nb_kernel114_x86_64_sse.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,123 行 · 第 1/5 页
S
2,123 行
mulss nb114_c6(%rsp),%xmm1 mulss nb114_c12(%rsp),%xmm2 movss %xmm2,%xmm3 subss %xmm1,%xmm3 addss nb114_Vvdwtot(%rsp),%xmm3 movss %xmm3,nb114_Vvdwtot(%rsp) mulss nb114_six(%rsp),%xmm1 mulss nb114_twelve(%rsp),%xmm2 subss %xmm1,%xmm2 mulss %xmm2,%xmm0 ## fscal movss %xmm0,%xmm1 movss %xmm0,%xmm2 mulss nb114_dxOO(%rsp),%xmm0 mulss nb114_dyOO(%rsp),%xmm1 mulss nb114_dzOO(%rsp),%xmm2 xorps %xmm3,%xmm3 xorps %xmm4,%xmm4 xorps %xmm5,%xmm5 addss %xmm0,%xmm3 addss %xmm1,%xmm4 addss %xmm2,%xmm5 movaps %xmm3,nb114_fjxO(%rsp) movaps %xmm4,nb114_fjyO(%rsp) movaps %xmm5,nb114_fjzO(%rsp) addss nb114_fixO(%rsp),%xmm0 addss nb114_fiyO(%rsp),%xmm1 addss nb114_fizO(%rsp),%xmm2 movss %xmm0,nb114_fixO(%rsp) movss %xmm1,nb114_fiyO(%rsp) movss %xmm2,nb114_fizO(%rsp) ## do M coulomb interaction movaps nb114_rinvMM(%rsp),%xmm0 movaps %xmm0,%xmm4 ## xmm4=rinv mulps %xmm0,%xmm0 ## xmm0=rinvsq ## fetch charges to xmm3 (temporary) xorps %xmm3,%xmm3 movss nb114_qqMH(%rsp),%xmm3 movhps nb114_qqMM(%rsp),%xmm3 shufps $193,%xmm3,%xmm3 ## 11000001 mulps %xmm3,%xmm4 ## xmm4=voul mulps %xmm4,%xmm0 addps nb114_vctot(%rsp),%xmm4 movaps %xmm4,nb114_vctot(%rsp) movaps %xmm0,%xmm1 movaps %xmm0,%xmm2 mulps nb114_dxMM(%rsp),%xmm0 mulps nb114_dyMM(%rsp),%xmm1 mulps nb114_dzMM(%rsp),%xmm2 ## update forces M - j water movaps nb114_fjxO(%rsp),%xmm3 movaps nb114_fjyO(%rsp),%xmm4 movaps nb114_fjzO(%rsp),%xmm5 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm5 movaps %xmm3,nb114_fjxO(%rsp) movaps %xmm4,nb114_fjyO(%rsp) movaps %xmm5,nb114_fjzO(%rsp) addps nb114_fixM(%rsp),%xmm0 addps nb114_fiyM(%rsp),%xmm1 addps nb114_fizM(%rsp),%xmm2 movaps %xmm0,nb114_fixM(%rsp) movaps %xmm1,nb114_fiyM(%rsp) movaps %xmm2,nb114_fizM(%rsp) ## i H1 & H2 simultaneously first get i particle coords: movaps nb114_jxO(%rsp),%xmm0 movaps nb114_jyO(%rsp),%xmm1 movaps nb114_jzO(%rsp),%xmm2 movaps %xmm0,%xmm3 movaps %xmm1,%xmm4 movaps %xmm2,%xmm5 subps nb114_ixH1(%rsp),%xmm0 subps nb114_iyH1(%rsp),%xmm1 subps nb114_izH1(%rsp),%xmm2 subps nb114_ixH2(%rsp),%xmm3 subps nb114_iyH2(%rsp),%xmm4 subps nb114_izH2(%rsp),%xmm5 movaps %xmm0,nb114_dxH1H1(%rsp) movaps %xmm1,nb114_dyH1H1(%rsp) movaps %xmm2,nb114_dzH1H1(%rsp) movaps %xmm3,nb114_dxH2H2(%rsp) movaps %xmm4,nb114_dyH2H2(%rsp) movaps %xmm5,nb114_dzH2H2(%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 movaps %xmm0,nb114_rsqH1H1(%rsp) movaps %xmm4,nb114_rsqH2H2(%rsp) ## start doing invsqrt use rsq values in xmm0, xmm4 