nb_kernel102_x86_64_sse2.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,225 行 · 第 1/5 页
S
2,225 行
subpd %xmm5,%xmm4 subpd %xmm8,%xmm7 ## 3-rsq*lu*lu mulpd %xmm1,%xmm9 mulpd %xmm4,%xmm10 mulpd %xmm7,%xmm11 ## lu*(3-rsq*lu*lu) movapd nb102_half(%rsp),%xmm15 mulpd %xmm15,%xmm9 ## rinvOH2 mulpd %xmm15,%xmm10 ## rinvH1H2 mulpd %xmm15,%xmm11 ## rinvH2H2 ## H2 interactions movapd %xmm9,%xmm0 movapd %xmm10,%xmm1 movapd %xmm11,%xmm2 mulpd %xmm9,%xmm9 mulpd %xmm10,%xmm10 mulpd %xmm11,%xmm11 mulpd nb102_qqOH(%rsp),%xmm0 mulpd nb102_qqHH(%rsp),%xmm1 mulpd nb102_qqHH(%rsp),%xmm2 mulpd %xmm0,%xmm9 mulpd %xmm1,%xmm10 mulpd %xmm2,%xmm11 addpd nb102_vctot(%rsp),%xmm0 addpd %xmm2,%xmm1 addpd %xmm1,%xmm0 movapd %xmm0,nb102_vctot(%rsp) ## move j H2 forces to xmm0-xmm2 movq nb102_faction(%rbp),%rdi movlpd 48(%rdi,%rax,8),%xmm0 movlpd 56(%rdi,%rax,8),%xmm1 movlpd 64(%rdi,%rax,8),%xmm2 movhpd 48(%rdi,%rbx,8),%xmm0 movhpd 56(%rdi,%rbx,8),%xmm1 movhpd 64(%rdi,%rbx,8),%xmm2 movapd %xmm9,%xmm7 movapd %xmm9,%xmm8 movapd %xmm11,%xmm13 movapd %xmm11,%xmm14 movapd %xmm11,%xmm15 movapd %xmm10,%xmm11 movapd %xmm10,%xmm12 mulpd nb102_dxOH2(%rsp),%xmm7 mulpd nb102_dyOH2(%rsp),%xmm8 mulpd nb102_dzOH2(%rsp),%xmm9 mulpd nb102_dxH1H2(%rsp),%xmm10 mulpd nb102_dyH1H2(%rsp),%xmm11 mulpd nb102_dzH1H2(%rsp),%xmm12 mulpd nb102_dxH2H2(%rsp),%xmm13 mulpd nb102_dyH2H2(%rsp),%xmm14 mulpd nb102_dzH2H2(%rsp),%xmm15 addpd %xmm7,%xmm0 addpd %xmm8,%xmm1 addpd %xmm9,%xmm2 addpd nb102_fixO(%rsp),%xmm7 addpd nb102_fiyO(%rsp),%xmm8 addpd nb102_fizO(%rsp),%xmm9 addpd %xmm10,%xmm0 addpd %xmm11,%xmm1 addpd %xmm12,%xmm2 addpd nb102_fixH1(%rsp),%xmm10 addpd nb102_fiyH1(%rsp),%xmm11 addpd nb102_fizH1(%rsp),%xmm12 addpd %xmm13,%xmm0 addpd %xmm14,%xmm1 addpd %xmm15,%xmm2 addpd nb102_fixH2(%rsp),%xmm13 addpd nb102_fiyH2(%rsp),%xmm14 addpd nb102_fizH2(%rsp),%xmm15 movapd %xmm7,nb102_fixO(%rsp) movapd %xmm8,nb102_fiyO(%rsp) movapd %xmm9,nb102_fizO(%rsp) movapd %xmm10,nb102_fixH1(%rsp) movapd %xmm11,nb102_fiyH1(%rsp) movapd %xmm12,nb102_fizH1(%rsp) movapd %xmm13,nb102_fixH2(%rsp) movapd %xmm14,nb102_fiyH2(%rsp) movapd %xmm15,nb102_fizH2(%rsp) ## store back j H2 forces from xmm0-xmm2 movlpd %xmm0,48(%rdi,%rax,8) movlpd %xmm1,56(%rdi,%rax,8) movlpd %xmm2,64(%rdi,%rax,8) movhpd %xmm0,48(%rdi,%rbx,8) movhpd %xmm1,56(%rdi,%rbx,8) movhpd %xmm2,64(%rdi,%rbx,8) ## should we do one more iteration? subl $2,nb102_innerk(%rsp) jl _nb_kernel102_x86_64_sse2.nb102_checksingle jmp _nb_kernel102_x86_64_sse2.nb102_unroll_loop_nb_kernel102_x86_64_sse2.nb102_checksingle: movl nb102_innerk(%rsp),%edx andl $1,%edx jnz _nb_kernel102_x86_64_sse2.nb102_dosingle jmp _nb_kernel102_x86_64_sse2.nb102_updateouterdata_nb_kernel102_x86_64_sse2.nb102_dosingle: movq nb102_innerjjnr(%rsp),%rdx ## pointer to jjnr[k] movl (%rdx),%eax movq nb102_pos(%rbp),%rsi lea (%rax,%rax,2),%rax ## move j O coordinates to local temp variables movsd (%rsi,%rax,8),%xmm0 movsd 8(%rsi,%rax,8),%xmm1 movsd 16(%rsi,%rax,8),%xmm2 ## xmm0 = Ox ## xmm1 = Oy ## xmm2 = Oz movsd %xmm0,%xmm3 movsd %xmm1,%xmm4 movsd %xmm2,%xmm5 movsd %xmm0,%xmm6 movsd %xmm1,%xmm7 movsd %xmm2,%xmm8 subsd nb102_ixO(%rsp),%xmm0 subsd nb102_iyO(%rsp),%xmm1 subsd nb102_izO(%rsp),%xmm2 subsd nb102_ixH1(%rsp),%xmm3 subsd nb102_iyH1(%rsp),%xmm4 subsd nb102_izH1(%rsp),%xmm5 subsd nb102_ixH2(%rsp),%xmm6 subsd nb102_iyH2(%rsp),%xmm7 subsd nb102_izH2(%rsp),%xmm8 movsd %xmm0,nb102_dxOO(%rsp) movsd %xmm1,nb102_dyOO(%rsp) movsd %xmm2,nb102_dzOO(%rsp) mulsd %xmm0,%xmm0 mulsd %xmm1,%xmm1 mulsd %xmm2,%xmm2 movsd %xmm3,nb102_dxH1O(%rsp) movsd %xmm4,nb102_dyH1O(%rsp) movsd %xmm5,nb102_dzH1O(%rsp) mulsd %xmm3,%xmm3 mulsd %xmm4,%xmm4 mulsd %xmm5,%xmm5 movsd %xmm6,nb102_dxH2O(%rsp) movsd %xmm7,nb102_dyH2O(%rsp) movsd %xmm8,nb102_dzH2O(%rsp) mulsd %xmm6,%xmm6 mulsd %xmm7,%xmm7 mulsd %xmm8,%xmm8 addsd %xmm1,%xmm0 addsd %xmm2,%xmm0 addsd %xmm4,%xmm3 addsd %xmm5,%xmm3 addsd %xmm7,%xmm6 addsd %xmm8,%xmm6 ## start doing invsqrt for jO atoms cvtsd2ss %xmm0,%xmm1 cvtsd2ss %xmm3,%xmm4 cvtsd2ss %xmm6,%xmm7 rsqrtss %xmm1,%xmm1 rsqrtss %xmm4,%xmm4 rsqrtss %xmm7,%xmm7 cvtss2sd %xmm1,%xmm1 cvtss2sd %xmm4,%xmm4 cvtss2sd %xmm7,%xmm7 movsd %xmm1,%xmm2 movsd %xmm4,%xmm5 movsd %xmm7,%xmm8 mulsd %xmm1,%xmm1 ## lu*lu mulsd %xmm4,%xmm4 ## lu*lu mulsd %xmm7,%xmm7 ## lu*lu movsd nb102_three(%rsp),%xmm9 movsd %xmm9,%xmm10 movsd %xmm9,%xmm11 mulsd %xmm0,%xmm1 ## rsq*lu*lu mulsd %xmm3,%xmm4 ## rsq*lu*lu mulsd %xmm6,%xmm7 ## rsq*lu*lu subsd %xmm1,%xmm9 subsd %xmm4,%xmm10 subsd %xmm7,%xmm11 ## 3-rsq*lu*lu mulsd %xmm2,%xmm9 