3 #include "multicoreruntime.h"
4 #include "methodheaders.h"
5 #include "multicoregarbage.h"
7 #include "multicoregcprofile.h"
8 #include "multicoregc.h"
9 #include "pmc_garbage.h"
11 #include "multicore_arch.h"
14 #include "bme_perf_counter.h"
20 extern int classsize[];
21 extern int typearray[];
22 extern int typearray2[];
23 extern int* supertypes[];
26 extern struct genhashtable * activetasks;
33 int instanceofif(int otype, int type) {
40 int num = supertypes[otype][0];
41 for(int i = 1; i < num + 1; i++) {
42 int t = supertypes[otype][i];
43 if(instanceofif(t, type) == 1) {
50 int instanceof(struct ___Object___ *ptr, int type) {
55 if(instanceofif(i, type) == 1) {
62 i=typearray2[i-NUMCLASSES];
68 void initializeexithandler() {
71 /* This function inject failures */
72 void injectinstructionfailure() {
73 // not supported in MULTICORE version
77 #ifdef D___Double______nativeparsedouble____L___String___
78 double CALL01(___Double______nativeparsedouble____L___String___,
79 struct ___String___ * ___str___) {
80 int length=VAR(___str___)->___count___;
81 int maxlength=(length>60) ? 60 : length;
82 char str[maxlength+1];
83 struct ArrayObject * chararray=VAR(___str___)->___value___;
85 int offset=VAR(___str___)->___offset___;
86 for(i=0; i<maxlength; i++) {
88 ((short *)(((char *)&chararray->___length___)+sizeof(int)))[i+offset];
91 double d=0.0; //atof(str); TODO Unimplemented nativeparsedoulbe
96 #ifdef D___Double______nativeparsedouble_____AR_B_I_I
97 double CALL23(___Double______nativeparsedouble_____AR_B_I_I,
102 struct ArrayObject * ___str___) {
103 int maxlength=(length>60)?60:length;
104 char str[maxlength+1];
105 struct ArrayObject * bytearray=VAR(___str___);
107 for(i=0; i<maxlength; i++) {
108 str[i]=(((char *)&bytearray->___length___)+sizeof(int))[i+start];
111 double d=0.0; //atof(str); TODO Unimplemented nativeparsedouble
116 typedef union jvalue {
126 #ifdef D___Double______doubleToRawLongBits____D
127 long long CALL11(___Double______doubleToRawLongBits____D,
129 double ___value___) {
133 #if defined(__IEEE_BYTES_LITTLE_ENDIAN)
134 /* On little endian ARM processors when using FPA, word order of
135 doubles is still big endian. So take that into account here. When
136 using VFP, word order of doubles follows byte order. */
137 #define SWAP_DOUBLE(a) (((a) << 32) | (((a) >> 32) & 0x00000000ffffffff))
138 val.j = SWAP_DOUBLE(val.j);
145 #ifdef D___Double______longBitsToDouble____J
146 double CALL11(___Double______longBitsToDouble____J,
147 long long ___bits___,
148 long long ___bits___) {
152 #if defined(__IEEE_BYTES_LITTLE_ENDIAN)
154 #define SWAP_DOUBLE(a) (((a) << 32) | (((a) >> 32) & 0x00000000ffffffff))
156 val.j = SWAP_DOUBLE(val.j);
163 #ifdef D___String______convertdoubletochar____D__AR_C
164 int CALL12(___String______convertdoubletochar____D__AR_C,
167 struct ArrayObject * ___chararray___) {
168 int length=VAR(___chararray___)->___length___;
171 int num=snprintf(str, length, "%f",___val___);
174 for(i=0; i<length; i++) {
175 ((short *)(((char *)&VAR(___chararray___)->___length___)+sizeof(int)))[i]=
182 #ifdef D___System______deepArrayCopy____L___Object____L___Object___
183 void deepArrayCopy(struct ___Object___ * dst,
184 struct ___Object___ * src) {
185 int dsttype=((int *)dst)[0];
186 int srctype=((int *)src)[0];
187 if (dsttype<NUMCLASSES||srctype<NUMCLASSES||srctype!=dsttype)
189 struct ArrayObject *aodst=(struct ArrayObject *)dst;
190 struct ArrayObject *aosrc=(struct ArrayObject *)src;
191 int dstlength=aodst->___length___;
192 int srclength=aosrc->___length___;
193 if (dstlength!=srclength)
