CN1996268A

CN1996268A - Method for implementing on-chip command cache

Info

Publication number: CN1996268A
Application number: CNA2006101697244A
Authority: CN
Inventors: 车德亮; 黄玮; 权海洋
Original assignee: Mxtronics Corp
Current assignee: Beijing times people core technology Co., Ltd.; China Aerospace Modern Electronic Company 772nd Institute
Priority date: 2006-12-28
Filing date: 2006-12-28
Publication date: 2007-07-11
Anticipated expiration: 2026-12-28
Also published as: CN100428200C

Abstract

The cache is made by cache control register, cache starting address register, matching mark P position, cache storage and LRU substitute stack. The cache is simple in structure, small in size, low in energy consumption, simple working process based on the structure, easy coordination with other functional parts, improving the reliability of the instruction cache, using hardware substituting stack selecting substituting section name, avoiding software judgment, improving working speed of the instruction cache.

Description

A kind of implementation method of on-chip command cache

Technical field

The invention belongs to field of computer technology, relate to the design and the manufacturing of high performance signal processor, specially refer to a kind of hardware implementation method of on-chip command cache control system.

Background technology

Since invention on computer, computing velocity and the unmatched contradiction of I/O speed just exist, and the cache technology is exactly one of technology that solves this contradiction.Traditional cache control system is normally finished by software, just be responsible for finishing the control and management of cache as the operating system of PC, the store status of whole cache, data dispatch, replacement policy, abnormality processing etc. all are to be responsible for finishing by the software program in the operating system.Along with the progress of microelectronics manufacture, computer system can realize on a chip that cache also is integrated on the sheet, how the cache on the sheet is carried out effective control and management, is the important content in the high-performance treatments chip design.

The implementation method of relevant command cache at present both domestic and external mainly is the method that adopts hardware and software to combine, and realizes by mainframe computer hardware and software operating system.This method realizes that the control of cache instruction has that power consumption is big, the time of instruction fetch is long and the shortcoming of poor reliability.

Summary of the invention

Technology of the present invention is dealt with problems: overcome the deficiencies in the prior art, a kind of implementation method of on-chip command cache is provided, adopt this method to improve the operating rate of command cache, simultaneously simple in structure, area is little, and is low in energy consumption, be convenient on chip, realize, and the reliability height.

Technical solution of the present invention: a kind of implementation method of on-chip command cache, its characteristics are: the structure of command cache is by a cache control register, cache segment base register, matching mark P position, cache section word memory and LRU replace storehouse and constitute, the cache control register is used to control or represent the state of cache, cache segment base register is used for the sector address of storage instruction address, whether matching mark P position is used for sign has the interior word of a section to aim at, cache section word memory is used for storage instruction, and LRU replaces storehouse and is used to write down the order that the cache section is replaced; When request external memory storage when providing instruction word, there will be two kinds of situations, cache hit or cache miss, if cache hits, sense order from cache, this fragment number are pressed into the top that LRU replaces storehouse, dispose corresponding P mark; If cache is miss, have two kinds of situations: first kind of situation is that cache segment base register and this instruction address are complementary, but corresponding matching mark P position is not set, carry out following operation this moment simultaneously: sense order and copy cache to from storer, this fragment number is pressed into the top that LRU replaces storehouse, disposes corresponding matching mark P position; Second kind of situation is that cache segment base register and instruction address does not match, carry out following operation this moment simultaneously: replace from LRU and select the storehouse with the fragment number that is replaced, all matching mark P bit clears in this section, the corresponding bit wide value of address of instruction is downloaded in the cache segment base register of segmentation of replacement, get instruction and copy cache to and, the fragment number of replacing is pressed into the stack top that LRU replaces storehouse the P flag set.

The present invention's advantage compared with prior art is:

(1) go up the cache order structure for of the present invention, have simple in structurely, area is little, and is low in energy consumption, is convenient to realize on chip.

(2) workflow that designs according to the structure of on-chip command cache of the present invention is simple, has improved the reliability of command cache work.

(3) adopt the hardware replacement storehouse to select to replace segment number, avoided the software of prior art to judge, improved the operating rate of command cache.

