WO1981001894A1 - Cache memory in which the data block size is variable - Google Patents
Cache memory in which the data block size is variable Download PDFInfo
- Publication number
- WO1981001894A1 WO1981001894A1 PCT/US1980/001666 US8001666W WO8101894A1 WO 1981001894 A1 WO1981001894 A1 WO 1981001894A1 US 8001666 W US8001666 W US 8001666W WO 8101894 A1 WO8101894 A1 WO 8101894A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- data
- index
- bits
- main memory
- block
- Prior art date
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F12/00—Accessing, addressing or allocating within memory systems or architectures
- G06F12/02—Addressing or allocation; Relocation
- G06F12/08—Addressing or allocation; Relocation in hierarchically structured memory systems, e.g. virtual memory systems
- G06F12/0802—Addressing of a memory level in which the access to the desired data or data block requires associative addressing means, e.g. caches
- G06F12/0864—Addressing of a memory level in which the access to the desired data or data block requires associative addressing means, e.g. caches using pseudo-associative means, e.g. set-associative or hashing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2212/00—Indexing scheme relating to accessing, addressing or allocation within memory systems or architectures
- G06F2212/60—Details of cache memory
- G06F2212/601—Reconfiguration of cache memory
Definitions
- the present invention relates to data proces ⁇ sing systems and, more particularly, to cache memories used in data processing systems.
- a cache memory is essentially a data buffer or fast memory that quickly provides data in response to memory address signals.
- the cache memory generally resides within a processor and acts as an intermediate storage between the main memory of the system and the processing elements of the processor.
- a processor often has its processing speed affected by the time that it takes to fetch data or instructions from the main memory. With a fast cache memory within the processor storing the most recently used data from main memory, the processor will, in most circumstances, be able to quickly get its needed data and instructions from the cache memory.
- the cache memory since the cache memory resides within the processor, the data and instructions in the cache memory can be provided much more quickly than they could if they needed to be fetched along a bus from main memory.
- LRU least recently used
- the present invention provides a data proces ⁇ sing system having a main memory for storing data, a processor for processing data from said main memory, and
- said cache memory stores data blocks having a variable number of data words and comprises a data buffer for receiving and storing data blocks from said main memory, said data buffer arranged in banks with each bank having a plurality of data blocks; and index for storing an address block corresponding to each data block stored in said data buffer and comprised of bits in the main memory address of the corresponding data block, said index including an index data input for receiving main memory address bits for comparison with the main memory address bits stored in said address blocks, and an index address input for receiving index address input bits selecting a data block in each of the banks for comparison with the main memory address bits at said index data input; and inhibit means for inhibit ⁇ ing selected ones of the index address input bits at said index address input in order to reduce the number of addressable address blocks in said index so that the number of data words in the corresponding data blocks in said data buffer may be increased.
- the number of words in a data block of the cache memory may be selected in advance of a task to be performed in order to optimize the hit ratio; for example, where the task involves the use of small clusters of data that are widely spread throughout the main memory the data blocks will be preselected to be small.
- Fig. 1 is a simplified block diagram of a data processing system, including a processor having a cache memory.
- Fig. 2 is a block diagram illustrating the index and data buffer of a cache memory, such as the -4-
- Figs. 3A, 3B and 3C illustrate the contents of a main memory, and the contents of an index and data buffer of a cache memory for storing blocks of data from the main memory.
- Fig. 4 is a block diagram of a cache memory, including an index and data buffer, having data blocks of variable size in accordance with the present inven- tion.
- Figs. 5A and 5B illustrate the contents of the index and data buffer of Fig. 4.
- Fig. 6 illustrates the bits of a main memory address and their use in varying the block size of the cache memory shown in Fig. 4.
- Fig. 7 is a detailed block diagram of the address inhibit circuit shown in Fig. 4.
- a data processing system 10 including a processor 12, a main memory 14, and an internal transfer bus 16 for connecting the processor 12 to the main memory 14.
- the bus 16 also connects other subsystems 18 to each of the main memory 14 and processor 12.
- the processor 12 can include an arithmetic-logic unit (ALU) 22, an instruction storage unit (ISU) 24, a cache memory 26, as well as other well-known and conventional components (not shown), all connected to each other and to the in ⁇ ternal transfer bus 16 by a processor data bus 28.
- ALU arithmetic-logic unit
- ISU instruction storage unit
- cache memory 26 as well-known and conventional components (not shown)
- the main memory 14 normally holds data and instructions that are needed by the processor 12, such data and instructions initially loaded into the main memory by card readers, disk units, or other peripheral devices that might be found in the other subsystems 18.
- the data and instructions stored in main memory 14 are provided to the processor 12 over internal transfer bus 16.
- the ISU 24 will normally provide a microprogram for execution and, in accordance with microinstructions in that microprogram, data will be fetched from main memory 14 to be operated on by the ALU 22.
- the data in the main memory 14 is organized in blocks, each block normally comprising a number of data words.
- the block in which it resides is stored in the cache memory 26. Since there is a likelihood that, when the processor accesses one word in a block and then subsequently needs a different word, the different word will be adjacent the first in the same block, the proces ⁇ sor will in many cases only need to access the cache memory 26.
- the blocks of data held in the cache memory are stored in accordance with a least recently used (LRU) algorithm so that when the processor needs a new data word not in the cache memory 26, and a new block of data fetched from main memory 14 is to be stored in the cache memory 26, the block least recently used by the processor in the cache memory is removed and the new block of data is stored in its place.
- LRU least recently used
- Processor hardware for implementing an LRU algorithm, including the transfer of data blocks between the main memory and cache memory is well-known in the art.
- Commercially available systems having cache memories and using LRU algorithms include, among others, the Class 370/Model 168 system available from IBM Corporation, Armonk, New York.
- the cache memory 26 is generally comprised of two memory components, an index 32 and a data buffer 34.
- the data buffer 34 stores the blocks of data from main memory 14, and the index 32 stores the main memory addresses associated with the data blocks in the data buffer 34.
- the processor checks for a given data word in the cache memory 26, it checks the addresses stored in the index 32 and, if there is a match, the selected data word is read from the data buffer 34 to the bus 28 for use by the processor.
- the cache memory 26 further includes a fetch generate counter 60 and a fetch return counter 62.
- a fetch generate counter 60 When a "miss" occurs, i.e., the data word sought by the processor is not stored in the data buffer 34, and the main memory 14 must transfer a new block having the data word to the cache memory, a FETCH signal is generated by the fetch generate counter 60 to cause each word of the block to be fetched from the main memory.
- a RETURN signal returns to the fetch return counter 62 when the data word is returned to the cache memory.
- the fetch generate and return counters 60 and 62 can be conventional binary counters of fixed size, depending on the size of the blocks of data stored in the cache memory.
- the fetch generate counter 60 decre ⁇ ments with each FETCH signal and the fetch return counter decrements with each RETURN signal.
- counter 60 pro- vides a GENERATE COMPLETE signal and counter 62 provides a RETURN COMPLETE signal, indicating that the complete transfer of the data block has occurred.
- Each word of the new data block from main memory 14 is received at a data buffer data input 36 of the data buffer.34.
- the conventional pro ⁇ cessor hardware implementing the LRU algorithm deter ⁇ mines which data block in data buffer 34 and correspond ⁇ ing address block in the index 32 are replaced when a new data block is needed in the cache memory 26.
- the index 32 is a "set associative" memory. As with any associative memory, the index 32 compares data (a main memory address) received at its data input 38 with its stored data in order to indicate a match. In addition, index 32 receives an address or index address input bits at its address input 40 so that only stored data in a set defined by the address at input 40 is compared with the data received at the data input 38. This can be more fully understood by referring to Figs. 3A, 3B and 3C in conjunction with Fig. 2. Fig. 3A illustrates the contents of main memory 14, Fig.
