void CreateProfilingClocks(void){
hins_ticks=0;hdel_ticks=0;hupdate_ticks=0;
dual_ticks=0;
hins_clk = CreateClock(); hdel_clk = CreateClock();
hdel_clk = CreateClock(); hupdate_clk = CreateClock();
dual_clk = CreateClock();
}
/** Parse a clock tray component. */
void ParseClock(const TokenNode *tp, TrayType *tray)
{
TrayComponentType *cp;
const char *format;
const char *zone;
const char *temp;
int width, height;
Assert(tp);
Assert(tray);
format = FindAttribute(tp->attributes, "format");
zone = FindAttribute(tp->attributes, "zone");
temp = FindAttribute(tp->attributes, WIDTH_ATTRIBUTE);
if(temp) {
width = ParseUnsigned(tp, temp);
} else {
width = 0;
}
temp = FindAttribute(tp->attributes, HEIGHT_ATTRIBUTE);
if(temp) {
height = ParseUnsigned(tp, temp);
} else {
height = 0;
}
cp = CreateClock(format, zone, width, height);
if(JLIKELY(cp)) {
ParseTrayComponentActions(tp, cp, AddClockAction);
AddTrayComponent(tray, cp);
}
}
int main(int argc,char **argv){
struct tms s1;
struct tms s2;
volatile int count,count1;
volatile int i;
volatile double X=1024;
Clock *test_clk = CreateClock();
StartClock(test_clk);
for(i=0;i<1000000;i++){
count1+=(i+1);
count1 = (count1*i)/count1;
X += ((X*X) + X/0.331)*(X*X*X);
}
StopClock(test_clk);
printf("PROBE1: Clock Ticks Elapsed is %Lu \n",(clock_t)GetClockTicks(test_clk));
StartClock(test_clk);
for(i=0;i<1000000;i++){
count1+=(i+1);
count1 = (count1*i)/count1;
X += ((X*X) + X/0.331)*(X*X*X);
}
StopClock(test_clk);
printf("PROBE2: Clock Ticks Elapsed is %Lu \n",(clock_t)GetClockTicks(test_clk));
printf("size of clock_t is %d \n",sizeof(clock_t));
}
void LogTransformT(RowStarts *rowptr,ColIndices *colind,ValueType *val,Indices n,ValueType *u,
ValueType *v,ValueType *dist,Indices *m,Indices *m_inv){
ValueType *max_c = new ValueType[n];
rowptr--;colind--;val--;max_c--; m--; m_inv--; u--; dist--;
Indices k,i,j,j0;
Clock *anst_clk;
anst_clk = CreateClock();
StartClock(anst_clk);
/*TODO: Avoid HUGE_VAL directly in templated code*/
for(j=1;j<=n;j++){
/*Find the maximum in the column*/
max_c[j] = (ValueType)0.0;
m[j] = (Indices)0; m_inv[j]=0;
u[j] = HUGE_VAL;dist[j]=HUGE_VAL;
}
for(i=1;i<=n;i++){
for(k=rowptr[i]+1;k<(rowptr[i+1]+1);k++){
val[k] = fabs(val[k]);
j = colind[k]+1;
if(val[k] > max_c[j]){
max_c[j] = val[k];
}
}
}
#if 0
for(k=1;k<=n;k++){
printf("max in col %u is %e \n",k,max_c[k]);
}
#endif
for(i=1;i<=n;i++){
for(k=rowptr[i]+1;k<(rowptr[i+1]+1);k++){
j = colind[k]+1;
if(val[k]<=1.0e-30){
val[k] = HUGE_VAL;
}else{
if(max_c[j]<=1.0e-30){
val[k] = HUGE_VAL;//MAX_DOUBLE_VALUE/n - (log(val[k])/M_LN10);
}else{
val[k] = log10(max_c[j]/val[k]);
val[k] = fabs(val[k]);
if(val[k]< u[j]){
u[j] = val[k];
}
}
}
/*take care of -0.0*/
/*assert(val[k]>=0.0);*/
}
}
StopClock(anst_clk);
anst_ticks = GetClockTicks(anst_clk);