Description of drawings

Fig. 1 is the structural representation of a kind of on-chip command cache of the present invention;

Fig. 2 is the workflow of a kind of on-chip command cache of the present invention;

Fig. 3 is the structural drawing that LRU of the present invention replaces storehouse.

Embodiment

As shown in Figure 1, in the SMDSP signal processor, the command cache capacity is 64 * 32 words, and it is divided into two 32 fields, and one 19 segment base register and each section are associated.Each word among the cache all has corresponding matching mark P position, otherwise P is that word is effectively then invalid among the 1 expression cache.

The structure of command cache is replaced storehouse by a cache control register, cache segment base register, matching mark P position, cache section word memory and LRU and is constituted, the cache control register is used to control or represent the state of cache, cache segment base register is used for the sector address of storage instruction address, whether matching mark P position is used for sign has the interior word of a section to aim at, cache section word memory is used for storage instruction, and LRU replaces storehouse and is used to write down the order that the cache section is replaced.

As shown in Figure 2, when request external memory storage when providing instruction word, there will be two kinds of situations, cache hit or cache miss, if cache hits, sense order from cache, this fragment number are pressed into the top that LRU replaces storehouse, dispose corresponding P mark; If cache is miss, have two kinds of situations: first kind of situation is that cache segment base register and this instruction address are complementary, but corresponding matching mark P position is not set, carry out following operation this moment simultaneously: sense order and copy cache to from storer, this fragment number is pressed into the top that LRU replaces storehouse, disposes corresponding matching mark P position; Second kind of situation is that cache segment base register and instruction address does not match, carry out following operation this moment simultaneously: replace from LRU and select the storehouse with the fragment number that is replaced, all matching mark P bit clears in this section, the corresponding bit wide value of address of instruction is downloaded in the cache segment base register of segmentation of replacement, get instruction and copy cache to and, the fragment number of replacing is pressed into the stack top that LRU replaces storehouse the P flag set.

LRU replaces storehouse and is used for determining which section of two sections is least recently used.When visiting one section, fragment number is left the LRU storehouse and is advanced to the top that LRU replaces storehouse at every turn.Therefore, the sequence number at storehouse top is the fragment number of using recently at most, and the sequence number of storehouse bottom is least-recently-used fragment number, and when resetting, the LRU storehouse is with 0 initialization top with 1 initialization lower curtate.

As shown in Figure 3, the replacement stack architecture of LRU of the present invention: significance bit CE (cache enable) and read signal position R (Read) process and door AND_1, the output and the RESET or non-that resets, through rejection gate NOR_2, output to the control end of transmission gate T1, the output of NOR_2 is connected to the complementary control end of T1 through reverser INV_1, clock CLK1 is connected to an input end with door And_2 behind T1, the output of Reset and SSA0 (section 0) warp or door or_1 and Reset and SSA1 (section 1) are connected to Sheffer stroke gate Nand_1 through the output of rejection gate Nor_1, output is linked and door And_2, be connected to reverser INV_2 with the output of door And_2, output is connected to reverser INV_3 and transmission gate T3.Reverser INV_3 is output as MRU_SN (using maximum segment numbers in the recent period), connects transmission gate T2 simultaneously.The output of transmission gate T2 is connected to reverser INV_2; Transmission gate T3 is output as LRU_SN (using minimum segment number in the recent period), and the control end of transmission gate T2 and T3 all meets CLK2, and complementary control signal connects the output of CLK2 through reverser INV_1.

The cache control register has 3: (1) clear cache position CC, work as CC=1, and make all passages among the Cache invalid.This position is changed to 0 after writing cache, and this position is 0 when resetting.(2) cache significance bit CE works as CE=1, enables cache, allows to use cache according to lru algorithm; Work as CE=0, forbid cache, can prevent that cache is updated modification; This position 0 when resetting.(3) freeze Cache position CF, freeze the operation of cache and LRU replacement storehouse as CF=1.If when CE=1 and CF=1, allow from cache, to read but do not allow to revise cache.When CF=1 or 0, allow cache to remove (CC=1), the CF position is made as 0 when resetting.

