void av_change(struct IC *p,bvtype *use,bvtype *def)
/* Berechnet die Aenderungen, die sich durch IC p an use und def ergeben. */
{
int i,j,n=-1;
int g1,g2;
/* Wenn eine Quelle==Ziel, dann wird dadurch kein neuer use erzeugt, */
/* um z.B. unbenutzte Induktionsvariablen in Schleifen zu eliminieren. */
g1=compare_objs(&p->q1,&p->z,p->typf);
g2=compare_objs(&p->q2,&p->z,p->typf);
if(!g1&&(p->q1.flags&(VAR|DREFOBJ))==VAR) n=p->q1.v->index;
if(!g2&&(p->q2.flags&(VAR|DREFOBJ))==VAR) n=p->q2.v->index;
for(j=0;j<p->use_cnt;j++){
i=p->use_list[j].v->index;
if(p->use_list[j].flags&DREFOBJ) i+=vcount-rcount;
if(i>=vcount) continue;
if(i!=n&&!BTST(def,i)) BSET(use,i);
}
/* Ein Wert wird nicht zerstoert, wenn es kein elementarer Typ ist und */
/* die Groesse kleiner als die Variable (steht in alle solchen ICs in */
/* q2.val.max. */
if((p->z.flags&(VAR|DREFOBJ))==VAR&&(ISSCALAR(p->z.v->vtyp->flags)||p->z.v->vtyp->flags==0||zmeqto(p->q2.val.vmax,szof(p->z.v->vtyp)))){
i=p->z.v->index;
if(i>=vcount) ierror(0);
if(g1&&g2&&!BTST(use,i)) BSET(def,i);
/* Wenn p geaendert wird, wird auch *p geaendert */
if(i<rcount&&!BTST(def,i+vcount-rcount)) BSET(use,i+vcount-rcount);
}
if((p->z.flags&(VAR|DREFOBJ))==(VAR|DREFOBJ)&&g1&&g2&&!(p->z.v->flags&DNOTTYPESAFE)){
i=p->z.v->index+vcount-rcount;
if(i>=vcount) ierror(0);
if(!BTST(use,i)) BSET(def,i);
}
}
int dead_assignments(struct flowgraph *fg)
/* Findet Zuweisungen, die unnoetig sind, da die Variable nie mehr */
/* benutzt werden kann. */
{
int changed=0;struct IC *p;bvtype *isused;
int i,j;
if(DEBUG&1024) printf("searching for dead assignments\n");
isused=mymalloc(vsize);
while(fg){
memcpy(isused,fg->av_out,vsize);
p=fg->end;
while(p){
if(p->z.flags&VAR){
i=p->z.v->index;
if(p->z.flags&DREFOBJ) i+=vcount-rcount;
if(!BTST(isused,i)&&!is_volatile_ic(p)&&!(disable&1)){
if(DEBUG&1024){printf("dead assignment deleted:\n");pric2(stdout,p);}
if(*p->z.v->identifier&&p->code!=ASSIGN){ err_ic=p;error(170,i>=vcount-rcount?"*":"",p->z.v->identifier);err_ic=0;}
if(p->code!=GETRETURN) changed=1;
if(p==fg->start){remove_IC_fg(fg,p);break;}
p=p->prev;remove_IC_fg(fg,p->next);
continue;
}
}
if(p->code!=SETRETURN&&p->code!=TEST&&p->code!=COMPARE&&(p->q1.flags&VAR)&&!BTST(isused,p->q1.v->index)&&(!(p->z.flags&VAR)||!p->z.v->reg||p->z.v->identifier)){
struct IC *m,*a;int f=p->q1.flags,dt=p->q1.dtyp;
p->q1.flags&=~DREFOBJ;
a=p->prev;if(a) m=a->prev; else m=0;
if(m&&a&&m->code==ASSIGN&&(a->q1.flags&(VAR|DREFOBJ))==VAR&&!compare_objs(&p->q1,&m->z,0)&&!compare_objs(&a->q1,&a->z,0)&&!compare_objs(&m->q1,&a->z,0)&&(a->q2.flags&KONST)&&!var_conflicts(a->q1.v,p)){