195 unsigned INTPTR *pointer=pointerarray[srctype];
197 int elementsize=classsize[srctype];
198 int size=srclength*elementsize;
200 memcpy(((char *)&aodst->___length___)+sizeof(int) ,
201 ((char *)&aosrc->___length___)+sizeof(int), size);
205 for(i=0;i<srclength;i++) {
206 struct ___Object___ * ptr=
207 ((struct ___Object___**)(((char*)&aosrc->___length___)+sizeof(int)))[i];
208 int ptrtype=((int *)ptr)[0];
209 if (ptrtype>=NUMCLASSES) {
210 struct ___Object___ * dstptr=((struct ___Object___**)
211 (((char*)&aodst->___length___)+sizeof(int)))[i];
212 deepArrayCopy(dstptr,ptr);
215 ((struct ___Object___ **)
216 (((char*) &aodst->___length___)+sizeof(int)))[i]=ptr;
222 void CALL02(___System______deepArrayCopy____L___Object____L___Object___,
223 struct ___Object___ * ___dst___,
224 struct ___Object___ * ___src___) {
225 deepArrayCopy(VAR(___dst___), VAR(___src___));
229 #ifdef D___System______arraycopy____L___Object____I_L___Object____I_I
230 void arraycopy(struct ___Object___ *src,
232 struct ___Object___ *dst,
235 int dsttype=((int *)dst)[0];
236 int srctype=((int *)src)[0];
238 //not an array or type mismatch
239 if (dsttype<NUMCLASSES||srctype<NUMCLASSES/*||srctype!=dsttype*/)
242 struct ArrayObject *aodst=(struct ArrayObject *)dst;
243 struct ArrayObject *aosrc=(struct ArrayObject *)src;
244 int dstlength=aodst->___length___;
245 int srclength=aosrc->___length___;
249 if (srcPos+length>srclength)
251 if (destPos+length>dstlength)
254 unsigned INTPTR *pointer=pointerarray[srctype];
256 int elementsize=classsize[srctype];
257 int size=length*elementsize;
259 memcpy(((char *)&aodst->___length___)+sizeof(int)+destPos*elementsize,
260 ((char *)&aosrc->___length___)+sizeof(int)+srcPos*elementsize, size);
264 for(i=0;i<length;i++) {
265 struct ___Object___ * ptr=((struct ___Object___**)
266 (((char*)&aosrc->___length___)+sizeof(int)))[i+srcPos];
267 int ptrtype=((int *)ptr)[0];
269 ((struct ___Object___ **)
270 (((char*) &aodst->___length___)+sizeof(int)))[i+destPos]=ptr;
275 void CALL35(___System______arraycopy____L___Object____I_L___Object____I_I,
279 struct ___Object___ * ___src___,
281 struct ___Object___ * ___dst___,
284 arraycopy(VAR(___src___), ___srcPos___, VAR(___dst___), ___destPos___,
289 #ifdef D___System______exit____I
290 void CALL11(___System______exit____I,
293 // gc_profile mode, output gc prfiling data
294 #if defined(MULTICORE_GC)||defined(PMC_GC)
295 if(STARTUPCORE == BAMBOO_NUM_OF_CORE) {
296 BAMBOO_PRINT(BAMBOO_GET_EXE_TIME());
297 BAMBOO_PRINT(0xbbbbbbbb);
298 CACHEADAPT_DISABLE_TIMER();
299 GC_OUTPUT_PROFILE_DATA();
303 gc_outputProfileDataReadable();
304 tprintf("FINISH_EXECUTION\n");
307 BAMBOO_EXIT_APP(___status___);
311 #ifdef D___Vector______removeElement_____AR_L___Object____I_I
312 void CALL23(___Vector______removeElement_____AR_L___Object____I_I,
315 struct ArrayObject * ___array___,
318 char* offset=((char *)(&VAR(___array___)->___length___))
319 +sizeof(unsigned int)+sizeof(void *)*___index___;
320 memmove(offset, offset+sizeof(void *),
321 (___size___-___index___-1)*sizeof(void *));
325 #ifdef D___System______printI____I
326 void CALL11(___System______printI____I,
329 BAMBOO_PRINT(0x1111);
330 BAMBOO_PRINT_REG(___status___);
334 #ifdef D___System______currentTimeMillis____
335 long long CALL00(___System______currentTimeMillis____) {
336 //TilePro64 is 700mHz
337 return ((unsigned long long)BAMBOO_GET_EXE_TIME())/700000;
341 #ifdef D___System______numGCs____
342 long long ___System______numGCs____(struct ___System______numGCs_____params * ___params___) {
351 #ifdef D___System______milliGcTime____
