nb_kernel113_x86_64_sse2.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,022 行 · 第 1/5 页
S
2,022 行
subsd %xmm2,%xmm1 ## 30-rsq*lu*lu mulsd %xmm3,%xmm1 ## lu*(3-rsq*lu*lu) mulsd nb113_half(%rsp),%xmm1 ## iter1 ( new lu) movapd %xmm1,%xmm3 mulsd %xmm1,%xmm1 ## lu*lu mulsd %xmm1,%xmm6 ## rsq*lu*lu movapd nb113_three(%rsp),%xmm1 subsd %xmm6,%xmm1 ## 3-rsq*lu*lu mulsd %xmm3,%xmm1 ## lu*( 3-rsq*lu*lu) mulsd nb113_half(%rsp),%xmm1 ## rinv movapd %xmm1,nb113_rinvH1(%rsp) ## rsqH2 - seed in xmm2 cvtsd2ss %xmm5,%xmm2 rsqrtss %xmm2,%xmm2 cvtss2sd %xmm2,%xmm2 movapd %xmm2,%xmm3 mulsd %xmm2,%xmm2 movapd nb113_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 nb113_half(%rsp),%xmm1 ## iter1 ( new lu) movapd %xmm1,%xmm3 mulsd %xmm1,%xmm1 ## lu*lu mulsd %xmm1,%xmm5 ## rsq*lu*lu movapd nb113_three(%rsp),%xmm1 subsd %xmm5,%xmm1 ## 3-rsq*lu*lu mulsd %xmm3,%xmm1 ## lu*( 3-rsq*lu*lu) mulsd nb113_half(%rsp),%xmm1 ## rinv movapd %xmm1,nb113_rinvH2(%rsp) ## rsqM - seed in xmm2 cvtsd2ss %xmm4,%xmm2 rsqrtss %xmm2,%xmm2 cvtss2sd %xmm2,%xmm2 movapd %xmm2,%xmm3 mulsd %xmm2,%xmm2 movapd nb113_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 nb113_half(%rsp),%xmm1 ## iter1 ( new lu) movapd %xmm1,%xmm3 mulsd %xmm1,%xmm1 ## lu*lu mulsd %xmm1,%xmm4 ## rsq*lu*lu movapd nb113_three(%rsp),%xmm1 subsd %xmm4,%xmm1 ## 3-rsq*lu*lu mulsd %xmm3,%xmm1 ## lu*( 3-rsq*lu*lu) mulsd nb113_half(%rsp),%xmm1 ## rinv movapd %xmm1,nb113_rinvM(%rsp) ## do O interactions directly. xmm7=rsq cvtsd2ss %xmm7,%xmm2 movapd %xmm7,%xmm6 rcpps %xmm2,%xmm2 cvtss2sd %xmm2,%xmm2 movapd nb113_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 nb113_c6(%rsp),%xmm1 mulsd nb113_c12(%rsp),%xmm2 movapd %xmm2,%xmm3 subsd %xmm1,%xmm3 ## Vvdw=Vvdw12-Vvdw6 addsd nb113_Vvdwtot(%rsp),%xmm3 mulsd nb113_six(%rsp),%xmm1 mulsd nb113_twelve(%rsp),%xmm2 subsd %xmm1,%xmm2 mulsd %xmm0,%xmm2 movapd %xmm2,%xmm4 ## total fsO movsd %xmm3,nb113_Vvdwtot(%rsp) movapd nb113_dxO(%rsp),%xmm0 movapd nb113_dyO(%rsp),%xmm1 movapd nb113_dzO(%rsp),%xmm2 mulsd %xmm4,%xmm0 mulsd %xmm4,%xmm1 mulsd %xmm4,%xmm2 ## update O forces movapd nb113_fixO(%rsp),%xmm3 movapd nb113_fiyO(%rsp),%xmm4 movapd nb113_fizO(%rsp),%xmm7 addsd %xmm0,%xmm3 addsd %xmm1,%xmm4 addsd %xmm2,%xmm7 movsd %xmm3,nb113_fixO(%rsp) movsd %xmm4,nb113_fiyO(%rsp) movsd %xmm7,nb113_fizO(%rsp) ## update j forces with water O movsd %xmm0,nb113_fjx(%rsp) movsd %xmm1,nb113_fjy(%rsp) movsd %xmm2,nb113_fjz(%rsp) ## H1 interactions movapd nb113_rinvH1(%rsp),%xmm6 movapd %xmm6,%xmm4 mulsd %xmm4,%xmm4 ## xmm6=rinv, xmm4=rinvsq mulsd nb113_qqH(%rsp),%xmm6 ## xmm6=vcoul mulsd %xmm6,%xmm4 ## total fsH1 in xmm4 addsd nb113_vctot(%rsp),%xmm6 movapd nb113_dxH1(%rsp),%xmm0 movapd nb113_dyH1(%rsp),%xmm1 movapd nb113_dzH1(%rsp),%xmm2 movsd %xmm6,nb113_vctot(%rsp) mulsd %xmm4,%xmm0 mulsd %xmm4,%xmm1 mulsd %xmm4,%xmm2 ## update H1 forces movapd nb113_fixH1(%rsp),%xmm3 movapd nb113_fiyH1(%rsp),%xmm4 movapd nb113_fizH1(%rsp),%xmm7 addsd %xmm0,%xmm3 addsd %xmm1,%xmm4 addsd %xmm2,%xmm7 movsd %xmm3,nb113_fixH1(%rsp) movsd %xmm4,nb113_fiyH1(%rsp) movsd %xmm7,nb113_fizH1(%rsp) ## update j forces with water H1 addsd nb113_fjx(%rsp),%xmm0 addsd nb113_fjy(%rsp),%xmm1 addsd nb113_fjz(%rsp),%xmm2 movsd %xmm0,nb113_fjx(%rsp) movsd %xmm1,nb113_fjy(%rsp) movsd %xmm2,nb113_fjz(%rsp) ## H2 interactions movapd nb113_rinvH2(%rsp),%xmm5 movapd %xmm5,%xmm4 mulsd %xmm4,%xmm4 ## xmm5=rinv, xmm4=rinvsq mulsd nb113_qqH(%rsp),%xmm5 ## xmm5=vcoul mulsd %xmm5,%xmm4 ## total fsH1 in xmm4 addsd nb113_vctot(%rsp),%xmm5 movapd nb113_dxH2(%rsp),%xmm0 movapd nb113_dyH2(%rsp),%xmm1 movapd nb113_dzH2(%rsp),%xmm2 movsd %xmm5,nb113_vctot(%rsp) mulsd %xmm4,%xmm0 mulsd %xmm4,%xmm1 mulsd %xmm4,%xmm2 ## update H2 forces movapd nb113_fixH2(%rsp),%xmm3 movapd nb113_fiyH2(%rsp),%xmm4 movapd nb113_fizH2(%rsp),%xmm7 addsd %xmm0,%xmm3 addsd %xmm1,%xmm4 addsd %xmm2,%xmm7 movsd %xmm3,nb113_fixH2(%rsp) movsd %xmm4,nb113_fiyH2(%rsp) movsd %xmm7,nb113_fizH2(%rsp) ## update j forces with water H2 addsd nb113_fjx(%rsp),%xmm0 addsd nb113_fjy(%rsp),%xmm1 addsd nb113_fjz(%rsp),%xmm2 movsd %xmm0,nb113_fjx(%rsp) movsd %xmm1,nb113_fjy(%rsp) movsd %xmm2,nb113_fjz(%rsp) ## M interactions movapd nb113_rinvM(%rsp),%xmm5 movapd %xmm5,%xmm4 mulsd %xmm4,%xmm4 ## xmm5=rinv, xmm4=rinvsq mulsd nb113_qqM(%rsp),%xmm5 ## xmm5=vcoul mulsd %xmm5,%xmm4 ## total fsH1 in xmm4 addsd nb113_vctot(%rsp),%xmm5 movapd nb113_dxM(%rsp),%xmm0 movapd nb113_dyM(%rsp),%xmm1 movapd nb113_dzM(%rsp),%xmm2 movsd %xmm5,nb113_vctot(%rsp) mulsd %xmm4,%xmm0 mulsd %xmm4,%xmm1 mulsd %xmm4,%xmm2 ## update M forces movapd nb113_fixM(%rsp),%xmm3 movapd nb113_fiyM(%rsp),%xmm4 movapd nb113_fizM(%rsp),%xmm7 addsd %xmm0,%xmm3 addsd %xmm1,%xmm4 addsd %xmm2,%xmm7 movsd %xmm3,nb113_fixM(%rsp) movsd %xmm4,nb113_fiyM(%rsp) movsd %xmm7,nb113_fizM(%rsp) movq nb113_faction(%rbp),%rdi ## update j forces addsd nb113_fjx(%rsp),%xmm0 addsd nb113_fjy(%rsp),%xmm1 addsd nb113_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_kernel113_x86_64_sse2.nb113_updateouterdata: movl nb113_ii3(%rsp),%ecx movq nb113_faction(%rbp),%rdi movq nb113_fshift(%rbp),%rsi movl nb113_is3(%rsp),%edx ## accumulate Oi forces in xmm0, xmm1, xmm2 movapd nb113_fixO(%rsp),%xmm0 movapd nb113_fiyO(%rsp),%xmm1 movapd nb113_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 nb113_fixH1(%rsp),%xmm0 movapd nb113_fiyH1(%rsp),%xmm1 movapd nb113_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 nb113_fixH2(%rsp),%xmm0 movapd nb113_fiyH2(%rsp),%xmm1 movapd nb113_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 nb113_fixM(%rsp),%xmm0 movapd nb113_fiyM(%rsp),%xmm1 movapd nb113_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 nb113_n(%rsp),%esi ## get group index for i particle movq nb113_gid(%rbp),%rdx ## base of gid[] movl (%rdx,%rsi,4),%edx ## ggid=gid[n] ## accumulate total potential energy and update it movapd nb113_vctot(%rsp),%xmm7 ## accumulate movhlps %xmm7,%xmm6 addsd %xmm6,%xmm7 ## low xmm7 has the sum now ## add earlier value from mem movq nb113_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 nb113_Vvdwtot(%rsp),%xmm7 ## accumulate movhlps %xmm7,%xmm6 addsd %xmm6,%xmm7 ## low xmm7 has the sum now ## add earlier value from mem movq nb113_Vvdw(%rbp),%rax addsd (%rax,%rdx,8),%xmm7 ## move back to mem movsd %xmm7,(%rax,%rdx,8) ## finish if last movl nb113_nn1(%rsp),%ecx ## esi already loaded with n incl %esi subl %esi,%ecx jz _nb_kernel113_x86_64_sse2.nb113_outerend ## not last, iterate outer loop once more! movl %esi,nb113_n(%rsp) jmp _nb_kernel113_x86_64_sse2.nb113_outer_nb_kernel113_x86_64_sse2.nb113_outerend: ## check if more outer neighborlists remain movl nb113_nri(%rsp),%ecx ## esi already loaded with n above subl %esi,%ecx jz _nb_kernel113_x86_64_sse2.nb113_end ## non-zero, do one more workunit jmp _nb_kernel113_x86_64_sse2.nb113_threadloop_nb_kernel113_x86_64_sse2.nb113_end: movl nb113_nouter(%rsp),%eax movl nb113_ninner(%rsp),%ebx movq nb113_outeriter(%rbp),%rcx movq nb113_inneriter(%rbp),%rdx movl %eax,(%rcx) movl %ebx,(%rdx)⌨️ 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?