if(DEBUG&1024){
printf("reorder post-op(q1):\n");
pric2(stdout,m);pric2(stdout,a);pric(stdout,p);
}
p->q1=a->q1;
m->next=p;p->prev=m;
if(p->next) p->next->prev=a;
a->next=p->next;
a->prev=p;p->next=a;
if(fg->end==p) fg->end=a;
if(p==last_ic) last_ic=a;
remove_IC_fg(fg,m);
av_update(a,isused);
p->use_list=myrealloc(p->use_list,(p->use_cnt+a->use_cnt)*VLS);
memcpy(&p->use_list[p->use_cnt],a->use_list,a->use_cnt*VLS);
p->use_cnt+=a->use_cnt;
changed=1;
if((f&DREFOBJ)&&p->q1.v->index>=rcount)
ierror(0);
}
p->q1.flags=f;
p->q1.dtyp=dt;
}
if(p->code!=TEST&&p->code!=COMPARE&&(p->q2.flags&VAR)&&!BTST(isused,p->q2.v->index)&&(!(p->z.flags&VAR)||!p->z.v->reg||p->z.v->identifier)){
struct IC *m,*a;int f=p->q2.flags,dt=p->q2.dtyp;
p->q2.flags&=~DREFOBJ;
a=p->prev;if(a) m=a->prev; else m=0;
if(m&&a&&m->code==ASSIGN&&(a->q1.flags&(VAR|DREFOBJ))==VAR&&!compare_objs(&p->q2,&m->z,0)&&!compare_objs(&a->q1,&a->z,0)&&!compare_objs(&m->q1,&a->z,0)&&(a->q2.flags&KONST)&&!var_conflicts(a->q1.v,p)){
if(DEBUG&1024){
printf("reorder post-op(q2):\n");
pric2(stdout,m);pric2(stdout,a);pric(stdout,p);
}
p->q2=a->q1;
m->next=p;p->prev=m;
if(p->next) p->next->prev=a;
a->next=p->next;
a->prev=p;p->next=a;
if(fg->end==p) fg->end=a;
if(p==last_ic) last_ic=a;
remove_IC_fg(fg,m);
av_update(a,isused);
p->use_list=myrealloc(p->use_list,(p->use_cnt+a->use_cnt)*VLS);
memcpy(&p->use_list[p->use_cnt],a->use_list,a->use_cnt*VLS);
p->use_cnt+=a->use_cnt;
changed=1;
if((f&DREFOBJ)&&p->q2.v->index>=rcount)
ierror(0);
}
p->q2.flags=f;
p->q2.dtyp=dt;
}
if(p->code!=TEST&&p->code!=COMPARE&&(p->z.flags&(VAR|DREFOBJ))==(VAR|DREFOBJ)&&!BTST(isused,p->z.v->index)){
struct IC *m,*a;int f=p->z.flags,dt=p->z.dtyp;
p->z.flags&=~DREFOBJ;
a=p->prev;if(a) m=a->prev; else m=0;
if(m&&a&&m->code==ASSIGN&&(a->q1.flags&(VAR|DREFOBJ))==VAR&&!compare_objs(&p->z,&m->z,0)&&!compare_objs(&a->q1,&a->z,0)&&!compare_objs(&m->q1,&a->z,0)&&(a->q2.flags&KONST)&&!var_conflicts(a->q1.v,p)){
if(DEBUG&1024){
printf("reorder post-op(z):\n");
pric2(stdout,m);pric2(stdout,a);pric2(stdout,p);
printf("--");
}
p->z=a->q1;
m->next=p;p->prev=m;
if(p->next) p->next->prev=a;
a->next=p->next;
a->prev=p;p->next=a;
if(fg->end==p) fg->end=a;
if(p==last_ic) last_ic=a;
remove_IC_fg(fg,m);
av_update(a,isused);
p->use_list=myrealloc(p->use_list,(p->use_cnt+a->use_cnt)*VLS);
memcpy(&p->use_list[p->use_cnt],a->use_list,a->use_cnt*VLS);
p->use_cnt+=a->use_cnt;
changed=1;
if((f&DREFOBJ)&&p->z.v->index>=rcount)
ierror(0);
}
p->z.flags=f;
p->z.dtyp=dt;
}
av_update(p,isused);
if(p==fg->start) break;
p=p->prev;
}
fg=fg->normalout;
