cache- Cache memory
In main memory and CPU Insert one or more levels between SRAM Composed cache memory . extend cache Limited , because SRAM It's expensive .
cache effect ： In order to solve the problem CPU An important technique for speed mismatch between and main memory .
cache characteristic ： It has spatial locality and temporal locality .
cache Composition of ：SRAM And control logic . If cache stay CPU Off chip , Its control logic is generally combined with the main memory control logic , Called main storage /cache controller . if cache stay CPU within , By CPU Provide its control logic .
CPU And cache The data exchange between them is based on words , and cache Data exchange with main memory is in blocks . A block consists of several words , It's fixed length .
cache schematic diagram
cache Hit rate of ： In order to make the average read time of main memory as close as possible cache Read out time of ,cache Hit rate should be close to 1.
Address mapping
meaning ： In order to put the main memory block in the cache in , Some method must be used to locate the main memory address to cache in , It's called address mapping .
Address mapping method ： Fully associative mapping , Direct mapping and group associative mapping .
Fully associative mapping
Summary ：
（1） In full connection cache in , All tags are implemented with an associated memory , All data with a common RAM To achieve .
（2） advantage ” Low conflict rate ,cache High utilization rate
（3） shortcoming ： Comparator is difficult to design and implement
（4） Small capacity only cache.
Direct mapping
Summary ：
（1） advantage ： The hardware is simple , Low cost .
（2） shortcoming ： Each main memory block has only one fixed row location to store .
（3） High conflict rate .（ If block numbers are spaced apart m Two blocks are stored in the same integer multiple cache Line time ）
（4） Suitable for large capacity cache.
Group associative mapping
Summary ：
Moderate consideration “ Fully associative mapping ” and “ Direct mapping ” And try to avoid the disadvantages of both .
Replacement strategy
meaning ： When a new main memory block needs to be copied to cache, When the row position of the block is occupied by other main memory , It's going to take place .
Suitable address mapping ： Fully associative mapping and group associative mapping
（1） Least frequently used （LFU） algorithm
meaning ： Swap out the row of data that has been accessed the least in a period of time . Set one counter per line , from 0 Start counting , Each time a row is accessed, the counter is incremented 1. When replacement is needed , Swap out the row with the smallest count , At the same time, all counters of these lines are cleared .
characteristic ： In this algorithm, the counting period is limited to the interval between two substitutions of these fixed lines （ Replace once , The counter is cleared once ）, It can not strictly reflect the recent visit .
（2） Minimum use in the near future （LRU） algorithm
meaning ： Replace those that have not been visited for a long time in the near future . Set one counter per line ,Cache Every hit , Hit line counter reset , Other lines of counter increase 1. When replacement is needed , Swap out the line with the largest count .
characteristic ： This algorithm protects just copied to cache New data row in , Has a high hit rate .
（3） Random substitution
meaning ： from cache Select a line randomly from the row position of .
characteristic ： It is easy to implement in hardware , And it's faster than those strategies . But it may decrease cache Hit rate and work efficiency .
（4） First in, first out （FIFO） algorithm
meaning ： It's always the first to be called in cache Replace the contents of , It is not necessary to record the usage of each block at any time .
characteristic ： Easy to implement , The circuit is simple . However, some frequently used programs may be added （ Such as loop program ） As the earliest cache Block and replace it .
cache Write back operation strategy of
meaning ：CPU Yes Cache The write of has changed Cache Content of . When the content is changed Cache The block is replaced Cache Time , Select the write back operation replacement strategy
Cache The content is consistent with the main memory content .
（1） Write back method
When CPU write Cache Hit time , Modify only Cache Content of , It does not write to main memory immediately ; Write back to main memory only when the row is replaced .
advantage ： Reduced access to main memory .
shortcoming ： There are hidden dangers of inconsistency .
solve the problem ： each Cache Row must be configured with a modify bit , To reflect whether the trip was CPU Modified .
（2） Complete writing
When CPU write Cache Hit time ,Cache Write changes occur simultaneously with main memory . So it is well maintained Cache Consistency with main memory content . When CPU write Cache On Miss , Write directly to main memory .
advantage ：Cache There is no need to set a modification bit and corresponding judgment logic for each line in .
shortcoming ：cache Yes CPU Write to main memory without caching , Reduced Cache The efficacy of .
（3） Write once
Writing strategy based on writeback and full write , The processing method of write hit and write miss is basically the same as that of writeback , Only the first write hit is written to main memory at the same time （ Complete writing ）.
advantage ： Easy to maintain the whole system Cache Consistency of .
Technology