- each main memory address has twenty-two bits. These include a first portion comprised of twelve of the most signifi ⁇ cant bits (MSB's) of the main memory address (A-.-.- pp ) received at the index data input 38, a second portion comprised of eight of the least significant bits (LSB's) of the main memory address (Ag-A-,g) received at the index address input 40 and defining the "set" of data to which the bits at the data input 38 are to be com ⁇ pared, and two remaining bits ( - ] _- 2) that are received at a word select input 42 of the cache memory, which will be described in greater detail later.
- MSB's most signifi ⁇ cant bits
- LSB's least significant bits
- index 32 and data buffer 34 are arranged in banks (horizontal rows), with each bank having 256 blocks of data or main memory addresses (numbered “0” through “255” in vertical columns). There are 4,096 banks (numbered “0” through “4,095") in the main memory 14, and sixteen banks (numbered “0” through “15") in each of the index 32 and data buffer 34.
- Each vertical column of blocks seen in the index 32 in Fig. 3B is one of the sets that is defined by the eight bits of the main memory address ( A 3-A-, Q ) received at the index ad ⁇ dress input 40 of the index 32 in Fig. 2.
- the data buffer 34 stores blocks of data from the main memory 14, and each data block in the data buffer has a corresponding address block in the index 32.
- Each address block in index 32 stores the twelve most significant bits ( A ⁇ A 22 ⁇ ° ⁇ t ⁇ ⁇ e main memory address of its corresponding data block.
- the eight least significant bits ( A 3 - -, 0 ) of the main memory address of each ' data block determine the vertical column of blocks in which the address block and its cor- responding data block will be located in the index 32 and data buffer 34.
- the new data block When a new block of data from one of the ver ⁇ tical column of blocks in the main memory is to be written into data buffer 34, the new data block is put in the same vertical column of blocks in the data buf ⁇ fer where it resided in the main memory 14, and at the bank having the least recently used data block. The most significant bits of the main memory address of the new data block are placed in the corresponding address block in the index 32.
- one of the vertical columns of address blocks in the index is selected by the bits at the index address input 40 (Fig. 2), and each group of most sig- nificant bits of the main memory addresses stored in each of the sixteen address blocks in that column or set of the index 32 is compared to the most significant bits of the main memory address received at the index data input 38. If there is no match, then ' a signal at a no match output 46 of the index is delivered to the proces ⁇ sor hardware implementing the LRU algorithm in order to bring into the cache memory the needed block of data. If there is a match, then an appropriate signal indi ⁇ cating the bank where the match occurred is generated by the index and is received at a buffer bank select input 48 of the data buffer 34.
- the least significant bits (- ⁇ -A ⁇ g) on the index address input 40 and the remaining two bits of the main memory address (h- ⁇ -K j ) on the word select input 42 are received at a block and word select input 49 of the data buffer 34 to select the appropriate vertical column of blocks in the data buffer and one of the words in the selected block.
- the resulting word appears at the buffer data output 50 and is provided to the processor data bus 28 for use by the processor.
- a cache memory 26' for storing variable size blocks of data.
- the cache memory 26' has an index 32' and a data buffer 34'.
- the index 32' is a set associative memory and, still assuming twenty-two bit main memory addresses, has an index address input 38' for receiving the most significant main memory address bits (A,,-A22. and an index data input 40' for receiving eight of the least significant main memory address bits ( 3 -A 10 ).
- the index 32' provides bits to a buffer bank select input 48' of the data buffer 34' for selecting one of the banks in the data buffer 34' when a match occurs, and has a no match output 46' for providing a signal to the hardware implementing the LRU algorithm when the cache memory 26' does not have the data needed by the processor.
- the least significant main memory address bits ( ⁇ -A- ⁇ g) and the two remaining bits (A ⁇ -A 2 ) on a word select input 42' are provided to a block and word select input 49' of the data buffer 34* in order to select a block and word, with the selected word appearing at a data buffer output 50'.
- a block size register 54 for storing information or bits representing the block size desired for the cache memory.
- the block size register 54 may be a conventional two bit binary register.
- the block size information in the block size register 54 comes from an input line 56 that may, in turn, originate either from manual switches set by an operator of the system, or that may be dynamically set by software or firmware.
- the block size information in the block size register 54 is provided to an address inhibit circuit 58, a fetch generate counter 60', and a fetch return counter 62'.
- the fetch generate counter 60' and fetch return counter 62' are provided with the block size information in order that the cache memory 26' fetches and receives the proper number of bytes or words from main memory when a new block of data is moved from the main memory to the cache memory. That is, the fetch generate counter 60' is provided with a binary value representing the block size, it decrements, and, as it decrements it issues a FETCH signal to the main memory to cause one word to be returned to the cache memory. When the counter 60' decrements to zero, it issues a GENERATE COMPLETE signal.
- the " fetch return counter 62* is also provided with a binary value representing the block size and it decrements as each byte or word of the new block is returned from the main memory with a RETURN signal. When it decrements to a value of zero, it issues a RETURN COMPLETE signal, indicating the transfer of the new block of data to the cache memory has been completed.
- the words of the new data block are received at a data buffer data input 36'.
- the new block size information is provided into both the fetch generate counter 60' and the fetch return counter 62', and blocks of data having the new size are fetched and returned from the main memory.
- the fetch generate counter 60' and fetch return counter 62' may each be implemented by a conventional binary register/ counter.
- counters 60' and 62' are each 4 bit counters.
- the address inhibit circuit 58 affects the number of bits received at the address input 40' of the index 32'. As will be more fully described later, as the size of the data blocks in the data buffer 34 » increases, fewer data blocks are stored in the data buffer 34', and less memory space (and fewer address blocks) is required in the index 32' for storing main memory address bits. With less memory space required in index 32', fewer of the bits ( 2-A-,g) are required at the address input 40' to address the index.
- Figs. 5A and 5B show the contents of the index 32' and the data buffer 34', respectively.
- the block sizes of data stored in the data buffer 34' may have any one of three sizes: 16 words per block, 8 words per block, or 4 words per block.
- fewer blocks in the index 32' are required to store main memory address bits. Accordingly, as il ⁇ lustrated in Fig. 5A, for sixteen-word blocks in the data buffer 34', only one-fourth of the index 32' is needed and, accordingly, only BLOCKS 0-63 are loaded with the main memory address bits.
- each block doubles when the block size changes from four words to eight words, and again doubles when the block size changes from eight words sixteen words.
- the sizes of the blocks stored in data buffer 34' do change, the structure and operation of the data buffer 34' is no different from the single block size data buffer 34 of Fig. 2. Still referring to Figs. 5A and 5B, and also referring to Fig.
- main memory address bits A 11 ⁇ A 22 are storec 3 i n each address block in the index 32', regardless of the size of the data blocks stored in data buffer 34'. However, as the data block size in ⁇ creases, fewer bits are needed at the index address input 40* (Fig. 4), since there are fewer blocks in the index 32'. Since there are fewer data blocks in data buffer 34', but a larger number of words per block, the same number of bits (A, -A, n ) are needed, regardless of block size, at the data buffer block and word select input 49' to select the desired word in each block. As seen in Fig.
- Fig. 7 illustrates the circuitry within the address inhibit circuit 58 of Fig. 4.
- the two lowest order bits (A., and A 4 ) of the bits pro ⁇ vided to the index address input 40' are passed through OR gates 62 and 64.
- the remaining bits 5 -A 1Q are passed directly through the address inhibit circuit 58 to the index address input 40".
- the block size register 54 supplies two bits to the address inhibit circuit 58.
- both OR gates 62 and 64 receive a "1" from block size register 54, both A 3 and A 4 are held at a "1" level, and only one-fourth of the index 32' (BLOCKS 0 through 63 in Fig. 5) is addressed.
- the two bits from block size register 54 are provided to the two highest order bit locations in the fetch generate counter 60' and fetch return counter 62', in order to provide for the transfer of blocks having 4, 8 or 16 word block sizes.
- the block size of the data in the cache memory can be varied.
