nb_kernel113_ppc_altivec.c

来自「最著名最快的分子模拟软件」· C语言 代码 · 共 873 行 · 第 1/3 页

				qqH             = vec_madd(iqH,jq,nul);
				Vvdw6           = vec_madd(c6,rinvsix,nul);
				vcoulM          = vec_madd(qqM,rinvM,nul);
				Vvdw12           = vec_madd(c12,vec_madd(rinvsix,rinvsix,nul),nul);
				vcoulH1         = vec_madd(qqH,rinvH1,nul);
				Vvdwtot          = vec_add(Vvdwtot,Vvdw12);
				fsO             = vec_madd(vec_twelve(),Vvdw12,nul);
				vcoulH2         = vec_madd(qqH,rinvH2,nul);
				Vvdwtot          = vec_sub(Vvdwtot,Vvdw6);
				fsO             = vec_nmsub(vec_six(),Vvdw6,fsO);
				fsH1            = vec_madd(vcoulH1,rinvsqH1,nul);
				fsH2            = vec_madd(vcoulH2,rinvsqH2,nul);
				fsM             = vec_madd(vcoulM,rinvsqM,nul);
				fsO             = vec_madd(fsO,rinvsqO,nul);
				vctot           = vec_add(vctot,vcoulM);
				vcoulH1         = vec_add(vcoulH1,vcoulH2);
				vctot           = vec_add(vctot,vcoulH1);

				fiOx            = vec_madd(fsO,dOx,fiOx); /* +=fx */
				dOx             = vec_nmsub(fsO,dOx,nul); /* -fx */
				fiOy            = vec_madd(fsO,dOy,fiOy); /* +=fy */
				dOy             = vec_nmsub(fsO,dOy,nul); /* -fy */
				fiOz            = vec_madd(fsO,dOz,fiOz); /* +=fz */
				dOz             = vec_nmsub(fsO,dOz,nul); /* -fz */
				fiH1x           = vec_madd(fsH1,dH1x,fiH1x); /* +=fx */
				dOx             = vec_nmsub(fsH1,dH1x,dOx); /* -fx */
				fiH1y           = vec_madd(fsH1,dH1y,fiH1y); /* +=fy */
				dOy             = vec_nmsub(fsH1,dH1y,dOy); /* -fy */
				fiH1z           = vec_madd(fsH1,dH1z,fiH1z); /* +=fz */
				dOz             = vec_nmsub(fsH1,dH1z,dOz); /* -fz */
				fiH2x           = vec_madd(fsH2,dH2x,fiH2x); /* +=fx */
				dOx             = vec_nmsub(fsH2,dH2x,dOx); /* -fx */
				fiH2y           = vec_madd(fsH2,dH2y,fiH2y); /* +=fy */
				dOy             = vec_nmsub(fsH2,dH2y,dOy); /* -fy */
				fiH2z           = vec_madd(fsH2,dH2z,fiH2z); /* +=fz */
				dOz             = vec_nmsub(fsH2,dH2z,dOz); /* -fz */
				fiMx            = vec_madd(fsM,dMx,fiMx); /* +=fx */
				dOx             = vec_nmsub(fsM,dMx,dOx); /* -fx */
				fiMy            = vec_madd(fsM,dMy,fiMy); /* +=fy */
				dOy             = vec_nmsub(fsM,dMy,dOy); /* -fy */
				fiMz            = vec_madd(fsM,dMz,fiMz); /* +=fz */
				dOz             = vec_nmsub(fsM,dMz,dOz); /* -fz */
      
				transpose_4_to_3(dOx,dOy,dOz,nul,&tmp1,&tmp2,&tmp3);
				add_xyz_to_mem(faction+j3a,tmp1);
				add_xyz_to_mem(faction+j3b,tmp2);
				add_xyz_to_mem(faction+j3c,tmp3);
			}
            else if(k<(nj1-1))
            {
				jnra            = jjnr[k];
				jnrb            = jjnr[k+1];
				j3a             = 3*jnra;
				j3b             = 3*jnrb;
				transpose_2_to_3(load_xyz(pos+j3a),
								 load_xyz(pos+j3b),&dMx,&dMy,&dMz);
				dOx             = vec_sub(iOx,dMx);
				dOy             = vec_sub(iOy,dMy);
				dOz             = vec_sub(iOz,dMz);
				dH1x            = vec_sub(iH1x,dMx);
				dH1y            = vec_sub(iH1y,dMy);
				dH1z            = vec_sub(iH1z,dMz);
				dH2x            = vec_sub(iH2x,dMx);
				dH2y            = vec_sub(iH2y,dMy);
				dH2z            = vec_sub(iH2z,dMz);
				dMx             = vec_sub(iMx,dMx);
				dMy             = vec_sub(iMy,dMy);
				dMz             = vec_sub(iMz,dMz);
      
