Merge release-5-0 into master

[alexxy/gromacs.git] / src / gromacs / mdlib / nbnxn_cuda / nbnxn_cuda_data_mgmt.cu

diff --git a/src/gromacs/mdlib/nbnxn_cuda/nbnxn_cuda_data_mgmt.cu b/src/gromacs/mdlib/nbnxn_cuda/nbnxn_cuda_data_mgmt.cu
index dfdafbb2fd9335d21e1501cd095144bfc606c576..6612d13097c1625da9b67b864b5581ec51179fa3 100644 (file)
--- a/src/gromacs/mdlib/nbnxn_cuda/nbnxn_cuda_data_mgmt.cu
+++ b/src/gromacs/mdlib/nbnxn_cuda/nbnxn_cuda_data_mgmt.cu
@@ -32,34 +32,37 @@
  * To help us fund GROMACS development, we humbly ask that you cite
  * the research papers on the package. Check out http://www.gromacs.org.
  */
-#ifdef HAVE_CONFIG_H
-#include <config.h>
-#endif
+#include "gmxpre.h"
+
+#include "nbnxn_cuda_data_mgmt.h"
+
+#include "config.h"
 
-#include <stdlib.h>
-#include <stdio.h>
 #include <assert.h>
+#include <stdarg.h>
+#include <stdio.h>
+#include <stdlib.h>
 
 #include <cuda.h>
 
-#include "gmx_fatal.h"
-#include "smalloc.h"
-#include "tables.h"
-#include "typedefs.h"
-#include "types/enums.h"
-#include "types/nb_verlet.h"
-#include "types/interaction_const.h"
-#include "types/force_flags.h"
-#include "../nbnxn_consts.h"
-#include "gmx_detect_hardware.h"
+#include "gromacs/gmxlib/cuda_tools/cudautils.cuh"
+#include "gromacs/legacyheaders/gmx_detect_hardware.h"
+#include "gromacs/legacyheaders/gpu_utils.h"
+#include "gromacs/legacyheaders/pmalloc_cuda.h"
+#include "gromacs/legacyheaders/tables.h"
+#include "gromacs/legacyheaders/typedefs.h"
+#include "gromacs/legacyheaders/types/enums.h"
+#include "gromacs/legacyheaders/types/force_flags.h"
+#include "gromacs/legacyheaders/types/interaction_const.h"
+#include "gromacs/mdlib/nb_verlet.h"
+#include "gromacs/mdlib/nbnxn_consts.h"
+#include "gromacs/pbcutil/ishift.h"
+#include "gromacs/utility/common.h"
+#include "gromacs/utility/cstringutil.h"
+#include "gromacs/utility/fatalerror.h"
+#include "gromacs/utility/smalloc.h"
 
 #include "nbnxn_cuda_types.h"
-#include "../../gmxlib/cuda_tools/cudautils.cuh"
-#include "nbnxn_cuda_data_mgmt.h"
-#include "pmalloc_cuda.h"
-#include "gpu_utils.h"
-
-#include "gromacs/utility/common.h"
 
 static bool bUseCudaEventBlockingSync = false; /* makes the CPU thread block */
 
@@ -126,7 +129,7 @@ static void init_ewald_coulomb_force_table(cu_nbparam_t          *nbp,
     pmalloc((void**)&ftmp, tabsize*sizeof(*ftmp));
 
     table_spline3_fill_ewald_lr(ftmp, NULL, NULL, tabsize,
-                                1/tabscale, nbp->ewald_beta);
+                                1/tabscale, nbp->ewald_beta, v_q_ewald_lr);
 
     /* If the table pointer == NULL the table is generated the first time =>
        the array pointer will be saved to nbparam and the texture is bound.
@@ -259,18 +262,10 @@ static int pick_ewald_kernel_type(bool                   bTwinCut,
     return kernel_type;
 }
 
