#include "gromacs/utility/gmxassert.h"
#ifdef GMX_NBNXN_SIMD_4XN
-#define GMX_SIMD_J_UNROLL_SIZE 1
-#include "kernel_common.h"
+# define GMX_SIMD_J_UNROLL_SIZE 1
+# include "kernel_common.h"
#endif
/* Prune a single nbnxn_pairtlist_t entry with distance rlistInner */
-void
-nbnxn_kernel_prune_4xn(NbnxnPairlistCpu * nbl,
- const nbnxn_atomdata_t * nbat,
- const rvec * gmx_restrict shift_vec,
- real rlistInner)
+void nbnxn_kernel_prune_4xn(NbnxnPairlistCpu* nbl,
+ const nbnxn_atomdata_t* nbat,
+ const rvec* gmx_restrict shift_vec,
+ real rlistInner)
{
#ifdef GMX_NBNXN_SIMD_4XN
using namespace gmx;
nbl->ci.resize(nbl->ciOuter.size());
nbl->cj.resize(nbl->cjOuter.size());
- const nbnxn_ci_t * gmx_restrict ciOuter = nbl->ciOuter.data();
- nbnxn_ci_t * gmx_restrict ciInner = nbl->ci.data();
+ const nbnxn_ci_t* gmx_restrict ciOuter = nbl->ciOuter.data();
+ nbnxn_ci_t* gmx_restrict ciInner = nbl->ci.data();
- const nbnxn_cj_t * gmx_restrict cjOuter = nbl->cjOuter.data();
- nbnxn_cj_t * gmx_restrict cjInner = nbl->cj.data();
+ const nbnxn_cj_t* gmx_restrict cjOuter = nbl->cjOuter.data();
+ nbnxn_cj_t* gmx_restrict cjInner = nbl->cj.data();
- const real * gmx_restrict shiftvec = shift_vec[0];
- const real * gmx_restrict x = nbat->x().data();
+ const real* gmx_restrict shiftvec = shift_vec[0];
+ const real* gmx_restrict x = nbat->x().data();
- const SimdReal rlist2_S(rlistInner*rlistInner);
+ const SimdReal rlist2_S(rlistInner * rlistInner);
/* Initialize the new list count as empty and add pairs that are in range */
int nciInner = 0;
const int nciOuter = nbl->ciOuter.size();
for (int i = 0; i < nciOuter; i++)
{
- const nbnxn_ci_t * gmx_restrict ciEntry = &ciOuter[i];
+ const nbnxn_ci_t* gmx_restrict ciEntry = &ciOuter[i];
/* Copy the original list entry to the pruned entry */
ciInner[nciInner].ci = ciEntry->ci;
ciInner[nciInner].cj_ind_start = ncjInner;
/* Extract shift data */
- int ish = (ciEntry->shift & NBNXN_CI_SHIFT);
- int ish3 = ish*3;
- int ci = ciEntry->ci;
-
- SimdReal shX_S = SimdReal(shiftvec[ish3 ]);
- SimdReal shY_S = SimdReal(shiftvec[ish3 + 1]);
- SimdReal shZ_S = SimdReal(shiftvec[ish3 + 2]);
-
-#if UNROLLJ <= 4
- int scix = ci*STRIDE*DIM;
-#else
- int scix = (ci >> 1)*STRIDE*DIM + (ci & 1)*(STRIDE >> 1);
-#endif
+ int ish = (ciEntry->shift & NBNXN_CI_SHIFT);
+ int ish3 = ish * 3;
+ int ci = ciEntry->ci;
+
+ SimdReal shX_S = SimdReal(shiftvec[ish3]);
+ SimdReal shY_S = SimdReal(shiftvec[ish3 + 1]);
+ SimdReal shZ_S = SimdReal(shiftvec[ish3 + 2]);
+
+# if UNROLLJ <= 4
+ int scix = ci * STRIDE * DIM;
+# else
+ int scix = (ci >> 1) * STRIDE * DIM + (ci & 1) * (STRIDE >> 1);
+# endif
/* Load i atom data */
- int sciy = scix + STRIDE;
- int sciz = sciy + STRIDE;
- SimdReal ix_S0 = SimdReal(x[scix ]) + shX_S;
- SimdReal ix_S1 = SimdReal(x[scix + 1]) + shX_S;
- SimdReal ix_S2 = SimdReal(x[scix + 2]) + shX_S;
- SimdReal ix_S3 = SimdReal(x[scix + 3]) + shX_S;
- SimdReal iy_S0 = SimdReal(x[sciy ]) + shY_S;
- SimdReal iy_S1 = SimdReal(x[sciy + 1]) + shY_S;
- SimdReal iy_S2 = SimdReal(x[sciy + 2]) + shY_S;
- SimdReal iy_S3 = SimdReal(x[sciy + 3]) + shY_S;
- SimdReal iz_S0 = SimdReal(x[sciz ]) + shZ_S;