- the "hit ratio" of the cache memory can be increased, resulting in increased speed of the processor.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Memory System Of A Hierarchy Structure (AREA)
Abstract
Increasing the speed and hit ratio for data fetches in data processing systems is essential. Increased data fetch speed normally results from using a fast cache memory with a slower main memory. The hit ratio for such fetches can be increased by using cache memory (26') which incorporates data buffer (34') that stores blocks of data that are varied in size and a set associative memory as index (32') which stores block addresses for main memory (14), which addresses are associated with the data blocks stored in buffer (34'). The block sizes are varied by selectively inhibiting address bits provided to an input (40') of the index (32') by address inhibit circuit (54) in response to information stored in block size register (54). Such block size information is also provided to a fetch generate counter (60') and a fetch return counter (62') for controlling the number of words transferred as a block from main memory (14) to cache memory (26').
Description
CACHEMEMORYINWHICHTHEDATABLOCKSIZEISVARIABLE
Technical Field
The present invention relates to data proces¬ sing systems and, more particularly, to cache memories used in data processing systems.
Background Art
Cache memories are frequently used in data processing systems in order to increase the speed of such systems. A cache memory is essentially a data buffer or fast memory that quickly provides data in response to memory address signals. The cache memory generally resides within a processor and acts as an intermediate storage between the main memory of the system and the processing elements of the processor. A processor often has its processing speed affected by the time that it takes to fetch data or instructions from the main memory. With a fast cache memory within the processor storing the most recently used data from main memory, the processor will, in most circumstances, be able to quickly get its needed data and instructions from the cache memory. In addition, since the cache memory resides within the processor, the data and instructions in the cache memory can be provided much more quickly than they could if they needed to be fetched along a bus from main memory.
There are a number of conditions that affect the likelihood of data being in the cache memory when needed by the processor. Data blocks that include words that have been recently used by the processor are much more likely to be needed by the processor than other blocks of data in the main memory. Accordingly, a least recently used (LRU) algorithm is often implemented in the processor for controlling the transfer of the blocks of data between the cache memory and the main memory. The LRU algorithm assures that only the most frequently used blocks of data are stored in the cache memory, and,
if a block of data is needed that is not in the cache memory, the needed block replaces a block in the cache memory that is the least recently used. Even with the use of LRU algorithms, however, there are still some instances during the operation of the processor when the processor must wait for data from the main memory because of the need for a block of data not stored in the cache memory. This condition or circumstance is referred to as a "miss". During any given period of time, the number of "misses" in relation to the total number of cache memory accesses is referred to as the "hit ratio".
A full discussion of cache memories, in par¬ ticular the factors affecting the "hit ratio" (fault ratio) is set forth in Performance Analysis of Cache
Memories by Guraraj S. Rao, 25 Journal of the Association for Computing Machinery 378 (July 1978).
It has been experienced that the type of task or program being performed by a data processing system affects the hit ratios of the cache memory. For example, if the task involves the use of small clusters of data that are widely spread throughout the main memory, the processor may frequently have to access the main memory as each small cluster of data is needed. A problem experienced with prior art cache memories is that of obtaining a consistently high "hit ratio" for varying types of tasks or programs being performed by the data processing system.
Disclosure of the Invention It is an object of the invention to provide a cache memory in a data processing system wherein the dependence of the "hit ratio" on the type of task being performed is reduced, so that a consistently high "hit ratio" can be obtained. The present invention provides a data proces¬ sing system having a main memory for storing data, a processor for processing data from said main memory, and
c~
a cache memory within said processor, characterized in that said cache memory stores data blocks having a variable number of data words and comprises a data buffer for receiving and storing data blocks from said main memory, said data buffer arranged in banks with each bank having a plurality of data blocks; and index for storing an address block corresponding to each data block stored in said data buffer and comprised of bits in the main memory address of the corresponding data block, said index including an index data input for receiving main memory address bits for comparison with the main memory address bits stored in said address blocks, and an index address input for receiving index address input bits selecting a data block in each of the banks for comparison with the main memory address bits at said index data input; and inhibit means for inhibit¬ ing selected ones of the index address input bits at said index address input in order to reduce the number of addressable address blocks in said index so that the number of data words in the corresponding data blocks in said data buffer may be increased.
In accordance with the invention, the number of words in a data block of the cache memory may be selected in advance of a task to be performed in order to optimize the hit ratio; for example, where the task involves the use of small clusters of data that are widely spread throughout the main memory the data blocks will be preselected to be small.
Brief Description of the Drawings One embodiment of the invention will now be described, by way of example, with reference to the accompanying drawings, in which:
Fig. 1 is a simplified block diagram of a data processing system, including a processor having a cache memory.
Fig. 2 is a block diagram illustrating the index and data buffer of a cache memory, such as the
-4-
cache memory illustrated in the data processing system of Fig. 1.
Figs. 3A, 3B and 3C illustrate the contents of a main memory, and the contents of an index and data buffer of a cache memory for storing blocks of data from the main memory.
Fig. 4 is a block diagram of a cache memory, including an index and data buffer, having data blocks of variable size in accordance with the present inven- tion.
Figs. 5A and 5B illustrate the contents of the index and data buffer of Fig. 4.
Fig. 6 illustrates the bits of a main memory address and their use in varying the block size of the cache memory shown in Fig. 4.
Fig. 7 is a detailed block diagram of the address inhibit circuit shown in Fig. 4.
Best Mode for Carrying Out the Invention
Turning first to Fig. 1, there is shown a data processing system 10, including a processor 12, a main memory 14, and an internal transfer bus 16 for connecting the processor 12 to the main memory 14. The bus 16 also connects other subsystems 18 to each of the main memory 14 and processor 12. As is conventional, the processor 12 can include an arithmetic-logic unit (ALU) 22, an instruction storage unit (ISU) 24, a cache memory 26, as well as other well-known and conventional components (not shown), all connected to each other and to the in¬ ternal transfer bus 16 by a processor data bus 28. The operation of a data processing system having a cache memory, such as data processing system 10 of Fig. 1, is well-known in the art. The main memory 14 normally holds data and instructions that are needed by the processor 12, such data and instructions initially loaded into the main memory by card readers, disk units, or other peripheral devices that might be found in the other subsystems 18. The data and instructions stored
in main memory 14 are provided to the processor 12 over internal transfer bus 16. In response to each instruc¬ tion from main memory 14, the ISU 24 will normally provide a microprogram for execution and, in accordance with microinstructions in that microprogram, data will be fetched from main memory 14 to be operated on by the ALU 22.
The data in the main memory 14 is organized in blocks, each block normally comprising a number of data words. When a specific word is needed from the main memory 14, the block in which it resides is stored in the cache memory 26. Since there is a likelihood that, when the processor accesses one word in a block and then subsequently needs a different word, the different word will be adjacent the first in the same block, the proces¬ sor will in many cases only need to access the cache memory 26. In addition, the blocks of data held in the cache memory are stored in accordance with a least recently used (LRU) algorithm so that when the processor needs a new data word not in the cache memory 26, and a new block of data fetched from main memory 14 is to be stored in the cache memory 26, the block least recently used by the processor in the cache memory is removed and the new block of data is stored in its place. Processor hardware for implementing an LRU algorithm, including the transfer of data blocks between the main memory and cache memory, is well-known in the art. Commercially available systems having cache memories and using LRU algorithms include, among others, the Class 370/Model 168 system available from IBM Corporation, Armonk, New York.
