KR100546295B1 - 2-level cache memory system reduces data transfer time - Google Patents

Abstract

A two-level cache memory system having reduced data transfer time according to the present invention includes a first level instruction cache storing a tag and a plurality of instructions associated with the tag, and a first level data cache storing a plurality of data associated with the tag and the tags. A first multiplexer for selectively outputting a tag of the first level instruction cache or a tag of the first level data cache in response to the first selection signal, an instruction stored in the first level instruction cache in response to the first selection signal, or a first A second multiplexer for selectively outputting data stored in the level data cache, a tag and a plurality of instructions associated with the tag, or a second level cache for storing data, a tag and a tag, and a first level instruction / data cache or a second level Big Buffer, First Level Instruction / Data Cache, or Second Level, which buffers a number of BigTek instructions or data generated by the cache. Compares the status bits of the cache to generate the Victim, and receives the Victim controller that transmits the Victim data buffered in the Victim buffer to the second level cache or main memory, and receives data or instructions read from the second level cache and the main memory. And a third multiplexer selectively outputting the first level instruction cache or the first level data cache in response to the second selection signal.

Description

2-Level Cache Memory System Reducing Data Transfer Time

1 is a block diagram illustrating a conventional two-level cache memory system.

2 is another block diagram illustrating a conventional two-level cache memory system.

Figure 3 is a block diagram of an embodiment for explaining a two-level cache memory system with reduced data transfer time in accordance with the present invention.

FIG. 4 is a diagram for describing a tag area of a cache memory in the cache memory system shown in FIG. 3.

5A and 5B are flowcharts for describing a memory access method performed in the cache memory system illustrated in FIG. 3.

The present invention relates to a cache memory system, and more particularly, to a two-level cache memory system with reduced data transfer time.

1 is a block diagram illustrating a conventional two-level cache memory system, including a processor 100, a level 1 instruction cache 110, a level 1 data cache 120, and multiplexers 130, 140, and 150. And level 2 cache 160.

Level 1 instruction cache 110 is composed of a tag (TAGS), and a cache block for storing a plurality of instructions associated with the tag. Here, the tag TAG represents a part of the instruction address INS_ADD output from the processor 100. In addition, the level 1 data cache 120 is composed of a tag and a cache block for storing a plurality of data associated with the tag. The multiplexer 130 selectively outputs an instruction address INS_ADD or a data address DT_ADD in response to the first selection signal S1.

Level 2 cache 160 consists of a tag and a cache block that stores a plurality of instructions or data associated with the tag. Here, each cache block stores data or instructions applied from main memory (not shown) or data transmitted from the level 1 data cache 120.

That is, a processor using a two-level cache memory, such as the processor 100 shown in FIG. 1, retrieves the level 1 data cache memory 120 when data is requested by a memory instruction, and if desired data exists. If not, the level 2 cache 160 is retrieved. At this time, if there is no data in the level 2 cache 160, the data is read from the external main memory, and the read data is read through the level 2 cache 160 and the level 1 cache 120. When data is read in this manner, the data stored in the level 1 instruction cache 110 and the level 1 data cache 120 is established in the level 2 cache 160. Therefore, a system configuration of a multiprocessor in which several processors share cache memory can be simplified. However, since the data applied from the external main memory is always written to the level 1 cache 120 via the level 2 cache 160 and output to the processor 100 through the level 1 cache 120, the data There is a disadvantage in that the transmission time of is long.

2 is another block diagram illustrating a conventional two-level cache memory system, which includes a processor 200, a level 1 instruction cache 210, a level 1 data cache 220, multiplexers 230-260, and a level. Two caches 270.

