nb_kernel213_x86_64_sse2.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,121 行 · 第 1/5 页
S
2,121 行
## calculate krsq movsd nb213_krf(%rsp),%xmm0 movsd %xmm0,%xmm1 movsd %xmm0,%xmm2 mulsd %xmm4,%xmm0 mulsd %xmm5,%xmm1 mulsd %xmm6,%xmm2 movsd %xmm0,nb213_krsqM(%rsp) movsd %xmm1,nb213_krsqH2(%rsp) movsd %xmm2,nb213_krsqH1(%rsp) ## start with rsqH1 - put seed in xmm2 cvtsd2ss %xmm6,%xmm2 rsqrtss %xmm2,%xmm2 cvtss2sd %xmm2,%xmm2 movapd %xmm2,%xmm3 mulsd %xmm2,%xmm2 movapd nb213_three(%rsp),%xmm1 mulsd %xmm6,%xmm2 ## rsq*lu*lu subsd %xmm2,%xmm1 ## 30-rsq*lu*lu mulsd %xmm3,%xmm1 ## lu*(3-rsq*lu*lu) mulsd nb213_half(%rsp),%xmm1 ## iter1 ( new lu) movapd %xmm1,%xmm3 mulsd %xmm1,%xmm1 ## lu*lu mulsd %xmm1,%xmm6 ## rsq*lu*lu movapd nb213_three(%rsp),%xmm1 subsd %xmm6,%xmm1 ## 3-rsq*lu*lu mulsd %xmm3,%xmm1 ## lu*( 3-rsq*lu*lu) mulsd nb213_half(%rsp),%xmm1 ## rinv movapd %xmm1,nb213_rinvH1(%rsp) ## rsqH2 - seed in xmm2 cvtsd2ss %xmm5,%xmm2 rsqrtss %xmm2,%xmm2 cvtss2sd %xmm2,%xmm2 movapd %xmm2,%xmm3 mulsd %xmm2,%xmm2 movapd nb213_three(%rsp),%xmm1 mulsd %xmm5,%xmm2 ## rsq*lu*lu subsd %xmm2,%xmm1 ## 30-rsq*lu*lu mulsd %xmm3,%xmm1 ## lu*(3-rsq*lu*lu) mulsd nb213_half(%rsp),%xmm1 ## iter1 ( new lu) movapd %xmm1,%xmm3 mulsd %xmm1,%xmm1 ## lu*lu mulsd %xmm1,%xmm5 ## rsq*lu*lu movapd nb213_three(%rsp),%xmm1 subsd %xmm5,%xmm1 ## 3-rsq*lu*lu mulsd %xmm3,%xmm1 ## lu*( 3-rsq*lu*lu) mulsd nb213_half(%rsp),%xmm1 ## rinv movapd %xmm1,nb213_rinvH2(%rsp) ## rsqM - seed in xmm2 cvtsd2ss %xmm4,%xmm2 rsqrtss %xmm2,%xmm2 cvtss2sd %xmm2,%xmm2 movapd %xmm2,%xmm3 mulsd %xmm2,%xmm2 movapd nb213_three(%rsp),%xmm1 mulsd %xmm4,%xmm2 ## rsq*lu*lu subsd %xmm2,%xmm1 ## 30-rsq*lu*lu mulsd %xmm3,%xmm1 ## lu*(3-rsq*lu*lu) mulsd nb213_half(%rsp),%xmm1 ## iter1 ( new lu) movapd %xmm1,%xmm3 mulsd %xmm1,%xmm1 ## lu*lu mulsd %xmm1,%xmm4 ## rsq*lu*lu movapd nb213_three(%rsp),%xmm1 subsd %xmm4,%xmm1 ## 3-rsq*lu*lu mulsd %xmm3,%xmm1 ## lu*( 3-rsq*lu*lu) mulsd nb213_half(%rsp),%xmm1 ## rinv movapd %xmm1,nb213_rinvM(%rsp) ## do O interactions directly. xmm7=rsq cvtsd2ss %xmm7,%xmm2 movapd %xmm7,%xmm6 rcpps %xmm2,%xmm2 cvtss2sd %xmm2,%xmm2 movapd nb213_two(%rsp),%xmm1 movapd %xmm1,%xmm0 mulsd %xmm2,%xmm7 subsd %xmm7,%xmm1 mulsd %xmm1,%xmm2 ## iter1 mulsd %xmm2,%xmm6 subsd %xmm6,%xmm0 mulsd %xmm2,%xmm0 ## xmm0=rinvsq movapd %xmm0,%xmm1 mulsd %xmm1,%xmm1 ## rinv4 mulsd %xmm0,%xmm1 ##rinvsix movapd %xmm1,%xmm2 mulsd %xmm2,%xmm2 ## rinvtwelve mulsd nb213_c6(%rsp),%xmm1 mulsd nb213_c12(%rsp),%xmm2 movapd %xmm2,%xmm3 subsd %xmm1,%xmm3 ## Vvdw=Vvdw12-Vvdw6 addsd nb213_Vvdwtot(%rsp),%xmm3 