Because the data needed by the ALU 22 will, in most cases, be stored in the cache memory 26, the data can be provided much more quickly to the ALU than if it were coming directly from main memory 14. As illustrated in Fig. 2, the cache memory 26 is generally comprised of two memory components, an index 32 and a data buffer 34. The data buffer 34 stores the blocks of data from main memory 14, and the
index 32 stores the main memory addresses associated with the data blocks in the data buffer 34. When the processor checks for a given data word in the cache memory 26, it checks the addresses stored in the index 32 and, if there is a match, the selected data word is read from the data buffer 34 to the bus 28 for use by the processor. •
As conventional, the cache memory 26 further includes a fetch generate counter 60 and a fetch return counter 62. When a "miss" occurs, i.e., the data word sought by the processor is not stored in the data buffer 34, and the main memory 14 must transfer a new block having the data word to the cache memory, a FETCH signal is generated by the fetch generate counter 60 to cause each word of the block to be fetched from the main memory. A RETURN signal returns to the fetch return counter 62 when the data word is returned to the cache memory. The fetch generate and return counters 60 and 62 can be conventional binary counters of fixed size, depending on the size of the blocks of data stored in the cache memory. The fetch generate counter 60 decre¬ ments with each FETCH signal and the fetch return counter decrements with each RETURN signal. When the counters 60 and 62 decrement to zero, counter 60 pro- vides a GENERATE COMPLETE signal and counter 62 provides a RETURN COMPLETE signal, indicating that the complete transfer of the data block has occurred. Each word of the new data block from main memory 14 is received at a data buffer data input 36 of the data buffer.34. As mentioned earlier, the conventional pro¬ cessor hardware implementing the LRU algorithm deter¬ mines which data block in data buffer 34 and correspond¬ ing address block in the index 32 are replaced when a new data block is needed in the cache memory 26. Such hardware generates the necessary control signals (not shown) to write the new block of data and corresponding address block in the cache memory.
In the cache memory 26 illustrated in Fig. 2, the index 32 is a "set associative" memory. As with any associative memory, the index 32 compares data (a main memory address) received at its data input 38 with its stored data in order to indicate a match. In addition, index 32 receives an address or index address input bits at its address input 40 so that only stored data in a set defined by the address at input 40 is compared with the data received at the data input 38. This can be more fully understood by referring to Figs. 3A, 3B and 3C in conjunction with Fig. 2. Fig. 3A illustrates the contents of main memory 14, Fig. 3B illustrates the contents of index 32, and Fig. 3C illus¬ trates the contents of data buffer 34. It is assumed for purposes of the present discussion that each main memory address has twenty-two bits. These include a first portion comprised of twelve of the most signifi¬ cant bits (MSB's) of the main memory address (A-.-.- pp) received at the index data input 38, a second portion comprised of eight of the least significant bits (LSB's) of the main memory address (Ag-A-,g) received at the index address input 40 and defining the "set" of data to which the bits at the data input 38 are to be com¬ pared, and two remaining bits ( -]_- 2) that are received at a word select input 42 of the cache memory, which will be described in greater detail later.
As seen in Figs. 3A through 3C7 each of the main memory 14, index 32 and data buffer 34 are arranged in banks (horizontal rows), with each bank having 256 blocks of data or main memory addresses (numbered "0" through "255" in vertical columns). There are 4,096 banks (numbered "0" through "4,095") in the main memory 14, and sixteen banks (numbered "0" through "15") in each of the index 32 and data buffer 34. Each vertical column of blocks seen in the index 32 in Fig. 3B is one of the sets that is defined by the eight bits of the main memory address (A3-A-,Q) received at the index ad¬ dress input 40 of the index 32 in Fig. 2.
As mentioned above, the data buffer 34 stores blocks of data from the main memory 14, and each data block in the data buffer has a corresponding address block in the index 32. Each address block in index 32 stores the twelve most significant bits (Aιι~A22^ °^ t^ιe main memory address of its corresponding data block. The eight least significant bits (A 3- -,0) of the main memory address of each' data block determine the vertical column of blocks in which the address block and its cor- responding data block will be located in the index 32 and data buffer 34.
When a new block of data from one of the ver¬ tical column of blocks in the main memory is to be written into data buffer 34, the new data block is put in the same vertical column of blocks in the data buf¬ fer where it resided in the main memory 14, and at the bank having the least recently used data block. The most significant bits of the main memory address of the new data block are placed in the corresponding address block in the index 32.
When the processor checks for a data block in the cache memory, one of the vertical columns of address blocks in the index is selected by the bits at the index address input 40 (Fig. 2), and each group of most sig- nificant bits of the main memory addresses stored in each of the sixteen address blocks in that column or set of the index 32 is compared to the most significant bits of the main memory address received at the index data input 38. If there is no match, then' a signal at a no match output 46 of the index is delivered to the proces¬ sor hardware implementing the LRU algorithm in order to bring into the cache memory the needed block of data. If there is a match, then an appropriate signal indi¬ cating the bank where the match occurred is generated by the index and is received at a buffer bank select input 48 of the data buffer 34. At the same time, the least significant bits (-^-A^g) on the index address input 40
and the remaining two bits of the main memory address (h-^-Kj ) on the word select input 42 are received at a block and word select input 49 of the data buffer 34 to select the appropriate vertical column of blocks in the data buffer and one of the words in the selected block. The resulting word appears at the buffer data output 50 and is provided to the processor data bus 28 for use by the processor.
The above is, of course, only a brief summary of the operation of a cache memory having an index which is set associative, such as the cache memory 26. This type of cache memory is well-known in the art, and for a more detailed discussion reference can be had to Per¬ formance Analysis of Cache Memories, by Gururaj S. Rao, 25 Journal of the Association for Computing Machinery 378 (July, 1978).
In accordance with the present invention, there is provided in Fig. 4 a cache memory 26' for storing variable size blocks of data. Like the cache memory 26 of Fig. 2, the cache memory 26' has an index 32' and a data buffer 34'. The index 32' is a set associative memory and, still assuming twenty-two bit main memory addresses, has an index address input 38' for receiving the most significant main memory address bits (A,,-A22. and an index data input 40' for receiving eight of the least significant main memory address bits ( 3-A10). The index 32' provides bits to a buffer bank select input 48' of the data buffer 34' for selecting one of the banks in the data buffer 34' when a match occurs, and has a no match output 46' for providing a signal to the hardware implementing the LRU algorithm when the cache memory 26' does not have the data needed by the processor. The least significant main memory address bits ( β-A-^g) and the two remaining bits (Aτ-A2) on a word select input 42' are provided to a block and word select input 49' of the data buffer 34* in order to select a block and word, with the selected word appearing at a data buffer output 50'.
In order to control the size of the blocks of data stored in the cache memory 26' of Fig. 4, there is provided a block size register 54 for storing information or bits representing the block size desired for the cache memory. In the illustrated embodiment, the block size register 54 may be a conventional two bit binary register. The block size information in the block size register 54 comes from an input line 56 that may, in turn, originate either from manual switches set by an operator of the system, or that may be dynamically set by software or firmware.
The only drawback to software or firmware dynamically setting the block size is that each time there is a change in the block size, the cache memory must have all its previous blocks destroyed in prepara¬ tion for the new size blocks. Accordingly, frequent changes in block size will increase the inefficiency of the system. As a result, one must be careful in changing the block size so that the desired increase in efficiency from changing the block size will not be offset by any decrease in efficiency caused by changing the block size at frequent intervals. It is anticipated that in the practice of this invention an operator of a data proces¬ sing system will change the block size only a few times each day, at the most, depending upon the tasks that are being performed. It may be trial and error proce¬ dure, with the operator changing the block sizes for each task and noting any increase in operating speed. When the same task or a similar task is run in the future, the operator will set the block size register 54 to the most favorable block size information previously noted.
The block size information in the block size register 54 is provided to an address inhibit circuit 58, a fetch generate counter 60', and a fetch return counter 62'. The fetch generate counter 60' and fetch return counter 62' are provided with the block size information in order that the cache memory 26' fetches
and receives the proper number of bytes or words from main memory when a new block of data is moved from the main memory to the cache memory. That is, the fetch generate counter 60' is provided with a binary value representing the block size, it decrements, and, as it decrements it issues a FETCH signal to the main memory to cause one word to be returned to the cache memory. When the counter 60' decrements to zero, it issues a GENERATE COMPLETE signal. The" fetch return counter 62* is also provided with a binary value representing the block size and it decrements as each byte or word of the new block is returned from the main memory with a RETURN signal. When it decrements to a value of zero, it issues a RETURN COMPLETE signal, indicating the transfer of the new block of data to the cache memory has been completed. The words of the new data block are received at a data buffer data input 36'. When the block size information changes, the new block size information is provided into both the fetch generate counter 60' and the fetch return counter 62', and blocks of data having the new size are fetched and returned from the main memory. The fetch generate counter 60' and fetch return counter 62' may each be implemented by a conventional binary register/ counter. If the largest block size is sixteen words per block, then counters 60' and 62' are each 4 bit counters. The address inhibit circuit 58 affects the number of bits received at the address input 40' of the index 32'. As will be more fully described later, as the size of the data blocks in the data buffer 34» increases, fewer data blocks are stored in the data buffer 34', and less memory space (and fewer address blocks) is required in the index 32' for storing main memory address bits. With less memory space required in index 32', fewer of the bits ( 2-A-,g) are required at the address input 40' to address the index.
