nb_kernel232_x86_64_sse.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,138 行 · 第 1/5 页
S
2,138 行
## do invsqrt rsqrtps %xmm0,%xmm1 mulps nb232_krf(%rsp),%xmm6 ## xmm6=krsq movaps %xmm1,%xmm2 movaps %xmm6,%xmm7 mulps %xmm1,%xmm1 movaps nb232_three(%rsp),%xmm3 mulps %xmm0,%xmm1 subps %xmm1,%xmm3 mulps %xmm2,%xmm3 mulps nb232_half(%rsp),%xmm3 ## rinv iO - j water in xmm3 movaps %xmm3,nb232_rinvOO(%rsp) addps %xmm3,%xmm6 ## xmm6=rinv+ krsq mulps nb232_two(%rsp),%xmm7 subps nb232_crf(%rsp),%xmm6 ## xmm6=rinv+ krsq-crf movaps %xmm3,%xmm0 ## rinv subps %xmm7,%xmm3 ## xmm3=rinv-2*krsq xorps %xmm4,%xmm4 ## fetch charges to xmm4 (temporary) movss nb232_qqOO(%rsp),%xmm4 movhps nb232_qqOH(%rsp),%xmm4 mulps %xmm4,%xmm6 ## vcoul mulps %xmm4,%xmm3 ## coul part of fs mulps %xmm0,%xmm3 movaps %xmm3,nb232_fstmp(%rsp) ## save it addps nb232_vctot(%rsp),%xmm6 movaps %xmm6,nb232_vctot(%rsp) movaps nb232_rinvOO(%rsp),%xmm0 movss %xmm0,%xmm1 mulss nb232_rsqOO(%rsp),%xmm1 ## xmm1=r mulss nb232_tsc(%rsp),%xmm1 cvttps2pi %xmm1,%mm6 cvtpi2ps %mm6,%xmm3 subss %xmm3,%xmm1 ## xmm1=eps movss %xmm1,%xmm2 mulss %xmm2,%xmm2 ## xmm2=eps2 pslld $3,%mm6 movq nb232_VFtab(%rbp),%rsi movd %mm6,%r8d ## dispersion movlps (%rsi,%r8,4),%xmm5 movaps %xmm5,%xmm4 shufps $136,%xmm7,%xmm4 ## constant 10001000 shufps $221,%xmm7,%xmm5 ## constant 11011101 movlps 8(%rsi,%r8,4),%xmm7 movaps %xmm7,%xmm6 shufps $136,%xmm3,%xmm6 ## constant 10001000 shufps $221,%xmm3,%xmm7 ## constant 11011101 ## 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 nb232_two(%rsp),%xmm7 ## two*Heps2 addss %xmm6,%xmm7 addss %xmm5,%xmm7 ## xmm7=FF mulss %xmm1,%xmm5 ## xmm5=eps*Fp addss %xmm4,%xmm5 ## xmm5=VV movss nb232_c6(%rsp),%xmm4 mulss %xmm4,%xmm7 ## fijD mulss %xmm4,%xmm5 ## Vvdw6 movss nb232_fstmp(%rsp),%xmm3 mulps nb232_tsc(%rsp),%xmm7 subss %xmm7,%xmm3 movss %xmm3,nb232_fstmp(%rsp) addss nb232_Vvdwtot(%rsp),%xmm5 movss %xmm5,nb232_Vvdwtot(%rsp) ## repulsion movlps 16(%rsi,%r8,4),%xmm5 movaps %xmm5,%xmm4 shufps $136,%xmm7,%xmm4 ## constant 10001000 shufps $221,%xmm7,%xmm5 ## constant 11011101 movlps 24(%rsi,%r8,4),%xmm7 movaps %xmm7,%xmm6 shufps $136,%xmm3,%xmm6 ## constant 10001000 shufps $221,%xmm3,%xmm7 ## constant 11011101 ## table ready, in xmm4-xmm7 mulss %xmm1,%xmm6 ## xmm6=Geps mulss %xmm2,%xmm7 ## xmm7=Heps2 addss %xmm6,%xmm5 addss %xmm7,%xmm5 ## xmm5=Fp mulss nb232_two(%rsp),%xmm7 ## two*Heps2 addss %xmm6,%xmm7 addss %xmm5,%xmm7 ## xmm7=FF mulss %xmm1,%xmm5 ## xmm5=eps*Fp addss %xmm4,%xmm5 ## xmm5=VV movss nb232_c12(%rsp),%xmm4 mulss %xmm4,%xmm7 ## fijR mulss %xmm4,%xmm5 ## Vvdw12 movaps nb232_fstmp(%rsp),%xmm3 mulss nb232_tsc(%rsp),%xmm7 subss %xmm7,%xmm3 addss nb232_Vvdwtot(%rsp),%xmm5 movss %xmm5,nb232_Vvdwtot(%rsp) mulps %xmm3,%xmm0 movaps %xmm0,%xmm1 movaps %xmm0,%xmm2 mulps nb232_dxOO(%rsp),%xmm0 mulps nb232_dyOO(%rsp),%xmm1 mulps nb232_dzOO(%rsp),%xmm2 ## initial update for j forces xorps %xmm3,%xmm3 xorps %xmm4,%xmm4 xorps %xmm5,%xmm5 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm5 movaps %xmm3,nb232_fjxO(%rsp) movaps %xmm4,nb232_fjyO(%rsp) movaps %xmm5,nb232_fjzO(%rsp) addps nb232_fixO(%rsp),%xmm0 addps nb232_fiyO(%rsp),%xmm1 addps nb232_fizO(%rsp),%xmm2 movaps %xmm0,nb232_fixO(%rsp) movaps %xmm1,nb232_fiyO(%rsp) movaps %xmm2,nb232_fizO(%rsp) ## done with i O Now do i H1 & H2 simultaneously first get i particle coords: movaps nb232_jxO(%rsp),%xmm0 movaps nb232_jyO(%rsp),%xmm1 movaps nb232_jzO(%rsp),%xmm2 movaps %xmm0,%xmm3 movaps %xmm1,%xmm4 movaps %xmm2,%xmm5 subps nb232_ixH1(%rsp),%xmm0 subps nb232_iyH1(%rsp),%xmm1 subps nb232_izH1(%rsp),%xmm2 subps nb232_ixH2(%rsp),%xmm3 subps nb232_iyH2(%rsp),%xmm4 subps nb232_izH2(%rsp),%xmm5 movaps %xmm0,nb232_dxH1O(%rsp) movaps %xmm1,nb232_dyH1O(%rsp) movaps %xmm2,nb232_dzH1O(%rsp) movaps %xmm3,nb232_dxH2O(%rsp) movaps %xmm4,nb232_dyH2O(%rsp) movaps %xmm5,nb232_dzH2O(%rsp) 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 ## do invsqrt rsqrtps %xmm0,%xmm1 rsqrtps %xmm4,%xmm5 movaps %xmm1,%xmm2 movaps %xmm5,%xmm6 mulps %xmm1,%xmm1 mulps %xmm5,%xmm5 movaps nb232_three(%rsp),%xmm3 movaps %xmm3,%xmm7 mulps %xmm0,%xmm1 mulps %xmm4,%xmm5 subps %xmm1,%xmm3 subps %xmm5,%xmm7 mulps %xmm2,%xmm3 mulps %xmm6,%xmm7 mulps nb232_half(%rsp),%xmm3 ## rinv H1 - j water mulps nb232_half(%rsp),%xmm7 ## rinv H2 - j water mulps nb232_krf(%rsp),%xmm0 ## krsq mulps nb232_krf(%rsp),%xmm4 ## krsq ## assemble charges in xmm6 xorps %xmm6,%xmm6 movss nb232_qqOH(%rsp),%xmm6 movhps nb232_qqHH(%rsp),%xmm6 movaps %xmm0,%xmm1 movaps %xmm4,%xmm5 addps %xmm3,%xmm0 ## krsq+ rinv addps %xmm7,%xmm4 ## krsq+ rinv subps nb232_crf(%rsp),%xmm0 subps nb232_crf(%rsp),%xmm4 mulps nb232_two(%rsp),%xmm1 mulps nb232_two(%rsp),%xmm5 mulps %xmm6,%xmm0 ## vcoul mulps %xmm6,%xmm4 ## vcoul addps %xmm0,%xmm4 addps nb232_vctot(%rsp),%xmm4 movaps %xmm4,nb232_vctot(%rsp) movaps %xmm3,%xmm0 movaps %xmm7,%xmm4 mulps %xmm3,%xmm3 mulps %xmm7,%xmm7 subps %xmm1,%xmm0 subps %xmm5,%xmm4 mulps %xmm6,%xmm0 mulps %xmm6,%xmm4 mulps %xmm3,%xmm0 ## fscal mulps %xmm4,%xmm7 ## fscal movaps %xmm0,%xmm1 movaps %xmm0,%xmm2 mulps nb232_dxH1O(%rsp),%xmm0 