Avoid using function calls in OpenMP directives

[alexxy/gromacs.git] / src / gromacs / mdlib / nbnxn_kernels / nbnxn_kernel_ref.c

diff --git a/src/gromacs/mdlib/nbnxn_kernels/nbnxn_kernel_ref.c b/src/gromacs/mdlib/nbnxn_kernels/nbnxn_kernel_ref.c
index e659e5a262d33e7c131f034c1ead81ae00c7dcec..a10465296569d932b4b90a2b34ca131d98c2fe58 100644 (file)
--- a/src/gromacs/mdlib/nbnxn_kernels/nbnxn_kernel_ref.c
+++ b/src/gromacs/mdlib/nbnxn_kernels/nbnxn_kernel_ref.c
@@ -37,10 +37,11 @@
 #endif
 
 #include <math.h>
+#include <assert.h>
 
 #include "typedefs.h"
 #include "vec.h"
-#include "smalloc.h"
+#include "gromacs/utility/smalloc.h"
 #include "force.h"
 #include "gmx_omp_nthreads.h"
 #include "nbnxn_kernel_ref.h"
@@ -68,34 +69,70 @@ typedef void (*p_nbk_func_ener)(const nbnxn_pairlist_t     *nbl,
 
 /* Analytical reaction-field kernels */
 #define CALC_COUL_RF
+#define LJ_CUT
 #include "nbnxn_kernel_ref_includes.h"
-#define VDW_FORCE_SWITCH
+#undef LJ_CUT
+#define LJ_FORCE_SWITCH
 #include "nbnxn_kernel_ref_includes.h"
-#undef VDW_FORCE_SWITCH
-#define VDW_POT_SWITCH
+#undef LJ_FORCE_SWITCH
+#define LJ_POT_SWITCH
 #include "nbnxn_kernel_ref_includes.h"
-#undef VDW_POT_SWITCH
+#undef LJ_POT_SWITCH
+#define LJ_EWALD
+#define LJ_CUT
+#define LJ_EWALD_COMB_GEOM
+#include "nbnxn_kernel_ref_includes.h"
+#undef LJ_EWALD_COMB_GEOM
+#define LJ_EWALD_COMB_LB
+#include "nbnxn_kernel_ref_includes.h"
+#undef LJ_EWALD_COMB_LB
+#undef LJ_CUT
+#undef LJ_EWALD
 #undef CALC_COUL_RF
 
 
 /* Tabulated exclusion interaction electrostatics kernels */
 #define CALC_COUL_TAB
+#define LJ_CUT
+#include "nbnxn_kernel_ref_includes.h"
+#undef LJ_CUT
+#define LJ_FORCE_SWITCH
 #include "nbnxn_kernel_ref_includes.h"
-#define VDW_FORCE_SWITCH
+#undef LJ_FORCE_SWITCH
+#define LJ_POT_SWITCH
 #include "nbnxn_kernel_ref_includes.h"
-#undef VDW_FORCE_SWITCH
-#define VDW_POT_SWITCH
+#undef LJ_POT_SWITCH
+#define LJ_EWALD
+#define LJ_CUT
+#define LJ_EWALD_COMB_GEOM
 #include "nbnxn_kernel_ref_includes.h"
-#undef VDW_POT_SWITCH
+#undef LJ_EWALD_COMB_GEOM
+#define LJ_EWALD_COMB_LB
+#include "nbnxn_kernel_ref_includes.h"
+#undef LJ_EWALD_COMB_LB
+#undef LJ_CUT
+#undef LJ_EWALD
 /* Twin-range cut-off kernels */
 #define VDW_CUTOFF_CHECK
+#define LJ_CUT
+#include "nbnxn_kernel_ref_includes.h"
+#undef LJ_CUT
+#define LJ_FORCE_SWITCH
 #include "nbnxn_kernel_ref_includes.h"
-#define VDW_FORCE_SWITCH
+#undef LJ_FORCE_SWITCH
+#define LJ_POT_SWITCH
 #include "nbnxn_kernel_ref_includes.h"
-#undef VDW_FORCE_SWITCH
-#define VDW_POT_SWITCH
+#undef LJ_POT_SWITCH
+#define LJ_EWALD
+#define LJ_CUT
+#define LJ_EWALD_COMB_GEOM
 #include "nbnxn_kernel_ref_includes.h"
-#undef VDW_POT_SWITCH
+#undef LJ_EWALD_COMB_GEOM
+#define LJ_EWALD_COMB_LB
+#include "nbnxn_kernel_ref_includes.h"
+#undef LJ_EWALD_COMB_LB
+#undef LJ_CUT
+#undef LJ_EWALD
 #undef VDW_CUTOFF_CHECK
 #undef CALC_COUL_TAB
 