This is illustrated more clearly by Figs. 5A and 5B, which show the contents of the index 32' and the data buffer 34', respectively. For purposes of the
present description, it is assumed that the block sizes of data stored in the data buffer 34' may have any one of three sizes: 16 words per block, 8 words per block, or 4 words per block. As mentioned earlier, as the num- ber of words or bytes per data block in the data buffer 34' increase, fewer blocks in the index 32' are required to store main memory address bits. Accordingly, as il¬ lustrated in Fig. 5A, for sixteen-word blocks in the data buffer 34', only one-fourth of the index 32' is needed and, accordingly, only BLOCKS 0-63 are loaded with the main memory address bits. For an eight-word block size, one-half of the index 32' is needed and, accordingly, only BLOCKS 0-127 are loaded with main memory address bits. For a four-word block size, the entire index 32' is needed, and BLOCKS 0-255 (all 256 blocks) of the index 32' are loaded with main memory address bits. It should be apparent, of course, that in actual practice any size block can be chosen in accord¬ ance with the present invention. If block sizes smaller than four words per block are needed, then a larger index than the index 32' is required. The theoretical minimum size for the data blocks is, of course, one byte or word per block. If larger data block sizes are needed, then the index 32' can have a smaller portion used for addressing the blocks than that illustrated in Fig. 5A. The theoretical maximum size for the data blocks is, of course, a single block which would occupy the entire data buffer 34*.
Although the amount of memory space used in the index 32' decreases as the data block size increases, the data buffer 34' uses the same amount of memory space regardless of the block size. This is due, of course, to the fact that the number of words in each block increases as the number of blocks decreases. As is illustrated at the bottom of Fig. 5B for BLOCK 0 of data buffer 34', each block doubles when the block size changes from four words to eight words, and again doubles when the block size changes from eight words
sixteen words. Although the sizes of the blocks stored in data buffer 34' do change, the structure and operation of the data buffer 34' is no different from the single block size data buffer 34 of Fig. 2. Still referring to Figs. 5A and 5B, and also referring to Fig. 6, which illustrates the 22 bits of each main memory address, main memory address bits A11~A22 are storec3 in each address block in the index 32', regardless of the size of the data blocks stored in data buffer 34'. However, as the data block size in¬ creases, fewer bits are needed at the index address input 40* (Fig. 4), since there are fewer blocks in the index 32'. Since there are fewer data blocks in data buffer 34', but a larger number of words per block, the same number of bits (A, -A,n) are needed, regardless of block size, at the data buffer block and word select input 49' to select the desired word in each block. As seen in Fig. 6, when each data block has four words, address bits 3-A-,Q are received at the index address input 40'. When each data block has eight words, ad¬ dress bits .- -jg are received at the index address input 40'. Finally, when each data block has sixteen words, only address bits 5-A-,Q are received at the index address input 40'. For all three word block si- zes, however, address bits A,-A-,g are received at the data buffer block and word select input 49', since one word in the data buffer is selected to appear at the data buffer output 50', regardless of the block size and the resulting number of blocks in the data buffer 34'. Fig. 7 illustrates the circuitry within the address inhibit circuit 58 of Fig. 4. As can be seen, the two lowest order bits (A., and A4 ) of the bits pro¬ vided to the index address input 40' are passed through OR gates 62 and 64. The remaining bits 5-A1Q are passed directly through the address inhibit circuit 58 to the index address input 40". The block size register 54 supplies two bits to the address inhibit circuit 58.
c::?ι
P - ' . ^
If both bits supplied by the block size register 54 to the address inhibit circuit 58 are at "0", a four word block size is indicated, all of the bits 3-A-,g are passed through to the index address input 40', and the entire index 32' (BLOCKS 0 through 255 in Fig. 5) is addressed. However, if an eight word block size is in¬ dicated, OR gate 64 receives a "1" from block size register 54. A-, is held at a "1" logic level, so that only one-half of index 32' (BLOCKS 0 through 127 in Fig. 5) is addressed. If a sixteen word block size is indi¬ cated, both OR gates 62 and 64 receive a "1" from block size register 54, both A3 and A4 are held at a "1" level, and only one-fourth of the index 32' (BLOCKS 0 through 63 in Fig. 5) is addressed. As should be apparent, the two bits from block size register 54 are provided to the two highest order bit locations in the fetch generate counter 60' and fetch return counter 62', in order to provide for the transfer of blocks having 4, 8 or 16 word block sizes. It can be seen from the foregoing that by providing in a processor a conventional cache memory having a set associative index, a data buffer, and fetch generate and fetch return counters, and further providing in accordance with the present invention means for in- hibiting certain address bits to the index, as well as changing the block size information supplied to the fetch generate and fetch return counters, the block size of the data in the cache memory can be varied. By varying the block size, the "hit ratio" of the cache memory can be increased, resulting in increased speed of the processor.
Although the presently preferred embodiment of this invention has been described, it will be understood that within the purview of this invention various changes may be made within the scope of the appended claims.
Claims
1. A data processing system (10) having a main memory (14) for storing data, a processor (12) for processing data from said main memory (14), and a cache memory (26') within said processor (12), characterized in that said cache memory (26' ) stores data blocks having a variable number of data words and comprises a data buffer (34') for receiving and storing data blocks from said main memory (14), said data buffer (34') arranged in banks with each bank having a plurality of data blocks; an index (32') for storing an address block cor¬ responding to each data block stored in said data buffer (34') and comprised of bits in the main memory address of the corresponding data block, said index (32') including an index data input (38') for receiving main memory address bits for comparison with the main memory address bits stored in said address blocks, and' an index address input (40") for receiving index address input bits selecting a data block in each of the banks for comparison with the main memory address bits at said index data input (381); and inhibit means (58) for inhibiting selected ones of the index address input bits at said index address input (40') in order to reduce the number of addressable address blocks in said index- (32') so that the number of data words in the corresponding data blocks in said data buffer (34* ) may be increased.
2. A data processing system according to claim 1, characterized in that said cache memory (26') further comprises a block size register (54) for storing block size bits representing the size of the data blocks stored in said data buffer (34'), and wherein said in¬ hibit means (58) includes logic gate means (62, 64) for receiving the index address input bits and the block size bits in order to hold selected ones of the index address input bits at a predetermined logic level during 2. (concluded) the receipt of the index address input bits at said index address input (40').
3. A data processing system according to claim 2, characterized in that said cache memory further comprises a fetch generate counter (60') for receiving the block size bits from said block size register (54) and generating a FETCH signal in order to fetch each data word in each data block stored in said data buffer (34') from said main memory (14); and a fetch return counter (62') for receiving the block size bits from said block size register (54) and receiving a RETURN signal from said main memory (14) with each data word in each data block from said main memory (14); said fetch generate counter (60') decremented by each FETCH signal and said fetch return counter (62') decremented by each RETURN signal so that when said data buffer (34') receives and. stores a complete data block from said main memory (14), said fetch generate counter (60') and said fetch return counter (62') have been decremented to a zero value.
4. A data processing system (10) comprising a main memory (14); and a processor (12) for processing data stored in said main memory (14), including a cache memory (26'), said cache memory (261) characterized by: a data buffer (34') for storing data blocks from said main memory (14); a set associative index (32') for storing sets of address blocks, each address block having an associated one of the data blocks in said data buffer (34') and comprised of bits from the main memory address of its associated data block, said index (32') including an index data input (38') for receiving main memory address bits for comparison with the main memory address bits in the address blocks and an index address input (40') for receiving index address input bits addressing a set of address blocks to be compared with the main 4 . ( concluded ) memory address bits at said index data input (38'); and means for varying the size of the data blocks in said data buffer (34'), including means (58) for inhibiting selected ones of the index address input bits in order to reduce the number of sets of address blocks stored in said index (32').
