#endif
/* Without exclusions and energies we only need to mask the cut-off,
- * this can be faster with blendv (only available with SSE4.1 and later).
+ * this can be faster with blendv.
*/
-#if !(defined CHECK_EXCLS || defined CALC_ENERGIES) && defined GMX_X86_SSE4_1 && !defined COUNT_PAIRS
+#if !(defined CHECK_EXCLS || defined CALC_ENERGIES) && defined GMX_HAVE_SIMD_BLENDV && !defined COUNT_PAIRS
/* With RF and tabulated Coulomb we replace cmp+and with sub+blendv.
* With gcc this is slower, except for RF on Sandy Bridge.
* Tested with gcc 4.6.2, 4.6.3 and 4.7.1.
#ifdef CHECK_EXCLS
/* Interaction (non-exclusion) mask of all 1's or 0's */
- gmx_mm_pr int_SSE0;
- gmx_mm_pr int_SSE2;
+ gmx_mm_pr int_S0;
+ gmx_mm_pr int_S2;
#endif
- gmx_mm_pr jxSSE, jySSE, jzSSE;
- gmx_mm_pr dx_SSE0, dy_SSE0, dz_SSE0;
- gmx_mm_pr dx_SSE2, dy_SSE2, dz_SSE2;
- gmx_mm_pr tx_SSE0, ty_SSE0, tz_SSE0;
- gmx_mm_pr tx_SSE2, ty_SSE2, tz_SSE2;
- gmx_mm_pr rsq_SSE0, rinv_SSE0, rinvsq_SSE0;
- gmx_mm_pr rsq_SSE2, rinv_SSE2, rinvsq_SSE2;
+ gmx_mm_pr jx_S, jy_S, jz_S;
+ gmx_mm_pr dx_S0, dy_S0, dz_S0;
+ gmx_mm_pr dx_S2, dy_S2, dz_S2;
+ gmx_mm_pr tx_S0, ty_S0, tz_S0;
+ gmx_mm_pr tx_S2, ty_S2, tz_S2;
+ gmx_mm_pr rsq_S0, rinv_S0, rinvsq_S0;
+ gmx_mm_pr rsq_S2, rinv_S2, rinvsq_S2;
#ifndef CUTOFF_BLENDV
/* wco: within cut-off, mask of all 1's or 0's */
- gmx_mm_pr wco_SSE0;
- gmx_mm_pr wco_SSE2;
+ gmx_mm_pr wco_S0;
+ gmx_mm_pr wco_S2;
#endif
#ifdef VDW_CUTOFF_CHECK
- gmx_mm_pr wco_vdw_SSE0;
+ gmx_mm_pr wco_vdw_S0;
#ifndef HALF_LJ
- gmx_mm_pr wco_vdw_SSE2;
+ gmx_mm_pr wco_vdw_S2;
#endif
#endif
#ifdef CALC_COULOMB
#ifdef CHECK_EXCLS
/* 1/r masked with the interaction mask */
- gmx_mm_pr rinv_ex_SSE0;
- gmx_mm_pr rinv_ex_SSE2;
+ gmx_mm_pr rinv_ex_S0;
+ gmx_mm_pr rinv_ex_S2;
#endif
- gmx_mm_pr jq_SSE;
- gmx_mm_pr qq_SSE0;
- gmx_mm_pr qq_SSE2;
+ gmx_mm_pr jq_S;
+ gmx_mm_pr qq_S0;
+ gmx_mm_pr qq_S2;
#ifdef CALC_COUL_TAB
/* The force (PME mesh force) we need to subtract from 1/r^2 */
- gmx_mm_pr fsub_SSE0;
- gmx_mm_pr fsub_SSE2;
+ gmx_mm_pr fsub_S0;
+ gmx_mm_pr fsub_S2;
#endif
#ifdef CALC_COUL_EWALD
- gmx_mm_pr brsq_SSE0, brsq_SSE2;
- gmx_mm_pr ewcorr_SSE0, ewcorr_SSE2;
+ gmx_mm_pr brsq_S0, brsq_S2;
+ gmx_mm_pr ewcorr_S0, ewcorr_S2;
#endif
/* frcoul = (1/r - fsub)*r */
- gmx_mm_pr frcoul_SSE0;
- gmx_mm_pr frcoul_SSE2;
+ gmx_mm_pr frcoul_S0;
+ gmx_mm_pr frcoul_S2;
#ifdef CALC_COUL_TAB
/* For tables: r, rs=r/sp, rf=floor(rs), frac=rs-rf */
- gmx_mm_pr r_SSE0, rs_SSE0, rf_SSE0, frac_SSE0;
- gmx_mm_pr r_SSE2, rs_SSE2, rf_SSE2, frac_SSE2;
+ gmx_mm_pr r_S0, rs_S0, rf_S0, frac_S0;
+ gmx_mm_pr r_S2, rs_S2, rf_S2, frac_S2;
/* Table index: rs truncated to an int */
-#if !(defined GMX_MM256_HERE && defined GMX_DOUBLE)
- gmx_epi32 ti_SSE0, ti_SSE2;
-#else
- __m128i ti_SSE0, ti_SSE2;
-#endif
+ gmx_epi32 ti_S0, ti_S2;
/* Linear force table values */
- gmx_mm_pr ctab0_SSE0, ctab1_SSE0;
- gmx_mm_pr ctab0_SSE2, ctab1_SSE2;
+ gmx_mm_pr ctab0_S0, ctab1_S0;
+ gmx_mm_pr ctab0_S2, ctab1_S2;
#ifdef CALC_ENERGIES
/* Quadratic energy table value */
- gmx_mm_pr ctabv_SSE0;
- gmx_mm_pr ctabv_SSE2;
+ gmx_mm_pr ctabv_S0;
+ gmx_mm_pr ctabv_S2;
#endif
#endif
#if defined CALC_ENERGIES && (defined CALC_COUL_EWALD || defined CALC_COUL_TAB)
/* The potential (PME mesh) we need to subtract from 1/r */
- gmx_mm_pr vc_sub_SSE0;
- gmx_mm_pr vc_sub_SSE2;
+ gmx_mm_pr vc_sub_S0;
+ gmx_mm_pr vc_sub_S2;
#endif
#ifdef CALC_ENERGIES
/* Electrostatic potential */
- gmx_mm_pr vcoul_SSE0;
- gmx_mm_pr vcoul_SSE2;
+ gmx_mm_pr vcoul_S0;
+ gmx_mm_pr vcoul_S2;
#endif
#endif
/* The force times 1/r */
- gmx_mm_pr fscal_SSE0;
- gmx_mm_pr fscal_SSE2;
+ gmx_mm_pr fscal_S0;
+ gmx_mm_pr fscal_S2;
#ifdef CALC_LJ
#ifdef LJ_COMB_LB
/* LJ sigma_j/2 and sqrt(epsilon_j) */
- gmx_mm_pr hsig_j_SSE, seps_j_SSE;
+ gmx_mm_pr hsig_j_S, seps_j_S;
/* LJ sigma_ij and epsilon_ij */
- gmx_mm_pr sig_SSE0, eps_SSE0;
+ gmx_mm_pr sig_S0, eps_S0;
#ifndef HALF_LJ
- gmx_mm_pr sig_SSE2, eps_SSE2;
+ gmx_mm_pr sig_S2, eps_S2;
#endif
#ifdef CALC_ENERGIES
- gmx_mm_pr sig2_SSE0, sig6_SSE0;