++max_c;
delete max_c;
}
void ANSTMC21(RowStarts rowptr,ColIndices colind,Indices n,Indices *m){
Indices *col_marker;
Clock *anst_clock = CreateClock();
Indices *p; Indices *m_inv;
CreateMC21WorkSpace(n,&col_marker,&p,&m_inv);
StartClock(anst_clock);
FindPerfectMatch(rowptr,colind,n,m,p,m_inv,col_marker);
StopClock(anst_clock);
anst_ticks = GetClockTicks(anst_clock);
FreeMC21WorkSpace(col_marker,p,m_inv);
free(anst_clock);
}
/*
*takes a binary file, set the columns which you want to
*sort in ESortParams.col_start and ESortParams.col_end
*/
void ExSortKmerEdges(const char *bin_kedge, const char *run_file,
unsigned char KEY_SIZE, ExKeyCompare key_compare){
unsigned char *run_buffer = NULL;
size_t key_buf_len = (KEY_SIZE*KEYS_IN_RUN);
size_t buf_len = key_buf_len + sizeof(unsigned long);
size_t ret_len;
int bin_kedge_fd = open(bin_kedge, O_RDONLY);
int run_file_fd = fileno(fopen(run_file, "w"));
unsigned long *runlen;
unsigned long rcount = 0;
unsigned char *key_buf;
Clock *clk = CreateClock();
if(!(bin_kedge_fd > 0 && run_file_fd > 0)){
perror("FAILED TO OPEN FILES:");
assert(0);
}
run_buffer = malloc(sizeof(unsigned char)*buf_len);
assert(run_buffer);
runlen = (unsigned long *)run_buffer;
key_buf = run_buffer + sizeof(unsigned long);
StartClock(clk);
while((ret_len = SafeRead(bin_kedge_fd, key_buf, key_buf_len)) >= KEY_SIZE){
*runlen = (unsigned long) ret_len;
IntegerSort_SB(key_buf, key_buf + (*runlen) - KEY_SIZE, KEY_SIZE, 1,
ESortParams.col_start, ESortParams.col_end, 0, CharMap,
ESortParams.endian);
ret_len = SafeWrite(run_file_fd, run_buffer,
((*runlen + sizeof(unsigned long))< buf_len)?(*runlen +
sizeof(unsigned long)):buf_len);
rcount++;
assert(ret_len == ((*runlen + sizeof(unsigned long)
< buf_len)?(*runlen + sizeof(unsigned long)):buf_len));
}
StopClock(clk);
FreeISortBuckets();
/*create the final run*/
close(run_file_fd); close(bin_kedge_fd);
free(run_buffer);
printf("\n[EX-SORT CREATED %lu RUNS] took %ld ticks\n", rcount, GetClockTicks(clk));
StartClock(clk);
/*call the external rway merge*/
ExternalRWayMerge(run_file, RUNS_PER_MERGE, KEY_SIZE, key_compare, rcount);
StopClock(clk);
printf("[R-WAY MERGE] took %ld ticks\n", GetClockTicks(clk));
}
void LogTransform(SparseGraph *G){
node_t *rowptr = G->rowptr; rowptr--;
node_t *colind = G->colind; colind--;
double *val = G->nnz;val--;
node_t n = G->order; node_t k;
node_t nnz_size = G->nnz_size;
node_t k0,j0;
double *max_c = (double *)calloc(n,sizeof(double));max_c--;
#if 0
for(k=1;k<=n;k++){
/*Find the maximum in the column*/
max_c[k] = (double)0.0;
}
#endif
Clock *anst_clk = CreateClock();
StartClock(anst_clk);
for(k=1;k<=nnz_size;k++){
val[k] = fabs(val[k]);
if(val[k] > max_c[colind[k]]){
max_c[colind[k]] = val[k];
}
}
#if 0
for(k=1;k<=n;k++){
printf("max in col %u is %e \n",k,max_c[k]);
