#include "multicoremsg.h"
#include "multicoregcprofile.h"
-gc_cache_revise_info_t gc_cache_revise_information;
+void cacheadapt_finish_compact(void *toptr) {
+ unsigned int dstpage=(toptr-gcbaseva)>>BAMBOO_PAGE_SIZE_BITS;
+ unsigned int * newtable=&gccachesamplingtbl_r[dstpage*NUMCORESACTIVE];
+
+ for(int core = 0; core < NUMCORESACTIVE; core++) {
+ (*newtable)=(*newtable)>>6;
+ newtable++;
+ }
+}
+
+void cacheadapt_finish_src_page(void *srcptr, void *tostart, void *tofinish) {
+ unsigned int srcpage=(srcptr-gcbaseva)>>BAMBOO_PAGE_SIZE_BITS;
+ unsigned int dstpage=(tostart-gcbaseva)>>BAMBOO_PAGE_SIZE_BITS;
+ unsigned int numbytes=tofinish-tostart;
+
+ unsigned int * oldtable=&gccachesamplingtbl[srcpage*NUMCORESACTIVE];
+ unsigned int * newtable=&gccachesamplingtbl_r[dstpage*NUMCORESACTIVE];
+
+ unsigned int page64th=numbytes>>(BAMBOO_PAGE_SIZE_BITS-6);
+
+ for(int core = 0; core < NUMCORESACTIVE; core++) {
+ (*newtable)+=page64th*(*oldtable);
+ newtable++;
+ oldtable++;
+ }
+}
+
+/* Bytes needed equal to zero is a special case... It means that we should finish the dst page */
+
+void cacheadapt_finish_dst_page(void *origptr, void *tostart, void *toptr, unsigned int bytesneeded) {
+ unsigned int numbytes=toptr-tostart;
+
+ void *tobound=(void *)((((unsigned INTPTR)toptr-1)&~(BAMBOO_PAGE_SIZE-1))+BAMBOO_PAGE_SIZE);
+ void *origbound=(void *)((((unsigned INTPTR)origptr)&~(BAMBOO_PAGE_SIZE-1))+BAMBOO_PAGE_SIZE);
+
+ unsigned int topage=(toptr-1-gcbaseva)>>BAMBOO_PAGE_SIZE_BITS;
+ unsigned int origpage=(origptr-gcbaseva)>>BAMBOO_PAGE_SIZE_BITS;
+
+ unsigned int * totable=&gccachesamplingtbl_r[topage*NUMCORESACTIVE];
+ unsigned int * origtable=&gccachesamplingtbl[origpage*NUMCORESACTIVE];
+
+ //handler
+ unsigned int remaintobytes=(bytesneeded==0)?0:(tobound-toptr);
+ unsigned int remainorigbytes=origbound-origptr;
+
+ do {
+ //round source bytes down....don't want to close out page if not necessary
+ remainorigbytes=(remainorigbytes>bytesneeded)?bytesneeded:remainorigbytes;
+
+ if (remaintobytes<=remainorigbytes) {
+ //Need to close out to page
+
+ numbytes+=remaintobytes;
+ unsigned int page64th=numbytes>>(BAMBOO_PAGE_SIZE_BITS-6);
+
+ for(int core = 0; core < NUMCORESACTIVE; core++) {
+ (*totable)=(*totable+page64th*(*origtable))>>6;
+ totable++;
+ origtable++;
+ }
+ toptr+=remaintobytes;
+ origptr+=remaintobytes;
+ bytesneeded-=remaintobytes;
+ topage++;//to page is definitely done
+ tobound+=BAMBOO_PAGE_SIZE;
+ origpage=(origptr-gcbaseva)>>BAMBOO_PAGE_SIZE_BITS;//handle exact match case
+ origbound=(void *) ((((unsigned INTPTR)origptr)&~(BAMBOO_PAGE_SIZE-1))+BAMBOO_PAGE_SIZE);
+ } else {
+ //Finishing off orig page
+
+ numbytes+=remainorigbytes;
+ unsigned int page64th=numbytes>>(BAMBOO_PAGE_SIZE_BITS-6);