That is, in the cache memory system of FIG. 2, the processor 200, the level 1 instruction cache 210, and the level 1 data cache 220 have the same function as the block having the same name shown in FIG. 1. However, data or instructions applied from the memory are not stored in the level 2 cache 270, but are directly stored in the level 1 instruction cache or the level 1 data cache. That is, the multiplexer 260 selectively outputs data stored in the level 2 cache 270 or data RD_DATA read from a memory (not shown) to the level 1 cache in response to the data selection signal S3. Implementing the cache memory system in this manner has the advantage that the data transfer speed can be increased because the cache does not go through the level 2 cache 270 when data required by the processor 200 is read from the memory. However, the relationship that all data contents of the level 1 caches 210 and 220 should be included in the level 2 cache 270 may not be established. This can lead to performance degradation because the level 1 cache must be accessed in addition to the level 2 cache to achieve cache synchronization with an external processor when configuring a multiprocessor system. To prevent this, the external DTS (DUPLICATE TAG) STORE), that is, it is difficult to implement a duplicate tag storage. Therefore, there is a need for a new cache memory system capable of realizing a DTS by including a relationship between a level 1 cache and a level 2 cache while increasing the data transfer speed.

An object of the present invention is to provide a two-level cache memory system capable of reducing data transfer time while satisfying an inclusion relationship between a level 1 cache and a level 2 cache.

In order to achieve the above object, a two-level cache memory system having reduced data transmission time according to the present invention stores a first level instruction cache, a plurality of data associated with a tag and a tag, and a plurality of instructions associated with a tag, and a tag. A first multiplexer selectively outputting a tag of the first level instruction cache or a tag of the first level data cache in response to the first selection signal and a first level instruction cache in response to the first selection signal A second multiplexer for selectively outputting the data stored in the first level data cache or the instruction stored in the second multiplexer, the tag and the tag associated with the second level cache, the tag and the tag for storing the data; Victim buffer, first-level instruction that buffers a number of Victim instructions or data generated from a data cache or second-level cache Victims are generated by comparing the status bits of the data cache or the second level cache, and the Victim controller which transfers the buffered buffer data to the second level cache or the main memory to the second level cache or the main memory and reads from the second level cache and the main memory. A third multiplexer may be configured to receive data or instructions to be output and selectively output the first level instruction cache or the first level data cache in response to the second selection signal.

Hereinafter, a two-level cache memory system having a reduced data transfer time according to the present invention will be described with reference to the accompanying drawings.

Figure 3 is a block diagram of an embodiment for explaining a two-level cache memory system with reduced data transfer time in accordance with the present invention. Referring to FIG. 3, the cache memory system includes a level 1 instruction cache 310, a level 1 data cache 320, multiplexers 330, 335, and 440, a Victim buffer 360, a multiplexer 350, and a level 1. The cache tag storage unit 380, the Victim control unit 390, and the level 2 cache 370 are included. For convenience of description, the processor 300 is shown together with FIG. 3.

The processor 300 of FIG. 3 is a block for executing and processing a predetermined program. The processor 300 outputs an instruction address INS_ADD or a data address DT_ADD to obtain necessary instructions or data. Access a memory (not shown).

The level 1 instruction cache 310 is configured to store a tag and a plurality of instructions associated with the tag, and outputs an instruction corresponding to the tag of the instruction address INS_ADD applied by the processor 300 to the processor 300, Stores instructions received from level 2 cache 370 or instructions read from main memory. Here, the tag is connected to the instruction address bus 318 and the instruction is connected to the instruction bus 315.

The level 1 data cache 320 is configured to store a tag and a plurality of data associated with the tag, and outputs data corresponding to the tag of the data address DT_ADD applied by the processor 300 to the processor 300, Stores data read from level 2 cache 370 or main memory. Here, the tag is connected to the data address bus 328 and the data is connected to the data bus 325.

The multiplexer 330 selectively outputs the tag of the level 1 instruction cache 310 or the tag of the level 1 data cache 320 to the tag area of the victor buffer 360 in response to the first selection signal S1.

The multiplexer 335 selectively outputs the instruction address INS_ADD or the data address DT_ADD applied from the processor 300 in response to the third selection signal S3, and the output result is the level of the Victim control unit 390. It is applied to the two caches 370 and is output as a memory read address MEM_R_ADD when a cache miss occurs in the level 1 cache and the level 2 cache.