}
#endif
for(k=1;k<=nnz_size;k++){
if(fabs(val[k]-0.0)<=1.0e-30){
val[k] = HUGE_VAL;
}else{
if(fabs(max_c[colind[k]]-0.0)<=1.0e-30){
val[k] = HUGE_VAL;//MAX_DOUBLE_VALUE/n - (log(val[k])/M_LN10);
}else{
val[k] = log10(max_c[colind[k]]/val[k]);
val[k] = fabs(val[k]);
}
}
/*take care of -0.0*/
assert(val[k]>=0.0);
}
StopClock(anst_clk);
anst_ticks = GetClockTicks(anst_clk);
free(++max_c);
}
/** Parse a clock tray component. */
void ParseClock(const TokenNode *tp, TrayType *tray) {
TrayComponentType *cp;
const char *format;
const char *zone;
const char *command;
const char *temp;
int width, height;
Assert(tp);
Assert(tray);
format = FindAttribute(tp->attributes, FORMAT_ATTRIBUTE);
zone = FindAttribute(tp->attributes, ZONE_ATTRIBUTE);
if(tp->value && strlen(tp->value) > 0) {
command = tp->value;
} else {
command = NULL;
}
temp = FindAttribute(tp->attributes, WIDTH_ATTRIBUTE);
if(temp) {
width = ParseUnsigned(tp, temp);
} else {
width = 0;
}
temp = FindAttribute(tp->attributes, HEIGHT_ATTRIBUTE);
if(temp) {
height = ParseUnsigned(tp, temp);
} else {
height = 0;
}
cp = CreateClock(format, zone, command, width, height);
if(JLIKELY(cp)) {
AddTrayComponent(tray, cp);
}
}
size_t WeightedMatching(RowStarts rowptr,ColIndices colind,ValueTypePtr C,
ValueTypePtr dist,ValueTypePtr u,ValueTypePtr v,Indices *p,
Indices *m_inv,Indices *m, Indices n,CompareFunction cmpFunc){
typedef typename std::iterator_traits<ValueTypePtr>::value_type ValueType;
Indices i,j,i1,jend,k,m_inv_prev;
Indices match_size=0; Indices k0,j0;
ValueType curr_shortest_path = (ValueType)0;
ValueType curr_aug_path = GetMaxTypeValue<ValueType>(curr_aug_path);
ValueType dist1; Indices itrace;
/*Cost of the edges in the match if
*$(i,j) \in M$ then $clabel[i] = C[i][j]$*/
Indices *clabel = new Indices[n];
Indices *aug_label = new Indices[n];
Indices *update_stack = new Indices[n];
Indices update_stack_index;
/*Save The Operations on the Heap.*/
Indices save_heap_index;
Indices *save_heap_op = new Indices[n];
#ifdef TURN_ON_SAVE_HEAP
double close_factor = (double)1.0 + (double)1.0e-16;
#endif
/*Force the write back to memory to avoid floating point issues*/
ValueType force_mem_write[3];
#ifndef NO_LOCAL_PROFILING
CreateProfilingClocks();
#endif
/*Core Profiling Clock*/
Clock *core_clk = CreateClock();
#ifdef USE_BIT_ARRAY
BitArray_t *col_marker = CreateBitArray(n);
BitArray_t *heap_marker = CreateBitArray(n);
#else
Indices *col_marker = new Indices[n];
unsigned int *heap_marker = NULL;
col_marker--;
for(i=1;i<=n;i++){ /*Do we need Initialization?*/
col_marker[i] = (Indices)0;
}
#endif
#if BINARY_HEAP
Heap *bin_heap = NewBinaryHeap(cmpFunc,n,GetDistKeyID);
ValueType *dist_ptr = NULL;
heap_marker = bin_heap->keyMap;
#endif
/*Algorithm Uses 1-Indexing to be consistent*/
