Commit 1f21b6a0 authored by Mulas, Giacomo's avatar Mulas, Giacomo
Browse files

Correct some bugs in parallelised scr2

parent f1b1bf1a

src/libnptm/clu_subs.cpp

+4 −4
Original line number Diff line number Diff line
@@ -2109,6 +2109,7 @@ void scr2( 
  double cfsq = 4.0 / (pi4sq * ccs * ccs);

  cph = uim * exri * vkarg;

  int ls = (c4->li < c4->le) ? c4->li : c4->le;

  int kmax = (ls+1)*(ls+1)-1;

  c3->tsas[0][0] = cc0;

  c3->tsas[1][0] = cc0;

  c3->tsas[0][1] = cc0;
@@ -2119,7 +2120,7 @@ void scr2( 
#ifdef USE_NVTX

  nvtxRangePush("scr2 outer loop 1");

#endif

#pragma omp parallel for

  //#pragma omp parallel for

  for (int i14 = 1; i14 <= c4->nsph; i14++) {

    int i = i14 - 1;

    int iogi = c1->iog[i14 - 1];
@@ -2137,7 +2138,6 @@ void scr2( 
      // but if it results im = 0, then we set l10 = l10-1 and im10 = 2*l10+1

      // furthermore if it results im10 > 2*l10+1, then we set

      // im10 = im10 -(2*l10+1) and l10 = l10+1 (there was a rounding error in a nearly exact root)

      int kmax = (ls+1)*(ls+1)-1;

#ifdef USE_NVTX

  nvtxRangePush("scr2 inner loop 1");

#endif
@@ -2190,7 +2190,7 @@ void scr2( 
      c1->sas[i][1][1] = s22 * csam;

    }

    // label 12

    phas = cexp(cph * (duk[0] * c1->rxx[i] + duk[1] * c1->ryy[i] + duk[2] * c1->rzz[i]));

    dcomplex phas = cexp(cph * (duk[0] * c1->rxx[i] + duk[1] * c1->ryy[i] + duk[2] * c1->rzz[i]));

    c3->tsas[0][0] += (c1->sas[iogi - 1][0][0] * phas);

    c3->tsas[1][0] += (c1->sas[iogi - 1][1][0] * phas);

    c3->tsas[0][1] += (c1->sas[iogi - 1][0][1] * phas);
@@ -2211,7 +2211,7 @@ void scr2( 
#ifdef USE_NVTX

  nvtxRangePush("scr2 inner loop 2");

#endif

#pragma omp target parallel for collapse(4)

#pragma omp target teams distribute parallel for simd collapse(4)

      for (int ipo1 = 1; ipo1 <=2; ipo1++) {

	for (int jpo1 = 1; jpo1 <= 2; jpo1++) {

	  // cc = dconjg(c1->sas[i24 - 1][jpo1 - 1][ipo1 - 1]);