The level 2 cache 370 stores a tag and a plurality of instructions or data connected to the tag, and outputs the instruction or data to the processor 300 through the level 1 cache at a memory address.

The Victim buffer 360 is connected to the tag and the tag to buffer a plurality of Victim instructions or data generated in the first level instruction / data cache 310/320 or the second level cache 370.

The level 1 cache tag storage unit 380 stores duplicate tags of the level 1 data cache 320 in order to efficiently control the big teams generated in the level 1 and 2 caches. In this way, by separately storing the tag of the level 1 data cache 320 is redundantly stored so that the data read / write operation between the processor 300 and the level 1 data cache 320 is not prevented.

The Victim controller 390 compares the status bits of the Level 1 data cache 320 or the Level 2 cache 370 to generate or invalidate the Vic Team, and to convert the Victim Data buffered in the Victim Buffer 360 to the Level 2 Cache 370. ) Or transfer to main memory.

The multiplexer 350 selectively stores the data stored in the level 2 cache 370 or the data RD_DATA read from the main memory in response to the second selection signal S2 as the level 1 command cache 310 or the level 1 data cache. Output

FIG. 4 is a diagram for describing tag areas of level 1 and 2 caches of the two-level cache memory system shown in FIG. 3. Referring to FIG. 4, the tag area of the level 1 and level 2 caches includes a valid bit, a back up bit, a dirty bit, and a shared bit in addition to the actual tag part. Contains the status bits.

In FIG. 4, the valid bit is an information bit indicating whether each cache block of the level 1 cache or the level 2 cache, that is, each data storage area is valid. A valid bit indicates that 0 is invalid and 1 is valid. (valid).

The BACK UP BIT indicates whether each cache block of the level 1 cache is backed up to the level 2 cache. For example, the backup bit of the level 1 cache 320 is a bit indicating whether a predetermined data block is backed up to the level 2 cache 370, and is written as 0 when data is read from the main memory. When data is read from the level 2 cache 370, the data is read to the level 2 cache 370.

In addition, the backup bit of the level 2 cache 370 indicates whether each data block of the level 2 cache 370 exists in the level 1 cache. Therefore, the backup bit is written as 0 when data is applied from the level 1 cache 310 or 320, and set to 1 when data is transmitted in the level 1 cache. In this case, when the processor 300 accesses the memory to update the data blocks of the level 1 caches 310 and 320, that is, during the data write operation, the backup bit of the level 1 cache is checked to be 1, and the level 2 cache 370 is checked. Invalidate the block of data and set the backup bit of the level 1 cache to zero. However, if the backup bit of the level 1 cache is zero, the data block is updated as it is.

If the number of sets of cache memories is one or more, that is, if the number of set combinations is larger than one, a Least Recently Used (LRU) bit may be added for more efficient cache access. Here, the LRU bit may be referred to as a status bit indicating which block has not been used recently in the set of data storage areas.

Also, the DIRTY BIT is information indicating that the data stored in the data block is different from the data stored in the memory block of the same address. That is, a dirty bit of 1 indicates that the data is different from the data in the memory block, and a zero bit indicates the same as the data in the memory block (CLEAN).

In addition, SHARED BIT is information indicating whether other processors are sharing a cache in a multiprocessor system. A shared bit indicates that the shared bit is 0, and 0 is not shared.

As described above, in the present invention, by adding the backup bit to the status bit indicating the cache state, it is possible to determine whether data is backed up between the level 1 cache or the level 2 cache. That is, based on the determination result, it is possible to efficiently control the Victim generated in the level 1 cache 310 or 320 or the level 2 cache 370, and as a result, the sum of the data stored in the level 1 cache and the level 2 cache. It can be implemented to match the data of level 2 cache using this conventional method.

