nb_kernel334_ia32_sse.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,029 行 · 第 1/5 页
S
2,029 行
movd %mm6,%eax psrlq $32,%mm6 movd %mm7,%ecx psrlq $32,%mm7 movd %mm6,%ebx movd %mm7,%edx leal (%eax,%eax,2),%eax leal (%ebx,%ebx,2),%ebx leal (%ecx,%ecx,2),%ecx leal (%edx,%edx,2),%edx movlps (%esi,%eax,4),%xmm5 movlps (%esi,%ecx,4),%xmm7 movhps (%esi,%ebx,4),%xmm5 movhps (%esi,%edx,4),%xmm7 ## got half coulomb table movaps %xmm5,%xmm4 shufps $136,%xmm7,%xmm4 ## 10001000 shufps $221,%xmm7,%xmm5 ## 11011101 movlps 8(%esi,%eax,4),%xmm7 movlps 8(%esi,%ecx,4),%xmm3 movhps 8(%esi,%ebx,4),%xmm7 movhps 8(%esi,%edx,4),%xmm3 ## other half of coulomb table movaps %xmm7,%xmm6 shufps $136,%xmm3,%xmm6 ## 10001000 shufps $221,%xmm3,%xmm7 ## 11011101 ## coulomb table ready, in xmm4-xmm7 mulps %xmm1,%xmm6 ## xmm6=Geps mulps %xmm2,%xmm7 ## xmm7=Heps2 addps %xmm6,%xmm5 addps %xmm7,%xmm5 ## xmm5=Fp mulps nb334_two(%esp),%xmm7 ## two*Heps2 movaps nb334_qqMM(%esp),%xmm3 addps %xmm6,%xmm7 addps %xmm5,%xmm7 ## xmm7=FF mulps %xmm1,%xmm5 ## xmm5=eps*Fp addps %xmm4,%xmm5 ## xmm5=VV mulps %xmm3,%xmm5 ## vcoul=qq*VV mulps %xmm7,%xmm3 ## fijC=FF*qq ## at this point mm5 contains vcoul and mm3 fijC addps nb334_vctot(%esp),%xmm5 movaps %xmm5,nb334_vctot(%esp) xorps %xmm1,%xmm1 mulps nb334_tsc(%esp),%xmm3 mulps %xmm0,%xmm3 subps %xmm3,%xmm1 movaps %xmm1,%xmm0 movaps %xmm1,%xmm2 movaps nb334_fjxM(%esp),%xmm3 movaps nb334_fjyM(%esp),%xmm4 movaps nb334_fjzM(%esp),%xmm5 mulps nb334_dxMM(%esp),%xmm0 mulps nb334_dyMM(%esp),%xmm1 mulps nb334_dzMM(%esp),%xmm2 subps %xmm0,%xmm3 subps %xmm1,%xmm4 subps %xmm2,%xmm5 addps nb334_fixM(%esp),%xmm0 addps nb334_fiyM(%esp),%xmm1 addps nb334_fizM(%esp),%xmm2 movaps %xmm3,nb334_fjxM(%esp) movaps %xmm4,nb334_fjyM(%esp) movaps %xmm5,nb334_fjzM(%esp) movaps %xmm0,nb334_fixM(%esp) movaps %xmm1,nb334_fiyM(%esp) movaps %xmm2,nb334_fizM(%esp) movl nb334_faction(%ebp),%edi movd %mm0,%eax movd %mm1,%ebx movd %mm2,%ecx movd %mm3,%edx ## Did all interactions - now update j forces ## 4 j waters with four atoms each. ## step 1 : transpose fjxO, fjyO, fjzO, fjxH1 movaps nb334_fjxO(%esp),%xmm0 movaps nb334_fjyO(%esp),%xmm1 movaps nb334_fjzO(%esp),%xmm2 movaps nb334_fjxH1(%esp),%xmm3 movaps %xmm0,%xmm4 movaps %xmm1,%xmm5 unpcklps %xmm2,%xmm4 unpcklps %xmm3,%xmm5 unpckhps %xmm2,%xmm0 unpckhps %xmm3,%xmm1 movaps %xmm4,%xmm2 movaps %xmm0,%xmm3 unpcklps %xmm5,%xmm4 unpckhps %xmm5,%xmm2 unpcklps %xmm1,%xmm0 unpckhps %xmm1,%xmm3 ## results are now in xmm4, xmm2, xmm0, xmm3 ## load the corresponding j forces from memory movlps (%edi,%eax,4),%xmm1 movlps (%edi,%ebx,4),%xmm5 movlps (%edi,%ecx,4),%xmm6 movlps (%edi,%edx,4),%xmm7 movhps 8(%edi,%eax,4),%xmm1 movhps 8(%edi,%ebx,4),%xmm5 movhps 8(%edi,%ecx,4),%xmm6 movhps 8(%edi,%edx,4),%xmm7 ## add addps %xmm4,%xmm1 addps %xmm2,%xmm5 addps %xmm0,%xmm6 addps %xmm3,%xmm7 ## store back movlps %xmm1,(%edi,%eax,4) movlps %xmm5,(%edi,%ebx,4) movlps %xmm6,(%edi,%ecx,4) movlps %xmm7,(%edi,%edx,4) movhps %xmm1,8(%edi,%eax,4) movhps %xmm5,8(%edi,%ebx,4) movhps %xmm6,8(%edi,%ecx,4) movhps %xmm7,8(%edi,%edx,4) ## step 2 : transpose fjyH1, fjzH1, fjxH2, fjyH2 movaps nb334_fjyH1(%esp),%xmm0 movaps nb334_fjzH1(%esp),%xmm1 movaps nb334_fjxH2(%esp),%xmm2 movaps nb334_fjyH2(%esp),%xmm3 movaps %xmm0,%xmm4 movaps %xmm1,%xmm5 unpcklps %xmm2,%xmm4 unpcklps %xmm3,%xmm5 unpckhps %xmm2,%xmm0 unpckhps %xmm3,%xmm1 movaps %xmm4,%xmm2 movaps %xmm0,%xmm3 unpcklps %xmm5,%xmm4 unpckhps %xmm5,%xmm2 unpcklps %xmm1,%xmm0 unpckhps %xmm1,%xmm3 ## results are now in xmm4, xmm2, xmm0, xmm3 ## load the corresponding j forces from memory movlps 16(%edi,%eax,4),%xmm1 movlps 16(%edi,%ebx,4),%xmm5 movlps 16(%edi,%ecx,4),%xmm6 movlps 16(%edi,%edx,4),%xmm7 movhps 24(%edi,%eax,4),%xmm1 movhps 24(%edi,%ebx,4),%xmm5 movhps 24(%edi,%ecx,4),%xmm6 movhps 24(%edi,%edx,4),%xmm7 ## add addps %xmm4,%xmm1 addps %xmm2,%xmm5 addps %xmm0,%xmm6 addps %xmm3,%xmm7 ## store back movlps %xmm1,16(%edi,%eax,4) movlps %xmm5,16(%edi,%ebx,4) movlps %xmm6,16(%edi,%ecx,4) movlps %xmm7,16(%edi,%edx,4) movhps %xmm1,24(%edi,%eax,4) movhps %xmm5,24(%edi,%ebx,4) movhps %xmm6,24(%edi,%ecx,4) movhps %xmm7,24(%edi,%edx,4) ## step 3 : transpose fjzH2, fjxM, fjyM, fjzM movaps nb334_fjzH2(%esp),%xmm0 movaps nb334_fjxM(%esp),%xmm1 movaps nb334_fjyM(%esp),%xmm2 movaps nb334_fjzM(%esp),%xmm3 movaps %xmm0,%xmm4 movaps %xmm1,%xmm5 unpcklps %xmm2,%xmm4 unpcklps %xmm3,%xmm5 unpckhps %xmm2,%xmm0 unpckhps %xmm3,%xmm1 movaps %xmm4,%xmm2 movaps %xmm0,%xmm3 unpcklps %xmm5,%xmm4 unpckhps %xmm5,%xmm2 unpcklps %xmm1,%xmm0 unpckhps %xmm1,%xmm3 ## results are now in xmm4, xmm2, xmm0, xmm3 ## load the corresponding j forces from memory movlps 32(%edi,%eax,4),%xmm1 movlps 32(%edi,%ebx,4),%xmm5 movlps 32(%edi,%ecx,4),%xmm6 movlps 32(%edi,%edx,4),%xmm7 movhps 40(%edi,%eax,4),%xmm1 movhps 40(%edi,%ebx,4),%xmm5 movhps 40(%edi,%ecx,4),%xmm6 movhps 40(%edi,%edx,4),%xmm7 ## add addps %xmm4,%xmm1 addps %xmm2,%xmm5 addps %xmm0,%xmm6 addps %xmm3,%xmm7 ## store back movlps %xmm1,32(%edi,%eax,4) movlps %xmm5,32(%edi,%ebx,4) movlps %xmm6,32(%edi,%ecx,4) movlps %xmm7,32(%edi,%edx,4) movhps %xmm1,40(%edi,%eax,4) movhps %xmm5,40(%edi,%ebx,4) movhps %xmm6,40(%edi,%ecx,4) movhps %xmm7,40(%edi,%edx,4) ## should we do one more iteration? subl $4,nb334_innerk(%esp) jl _nb_kernel334_ia32_sse.nb334_single_check jmp _nb_kernel334_ia32_sse.nb334_unroll_loop_nb_kernel334_ia32_sse.nb334_single_check: addl $4,nb334_innerk(%esp) jnz _nb_kernel334_ia32_sse.nb334_single_loop jmp _nb_kernel334_ia32_sse.nb334_updateouterdata_nb_kernel334_ia32_sse.nb334_single_loop: movl nb334_innerjjnr(%esp),%edx ## pointer to jjnr[k] movl (%edx),%eax addl $4,nb334_innerjjnr(%esp) movl nb334_pos(%ebp),%esi leal (%eax,%eax,2),%eax ## fetch j coordinates movlps (%esi,%eax,4),%xmm3 ## Ox Oy movlps 16(%esi,%eax,4),%xmm4 ## H1y H1z movlps 32(%esi,%eax,4),%xmm5 ## H2z Mx movhps 8(%esi,%eax,4),%xmm3 ## Ox Oy Oz H1x movhps 24(%esi,%eax,4),%xmm4 ## H1y H1z H2x H2y movhps 40(%esi,%eax,4),%xmm5 ## H2z Mx My Mz ## transpose movaps %xmm4,%xmm0 movaps %xmm3,%xmm1 movaps %xmm4,%xmm2 movaps %xmm3,%xmm6 shufps $18,%xmm5,%xmm4 ## (00010010) h2x - Mx - shufps $193,%xmm0,%xmm3 ## (11000001) Oy - H1y - shufps $35,%xmm5,%xmm2 ## (00100011) H2y - My - shufps $18,%xmm0,%xmm1 ## (00010010) Oz - H1z - ## xmm6: Ox - - H1x xmm5: H2z - - Mz shufps $140,%xmm4,%xmm6 ## (10001100) Ox H1x H2x Mx shufps $136,%xmm2,%xmm3 ## (10001000) Oy H1y H2y My shufps $200,%xmm5,%xmm1 ## (11001000) Oz H1z H2z Mz ## store all j coordinates in jO movaps %xmm6,nb334_jxO(%esp) movaps %xmm3,nb334_jyO(%esp) movaps %xmm1,nb334_jzO(%esp) ## do O and H1 in parallel movaps nb334_ixO(%esp),%xmm0 movaps nb334_iyO(%esp),%xmm1 movaps nb334_izO(%esp),%xmm2 movaps nb334_ixH1(%esp),%xmm3 movaps nb334_iyH1(%esp),%xmm4 movaps nb334_izH1(%esp),%xmm5 subps nb334_jxO(%esp),%xmm0 subps nb334_jyO(%esp),%xmm1 subps nb334_jzO(%esp),%xmm2 subps nb334_jxO(%esp),%xmm3 subps nb334_jyO(%esp),%xmm4 subps nb334_jzO(%esp),%xmm5 movaps %xmm0,nb334_dxOO(%esp) movaps %xmm1,nb334_dyOO(%esp) movaps %xmm2,nb334_dzOO(%esp) movaps %xmm3,nb334_dxH1H1(%esp) movaps %xmm4,nb334_dyH1H1(%esp) movaps %xmm5,nb334_dzH1H1(%esp) mulps %xmm0,%xmm0 mulps %xmm1,%xmm1 mulps %xmm2,%xmm2 addps %xmm1,%xmm0 addps %xmm2,%xmm0 ## have rsq in xmm0 mulps %xmm3,%xmm3 mulps %xmm4,%xmm4 mulps %xmm5,%xmm5 addps %xmm3,%xmm4 addps %xmm5,%xmm4 ## have rsq in xmm4 movaps %xmm0,nb334_rsqOO(%esp) movaps %xmm4,nb334_rsqH1H1(%esp) ## do 1/sqrt(x) for O and H1 rsqrtps %xmm0,%xmm1 rsqrtps %xmm4,%xmm5 movaps %xmm1,%xmm2 movaps %xmm5,%xmm6 mulps %xmm1,%xmm1 mulps %xmm5,%xmm5 movaps nb334_three(%esp),%xmm3 movaps %xmm3,%xmm7 mulps %xmm0,%xmm1 mulps %xmm4,%xmm5 subps %xmm1,%xmm3 subps %xmm5,%xmm7 mulps %xmm2,%xmm3 mulps %xmm6,%xmm7 mulps nb334_half(%esp),%xmm3 ## rinv O - j water mulps nb334_half(%esp),%xmm7 ## rinv H1 - j water movaps %xmm3,nb334_rinvOO(%esp) movaps %xmm7,nb334_rinvH1H1(%esp) movl nb334_VFtab(%ebp),%esi ## do O table LJ interaction movaps %xmm3,%xmm0 movaps %xmm0,%xmm1 mulss nb334_rsqOO(%esp),%xmm1 ## xmm1=r mulss nb334_tsc(%esp),%xmm1 cvttps2pi %xmm1,%mm6 cvtpi2ps %mm6,%xmm3 subss %xmm3,%xmm1 ## xmm1=eps movaps %xmm1,%xmm2 mulss %xmm2,%xmm2 ## xmm2=eps2 pslld $2,%mm6 movd %mm6,%ebx leal (%ebx,%ebx,2),%ebx ## load dispersion table data into xmm4 movlps 16(%esi,%ebx,4),%xmm4 movlps 24(%esi,%ebx,4),%xmm6 movaps %xmm4,%xmm5 movaps %xmm6,%xmm7 shufps $0x1,%xmm5,%xmm5 shufps $0x1,%xmm7,%xmm7 ## dispersion table YFGH ready in xmm4-xmm7 mulss %xmm1,%xmm6 ## xmm6=Geps mulss %xmm2,%xmm7 ## xmm7=Heps2 addss %xmm6,%xmm5 addss %xmm7,%xmm5 ## xmm5=Fp mulss nb334_two(%esp),%xmm7 ## two*Heps2 addss %xmm6,%xmm7 addss %xmm5,%xmm7 ## xmm7=FF mulss %xmm1,%xmm5 ## xmm5=eps*Fp addss %xmm4,%xmm5 ## xmm5=VV movaps nb334_c6(%esp),%xmm4 mulss %xmm4,%xmm7 ## fijD mulss %xmm4,%xmm5 ## Vvdw6 ## save scalar force in xmm3. Update Vvdwtot directly addss nb334_Vvdwtot(%esp),%xmm5 movaps %xmm7,%xmm3 ## fscal movss %xmm5,nb334_Vvdwtot(%esp) ## load repulsion table data into xmm4 movlps 32(%esi,%ebx,4),%xmm4 movlps 40(%esi,%ebx,4),%xmm6 movaps %xmm4,%xmm5 movaps %xmm6,%xmm7 shufps $0x1,%xmm5,%xmm5 shufps $0x1,%xmm7,%xmm7 ## repulsion table YFGH ready in xmm4-xmm7 mulss %xmm1,%xmm6 ## xmm6=Geps mulss %xmm2,%xmm7 ## xmm7=Heps2 addss %xmm6,%xmm5 addss %xmm7,%xmm5 ## xmm5=Fp mulss nb334_two(%esp),%xmm7 ## two*Heps2 addss %xmm6,%xmm7 addss %xmm5,%xmm7 ## xmm7=FF mulss %xmm1,%xmm5 ## xmm5=eps*Fp addss %xmm4,%xmm5 ## xmm5=VV movaps nb334_c12(%esp),%xmm4 mulss %xmm4,%xmm7 ## fijR mulss %xmm4,%xmm5 ## Vvdw12 addss %xmm3,%xmm7 addss nb334_Vvdwtot(%esp),%xmm5 movss %xmm5,nb334_Vvdwtot(%esp) xorps %xmm1,%xmm1 mulss nb334_tsc(%esp),%xmm7 mulss %xmm0,%xmm7 subss %xmm7,%xmm1 movaps %xmm1,%xmm0 movaps %xmm1,%xmm2 mulss nb334_dxOO(%esp),%xmm0 mulss nb334_dyOO(%esp),%xmm1 mulss nb334_dzOO(%esp),%xmm2 xorps %xmm3,%xmm3 xorps %xmm4,%xmm4 xorps %xmm5,%xmm5 subss %xmm0,%xmm3 subss %xmm1,%xmm4 subss %xmm2,%xmm5 movaps %xmm3,nb334_fjxO(%esp) movaps %xmm4,nb334_fjyO(%esp) movaps %xmm5,nb334_fjzO(%esp) addss nb334_fixO(%esp),%xmm0 addss nb334_fiyO(%esp),%xmm1 addss nb334_fizO(%esp),%xmm2 movss %xmm0,nb334_fixO(%esp) mov⌨️ 快捷键说明
复制代码Ctrl + C
搜索代码Ctrl + F
全屏模式F11
增大字号Ctrl + =
减小字号Ctrl + -
显示快捷键?