nb_kernel203_ppc_altivec.c

来自「最著名最快的分子模拟软件」· C语言 代码 · 共 774 行 · 第 1/3 页

				krsqH1          = vec_madd(vkrf,rsqH1,nul);
				krsqH2          = vec_madd(vkrf,rsqH2,nul);
				vcoulM          = vec_add(rinvM,krsqM);
				vcoulH1         = vec_add(rinvH1,krsqH1);
				vcoulH2         = vec_add(rinvH2,krsqH2);
				vcoulM          = vec_sub(vcoulM,vcrf);
				vcoulH1         = vec_sub(vcoulH1,vcrf);
				vcoulH2         = vec_sub(vcoulH2,vcrf);
				vctot           = vec_madd(qqM,vcoulM,vctot);
				fsM             = vec_nmsub(vec_two(),krsqM,rinvM);
				fsH1            = vec_nmsub(vec_two(),krsqH1,rinvH1);
				fsH2            = vec_nmsub(vec_two(),krsqH2,rinvH2);
				vctot           = vec_madd(qqH,vcoulH1,vctot);
				fsM             = vec_madd(fsM,qqM,nul);
				fsH1            = vec_madd(fsH1,qqH,nul);
				fsH2            = vec_madd(fsH2,qqH,nul);
				vctot           = vec_madd(qqH,vcoulH2,vctot);
				fsM             = vec_madd(fsM,rinvsqM,nul);
				fsH1            = vec_madd(fsH1,rinvsqH1,nul);
				fsH2            = vec_madd(fsH2,rinvsqH2,nul);

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

				transpose_4_to_3(dMx,dMy,dMz,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),&dH2x,&dH2y,&dH2z);
				dMx             = vec_sub(iMx,dH2x);
				dMy             = vec_sub(iMy,dH2y);
				dMz             = vec_sub(iMz,dH2z);
				dH1x            = vec_sub(iH1x,dH2x);
				dH1y            = vec_sub(iH1y,dH2y);
				dH1z            = vec_sub(iH1z,dH2z);
				dH2x            = vec_sub(iH2x,dH2x);
				dH2y            = vec_sub(iH2y,dH2y);
				dH2z            = vec_sub(iH2z,dH2z);
      
				rsqM            = vec_madd(dMx,dMx,nul);
				rsqH1           = vec_madd(dH1x,dH1x,nul);
				rsqH2           = vec_madd(dH2x,dH2x,nul);
				rsqM            = vec_madd(dMy,dMy,rsqM);
				rsqH1           = vec_madd(dH1y,dH1y,rsqH1);
				rsqH2           = vec_madd(dH2y,dH2y,rsqH2);
				rsqM            = vec_madd(dMz,dMz,rsqM);
				rsqH1           = vec_madd(dH1z,dH1z,rsqH1);
				rsqH2           = vec_madd(dH2z,dH2z,rsqH2);

				zero_highest_2_elements_in_3_vectors(&rsqM,&rsqH1,&rsqH2);
				do_3_invsqrt(rsqM,rsqH1,rsqH2,&rinvM,&rinvH1,&rinvH2);
				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);
				/* load 2 j charges and multiply by iq */
				jq=load_2_float(charge+jnra,charge+jnrb);
				qqM             = vec_madd(iqM,jq,nul);
				qqH             = vec_madd(iqH,jq,nul);
				krsqM           = vec_madd(vkrf,rsqM,nul);
				krsqH1          = vec_madd(vkrf,rsqH1,nul);
				krsqH2          = vec_madd(vkrf,rsqH2,nul);
				vcoulM          = vec_add(rinvM,krsqM);
				vcoulH1         = vec_add(rinvH1,krsqH1);
				vcoulH2         = vec_add(rinvH2,krsqH2);
				vcoulM          = vec_sub(vcoulM,vcrf);
				vcoulH1         = vec_sub(vcoulH1,vcrf);
				vcoulH2         = vec_sub(vcoulH2,vcrf);
				vctot           = vec_madd(qqM,vcoulM,vctot);
				fsM             = vec_nmsub(vec_two(),krsqM,rinvM);
				fsH1            = vec_nmsub(vec_two(),krsqH1,rinvH1);
				fsH2            = vec_nmsub(vec_two(),krsqH2,rinvH2);
				vctot           = vec_madd(qqH,vcoulH1,vctot);
				fsM             = vec_madd(fsM,qqM,nul);
				fsH1            = vec_madd(fsH1,qqH,nul);
				fsH2            = vec_madd(fsH2,qqH,nul);
				vctot           = vec_madd(qqH,vcoulH2,vctot);
				fsM             = vec_madd(fsM,rinvsqM,nul);
				fsH1            = vec_madd(fsH1,rinvsqH1,nul);
				fsH2            = vec_madd(fsH2,rinvsqH2,nul);