mulps nb232_dyH1O(%rsp),%xmm1 mulps nb232_dzH1O(%rsp),%xmm2 ## update forces H1 - j water movaps nb232_fjxO(%rsp),%xmm3 movaps nb232_fjyO(%rsp),%xmm4 movaps nb232_fjzO(%rsp),%xmm5 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm5 movaps %xmm3,nb232_fjxO(%rsp) movaps %xmm4,nb232_fjyO(%rsp) movaps %xmm5,nb232_fjzO(%rsp) addps nb232_fixH1(%rsp),%xmm0 addps nb232_fiyH1(%rsp),%xmm1 addps nb232_fizH1(%rsp),%xmm2 movaps %xmm0,nb232_fixH1(%rsp) movaps %xmm1,nb232_fiyH1(%rsp) movaps %xmm2,nb232_fizH1(%rsp) ## do forces H2 - j water movaps %xmm7,%xmm0 movaps %xmm7,%xmm1 movaps %xmm7,%xmm2 mulps nb232_dxH2O(%rsp),%xmm0 mulps nb232_dyH2O(%rsp),%xmm1 mulps nb232_dzH2O(%rsp),%xmm2 movaps nb232_fjxO(%rsp),%xmm3 movaps nb232_fjyO(%rsp),%xmm4 movaps nb232_fjzO(%rsp),%xmm5 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm5 movq nb232_faction(%rbp),%rsi movaps %xmm3,nb232_fjxO(%rsp) movaps %xmm4,nb232_fjyO(%rsp) movaps %xmm5,nb232_fjzO(%rsp) addps nb232_fixH2(%rsp),%xmm0 addps nb232_fiyH2(%rsp),%xmm1 addps nb232_fizH2(%rsp),%xmm2 movaps %xmm0,nb232_fixH2(%rsp) movaps %xmm1,nb232_fiyH2(%rsp) movaps %xmm2,nb232_fizH2(%rsp) ## update j water forces from local variables movlps (%rsi,%rax,4),%xmm0 movlps 12(%rsi,%rax,4),%xmm1 movhps 24(%rsi,%rax,4),%xmm1 movaps nb232_fjxO(%rsp),%xmm3 movaps nb232_fjyO(%rsp),%xmm4 movaps nb232_fjzO(%rsp),%xmm5 movaps %xmm5,%xmm6 movaps %xmm5,%xmm7 shufps $2,%xmm6,%xmm6 ## constant 00000010 shufps $3,%xmm7,%xmm7 ## constant 00000011 addss 8(%rsi,%rax,4),%xmm5 addss 20(%rsi,%rax,4),%xmm6 addss 32(%rsi,%rax,4),%xmm7 movss %xmm5,8(%rsi,%rax,4) movss %xmm6,20(%rsi,%rax,4) movss %xmm7,32(%rsi,%rax,4) movaps %xmm3,%xmm5 unpcklps %xmm4,%xmm3 unpckhps %xmm4,%xmm5 addps %xmm3,%xmm0 addps %xmm5,%xmm1 movlps %xmm0,(%rsi,%rax,4) movlps %xmm1,12(%rsi,%rax,4) movhps %xmm1,24(%rsi,%rax,4) decl nb232_innerk(%rsp) jz _nb_kernel232_x86_64_sse.nb232_updateouterdata jmp _nb_kernel232_x86_64_sse.nb232_single_loop_nb_kernel232_x86_64_sse.nb232_updateouterdata: movl nb232_ii3(%rsp),%ecx movq nb232_faction(%rbp),%rdi movq nb232_fshift(%rbp),%rsi movl nb232_is3(%rsp),%edx ## accumulate Oi forces in xmm0, xmm1, xmm2 movaps nb232_fixO(%rsp),%xmm0 movaps nb232_fiyO(%rsp),%xmm1 movaps nb232_fizO(%rsp),%xmm2 movhlps %xmm0,%xmm3 movhlps %xmm1,%xmm4 movhlps %xmm2,%xmm5 addps %xmm3,%xmm0 addps %xmm4,%xmm1 addps %xmm5,%xmm2 ## sum is in 1/2 in xmm0-xmm2 movaps %xmm0,%xmm3 movaps %xmm1,%xmm4 movaps %xmm2,%xmm5 shufps $1,%xmm3,%xmm3 shufps $1,%xmm4,%xmm4 shufps $1,%xmm5,%xmm5 addss %xmm3,%xmm0 addss %xmm4,%xmm1 addss %xmm5,%xmm2 ## xmm0-xmm2 has single force in pos0 ## increment i force movss (%rdi,%rcx,4),%xmm3 movss 4(%rdi,%rcx,4),%xmm4 movss 8(%rdi,%rcx,4),%xmm5 subss %xmm0,%xmm3 subss %xmm1,%xmm4 subss %xmm2,%xmm5 movss %xmm3,(%rdi,%rcx,4) movss %xmm4,4(%rdi,%rcx,4) movss %xmm5,8(%rdi,%rcx,4) ## accumulate force in xmm6/xmm7 for fshift movaps %xmm0,%xmm6 movss %xmm2,%xmm7 movlhps %xmm1,%xmm6 shufps $8,%xmm6,%xmm6 ## constant 00001000 ## accumulate H1i forces in xmm0, xmm1, xmm2 movaps nb232_fixH1(%rsp),%xmm0 movaps nb232_fiyH1(%rsp),%xmm1 movaps nb232_fizH1(%rsp),%xmm2 movhlps %xmm0,%xmm3 movhlps %xmm1,%xmm4 movhlps %xmm2,%xmm5 addps %xmm3,%xmm0 addps %xmm4,%xmm1 addps %xmm5,%xmm2 ## sum is in 1/2 in xmm0-xmm2 movaps %xmm0,%xmm3 movaps %xmm1,%xmm4 movaps %xmm2,%xmm5 shufps $1,%xmm3,%xmm3 shufps $1,%xmm4,%xmm4 shufps $1,%xmm5,%xmm5 addss %xmm3,%xmm0 addss %xmm4,%xmm1 addss %xmm5,%xmm2 ## xmm0-xmm2 has single force in pos0 ## increment i force movss 12(%rdi,%rcx,4),%xmm3 movss 16(%rdi,%rcx,4),%xmm4 movss 20(%rdi,%rcx,4),%xmm5 subss %xmm0,%xmm3 subss %xmm1,%xmm4 subss %xmm2,%xmm5 movss %xmm3,12(%rdi,%rcx,4) movss %xmm4,16(%rdi,%rcx,4) movss %xmm5,20(%rdi,%rcx,4) ## accumulate force in xmm6/xmm7 for fshift addss %xmm2,%xmm7 movlhps %xmm1,%xmm0 shufps $8,%xmm0,%xmm0 ## constant 00001000 addps %xmm0,%xmm6 ## accumulate H2i forces in xmm0, xmm1, xmm2 movaps nb232_fixH2(%rsp),%xmm0 movaps nb232_fiyH2(%rsp),%xmm1 movaps nb232_fizH2(%rsp),%xmm2 movhlps %xmm0,%xmm3 movhlps %xmm1,%xmm4 movhlps %xmm2,%xmm5 addps %xmm3,%xmm0 addps %xmm4,%xmm1 addps %xmm5,%xmm2 ## sum is in 1/2 in xmm0-xmm2 movaps %xmm0,%xmm3 movaps %xmm1,%xmm4 movaps %xmm2,%xmm5 shufps $1,%xmm3,%xmm3 shufps $1,%xmm4,%xmm4 shufps $1,%xmm5,%xmm5 addss %xmm3,%xmm0 addss %xmm4,%xmm1 addss %xmm5,%xmm2 ## xmm0-xmm2 has single force in pos0 ## increment i force movss 24(%rdi,%rcx,4),%xmm3 movss 28(%rdi,%rcx,4),%xmm4 movss 32(%rdi,%rcx,4),%xmm5 subss %xmm0,%xmm3 subss %xmm1,%xmm4 subss %xmm2,%xmm5 movss %xmm3,24(%rdi,%rcx,4) movss %xmm4,28(%rdi,%rcx,4) movss %xmm5,32(%rdi,%rcx,4) ## accumulate force in xmm6/xmm7 for fshift addss %xmm2,%xmm7 movlhps %xmm1,%xmm0 shufps $8,%xmm0,%xmm0 ## constant 00001000 addps %xmm0,%xmm6 ## increment fshift force movlps (%rsi,%rdx,4),%xmm3 movss 8(%rsi,%rdx,4),%xmm4 subps %xmm6,%xmm3 subss %xmm7,%xmm4 movlps %xmm3,(%rsi,%rdx,4) movss %xmm4,8(%rsi,%rdx,4) ## get n from stack movl nb232_n(%rsp),%esi ## get group index for i particle movq nb232_gid(%rbp),%rdx ## base of gid[]⌨️ 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?