+ gmx_mm_pr sig2_S0, sig6_S0;
#ifndef HALF_LJ
- gmx_mm_pr sig2_SSE2, sig6_SSE2;
+ gmx_mm_pr sig2_S2, sig6_S2;
#endif
#endif /* LJ_COMB_LB */
#endif /* CALC_LJ */
#ifdef LJ_COMB_GEOM
- gmx_mm_pr c6s_j_SSE, c12s_j_SSE;
+ gmx_mm_pr c6s_j_S, c12s_j_S;
#endif
#if defined LJ_COMB_GEOM || defined LJ_COMB_LB
#ifndef FIX_LJ_C
/* LJ C6 and C12 parameters, used with geometric comb. rule */
- gmx_mm_pr c6_SSE0, c12_SSE0;
+ gmx_mm_pr c6_S0, c12_S0;
#ifndef HALF_LJ
- gmx_mm_pr c6_SSE2, c12_SSE2;
+ gmx_mm_pr c6_S2, c12_S2;
#endif
#endif
/* Intermediate variables for LJ calculation */
#ifndef LJ_COMB_LB
- gmx_mm_pr rinvsix_SSE0;
+ gmx_mm_pr rinvsix_S0;
#ifndef HALF_LJ
- gmx_mm_pr rinvsix_SSE2;
+ gmx_mm_pr rinvsix_S2;
#endif
#endif
#ifdef LJ_COMB_LB
- gmx_mm_pr sir_SSE0, sir2_SSE0, sir6_SSE0;
+ gmx_mm_pr sir_S0, sir2_S0, sir6_S0;
#ifndef HALF_LJ
- gmx_mm_pr sir_SSE2, sir2_SSE2, sir6_SSE2;
+ gmx_mm_pr sir_S2, sir2_S2, sir6_S2;
#endif
#endif
- gmx_mm_pr FrLJ6_SSE0, FrLJ12_SSE0;
+ gmx_mm_pr FrLJ6_S0, FrLJ12_S0;
#ifndef HALF_LJ
- gmx_mm_pr FrLJ6_SSE2, FrLJ12_SSE2;
+ gmx_mm_pr FrLJ6_S2, FrLJ12_S2;
#endif
#ifdef CALC_ENERGIES
- gmx_mm_pr VLJ6_SSE0, VLJ12_SSE0, VLJ_SSE0;
+ gmx_mm_pr VLJ6_S0, VLJ12_S0, VLJ_S0;
#ifndef HALF_LJ
- gmx_mm_pr VLJ6_SSE2, VLJ12_SSE2, VLJ_SSE2;
+ gmx_mm_pr VLJ6_S2, VLJ12_S2, VLJ_S2;
#endif
#endif
#endif /* CALC_LJ */
+ gmx_mm_hpr fjx_S, fjy_S, fjz_S;
+
/* j-cluster index */
cj = l_cj[cjind].cj;
#ifdef CHECK_EXCLS
{
/* Load integer interaction mask */
- /* With AVX there are no integer operations, so cast to real */
- gmx_mm_pr mask_pr = gmx_mm_castsi256_pr(_mm256_set1_epi32(l_cj[cjind].excl));
- /* Intel Compiler version 12.1.3 20120130 is buggy: use cast.
- * With gcc we don't need the cast, but it's faster.
- */
-#define cast_cvt(x) _mm256_cvtepi32_ps(_mm256_castps_si256(x))
- int_SSE0 = gmx_cmpneq_pr(cast_cvt(gmx_and_pr(mask_pr, mask0)), zero_SSE);
- int_SSE2 = gmx_cmpneq_pr(cast_cvt(gmx_and_pr(mask_pr, mask2)), zero_SSE);
-#undef cast_cvt
+ gmx_mm_pr mask_pr_S = gmx_castsi_pr(gmx_set1_epi32(l_cj[cjind].excl));
+
+ int_S0 = gmx_checkbitmask_pr(mask_pr_S, mask_S0);
+ int_S2 = gmx_checkbitmask_pr(mask_pr_S, mask_S2);
}
#endif
+
/* load j atom coordinates */
- jxSSE = gmx_loaddh_pr(x+ajx);
- jySSE = gmx_loaddh_pr(x+ajy);
- jzSSE = gmx_loaddh_pr(x+ajz);
+ gmx_loaddh_pr(jx_S, x+ajx);
+ gmx_loaddh_pr(jy_S, x+ajy);
+ gmx_loaddh_pr(jz_S, x+ajz);
/* Calculate distance */
- dx_SSE0 = gmx_sub_pr(ix_SSE0, jxSSE);
- dy_SSE0 = gmx_sub_pr(iy_SSE0, jySSE);
- dz_SSE0 = gmx_sub_pr(iz_SSE0, jzSSE);
- dx_SSE2 = gmx_sub_pr(ix_SSE2, jxSSE);
- dy_SSE2 = gmx_sub_pr(iy_SSE2, jySSE);
- dz_SSE2 = gmx_sub_pr(iz_SSE2, jzSSE);
+ dx_S0 = gmx_sub_pr(ix_S0, jx_S);
+ dy_S0 = gmx_sub_pr(iy_S0, jy_S);
+ dz_S0 = gmx_sub_pr(iz_S0, jz_S);
+ dx_S2 = gmx_sub_pr(ix_S2, jx_S);
+ dy_S2 = gmx_sub_pr(iy_S2, jy_S);
+ dz_S2 = gmx_sub_pr(iz_S2, jz_S);
/* rsq = dx*dx+dy*dy+dz*dz */
- rsq_SSE0 = gmx_calc_rsq_pr(dx_SSE0, dy_SSE0, dz_SSE0);
- rsq_SSE2 = gmx_calc_rsq_pr(dx_SSE2, dy_SSE2, dz_SSE2);
+ rsq_S0 = gmx_calc_rsq_pr(dx_S0, dy_S0, dz_S0);
+ rsq_S2 = gmx_calc_rsq_pr(dx_S2, dy_S2, dz_S2);
#ifndef CUTOFF_BLENDV
- wco_SSE0 = gmx_cmplt_pr(rsq_SSE0, rc2_SSE);
- wco_SSE2 = gmx_cmplt_pr(rsq_SSE2, rc2_SSE);
+ wco_S0 = gmx_cmplt_pr(rsq_S0, rc2_S);
+ wco_S2 = gmx_cmplt_pr(rsq_S2, rc2_S);
#endif
#ifdef CHECK_EXCLS
#if UNROLLJ == UNROLLI
if (cj == ci_sh)
{
- wco_SSE0 = gmx_and_pr(wco_SSE0, diag_SSE0);
- wco_SSE2 = gmx_and_pr(wco_SSE2, diag_SSE2);
+ wco_S0 = gmx_and_pr(wco_S0, diag_S0);
+ wco_S2 = gmx_and_pr(wco_S2, diag_S2);
}
#else
-#error "only UNROLLJ == UNROLLI currently supported in the joined kernels"
+#if UNROLLJ == 2*UNROLLI
+ if (cj*2 == ci_sh)
+ {
+ wco_S0 = gmx_and_pr(wco_S0, diag0_S0);
+ wco_S2 = gmx_and_pr(wco_S2, diag0_S2);
+ }
+ else if (cj*2 + 1 == ci_sh)
+ {
+ wco_S0 = gmx_and_pr(wco_S0, diag1_S0);
+ wco_S2 = gmx_and_pr(wco_S2, diag1_S2);
+ }
+#else
+#error "only UNROLLJ == UNROLLI*(1 or 2) currently supported in 2xnn kernels"
+#endif
#endif
#else /* EXCL_FORCES */
- /* Remove all excluded atom pairs from the list */