+
+ for(int core = 0; core < NUMCORESACTIVE; core++) {
+ (*totable)+=page64th*(*origtable);
+ totable++;
+ origtable++;
+ }
+ toptr+=remainorigbytes;
+ origptr+=remainorigbytes;
+ bytesneeded-=remainorigbytes;
+ origpage++;//just orig page is done
+ origbound+=BAMBOO_PAGE_SIZE;
+ }
+ totable=&gccachesamplingtbl_r[topage*NUMCORESACTIVE];
+ origtable=&gccachesamplingtbl[origpage*NUMCORESACTIVE];
+
+ remaintobytes=tobound-toptr;
+ remainorigbytes=origbound-origptr;
+
+ numbytes=0;
+ } while(bytesneeded!=0);
+}
// prepare for cache adaption:
// -- flush the shared heap
// clean the dtlb entries
BAMBOO_CLEAN_DTLB();
- // change the cache strategy
- gccachestage = isgccachestage;
+ if(isgccachestage) {
+ bamboo_install_dtlb_handler_for_gc();
+ } else {
+ bamboo_install_dtlb_handler_for_mutator();
+ }
}
// the master core decides how to adapt cache strategy for the mutator
// find the core that accesses the page #page_index most
#define CACHEADAPT_FIND_HOTTEST_CORE(page_index,hottestcore,hotfreq) \
{ \
- int *local_tbl=&gccachesamplingtbl_r[page_index]; \
+ unsigned int *local_tbl=&gccachesamplingtbl_r[page_index*NUMCORESACTIVE]; \
for(int i = 0; i < NUMCORESACTIVE; i++) { \
int freq = *local_tbl; \
- local_tbl=(int *)(((char *)local_tbl)+size_cachesamplingtbl_local_r); \
+ local_tbl++; \
if(hotfreq < freq) { \
hotfreq = freq; \
hottestcore = i; \
// access time of the page at the same time
#define CACHEADAPT_FIND_HOTTEST_CORE_W_TOTALFREQ(page_index,hottestcore,hotfreq,totalfreq) \
{ \
- int *local_tbl=&gccachesamplingtbl_r[page_index]; \
+ unsigned int *local_tbl=&gccachesamplingtbl_r[page_index*NUMCORESACTIVE]; \
for(int i = 0; i < NUMCORESACTIVE; i++) { \
int freq = *local_tbl; \
- local_tbl=(int *)(((char *)local_tbl)+size_cachesamplingtbl_local_r); \
+ local_tbl++; \
totalfreq += freq; \
if(hotfreq < freq) { \
hotfreq = freq; \
// make all pages hfh
void cacheAdapt_policy_h4h(int coren){
- unsigned int page_num=(BAMBOO_SHARED_MEM_SIZE)/(BAMBOO_PAGE_SIZE);
+ unsigned int page_num=(BAMBOO_SHARED_MEM_SIZE)>>(BAMBOO_PAGE_SIZE_BITS);
unsigned int page_gap=page_num/NUMCORESACTIVE;
unsigned int page_index=page_gap*coren;
unsigned int page_index_end=(coren==NUMCORESACTIVE-1)?page_num:(page_index+page_gap);
VA page_sva = gcbaseva+(BAMBOO_PAGE_SIZE)*page_index;
- int * tmp_p = gccachepolicytbl;
+ unsigned int * tmp_p = gccachepolicytbl;
for(; page_index < page_index_end; page_index++) {
bamboo_cache_policy_t policy = {0};
policy.cache_mode = BAMBOO_CACHE_MODE_HASH;
// make all pages local as non-cache-adaptable gc local mode
void cacheAdapt_policy_local(int coren){
- unsigned int page_num=(BAMBOO_SHARED_MEM_SIZE)/(BAMBOO_PAGE_SIZE);
+ unsigned int page_num=(BAMBOO_SHARED_MEM_SIZE)>>(BAMBOO_PAGE_SIZE_BITS);