5. A data processing system according to claim 4, characterized by a block size register (54) for storing bits representing the size of the data blocks stored in said data buffer (34'), and wherein said means for inhibiting (58) is characterized by logic gate means (62, 64) for logically combining the index address input bits and the bits from said block size register (54) in order to hold the selected ones of the index address input bits (40') at a predetermined binary value in order to reduce the addressable size of said index (32').
6. A data processing system according to claim 4, characterized by a block size register (54) for storing bits representing the size of the data blocks stored in said data buffer (34'); a fetch generate count- er (60') connected for receiving the bits in said block size register (54) and generating a FETCH signal in order to fetch each data word of each data block stored in said data buffer (34') from said main memory (14); and a fetch return counter (621) connected for receiving the bits in said block size register (54) and receiving a
RETURN signal upon the storing of each data word in said data buffer (34') from said main memory (14); said fetch generate counter (60') decremented upon each FETCH sig¬ nal and said fetch return counter (62*) decremented upon each RETURN signal so that said fetch generate counter
(60') and said fetch return counter (62') are both decre¬ mented to a zero value when a complete data block is stored in said data buffer (34') from said main memory (14).
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
JP81500394A JPS56501548A (en) | 1979-12-19 | 1980-12-10 | |
DE8181900188T DE3072078D1 (en) | 1979-12-19 | 1980-12-10 | Cache memory in which the data block size is variable |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US105186 | 1979-12-19 | ||
US06/105,186 US4315312A (en) | 1979-12-19 | 1979-12-19 | Cache memory having a variable data block size |
Publications (1)
Publication Number | Publication Date |
---|---|
WO1981001894A1 true WO1981001894A1 (en) | 1981-07-09 |
Family
ID=22304521
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/US1980/001666 WO1981001894A1 (en) | 1979-12-19 | 1980-12-10 | Cache memory in which the data block size is variable |
Country Status (5)
Country | Link |
---|---|
US (1) | US4315312A (en) |
EP (1) | EP0042000B1 (en) |
JP (1) | JPS56501548A (en) |
DE (1) | DE3072078D1 (en) |
WO (1) | WO1981001894A1 (en) |
Cited By (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP0285172A2 (en) * | 1987-03-31 | 1988-10-05 | Nec Corporation | Cache controller with a variable mapping mode |
EP0325419A2 (en) * | 1988-01-20 | 1989-07-26 | Advanced Micro Devices, Inc. | Method and apparatus for caching interlock variables in an integrated cache memory |
EP0325420A2 (en) * | 1988-01-20 | 1989-07-26 | Advanced Micro Devices, Inc. | Organization of an integrated cache unit for flexible usage in cache system design |
EP0325422A2 (en) * | 1988-01-20 | 1989-07-26 | Advanced Micro Devices, Inc. | Integrated cache unit |
EP0325421A2 (en) * | 1988-01-20 | 1989-07-26 | Advanced Micro Devices, Inc. | Organization of an integrated cache unit for flexible usage in supporting multiprocessor operations |
EP0342631A2 (en) * | 1988-05-18 | 1989-11-23 | Nec Corporation | Buffer memory device capable of memorizing operand and instruction data blocks at different block sizes |
EP0448205A2 (en) * | 1990-03-23 | 1991-09-25 | Advanced Micro Devices, Inc. | Re-configurable block length cache |
FR2733066A1 (en) * | 1995-04-14 | 1996-10-18 | Nec Corp | Controller for variable size cache memory |
WO2014155073A1 (en) * | 2013-03-25 | 2014-10-02 | Lancaster University Business Enterprises Limited | Cache |
EP2911062A4 (en) * | 2012-10-18 | 2015-10-14 | Zte Corp | Method and device for adjusting cache block length of cache memory |
GB2538055A (en) * | 2015-04-28 | 2016-11-09 | Advanced Risc Mach Ltd | Data processing apparatus having a cache |
US10250709B2 (en) | 2015-04-28 | 2019-04-02 | Arm Limited | Data processing apparatus, controller, cache and method |
Families Citing this family (75)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
JPS5687282A (en) * | 1979-12-14 | 1981-07-15 | Nec Corp | Data processor |
US4489378A (en) * | 1981-06-05 | 1984-12-18 | International Business Machines Corporation | Automatic adjustment of the quantity of prefetch data in a disk cache operation |
US4490782A (en) * | 1981-06-05 | 1984-12-25 | International Business Machines Corporation | I/O Storage controller cache system with prefetch determined by requested record's position within data block |
US4503501A (en) * | 1981-11-27 | 1985-03-05 | Storage Technology Corporation | Adaptive domain partitioning of cache memory space |
JPS58147879A (en) * | 1982-02-26 | 1983-09-02 | Toshiba Corp | Control system of cache memory |
US4510581A (en) * | 1983-02-14 | 1985-04-09 | Prime Computer, Inc. | High speed buffer allocation apparatus |
US4897783A (en) * | 1983-03-14 | 1990-01-30 | Nay Daniel L | Computer memory system |
US4620274A (en) * | 1983-04-01 | 1986-10-28 | Honeywell Information Systems Inc. | Data available indicator for an exhausted operand string |
JPH0644245B2 (en) * | 1983-12-29 | 1994-06-08 | 富士通株式会社 | Store buffer device |
US4736293A (en) * | 1984-04-11 | 1988-04-05 | American Telephone And Telegraph Company, At&T Bell Laboratories | Interleaved set-associative memory |
US4646233A (en) * | 1984-06-20 | 1987-02-24 | Weatherford James R | Physical cache unit for computer |
US4942518A (en) * | 1984-06-20 | 1990-07-17 | Convex Computer Corporation | Cache store bypass for computer |
US4899275A (en) * | 1985-02-22 | 1990-02-06 | Intergraph Corporation | Cache-MMU system |
US4933837A (en) * | 1986-12-01 | 1990-06-12 | Advanced Micro Devices, Inc. | Methods and apparatus for optimizing instruction processing in computer systems employing a combination of instruction cache and high speed consecutive transfer memories |
US5218685A (en) * | 1987-01-02 | 1993-06-08 | General Electric Company | System for write once read many optical storage devices to appear rewritable |
US4914573A (en) * | 1987-10-05 | 1990-04-03 | Motorola, Inc. | Bus master which selectively attempts to fill complete entries in a cache line |
US4947319A (en) * | 1988-09-15 | 1990-08-07 | International Business Machines Corporation | Arbitral dynamic cache using processor storage |
US4994962A (en) * | 1988-10-28 | 1991-02-19 | Apollo Computer Inc. | Variable length cache fill |
US5214777A (en) * | 1989-03-27 | 1993-05-25 | Ncr Corporation | High speed read/modify/write memory system and method |
US5394531A (en) * | 1989-04-03 | 1995-02-28 | International Business Machines Corporation | Dynamic storage allocation system for a prioritized cache |
GB9008145D0 (en) * | 1989-05-31 | 1990-06-06 | Ibm | Microcomputer system employing address offset mechanism to increase the supported cache memory capacity |
US5226141A (en) * | 1989-07-14 | 1993-07-06 | Touch Technologies, Inc. | Variable capacity cache memory |
AU6075790A (en) * | 1989-07-14 | 1991-02-22 | Touch Technologies, Inc. | Variable capacity cache memory |
EP0408810B1 (en) * | 1989-07-20 | 1996-03-20 | Kabushiki Kaisha Toshiba | Multi processor computer system |
JP2504206B2 (en) * | 1989-07-27 | 1996-06-05 | 三菱電機株式会社 | Bus controller |
US5319769A (en) * | 1989-09-11 | 1994-06-07 | Sharp Kabushiki Kaisha | Memory access circuit for handling data pockets including data having misaligned addresses and different widths |
US5073851A (en) * | 1990-02-21 | 1991-12-17 | Apple Computer, Inc. | Apparatus and method for improved caching in a computer system |