- SimdReal iz_S1 = SimdReal(x[sciz + 1]) + shZ_S;
- SimdReal iz_S2 = SimdReal(x[sciz + 2]) + shZ_S;
- SimdReal iz_S3 = SimdReal(x[sciz + 3]) + shZ_S;
+ int sciy = scix + STRIDE;
+ int sciz = sciy + STRIDE;
+ SimdReal ix_S0 = SimdReal(x[scix]) + shX_S;
+ SimdReal ix_S1 = SimdReal(x[scix + 1]) + shX_S;
+ SimdReal ix_S2 = SimdReal(x[scix + 2]) + shX_S;
+ SimdReal ix_S3 = SimdReal(x[scix + 3]) + shX_S;
+ SimdReal iy_S0 = SimdReal(x[sciy]) + shY_S;
+ SimdReal iy_S1 = SimdReal(x[sciy + 1]) + shY_S;
+ SimdReal iy_S2 = SimdReal(x[sciy + 2]) + shY_S;
+ SimdReal iy_S3 = SimdReal(x[sciy + 3]) + shY_S;
+ SimdReal iz_S0 = SimdReal(x[sciz]) + shZ_S;
+ SimdReal iz_S1 = SimdReal(x[sciz + 1]) + shZ_S;
+ SimdReal iz_S2 = SimdReal(x[sciz + 2]) + shZ_S;
+ SimdReal iz_S3 = SimdReal(x[sciz + 3]) + shZ_S;
for (int cjind = ciEntry->cj_ind_start; cjind < ciEntry->cj_ind_end; cjind++)
{
/* j-cluster index */
- int cj = cjOuter[cjind].cj;
+ int cj = cjOuter[cjind].cj;
/* Atom indices (of the first atom in the cluster) */
-#if UNROLLJ == STRIDE
- int aj = cj*UNROLLJ;
- int ajx = aj*DIM;
-#else
- int ajx = (cj >> 1)*DIM*STRIDE + (cj & 1)*UNROLLJ;
-#endif
- int ajy = ajx + STRIDE;
- int ajz = ajy + STRIDE;
+# if UNROLLJ == STRIDE
+ int aj = cj * UNROLLJ;
+ int ajx = aj * DIM;
+# else
+ int ajx = (cj >> 1) * DIM * STRIDE + (cj & 1) * UNROLLJ;
+# endif
+ int ajy = ajx + STRIDE;
+ int ajz = ajy + STRIDE;
/* load j atom coordinates */
- SimdReal jx_S = load<SimdReal>(x + ajx);
- SimdReal jy_S = load<SimdReal>(x + ajy);
- SimdReal jz_S = load<SimdReal>(x + ajz);
+ SimdReal jx_S = load<SimdReal>(x + ajx);
+ SimdReal jy_S = load<SimdReal>(x + ajy);
+ SimdReal jz_S = load<SimdReal>(x + ajz);
/* Calculate distance */
- SimdReal dx_S0 = ix_S0 - jx_S;
- SimdReal dy_S0 = iy_S0 - jy_S;
- SimdReal dz_S0 = iz_S0 - jz_S;
- SimdReal dx_S1 = ix_S1 - jx_S;
- SimdReal dy_S1 = iy_S1 - jy_S;
- SimdReal dz_S1 = iz_S1 - jz_S;
- SimdReal dx_S2 = ix_S2 - jx_S;
- SimdReal dy_S2 = iy_S2 - jy_S;
- SimdReal dz_S2 = iz_S2 - jz_S;
- SimdReal dx_S3 = ix_S3 - jx_S;
- SimdReal dy_S3 = iy_S3 - jy_S;
- SimdReal dz_S3 = iz_S3 - jz_S;
+ SimdReal dx_S0 = ix_S0 - jx_S;
+ SimdReal dy_S0 = iy_S0 - jy_S;
+ SimdReal dz_S0 = iz_S0 - jz_S;
+ SimdReal dx_S1 = ix_S1 - jx_S;
+ SimdReal dy_S1 = iy_S1 - jy_S;
+ SimdReal dz_S1 = iz_S1 - jz_S;
+ SimdReal dx_S2 = ix_S2 - jx_S;
+ SimdReal dy_S2 = iy_S2 - jy_S;
+ SimdReal dz_S2 = iz_S2 - jz_S;
+ SimdReal dx_S3 = ix_S3 - jx_S;
+ SimdReal dy_S3 = iy_S3 - jy_S;
+ SimdReal dz_S3 = iz_S3 - jz_S;
/* rsq = dx*dx+dy*dy+dz*dz */
SimdReal rsq_S0 = norm2(dx_S0, dy_S0, dz_S0);
SimdBool wco_S2 = (rsq_S2 < rlist2_S);
SimdBool wco_S3 = (rsq_S3 < rlist2_S);
- wco_S0 = wco_S0 || wco_S1;
- wco_S2 = wco_S2 || wco_S3;
- wco_S0 = wco_S0 || wco_S2;
+ wco_S0 = wco_S0 || wco_S1;
+ wco_S2 = wco_S2 || wco_S3;
+ wco_S0 = wco_S0 || wco_S2;
/* Putting the assignment inside the conditional is slower */
cjInner[ncjInner] = cjOuter[cjind];
nbl->ci.resize(nciInner);
nbl->cj.resize(ncjInner);
-#else /* GMX_NBNXN_SIMD_4XN */
+#else /* GMX_NBNXN_SIMD_4XN */
GMX_RELEASE_ASSERT(false, "4xN kernel called without 4xN support");
biod.pnpi.spb.ru / alexxy / gromacs.git / blobdiff