- wco_SSE0 = gmx_and_pr(wco_SSE0, int_SSE0);
- wco_SSE2 = gmx_and_pr(wco_SSE2, int_SSE2);
+ /* No exclusion forces: remove all excluded atom pairs from the list */
+ wco_S0 = gmx_and_pr(wco_S0, int_S0);
+ wco_S2 = gmx_and_pr(wco_S2, int_S2);
#endif
#endif
#ifdef COUNT_PAIRS
{
int i, j;
- real tmp[UNROLLJ];
- for (i = 0; i < UNROLLI; i++)
+ real tmpa[2*GMX_SIMD_WIDTH_HERE], *tmp;
+ tmp = gmx_simd_align_real(tmpa);
+ for (i = 0; i < UNROLLI; i+=2)
{
- gmx_storeu_pr(tmp, i == 0 ? wco_SSE0 : (i == 1 ? wco_SSE1 : (i == 2 ? wco_SSE2 : wco_SSE3)));
- for (j = 0; j < UNROLLJ; j++)
+ gmx_store_pr(tmp, i == 0 ? wco_S0 : wco_S2);
+ for (j = 0; j < 2*UNROLLJ; j++)
{
if (!(tmp[j] == 0))
{
#ifdef CHECK_EXCLS
/* For excluded pairs add a small number to avoid r^-6 = NaN */
- rsq_SSE0 = gmx_add_pr(rsq_SSE0, gmx_andnot_pr(int_SSE0, avoid_sing_SSE));
- rsq_SSE2 = gmx_add_pr(rsq_SSE2, gmx_andnot_pr(int_SSE2, avoid_sing_SSE));
+ rsq_S0 = gmx_add_pr(rsq_S0, gmx_andnot_pr(int_S0, avoid_sing_S));
+ rsq_S2 = gmx_add_pr(rsq_S2, gmx_andnot_pr(int_S2, avoid_sing_S));
#endif
/* Calculate 1/r */
- rinv_SSE0 = gmx_invsqrt_pr(rsq_SSE0);
- rinv_SSE2 = gmx_invsqrt_pr(rsq_SSE2);
+ rinv_S0 = gmx_invsqrt_pr(rsq_S0);
+ rinv_S2 = gmx_invsqrt_pr(rsq_S2);
#ifdef CALC_COULOMB
/* Load parameters for j atom */
- jq_SSE = gmx_loaddh_pr(q+aj);
- qq_SSE0 = gmx_mul_pr(iq_SSE0, jq_SSE);
- qq_SSE2 = gmx_mul_pr(iq_SSE2, jq_SSE);
+ gmx_loaddh_pr(jq_S, q+aj);
+ qq_S0 = gmx_mul_pr(iq_S0, jq_S);
+ qq_S2 = gmx_mul_pr(iq_S2, jq_S);
#endif
#ifdef CALC_LJ
#if !defined LJ_COMB_GEOM && !defined LJ_COMB_LB && !defined FIX_LJ_C
- load_lj_pair_params2(nbfp0, nbfp1, type, aj, c6_SSE0, c12_SSE0);
+ load_lj_pair_params2(nbfp0, nbfp1, type, aj, c6_S0, c12_S0);
#ifndef HALF_LJ
- load_lj_pair_params2(nbfp2, nbfp3, type, aj, c6_SSE2, c12_SSE2);
+ load_lj_pair_params2(nbfp2, nbfp3, type, aj, c6_S2, c12_S2);
#endif
#endif /* not defined any LJ rule */
#ifdef LJ_COMB_GEOM
- c6s_j_SSE = gmx_loaddh_pr(ljc+aj2+0);
- c12s_j_SSE = gmx_loaddh_pr(ljc+aj2+STRIDE);
- c6_SSE0 = gmx_mul_pr(c6s_SSE0, c6s_j_SSE );
+ gmx_loaddh_pr(c6s_j_S, ljc+aj2+0);
+ gmx_loaddh_pr(c12s_j_S, ljc+aj2+STRIDE);
+ c6_S0 = gmx_mul_pr(c6s_S0, c6s_j_S );
#ifndef HALF_LJ
- c6_SSE2 = gmx_mul_pr(c6s_SSE2, c6s_j_SSE );
+ c6_S2 = gmx_mul_pr(c6s_S2, c6s_j_S );
#endif
- c12_SSE0 = gmx_mul_pr(c12s_SSE0, c12s_j_SSE);
+ c12_S0 = gmx_mul_pr(c12s_S0, c12s_j_S);
#ifndef HALF_LJ
- c12_SSE2 = gmx_mul_pr(c12s_SSE2, c12s_j_SSE);
+ c12_S2 = gmx_mul_pr(c12s_S2, c12s_j_S);
#endif
#endif /* LJ_COMB_GEOM */
#ifdef LJ_COMB_LB
- hsig_j_SSE = gmx_loaddh_pr(ljc+aj2+0);
- seps_j_SSE = gmx_loaddh_pr(ljc+aj2+STRIDE);
+ gmx_loaddh_pr(hsig_j_S, ljc+aj2+0);
+ gmx_loaddh_pr(seps_j_S, ljc+aj2+STRIDE);
- sig_SSE0 = gmx_add_pr(hsig_i_SSE0, hsig_j_SSE);
- eps_SSE0 = gmx_mul_pr(seps_i_SSE0, seps_j_SSE);
+ sig_S0 = gmx_add_pr(hsig_i_S0, hsig_j_S);
+ eps_S0 = gmx_mul_pr(seps_i_S0, seps_j_S);
#ifndef HALF_LJ
- sig_SSE2 = gmx_add_pr(hsig_i_SSE2, hsig_j_SSE);
- eps_SSE2 = gmx_mul_pr(seps_i_SSE2, seps_j_SSE);
+ sig_S2 = gmx_add_pr(hsig_i_S2, hsig_j_S);
+ eps_S2 = gmx_mul_pr(seps_i_S2, seps_j_S);
#endif
#endif /* LJ_COMB_LB */
#endif /* CALC_LJ */
#ifndef CUTOFF_BLENDV
- rinv_SSE0 = gmx_and_pr(rinv_SSE0, wco_SSE0);
- rinv_SSE2 = gmx_and_pr(rinv_SSE2, wco_SSE2);
+ rinv_S0 = gmx_blendzero_pr(rinv_S0, wco_S0);
+ rinv_S2 = gmx_blendzero_pr(rinv_S2, wco_S2);
#else
/* We only need to mask for the cut-off: blendv is faster */
- rinv_SSE0 = gmx_blendv_pr(rinv_SSE0, zero_SSE, gmx_sub_pr(rc2_SSE, rsq_SSE0));
- rinv_SSE2 = gmx_blendv_pr(rinv_SSE2, zero_SSE, gmx_sub_pr(rc2_SSE, rsq_SSE2));
+ rinv_S0 = gmx_blendv_pr(rinv_S0, zero_S, gmx_sub_pr(rc2_S, rsq_S0));
+ rinv_S2 = gmx_blendv_pr(rinv_S2, zero_S, gmx_sub_pr(rc2_S, rsq_S2));
#endif
- rinvsq_SSE0 = gmx_mul_pr(rinv_SSE0, rinv_SSE0);
- rinvsq_SSE2 = gmx_mul_pr(rinv_SSE2, rinv_SSE2);
+ rinvsq_S0 = gmx_mul_pr(rinv_S0, rinv_S0);
+ rinvsq_S2 = gmx_mul_pr(rinv_S2, rinv_S2);
#ifdef CALC_COULOMB
/* Note that here we calculate force*r, not the usual force/r.
