void CTimer::Stop(void)
{
#ifdef _WIN32
QueryPerformanceCounter(&m_t1);
#endif
m_c1 = ippGetCpuClocks();
return;
} // CTimer::Stop()
void CTimer::Start(void)
{
#ifdef _WIN32
QueryPerformanceCounter(&m_t0);
#endif
#ifdef linux
m_t0 = clock();
#endif
m_c0 = ippGetCpuClocks();
return;
} // CTimer::Start()
static void measure_end(MeasureIt *m)
{
m->p_end = ippGetCpuClocks();
}
// START FUNC DECL
int
srt_uniform_I8(
long long *X, /* input */
long long nR,
long long *Y, /* output */
int shift,
char *mode /* "asc" or "dsc" */
)
// STOP FUNC DECL
{
int status = 0;
long long **t_cnt = NULL, *cnt = NULL, *bin_offset = NULL; int nT = 0;
long long t[16];
t[0] = ippGetCpuClocks();
if ( shift < 0 ) { go_BYE(-1); }
if ( X == NULL ) { go_BYE(-1); }
if ( mode == NULL ) { go_BYE(-1); }
if ( nR < 16*1048576 ) {
fprintf(stderr, "Too few elements to sort \n"); go_BYE(-1);
}
if ( ( strcmp(mode, "asc") != 0 ) && ( strcmp(mode, "dsc") != 0 ) ) {
go_BYE(-1);
}
nT = NUM_THREADS;
long long block_size = nR / nT;
cnt = malloc(nT * sizeof(long long));
return_if_malloc_failed(cnt);
assign_const_I8(cnt, nT, 0);
bin_offset = malloc(nT * sizeof(long long));
return_if_malloc_failed(bin_offset);
t_cnt = malloc(nT * sizeof(long long *));
return_if_malloc_failed(t_cnt);
for ( int i = 0; i < nT; i++ ) {
t_cnt[i] = malloc(nT * sizeof(long long));
return_if_malloc_failed(t_cnt[i]);
assign_const_I8(t_cnt[i], nT, 0);
}
t[1] = ippGetCpuClocks();
// Figure out how many elements in each bin
cilkfor ( int tid = 0; tid < nT; tid++ ) {
long long lb = tid * block_size;
long long ub = lb + block_size;
if ( tid == (nT-1) ) { ub = nR; }
long long *l_cnt = t_cnt[tid];
for ( long long i = lb; i < ub; i++ ) {
unsigned long long inval = (unsigned long long) X[i];
int bin = inval >> shift;
if ( bin >= NUM_THREADS ) { status = -1; continue; }
l_cnt[bin]++;
}
}
t[2] = ippGetCpuClocks();
// Sum up over local counts
for ( int i = 0; i < nT; i++ ) {
long long tempsum = 0;
for ( int j = 0; j < nT; j++ ) {
tempsum += t_cnt[j][i];
}
cnt[i] = tempsum;
}
// Quick debugging check
long long chk_nX = 0;
for ( int i = 0; i < nT; i++ ) {
chk_nX += cnt[i];
}
// Create cumulative counts
bin_offset[0] = 0;
for ( int i = 1; i < nT; i++ ) {
bin_offset[i] = bin_offset[i-1] + cnt[i-1];
}
/*
for ( int i = 1; i < nT; i++ ) {
fprintf(stderr, "%d --> %lld \n", i, cnt[i]);
}
*/
t[3] = ippGetCpuClocks();
if ( chk_nX != nR ) { go_BYE(-1); }
// Now let us get elements to their correct bins
for ( int i = 0; i < nR; i++ ) {
unsigned long long inval = (unsigned long long) X[i];
int bin = inval >> shift;
// if ( bin >= NUM_THREADS ) { status = -1; continue; }
Y[bin_offset[bin]] = X[i];
bin_offset[bin]++;
}
cBYE(status);
t[4] = ippGetCpuClocks();
// Re-Create cumulative counts
bin_offset[0] = 0;
for ( int i = 1; i < nT; i++ ) {
bin_offset[i] = bin_offset[i-1] + cnt[i-1];
}
// ------------------------------------------------------------
// Now we can sort individual bins in parallel
cilkfor ( int tid = 0; tid < nT; tid++ ) {
long long *lY = (long long *)Y; lY += bin_offset[tid];
if ( cnt[tid] == 0 ) { continue; }
if ( strcmp(mode, "asc") == 0 ) {
qsort_asc_I8(lY, cnt[tid], sizeof(long long), NULL);
}
else if ( strcmp(mode, "dsc") == 0 ) {
qsort_dsc_I8(lY, cnt[tid], sizeof(long long), NULL);
}
else { status = -1; continue; }
}
cBYE(status);
t[5] = ippGetCpuClocks();
/*
for ( int i = 1; i < 5; i++ ) {
fprintf(stderr, "%d --> %lld \n", i, t[i] - t[i-1]);
}
*/
// ------------------------------------------------------------
BYE:
if ( t_cnt != NULL ) {
for ( int i = 0; i < nT; i++ ) {
free_if_non_null(t_cnt[i]);
}
free_if_non_null(t_cnt);
}
free_if_non_null(cnt);
free_if_non_null(bin_offset);
return(status);
}
static void measure_start(MeasureIt *m)
{
m->p_start = ippGetCpuClocks();
}