mulsd nb213_six(%rsp),%xmm1 mulsd nb213_twelve(%rsp),%xmm2 subsd %xmm1,%xmm2 mulsd %xmm0,%xmm2 movapd %xmm2,%xmm4 ## total fsO movsd %xmm3,nb213_Vvdwtot(%rsp) movapd nb213_dxO(%rsp),%xmm0 movapd nb213_dyO(%rsp),%xmm1 movapd nb213_dzO(%rsp),%xmm2 mulsd %xmm4,%xmm0 mulsd %xmm4,%xmm1 mulsd %xmm4,%xmm2 ## update O forces movapd nb213_fixO(%rsp),%xmm3 movapd nb213_fiyO(%rsp),%xmm4 movapd nb213_fizO(%rsp),%xmm7 addsd %xmm0,%xmm3 addsd %xmm1,%xmm4 addsd %xmm2,%xmm7 movsd %xmm3,nb213_fixO(%rsp) movsd %xmm4,nb213_fiyO(%rsp) movsd %xmm7,nb213_fizO(%rsp) ## update j forces with water O movsd %xmm0,nb213_fjx(%rsp) movsd %xmm1,nb213_fjy(%rsp) movsd %xmm2,nb213_fjz(%rsp) ## H1 interactions movsd nb213_rinvH1(%rsp),%xmm6 movsd %xmm6,%xmm4 mulsd %xmm4,%xmm4 ## xmm6=rinv, xmm4=rinvsq movsd %xmm6,%xmm7 movsd nb213_krsqH1(%rsp),%xmm0 addsd %xmm0,%xmm6 ## xmm6=rinv+ krsq mulsd nb213_two(%rsp),%xmm0 subsd nb213_crf(%rsp),%xmm6 subsd %xmm0,%xmm7 ## xmm7=rinv-2*krsq mulsd nb213_qqH(%rsp),%xmm6 ## vcoul mulsd nb213_qqH(%rsp),%xmm7 mulsd %xmm7,%xmm4 ## total fsH1 in xmm4 addsd nb213_vctot(%rsp),%xmm6 movapd nb213_dxH1(%rsp),%xmm0 movapd nb213_dyH1(%rsp),%xmm1 movapd nb213_dzH1(%rsp),%xmm2 movsd %xmm6,nb213_vctot(%rsp) mulsd %xmm4,%xmm0 mulsd %xmm4,%xmm1 mulsd %xmm4,%xmm2 ## update H1 forces movapd nb213_fixH1(%rsp),%xmm3 movapd nb213_fiyH1(%rsp),%xmm4 movapd nb213_fizH1(%rsp),%xmm7 addsd %xmm0,%xmm3 addsd %xmm1,%xmm4 addsd %xmm2,%xmm7 movsd %xmm3,nb213_fixH1(%rsp) movsd %xmm4,nb213_fiyH1(%rsp) movsd %xmm7,nb213_fizH1(%rsp) ## update j forces with water H1 addsd nb213_fjx(%rsp),%xmm0 addsd nb213_fjy(%rsp),%xmm1 addsd nb213_fjz(%rsp),%xmm2 movsd %xmm0,nb213_fjx(%rsp) movsd %xmm1,nb213_fjy(%rsp) movsd %xmm2,nb213_fjz(%rsp) ## H2 interactions movsd nb213_rinvH2(%rsp),%xmm5 movsd %xmm5,%xmm4 mulsd %xmm4,%xmm4 ## xmm5=rinv, xmm4=rinvsq movsd %xmm5,%xmm7 movsd nb213_krsqH2(%rsp),%xmm0 addsd %xmm0,%xmm5 ## xmm5=rinv+ krsq mulsd nb213_two(%rsp),%xmm0 subsd nb213_crf(%rsp),%xmm5 subsd %xmm0,%xmm7 ## xmm7=rinv-2*krsq mulsd nb213_qqH(%rsp),%xmm5 ## vcoul mulsd nb213_qqH(%rsp),%xmm7 mulsd %xmm7,%xmm4 ## total fsH2 in xmm4 addsd nb213_vctot(%rsp),%xmm5 movapd nb213_dxH2(%rsp),%xmm0 movapd nb213_dyH2(%rsp),%xmm1 movapd nb213_dzH2(%rsp),%xmm2 movsd %xmm5,nb213_vctot(%rsp) mulsd %xmm4,%xmm0 mulsd %xmm4,%xmm1 mulsd %xmm4,%xmm2 ## update H2 forces movapd nb213_fixH2(%rsp),%xmm3 movapd nb213_fiyH2(%rsp),%xmm4 movapd nb213_fizH2(%rsp),%xmm7 addsd %xmm0,%xmm3 addsd %xmm1,%xmm4 addsd %xmm2,%xmm7 movsd %xmm3,nb213_fixH2(%rsp) movsd %xmm4,nb213_fiyH2(%rsp) movsd %xmm7,nb213_fizH2(%rsp) ## update j forces with water H2 addsd nb213_fjx(%rsp),%xmm0 addsd nb213_fjy(%rsp),%xmm1 addsd nb213_fjz(%rsp),%xmm2 movsd %xmm0,nb213_fjx(%rsp) movsd %xmm1,nb213_fjy(%rsp) movsd %xmm2,nb213_fjz(%rsp) ## M interactions movsd nb213_rinvM(%rsp),%xmm5 movsd %xmm5,%xmm4 mulsd %xmm4,%xmm4 ## xmm5=rinv, xmm4=rinvsq movsd %xmm5,%xmm7 movsd nb213_krsqM(%rsp),%xmm0 addsd %xmm0,%xmm5 ## xmm5=rinv+ krsq mulsd nb213_two(%rsp),%xmm0 subsd nb213_crf(%rsp),%xmm5 subsd %xmm0,%xmm7 ## xmm7=rinv-2*krsq mulsd nb213_qqM(%rsp),%xmm5 ## vcoul mulsd nb213_qqM(%rsp),%xmm7 mulsd %xmm7,%xmm4 ## total fsH2 in xmm4 addsd nb213_vctot(%rsp),%xmm5 movapd nb213_dxM(%rsp),%xmm0 movapd nb213_dyM(%rsp),%xmm1 movapd nb213_dzM(%rsp),%xmm2 movsd %xmm5,nb213_vctot(%rsp) mulsd %xmm4,%xmm0 mulsd %xmm4,%xmm1 mulsd %xmm4,%xmm2 ## update M forces movapd nb213_fixM(%rsp),%xmm3 movapd nb213_fiyM(%rsp),%xmm4 movapd nb213_fizM(%rsp),%xmm7 addsd %xmm0,%xmm3 addsd %xmm1,%xmm4 addsd %xmm2,%xmm7 movsd %xmm3,nb213_fixM(%rsp) movsd %xmm4,nb213_fiyM(%rsp) movsd %xmm7,nb213_fizM(%rsp) movq nb213_faction(%rbp),%rdi ## update j forces addsd nb213_fjx(%rsp),%xmm0 addsd nb213_fjy(%rsp),%xmm1 addsd nb213_fjz(%rsp),%xmm2 movlpd (%rdi,%rax,8),%xmm3 movlpd 8(%rdi,%rax,8),%xmm4 movlpd 16(%rdi,%rax,8),%xmm5 addsd %xmm0,%xmm3 addsd %xmm1,%xmm4 addsd %xmm2,%xmm5 movlpd %xmm3,(%rdi,%rax,8) movlpd %xmm4,8(%rdi,%rax,8) movlpd %xmm5,16(%rdi,%rax,8)_nb_kernel213_x86_64_sse2.nb213_updateouterdata: movl nb213_ii3(%rsp),%ecx movq nb213_faction(%rbp),%rdi movq nb213_fshift(%rbp),%rsi movl nb213_is3(%rsp),%edx ## accumulate Oi forces in xmm0, xmm1, xmm2 movapd nb213_fixO(%rsp),%xmm0 movapd nb213_fiyO(%rsp),%xmm1 movapd nb213_fizO(%rsp),%xmm2 movhlps %xmm0,%xmm3 movhlps %xmm1,%xmm4 movhlps %xmm2,%xmm5 addsd %xmm3,%xmm0 addsd %xmm4,%xmm1 addsd %xmm5,%xmm2 ## sum is in low xmm0-xmm2 ## increment i force movsd (%rdi,%rcx,8),%xmm3 movsd 8(%rdi,%rcx,8),%xmm4 movsd 16(%rdi,%rcx,8),%xmm5 subsd %xmm0,%xmm3 subsd %xmm1,%xmm4 subsd %xmm2,%xmm5 movsd %xmm3,(%rdi,%rcx,8) movsd %xmm4,8(%rdi,%rcx,8) movsd %xmm5,16(%rdi,%rcx,8) ## accumulate force in xmm6/xmm7 for fshift movapd %xmm0,%xmm6 movsd %xmm2,%xmm7 unpcklpd %xmm1,%xmm6 ## accumulate H1i forces in xmm0, xmm1, xmm2 movapd nb213_fixH1(%rsp),%xmm0 movapd nb213_fiyH1(%rsp),%xmm1 movapd nb213_fizH1(%rsp),%xmm2 movhlps %xmm0,%xmm3 movhlps %xmm1,%xmm4 movhlps %xmm2,%xmm5 addsd %xmm3,%xmm0 addsd %xmm4,%xmm1 addsd %xmm5,%xmm2 ## sum is in low xmm0-xmm2 ## increment i force movsd 24(%rdi,%rcx,8),%xmm3 movsd 32(%rdi,%rcx,8),%xmm4 movsd 40(%rdi,%rcx,8),%xmm5 subsd %xmm0,%xmm3 subsd %xmm1,%xmm4 subsd %xmm2,%xmm5 movsd %xmm3,24(%rdi,%rcx,8) movsd %xmm4,32(%rdi,%rcx,8) movsd %xmm5,40(%rdi,%rcx,8) ## accumulate force in xmm6/xmm7 for fshift addsd %xmm2,%xmm7 unpcklpd %xmm1,%xmm0 addpd %xmm0,%xmm6 ## accumulate H2i forces in xmm0, xmm1, xmm2 movapd nb213_fixH2(%rsp),%xmm0 movapd nb213_fiyH2(%rsp),%xmm1 movapd nb213_fizH2(%rsp),%xmm2 movhlps %xmm0,%xmm3 movhlps %xmm1,%xmm4 movhlps %xmm2,%xmm5 addsd %xmm3,%xmm0 addsd %xmm4,%xmm1 addsd %xmm5,%xmm2 ## sum is in low xmm0-xmm2 ## increment i force movsd 48(%rdi,%rcx,8),%xmm3 movsd 56(%rdi,%rcx,8),%xmm4 movsd 64(%rdi,%rcx,8),%xmm5 subsd %xmm0,%xmm3 subsd %xmm1,%xmm4 subsd %xmm2,%xmm5 movsd %xmm3,48(%rdi,%rcx,8) movsd %xmm4,56(%rdi,%rcx,8) movsd %xmm5,64(%rdi,%rcx,8) ## accumulate force in xmm6/xmm7 for fshift addsd %xmm2,%xmm7 unpcklpd %xmm1,%xmm0 addpd %xmm0,%xmm6 ## accumulate Mi forces in xmm0, xmm1, xmm2 movapd nb213_fixM(%rsp),%xmm0 movapd nb213_fiyM(%rsp),%xmm1 movapd nb213_fizM(%rsp),%xmm2 movhlps %xmm0,%xmm3 movhlps %xmm1,%xmm4 movhlps %xmm2,%xmm5 addsd %xmm3,%xmm0 addsd %xmm4,%xmm1 addsd %xmm5,%xmm2 ## sum is in low xmm0-xmm2 ## increment i force movsd 72(%rdi,%rcx,8),%xmm3 movsd 80(%rdi,%rcx,8),%xmm4 movsd 88(%rdi,%rcx,8),%xmm5 subsd %xmm0,%xmm3 subsd %xmm1,%xmm4 subsd %xmm2,%xmm5 movsd %xmm3,72(%rdi,%rcx,8) movsd %xmm4,80(%rdi,%rcx,8) movsd %xmm5,88(%rdi,%rcx,8) ## accumulate force in xmm6/xmm7 for fshift addsd %xmm2,%xmm7 unpcklpd %xmm1,%xmm0 addpd %xmm0,%xmm6 ## increment fshift force movlpd (%rsi,%rdx,8),%xmm3 movhpd 8(%rsi,%rdx,8),%xmm3 movsd 16(%rsi,%rdx,8),%xmm4 subpd %xmm6,%xmm3 subsd %xmm7,%xmm4 movlpd %xmm3,(%rsi,%rdx,8) movhpd %xmm3,8(%rsi,%rdx,8) movsd %xmm4,16(%rsi,%rdx,8) ## get n from stack movl nb213_n(%rsp),%esi ## get group index for i particle movq nb213_gid(%rbp),%rdx ## base of gid[] movl (%rdx,%rsi,4),%edx ## ggid=gid[n] ## accumulate total potential energy and update it movapd nb213_vctot(%rsp),%xmm7 ## accumulate movhlps %xmm7,%xmm6 addsd %xmm6,%xmm7 ## low xmm7 has the sum now ## add earlier value from mem movq nb213_Vc(%rbp),%rax addsd (%rax,%rdx,8),%xmm7 ## move back to mem movsd %xmm7,(%rax,%rdx,8) ## accumulate total lj energy and update it movapd nb213_Vvdwtot(%rsp),%xmm7 ## accumulate movhlps %xmm7,%xmm6 addsd %xmm6,%xmm7 ## low xmm7 has the sum now ## add earlier value from mem movq nb213_Vvdw(%rbp),%rax addsd (%rax,%rdx,8),%xmm7 ## move back to mem movsd %xmm7,(%rax,%rdx,8) ## finish if last movl nb213_nn1(%rsp),%ecx ## esi already loaded with n incl %esi⌨️ 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?