#ifdef EXCL_FORCES
/* Only add 1/r for non-excluded atom pairs */
- rinv_ex_SSE0 = gmx_and_pr(rinv_SSE0, int_SSE0);
- rinv_ex_SSE2 = gmx_and_pr(rinv_SSE2, int_SSE2);
+ rinv_ex_S0 = gmx_blendzero_pr(rinv_S0, int_S0);
+ rinv_ex_S2 = gmx_blendzero_pr(rinv_S2, int_S2);
#else
/* No exclusion forces, we always need 1/r */
-#define rinv_ex_SSE0 rinv_SSE0
-#define rinv_ex_SSE2 rinv_SSE2
+#define rinv_ex_S0 rinv_S0
+#define rinv_ex_S2 rinv_S2
#endif
#ifdef CALC_COUL_RF
/* Electrostatic interactions */
- frcoul_SSE0 = gmx_mul_pr(qq_SSE0, gmx_add_pr(rinv_ex_SSE0, gmx_mul_pr(rsq_SSE0, mrc_3_SSE)));
- frcoul_SSE2 = gmx_mul_pr(qq_SSE2, gmx_add_pr(rinv_ex_SSE2, gmx_mul_pr(rsq_SSE2, mrc_3_SSE)));
+ frcoul_S0 = gmx_mul_pr(qq_S0, gmx_add_pr(rinv_ex_S0, gmx_mul_pr(rsq_S0, mrc_3_S)));
+ frcoul_S2 = gmx_mul_pr(qq_S2, gmx_add_pr(rinv_ex_S2, gmx_mul_pr(rsq_S2, mrc_3_S)));
#ifdef CALC_ENERGIES
- vcoul_SSE0 = gmx_mul_pr(qq_SSE0, gmx_add_pr(rinv_ex_SSE0, gmx_add_pr(gmx_mul_pr(rsq_SSE0, hrc_3_SSE), moh_rc_SSE)));
- vcoul_SSE2 = gmx_mul_pr(qq_SSE2, gmx_add_pr(rinv_ex_SSE2, gmx_add_pr(gmx_mul_pr(rsq_SSE2, hrc_3_SSE), moh_rc_SSE)));
+ vcoul_S0 = gmx_mul_pr(qq_S0, gmx_add_pr(rinv_ex_S0, gmx_add_pr(gmx_mul_pr(rsq_S0, hrc_3_S), moh_rc_S)));
+ vcoul_S2 = gmx_mul_pr(qq_S2, gmx_add_pr(rinv_ex_S2, gmx_add_pr(gmx_mul_pr(rsq_S2, hrc_3_S), moh_rc_S)));
#endif
#endif
* as large distances can cause an overflow in gmx_pmecorrF/V.
*/
#ifndef CUTOFF_BLENDV
- brsq_SSE0 = gmx_mul_pr(beta2_SSE, gmx_and_pr(rsq_SSE0, wco_SSE0));
- brsq_SSE2 = gmx_mul_pr(beta2_SSE, gmx_and_pr(rsq_SSE2, wco_SSE2));
+ brsq_S0 = gmx_mul_pr(beta2_S, gmx_blendzero_pr(rsq_S0, wco_S0));
+ brsq_S2 = gmx_mul_pr(beta2_S, gmx_blendzero_pr(rsq_S2, wco_S2));
#else
/* Strangely, putting mul on a separate line is slower (icc 13) */
- brsq_SSE0 = gmx_mul_pr(beta2_SSE, gmx_blendv_pr(rsq_SSE0, zero_SSE, gmx_sub_pr(rc2_SSE, rsq_SSE0)));
- brsq_SSE2 = gmx_mul_pr(beta2_SSE, gmx_blendv_pr(rsq_SSE2, zero_SSE, gmx_sub_pr(rc2_SSE, rsq_SSE2)));
+ brsq_S0 = gmx_mul_pr(beta2_S, gmx_blendv_pr(rsq_S0, zero_S, gmx_sub_pr(rc2_S, rsq_S0)));
+ brsq_S2 = gmx_mul_pr(beta2_S, gmx_blendv_pr(rsq_S2, zero_S, gmx_sub_pr(rc2_S, rsq_S2)));
#endif
- ewcorr_SSE0 = gmx_mul_pr(gmx_pmecorrF_pr(brsq_SSE0), beta_SSE);
- ewcorr_SSE2 = gmx_mul_pr(gmx_pmecorrF_pr(brsq_SSE2), beta_SSE);
- frcoul_SSE0 = gmx_mul_pr(qq_SSE0, gmx_add_pr(rinv_ex_SSE0, gmx_mul_pr(ewcorr_SSE0, brsq_SSE0)));
- frcoul_SSE2 = gmx_mul_pr(qq_SSE2, gmx_add_pr(rinv_ex_SSE2, gmx_mul_pr(ewcorr_SSE2, brsq_SSE2)));
+ ewcorr_S0 = gmx_mul_pr(gmx_pmecorrF_pr(brsq_S0), beta_S);
+ ewcorr_S2 = gmx_mul_pr(gmx_pmecorrF_pr(brsq_S2), beta_S);
+ frcoul_S0 = gmx_mul_pr(qq_S0, gmx_add_pr(rinv_ex_S0, gmx_mul_pr(ewcorr_S0, brsq_S0)));
+ frcoul_S2 = gmx_mul_pr(qq_S2, gmx_add_pr(rinv_ex_S2, gmx_mul_pr(ewcorr_S2, brsq_S2)));
#ifdef CALC_ENERGIES
- vc_sub_SSE0 = gmx_mul_pr(gmx_pmecorrV_pr(brsq_SSE0), beta_SSE);
- vc_sub_SSE2 = gmx_mul_pr(gmx_pmecorrV_pr(brsq_SSE2), beta_SSE);
+ vc_sub_S0 = gmx_mul_pr(gmx_pmecorrV_pr(brsq_S0), beta_S);
+ vc_sub_S2 = gmx_mul_pr(gmx_pmecorrV_pr(brsq_S2), beta_S);
#endif
#endif /* CALC_COUL_EWALD */
#ifdef CALC_COUL_TAB
/* Electrostatic interactions */
- r_SSE0 = gmx_mul_pr(rsq_SSE0, rinv_SSE0);
- r_SSE2 = gmx_mul_pr(rsq_SSE2, rinv_SSE2);
+ r_S0 = gmx_mul_pr(rsq_S0, rinv_S0);
+ r_S2 = gmx_mul_pr(rsq_S2, rinv_S2);
/* Convert r to scaled table units */
- rs_SSE0 = gmx_mul_pr(r_SSE0, invtsp_SSE);
- rs_SSE2 = gmx_mul_pr(r_SSE2, invtsp_SSE);
+ rs_S0 = gmx_mul_pr(r_S0, invtsp_S);
+ rs_S2 = gmx_mul_pr(r_S2, invtsp_S);