When the request external memory storage provides when instruction, the workflow of command cache as shown in Figure 2, promptly when asking external memory storage to provide instruction word, there will be two kinds of situations, cache hit or cache miss.If cache hits, sense order from cache; This fragment number is pressed into the top that LRU replaces storehouse; Dispose corresponding P mark.If cache is miss, there are two kinds of situations: first kind of situation, be that cache section beginning address register and instruction address is complementary, but corresponding matching mark P position is not set, carry out 3 operations this moment simultaneously, sense order and copy Cache to from storer, this fragment number are pressed into the top that LRU replaces storehouse, dispose corresponding matching mark P position; Second kind of situation, be that caohe segment base register and instruction address does not match, carry out 4 operations this moment simultaneously, from LRU replacement storehouse, select the fragment number that is replaced, 32 matching mark P bit clears in this section, 19 high addresss of request instruction are downloaded in the cache segment base register of segmentation of LRU_SN signal indication that LRU replaces storehouse, get instruction and copy cache to and, the fragment number of replacing is pressed into the stack top that LRU replaces storehouse the P flag set.

Claims

1, a kind of implementation method of on-chip command cache, it is characterized in that: the structure of command cache is by a cache control register, cache segment base register, matching mark P position, cache section word memory and LRU replace storehouse and constitute, the cache control register is used to control or represent the state of cache, cache segment base register is used for the sector address of storage instruction address, whether matching mark P position is used for sign has the interior word of a section to aim at, cache section word memory is used for storage instruction, and LRU replaces storehouse and is used to write down the order that the cache section is replaced; When request external memory storage when providing instruction word, there will be two kinds of situations, cache hit or cache miss, if cache hits, sense order from cache, this fragment number are pressed into the top that LRU replaces storehouse, dispose corresponding P mark; If cache is miss, have two kinds of situations: first kind of situation is that cache segment base register and this instruction address are complementary, but corresponding matching mark P position is not set, carry out following operation this moment simultaneously: sense order and copy cache to from storer, this fragment number is pressed into the top that LRU replaces storehouse, disposes corresponding matching mark P position; Second kind of situation is that cache segment base register and instruction address does not match, carry out following operation this moment simultaneously: replace from LRU and select the storehouse with the fragment number that is replaced, all matching mark P bit clears in this section, the value of corresponding bit wide of the address of instruction is downloaded in the cache section beginning address register of segmentation of replacement, get instruction and copy cache to and, the fragment number of replacing is pressed into the stack top that LRU replaces storehouse the P flag set.

2, the implementation method of a kind of on-chip command cache according to claim 1, it is characterized in that: the replacement stack architecture of described LRU: the significance bit CE of cache and read signal position R process and door AND_1, the output and the RESET or non-that resets, through rejection gate NOR_2, output to the control end of transmission gate T1, the output of NOR_2 is connected to the complementary control end of T1 through reverser INV_1, CLK1 is connected to an input end with door And_2 behind T1, Reset and SSA0 (segment base register 0, SSA0) output and Reset and SSA1 (the segment base register 1 of warp or door or_1, SSA1) output through rejection gate Nor_1 is connected to Sheffer stroke gate Nand_1, output is linked and door And_2, be connected to reverser INV_2 with the output of door And_2, output is connected to reverser INV_3 and transmission gate T3, reverser INV_3 is output as the maximum segment number MRU_SN of recent use, connect transmission gate T2 simultaneously, the output of transmission gate T2 is connected to reverser INV_2, transmission gate T3 is output as the minimum segment number LRU_SN of recent use, the control end of transmission gate T2 and T3 all meets clock CLK2, and complementary control signal connects the output of CLK2 through reverser INV_1.

3, the implementation method of a kind of on-chip command cache according to claim 1 is characterized in that: described cache control register has 3, (1) clear cache position CC, work as CC=1, make all passages among the cache invalid, write that this position is changed to 0 behind the cache, this position is 0 when resetting; (2) cache significance bit CE works as CE=1, enables cache, allows to use cache according to lru algorithm, works as CE=0, forbids cache, prevents that cache is updated modification, this position 0 when resetting; (3) freeze cache position CF, when CF=1 freezes the operation that cache and LRU replace storehouse, if CE=1, and during CF=1, allow from cache, to read but do not allow to revise cache; When CF=1 or 0, allow cache to remove (CC=1), the CF position is made as 0 when resetting.