5A and 5B are flowcharts illustrating a cache access method performed in a two-level cache memory system according to the present invention. 5A to 5B, the cache access method may include determining whether a hit has occurred by transmitting data of a level 1 cache to a processor or accessing a level 2 cache according to whether a level 1 cache cache hit occurs. Step 500 to 510), if a hit occurs in the level 2 cache, processing the big team of the level 1 cache and transmitting data of the level 2 cache to the processor (steps 520 to 530), and if the hit does not occur in the level 2 cache Processing a big team generated in the level 1 cache and the level 2 cache, and accessing the main memory to output the read data to the processor through the level 1 cache (steps 550 to 560).

Hereinafter, an operation and a cache access method of a two-level cache memory system having reduced data transfer time according to the present invention will be described in detail with reference to FIGS. 3 and 5. Here, it is assumed that it is the case of cache access for reading data.

First, it is determined whether a cache hit has occurred by the data address DT_ADD applied by the processor 300 (operation 500). In this case, determining whether a cache hit is generated is performed by determining whether a tag of the data address DT_ADD applied from the processor 300 and a tag of the cache block read from the level 1 data cache 320 match. If it is determined in step 500 that the cache hit occurs, the data associated with the matched tag of the level 1 data cache 320 is transmitted to the processor 300 (step 508).

On the other hand, if a hit does not occur in the level 1 cache 320, the level 2 cache 370 is accessed (step 505). At this time, the tag of the data address DT_ADD and the tag of the data block read from the level 2 cache 370 are compared to determine whether the hit occurs in the level 2 cache 370 to determine whether a hit has occurred (step 510). . If it is determined in step 510 that the cache hit is generated, it is determined whether a big team VICTIM is generated in the level 1 cache 320 and the generated big team is processed (step 520). That is, it is determined whether a big team is generated in the level 1 cache 320 (step 522). If it is determined that a big team is generated, the big team generated in step 522 is buffered in the big team buffer 360 and then level 2 is generated. The cache 370 is stored in operation 524. Here, the generation or processing of the big team is performed by the big team controller 390. The Victim control unit 390 uses the status bits of the tag area stored in the level 1 cache tag storage unit 380 and the previous data block existing in the data block where the new block is to be located using the backup bit. If 0, big team is created. If the big team is processed in operation 524, the data associated with the matched tag is transmitted from the level 2 cache 370 to the level 1 cache 320 and the backup bit of the level 2 cache 370 is set to 1 ( Step 532). Thereafter, the data received by the level 1 cache 300 is transmitted to the processor 300 through the data line 325 (operation 534). In addition, when the big team is not generated in the level 1 cache 320 by the step 522, the process proceeds to a step 532.

On the other hand, if it is determined in step 510 that the cache hit has not occurred in the level 2 cache 370, the big team generated in the level 1 cache 320 and the level 2 cache 360 are processed (step 550). In detail, it is determined whether a big team is generated in the level 1 cache 320 (operation 552). In this case, when it is determined that the big team is generated in the level 1 cache 320, the generated big team is buffered in the big team buffer 360 and stored in the level 2 cache 370 or a corresponding data block of the main memory (step 554). ). As such, when the big team generated in the level 1 cache 320 is processed, it is determined whether the big team is generated in the level 2 cache 370 (step 556). If a big team is generated in the level 2 cache 370, the generated big team is buffered in the big team buffer 360 and stored in the main memory (not shown) (step 558). As described above, the Victim generated in the Level 1 data cache 320 or the Level 2 cache 370 is buffered and processed by the Victim controller 390. At this time, when the Victim processing is completed, the main memory is accessed to obtain required data (step 560). As such, the processor 300 accesses the main memory to read desired data. Data read from the memory is stored in the level 1 cache 320.

As described above, in the present invention, since data read from the main memory is directly transferred to the level 1 cache without passing through the level 2 cache 370, the data transfer time is shortened. In addition, the Victim controller 390 determines whether the data is backed up between the Level 1 cache 320 and the Level 2 cache 370 and controls the Vic Team so that an inclusion relationship between the caches is established.