/* Truncate scaled r to an int */
- ti_SSE0 = gmx_cvttpr_epi32(rs_SSE0);
- ti_SSE2 = gmx_cvttpr_epi32(rs_SSE2);
-#ifdef GMX_X86_SSE4_1
- /* SSE4.1 floor is faster than gmx_cvtepi32_ps int->float cast */
- rf_SSE0 = gmx_floor_pr(rs_SSE0);
- rf_SSE2 = gmx_floor_pr(rs_SSE2);
+ ti_S0 = gmx_cvttpr_epi32(rs_S0);
+ ti_S2 = gmx_cvttpr_epi32(rs_S2);
+#ifdef GMX_HAVE_SIMD_FLOOR
+ rf_S0 = gmx_floor_pr(rs_S0);
+ rf_S2 = gmx_floor_pr(rs_S2);
#else
- rf_SSE0 = gmx_cvtepi32_pr(ti_SSE0);
- rf_SSE2 = gmx_cvtepi32_pr(ti_SSE2);
+ rf_S0 = gmx_cvtepi32_pr(ti_S0);
+ rf_S2 = gmx_cvtepi32_pr(ti_S2);
#endif
- frac_SSE0 = gmx_sub_pr(rs_SSE0, rf_SSE0);
- frac_SSE2 = gmx_sub_pr(rs_SSE2, rf_SSE2);
+ frac_S0 = gmx_sub_pr(rs_S0, rf_S0);
+ frac_S2 = gmx_sub_pr(rs_S2, rf_S2);
/* Load and interpolate table forces and possibly energies.
* Force and energy can be combined in one table, stride 4: FDV0
* Currently single precision uses FDV0, double F and V.
*/
#ifndef CALC_ENERGIES
- load_table_f(tab_coul_F, ti_SSE0, ti0, ctab0_SSE0, ctab1_SSE0);
- load_table_f(tab_coul_F, ti_SSE2, ti2, ctab0_SSE2, ctab1_SSE2);
+ load_table_f(tab_coul_F, ti_S0, ti0, ctab0_S0, ctab1_S0);
+ load_table_f(tab_coul_F, ti_S2, ti2, ctab0_S2, ctab1_S2);
#else
#ifdef TAB_FDV0
- load_table_f_v(tab_coul_F, ti_SSE0, ti0, ctab0_SSE0, ctab1_SSE0, ctabv_SSE0);
- load_table_f_v(tab_coul_F, ti_SSE2, ti2, ctab0_SSE2, ctab1_SSE2, ctabv_SSE2);
+ load_table_f_v(tab_coul_F, ti_S0, ti0, ctab0_S0, ctab1_S0, ctabv_S0);
+ load_table_f_v(tab_coul_F, ti_S2, ti2, ctab0_S2, ctab1_S2, ctabv_S2);
#else
- load_table_f_v(tab_coul_F, tab_coul_V, ti_SSE0, ti0, ctab0_SSE0, ctab1_SSE0, ctabv_SSE0);
- load_table_f_v(tab_coul_F, tab_coul_V, ti_SSE2, ti2, ctab0_SSE2, ctab1_SSE2, ctabv_SSE2);
+ load_table_f_v(tab_coul_F, tab_coul_V, ti_S0, ti0, ctab0_S0, ctab1_S0, ctabv_S0);
+ load_table_f_v(tab_coul_F, tab_coul_V, ti_S2, ti2, ctab0_S2, ctab1_S2, ctabv_S2);
#endif
#endif
- fsub_SSE0 = gmx_add_pr(ctab0_SSE0, gmx_mul_pr(frac_SSE0, ctab1_SSE0));
- fsub_SSE2 = gmx_add_pr(ctab0_SSE2, gmx_mul_pr(frac_SSE2, ctab1_SSE2));
- frcoul_SSE0 = gmx_mul_pr(qq_SSE0, gmx_sub_pr(rinv_ex_SSE0, gmx_mul_pr(fsub_SSE0, r_SSE0)));
- frcoul_SSE2 = gmx_mul_pr(qq_SSE2, gmx_sub_pr(rinv_ex_SSE2, gmx_mul_pr(fsub_SSE2, r_SSE2)));
+ fsub_S0 = gmx_add_pr(ctab0_S0, gmx_mul_pr(frac_S0, ctab1_S0));
+ fsub_S2 = gmx_add_pr(ctab0_S2, gmx_mul_pr(frac_S2, ctab1_S2));
+ frcoul_S0 = gmx_mul_pr(qq_S0, gmx_sub_pr(rinv_ex_S0, gmx_mul_pr(fsub_S0, r_S0)));
+ frcoul_S2 = gmx_mul_pr(qq_S2, gmx_sub_pr(rinv_ex_S2, gmx_mul_pr(fsub_S2, r_S2)));
#ifdef CALC_ENERGIES
- vc_sub_SSE0 = gmx_add_pr(ctabv_SSE0, gmx_mul_pr(gmx_mul_pr(mhalfsp_SSE, frac_SSE0), gmx_add_pr(ctab0_SSE0, fsub_SSE0)));
- vc_sub_SSE2 = gmx_add_pr(ctabv_SSE2, gmx_mul_pr(gmx_mul_pr(mhalfsp_SSE, frac_SSE2), gmx_add_pr(ctab0_SSE2, fsub_SSE2)));
+ vc_sub_S0 = gmx_add_pr(ctabv_S0, gmx_mul_pr(gmx_mul_pr(mhalfsp_S, frac_S0), gmx_add_pr(ctab0_S0, fsub_S0)));
+ vc_sub_S2 = gmx_add_pr(ctabv_S2, gmx_mul_pr(gmx_mul_pr(mhalfsp_S, frac_S2), gmx_add_pr(ctab0_S2, fsub_S2)));
#endif
#endif /* CALC_COUL_TAB */
#ifndef NO_SHIFT_EWALD
/* Add Ewald potential shift to vc_sub for convenience */
#ifdef CHECK_EXCLS
- vc_sub_SSE0 = gmx_add_pr(vc_sub_SSE0, gmx_and_pr(sh_ewald_SSE, int_SSE0));
- vc_sub_SSE2 = gmx_add_pr(vc_sub_SSE2, gmx_and_pr(sh_ewald_SSE, int_SSE2));
+ vc_sub_S0 = gmx_add_pr(vc_sub_S0, gmx_blendzero_pr(sh_ewald_S, int_S0));
+ vc_sub_S2 = gmx_add_pr(vc_sub_S2, gmx_blendzero_pr(sh_ewald_S, int_S2));
#else
- vc_sub_SSE0 = gmx_add_pr(vc_sub_SSE0, sh_ewald_SSE);