				rsqO            = vec_madd(dOx,dOx,nul);
				rsqH1           = vec_madd(dH1x,dH1x,nul);
				rsqH2           = vec_madd(dH2x,dH2x,nul);
				rsqM            = vec_madd(dMx,dMx,nul);
				rsqO            = vec_madd(dOy,dOy,rsqO);
				rsqH1           = vec_madd(dH1y,dH1y,rsqH1);
				rsqH2           = vec_madd(dH2y,dH2y,rsqH2);
				rsqM            = vec_madd(dMy,dMy,rsqM);
				rsqO            = vec_madd(dOz,dOz,rsqO);
				rsqH1           = vec_madd(dH1z,dH1z,rsqH1);
				rsqH2           = vec_madd(dH2z,dH2z,rsqH2);
				rsqM            = vec_madd(dMz,dMz,rsqM);
				rinvsqO         = do_recip(rsqO);
				do_3_invsqrt(rsqM,rsqH1,rsqH2,&rinvM,&rinvH1,&rinvH2);
				zero_highest_2_elements_in_vector(&rinvsqO);
				zero_highest_2_elements_in_3_vectors(&rinvM,&rinvH1,&rinvH2);

				rinvsqM         = vec_madd(rinvM,rinvM,nul);
				rinvsqH1        = vec_madd(rinvH1,rinvH1,nul);
				rinvsqH2        = vec_madd(rinvH2,rinvH2,nul);
				rinvsix         = vec_madd(rinvsqO,rinvsqO,nul);
				rinvsix         = vec_madd(rinvsix,rinvsqO,nul);
				jq=load_2_float(charge+jnra,charge+jnrb);
				tja             = ntiA+2*type[jnra];
				tjb             = ntiA+2*type[jnrb];
				/* load 2 j charges and multiply by iq */
				load_2_pair(vdwparam+tja,vdwparam+tjb,&c6,&c12);
				qqM             = vec_madd(iqM,jq,nul);
				qqH             = vec_madd(iqH,jq,nul);
				Vvdw6           = vec_madd(c6,rinvsix,nul);
				vcoulM          = vec_madd(qqM,rinvM,nul);
				Vvdw12          = vec_madd(c12,vec_madd(rinvsix,rinvsix,nul),nul);
				vcoulH1         = vec_madd(qqH,rinvH1,nul);
				Vvdwtot         = vec_add(Vvdwtot,Vvdw12);
				fsO             = vec_madd(vec_twelve(),Vvdw12,nul);
				vcoulH2         = vec_madd(qqH,rinvH2,nul);
				Vvdwtot         = vec_sub(Vvdwtot,Vvdw6);
				fsO             = vec_nmsub(vec_six(),Vvdw6,fsO);
				fsH1            = vec_madd(vcoulH1,rinvsqH1,nul);
				fsH2            = vec_madd(vcoulH2,rinvsqH2,nul);
				fsM             = vec_madd(vcoulM,rinvsqM,nul);
				fsO             = vec_madd(fsO,rinvsqO,nul);
				vctot           = vec_add(vctot,vcoulM);
				vcoulH1         = vec_add(vcoulH1,vcoulH2);
				vctot           = vec_add(vctot,vcoulH1);

				fiOx            = vec_madd(fsO,dOx,fiOx); /* +=fx */
				dOx             = vec_nmsub(fsO,dOx,nul); /* -fx */
				fiOy            = vec_madd(fsO,dOy,fiOy); /* +=fy */
				dOy             = vec_nmsub(fsO,dOy,nul); /* -fy */
				fiOz            = vec_madd(fsO,dOz,fiOz); /* +=fz */
				dOz             = vec_nmsub(fsO,dOz,nul); /* -fz */
				fiH1x           = vec_madd(fsH1,dH1x,fiH1x); /* +=fx */
				dOx             = vec_nmsub(fsH1,dH1x,dOx); /* -fx */
				fiH1y           = vec_madd(fsH1,dH1y,fiH1y); /* +=fy */
				dOy             = vec_nmsub(fsH1,dH1y,dOy); /* -fy */
				fiH1z           = vec_madd(fsH1,dH1z,fiH1z); /* +=fz */
				dOz             = vec_nmsub(fsH1,dH1z,dOz); /* -fz */
				fiH2x           = vec_madd(fsH2,dH2x,fiH2x); /* +=fx */
				dOx             = vec_nmsub(fsH2,dH2x,dOx); /* -fx */
				fiH2y           = vec_madd(fsH2,dH2y,fiH2y); /* +=fy */
				dOy             = vec_nmsub(fsH2,dH2y,dOy); /* -fy */
				fiH2z           = vec_madd(fsH2,dH2z,fiH2z); /* +=fz */
				dOz             = vec_nmsub(fsH2,dH2z,dOz); /* -fz */
				fiMx            = vec_madd(fsM,dMx,fiMx); /* +=fx */
				dOx             = vec_nmsub(fsM,dMx,dOx); /* -fx */
				fiMy            = vec_madd(fsM,dMy,fiMy); /* +=fy */
				dOy             = vec_nmsub(fsM,dMy,dOy); /* -fy */
				fiMz            = vec_madd(fsM,dMz,fiMz); /* +=fz */
				dOz             = vec_nmsub(fsM,dMz,dOz); /* -fz */

