also fillIdx

gridder_cxx.h

@@ -1587,8 +1587,19 @@ void fillIdx(const Baselines &baselines,
   size_t nbu = (gconf.Nu()+1+side-1) >> logsquare,
          nbv = (gconf.Nv()+1+side-1) >> logsquare;
   vector<idx_t> acc(nbu*nbv+1, 0);
-  vector<idx_t> tmp(nrow*(chend-chbegin));
-  for (idx_t irow=0, idx=0; irow<nrow; ++irow)
+  vector<idx_t> tmp(nrow*(chend-chbegin),~idx_t(0));
+
+#pragma omp parallel
+{
+  idx_t nthr = my_num_threads();
+  idx_t id = my_thread_num();
+  idx_t nbase = nrow/nthr;
+  idx_t additional = nrow%nthr;
+  idx_t lo = id*nbase + ((id<additional) ? id : additional);
+  idx_t hi = lo+nbase+(id<additional);
+  vector<idx_t> lacc(nbu*nbv+1, 0);
+
+  for (idx_t irow=lo, idx=lo*(chend-chbegin); irow<hi; ++irow)
     for (int ichan=chbegin; ichan<chend; ++ichan)
       if (!flags(irow, ichan))
         {
@@ -1601,10 +1612,15 @@ void fillIdx(const Baselines &baselines,
           gconf.getpix(uvw.u, uvw.v, u, v, iu0, iv0);
           iu0 = (iu0+nsafe)>>logsquare;
           iv0 = (iv0+nsafe)>>logsquare;
-          ++acc[nbv*iu0 + iv0 + 1];
+          ++lacc[nbv*iu0 + iv0 + 1];
           tmp[idx++] = nbv*iu0 + iv0;
           }
         }
+#pragma omp barrier
+#pragma omp critical(xyz)
+  for (size_t i=0; i<acc.size(); ++i)
+    acc[i]+=lacc[i];
+}
 
   for (size_t i=1; i<acc.size(); ++i)
     acc[i] += acc[i-1];
@@ -1613,6 +1629,7 @@ void fillIdx(const Baselines &baselines,
     for (int ichan=chbegin; ichan<chend; ++ichan)
       if (!flags(irow, ichan))
         {
+        while (tmp[idx]==idx_t(~0)) ++idx;
         auto w = abs(baselines.effectiveCoord(RowChan{irow,idx_t(ichan)}).w);
         if ((w>=wmin) && (w<wmax))
           res[acc[tmp[idx++]]++] = baselines.getIdx(irow, ichan);