diff --git a/labs/malloclab_lab/mdriver b/labs/malloclab_lab/mdriver index 1ca2a70..7216ced 100644 Binary files a/labs/malloclab_lab/mdriver and b/labs/malloclab_lab/mdriver differ diff --git a/labs/malloclab_lab/mm.c b/labs/malloclab_lab/mm.c index dbadeba..f6b3b50 100644 --- a/labs/malloclab_lab/mm.c +++ b/labs/malloclab_lab/mm.c @@ -1,5 +1,5 @@ /* - * mm-naive.c - The clear list, first fit malloc package. + * mm-naive.c - The clear list(LIFO), best fit, coalesce immediately malloc package. */ #include @@ -19,7 +19,7 @@ team_t team = { "" }; -/* single word (4) or double word (8) alignment */ +/* single size_t (4) or double size_t (8) alignment */ #define ALIGNMENT 8 /* rounds up to the nearest multiple of ALIGNMENT */ #define ALIGN(size) (((size) + (ALIGNMENT - 1)) & ~0x7) @@ -28,7 +28,7 @@ team_t team = { #define WSIZE 4 #define DSIZE 8 #define FSIZE 16 -#define ADDRESS (sizeof(unsigned long)) +#define ADDRESS (sizeof(size_t)) #define CHUNK 1 << 12 #define MIN_BLOCK (2 * ADDRESS + DSIZE) @@ -50,10 +50,10 @@ team_t team = { #define POS_PRED(bp) ((byte *)(bp)) #define POS_SUCC(bp) (((byte *)(bp) + ADDRESS)) -#define GET_PRED(bp) (*(word *)POS_PRED(bp)) -#define GET_SUCC(bp) (*(word *)POS_SUCC(bp)) +#define GET_PRED(bp) (*(size_t *)POS_PRED(bp)) +#define GET_SUCC(bp) (*(size_t *)POS_SUCC(bp)) -typedef unsigned long word; +typedef unsigned int word; typedef char byte; // mark the front and tail pos @@ -113,12 +113,20 @@ static void *_next_best_fit(size_t size); /** * traverse blank block only and find first fit, then place in + * @deprecated for the memory loss * @param size align by 8, excluding head and foot * @return */ static void *_first_fit_of_clear(size_t size); /** + * best fit for clear list + * @param size align by 8, excluding head and foot + * @return + */ +static void *_best_fit_of_clear(size_t size); + +/** * allocate the block and cut sometimes * @param size align by 8, excluding head and foot */ @@ -157,7 +165,7 @@ void *mm_malloc(size_t size) { size_t adjust_size = ALIGN(size); size_t extend_size; void *bp; - if ((bp = _first_fit_of_clear(adjust_size)) != NULL) { + if ((bp = _best_fit_of_clear(adjust_size)) != NULL) { // fitted_p = bp; return bp; } else { @@ -208,15 +216,35 @@ void *mm_realloc(void *ptr, size_t size) { if (adjust_size <= total_size) { void *next = NEXT(ptr); // remove - SET(POS_SUCC(GET_PRED(next)), GET_SUCC(next)); - SET(POS_PRED(GET_SUCC(next)), GET_PRED(next)); - if (next == list_p) { - if (GET_SUCC(next) == (word)next) list_p = NULL; - else list_p = (void *)GET_SUCC(next); + if (total_size - adjust_size >= MIN_BLOCK) { + SET(POS_SUCC(GET_PRED(next)), GET_SUCC(next)); + SET(POS_PRED(GET_SUCC(next)), GET_PRED(next)); + SET(HEAD(ptr), PACK(total_size, 1)); + SET(FOOT(ptr), PACK(total_size, 1)); + if (next == list_p) { + if (GET_SUCC(next) == (size_t) next) list_p = NULL; + else list_p = (void *) GET_SUCC(next); + } + if (next == tail_p) tail_p = ptr; + } else { // replace + void *pred = (void *)GET_PRED(next); + void *succ = (void *) GET_SUCC(next); + SET(HEAD(ptr), PACK(adjust_size, 1)); + SET(FOOT(ptr), PACK(adjust_size, 1)); + size_t new_size = total_size - adjust_size - DSIZE; + void *new = NEXT(ptr); + SET(HEAD(new), PACK(new_size, 0)); + SET(FOOT(new), PACK(new_size, 0)); + if (pred == next) { + SET(POS_PRED(new), (size_t)new); + SET(POS_SUCC(new), (size_t)new); + } else { + SET(POS_PRED(succ), (size_t)new); + SET(POS_SUCC(pred), (size_t)new); + } + if (list_p == next) list_p = new; + if (next == tail_p) tail_p = new; } - SET(HEAD(ptr), PACK(total_size, 1)); - SET(FOOT(ptr), PACK(total_size, 1)); - if (next == tail_p) tail_p = ptr; return ptr; } else { if ((new_ptr = mm_malloc(size)) == NULL) return NULL; @@ -315,14 +343,14 @@ static void *__coalesce_none(void *bp) { // tweak list if (list_p == NULL) { list_p = bp; - SET(POS_SUCC(list_p), (word)list_p); - SET(POS_PRED(list_p), (word)list_p); + SET(POS_SUCC(list_p), (size_t)list_p); + SET(POS_PRED(list_p), (size_t)list_p); } else { // add to list SET(POS_SUCC(bp), GET_SUCC(list_p)); - SET(POS_PRED(bp), (word)list_p); - SET(POS_PRED(GET_SUCC(list_p)), (word)bp); - SET(POS_SUCC(list_p), (word)bp); + SET(POS_PRED(bp), (size_t)list_p); + SET(POS_PRED(GET_SUCC(list_p)), (size_t)bp); + SET(POS_SUCC(list_p), (size_t)bp); list_p = bp; } #ifdef DEBUG @@ -407,6 +435,25 @@ static void *_first_fit_of_clear(size_t size) { return NULL; } +static void *_best_fit_of_clear(size_t size) { + void *bp = list_p; + if (bp == NULL) return NULL; + size_t min = 0; + void *min_p = NULL; + do { + if (SIZE(bp) >= size) { + if (min_p == NULL || SIZE(bp) < min) { + min = SIZE(bp); + min_p = bp; + } + } + bp = (void *)GET_SUCC(bp); + } while (bp != list_p); + if (min_p == NULL) return NULL; + _place(min_p, size); + return min_p; +} + static void _place(void *ptr, size_t size) { size_t p_size = SIZE(ptr); if (p_size - size >= MIN_BLOCK) { @@ -429,7 +476,7 @@ static void _place(void *ptr, size_t size) { SET(POS_SUCC(GET_PRED(ptr)), GET_SUCC(ptr)); SET(POS_PRED(GET_SUCC(ptr)), GET_PRED(ptr)); if (ptr == list_p) { - if (GET_SUCC(ptr) == (word)ptr) list_p = NULL; + if (GET_SUCC(ptr) == (size_t)ptr) list_p = NULL; else list_p = (void *)GET_SUCC(ptr); } } @@ -441,14 +488,14 @@ static void _place(void *ptr, size_t size) { } static void _fix_list(void *in, void *out) { - if (GET_SUCC(in) == (word)in) { - SET(POS_SUCC(out), (word)out); - SET(POS_PRED(out), (word)out); + if (GET_SUCC(in) == (size_t)in) { + SET(POS_SUCC(out), (size_t)out); + SET(POS_PRED(out), (size_t)out); } else { SET(POS_SUCC(out), GET_SUCC(in)); SET(POS_PRED(out), GET_PRED(in)); - SET(POS_SUCC(GET_PRED(in)), (word)out); - SET(POS_PRED(GET_SUCC(in)), (word)out); + SET(POS_SUCC(GET_PRED(in)), (size_t)out); + SET(POS_PRED(GET_SUCC(in)), (size_t)out); } } diff --git a/labs/malloclab_lab/mm.o b/labs/malloclab_lab/mm.o index 362b601..c9d6dd6 100644 Binary files a/labs/malloclab_lab/mm.o and b/labs/malloclab_lab/mm.o differ diff --git a/labs/malloclab_lab/solve_note.md b/labs/malloclab_lab/solve_note.md index b10c2ad..088887b 100644 --- a/labs/malloclab_lab/solve_note.md +++ b/labs/malloclab_lab/solve_note.md @@ -13,7 +13,7 @@ 理解逻辑后，接下来将实现自己的版本 * 为方便理解，定义了两个类型别名 ```c - typedef unsigned long word; + typedef unsigned int word; typedef char byte; ``` * 不同于示例代码用序言块和尾块标记，本人仅用两个指针标记头尾，来提高内存利用率 @@ -152,7 +152,7 @@ Perf index = 48 (util) + 40 (thru) = 88/100 ## Version 2 #### 显示 LIFO 空闲链表 * 经过多次尝试，88 分基本上是隐式链表的极限了，为进一步提高吞吐率，我们维护一个双向链表将空闲块串起来，这里选择的则是 LIFO 方式 -* 首先由于 32/64 位的地址所需位数不同，为了保持 64 位干净，我们需要定义地址的位数为 `sizeof(unsigned long)`，同时还需定义最小的空闲块大小 +* 首先由于 32/64 位的地址所需位数不同，为了保持 64 位干净，我们需要定义地址的位数为 `sizeof(size_t)`，同时还需定义最小的空闲块大小 * 需要在合并、切割处着重维护显示链表，同时为了减少不必要的判断，本人将选择双向循环列表，并由 `list_p` 标记头部 * 首先，每次调用 `_coalesce` 之前都会产生一个未入链表的空闲块，我们只需要修改 `__coalesce_next, __coalesce_all`，以及没有合并的情况，并将 `list_p` 指向新块 * 初始化时记得将第一个空闲块的头尾指向自己 @@ -185,7 +185,32 @@ Total 80% 112372 0.008497 13225 Perf index = 48 (util) + 40 (thru) = 88/100 ``` +## Version 2.1 +#### 最佳适配 +* 可能也是为了鼓励提升内存利用率，而我们的吞吐率已经极大提高了，所以可以用最佳适配来换取内存利用率 +* `realloc` 也添加了一些情况的判断 +#### 得分 +* 我们得到了最终的得分，可以看到牺牲了很多吞吐率才换来了内存利用率百分之三的提升 +```c +Results for mm malloc: +trace valid util ops secs Kops + 0 yes 99% 5694 0.000145 39405 + 1 yes 100% 5848 0.000166 35314 + 2 yes 100% 6648 0.000207 32131 + 3 yes 100% 5380 0.000140 38374 + 4 yes 100% 14400 0.000089162528 + 5 yes 95% 4800 0.002331 2059 + 6 yes 95% 4800 0.002358 2035 + 7 yes 55% 12000 0.027697 433 + 8 yes 51% 24000 0.063180 380 + 9 yes 40% 14401 0.000257 55970 +10 yes 73% 14401 0.000077187513 +Total 83% 112372 0.096647 1163 + +Perf index = 50 (util) + 40 (thru) = 90/100 +``` ## 最后 +* 最终的版本是显示空闲链表LIFO，最佳适配，立即合并 * 所有用到的辅助函数均在开头声明并注释，被淘汰的函数用 `@deprecated` 标记 *** 2022.12.29 ~ 2022.12.31 \ No newline at end of file