US5287512A (en) * | 1990-08-06 | 1994-02-15 | Ncr Corporation | Computer memory system and method for cleaning data elements |
US5247653A (en) * | 1990-08-17 | 1993-09-21 | Seagate Technology, Inc. | Adaptive segment control and method for simulating a multi-segment cache |
US5537624A (en) * | 1991-02-12 | 1996-07-16 | The United States Of America As Represented By The Secretary Of The Navy | Data repacking circuit having toggle buffer for transferring digital data from P1Q1 bus width to P2Q2 bus width |
US5530835A (en) * | 1991-09-18 | 1996-06-25 | Ncr Corporation | Computer memory data merging technique for computers with write-back caches |
US5586303A (en) * | 1992-02-12 | 1996-12-17 | Integrated Device Technology, Inc. | Structure and method for providing a cache memory of selectable sizes |
FR2688612A1 (en) * | 1992-03-13 | 1993-09-17 | Inst Nat Rech Inf Automat | ANTEMEMORY DEVICE. |
JPH0683712A (en) * | 1992-09-03 | 1994-03-25 | Nec Ic Microcomput Syst Ltd | Microprocessor |
GB2271202B (en) * | 1992-10-01 | 1995-12-13 | Digital Equipment Int | Dynamic non-coherent cache memory resizing mechanism |
JP3516963B2 (en) * | 1993-03-12 | 2004-04-05 | 株式会社東芝 | Memory access control device |
US5664106A (en) * | 1993-06-04 | 1997-09-02 | Digital Equipment Corporation | Phase-space surface representation of server computer performance in a computer network |
JPH07175698A (en) * | 1993-12-17 | 1995-07-14 | Fujitsu Ltd | File system |
GB2286910B (en) * | 1994-02-24 | 1998-11-25 | Intel Corp | Apparatus and method for prefetching data to load buffers in a bridge between two buses in a computer |
US5721874A (en) * | 1995-06-16 | 1998-02-24 | International Business Machines Corporation | Configurable cache with variable, dynamically addressable line sizes |
JPH0950401A (en) * | 1995-08-09 | 1997-02-18 | Toshiba Corp | Cache memory and information processor provided with the same |
US5857214A (en) * | 1995-12-18 | 1999-01-05 | Advanced Micro Devices, Inc. | Microprocessor with a fixed cache size selected from a predesigned set of sizes |
US5978893A (en) * | 1996-06-19 | 1999-11-02 | Apple Computer, Inc. | Method and system for memory management |
US5983322A (en) * | 1997-04-14 | 1999-11-09 | International Business Machines Corporation | Hardware-managed programmable congruence class caching mechanism |
US6026470A (en) * | 1997-04-14 | 2000-02-15 | International Business Machines Corporation | Software-managed programmable associativity caching mechanism monitoring cache misses to selectively implement multiple associativity levels |
US6000014A (en) * | 1997-04-14 | 1999-12-07 | International Business Machines Corporation | Software-managed programmable congruence class caching mechanism |
US5916309A (en) * | 1997-05-12 | 1999-06-29 | Lexmark International Inc. | System for dynamically determining the size and number of communication buffers based on communication parameters at the beginning of the reception of message |
US6046817A (en) * | 1997-05-12 | 2000-04-04 | Lexmark International, Inc. | Method and apparatus for dynamic buffering of input/output ports used for receiving and transmitting print data at a printer |
US6031624A (en) * | 1997-09-08 | 2000-02-29 | Lexmark International, Inc. | Method and apparatus for adaptive data buffering in a parallelized printing system |
US6434671B2 (en) * | 1997-09-30 | 2002-08-13 | Intel Corporation | Software-controlled cache memory compartmentalization |
US20020042861A1 (en) * | 1997-11-07 | 2002-04-11 | Kavipurapu Gautam Nag | Apparatus and method for implementing a variable block size cache |
US6349364B1 (en) * | 1998-03-20 | 2002-02-19 | Matsushita Electric Industrial Co., Ltd. | Cache memory system with variable block-size mechanism |
US6496916B1 (en) * | 1998-04-17 | 2002-12-17 | Agere Systems Inc. | System for flexible memory paging in partitioning memory |
EP1095373A2 (en) | 1998-05-15 | 2001-05-02 | Storage Technology Corporation | Caching method for data blocks of variable size |
JP2002189607A (en) * | 2000-12-22 | 2002-07-05 | Nec Corp | Memory managing method and information processor |
US6745266B1 (en) * | 2001-12-21 | 2004-06-01 | Unisys Corporation | Method and apparatus for disk cache translation between systems |
CN1714347A (en) * | 2002-12-17 | 2005-12-28 | 国际商业机器公司 | Selectively changeable line width memory |
US7406579B2 (en) * | 2002-12-17 | 2008-07-29 | International Business Machines Corporation | Selectively changeable line width memory |
JP3694005B2 (en) * | 2003-05-21 | 2005-09-14 | 沖電気工業株式会社 | Digital signal processing apparatus and digital signal processing method |
US7257691B2 (en) * | 2003-09-26 | 2007-08-14 | International Business Machines Corporation | Writing and reading of data in probe-based data storage devices |
US7502887B2 (en) * | 2003-11-12 | 2009-03-10 | Panasonic Corporation | N-way set associative cache memory and control method thereof |
US7284093B2 (en) * | 2004-01-16 | 2007-10-16 | International Business Machines Corporation | Self-tuning cache |
JP4113524B2 (en) * | 2004-09-13 | 2008-07-09 | インターナショナル・ビジネス・マシーンズ・コーポレーション | Cache memory system and control method thereof |
JP2006113882A (en) * | 2004-10-15 | 2006-04-27 | Fujitsu Ltd | Data management device |
JP4535047B2 (en) * | 2006-09-06 | 2010-09-01 | ソニー株式会社 | Image data processing method, program for image data processing method, recording medium recording program for image data processing method, and image data processing apparatus |
KR100843218B1 (en) * | 2006-12-18 | 2008-07-02 | 삼성전자주식회사 | Flash memory device and method for changing a block size using address shifting |
US8266409B2 (en) | 2009-03-03 | 2012-09-11 | Qualcomm Incorporated | Configurable cache and method to configure same |
US9514055B2 (en) * | 2009-12-31 | 2016-12-06 | Seagate Technology Llc | Distributed media cache for data storage systems |
KR101710116B1 (en) * | 2010-08-25 | 2017-02-24 | 삼성전자주식회사 | Processor, Apparatus and Method for memory management |
US8775506B2 (en) | 2011-10-19 | 2014-07-08 | Oracle International Corporation | Eager block fetching for web-based data grids |
US9098595B2 (en) | 2011-10-19 | 2015-08-04 | Oracle International Corporation | Adaptive navigation functionality in a web-based data-entry grid |
US9411735B2 (en) * | 2014-04-15 | 2016-08-09 | International Business Machines Corporation | Counter-based wide fetch management |
CN106920279B (en) * | 2017-03-07 | 2018-06-19 | 百度在线网络技术(北京)有限公司 | Three-dimensional map construction method and device |
JP7468218B2 (en) * | 2020-07-22 | 2024-04-16 | 富士通株式会社 | Semiconductor device and cache control method |
US20240143510A1 (en) * | 2022-11-01 | 2024-05-02 | Arm Limited | Apparatus and method for cache invalidation |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US3764996A (en) * | 1971-12-23 | 1973-10-09 | Ibm | Storage control and address translation |
US4161024A (en) * | 1977-12-22 | 1979-07-10 | Honeywell Information Systems Inc. | Private cache-to-CPU interface in a bus oriented data processing system |
Family Cites Families (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US3820078A (en) * | 1972-10-05 | 1974-06-25 | Honeywell Inf Systems | Multi-level storage system having a buffer store with variable mapping modes |
US3840863A (en) * | 1973-10-23 | 1974-10-08 | Ibm | Dynamic storage hierarchy system |
JPS5177038A (en) * | 1974-12-27 | 1976-07-03 | Fujitsu Ltd | |
US4047243A (en) * | 1975-05-27 | 1977-09-06 | Burroughs Corporation | Segment replacement mechanism for varying program window sizes in a data processing system having virtual memory |