unsigned int page_gap=page_num/NUMCORESACTIVE;
unsigned int page_index=page_gap*coren;
unsigned int page_index_end=(coren==NUMCORESACTIVE-1)?page_num:(page_index+page_gap);
VA page_sva = gcbaseva+(BAMBOO_PAGE_SIZE)*page_index;
- int * tmp_p = gccachepolicytbl;
+ unsigned int * tmp_p = gccachepolicytbl;
for(; page_index < page_index_end; page_index++) {
bamboo_cache_policy_t policy = {0};
unsigned int block = 0;
- BLOCKINDEX((void *) page_sva, block);
+ BLOCKINDEX(block, (void *) page_sva);
unsigned int coren = gc_block2core[block%(NUMCORES4GC*2)];
CACHEADAPT_POLICY_SET_HOST_CORE(policy, coren);
CACHEADAPT_CHANGE_POLICY_4_PAGE(tmp_p,page_index,policy);
}
void cacheAdapt_policy_hottest(int coren){
- unsigned int page_num=(BAMBOO_SHARED_MEM_SIZE)/(BAMBOO_PAGE_SIZE);
+ unsigned int page_num=(BAMBOO_SHARED_MEM_SIZE)>>(BAMBOO_PAGE_SIZE_BITS);
unsigned int page_gap=page_num/NUMCORESACTIVE;
unsigned int page_index=page_gap*coren;
unsigned int page_index_end=(coren==NUMCORESACTIVE-1)?page_num:(page_index+page_gap);
VA page_sva = gcbaseva+(BAMBOO_PAGE_SIZE)*page_index;
- int * tmp_p = gccachepolicytbl;
+ unsigned int * tmp_p = gccachepolicytbl;
for(; page_index < page_index_end; page_index++) {
bamboo_cache_policy_t policy = {0};
unsigned int hottestcore = 0;
}
}
-#define GC_CACHE_ADAPT_DOMINATE_THRESHOLD 64
+#define GC_CACHE_ADAPT_DOMINATE_THRESHOLD 1
// cache the page on the core that accesses it the most if that core accesses
// it more than (GC_CACHE_ADAPT_DOMINATE_THRESHOLD)% of the total. Otherwise,
// h4h the page.
void cacheAdapt_policy_dominate(int coren){
- unsigned int page_num=(BAMBOO_SHARED_MEM_SIZE)/(BAMBOO_PAGE_SIZE);
+ unsigned int page_num=(BAMBOO_SHARED_MEM_SIZE)>>(BAMBOO_PAGE_SIZE_BITS);
unsigned int page_gap=page_num/NUMCORESACTIVE;
unsigned int page_index=page_gap*coren;
unsigned int page_index_end=(coren==NUMCORESACTIVE-1)?page_num:(page_index+page_gap);
VA page_sva = gcbaseva+(BAMBOO_PAGE_SIZE)*page_index;
- int * tmp_p = gccachepolicytbl;
+ unsigned int * tmp_p = gccachepolicytbl;
for(; page_index < page_index_end; page_index++) {
bamboo_cache_policy_t policy = {0};
unsigned int hottestcore = 0;
- unsigned long long totalfreq = 0;
+ unsigned int totalfreq = 0;
unsigned int hotfreq = 0;
CACHEADAPT_FIND_HOTTEST_CORE_W_TOTALFREQ(page_index,hottestcore,hotfreq,totalfreq);
// Decide the cache strategy for this page
// the gcpolicytbl
// Format: page start va + cache policy
if(hotfreq != 0) {
- totalfreq=(totalfreq*GC_CACHE_ADAPT_DOMINATE_THRESHOLD)>>7;
- if(hotfreq < totalfreq) {
+ totalfreq=totalfreq>>GC_CACHE_ADAPT_DOMINATE_THRESHOLD;
+ if((unsigned int)hotfreq < (unsigned int)totalfreq) {
// use hfh
policy.cache_mode = BAMBOO_CACHE_MODE_HASH;
