nb_kernel332_ia32_sse.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,043 行 · 第 1/5 页
S
2,043 行
## First hydrogen Y & Z forces for 4 j particles movaps nb332_fjyH1(%esp),%xmm0 ## xmm0= fjyH1a fjyH1b fjyH1c fjyH1d movaps nb332_fjzH1(%esp),%xmm2 ## xmm1= fjzH1a fjzH1b fjzH1c fjzH1d movlps 16(%edi,%eax,4),%xmm3 movlps 16(%edi,%ecx,4),%xmm4 movaps %xmm0,%xmm1 unpcklps %xmm2,%xmm0 ## xmm0= fjyH1a fjzH1a fjyH1b fjzH1b unpckhps %xmm2,%xmm1 ## xmm1= fjyH1c fjzH1c fjyH1d fjzH1d movhps 16(%edi,%ebx,4),%xmm3 movhps 16(%edi,%edx,4),%xmm4 addps %xmm0,%xmm3 addps %xmm1,%xmm4 movlps %xmm3,16(%edi,%eax,4) movlps %xmm4,16(%edi,%ecx,4) movhps %xmm3,16(%edi,%ebx,4) movhps %xmm4,16(%edi,%edx,4) ## Second hydrogen X & Y forces for 4 j particles movaps nb332_fjxH2(%esp),%xmm0 ## xmm0= fjxH2a fjxH2b fjxH2c fjxH2d movaps nb332_fjyH2(%esp),%xmm2 ## xmm1= fjyH2a fjyH2b fjyH2c fjyH2d movlps 24(%edi,%eax,4),%xmm3 movlps 24(%edi,%ecx,4),%xmm4 movaps %xmm0,%xmm1 unpcklps %xmm2,%xmm0 ## xmm0= fjxH2a fjyH2a fjxH2b fjyH2b unpckhps %xmm2,%xmm1 ## xmm1= fjxH2c fjyH2c fjxH2d fjyH2d movhps 24(%edi,%ebx,4),%xmm3 movhps 24(%edi,%edx,4),%xmm4 addps %xmm0,%xmm3 addps %xmm1,%xmm4 movlps %xmm3,24(%edi,%eax,4) movlps %xmm4,24(%edi,%ecx,4) movhps %xmm3,24(%edi,%ebx,4) movhps %xmm4,24(%edi,%edx,4) ## Second hydrogen Z forces for 4 j particles ## Just load the four Z coords into one reg. each movss 32(%edi,%eax,4),%xmm4 movss 32(%edi,%ebx,4),%xmm5 movss 32(%edi,%ecx,4),%xmm6 movss 32(%edi,%edx,4),%xmm7 ## add what we have on the stack addss nb332_fjzH2(%esp),%xmm4 addss nb332_fjzH2b(%esp),%xmm5 addss nb332_fjzH2c(%esp),%xmm6 addss nb332_fjzH2d(%esp),%xmm7 ## store back movss %xmm4,32(%edi,%eax,4) movss %xmm5,32(%edi,%ebx,4) movss %xmm6,32(%edi,%ecx,4) movss %xmm7,32(%edi,%edx,4) ## should we do one more iteration? subl $4,nb332_innerk(%esp) jl _nb_kernel332_ia32_sse.nb332_single_check jmp _nb_kernel332_ia32_sse.nb332_unroll_loop_nb_kernel332_ia32_sse.nb332_single_check: addl $4,nb332_innerk(%esp) jnz _nb_kernel332_ia32_sse.nb332_single_loop jmp _nb_kernel332_ia32_sse.nb332_updateouterdata_nb_kernel332_ia32_sse.nb332_single_loop: movl nb332_innerjjnr(%esp),%edx ## pointer to jjnr[k] movl (%edx),%eax addl $4,nb332_innerjjnr(%esp) movl nb332_pos(%ebp),%esi leal (%eax,%eax,2),%eax ## fetch j coordinates xorps %xmm3,%xmm3 xorps %xmm4,%xmm4 xorps %xmm5,%xmm5 movss (%esi,%eax,4),%xmm3 ## jxO - - - movss 4(%esi,%eax,4),%xmm4 ## jyO - - - movss 8(%esi,%eax,4),%xmm5 ## jzO - - - movlps 12(%esi,%eax,4),%xmm6 ## xmm6 = jxH1 jyH1 - - movss 20(%esi,%eax,4),%xmm7 ## xmm7 = jzH1 - - - movhps 24(%esi,%eax,4),%xmm6 ## xmm6 = jxH1 jyH1 jxH2 jyH2 movss 32(%esi,%eax,4),%xmm2 ## xmm2 = jzH2 - - - ## have all coords, time for some shuffling. shufps $216,%xmm6,%xmm6 ## constant 11011000 ;# xmm6 = jxH1 jxH2 jyH1 jyH2 unpcklps %xmm2,%xmm7 ## xmm7 = jzH1 jzH2 - - movaps nb332_ixO(%esp),%xmm0 movaps nb332_iyO(%esp),%xmm1 movaps nb332_izO(%esp),%xmm2 movlhps %xmm6,%xmm3 ## xmm3 = jxO 0 jxH1 jxH2 shufps $228,%xmm6,%xmm4 ## constant 11100100 ;# xmm4 = jyO 0 jyH1 jyH2 shufps $68,%xmm7,%xmm5 ## constant 01000100 ;# xmm5 = jzO 0 jzH1 jzH2 ## store all j coordinates in jO movaps %xmm3,nb332_jxO(%esp) movaps %xmm4,nb332_jyO(%esp) movaps %xmm5,nb332_jzO(%esp) subps %xmm3,%xmm0 subps %xmm4,%xmm1 subps %xmm5,%xmm2 movaps %xmm0,nb332_dxOO(%esp) movaps %xmm1,nb332_dyOO(%esp) movaps %xmm2,nb332_dzOO(%esp) mulps %xmm0,%xmm0 mulps %xmm1,%xmm1 mulps %xmm2,%xmm2 addps %xmm1,%xmm0 addps %xmm2,%xmm0 ## have rsq in xmm0 ## do invsqrt rsqrtps %xmm0,%xmm1 movaps %xmm1,%xmm2 mulps %xmm1,%xmm1 movaps nb332_three(%esp),%xmm3 mulps %xmm0,%xmm1 subps %xmm1,%xmm3 mulps %xmm2,%xmm3 mulps nb332_half(%esp),%xmm3 ## rinv iO - j water movaps %xmm3,%xmm1 mulps %xmm0,%xmm1 ## xmm1=r movaps %xmm3,%xmm0 ## xmm0=rinv mulps nb332_tsc(%esp),%xmm1 movhlps %xmm1,%xmm2 cvttps2pi %xmm1,%mm6 cvttps2pi %xmm2,%mm7 ## mm6/mm7 contain lu indices cvtpi2ps %mm6,%xmm3 cvtpi2ps %mm7,%xmm2 movlhps %xmm2,%xmm3 subps %xmm3,%xmm1 ## xmm1=eps movaps %xmm1,%xmm2 mulps %xmm2,%xmm2 ## xmm2=eps2 pslld $2,%mm6 pslld $2,%mm7 movd %mm6,%ebx movd %mm7,%ecx psrlq $32,%mm7 movd %mm7,%edx ## table indices in ebx,ecx,edx movl nb332_VFtab(%ebp),%esi leal (%ebx,%ebx,2),%ebx leal (%ecx,%ecx,2),%ecx leal (%edx,%edx,2),%edx movlps (%esi,%ebx,4),%xmm5 movlps (%esi,%ecx,4),%xmm7 movhps (%esi,%edx,4),%xmm7 ## got half coulomb table movaps %xmm5,%xmm4 shufps $136,%xmm7,%xmm4 ## constant 10001000 shufps $221,%xmm7,%xmm5 ## constant 11011101 movlps 8(%esi,%ebx,4),%xmm7 movlps 8(%esi,%ecx,4),%xmm3 movhps 8(%esi,%edx,4),%xmm3 ## other half of coulomb table movaps %xmm7,%xmm6 shufps $136,%xmm3,%xmm6 ## constant 10001000 shufps $221,%xmm3,%xmm7 ## constant 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 nb332_two(%esp),%xmm7 ## two*Heps2 xorps %xmm3,%xmm3 ## fetch charges to xmm3 (temporary) movss nb332_qqOO(%esp),%xmm3 movhps nb332_qqOH(%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 xmm5 contains vcoul and xmm3 fijC addps nb332_vctot(%esp),%xmm5 movaps %xmm5,nb332_vctot(%esp) ## put scalar force on stack temporarily movaps %xmm3,nb332_fstmp(%esp) ## dispersion movss 16(%esi,%ebx,4),%xmm4 