rsqrtps %xmm0,%xmm1 rsqrtps %xmm4,%xmm5 movaps %xmm1,%xmm2 movaps %xmm5,%xmm6 mulps %xmm1,%xmm1 mulps %xmm5,%xmm5 movaps nb114_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 nb114_half(%rsp),%xmm3 ## rinvH1H1 mulps nb114_half(%rsp),%xmm7 ## rinvH2H2 movaps %xmm3,nb114_rinvH1H1(%rsp) movaps %xmm7,nb114_rinvH2H2(%rsp) ## Do H1 coulomb interaction movaps nb114_rinvH1H1(%rsp),%xmm0 movaps %xmm0,%xmm4 ## xmm4=rinv mulps %xmm0,%xmm0 ## xmm0=rinvsq ## fetch charges to xmm3 (temporary) xorps %xmm3,%xmm3 movss nb114_qqHH(%rsp),%xmm3 movhps nb114_qqMH(%rsp),%xmm3 shufps $193,%xmm3,%xmm3 ## 11000001 mulps %xmm3,%xmm4 ## xmm4=voul mulps %xmm4,%xmm0 addps nb114_vctot(%rsp),%xmm4 movaps %xmm4,nb114_vctot(%rsp) movaps %xmm0,%xmm1 movaps %xmm0,%xmm2 mulps nb114_dxH1H1(%rsp),%xmm0 mulps nb114_dyH1H1(%rsp),%xmm1 mulps nb114_dzH1H1(%rsp),%xmm2 ## update forces H1 - j water movaps nb114_fjxO(%rsp),%xmm3 movaps nb114_fjyO(%rsp),%xmm4 movaps nb114_fjzO(%rsp),%xmm5 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm5 movaps %xmm3,nb114_fjxO(%rsp) movaps %xmm4,nb114_fjyO(%rsp) movaps %xmm5,nb114_fjzO(%rsp) addps nb114_fixH1(%rsp),%xmm0 addps nb114_fiyH1(%rsp),%xmm1 addps nb114_fizH1(%rsp),%xmm2 movaps %xmm0,nb114_fixH1(%rsp) movaps %xmm1,nb114_fiyH1(%rsp) movaps %xmm2,nb114_fizH1(%rsp) ## H2 Coulomb movaps nb114_rinvH2H2(%rsp),%xmm0 movaps %xmm0,%xmm4 ## xmm4=rinv mulps %xmm0,%xmm0 ## xmm0=rinvsq ## fetch charges to xmm3 (temporary) xorps %xmm3,%xmm3 movss nb114_qqHH(%rsp),%xmm3 movhps nb114_qqMH(%rsp),%xmm3 shufps $193,%xmm3,%xmm3 ## 11000001 mulps %xmm3,%xmm4 ## xmm4=voul mulps %xmm4,%xmm0 addps nb114_vctot(%rsp),%xmm4 ## local vctot summation variable movaps %xmm4,nb114_vctot(%rsp) movaps %xmm0,%xmm1 movaps %xmm0,%xmm2 mulps nb114_dxH2H2(%rsp),%xmm0 mulps nb114_dyH2H2(%rsp),%xmm1 mulps nb114_dzH2H2(%rsp),%xmm2 ## update forces H2 - j water movaps nb114_fjxO(%rsp),%xmm3 movaps nb114_fjyO(%rsp),%xmm4 movaps nb114_fjzO(%rsp),%xmm5 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm5 movaps %xmm3,nb114_fjxO(%rsp) movaps %xmm4,nb114_fjyO(%rsp) movaps %xmm5,nb114_fjzO(%rsp) addps nb114_fixH2(%rsp),%xmm0 addps nb114_fiyH2(%rsp),%xmm1 addps nb114_fizH2(%rsp),%xmm2 movaps %xmm0,nb114_fixH2(%rsp) movaps %xmm1,nb114_fiyH2(%rsp) movaps %xmm2,nb114_fizH2(%rsp) movq nb114_faction(%rbp),%rsi ## update j water forces from local variables. ## transpose back first movaps nb114_fjxO(%rsp),%xmm0 ## Ox H1x H2x Mx movaps nb114_fjyO(%rsp),%xmm1 ## Oy H1y H2y My movaps nb114_fjzO(%rsp),%xmm2 ## Oz H1z H2z Mz movaps %xmm0,%xmm3 movaps %xmm0,%xmm4 unpcklps %xmm1,%xmm3 ## Ox Oy - - shufps $0x1,%xmm2,%xmm4 ## h1x - Oz - movaps %xmm1,%xmm5 movaps %xmm0,%xmm6 unpcklps %xmm2,%xmm5 ## - - H1y H1z unpckhps %xmm1,%xmm6 ## h2x h2y - - unpckhps %xmm2,%xmm1 ## - - My Mz shufps $0x32,%xmm0,%xmm2 ## (00110010) h2z - Mx - shufps $36,%xmm4,%xmm3 ## 00100100 ;# Ox Oy Oz H1x shufps $78,%xmm6,%xmm5 ## 01001110 ;# h1y h1z h2x h2y shufps $232,%xmm1,%xmm2 ## 11101000 ;# h2z mx my mz movlps (%rsi,%rax,4),%xmm0 movlps 16(%rsi,%rax,4),%xmm1 movlps 32(%rsi,%rax,4),%xmm4 movhps 8(%rsi,%rax,4),%xmm0 movhps 24(%rsi,%rax,4),%xmm1 movhps 40(%rsi,%rax,4),%xmm4 addps %xmm3,%xmm0 addps %xmm5,%xmm1 addps %xmm2,%xmm4 movlps %xmm0,(%rsi,%rax,4) movlps %xmm1,16(%rsi,%rax,4) movlps %xmm4,32(%rsi,%rax,4) movhps %xmm0,8(%rsi,%rax,4) movhps %xmm1,24(%rsi,%rax,4) movhps %xmm4,40(%rsi,%rax,4) decl nb114_innerk(%rsp) jz _nb_kernel114_x86_64_sse.nb114_updateouterdata jmp _nb_kernel114_x86_64_sse.nb114_single_loop_nb_kernel114_x86_64_sse.nb114_updateouterdata: movl nb114_ii3(%rsp),%ecx movq nb114_faction(%rbp),%rdi movq nb114_fshift(%rbp),%rsi movl nb114_is3(%rsp),%edx ## accumulate Oi forces in xmm0, xmm1, xmm2 movaps nb114_fixO(%rsp),%xmm0 movaps nb114_fiyO(%rsp),%xmm1 movaps nb114_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 ## 00001000 ## accumulate H1i forces in xmm0, xmm1, xmm2 movaps nb114_fixH1(%rsp),%xmm0 movaps nb114_fiyH1(%rsp),%xmm1 movaps nb114_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 ## 00001000 addps %xmm0,%xmm6 ## accumulate H2i forces in xmm0, xmm1, xmm2 movaps nb114_fixH2(%rsp),%xmm0 movaps nb114_fiyH2(%rsp),%xmm1 movaps nb114_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 ## 00001000 addps %xmm0,%xmm6 ## accumulate Mi forces in xmm0, xmm1, xmm2 movaps nb114_fixM(%rsp),%xmm0 movaps nb114_fiyM(%rsp),%xmm1 movaps nb114_fizM(%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 36(%rdi,%rcx,4),%xmm3 movss 40(%rdi,%rcx,4),%xmm4 movss 44(%rdi,%rcx,4),%xmm5 subss %xmm0,%xmm3 subss %xmm1,%xmm4 subss %xmm2,%xmm5 movss %xmm3,36(%rdi,%rcx,4) movss %xmm4,40(%rdi,%rcx,4) movss %xmm5,44(%rdi,%rcx,4) ## accumulate force in xmm6/xmm7 for fshift addss %xmm2,%xmm7 movlhps %xmm1,%xmm0 shufps $8,%xmm0,%xmm0 ## 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 nb114_n(%rsp),%esi ## get group index for i particle movq nb114_gid(%rbp),%rdx ## base of gid[] movl (%rdx,%rsi,4),%edx ## ggid=gid[n] ## accumulate total potential energy and update it movaps nb114_vctot(%rsp),%xmm7 ## accumulate ⌨️ 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?