+ /*unsigned int block = 0;
+ BLOCKINDEX(block, (void *) page_sva);
+ unsigned int coren = gc_block2core[block%(NUMCORES4GC*2)];
+ CACHEADAPT_POLICY_SET_HOST_CORE(policy, coren);*/
} else {
// locally cache the page in the hottest core
CACHEADAPT_POLICY_SET_HOST_CORE(policy, hottestcore);
}
}
-#if 0
-#define GC_CACHE_ADAPT_OVERLOAD_THRESHOLD 10
-// record the worklocad of the hottestcore into core2heavypages
-#define CACHEADAPT_RECORD_PAGE_WORKLOAD(hottestcore,totalfreq,hotfreq,remoteaccess,tmp_p) \
- { \
- workload[hottestcore] += (totalfreq); \
- total_workload += (totalfreq); \
- unsigned long long remoteaccess = (totalfreq) - (hotfreq); \
- unsigned int index = (unsigned int)core2heavypages[hottestcore][0]; \
- core2heavypages[hottestcore][3*index+3] = (remoteaccess); \
- core2heavypages[hottestcore][3*index+2] = (totalfreq); \
- core2heavypages[hottestcore][3*index+1] = (unsigned long long)((tmp_p)-1); \
- core2heavypages[hottestcore][0]++; \
- }
-
-void gc_quicksort(unsigned long long *array,unsigned int left,unsigned int right,unsigned int offset) {
- unsigned int pivot = 0;;
- unsigned int leftIdx = left;
- unsigned int rightIdx = right;
- if((right-left+1) >= 1) {
- pivot = (left+right)/2;
- while((leftIdx <= pivot) && (rightIdx >= pivot)) {
- unsigned long long pivotValue = array[pivot*3-offset];
- while((array[leftIdx*3-offset] > pivotValue) && (leftIdx <= pivot)) {
- leftIdx++;
- }
- while((array[rightIdx*3-offset] < pivotValue) && (rightIdx >= pivot)) {
- rightIdx--;
- }
- // swap [leftIdx] & [rightIdx]
- for(int k = 0; k < 3; k++) {
- unsigned long long tmp = array[3*rightIdx-k];
- array[3*rightIdx-k] = array[3*leftIdx-k];
- array[3*leftIdx-k] = tmp;
- }
- leftIdx++;
- rightIdx--;
- if((leftIdx-1) == pivot) {
- pivot = rightIdx = rightIdx + 1;
- } else if((leftIdx+1) == pivot) {
- pivot = leftIdx = leftIdx-1;
- }
- }
- gc_quicksort(array, left, pivot-1, offset);
- gc_quicksort(array, pivot+1, right, offset);
- }
- return;
-}
-
-INLINE int cacheAdapt_h4h_remote_accesses(unsigned long long workload_threshold,unsigned long long ** core2heavypages, unsigned long long * workload,int i) {
- int j = 1;
- unsigned int index = (unsigned int)core2heavypages[i][0];
- if(workload[i] > workload_threshold) {
- // sort according to the remoteaccess
- gc_quicksort(&core2heavypages[i][0], 1, index, 0);
- while((workload[i] > workload_threshold) && (j<index*3)) {
- // hfh those pages with more remote accesses
- bamboo_cache_policy_t policy = {0};
- policy.cache_mode = BAMBOO_CACHE_MODE_HASH;
- *((unsigned int*)core2heavypages[i][j]) = policy.word;
- workload[i] -= core2heavypages[i][j+1];
- j += 3;
- }
- }
- return j;
-}
-
-// Every page cached on the core that accesses it the most.
