1. ECE 353 Lab 1: Cache Simulation

2. Purpose Introduce C programming by means of a simple example Reinforce your knowledge of set associative caches

3. Caches Motivation: The speed differential between main memory and processors is increasing. Main memory may take over a hundred CPU cycles to access. Aim: Achieve the speed of high-speed memory, while keeping the cost per bit close to that of low-speed memory Exploits locality:  Temporal locality  Spatial locality Many machines have two levels of cache

4. Units of Movement The contents of  Main memory: Organized into pages.  Cache: Organized into blocks or lines. Material brought into main memory is moved in these units

5. Cache Blocks or Lines Each memory address belongs to a cache block. The cache block address is the byte address with the rightmost lg B bits removed, where B is the size of the block

6. Parts of a Cache Directory: Where the tags associated with the cache contents are stored.  Each directory entry corresponds to a matching line in the cache  When looking for an item in cache, we seek to match the tag of the desired item against the contents of the directory.  In addition to the tag, we also have a valid bit. Only return values from the cache if the tag matches and the valid bit indicates the line is valid. Data Array: Where the contents of the cache are stored.

7. Fully Associative Cache Any block can be placed anywhere in the cache  This flexibility may reduce the miss rate When looking for an item, we need to search through the entire directory looking for a tag match  This parallel search can increase hardware complexity and the time taken to execute the match

8. Direct-Mapped Cache Each cache set consists of exactly one line  This reduces flexibility, which may increase the miss rate To check if the desired item is in the cache,  Identify the set that the item maps to  Check the contents of the single directory entry corresponding to that set.  Only one match needs to be done: can be done faster.

9. K-way Caches Each set consists of K lines When looking for an item in cache,  Identify the set to which it belongs  Do a parallel search among all the K tags in the directory corresponding to the K lines in that set

10. Set Associative Organization © 2004 Morgan-Kaufman Publishers

11. Typical Performance Use split caches because there is more spatial locality in code: © 2004 Morgan-Kaufman Publishers

12. Cache Simulation Used to determine the miss rate for a given trace and cache configuration. The trace is a sequence of memory addresses The cache configuration is  Data capacity of the cache  Number of lines (or blocks) per set  Number of bytes per line

13. C Many of the constructs are very similar to those found in Java In many other respects, C is more primitive:  Not object-oriented  Lacks inbuilt garbage collection, some checks (e.g. array bounds)  C carries out implicit type conversions when appropriate

14. Some C Constructs Relational, logical , >=, ==, !=  &&, ||, ! Arithmetic  +, -, *, /, %

15. C Constructs (contd.) Some control constructs:  while( ) { };  if ( ) { };  for (i=0; i<100; i++) { };

16. Data Types int char float double long short

17. Arrays x[1000] is a linear array of 1000 elements Y[100][100] is a two-dimensional 100x100 array Variable names (like other things in C) are case- sensitive: Y is not the same as y.

18. Example #include #define SIZE 100 main(){ int iter; int a[SIZE], b[SIZE], c[SIZE]; for (iter=0; iter<SIZE; iter++){ c[iter] = iter*3; b[iter] = iter/2; a[iter] = b[iter]+c[iter]; } printf(“My first C program\n”); for (iter=0; iter<SIZE; iter++) printf(“a[%d] = %d; b[%d]= %d; c[%d]= %d\n”, iter, a[iter], iter, b[iter], iter, c[iter]); }

19. Pointers A pointer is a variable which contains the address of some other variable.  int *i is the declaration of an integer whose address is stored in i.  int j is the declaration of another integer; its address is obtained by using the address operator, &. That is, the location of j is given by &j.  The pointer to an array is just the name of the array.  Example: For some array a[100], *a is a[0] and *(a+i) is a[i]

20. Opening and Reading Files: Example #include main(){ int x, y; FILE *ifp; //Pointer to a file is declared ifp = fopen(“trace.txt”, “r”); // ifp points to file // trace.txt, opened for // reading while (!feof(ifp) { // exit if end-of-file reached fscanf(ifp, “%d %x”, &x, &y); // read next line printf(“Address read = %x\n”, y);// print as required } fclose(ifp); // Close file } If individual file lines consist of an integer followed by a hex quantity, and we want to print out the hex quantity:

21. Opening and Writing into Files: Example #include main(){ int a[100], i; FILE *output; for (i=0; i<100; i++) a[i]=i; output=fopen(“example.txt”,”w”); for (i=0; i<100; i++) fprintf(output, “%5d\n”, a[i]); fclose(output) }

22. Cache Simulation (contd.) Given the following:  Cache parameters  Memory access trace Obtain:  Cache hit rate

23. Simulation Maintain the cache directory and LRU status of the lines within each set When an access is made, update LRU status.  If a hit, record it as such  If a miss, update the contents of the directory

24. Cache Directory Implementation 1: Implement the directory as an array, with array entries corresponding to directory entries.  Advantage: Simpler to program if you don’t have much C experience  Disadvantage: Array sizes are static. The array would have to be large enough to accommodate the largest sized directory that the simulator is designed for

25. Cache Directory (contd.) Implementation 2: Use pointers and malloc  Advantage: Is flexible and assigns memory dynamically.  Disadvantage: If this is your first non-trivial C program, assigning and freeing memory dynamically can be difficult. malloc: Dynamically assigns memory.  The call malloc(b)  Allocates memory that is sufficient to hold b bytes.  Returns a pointer to the starting address of that chunk of memory  To allocate memory sufficient to hold b integers, use malloc(b*sizeof(int))

26. Example of malloc int *dir_ptr; int k; //value of k can change dir_ptr = (int *) malloc(k * sizeof(int));

27. Freeing Memory Memory that is allocated using malloc must be freed once it is no longer needed. To free the memory that was allocated in the previous example, set dir_ptr to point to the starting point of the allocated chunk and do the call free(dir_ptr)