@@ -105,28 +142,28 @@ enum {
 };
 
 enum {
-    vdwtCUT, vdwtFSWITCH, vdwtPSWITCH, vdwtNR
+    vdwtCUT, vdwtFSWITCH, vdwtPSWITCH, vdwtEWALDGEOM, vdwtEWALDLB, vdwtNR
 };
 
 p_nbk_func_noener p_nbk_c_noener[coultNR][vdwtNR] =
 {
-    { nbnxn_kernel_ElecRF_VdwLJ_F_ref,           nbnxn_kernel_ElecRF_VdwLJFsw_F_ref,           nbnxn_kernel_ElecRF_VdwLJPsw_F_ref           },
-    { nbnxn_kernel_ElecQSTab_VdwLJ_F_ref,        nbnxn_kernel_ElecQSTab_VdwLJFsw_F_ref,        nbnxn_kernel_ElecQSTab_VdwLJPsw_F_ref        },
-    { nbnxn_kernel_ElecQSTabTwinCut_VdwLJ_F_ref, nbnxn_kernel_ElecQSTabTwinCut_VdwLJFsw_F_ref, nbnxn_kernel_ElecQSTabTwinCut_VdwLJPsw_F_ref }
+    { nbnxn_kernel_ElecRF_VdwLJ_F_ref,           nbnxn_kernel_ElecRF_VdwLJFsw_F_ref,           nbnxn_kernel_ElecRF_VdwLJPsw_F_ref,           nbnxn_kernel_ElecRF_VdwLJEwCombGeom_F_ref,           nbnxn_kernel_ElecRF_VdwLJEwCombLB_F_ref           },
+    { nbnxn_kernel_ElecQSTab_VdwLJ_F_ref,        nbnxn_kernel_ElecQSTab_VdwLJFsw_F_ref,        nbnxn_kernel_ElecQSTab_VdwLJPsw_F_ref,        nbnxn_kernel_ElecQSTab_VdwLJEwCombGeom_F_ref,        nbnxn_kernel_ElecQSTab_VdwLJEwCombLB_F_ref        },
+    { nbnxn_kernel_ElecQSTabTwinCut_VdwLJ_F_ref, nbnxn_kernel_ElecQSTabTwinCut_VdwLJFsw_F_ref, nbnxn_kernel_ElecQSTabTwinCut_VdwLJPsw_F_ref, nbnxn_kernel_ElecQSTabTwinCut_VdwLJEwCombGeom_F_ref, nbnxn_kernel_ElecQSTabTwinCut_VdwLJEwCombLB_F_ref }
 };
 
 p_nbk_func_ener p_nbk_c_ener[coultNR][vdwtNR] =
 {
-    { nbnxn_kernel_ElecRF_VdwLJ_VF_ref,           nbnxn_kernel_ElecRF_VdwLJFsw_VF_ref,           nbnxn_kernel_ElecRF_VdwLJPsw_VF_ref           },
-    { nbnxn_kernel_ElecQSTab_VdwLJ_VF_ref,        nbnxn_kernel_ElecQSTab_VdwLJFsw_VF_ref,        nbnxn_kernel_ElecQSTab_VdwLJPsw_VF_ref        },
-    { nbnxn_kernel_ElecQSTabTwinCut_VdwLJ_VF_ref, nbnxn_kernel_ElecQSTabTwinCut_VdwLJFsw_VF_ref, nbnxn_kernel_ElecQSTabTwinCut_VdwLJPsw_VF_ref }
+    { nbnxn_kernel_ElecRF_VdwLJ_VF_ref,           nbnxn_kernel_ElecRF_VdwLJFsw_VF_ref,           nbnxn_kernel_ElecRF_VdwLJPsw_VF_ref,           nbnxn_kernel_ElecRF_VdwLJEwCombGeom_VF_ref,           nbnxn_kernel_ElecRF_VdwLJEwCombLB_VF_ref            },
+    { nbnxn_kernel_ElecQSTab_VdwLJ_VF_ref,        nbnxn_kernel_ElecQSTab_VdwLJFsw_VF_ref,        nbnxn_kernel_ElecQSTab_VdwLJPsw_VF_ref,        nbnxn_kernel_ElecQSTab_VdwLJEwCombGeom_VF_ref,        nbnxn_kernel_ElecQSTab_VdwLJEwCombLB_VF_ref         },
+    { nbnxn_kernel_ElecQSTabTwinCut_VdwLJ_VF_ref, nbnxn_kernel_ElecQSTabTwinCut_VdwLJFsw_VF_ref, nbnxn_kernel_ElecQSTabTwinCut_VdwLJPsw_VF_ref, nbnxn_kernel_ElecQSTabTwinCut_VdwLJEwCombGeom_VF_ref, nbnxn_kernel_ElecQSTabTwinCut_VdwLJEwCombLB_VF_ref  }
 };
 