An operation performed by the Victim controller 390 shown in FIG. 3 will be described in more detail with reference to the cache access method. This method is applicable regardless of the number of sets of caches, one example of which is implemented as a two way cache in which the level 1 cache 320 and the level 2 cache 370 are each combined into two sets. The operation of the case will be described. A new function different from the conventional cache is when a cache miss occurs in the level 1 and 2 caches to access the main memory (not shown).

First, if both sets of Level 1 cache 320 blocks are invalid and both Level 2 caches 370 are valid, then the LRU blocks of Level 2 cache 370, i.e. Select one data block that has not been used recently. At this time, if the selected LRU block is dirty, that is, the dirty bit of the status bit is 1, the LRU block is optimized into memory and invalidated when it is clear.

In addition, if both blocks of the level 1 cache 320 are valid, and the address to be read and the index of the level 2 cache 370 are one block and the other block is different, the following description will be given. . That is, when there is a block having a different level 2 cache index among the data blocks of the level 1 cache 320, if the backup bit is 0, the level 2 cache 370 is optimized, and if the backup bit is 1, it is invalidated. In addition, when the address to be read is the same as the level 2 cache index, and there is one of the level 2 cache blocks valid and the backup bit is 0, when the block whose backup bit is 0 is dirty, it is optimized into memory. If it is clear, it is invalidated.

On the other hand, if both blocks of the level 1 cache 320 are valid and both the address to be read and the index of the level 2 cache 370 are the same, the LRU block is selected from the level 1 cache block and is dirty. Victimize and invalidate if cleared. In addition, when a levelized / invalidated block exists in the level 2 cache 370 in the level 1 cache 320, the block is invalidated. At this time, it is determined by determining whether the backup bit is 1 in whether the levelized / invalidated block is present in the level 2 cache 370 in the level 1 cache 320.

When creating a big team in this way, there is only a maximum of two sets of level 1 indexes for the sum of the level 1 cache data and the level 2 cache data. In addition, it is possible to implement the level 2 cache in an n way, and to implement a DTS for cache synchronization in a multiprocessor system.

According to the present invention, a two-level cache memory system can be implemented such that a level 2 cache can virtually include a level 1 cache, thereby simplifying system configuration and reducing data transfer time in a multiprocessor system. In addition, since a virtual inclusion relationship is established between the level 1 cache and the level 2 cache, the level 2 cache can be implemented in an n way and a DTS for cache synchronization in a multiprocessor system can be implemented.

Claims (3)

A first level instruction cache that stores a tag and a plurality of instructions associated with the tag, wherein the tag is coupled by an instruction address bus, and wherein the instructions are coupled by an instruction bus; A first level data cache for storing a tag and a plurality of data associated with the tag, wherein the tag is connected by a data address bus and the data is connected by a data bus; A first multiplexer for selectively outputting a tag of the first level instruction cache or a tag of the first level data cache in response to a first selection signal; A second multiplexer for selectively outputting instructions stored in the first level instruction cache or data stored in the first level data cache in response to the first selection signal; A second level cache for storing a tag and a plurality of instructions or data associated with the tag; A tag buffer connected to the tag and the tag to buffer a plurality of VicT instruction or data generated in the first level instruction / data cache or the second level cache; A Victim controller for generating the Victim by comparing the status bits of the first level instruction / data cache or the second level cache and transmitting the Victim data buffered in the Victim buffer to the second level cache or main memory; And A third multiplexer which receives the second level cache and data or instructions read from the main memory and selectively outputs the second level cache to the first level instruction cache or the first level data cache in response to a second selection signal; A two-level cache memory system. The method of claim 1, And storing the tag of the first level data cache in duplicate. The method of claim 1, wherein the first level instruction / data cache and the second level cache are: A backup bit is included in each of the status bits, The backup bit of the first level command / data cache, Is set to a first state when read from the main memory and updated with new data, and is set to a second state when data is received from the second level cache, The backup bit of the second level cache is set to a first state when data or an instruction is received from the first level instruction / data cache, and when predetermined data is transmitted to the first level instruction / data cache. And a second level in the second state.

Images