352 long long ___System______milliGcTime____(struct ___System______milliGcTime_____params * ___params___) {
354 return GCtime/700000;
361 #ifdef D___System______nanoTime____
362 long long CALL00(___System______nanoTime____) {
363 //TilePro64 is 700mHz
364 return ((unsigned long long)BAMBOO_GET_EXE_TIME())/700;
368 #ifdef D___System______setgcprofileflag____
369 void CALL00(___System______setgcprofileflag____) {
372 extern volatile bool gc_profile_flag;
373 gc_profile_flag = true;
379 #ifdef D___System______resetgcprofileflag____
380 void CALL00(___System______resetgcprofileflag____) {
383 extern volatile bool gc_profile_flag;
384 gc_profile_flag = false;
390 #ifdef D___System______gc____
391 void CALL00(___System______gc____) {
393 if(BAMBOO_NUM_OF_CORE == STARTUPCORE) {
394 if(!gc_status_info.gcprocessing && !gcflag) {
397 for(int i = 0; i < NUMCORESACTIVE; i++) {
398 // reuse the gcnumsendobjs & gcnumreceiveobjs
399 gcnumsendobjs[0][i] = 0;
400 gcnumreceiveobjs[0][i] = 0;
402 for(int i = 0; i < NUMCORES4GC; i++) {
403 if(i != STARTUPCORE) {
404 send_msg_1(i,GCSTARTPRE);
409 // send msg to the startup core to start gc
410 send_msg_1(STARTUPCORE, GCINVOKE);
416 #ifdef D___System______printString____L___String___
417 void CALL01(___System______printString____L___String___, struct ___String___ * ___s___) {
418 #if defined(MGC)&&defined(TILERA_BME)
419 struct ArrayObject * chararray=VAR(___s___)->___value___;
421 int offset=VAR(___s___)->___offset___;
423 for(i=0; i<VAR(___s___)->___count___; i++) {
425 ((short *)(((char *)&chararray->___length___)+sizeof(int)))[i+offset];
432 /* Object allocation function */
434 #if defined(MULTICORE_GC)||defined(PMC_GC)
435 void * allocate_new(void * ptr,
437 struct ___Object___ * v=
438 (struct ___Object___*)FREEMALLOC((struct garbagelist*) ptr,classsize[type]);
449 /* Array allocation function */
451 struct ArrayObject * allocate_newarray(void * ptr,
454 struct ArrayObject * v=(struct ArrayObject *)FREEMALLOC(
455 (struct garbagelist*)ptr,
456 sizeof(struct ArrayObject)+length*classsize[type]);
465 v->___length___=length;
466 initlock((struct ___Object___ *)v);
471 void * allocate_new(int type) {
472 struct ___Object___ * v=FREEMALLOC(classsize[type]);
482 /* Array allocation function */
484 struct ArrayObject * allocate_newarray(int type,
486 struct ArrayObject * v=FREEMALLOC(
487 sizeof(struct ArrayObject)+length*classsize[type]);
493 v->___length___=length;
494 initlock((struct ___Object___ *) v);
499 /* Converts C character arrays into Java strings */
500 #if defined(MULTICORE_GC)||defined(PMC_GC)
501 __attribute__((malloc)) struct ___String___ * NewStringShort(void * ptr,
505 __attribute__((malloc)) struct ___String___ * NewStringShort(const short *str,
509 #if defined(MULTICORE_GC)||defined(PMC_GC)
510 struct ArrayObject * chararray=
511 allocate_newarray((struct garbagelist *)ptr, CHARARRAYTYPE, length);
512 INTPTR ptrarray[]={1, (INTPTR) ptr, (INTPTR) chararray};
513 struct ___String___ * strobj=
514 allocate_new((struct garbagelist *) &ptrarray, STRINGTYPE);
515 chararray=(struct ArrayObject *) ptrarray[2];
517 struct ArrayObject * chararray=allocate_newarray(CHARARRAYTYPE, length);
518 struct ___String___ * strobj=allocate_new(STRINGTYPE);
520 strobj->___value___=chararray;
521 strobj->___count___=length;
522 strobj->___offset___=0;
524 for(i=0; i<length; i++) {
525 ((short *)(((char *)&chararray->___length___)+sizeof(int)))[i]=str[i];
530 /* Converts C character arrays into Java strings */
531 #if defined(MULTICORE_GC)||defined(PMC_GC)