movss 20(%esi,%ebx,4),%xmm5 movss 24(%esi,%ebx,4),%xmm6 movss 28(%esi,%ebx,4),%xmm7 ## dispersion table ready, in xmm4-xmm7 mulss %xmm1,%xmm6 ## xmm6=Geps mulss %xmm2,%xmm7 ## xmm7=Heps2 addss %xmm6,%xmm5 addss %xmm7,%xmm5 ## xmm5=Fp mulss nb332_two(%esp),%xmm7 ## two*Heps2 addss %xmm6,%xmm7 addss %xmm5,%xmm7 ## xmm7=FF mulss %xmm1,%xmm5 ## xmm5=eps*Fp addss %xmm4,%xmm5 ## xmm5=VV xorps %xmm4,%xmm4 movss nb332_c6(%esp),%xmm4 mulps %xmm4,%xmm7 ## fijD mulps %xmm4,%xmm5 ## Vvdw6 addps nb332_fstmp(%esp),%xmm7 ## add to fscal ## put scalar force on stack Update Vvdwtot directly addps nb332_Vvdwtot(%esp),%xmm5 movaps %xmm7,nb332_fstmp(%esp) movaps %xmm5,nb332_Vvdwtot(%esp) ## repulsion movss 32(%esi,%ebx,4),%xmm4 movss 36(%esi,%ebx,4),%xmm5 movss 40(%esi,%ebx,4),%xmm6 movss 44(%esi,%ebx,4),%xmm7 ## table ready, in xmm4-xmm7 mulss %xmm1,%xmm6 ## xmm6=Geps mulss %xmm2,%xmm7 ## xmm7=Heps2 addss %xmm6,%xmm5 addss %xmm7,%xmm5 ## xmm5=Fp mulss nb332_two(%esp),%xmm7 ## two*Heps2 addss %xmm6,%xmm7 addss %xmm5,%xmm7 ## xmm7=FF mulss %xmm1,%xmm5 ## xmm5=eps*Fp addss %xmm4,%xmm5 ## xmm5=VV xorps %xmm4,%xmm4 movss nb332_c12(%esp),%xmm4 mulps %xmm4,%xmm7 ## fijR mulps %xmm4,%xmm5 ## Vvdw12 addps nb332_fstmp(%esp),%xmm7 addps nb332_Vvdwtot(%esp),%xmm5 movaps %xmm5,nb332_Vvdwtot(%esp) xorps %xmm1,%xmm1 mulps nb332_tsc(%esp),%xmm7 mulps %xmm0,%xmm7 subps %xmm7,%xmm1 movaps %xmm1,%xmm0 movaps %xmm1,%xmm2 mulps nb332_dxOO(%esp),%xmm0 mulps nb332_dyOO(%esp),%xmm1 mulps nb332_dzOO(%esp),%xmm2 ## initial update for j forces xorps %xmm3,%xmm3 xorps %xmm4,%xmm4 xorps %xmm5,%xmm5 subps %xmm0,%xmm3 subps %xmm1,%xmm4 subps %xmm2,%xmm5 movaps %xmm3,nb332_fjxO(%esp) movaps %xmm4,nb332_fjyO(%esp) movaps %xmm5,nb332_fjzO(%esp) addps nb332_fixO(%esp),%xmm0 addps nb332_fiyO(%esp),%xmm1 addps nb332_fizO(%esp),%xmm2 movaps %xmm0,nb332_fixO(%esp) movaps %xmm1,nb332_fiyO(%esp) movaps %xmm2,nb332_fizO(%esp) ## done with i O Now do i H1 & H2 simultaneously first get i particle coords: movaps nb332_ixH1(%esp),%xmm0 movaps nb332_iyH1(%esp),%xmm1 movaps nb332_izH1(%esp),%xmm2 movaps nb332_ixH2(%esp),%xmm3 movaps nb332_iyH2(%esp),%xmm4 movaps nb332_izH2(%esp),%xmm5 subps nb332_jxO(%esp),%xmm0 subps nb332_jyO(%esp),%xmm1 subps nb332_jzO(%esp),%xmm2 subps nb332_jxO(%esp),%xmm3 subps nb332_jyO(%esp),%xmm4 subps nb332_jzO(%esp),%xmm5 movaps %xmm0,nb332_dxH1O(%esp) movaps %xmm1,nb332_dyH1O(%esp) movaps %xmm2,nb332_dzH1O(%esp) movaps %xmm3,nb332_dxH2O(%esp) movaps %xmm4,nb332_dyH2O(%esp) movaps %xmm5,nb332_dzH2O(%esp) mulps %xmm0,%xmm0 mulps %xmm1,%xmm1 mulps %xmm2,%xmm2 mulps %xmm3,%xmm3 mulps %xmm4,%xmm4 mulps %xmm5,%xmm5 addps %xmm1,%xmm0 addps %xmm3,%xmm4 addps %xmm2,%xmm0 ## have rsqH1 in xmm0 addps %xmm5,%xmm4 ## have rsqH2 in xmm4 ## start with H1, save H2 data movaps %xmm4,nb332_rsqH2O(%esp) ## do invsqrt rsqrtps %xmm0,%xmm1 rsqrtps %xmm4,%xmm5 movaps %xmm1,%xmm2 movaps %xmm5,%xmm6 mulps %xmm1,%xmm1 mulps %xmm5,%xmm5 movaps nb332_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 nb332_half(%esp),%xmm3 ## rinv H1 - j water mulps nb332_half(%esp),%xmm7 ## rinv H2 - j water ## start with H1, save H2 data movaps %xmm7,nb332_rinvH2O(%esp) movaps %xmm3,%xmm1 mulps %xmm0,%xmm1 ## xmm1=r movaps %xmm3,%xmm0 ## xmm0=rinv mulps nb332_tsc(%esp),%xmm1 movhlps %xmm1,%xmm2 cvttps2pi %xmm1,%mm6 cvttps2pi %xmm2,%mm7 ## mm6/mm7 contain lu indices cvtpi2ps %mm6,%xmm3 cvtpi2ps %mm7,%xmm2 movlhps %xmm2,%xmm3 subps %xmm3,%xmm1 ## xmm1=eps movaps %xmm1,%xmm2 mulps %xmm2,%xmm2 ## xmm2=eps2 pslld $2,%mm6 pslld $2,%mm7 movd %mm6,%ebx movd %mm7,%ecx psrlq $32,%mm7 movd %mm7,%edx ## table indices in ebx,ecx,edx leal (%ebx,%ebx,2),%ebx leal (%ecx,%ecx,2),%ecx leal (%edx,%edx,2),%edx movlps (%esi,%ebx,4),%xmm5 movlps (%esi,%ecx,4),%xmm7 movhps (%esi,%edx,4),%xmm7 ## got half coulomb table movaps %xmm5,%xmm4 shufps $136,%xmm7,%xmm4 ## constant 10001000 shufps $221,%xmm7,%xmm5 ## constant 11011101 movlps 8(%esi,%ebx,4),%xmm7 movlps 8(%esi,%ecx,4),%xmm3 movhps 8(%esi,%edx,4),%xmm3 ## other half of coulomb table movaps %xmm7,%xmm6 shufps $136,%xmm3,%xmm6 ## constant 10001000 shufps $221,%xmm3,%xmm7 ## constant 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 nb332_two(%esp),%xmm7 ## two*Heps2 xorps %xmm3,%xmm3 ## fetch charges to xmm3 (temporary) movss nb332_qqOH(%esp),%xmm3 movhps nb332_qqHH(%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 xmm5 contains vcoul and xmm3 fijC addps nb332_vctot(%esp),%xmm5 movaps %xmm5,nb332_vctot(%esp) xorps %xmm1,%xmm1 mulps nb332_tsc(%esp),%xmm3 mulps %xmm0,%xmm3 subps %xmm3,%xmm1 movaps %xmm1,%xmm0 movaps %xmm1,%xmm2 mulps nb332_dxH1O(%esp),%xmm0 mulps nb332_dyH1O(%esp),%xmm1 mulps nb332_dzH1O(%esp),%xmm2 ## update forces H1 - j water movaps nb332_fjxO(%esp),%xmm3 movaps nb332_fjyO(%esp),%xmm4 movaps nb332_fjzO(%esp),%xmm5 subps %xmm0,%xmm3 subps %xmm1,%xmm4 subps %xmm2,%xmm5 movaps %xmm3,nb332_fjxO(%esp) movaps %xmm4,nb332_fjyO(%esp) movaps %xmm5,nb332_fjzO(%esp) addps nb332_fixH1(%esp),%xmm0 addps nb332_fiyH1(%esp),%xmm1 addps nb332_fizH1(%esp),%xmm2 movaps %xmm0,nb332_fixH1(%esp) movaps %xmm1,nb332_fiyH1(%esp) movaps ⌨️ 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?