711S120000, C711S129000, C711S168000

Reexamination Certificate

active

06694421

ABSTRACT:

FIELD
The present invention relates to processors. More particularly, the present invention relates to memory cache bank prediction in a processor.
BACKGROUND
To more efficiently access data, many processors move information from a main memory, which can be slow to access, to a memory “cache” which allows for faster access. In addition, modern processors schedule instructions “out of order” and execute multiple instructions per cycle to achieve high performance. Some instructions, however, need to access information stored in the memory cache. For example, about one-third of all micro-instructions executed may be load/store instructions which access the memory cache. In order to achieve high performance by executing multiple instructions in a single cycle, the system should therefore permit more than one concurrent memory cache access in a cycle.
There are several ways to accomplish this goal. A truly “multi-ported” cache, i.e. one that supports multiple simultaneous accesses, fulfills this goal, but this is a complex solution that can be costly to implement in terms of area, power and speed.
Another known solution is a “multi-banked” cache. In this scheme, the memory cache is split into several independently addressed banks, and each bank supports one access per cycle.
FIG. 1
illustrates a system having such a multi-bank memory cache. The system includes a first memory cache bank
240
and a second memory cache bank
340
. A scheduling unit
100
schedules instructions to one of two pipelines. For example, the scheduling unit
100
may schedule an instruction to a first pipeline such that the instruction is processed by an Address Generation Unit (AGU)
210
and ultimately by a cache access unit
230
. The scheduling unit
100
may instead schedule an instruction to a second pipeline such that the instruction is processed by another AGU
310
and ultimately by another cache access unit
330
.
This is a sub-ideal implementation because only accesses to different banks are possible concurrently. This is done in order to reduce the complexity and cost of the cache, while still allowing more than one memory access in a single cycle. As a result, an instruction being processed by the first pipeline that needs to access information in the second memory cache bank
340
may not be able to execute.
To solve that problem, each instruction pipeline may use a “cross-bar” to access information in the other memory cache bank. For example, a set up latency
220
,
320
may be incurred while the pipeline accesses information in the other memory cache bank. This delay, however, slows the operation of the pipeline.
If the memory cache bank associated with each load instruction was known, the processor could schedule load instructions in such a way so as to maximize the utilization of the banks
240
,
340
and approximate true multi-porting. However, in current processors this is not done because the scheduling precedes the bank determination.
SUMMARY
In accordance with an embodiment of the present invention, a memory cache bank prediction unit is provided for use in a processor having a plurality of memory cache banks. The memory cache bank prediction unit has an input port that receives an instruction. The memory cache bank prediction unit also has an evaluation unit, coupled to the input port, that predicts which of the plurality of memory cache banks is associated with the instruction.


REFERENCES:
patent: 5752259 (1998-05-01), Tran
patent: 5933860 (1999-08-01), Emer et al.
patent: 5953747 (1999-09-01), Steely et al.
patent: 6005592 (1999-12-01), Koizumi et al.
patent: 6173333 (2001-01-01), Jolitz et al.
patent: 6189083 (2001-02-01), Snyder, II

Affiliated with

Also associated with

Rating

Cache memory bank access prediction does not yet have a rating. At this time, there are no reviews or comments for this patent.

If you have personal experience with Cache memory bank access prediction, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Cache memory bank access prediction will most certainly appreciate the feedback.

Rate now

Comments { 0 }

Profile ID: LFUS-PAI-O-3290367