Assignment 4 Cache Simulator Solution

Overview



The goal of this assignment is to help you understand caches better. You are required to write a cache simulator using the C programming language. The programs have to run on iLab machines. We are providing real program memory traces as input to your cache simulator. The format and structure of the memory traces are described below.







Memory Access Traces



The input to the cache simulator is a memory access trace, which we have generated by executing real programs. The trace contains memory addresses accessed during program execution. Your cache simulator will have to use these addresses to determine if the access is a hit or a miss, and the actions to perform in each case. The memory trace le consists of multiple lines. Each line of the trace le corresponds to a memory accesses performed by the program. Each line consists of two columns, which are space separated. The second column reports 48-bit memory address that has been accessed by the program while the rst column indicates whether the memory access is a read (R) or a write (W). The trace le always ends with a #eof string. We have provided you three input trace les (some of them are larger in size). You can safely assume the trace les always have proper format. Here is a sample trace le.




0x9cb3d40 W 0x9cb3d40



0x9cb3d44 W 0x9cb3d44
0xbf8ef498



#eof

Cache Simulator



You will implement a cache simulator to evaluate di erent con gurations of caches. Your program should be able to support traces with any number of lines. The followings are the requirements for your cache simulator:




You simulate one level cache.




The cache size, associativity, the replacement policy, and the block size are the input parameters. Cache size and block size are speci ed in bytes.




Replacement algorithm: First In First Out (FIFO). You have to simulate a write through cache.




Cache Simulator Interface



You have to name your cache simulator C code rst. Your program should support the fol-lowing usage interface: ./ rst <cache size<block size<cache policy<associativity<prefetch size<trace le







where:




<cache sizeis the total size of the cache in bytes. This number should be a power of 2.



<block sizeis a power of 2 integer that speci es the size of the cache block in bytes.



<cache policyHere is valid cache policy is fo and lru if you do the extra credit.



<associativityis one of:



direct - simulate a direct mapped cache.




assoc - simulate a fully associative cache.




assoc:n - simulate an n way associative cache. n will be a power of 2.




<prefetch sizeis the number of ajacent blocks that should be prefetched by the prefetcher in case of a miss



<trace leis the name of the input trace le.






NOTE: Your program should check if all the inputs are in valid format, and the trace le exist. if not print error and then silently terminate the program.













Cache Prefetcher



Prefetching is a common technique to increase the spatial locality of the caches beyond the cache line. The idea of prefetching is to bring the data into the cache before it is needed (accessed). In a normal cache, you bring a block of data into the cache whenever you



experience a cache-miss. Now, we want you to explore a di erent type of cache that prefetches not only brings the block corresponding to the access but also prefetches some adjacent blocks. The number of adjacent blocks that has to be prefetched every time prefetcher enabled, is provided by the user as an input.




Here is some important assumptions about the prefetcher. First, the prefetcher is activated only on cache misses. Second, none of the counters (including the memory read) are updated if the prefetched block already exist in the cache. Third, with respect to cache replacement policies, if the prefetched block hits in the cache, the block replacement policy status should not be updated. Otherwise, it is treated similar to a block that missed the cache.







Sample Output



In your program, you should keep track of four counters. These four counters are memory reads (per cache block), memory writes (per cache block), cache hits, and cache misses. As the output, your program should print out these four counter in the format shown below. You should follow the exact same format (pay attention to case sensitivity of the letters), otherwise, the autograder can not grade your program properly and you do not get any points.




$./ rst 32 4 fo assoc:2 1 trace2.txt




no-prefetch




Memory reads: 3499




Memory writes: 2861




Cache hits: 6501




Cache misses: 3499




with-prefetch




Memory reads: 3521




Memory writes: 2861




Cache hits: 8124




Cache misses: 1876







In this example, we are simulating a 2-way set associate cache of size 32 bytes. Each cache block is 4 bytes. The prefetch size is 1, meaning that we prefetch only one adjacent block. The trace le name is trace2.txt.




NOTE: Some of the trace les are quite large. So it might take a few minutes for the autograder to grade for all the testcases.