- vc_sub_SSE2 = gmx_add_pr(vc_sub_SSE2, sh_ewald_SSE);
+ vc_sub_S0 = gmx_add_pr(vc_sub_S0, sh_ewald_S);
+ vc_sub_S2 = gmx_add_pr(vc_sub_S2, sh_ewald_S);
#endif
#endif
- vcoul_SSE0 = gmx_mul_pr(qq_SSE0, gmx_sub_pr(rinv_ex_SSE0, vc_sub_SSE0));
- vcoul_SSE2 = gmx_mul_pr(qq_SSE2, gmx_sub_pr(rinv_ex_SSE2, vc_sub_SSE2));
+ vcoul_S0 = gmx_mul_pr(qq_S0, gmx_sub_pr(rinv_ex_S0, vc_sub_S0));
+ vcoul_S2 = gmx_mul_pr(qq_S2, gmx_sub_pr(rinv_ex_S2, vc_sub_S2));
#endif
#ifdef CALC_ENERGIES
/* Mask energy for cut-off and diagonal */
- vcoul_SSE0 = gmx_and_pr(vcoul_SSE0, wco_SSE0);
- vcoul_SSE2 = gmx_and_pr(vcoul_SSE2, wco_SSE2);
+ vcoul_S0 = gmx_blendzero_pr(vcoul_S0, wco_S0);
+ vcoul_S2 = gmx_blendzero_pr(vcoul_S2, wco_S2);
#endif
#endif /* CALC_COULOMB */
/* Lennard-Jones interaction */
#ifdef VDW_CUTOFF_CHECK
- wco_vdw_SSE0 = gmx_cmplt_pr(rsq_SSE0, rcvdw2_SSE);
+ wco_vdw_S0 = gmx_cmplt_pr(rsq_S0, rcvdw2_S);
#ifndef HALF_LJ
- wco_vdw_SSE2 = gmx_cmplt_pr(rsq_SSE2, rcvdw2_SSE);
+ wco_vdw_S2 = gmx_cmplt_pr(rsq_S2, rcvdw2_S);
#endif
#else
/* Same cut-off for Coulomb and VdW, reuse the registers */
-#define wco_vdw_SSE0 wco_SSE0
-#define wco_vdw_SSE2 wco_SSE2
+#define wco_vdw_S0 wco_S0
+#define wco_vdw_S2 wco_S2
#endif
#ifndef LJ_COMB_LB
- rinvsix_SSE0 = gmx_mul_pr(rinvsq_SSE0, gmx_mul_pr(rinvsq_SSE0, rinvsq_SSE0));
+ rinvsix_S0 = gmx_mul_pr(rinvsq_S0, gmx_mul_pr(rinvsq_S0, rinvsq_S0));
#ifdef EXCL_FORCES
- rinvsix_SSE0 = gmx_and_pr(rinvsix_SSE0, int_SSE0);
+ rinvsix_S0 = gmx_blendzero_pr(rinvsix_S0, int_S0);
#endif
#ifndef HALF_LJ
- rinvsix_SSE2 = gmx_mul_pr(rinvsq_SSE2, gmx_mul_pr(rinvsq_SSE2, rinvsq_SSE2));
+ rinvsix_S2 = gmx_mul_pr(rinvsq_S2, gmx_mul_pr(rinvsq_S2, rinvsq_S2));
#ifdef EXCL_FORCES
- rinvsix_SSE2 = gmx_and_pr(rinvsix_SSE2, int_SSE2);
+ rinvsix_S2 = gmx_blendzero_pr(rinvsix_S2, int_S2);
#endif
#endif
#ifdef VDW_CUTOFF_CHECK
- rinvsix_SSE0 = gmx_and_pr(rinvsix_SSE0, wco_vdw_SSE0);
+ rinvsix_S0 = gmx_blendzero_pr(rinvsix_S0, wco_vdw_S0);
#ifndef HALF_LJ
- rinvsix_SSE2 = gmx_and_pr(rinvsix_SSE2, wco_vdw_SSE2);
+ rinvsix_S2 = gmx_blendzero_pr(rinvsix_S2, wco_vdw_S2);
#endif
#endif
- FrLJ6_SSE0 = gmx_mul_pr(c6_SSE0, rinvsix_SSE0);
+ FrLJ6_S0 = gmx_mul_pr(c6_S0, rinvsix_S0);
#ifndef HALF_LJ
- FrLJ6_SSE2 = gmx_mul_pr(c6_SSE2, rinvsix_SSE2);
+ FrLJ6_S2 = gmx_mul_pr(c6_S2, rinvsix_S2);
#endif
- FrLJ12_SSE0 = gmx_mul_pr(c12_SSE0, gmx_mul_pr(rinvsix_SSE0, rinvsix_SSE0));
+ FrLJ12_S0 = gmx_mul_pr(c12_S0, gmx_mul_pr(rinvsix_S0, rinvsix_S0));
#ifndef HALF_LJ
- FrLJ12_SSE2 = gmx_mul_pr(c12_SSE2, gmx_mul_pr(rinvsix_SSE2, rinvsix_SSE2));
+ FrLJ12_S2 = gmx_mul_pr(c12_S2, gmx_mul_pr(rinvsix_S2, rinvsix_S2));
#endif
#endif /* not LJ_COMB_LB */
#ifdef LJ_COMB_LB
- sir_SSE0 = gmx_mul_pr(sig_SSE0, rinv_SSE0);
+ sir_S0 = gmx_mul_pr(sig_S0, rinv_S0);
#ifndef HALF_LJ
- sir_SSE2 = gmx_mul_pr(sig_SSE2, rinv_SSE2);
+ sir_S2 = gmx_mul_pr(sig_S2, rinv_S2);
#endif
- sir2_SSE0 = gmx_mul_pr(sir_SSE0, sir_SSE0);
+ sir2_S0 = gmx_mul_pr(sir_S0, sir_S0);
#ifndef HALF_LJ
- sir2_SSE2 = gmx_mul_pr(sir_SSE2, sir_SSE2);
+ sir2_S2 = gmx_mul_pr(sir_S2, sir_S2);
#endif
- sir6_SSE0 = gmx_mul_pr(sir2_SSE0, gmx_mul_pr(sir2_SSE0, sir2_SSE0));
+ sir6_S0 = gmx_mul_pr(sir2_S0, gmx_mul_pr(sir2_S0, sir2_S0));
#ifdef EXCL_FORCES
- sir6_SSE0 = gmx_and_pr(sir6_SSE0, int_SSE0);
+ sir6_S0 = gmx_blendzero_pr(sir6_S0, int_S0);
#endif
#ifndef HALF_LJ
- sir6_SSE2 = gmx_mul_pr(sir2_SSE2, gmx_mul_pr(sir2_SSE2, sir2_SSE2));
+ sir6_S2 = gmx_mul_pr(sir2_S2, gmx_mul_pr(sir2_S2, sir2_S2));
#ifdef EXCL_FORCES
- sir6_SSE2 = gmx_and_pr(sir6_SSE2, int_SSE2);