				transpose_4_to_2(dOx,dOy,dOz,nul,&tmp1,&tmp2);
				add_xyz_to_mem(faction+j3a,tmp1);
				add_xyz_to_mem(faction+j3b,tmp2);
			} 
            else if(k<nj1) 
            {
				jnra            = jjnr[k];
				j3a             = 3*jnra;
				transpose_1_to_3(load_xyz(pos+j3a),
								 &dMx,&dMy,&dMz);
				dOx             = vec_sub(iOx,dMx);
				dOy             = vec_sub(iOy,dMy);
				dOz             = vec_sub(iOz,dMz);
				dH1x            = vec_sub(iH1x,dMx);
				dH1y            = vec_sub(iH1y,dMy);
				dH1z            = vec_sub(iH1z,dMz);
				dH2x            = vec_sub(iH2x,dMx);
				dH2y            = vec_sub(iH2y,dMy);
				dH2z            = vec_sub(iH2z,dMz);
				dMx             = vec_sub(iMx,dMx);
				dMy             = vec_sub(iMy,dMy);
				dMz             = vec_sub(iMz,dMz);

				rsqO            = vec_madd(dOx,dOx,nul);
				rsqH1           = vec_madd(dH1x,dH1x,nul);
				rsqH2           = vec_madd(dH2x,dH2x,nul);
				rsqM            = vec_madd(dMx,dMx,nul);
				rsqO            = vec_madd(dOy,dOy,rsqO);
				rsqH1           = vec_madd(dH1y,dH1y,rsqH1);
				rsqH2           = vec_madd(dH2y,dH2y,rsqH2);
				rsqM            = vec_madd(dMy,dMy,rsqM);
				rsqO            = vec_madd(dOz,dOz,rsqO);
				rsqH1           = vec_madd(dH1z,dH1z,rsqH1);
				rsqH2           = vec_madd(dH2z,dH2z,rsqH2);
				rsqM            = vec_madd(dMz,dMz,rsqM);
				rinvsqO         = do_recip(rsqO);
				do_3_invsqrt(rsqM,rsqH1,rsqH2,&rinvM,&rinvH1,&rinvH2);
				zero_highest_3_elements_in_vector(&rinvsqO);
				zero_highest_3_elements_in_3_vectors(&rinvM,&rinvH1,&rinvH2);
      
				rinvsqM         = vec_madd(rinvM,rinvM,nul);
				rinvsqH1        = vec_madd(rinvH1,rinvH1,nul);
				rinvsqH2        = vec_madd(rinvH2,rinvH2,nul);
				rinvsix         = vec_madd(rinvsqO,rinvsqO,nul);
				rinvsix         = vec_madd(rinvsix,rinvsqO,nul);
				tja             = ntiA+2*type[jnra];
				/* load 1 j charges and multiply by iq */
				jq=load_1_float(charge+jnra);
				load_1_pair(vdwparam+tja,&c6,&c12);
				qqM             = vec_madd(iqM,jq,nul);
				qqH             = vec_madd(iqH,jq,nul);
				Vvdw6           = vec_madd(c6,rinvsix,nul);
				vcoulM          = vec_madd(qqM,rinvM,nul);
				Vvdw12          = vec_madd(c12,vec_madd(rinvsix,rinvsix,nul),nul);
				vcoulH1         = vec_madd(qqH,rinvH1,nul);
				Vvdwtot         = vec_add(Vvdwtot,Vvdw12);
				fsO             = vec_madd(vec_twelve(),Vvdw12,nul);
				vcoulH2         = vec_madd(qqH,rinvH2,nul);
				Vvdwtot         = vec_sub(Vvdwtot,Vvdw6);
				fsO             = vec_nmsub(vec_six(),Vvdw6,fsO);
				fsH1            = vec_madd(vcoulH1,rinvsqH1,nul);
				fsH2            = vec_madd(vcoulH2,rinvsqH2,nul);
				fsM             = vec_madd(vcoulM,rinvsqM,nul);
				fsO             = vec_madd(fsO,rinvsqO,nul);
				vctot           = vec_add(vctot,vcoulM);
				vcoulH1         = vec_add(vcoulH1,vcoulH2);
				vctot           = vec_add(vctot,vcoulH1);