-// Check to see if any core's pages total more accesses than threshold
-// GC_CACHE_ADAPT_OVERLOAD_THRESHOLD. If so, find the pages with the
-// most remote accesses and hash for home them until we get below
-// GC_CACHE_ADAPT_OVERLOAD_THRESHOLD
-int cacheAdapt_policy_overload(int coren){
- unsigned int page_index = 0;
- VA page_sva = gcbaseva;
- unsigned int page_num = BAMBOO_SHARED_MEM_SIZE/BAMBOO_PAGE_SIZE;
- unsigned int numchanged = 0;
- int * tmp_p = gccachepolicytbl+1;
- unsigned long long workload[NUMCORESACTIVE];
- memset(workload, 0, NUMCORESACTIVE*sizeof(unsigned long long));
- unsigned long long total_workload = 0;
- unsigned long long core2heavypages[NUMCORESACTIVE][page_num*3+1];
- memset(core2heavypages,0,sizeof(unsigned long long)*(page_num*3+1)*NUMCORESACTIVE);
- for(page_index = 0; page_sva < gctopva; page_index++) {
- bamboo_cache_policy_t policy = {0};
- unsigned int hottestcore = 0;
- unsigned long long totalfreq = 0;
- unsigned int hotfreq = 0;
- CACHEADAPT_FIND_HOTTEST_CORE_W_TOTALFREQ(page_index,hottestcore,hotfreq,totalfreq);
- // Decide the cache strategy for this page
- // If decide to adapt a new cache strategy, write into the shared block of
- // the gcsharedsamplingtbl. The mem recording information that has been
- // written is enough to hold the information.
- // Format: page start va + cache strategy(hfh/(host core+[x,y]))
- if(hotfreq != 0) {
- totalfreq/=BAMBOO_PAGE_SIZE;
- hotfreq/=BAMBOO_PAGE_SIZE;
- // locally cache the page in the hottest core
- CACHEADAPT_POLICY_SET_HOST_CORE(policy, hottestcore);
- CACHEADAPT_CHANGE_POLICY_4_PAGE(tmp_p,page_index,policy,numchanged);
- CACHEADAPT_RECORD_PAGE_WORKLOAD(hottestcore,totalfreq,hotfreq,remoteaccess,tmp_p);
- }
- page_sva += BAMBOO_PAGE_SIZE;
- }
-
- unsigned long long workload_threshold=total_workload/GC_CACHE_ADAPT_OVERLOAD_THRESHOLD;
- // Check the workload of each core
- for(int i = 0; i < NUMCORESACTIVE; i++) {
- cacheAdapt_h4h_remote_accesses(workload_threshold,core2heavypages,workload,i);
- }
-
- return numchanged;
-}
-
-#define GC_CACHE_ADAPT_ACCESS_THRESHOLD 70
-#define GC_CACHE_ADAPT_CROWD_THRESHOLD 20
-// Every page cached on the core that accesses it the most.
-// Check to see if any core's pages total more accesses than threshold
-// GC_CACHE_ADAPT_OVERLOAD_THRESHOLD. If so, find the pages with the
-// most remote accesses and hash for home them until we get below
-// GC_CACHE_ADAPT_OVERLOAD_THRESHOLD.
-// Sort pages based on activity....
-// If more then GC_CACHE_ADAPT_ACCESS_THRESHOLD% of the accesses for a
-// core's pages are from more than GC_CACHE_ADAPT_CROWD_THRESHOLD pages,
-// then start hfh these pages(selecting the ones with the most remote
-// accesses first or fewest local accesses) until we get below
-// GC_CACHE_ADAPT_CROWD_THRESHOLD pages.