}
free(isused);
return(changed);
}
void ParallelPairs(void *objs, int Nmyobjs, int sizeobj,
int (*numget)(const void *),
void (*numset)(const void *, int ),
int (*procget)(const void *),
void (*marry)(const void *, const void *),
int (*compare_objs)(const void *, const void *)){
char *myobjs = (char*) objs;
int n, p, sk, cnt, num;
int maxind = 0;
int procid, nprocs;
MPI_Comm_rank(MPI_COMM_WORLD, &procid);
MPI_Comm_size(MPI_COMM_WORLD, &nprocs);
/* local sort */
qsort(myobjs, Nmyobjs, sizeobj, compare_objs);
/* TW: homework replace from here --------> */
maxind = 0;
for(n=0;n<Nmyobjs;++n){
num = numget(myobjs+sizeobj*n);
maxind = max(maxind, num);
}
int globalmaxind;
MPI_Allreduce(&maxind, &globalmaxind, 1, MPI_INT, MPI_MAX, MPI_COMM_WORLD);
int binsize = ceil( (double)(globalmaxind)/(double)nprocs ) + 10;
int *outN = (int*) calloc(nprocs, sizeof(int));
int *inN = (int*) calloc(nprocs, sizeof(int));
int *cumoutN = (int*) calloc(nprocs, sizeof(int));
int *cuminN = (int*) calloc(nprocs, sizeof(int));
sk = 0;
/* count the number of objs in each bin */
int binup = binsize;
for(p=0;p<nprocs;++p){
while( numget(myobjs+(sk*sizeobj) ) <= binup ){
++(outN[p]);
++sk;
if(sk==Nmyobjs){
break;
}
}
binup += binsize;
if(sk==Nmyobjs){
break;
}
}
/* TW: <---------- replace to here */
/* communicate numbers to be sent to each bin */
MPI_Alltoall(outN, 1, MPI_INT,
inN, 1, MPI_INT,
MPI_COMM_WORLD);
/* build incoming buffer */
int Notherobjs = 0;
for(p=0;p<nprocs;++p)
Notherobjs += inN[p];
for(p=0;p<nprocs;++p){
outN[p] *= sizeobj/sizeof(int);
inN[p] *= sizeobj/sizeof(int);
}
for(p=1;p<nprocs;++p){
cumoutN[p] = cumoutN[p-1]+outN[p-1];
cuminN[p] = cuminN[p-1] + inN[p-1];
}
/* fill up bins of objects from cloud */
char *otherobjs = (char*) calloc(Notherobjs*sizeobj, sizeof(char));
MPI_Alltoallv(myobjs, outN, cumoutN, MPI_INT,
otherobjs, inN, cuminN, MPI_INT,
MPI_COMM_WORLD);
/* sort the bin */
qsort(otherobjs, Notherobjs, sizeobj, compare_objs);
/* number unique objs consecutively in each bin */
for(n=1;n<Notherobjs;++n){
/* match ? */
if(!compare_objs(otherobjs+ n*sizeobj,
otherobjs+(n-1)*sizeobj)){
marry(otherobjs+n*sizeobj, otherobjs+(n-1)*sizeobj);
}
}
char *outobjs = (char*) calloc(Notherobjs*sizeobj, sizeof(char));
sk = 0;
for(p=0;p<nprocs;++p)
for(n=0;n<Notherobjs;++n)
if(procget(otherobjs+n*sizeobj)==p){
memcpy(outobjs+sk*sizeobj, otherobjs+n*sizeobj, sizeobj);
++sk;
}
/* send results out */
MPI_Alltoallv(outobjs, inN, cuminN, MPI_INT,
myobjs, outN, cumoutN, MPI_INT,
MPI_COMM_WORLD);
free(otherobjs);
free(outN); free(inN); free(cumoutN); free(cuminN);
}