				fiOx            = vec_madd(fsO,dOx,fiOx); /* +=fx */
				dOx             = vec_nmsub(fsO,dOx,nul); /* -fx */
				fiOy            = vec_madd(fsO,dOy,fiOy); /* +=fy */
				dOy             = vec_nmsub(fsO,dOy,nul); /* -fy */
				fiOz            = vec_madd(fsO,dOz,fiOz); /* +=fz */
				dOz             = vec_nmsub(fsO,dOz,nul); /* -fz */
				fiH1x           = vec_madd(fsH1,dH1x,fiH1x); /* +=fx */
				dOx             = vec_nmsub(fsH1,dH1x,dOx); /* -fx */
				fiH1y           = vec_madd(fsH1,dH1y,fiH1y); /* +=fy */
				dOy             = vec_nmsub(fsH1,dH1y,dOy); /* -fy */
				fiH1z           = vec_madd(fsH1,dH1z,fiH1z); /* +=fz */
				dOz             = vec_nmsub(fsH1,dH1z,dOz); /* -fz */
				fiH2x           = vec_madd(fsH2,dH2x,fiH2x); /* +=fx */
				dOx             = vec_nmsub(fsH2,dH2x,dOx); /* -fx */
				fiH2y           = vec_madd(fsH2,dH2y,fiH2y); /* +=fy */
				dOy             = vec_nmsub(fsH2,dH2y,dOy); /* -fy */
				fiH2z           = vec_madd(fsH2,dH2z,fiH2z); /* +=fz */
				dOz             = vec_nmsub(fsH2,dH2z,dOz); /* -fz */
				fiMx            = vec_madd(fsM,dMx,fiMx); /* +=fx */
				dOx             = vec_nmsub(fsM,dMx,dOx); /* -fx */
				fiMy            = vec_madd(fsM,dMy,fiMy); /* +=fy */
				dOy             = vec_nmsub(fsM,dMy,dOy); /* -fy */
				fiMz            = vec_madd(fsM,dMz,fiMz); /* +=fz */
				dOz             = vec_nmsub(fsM,dMz,dOz); /* -fz */

				transpose_4_to_1(dOx,dOy,dOz,nul,&tmp1);
				add_xyz_to_mem(faction+j3a,tmp1);
			}
			/* update outer data */
			update_i_4atoms_forces(faction+ii3,fshift+is3,
								   fiOx,fiOy,fiOz,fiH1x,fiH1y,fiH1z,
								   fiH2x,fiH2y,fiH2z,fiMx,fiMy,fiMz);

			add_vector_to_float(Vc+gid[n],vctot);
			add_vector_to_float(Vvdw+gid[n],Vvdwtot);
			ninner += nj1 - nj0;
		}
#ifdef GMX_THREADS
		nouter += nn1 - nn0;
	} while (nn1<nri);
#else
	nouter = nri;
#endif
	*outeriter = nouter;
	*inneriter = ninner;
}


void 
nb_kernel113nf_ppc_altivec(int *             p_nri,
                       int               iinr[],
                       int               jindex[],
                       int               jjnr[],
                       int               shift[],
                       float             shiftvec[],
                       float             fshift[],
                       int               gid[],
                       float             pos[],
                       float             faction[],
                       float             charge[],
                       float *           p_facel,
                       float *           p_krf,
                       float *           p_crf,
                       float             Vc[],
                       int               type[],
                       int *             p_ntype,
                       float             vdwparam[],
                       float             Vvdw[],
                       float *           p_tabscale,
                       float             VFtab[],
                       float             invsqrta[],
                       float             dvda[],
                       float *           p_gbtabscale,
                       float             GBtab[],
                       int *             p_nthreads,
                       int *             count,
                       void *            mtx,
                       int *             outeriter,
                       int *             inneriter,
					   float *           work)
{
	vector float iOx,iOy,iOz,iH1x,iH1y,iH1z,iH2x,iH2y,iH2z,iMx,iMy,iMz;
	vector float dOx,dOy,dOz,dH1x,dH1y,dH1z,dH2x,dH2y,dH2z,dMx,dMy,dMz;