-int cacheAdapt_policy_crowd(int coren){
- unsigned int page_index = 0;
- VA page_sva = gcbaseva;
- unsigned int page_num = BAMBOO_SHARED_MEM_SIZE/BAMBOO_PAGE_SIZE;
- unsigned int numchanged = 0;
- int * tmp_p = gccachepolicytbl+1;
- unsigned long long workload[NUMCORESACTIVE];
- memset(workload, 0, NUMCORESACTIVE*sizeof(unsigned long long));
- unsigned long long total_workload = 0;
- unsigned long long core2heavypages[NUMCORESACTIVE][page_num*3+1];
- memset(core2heavypages,0,sizeof(unsigned long long)*(page_num*3+1)*NUMCORESACTIVE);
- for(page_index = 0; page_sva < gctopva; page_index++) {
- bamboo_cache_policy_t policy = {0};
- unsigned int hottestcore = 0;
- unsigned long long totalfreq = 0;
- unsigned int hotfreq = 0;
- CACHEADAPT_FIND_HOTTEST_CORE_W_TOTALFREQ(page_index,hottestcore,hotfreq,totalfreq);
- // Decide the cache strategy for this page
- // If decide to adapt a new cache strategy, write into the shared block of
- // the gcsharedsamplingtbl. The mem recording information that has been
- // written is enough to hold the information.
- // Format: page start va + cache strategy(hfh/(host core+[x,y]))
- if(hotfreq != 0) {
- totalfreq/=BAMBOO_PAGE_SIZE;
- hotfreq/=BAMBOO_PAGE_SIZE;
- // locally cache the page in the hottest core
- CACHEADAPT_POLICY_SET_HOST_CORE(policy, hottestcore);
- CACHEADAPT_CHANGE_POLICY_4_PAGE(tmp_p,page_index,policy,numchanged);
- CACHEADAPT_RECORD_PAGE_WORKLOAD(hottestcore,totalfreq,hotfreq,remoteaccess,tmp_p);
- }
- page_sva += BAMBOO_PAGE_SIZE;
- }
-
- unsigned long long workload_threshold=total_workload/GC_CACHE_ADAPT_OVERLOAD_THRESHOLD;
- // Check the workload of each core
- for(int i = 0; i < NUMCORESACTIVE; i++) {
- unsigned int index=(unsigned int)core2heavypages[i][0];
- int j=cacheAdapt_h4h_remote_accesses(workload_threshold,core2heavypages,workload,i);
- // Check if the accesses are crowded on few pages
- // sort according to the total access
-inner_crowd:
- gc_quicksort(&core2heavypages[i][0], j/3+1, index, 1);
- unsigned long long threshold=GC_CACHE_ADAPT_ACCESS_THRESHOLD*workload[i]/100;
- int num_crowded = 0;
- unsigned long long t_workload = 0;
- do {
- t_workload += core2heavypages[i][j+num_crowded*3+1];
- num_crowded++;
- } while(t_workload < threshold);
- // num_crowded <= GC_CACHE_ADAPT_CROWD_THRESHOLD and if there are enough
- // items, it is always == GC_CACHE_ADAPT_CROWD_THRESHOLD
- if(num_crowded > GC_CACHE_ADAPT_CROWD_THRESHOLD) {
- // need to hfh these pages
- // sort the pages according to remote access
- gc_quicksort(&core2heavypages[i][0], j/3+1, j/3+num_crowded, 0);
- // h4h those pages with more remote accesses
- bamboo_cache_policy_t policy = {0};
- policy.cache_mode = BAMBOO_CACHE_MODE_HASH;
- *((unsigned int*)core2heavypages[i][j]) = policy.word;
- workload[i] -= core2heavypages[i][j+1];
- t_workload -= core2heavypages[i][j+1];
- j += 3;