532 struct ___String___ * NewString(void * ptr,
536 struct ___String___ * NewString(const char *str,
540 #if defined(MULTICORE_GC)||defined(PMC_GC)
541 struct ArrayObject * chararray=
542 allocate_newarray((struct garbagelist *)ptr, CHARARRAYTYPE, length);
543 int ptrarray[]={1, (int) ptr, (int) chararray};
544 struct ___String___ * strobj=
545 allocate_new((struct garbagelist *) &ptrarray, STRINGTYPE);
546 chararray=(struct ArrayObject *) ptrarray[2];
548 struct ArrayObject * chararray=allocate_newarray(CHARARRAYTYPE, length);
549 struct ___String___ * strobj=allocate_new(STRINGTYPE);
551 strobj->___value___=chararray;
552 strobj->___count___=length;
553 strobj->___offset___=0;
555 for(i=0; i<length; i++) {
556 ((short*)(((char*)&chararray->___length___)+sizeof(int)))[i]=(short)str[i];
561 /* Generated code calls this if we fail a bounds check */
563 void failedboundschk(int num, int index, struct ArrayObject * ao) {
565 printf("Array out of bounds at line %u with index %u of object %x with lengt\
566 h %u\n", num, index, ao, ao->___length___);
576 printf("Array out of bounds\n");
577 longjmp(error_handler,2);
584 /* Generated code calls this if we fail null ptr chk */
585 void failednullptr(void * ptr) {
586 #if defined(MULTICORE_GC)||defined(PMC_GC)
588 //print out current stack
591 struct garbagelist * stackptr = (struct garbagelist *)ptr;
592 while(stackptr!=NULL) {
593 tprintf("Stack %d: \n\t", j);
594 for(i=0; i<stackptr->size; i++) {
595 if(stackptr->array[i] != NULL) {
596 tprintf("%x, ", stackptr->array[i]);
602 stackptr=stackptr->next;
607 printf("NULL ptr\n");
617 printf("NULL ptr\n");
618 longjmp(error_handler,2);
625 /* Abort task call */
629 printf("Aborting\n");
630 longjmp(error_handler,4);
633 printf("Aborting\n");
638 void initruntimedata() {
639 // initialize the arrays
640 if(STARTUPCORE == BAMBOO_NUM_OF_CORE) {
641 // startup core to initialize corestatus[]
642 for(int i = 0; i < NUMCORESACTIVE; ++i) {
645 numreceiveobjs[i] = 0;
652 self_numsendobjs = 0;
653 self_numreceiveobjs = 0;
655 for(int i = 0; i < BAMBOO_MSG_BUF_LENGTH; ++i) {
660 //msglength = BAMBOO_MSG_BUF_LENGTH;
662 for(int i = 0; i < BAMBOO_OUT_BUF_LENGTH; ++i) {
668 isMsgHanging = false;
671 bamboo_cur_msp = NULL;
672 bamboo_smem_size = 0;
677 INITMULTICOREGCDATA();
681 bamboo_current_thread = NULL;
687 void disruntimedata() {
688 DISMULTICOREGCDATA();
690 BAMBOO_LOCAL_MEM_CLOSE();
691 BAMBOO_SHARE_MEM_CLOSE();
694 void recordtotalexetime() {
696 totalexetime = BAMBOO_GET_EXE_TIME()-bamboo_start_time;
698 unsigned long long timediff=BAMBOO_GET_EXE_TIME()-bamboo_start_time;
699 BAMBOO_PRINT(timediff);
700 #ifndef BAMBOO_MEMPROF
701 BAMBOO_PRINT(0xbbbbbbbb);
706 void getprofiledata_I() {
707 //profile mode, send msgs to other cores to request pouring out progiling data
709 // use numconfirm to check if all cores have finished output task profiling
710 // information. This is safe as when the execution reaches this phase there
711 // should have no other msgs except the PROFILEFINISH msg, there should be
713 numconfirm=NUMCORESACTIVE-1;
714 BAMBOO_ENTER_CLIENT_MODE_FROM_RUNTIME();
715 for(i = 1; i < NUMCORESACTIVE; ++i) {
716 // send profile request msg to core i
717 send_msg_2(i, PROFILEOUTPUT, totalexetime);
720 // pour profiling data on startup core
724 BAMBOO_ENTER_RUNTIME_MODE_FROM_CLIENT();
725 if(numconfirm != 0) {
727 BAMBOO_ENTER_CLIENT_MODE_FROM_RUNTIME();
731 BAMBOO_ENTER_CLIENT_MODE_FROM_RUNTIME();
738 void checkCoreStatus() {
742 (waitconfirm && (numconfirm == 0))) {
743 BAMBOO_ENTER_RUNTIME_MODE_FROM_CLIENT();