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

				transpose_3_to_2(dMx,dMy,dMz,&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),&dH2x,&dH2y,&dH2z);
				dMx             = vec_sub(iMx,dH2x);
				dMy             = vec_sub(iMy,dH2y);
				dMz             = vec_sub(iMz,dH2z);
				dH1x            = vec_sub(iH1x,dH2x);
				dH1y            = vec_sub(iH1y,dH2y);
				dH1z            = vec_sub(iH1z,dH2z);
				dH2x            = vec_sub(iH2x,dH2x);
				dH2y            = vec_sub(iH2y,dH2y);
				dH2z            = vec_sub(iH2z,dH2z);
      
				rsqM            = vec_madd(dMx,dMx,nul);
				rsqH1           = vec_madd(dH1x,dH1x,nul);
				rsqH2           = vec_madd(dH2x,dH2x,nul);
				rsqM            = vec_madd(dMy,dMy,rsqM);
				rsqH1           = vec_madd(dH1y,dH1y,rsqH1);
				rsqH2           = vec_madd(dH2y,dH2y,rsqH2);
				rsqM            = vec_madd(dMz,dMz,rsqM);
				rsqH1           = vec_madd(dH1z,dH1z,rsqH1);
				rsqH2           = vec_madd(dH2z,dH2z,rsqH2);

				zero_highest_3_elements_in_3_vectors(&rsqM,&rsqH1,&rsqH2);
				do_3_invsqrt(rsqM,rsqH1,rsqH2,&rinvM,&rinvH1,&rinvH2);
				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);
				/* load 1 j charges and multiply by iq */
				jq=load_1_float(charge+jnra);
				qqM             = vec_madd(iqM,jq,nul);
				qqH             = vec_madd(iqH,jq,nul);
				krsqM           = vec_madd(vkrf,rsqM,nul);
				krsqH1          = vec_madd(vkrf,rsqH1,nul);
				krsqH2          = vec_madd(vkrf,rsqH2,nul);
				vcoulM          = vec_add(rinvM,krsqM);
				vcoulH1         = vec_add(rinvH1,krsqH1);
				vcoulH2         = vec_add(rinvH2,krsqH2);
				vcoulM          = vec_sub(vcoulM,vcrf);
				vcoulH1         = vec_sub(vcoulH1,vcrf);
				vcoulH2         = vec_sub(vcoulH2,vcrf);
				vctot           = vec_madd(qqM,vcoulM,vctot);
				fsM             = vec_nmsub(vec_two(),krsqM,rinvM);
				fsH1            = vec_nmsub(vec_two(),krsqH1,rinvH1);
				fsH2            = vec_nmsub(vec_two(),krsqH2,rinvH2);
				vctot           = vec_madd(qqH,vcoulH1,vctot);
				fsM             = vec_madd(fsM,qqM,nul);
				fsH1            = vec_madd(fsH1,qqH,nul);
				fsH2            = vec_madd(fsH2,qqH,nul);
				vctot           = vec_madd(qqH,vcoulH2,vctot);
				fsM             = vec_madd(fsM,rinvsqM,nul);
				fsH1            = vec_madd(fsH1,rinvsqH1,nul);
				fsH2            = vec_madd(fsH2,rinvsqH2,nul);

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

				transpose_3_to_1(dMx,dMy,dMz,&tmp1);
				add_xyz_to_mem(faction+j3a,tmp1);
			}
			/* update outer data */
			update_i_3atoms_forces(faction+ii3+3,fshift+is3,
								   fiH1x,fiH1y,fiH1z,fiH2x,fiH2y,fiH2z,
								   fiMx,fiMy,fiMz);

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





void 
nb_kernel203nf_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,