- threshold = GC_CACHE_ADAPT_ACCESS_THRESHOLD*workload[i]/100;
- goto inner_crowd;
- }
- }
-
- return numchanged;
-}
-#endif
-
unsigned int cacheAdapt_decision(int coren) {
BAMBOO_CACHE_MF();
// check the statistic data
cacheAdapt_policy_hottest(coren);
#elif defined GC_CACHE_ADAPT_POLICY4
cacheAdapt_policy_dominate(coren);
-//#elif defined GC_CACHE_ADAPT_POLICY5
-// cacheAdapt_policy_overload(coren);
-//#elif defined GC_CACHE_ADAPT_POLICY6
-// cacheAdapt_policy_crowd(coren);
#endif
}
void cacheAdapt_mutator() {
BAMBOO_CACHE_MF();
// check the changes and adapt them
- int * tmp_p = gccachepolicytbl;
+ unsigned int * tmp_p = gccachepolicytbl;
unsigned int page_sva = gcbaseva;
for(; page_sva<gctopva; page_sva+=BAMBOO_PAGE_SIZE) {
// read out the policy
}
}
+// Cache adapt phase process for clients
void cacheAdapt_phase_client() {
WAITFORGCPHASE(CACHEPOLICYPHASE);
GC_PRINTF("Start cachepolicy phase\n");
//send init finish msg to core coordinator
send_msg_2(STARTUPCORE, GCFINISHPREF, BAMBOO_NUM_OF_CORE);
GC_PRINTF("Finish prefinish phase\n");
- CACHEADAPT_SAMPING_RESET();
+ CACHEADAPT_SAMPLING_RESET();
if(BAMBOO_NUM_OF_CORE < NUMCORESACTIVE) {
// zero out the gccachesamplingtbl
BAMBOO_MEMSET_WH(gccachesamplingtbl_local,0,size_cachesamplingtbl_local);
extern unsigned long long gc_output_cache_policy_time;
+// Cache adpat phase process for the master
void cacheAdapt_phase_master() {
GCPROFILE_ITEM();
unsigned long long tmpt = BAMBOO_GET_EXE_TIME();
GC_PRINTF("Start cachepolicy phase \n");
// cache adapt phase
cacheAdapt_decision(BAMBOO_NUM_OF_CORE);
- GC_CHECK_ALL_CORE_STATUS(CACHEPOLICYPHASE==gc_status_info.gcphase);
+ GC_CHECK_ALL_CORE_STATUS();
BAMBOO_CACHE_MF();
// let all cores to adopt new policies
// cache adapt phase
cacheAdapt_mutator();
cacheAdapt_gc(false);
- GC_CHECK_ALL_CORE_STATUS(PREFINISHPHASE==gc_status_info.gcphase);
-
- CACHEADAPT_SAMPING_RESET();
+ GC_CHECK_ALL_CORE_STATUS();
+
+ CACHEADAPT_SAMPLING_RESET();
if(BAMBOO_NUM_OF_CORE < NUMCORESACTIVE) {
// zero out the gccachesamplingtbl
BAMBOO_MEMSET_WH(gccachesamplingtbl_local,0,size_cachesamplingtbl_local);
}
}
+// output original cache sampling data for each page
void gc_output_cache_sampling() {
+ extern volatile bool gc_profile_flag;
+ if(!gc_profile_flag) return;
unsigned int page_index = 0;
VA page_sva = 0;
- unsigned int page_num = (BAMBOO_SHARED_MEM_SIZE) / (BAMBOO_PAGE_SIZE);
+ unsigned int page_num = (BAMBOO_SHARED_MEM_SIZE) >> (BAMBOO_PAGE_SIZE_BITS);
for(page_index = 0; page_index < page_num; page_index++) {
page_sva = gcbaseva + (BAMBOO_PAGE_SIZE) * page_index;
unsigned int block = 0;
- BLOCKINDEX((void *) page_sva, block);
+ BLOCKINDEX(block, (void *) page_sva);
unsigned int coren = gc_block2core[block%(NUMCORES4GC*2)];
- printf("%x, %d, %d, ",(int)page_sva,page_index,coren);
+ //printf("%x, %d, %d, ",(int)page_sva,page_index,coren);