C--;m--;dist--;u--;v--;p--;m_inv--;
rowptr--;colind--;clabel--;save_heap_op--;
update_stack--;aug_label--;
assert(dist && u && v && p);
ComputeInitialExtremeMatch<ValueType,Indices>(u,v,clabel,C,m,m_inv,colind,
rowptr,n,dist);
match_size=0;
StartClock(core_clk);
for(i=1;i<=n;i++){
if(m_inv[i]){
match_size++;
continue;
}
/*
*Aim is to find a value for jend such that the path
*from i-->jend is the shortest
*/
i1 = i; p[i1] = 0; jend=0; itrace=i;
#ifdef USE_BIT_ARRAY
ResetAllBits(col_marker);
ResetAllBits(heap_marker);
#endif
#if BINARY_HEAP
bin_heap->n = 0;
dist_base = (unsigned long)&(dist[1]);
#endif
curr_shortest_path=(ValueType)0;
curr_aug_path=GetMaxTypeValue<ValueType>(curr_aug_path);
save_heap_index = (Indices)0;
update_stack_index = (Indices)0;
while(1){
for(k=rowptr[i1]+1;k<(rowptr[i1+1]+1);k++){
j = colind[k]+1;
#ifdef USE_BIT_ARRAY
if(CheckBit(col_marker,j)){
#else
if(col_marker[j]==i){
#endif
continue;
}
force_mem_write[k%3] = C[k]-(v[i1]+u[j]);
dist1 = curr_shortest_path + force_mem_write[k%3];
/*Prune any dist1's > curr_aug_path, since
*all the costs>0
*/
if(dist1 < curr_aug_path){
if(!m[j]){
/*we need itrace because, the last i1 which
*we explore may not actually give the shortest
*augmenting path.*/
jend = j; itrace = i1;
curr_aug_path = dist1;
aug_label[j] = k;
}else if(dist1 < dist[j]){
/*Update the dist*/
dist[j] = dist1; p[m[j]] = i1;
aug_label[j] = k;
#if SIMPLE_HEAP
#ifdef USE_BIT_ARRAY
SetBit(heap_marker,j);
#else
heap_marker[j] = i;
#endif
#elif BINARY_HEAP /*SIMPLE_HEAP*/
#ifdef USE_BIT_ARRAY
if(CheckBit(heap_marker,j)){
#else
if(heap_marker[j]){
#endif
#ifndef NO_LOCAL_PROFILING
StartClock(hupdate_clk);
#endif
/*Call the decrease Key Operation*/
DecreaseKey(bin_heap,j);
#ifndef NO_LOCAL_PROFILING
StopClock(hupdate_clk);
hupdate_ticks += GetClockTicks(hupdate_clk);
#endif
}
#ifdef TURN_ON_SAVE_HEAP
else if(curr_shortest_path &&
dist[j] <= (curr_shortest_path)*(close_factor)){
/*If dist[j] is close to root push it in
*save_heap_op*/
assert(save_heap_index < n);
save_heap_op[++save_heap_index] = j;
}
#endif
else{
#ifndef NO_LOCAL_PROFILING
StartClock(hins_clk);
#endif
InsertHeap(bin_heap,&(dist[j]));
#ifndef NO_LOCAL_PROFILING
StopClock(hins_clk);
hins_ticks += GetClockTicks(hins_clk);
#endif
#ifdef USE_BIT_ARRAY
SetBit(heap_marker,j);
#endif
}
#endif /*SIMPLE_HEAP*/
}
}
}
if(curr_aug_path <= curr_shortest_path){
break;
}
/*We now have a heap of matched cols, so pick the min*/
#ifdef SIMPLE_HEAP
j = SimplePickMin(heap_marker,dist,n);
if(j){
curr_shortest_path = dist[j];
UnsetBit(heap_marker,j);
#elif BINARY_HEAP
#ifndef NO_LOCAL_PROFILING
StartClock(hdel_clk);
#endif
if(save_heap_index){
j = save_heap_op[save_heap_index];
save_heap_index--;
curr_shortest_path = dist[j];
#ifdef USE_BIT_ARRAY
SetBit(col_marker,j);