+ sir6_S2 = gmx_blendzero_pr(sir6_S2, int_S2);
#endif
#endif
#ifdef VDW_CUTOFF_CHECK
- sir6_SSE0 = gmx_and_pr(sir6_SSE0, wco_vdw_SSE0);
+ sir6_S0 = gmx_blendzero_pr(sir6_S0, wco_vdw_S0);
#ifndef HALF_LJ
- sir6_SSE2 = gmx_and_pr(sir6_SSE2, wco_vdw_SSE2);
+ sir6_S2 = gmx_blendzero_pr(sir6_S2, wco_vdw_S2);
#endif
#endif
- FrLJ6_SSE0 = gmx_mul_pr(eps_SSE0, sir6_SSE0);
+ FrLJ6_S0 = gmx_mul_pr(eps_S0, sir6_S0);
#ifndef HALF_LJ
- FrLJ6_SSE2 = gmx_mul_pr(eps_SSE2, sir6_SSE2);
+ FrLJ6_S2 = gmx_mul_pr(eps_S2, sir6_S2);
#endif
- FrLJ12_SSE0 = gmx_mul_pr(FrLJ6_SSE0, sir6_SSE0);
+ FrLJ12_S0 = gmx_mul_pr(FrLJ6_S0, sir6_S0);
#ifndef HALF_LJ
- FrLJ12_SSE2 = gmx_mul_pr(FrLJ6_SSE2, sir6_SSE2);
+ FrLJ12_S2 = gmx_mul_pr(FrLJ6_S2, sir6_S2);
#endif
#if defined CALC_ENERGIES
/* We need C6 and C12 to calculate the LJ potential shift */
- sig2_SSE0 = gmx_mul_pr(sig_SSE0, sig_SSE0);
+ sig2_S0 = gmx_mul_pr(sig_S0, sig_S0);
#ifndef HALF_LJ
- sig2_SSE2 = gmx_mul_pr(sig_SSE2, sig_SSE2);
+ sig2_S2 = gmx_mul_pr(sig_S2, sig_S2);
#endif
- sig6_SSE0 = gmx_mul_pr(sig2_SSE0, gmx_mul_pr(sig2_SSE0, sig2_SSE0));
+ sig6_S0 = gmx_mul_pr(sig2_S0, gmx_mul_pr(sig2_S0, sig2_S0));
#ifndef HALF_LJ
- sig6_SSE2 = gmx_mul_pr(sig2_SSE2, gmx_mul_pr(sig2_SSE2, sig2_SSE2));
+ sig6_S2 = gmx_mul_pr(sig2_S2, gmx_mul_pr(sig2_S2, sig2_S2));
#endif
- c6_SSE0 = gmx_mul_pr(eps_SSE0, sig6_SSE0);
+ c6_S0 = gmx_mul_pr(eps_S0, sig6_S0);
#ifndef HALF_LJ
- c6_SSE2 = gmx_mul_pr(eps_SSE2, sig6_SSE2);
+ c6_S2 = gmx_mul_pr(eps_S2, sig6_S2);
#endif
- c12_SSE0 = gmx_mul_pr(c6_SSE0, sig6_SSE0);
+ c12_S0 = gmx_mul_pr(c6_S0, sig6_S0);
#ifndef HALF_LJ
- c12_SSE2 = gmx_mul_pr(c6_SSE2, sig6_SSE2);
+ c12_S2 = gmx_mul_pr(c6_S2, sig6_S2);
#endif
#endif
#endif /* LJ_COMB_LB */
#ifdef CALC_COULOMB
#ifndef ENERGY_GROUPS
- vctotSSE = gmx_add_pr(vctotSSE, gmx_add_pr(vcoul_SSE0, vcoul_SSE2));
+ vctot_S = gmx_add_pr(vctot_S, gmx_add_pr(vcoul_S0, vcoul_S2));
#else
- add_ener_grp_halves(vcoul_SSE0, vctp[0], vctp[1], egp_jj);
- add_ener_grp_halves(vcoul_SSE2, vctp[2], vctp[3], egp_jj);
+ add_ener_grp_halves(vcoul_S0, vctp[0], vctp[1], egp_jj);
+ add_ener_grp_halves(vcoul_S2, vctp[2], vctp[3], egp_jj);
#endif
#endif
#ifdef CALC_LJ
/* Calculate the LJ energies */
- VLJ6_SSE0 = gmx_mul_pr(sixthSSE, gmx_sub_pr(FrLJ6_SSE0, gmx_mul_pr(c6_SSE0, sh_invrc6_SSE)));
+ VLJ6_S0 = gmx_mul_pr(sixth_S, gmx_sub_pr(FrLJ6_S0, gmx_mul_pr(c6_S0, sh_invrc6_S)));
#ifndef HALF_LJ
- VLJ6_SSE2 = gmx_mul_pr(sixthSSE, gmx_sub_pr(FrLJ6_SSE2, gmx_mul_pr(c6_SSE2, sh_invrc6_SSE)));
+ VLJ6_S2 = gmx_mul_pr(sixth_S, gmx_sub_pr(FrLJ6_S2, gmx_mul_pr(c6_S2, sh_invrc6_S)));
#endif
- VLJ12_SSE0 = gmx_mul_pr(twelvethSSE, gmx_sub_pr(FrLJ12_SSE0, gmx_mul_pr(c12_SSE0, sh_invrc12_SSE)));
+ VLJ12_S0 = gmx_mul_pr(twelveth_S, gmx_sub_pr(FrLJ12_S0, gmx_mul_pr(c12_S0, sh_invrc12_S)));
#ifndef HALF_LJ
- VLJ12_SSE2 = gmx_mul_pr(twelvethSSE, gmx_sub_pr(FrLJ12_SSE2, gmx_mul_pr(c12_SSE2, sh_invrc12_SSE)));
+ VLJ12_S2 = gmx_mul_pr(twelveth_S, gmx_sub_pr(FrLJ12_S2, gmx_mul_pr(c12_S2, sh_invrc12_S)));
#endif
- VLJ_SSE0 = gmx_sub_pr(VLJ12_SSE0, VLJ6_SSE0);
+ VLJ_S0 = gmx_sub_pr(VLJ12_S0, VLJ6_S0);
#ifndef HALF_LJ
- VLJ_SSE2 = gmx_sub_pr(VLJ12_SSE2, VLJ6_SSE2);
+ VLJ_S2 = gmx_sub_pr(VLJ12_S2, VLJ6_S2);
#endif
/* The potential shift should be removed for pairs beyond cut-off */
- VLJ_SSE0 = gmx_and_pr(VLJ_SSE0, wco_vdw_SSE0);
+ VLJ_S0 = gmx_blendzero_pr(VLJ_S0, wco_vdw_S0);
#ifndef HALF_LJ
- VLJ_SSE2 = gmx_and_pr(VLJ_SSE2, wco_vdw_SSE2);
+ VLJ_S2 = gmx_blendzero_pr(VLJ_S2, wco_vdw_S2);
#endif
#ifdef CHECK_EXCLS
/* The potential shift should be removed for excluded pairs */