744 corestatus[BAMBOO_NUM_OF_CORE] = 0;
745 numsendobjs[BAMBOO_NUM_OF_CORE] = self_numsendobjs;
746 numreceiveobjs[BAMBOO_NUM_OF_CORE] = self_numreceiveobjs;
747 // check the status of all cores
748 for(i = 0; i < NUMCORESACTIVE; ++i) {
749 if(corestatus[i] != 0) {
753 if(i == NUMCORESACTIVE) {
754 // check if the sum of send objs and receive obj are the same
755 // yes->check if the info is the latest; no->go on executing
757 for(i = 0; i < NUMCORESACTIVE; ++i) {
758 sumsendobj += numsendobjs[i];
760 for(i = 0; i < NUMCORESACTIVE; ++i) {
761 sumsendobj -= numreceiveobjs[i];
763 if(0 == sumsendobj) {
765 // the first time found all cores stall
766 // send out status confirm msg to all other cores
767 // reset the corestatus array too
768 corestatus[BAMBOO_NUM_OF_CORE] = 1;
770 numconfirm = NUMCORESACTIVE - 1;
771 BAMBOO_ENTER_CLIENT_MODE_FROM_RUNTIME();
772 for(i = 1; i < NUMCORESACTIVE; ++i) {
774 // send status confirm msg to core i
775 send_msg_1(i, STATUSCONFIRM);
779 // all the core status info are the latest
780 // terminate; for profiling mode, send request to all
781 // other cores to pour out profiling data
782 recordtotalexetime();
784 CACHEADAPT_DISABLE_TIMER();
785 GC_OUTPUT_PROFILE_DATA();
789 gc_outputProfileDataReadable();
791 tprintf("FINISH_EXECUTION\n");
792 BAMBOO_ENTER_CLIENT_MODE_FROM_RUNTIME();
793 terminate(); // All done.
796 // still some objects on the fly on the network
797 // reset the waitconfirm and numconfirm
802 // not all cores are stall, keep on waiting
806 BAMBOO_ENTER_CLIENT_MODE_FROM_RUNTIME();
810 // main function for each core
811 void run(int argc, char** argv) {
812 bool sendStall = false;
814 bool tocontinue = false;
816 corenum = BAMBOO_GET_NUM_OF_CORE();
817 // initialize runtime data structures
824 if (BAMBOO_NUM_OF_CORE==STARTUPCORE)
825 profile_init(_LOCAL_DRD_CNT,_LOCAL_WR_CNT, _REMOTE_DRD_CNT, _REMOTE_WR_CNT);
827 int offcore=4*(BAMBOO_NUM_OF_CORE-1);
828 profile_init(validevents[(offcore)%87], validevents[(offcore+1)%87], validevents[(offcore+2)%87], validevents[(offcore+3)%87]);
831 if (BAMBOO_NUM_OF_CORE==STARTUPCORE) {
832 numconfirm=NUMCORES-1;
833 for(int i=0;i<NUMCORES;i++) {
834 if (i!=STARTUPCORE) {
835 send_msg_1(i,REQNOTIFYSTART);
840 tprintf("START_EXECUTION\n");
841 bamboo_start_time = BAMBOO_GET_EXE_TIME();
847 bme_performance_counter_start();
850 CACHEADAPT_ENABLE_TIMER();
852 initializeexithandler();
854 // main process of the execution module
855 if(BAMBOO_NUM_OF_CORE > NUMCORESACTIVE - 1) {
857 // non-executing cores, only processing communications
863 /* Create queue of active tasks */
864 activetasks= genallocatehashtable((unsigned int (*)(void *)) &hashCodetpd,
865 (int (*)(void *,void *)) &comparetpd);
867 /* Process task information */
870 if(STARTUPCORE == BAMBOO_NUM_OF_CORE) {
871 /* Create startup object */
872 createstartupobject(argc, argv);
877 profile_start(APP_REGION);
879 if(STARTUPCORE == BAMBOO_NUM_OF_CORE) {
881 // run the initStaticAndGlobal method to initialize the static blocks and
883 initStaticAndGlobal();
885 // run the main method in the specified mainclass
886 mgc_main(argc, argv);
893 // check if there are new active tasks can be executed
899 while(receiveObject_I() != -1) {
902 // check if there are some pending objects,
903 // if yes, enqueue them and executetasks again
904 tocontinue = checkObjQueue();
906 tocontinue = trystartthread();
914 if(STARTUPCORE == BAMBOO_NUM_OF_CORE) {
925 // wait for some time
931 // send StallMsg to startup core
933 send_msg_4(STARTUPCORE,TRANSTALL,BAMBOO_NUM_OF_CORE,self_numsendobjs,self_numreceiveobjs);