#else
col_marker[j] = (Indices)i;
update_stack[++update_stack_index]=j;
#endif /*#ifdef USE_BIT_ARRAY*/
i1 = m[j];
}else if(dist_ptr = (ValueType *) HeapDelete(bin_heap)) {
#ifndef NO_LOCAL_PROFILING
StopClock(hdel_clk);
hdel_ticks += GetClockTicks(hdel_clk);
#endif
assert((unsigned long)dist_ptr >= (unsigned long)&dist[1]);
j = ((((unsigned long)dist_ptr -
(unsigned long)&dist[1]))/sizeof(ValueType))+1;
assert(j>=1 && j<=n);
curr_shortest_path = dist[j];
heap_marker[j] = 0; /*Setting the keyMap in Heap to 0*/
#endif /*#ifdef SIMPLE_HEAP */
#ifdef USE_BIT_ARRAY
SetBit(col_marker,j);
update_stack[++update_stack_index]=j;
#else
col_marker[j] = (Indices)i;
update_stack[++update_stack_index]=j;
#endif /*#ifdef USE_BIT_ARRAY*/
i1 = m[j];
}else{
break;
}
}
/*We found a shortest augmenting path*/
if(jend){
unsigned long **harray = bin_heap->heapArray;
#ifndef NO_LOCAL_PROFILING
StartClock(dual_clk);
#endif
/*NOTE1: We need a very fast way to update
*the dual variables and also reset the dist[]
*we avoid linear scan where ever we can to update
*these dual variables*/
while(update_stack_index){ /*Update u[j]: while*/
j=update_stack[update_stack_index--];
u[j] = (u[j]+dist[j])-curr_aug_path;
if(m[j]){ /*See NOTE1*/
i1 = m[j];
v[i1] = C[clabel[i1]] - u[j];
}
dist[j] = MAX_DOUBLE_VALUE;
if(bin_heap->n){
dist_ptr = (double *)harray[bin_heap->n];
j = ((((unsigned long)dist_ptr -
(unsigned long)&dist[1]))/sizeof(ValueType))+1;
heap_marker[j] = 0;
*((double *)harray[bin_heap->n]) = MAX_DOUBLE_VALUE;
bin_heap->n -= 1 ;
}
} /*Update u[j]: while*/
/*Uncomment if you need to print augmenting path*/
/*node_t itrace_prev;*/
/*printf("Shortest augmenting Path {");*/
j=jend;
while(itrace){
m_inv_prev = m_inv[itrace];
m[j] = itrace; m_inv[itrace]=j;
/*See NOTE1(above)*/
clabel[itrace] = aug_label[j];
v[itrace] = C[clabel[itrace]] - u[j];
/*printf("(%u,%u)",itrace,j);*/
j=m_inv_prev;
/*itrace_prev = itrace;*/
itrace = p[itrace];
/*
if(itrace){
printf("(%u,%u)",itrace_prev,m_inv_prev);
}*/
}
/*printf("}\n");*/
/*There may some dist[] still in the heap*/
while(bin_heap->n){
dist_ptr = (double *)harray[bin_heap->n];
j = ((((unsigned long)dist_ptr -
(unsigned long)&dist[1]))/sizeof(ValueType))+1;
heap_marker[j] = 0;
*((double *)harray[bin_heap->n]) = MAX_DOUBLE_VALUE;
bin_heap->n -= 1;
}
match_size++;
/*End Dual Update*/
#ifndef NO_LOCAL_PROFILING
StopClock(dual_clk);
dual_ticks += GetClockTicks(dual_clk);
#endif
} /*if(jend) : Found Augmeting Path*/
} /*for(i=1;i<=n;i++): Main Outer Loop*/
StopClock(core_clk);
WeightedMatchTicks = GetClockTicks(core_clk);
#ifndef NO_LOCAL_PROFILING
printf("Profile Summary\n");
printf("HINS=(%d) HDEL=(%d) HUPDATE=(%d)\n",(int)hins_ticks,(int)hdel_ticks,
(int)hupdate_ticks);
printf("DUAL=(%d) \n",(int)dual_ticks);
#endif