 p_nbk_func_ener p_nbk_c_energrp[coultNR][vdwtNR] =
 {
-    { nbnxn_kernel_ElecRF_VdwLJ_VgrpF_ref,           nbnxn_kernel_ElecRF_VdwLJFsw_VgrpF_ref,           nbnxn_kernel_ElecRF_VdwLJPsw_VgrpF_ref           },
-    { nbnxn_kernel_ElecQSTab_VdwLJ_VgrpF_ref,        nbnxn_kernel_ElecQSTab_VdwLJFsw_VgrpF_ref,        nbnxn_kernel_ElecQSTab_VdwLJPsw_VgrpF_ref        },
-    { nbnxn_kernel_ElecQSTabTwinCut_VdwLJ_VgrpF_ref, nbnxn_kernel_ElecQSTabTwinCut_VdwLJFsw_VgrpF_ref, nbnxn_kernel_ElecQSTabTwinCut_VdwLJPsw_VgrpF_ref }
+    { nbnxn_kernel_ElecRF_VdwLJ_VgrpF_ref,           nbnxn_kernel_ElecRF_VdwLJFsw_VgrpF_ref,           nbnxn_kernel_ElecRF_VdwLJPsw_VgrpF_ref,           nbnxn_kernel_ElecRF_VdwLJEwCombGeom_VgrpF_ref,           nbnxn_kernel_ElecRF_VdwLJEwCombLB_VgrpF_ref           },
+    { nbnxn_kernel_ElecQSTab_VdwLJ_VgrpF_ref,        nbnxn_kernel_ElecQSTab_VdwLJFsw_VgrpF_ref,        nbnxn_kernel_ElecQSTab_VdwLJPsw_VgrpF_ref,        nbnxn_kernel_ElecQSTab_VdwLJEwCombGeom_VgrpF_ref,        nbnxn_kernel_ElecQSTab_VdwLJEwCombLB_VgrpF_ref        },
+    { nbnxn_kernel_ElecQSTabTwinCut_VdwLJ_VgrpF_ref, nbnxn_kernel_ElecQSTabTwinCut_VdwLJFsw_VgrpF_ref, nbnxn_kernel_ElecQSTabTwinCut_VdwLJPsw_VgrpF_ref, nbnxn_kernel_ElecQSTabTwinCut_VdwLJEwCombGeom_VgrpF_ref, nbnxn_kernel_ElecQSTabTwinCut_VdwLJEwCombLB_VgrpF_ref }
 };
 
 void
@@ -145,6 +182,7 @@ nbnxn_kernel_ref(const nbnxn_pairlist_set_t *nbl_list,
     int                coult;
     int                vdwt;
     int                nb;
+    int                nthreads gmx_unused;
 
     nnbl = nbl_list->nnbl;
     nbl  = nbl_list->nbl;
@@ -165,24 +203,45 @@ nbnxn_kernel_ref(const nbnxn_pairlist_set_t *nbl_list,
         }
     }
 
-    switch (ic->vdw_modifier)
+    if (ic->vdwtype == evdwCUT)
+    {
+        switch (ic->vdw_modifier)
+        {
+            case eintmodPOTSHIFT:
+            case eintmodNONE:
+                vdwt = vdwtCUT;
+                break;
+            case eintmodFORCESWITCH:
+                vdwt = vdwtFSWITCH;
+                break;
+            case eintmodPOTSWITCH:
+                vdwt = vdwtPSWITCH;
+                break;
+            default:
+                gmx_incons("Unsupported VdW modifier");
+                break;
+        }
+    }
+    else if (ic->vdwtype == evdwPME)
+    {
+        if (ic->ljpme_comb_rule == ljcrGEOM)
+        {
+            assert(nbat->comb_rule == ljcrGEOM);
+            vdwt = vdwtEWALDGEOM;
+        }
+        else
+        {
+            assert(nbat->comb_rule == ljcrLB);
+            vdwt = vdwtEWALDLB;
+        }
+    }
+    else
     {
-        case eintmodPOTSHIFT:
-        case eintmodNONE:
-            vdwt = vdwtCUT;
-            break;
-        case eintmodFORCESWITCH:
-            vdwt = vdwtFSWITCH;
-            break;
-        case eintmodPOTSWITCH:
-            vdwt = vdwtPSWITCH;
-            break;
-        default:
-            gmx_incons("Unsupported VdW modifier");
-            break;
+        gmx_incons("Unsupported vdwtype in nbnxn reference kernel");
     }
 
-#pragma omp parallel for schedule(static) num_threads(gmx_omp_nthreads_get(emntNonbonded))
+    nthreads = gmx_omp_nthreads_get(emntNonbonded);
+#pragma omp parallel for schedule(static) num_threads(nthreads)
     for (nb = 0; nb < nnbl; nb++)
     {
         nbnxn_atomdata_output_t *out;