Cache Replacement Policy



he goal of the cache replacement policy is to decide which block has to be evicted in case there is no space in the set for an incoming cache block. It is always preferable to achieve the best performance to replace the block that will be re-referenced furthest in the future.



There are di erent ways one can implement cache replacement policy. Here we use FIFO replacement policy and LRU policy is extra credit.







7.1 FIFO




Using this algorithm, you can always evict the block accessed rst in the set without any regard to how often or how many times it was accessed before. So let us say that your cache is empty initially and that each set has two ways. Now suppose that you access blocks A, B, A, C. To make room for C, you would evict A since it was the rst block to be brought into the set.







7.2 LRU




Least Recently Used (LRU) replacement policy is used replace the cache line that is least recently used. Let us use the same example as the one for FIFO. If we access A,B,A,C. To make room for C, we should evict B since it is the least recently used block (in contrast to FIFO which we evicted A)







Other Details



(a) When your program starts, there is nothing in the cache. So, all cache lines are empty (invalid).



(b) you can assume that the memory size is 2pow48 . Therefore, memory addresses are 48 bit (zero extend the addresses in the trace le if theyre less than 48-bit in length).




(c) the number of bits in the tag, cache address, and byte address are determined by the cache size and the block size;




For a write-through cache, there is the question of what should happen in case of a write miss. In this assignment, the assumption is that the block is rst read from memory (one read memory), and then followed by a memory write.



3- You do not need to simulate the memory in this assignment. Because, the traces doesnt contain any information on data values transferred between the memory and the caches.




You have to compile your program with the following ags: -Wall -Werror -fsanitize=address



You should include a make le in you submission.






Extra credit (25 points)



As an extra credit, you should implement LRU (Least Recently Used) cache policy. Your program should output exactly the same format output as it shown before. Here is an ex-ample of running your program with LRU policy.



$./ rst 32 4 lru assoc:2 1 trace2.txt













10 Submission




You have to e-submit the assignment using Sakai . Put all les (source code + Make le) into a directory named rst, which itself is a sub-directory under pa4 . Then, create a tar le (follow the instructions in the previous assignments to create the tar le). Your submission should be only a tar le named pa4.tar. You have to e-submit the assignment using Sakai. Your submission should be a tar le named pa4.tar. To create this le, put everything that you are submitting into a directory named pa4. Then, cd into the directory containing pa4 (that is, pa4s parent directory) and run the following command: $tar cvf pa4.tar pa4




To check that you have correctly created the tar le, you should copy it (pa4.tar) into an empty directory and run the following command: $tar xvf pa4.tar







This is how the folder structure should be.




pa4




{ rst




rst.c rst.h Make le




Source code: all source code les necessary for building your programs. e.g. rst.c and rst.h.




Make le: There should be at least three rules in your Make le:




all: make a complete build of your program ( rst).




rst: build the executables ( rst).




clean: prepare for rebuilding from scratch.










Do NOT INCLUDE ANY OF THE TRACE FILES IN THE FOLDER







11 Autograder




First mode




Testing when you are writing code with a pa4 folder.



Lets say you have a pa4 folder with the directory structure as described in the assignment.



Copy the folder to the directory of the autograder



Run the autograder with the following command



$python auto grader.py







It will run the test cases and print your scores.




Second mode




This mode is to test your nal submission (i.e, pa4.tar) 1. Copy pa4.tar to the autograder directory




Run the autograder with pa4.tar as the argument as below: $python auto grader.py pa4.tar









12 Grading guidelines




We should be able build your program by just running make.



Your program should follow the format speci ed above for the usage in-terface.



Your program should strictly follow the input and output speci cations mentioned above. (Note: This is perhaps the most important guideline: fail-ing to follow it might result in you losing all or most of your points for this assignment. Make sure your programs output format is exactly as speci ed. Any deviation will cause the automated grader to mark your output as incor-rect. REQUESTS FOR RE-EVALUATIONS OF PROGRAMS REJECTED DUE TO IMPROPER FORMAT WILL NOT BE ENTERTAINED.)