-
-/*! Initializes the nonbonded parameter data structure. */
-static void init_nbparam(cu_nbparam_t              *nbp,
-                         const interaction_const_t *ic,
-                         const nbnxn_atomdata_t    *nbat,
-                         const cuda_dev_info_t     *dev_info)
+/*! Copies all parameters related to the cut-off from ic to nbp */
+static void set_cutoff_parameters(cu_nbparam_t              *nbp,
+                                  const interaction_const_t *ic)
 {
-    cudaError_t stat;
-    int         ntypes, nnbfp, nnbfp_comb;
-
-    ntypes  = nbat->ntype;
-
     nbp->ewald_beta       = ic->ewaldcoeff_q;
     nbp->sh_ewald         = ic->sh_ewald;
     nbp->epsfac           = ic->epsfac;
@@ -287,6 +282,20 @@ static void init_nbparam(cu_nbparam_t              *nbp,
     nbp->dispersion_shift = ic->dispersion_shift;
     nbp->repulsion_shift  = ic->repulsion_shift;
     nbp->vdw_switch       = ic->vdw_switch;
+}
+
+/*! Initializes the nonbonded parameter data structure. */
+static void init_nbparam(cu_nbparam_t              *nbp,
+                         const interaction_const_t *ic,
+                         const nbnxn_atomdata_t    *nbat,
+                         const cuda_dev_info_t     *dev_info)
+{
+    cudaError_t stat;
+    int         ntypes, nnbfp, nnbfp_comb;
+
+    ntypes  = nbat->ntype;
+
+    set_cutoff_parameters(nbp, ic);
 
     if (ic->vdwtype == evdwCUT)
     {
@@ -419,14 +428,17 @@ static void init_nbparam(cu_nbparam_t              *nbp,
 
 /*! Re-generate the GPU Ewald force table, resets rlist, and update the
  *  electrostatic type switching to twin cut-off (or back) if needed. */
-void nbnxn_cuda_pme_loadbal_update_param(nbnxn_cuda_ptr_t           cu_nb,
-                                         const interaction_const_t *ic)
+void nbnxn_cuda_pme_loadbal_update_param(const nonbonded_verlet_t    *nbv,
+                                         const interaction_const_t   *ic)
 {
-    cu_nbparam_t *nbp = cu_nb->nbparam;
+    if (!nbv || nbv->grp[0].kernel_type != nbnxnk8x8x8_CUDA)
+    {
+        return;
+    }
+    nbnxn_cuda_ptr_t cu_nb = nbv->cu_nbv;
+    cu_nbparam_t    *nbp   = cu_nb->nbparam;
 
-    nbp->rlist_sq       = ic->rlist * ic->rlist;
-    nbp->rcoulomb_sq    = ic->rcoulomb * ic->rcoulomb;
-    nbp->ewald_beta     = ic->ewaldcoeff_q;
+    set_cutoff_parameters(nbp, ic);
 
     nbp->eeltype        = pick_ewald_kernel_type(ic->rcoulomb != ic->rvdw,
                                                  cu_nb->dev_info);
@@ -567,7 +579,7 @@ void nbnxn_cuda_init(FILE                 *fplog,
          * priorities, because we are querying the priority range which in this
          * case will be a single value.
          */
-#if CUDA_VERSION >= 5500
+#if CUDA_VERSION >= 5050
         {
             int highest_priority;
             stat = cudaDeviceGetStreamPriorityRange(NULL, &highest_priority);
@@ -1072,11 +1084,11 @@ wallclock_gpu_t * nbnxn_cuda_get_timings(nbnxn_cuda_ptr_t cu_nb)
     return (cu_nb != NULL && cu_nb->bDoTime) ? cu_nb->timings : NULL;
 }
 
-void nbnxn_cuda_reset_timings(nbnxn_cuda_ptr_t cu_nb)
+void nbnxn_cuda_reset_timings(nonbonded_verlet_t* nbv)
 {
-    if (cu_nb->bDoTime)
+    if (nbv->cu_nbv && nbv->cu_nbv->bDoTime)
     {
-        init_timings(cu_nb->timings);
+        init_timings(nbv->cu_nbv->timings);
     }
 }