+ unsigned int * local_tbl = &gccachesamplingtbl[page_index*NUMCORESACTIVE];
+ int accesscore = 0;
for(int i = 0; i < NUMCORESACTIVE; i++) {
- int * local_tbl = (int *)((void *)gccachesamplingtbl+size_cachesamplingtbl_local*i);
- int freq = local_tbl[page_index];
- //if(freq != 0) {
- printf("%d, ", freq);
- //}
+ int freq = *local_tbl;
+ local_tbl++;
+ if(freq != 0) {
+ accesscore++;
+ //printf("%d, ", freq);
+ }
}
- printf("\n");
+ if(accesscore!=0) {
+ printf("%x, %d, %d, ",(int)page_sva,page_index,coren);
+ unsigned int * local_tbl = &gccachesamplingtbl[page_index*NUMCORESACTIVE];
+ for(int i = 0; i < NUMCORESACTIVE; i++) {
+ unsigned int freq = *local_tbl;
+ local_tbl++;
+ printf("%u, ", freq);
+ }
+ printf("\n");
+ }
+ //printf("\n");
}
printf("=================\n");
}
+// output revised cache sampling data for each page after compaction
void gc_output_cache_sampling_r() {
+ extern volatile bool gc_profile_flag;
+ if(!gc_profile_flag) return;
// TODO summary data
unsigned int sumdata[NUMCORESACTIVE][NUMCORESACTIVE];
for(int i = 0; i < NUMCORESACTIVE; i++) {
tprintf("cache sampling_r \n");
unsigned int page_index = 0;
VA page_sva = 0;
- unsigned int page_num = (BAMBOO_SHARED_MEM_SIZE) / (BAMBOO_PAGE_SIZE);
+ unsigned int page_num = (BAMBOO_SHARED_MEM_SIZE) >> (BAMBOO_PAGE_SIZE_BITS);
for(page_index = 0; page_index < page_num; page_index++) {
page_sva = gcbaseva + (BAMBOO_PAGE_SIZE) * page_index;
unsigned int block = 0;
- BLOCKINDEX((void *)page_sva, block);
+ BLOCKINDEX(block, (void *)page_sva);
unsigned int coren = gc_block2core[block%(NUMCORES4GC*2)];
- printf(" %x, %d, %d, ",(int)page_sva,page_index,coren);
+ //printf("%x, %d, %d, ",(int)page_sva,page_index,coren);
int accesscore = 0; // TODO
+ unsigned int * local_tbl = &gccachesamplingtbl_r[page_index*NUMCORESACTIVE];
for(int i = 0; i < NUMCORESACTIVE; i++) {
- int * local_tbl = (int *)((void *)gccachesamplingtbl_r+size_cachesamplingtbl_local_r*i);
- int freq = local_tbl[page_index]/BAMBOO_PAGE_SIZE;
- printf("%d, ", freq);
+ unsigned int freq = *local_tbl;
+ //printf("%d, ", freq);
if(freq != 0) {
accesscore++;// TODO
}
+ local_tbl++;
}
if(accesscore!=0) {
+ printf("%x, %d, %d, ",(int)page_sva,page_index,coren);
+ unsigned int * local_tbl = &gccachesamplingtbl_r[page_index*NUMCORESACTIVE];
for(int i = 0; i < NUMCORESACTIVE; i++) {
- int * local_tbl = (int *)((void *)gccachesamplingtbl_r+size_cachesamplingtbl_local_r*i);
- int freq = local_tbl[page_index]/BAMBOO_PAGE_SIZE;
+ unsigned int freq = *local_tbl;
+ printf("%u, ", freq);
sumdata[accesscore-1][i]+=freq;
+ local_tbl++;
}
- }
-
- printf("\n");
+ printf("\n");
+ }
+ //printf("\n");
}
+ printf("+++++\n");
// TODO printout the summary data
for(int i = 0; i < NUMCORESACTIVE; i++) {
printf("%d ", i);
for(int j = 0; j < NUMCORESACTIVE; j++) {
- printf(" %d ", sumdata[j][i]);
+ printf(" %u ", sumdata[j][i]);
}
printf("\n");
}