- VLJ_SSE0 = gmx_and_pr(VLJ_SSE0, int_SSE0);
+ VLJ_S0 = gmx_blendzero_pr(VLJ_S0, int_S0);
#ifndef HALF_LJ
- VLJ_SSE2 = gmx_and_pr(VLJ_SSE2, int_SSE2);
+ VLJ_S2 = gmx_blendzero_pr(VLJ_S2, int_S2);
#endif
#endif
#ifndef ENERGY_GROUPS
- VvdwtotSSE = gmx_add_pr(VvdwtotSSE,
+ Vvdwtot_S = gmx_add_pr(Vvdwtot_S,
#ifndef HALF_LJ
- gmx_add_pr(VLJ_SSE0, VLJ_SSE2)
+ gmx_add_pr(VLJ_S0, VLJ_S2)
#else
- VLJ_SSE0
+ VLJ_S0
#endif
- );
+ );
#else
- add_ener_grp_halves(VLJ_SSE0, vvdwtp[0], vvdwtp[1], egp_jj);
+ add_ener_grp_halves(VLJ_S0, vvdwtp[0], vvdwtp[1], egp_jj);
#ifndef HALF_LJ
- add_ener_grp_halves(VLJ_SSE2, vvdwtp[2], vvdwtp[3], egp_jj);
+ add_ener_grp_halves(VLJ_S2, vvdwtp[2], vvdwtp[3], egp_jj);
#endif
#endif
#endif /* CALC_LJ */
#endif /* CALC_ENERGIES */
#ifdef CALC_LJ
- fscal_SSE0 = gmx_mul_pr(rinvsq_SSE0,
+ fscal_S0 = gmx_mul_pr(rinvsq_S0,
#ifdef CALC_COULOMB
- gmx_add_pr(frcoul_SSE0,
+ gmx_add_pr(frcoul_S0,
#else
- (
+ (
#endif
- gmx_sub_pr(FrLJ12_SSE0, FrLJ6_SSE0)));
+ gmx_sub_pr(FrLJ12_S0, FrLJ6_S0)));
#else
- fscal_SSE0 = gmx_mul_pr(rinvsq_SSE0, frcoul_SSE0);
+ fscal_S0 = gmx_mul_pr(rinvsq_S0, frcoul_S0);
#endif /* CALC_LJ */
#if defined CALC_LJ && !defined HALF_LJ
- fscal_SSE2 = gmx_mul_pr(rinvsq_SSE2,
+ fscal_S2 = gmx_mul_pr(rinvsq_S2,
#ifdef CALC_COULOMB
- gmx_add_pr(frcoul_SSE2,
+ gmx_add_pr(frcoul_S2,
#else
- (
+ (
#endif
- gmx_sub_pr(FrLJ12_SSE2, FrLJ6_SSE2)));
+ gmx_sub_pr(FrLJ12_S2, FrLJ6_S2)));
#else
/* Atom 2 and 3 don't have LJ, so only add Coulomb forces */
- fscal_SSE2 = gmx_mul_pr(rinvsq_SSE2, frcoul_SSE2);
+ fscal_S2 = gmx_mul_pr(rinvsq_S2, frcoul_S2);
#endif
/* Calculate temporary vectorial force */
- tx_SSE0 = gmx_mul_pr(fscal_SSE0, dx_SSE0);
- tx_SSE2 = gmx_mul_pr(fscal_SSE2, dx_SSE2);
- ty_SSE0 = gmx_mul_pr(fscal_SSE0, dy_SSE0);
- ty_SSE2 = gmx_mul_pr(fscal_SSE2, dy_SSE2);
- tz_SSE0 = gmx_mul_pr(fscal_SSE0, dz_SSE0);
- tz_SSE2 = gmx_mul_pr(fscal_SSE2, dz_SSE2);
+ tx_S0 = gmx_mul_pr(fscal_S0, dx_S0);
+ tx_S2 = gmx_mul_pr(fscal_S2, dx_S2);
+ ty_S0 = gmx_mul_pr(fscal_S0, dy_S0);
+ ty_S2 = gmx_mul_pr(fscal_S2, dy_S2);
+ tz_S0 = gmx_mul_pr(fscal_S0, dz_S0);
+ tz_S2 = gmx_mul_pr(fscal_S2, dz_S2);
/* Increment i atom force */
- fix_SSE0 = gmx_add_pr(fix_SSE0, tx_SSE0);
- fix_SSE2 = gmx_add_pr(fix_SSE2, tx_SSE2);
- fiy_SSE0 = gmx_add_pr(fiy_SSE0, ty_SSE0);
- fiy_SSE2 = gmx_add_pr(fiy_SSE2, ty_SSE2);
- fiz_SSE0 = gmx_add_pr(fiz_SSE0, tz_SSE0);
- fiz_SSE2 = gmx_add_pr(fiz_SSE2, tz_SSE2);
+ fix_S0 = gmx_add_pr(fix_S0, tx_S0);
+ fix_S2 = gmx_add_pr(fix_S2, tx_S2);
+ fiy_S0 = gmx_add_pr(fiy_S0, ty_S0);
+ fiy_S2 = gmx_add_pr(fiy_S2, ty_S2);
+ fiz_S0 = gmx_add_pr(fiz_S0, tz_S0);
+ fiz_S2 = gmx_add_pr(fiz_S2, tz_S2);
/* Decrement j atom force */
- gmx_store_hpr(f+ajx,
- gmx_sub_hpr( gmx_load_hpr(f+ajx), gmx_sum4_hpr(tx_SSE0, tx_SSE2) ));
- gmx_store_hpr(f+ajy,
- gmx_sub_hpr( gmx_load_hpr(f+ajy), gmx_sum4_hpr(ty_SSE0, ty_SSE2) ));
- gmx_store_hpr(f+ajz,
- gmx_sub_hpr( gmx_load_hpr(f+ajz), gmx_sum4_hpr(tz_SSE0, tz_SSE2) ));
+ gmx_load_hpr(fjx_S, f+ajx);
+ gmx_load_hpr(fjy_S, f+ajy);
+ gmx_load_hpr(fjz_S, f+ajz);
+ gmx_store_hpr(f+ajx, gmx_sub_hpr(fjx_S, gmx_sum4_hpr(tx_S0, tx_S2)));
+ gmx_store_hpr(f+ajy, gmx_sub_hpr(fjy_S, gmx_sum4_hpr(ty_S0, ty_S2)));
+ gmx_store_hpr(f+ajz, gmx_sub_hpr(fjz_S, gmx_sum4_hpr(tz_S0, tz_S2)));
}
-#undef rinv_ex_SSE0
-#undef rinv_ex_SSE2
+#undef rinv_ex_S0
+#undef rinv_ex_S2
-#undef wco_vdw_SSE0
-#undef wco_vdw_SSE2
+#undef wco_vdw_S0
+#undef wco_vdw_S2
#undef CUTOFF_BLENDV
biod.pnpi.spb.ru / alexxy / gromacs.git / blobdiff