JPS5263038A (en) * | 1975-10-01 | 1977-05-25 | Hitachi Ltd | Data processing device |
JPS54128634A (en) * | 1978-03-30 | 1979-10-05 | Toshiba Corp | Cash memory control system |
US4234934A (en) * | 1978-11-30 | 1980-11-18 | Sperry Rand Corporation | Apparatus for scaling memory addresses |
-
1979
- 1979-12-19 US US06/105,186 patent/US4315312A/en not_active Expired - Lifetime
-
1980
- 1980-12-10 WO PCT/US1980/001666 patent/WO1981001894A1/en active IP Right Grant
- 1980-12-10 EP EP81900188A patent/EP0042000B1/en not_active Expired
- 1980-12-10 DE DE8181900188T patent/DE3072078D1/en not_active Expired
- 1980-12-10 JP JP81500394A patent/JPS56501548A/ja active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US3764996A (en) * | 1971-12-23 | 1973-10-09 | Ibm | Storage control and address translation |
US4161024A (en) * | 1977-12-22 | 1979-07-10 | Honeywell Information Systems Inc. | Private cache-to-CPU interface in a bus oriented data processing system |
Non-Patent Citations (2)
Title |
---|
FLORES, Computer Organization, 1969 Prentice Hall, Englewood Cliffs, N.J. Chapter 9, pages 228-237 * |
See also references of EP0042000A4 * |
Cited By (23)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP0285172A3 (en) * | 1987-03-31 | 1990-09-05 | Nec Corporation | Cache controller with a variable mapping mode |
EP0285172A2 (en) * | 1987-03-31 | 1988-10-05 | Nec Corporation | Cache controller with a variable mapping mode |
US5627992A (en) * | 1988-01-20 | 1997-05-06 | Advanced Micro Devices | Organization of an integrated cache unit for flexible usage in supporting microprocessor operations |
EP0325422A2 (en) * | 1988-01-20 | 1989-07-26 | Advanced Micro Devices, Inc. | Integrated cache unit |
EP0325421A2 (en) * | 1988-01-20 | 1989-07-26 | Advanced Micro Devices, Inc. | Organization of an integrated cache unit for flexible usage in supporting multiprocessor operations |
EP0325420A2 (en) * | 1988-01-20 | 1989-07-26 | Advanced Micro Devices, Inc. | Organization of an integrated cache unit for flexible usage in cache system design |
EP0325419A3 (en) * | 1988-01-20 | 1991-01-02 | Advanced Micro Devices, Inc. | Method and apparatus for caching interlock variables in an integrated cache memory |
EP0325422A3 (en) * | 1988-01-20 | 1991-01-09 | Advanced Micro Devices, Inc. | Integrated cache unit |
EP0325420A3 (en) * | 1988-01-20 | 1991-01-09 | Advanced Micro Devices, Inc. | Organization of an integrated cache unit for flexible usage in cache system design |
EP0325421A3 (en) * | 1988-01-20 | 1991-01-16 | Advanced Micro Devices, Inc. | Organization of an integrated cache unit for flexible usage in supporting multiprocessor operations |
EP0325419A2 (en) * | 1988-01-20 | 1989-07-26 | Advanced Micro Devices, Inc. | Method and apparatus for caching interlock variables in an integrated cache memory |
EP0342631A2 (en) * | 1988-05-18 | 1989-11-23 | Nec Corporation | Buffer memory device capable of memorizing operand and instruction data blocks at different block sizes |
EP0342631A3 (en) * | 1988-05-18 | 1991-09-18 | Nec Corporation | Buffer memory device capable of memorizing operand and instruction data blocks at different block sizes |
EP0448205A3 (en) * | 1990-03-23 | 1992-11-19 | Advanced Micro Devices, Inc. | Re-configurable block length cache |
EP0448205A2 (en) * | 1990-03-23 | 1991-09-25 | Advanced Micro Devices, Inc. | Re-configurable block length cache |
FR2733066A1 (en) * | 1995-04-14 | 1996-10-18 | Nec Corp | Controller for variable size cache memory |
US5809535A (en) * | 1995-04-14 | 1998-09-15 | Nec Corporation | Cache memory control apparatus utilizing a bit as a second valid bit in a tag in a first mode and as an additional address bit in a second mode |
EP2911062A4 (en) * | 2012-10-18 | 2015-10-14 | Zte Corp | Method and device for adjusting cache block length of cache memory |
WO2014155073A1 (en) * | 2013-03-25 | 2014-10-02 | Lancaster University Business Enterprises Limited | Cache |
GB2538055A (en) * | 2015-04-28 | 2016-11-09 | Advanced Risc Mach Ltd | Data processing apparatus having a cache |
GB2538055B (en) * | 2015-04-28 | 2017-04-19 | Advanced Risc Mach Ltd | Data processing apparatus having a cache |
US10250709B2 (en) | 2015-04-28 | 2019-04-02 | Arm Limited | Data processing apparatus, controller, cache and method |
US10467140B2 (en) | 2015-04-28 | 2019-11-05 | Arm Limited | Apparatus having main TLB and local TLBS, and configured to set selected size for new entry allocated to local TLB to a default size |
Also Published As
Publication number | Publication date |
---|---|
EP0042000A4 (en) | 1985-02-18 |
JPS56501548A (en) | 1981-10-22 |
DE3072078D1 (en) | 1988-03-31 |
US4315312A (en) | 1982-02-09 |
EP0042000B1 (en) | 1988-02-24 |
EP0042000A1 (en) | 1981-12-23 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
EP0042000B1 (en) | Cache memory in which the data block size is variable | |
US5737750A (en) | Partitioned single array cache memory having first and second storage regions for storing non-branch and branch instructions | |
US5091851A (en) | Fast multiple-word accesses from a multi-way set-associative cache memory | |
US4493026A (en) | Set associative sector cache | |
US5689679A (en) | Memory system and method for selective multi-level caching using a cache level code | |
US5465342A (en) | Dynamically adaptive set associativity for cache memories | |
US3979726A (en) | Apparatus for selectively clearing a cache store in a processor having segmentation and paging | |
US6772316B2 (en) | Method and apparatus for updating and invalidating store data | |
US5367653A (en) | Reconfigurable multi-way associative cache memory | |
US4707784A (en) | Prioritized secondary use of a cache with simultaneous access | |
US4928239A (en) | Cache memory with variable fetch and replacement schemes | |
US5018061A (en) | Microprocessor with on-chip cache memory with lower power consumption | |
US5754819A (en) | Low-latency memory indexing method and structure | |
US5835928A (en) | Circuitry and method for relating first and second memory locations where the second memory location stores information from the first memory location | |
US5706466A (en) | Von Neumann system with harvard processor and instruction buffer | |
CA2020275C (en) | Apparatus and method for reading, writing, and refreshing memory with direct virtual or physical access | |
EP0508577A1 (en) | Address translation mechanism | |
US6874077B2 (en) | Parallel distributed function translation lookaside buffer | |
US6745291B1 (en) | High speed LRU line replacement system for cache memories | |
GB2176918A (en) | Memory management for microprocessor system | |
US6571316B1 (en) | Cache memory array for multiple address spaces | |
US5539894A (en) | Method and apparatus for optimizing a sector cache tag, block and sub-block structure base on main memory size | |
US5511180A (en) | Method and circuit for determining the size of a cache memory | |
US5218687A (en) | Method and apparatus for fast memory access in a computer system | |
US4961135A (en) | Translation lookaside buffer control system |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
AK | Designated states |
Designated state(s): JP |
|
AL | Designated countries for regional patents |
Designated state(s): DE FR GB |
|
WWE | Wipo information: entry into national phase |
Ref document number: 1981900188 Country of ref document: EP |
|
WWP | Wipo information: published in national office |
Ref document number: 1981900188 Country of ref document: EP |
|
WWG | Wipo information: grant in national office |
Ref document number: 1981900188 Country of ref document: EP |