#ifdef USE_BIT_ARRAY
FreeBitArray(col_marker);
FreeBitArray(heap_marker);
#else
col_marker++;
delete col_marker;
#endif
#ifdef SIMPLE_HEAP
heap_marker++;
delete heap_marker;
#endif
aug_label++;
delete aug_label;
save_heap_op++;
delete save_heap_op;
clabel++;
delete clabel;
#ifdef BINARY_HEAP
FreeHeap(bin_heap);
#endif
return match_size;
}
/*O(n) time picking the maximum from the heap_marker */
node_t SimplePickMin(BitArray_t *bit_heap,double *dist,node_t n){
node_t min_j=0;node_t j;
double curr_min = HUGE_VAL;
for(j=1;j<=n;j++){
if(CheckBit(bit_heap,j) && dist[j] < curr_min){
min_j = j;
curr_min = dist[j];
}
}
return min_j;
}
#ifdef BINARY_HEAP
inline keyid_t GetDistKeyID(void *dist_ptr){ assert((unsigned long)dist_ptr >= dist_base);
return (((((unsigned long)dist_ptr-dist_base))/sizeof(double))+1);
}
#endif
BitArray_t* CreateBitArray(unsigned int size){
div_t d = div(size,SIZE_OF_BYTE_IN_BITS);
BitArray_t *bits = (BitArray_t *)malloc(sizeof(BitArray_t)*1);
assert(bits);
bits->size = (d.rem > 0)?(d.quot+1):d.quot;
bits->ba = (char *)malloc(sizeof(char)*(bits->size));
assert(bits->ba);
memset(bits->ba,'\0',bits->size);
return bits;
}
void ComputeInitialExtremeMatch(ValueType *u,ValueType *v,Indices *clabel,ValueType *C,
Indices *m,Indices *m_inv,Indices* colind,Indices* rowptr,Indices n,ValueType *dist){
Indices i,k,i1,k1;
Indices k0,k10,j0,j,j1,j10;
ValueType vmin;
ValueType C1k;
Clock *init_match_clk = CreateClock();
StartClock(init_match_clk);
#if 0
/*Compute u[j]*/
for(j=1;j<=n;j++){
u[j] = GetMaxTypeValue<ValueType>(u[j]);
dist[j] = u[j];
m[j]=0;m_inv[j]=0;
}
for(i=1;i<=n;i++){
for(k=rowptr[i];k<rowptr[i+1];k++){
if(C[k]<u[colind[k]]){
u[colind[k]] = C[k];
}
}
}
#endif
/*Compute v[i]*/
for(i=1;i<=n;i++){
v[i] = GetMaxTypeValue<ValueType>(v[i]);
for(k=rowptr[i]+1;k<(rowptr[i+1]+1);k++){
j = colind[k]+1;
vmin = (C[k]-u[j]);
if(vmin < v[i]){
v[i] = vmin;
}
}
}
/*Update Cost and match.*/
for(i=1;i<=n;i++){
for(k=rowptr[i]+1;k<(rowptr[i+1]+1);k++){
j = colind[k]+1;
C1k = C[k]-v[i]-u[j]; /*to handle -0.0*/
if(fabs(C1k-0.0) <= 1.0e-30 && (!m[j] && !m_inv[i])){
m[j] = i;
m_inv[i] = j;
clabel[i] = k;
}
}
}
/*1-Step Augmentation*/
for(i=1;i<=n;i++){
if(!m_inv[i]){ /*Unmatched Row*/
for(k=rowptr[i]+1;k<(rowptr[i+1]+1) && !(m_inv[i]);k++){
j = colind[k]+1;
C1k = fabs(C[k]-v[i]-u[j]);
if(C1k <= 1.0e-30){
/*assert(m[colind[k]]);*/
i1 = m[j];
/*assert(m_inv[i1] == j);*/
/*See if we can find any C1(i1,j1) == 0*/
for(k1=rowptr[i1]+1;k1<(rowptr[i1+1]+1);k1++){
j1 = colind[k1]+1;
C1k = fabs(C[k1] - v[i1]-u[j1]);
if(C1k <= 1.0e-30 && !(m[j1])){
/*augment the match.*/
m[j] = i;
m_inv[i] = j;
clabel[i] = k;
m[j1] = i1;
m_inv[i1] = j1;
clabel[i1] = k1;
break;
}
}
}
}
}
}
StopClock(init_match_clk);
InitialMatchTicks = GetClockTicks(init_match_clk);
}