mulsd %xmm5,%xmm10 mulsd %xmm8,%xmm11 ## lu*(3-rsq*lu*lu) movsd nb102_half(%rsp),%xmm15 mulsd %xmm15,%xmm9 ## first iteration for rinvOO mulsd %xmm15,%xmm10 ## first iteration for rinvH1O mulsd %xmm15,%xmm11 ## first iteration for rinvH2O ## second iteration step movsd %xmm9,%xmm2 movsd %xmm10,%xmm5 movsd %xmm11,%xmm8 mulsd %xmm2,%xmm2 ## lu*lu mulsd %xmm5,%xmm5 ## lu*lu mulsd %xmm8,%xmm8 ## lu*lu movsd nb102_three(%rsp),%xmm1 movsd %xmm1,%xmm4 movsd %xmm1,%xmm7 mulsd %xmm0,%xmm2 ## rsq*lu*lu mulsd %xmm3,%xmm5 ## rsq*lu*lu mulsd %xmm6,%xmm8 ## rsq*lu*lu subsd %xmm2,%xmm1 subsd %xmm5,%xmm4 subsd %xmm8,%xmm7 ## 3-rsq*lu*lu mulsd %xmm1,%xmm9 mulsd %xmm4,%xmm10 mulsd %xmm7,%xmm11 ## lu*(3-rsq*lu*lu) movsd nb102_half(%rsp),%xmm15 mulsd %xmm15,%xmm9 ## rinvOO mulsd %xmm15,%xmm10 ## rinvH1O mulsd %xmm15,%xmm11 ## rinvH2O ## O interactions movsd %xmm9,%xmm0 movsd %xmm10,%xmm1 movsd %xmm11,%xmm2 mulsd %xmm9,%xmm9 mulsd %xmm10,%xmm10 mulsd %xmm11,%xmm11 mulsd nb102_qqOO(%rsp),%xmm0 mulsd nb102_qqOH(%rsp),%xmm1 mulsd nb102_qqOH(%rsp),%xmm2 mulsd %xmm0,%xmm9 mulsd %xmm1,%xmm10 mulsd %xmm2,%xmm11 addsd nb102_vctot(%rsp),%xmm0 addsd %xmm2,%xmm1 addsd %xmm1,%xmm0 movsd %xmm0,nb102_vctot(%rsp) ## move j O forces to xmm0-xmm2 movsd (%rdi,%rax,8),%xmm0 movsd 8(%rdi,%rax,8),%xmm1 movsd 16(%rdi,%rax,8),%xmm2 movsd %xmm9,%xmm7 movsd %xmm9,%xmm8 movsd %xmm11,%xmm13 movsd %xmm11,%xmm14 movsd %xmm11,%xmm15 movsd %xmm10,%xmm11 movsd %xmm10,%xmm12 mulsd nb102_dxOO(%rsp),%xmm7 mulsd nb102_dyOO(%rsp),%xmm8 mulsd nb102_dzOO(%rsp),%xmm9 mulsd nb102_dxH1O(%rsp),%xmm10 mulsd nb102_dyH1O(%rsp),%xmm11 mulsd nb102_dzH1O(%rsp),%xmm12 mulsd nb102_dxH2O(%rsp),%xmm13 mulsd nb102_dyH2O(%rsp),%xmm14 mulsd nb102_dzH2O(%rsp),%xmm15 addsd %xmm7,%xmm0 addsd %xmm8,%xmm1 addsd %xmm9,%xmm2 addsd nb102_fixO(%rsp),%xmm7 addsd nb102_fiyO(%rsp),%xmm8 addsd nb102_fizO(%rsp),%xmm9 addsd %xmm10,%xmm0 addsd %xmm11,%xmm1 addsd %xmm12,%xmm2 addsd nb102_fixH1(%rsp),%xmm10 addsd nb102_fiyH1(%rsp),%xmm11 addsd nb102_fizH1(%rsp),%xmm12 addsd %xmm13,%xmm0 addsd %xmm14,%xmm1 addsd %xmm15,%xmm2 addsd nb102_fixH2(%rsp),%xmm13 addsd nb102_fiyH2(%rsp),%xmm14 addsd nb102_fizH2(%rsp),%xmm15 movsd %xmm7,nb102_fixO(%rsp) movsd %xmm8,nb102_fiyO(%rsp) movsd %xmm9,nb102_fizO(%rsp) movsd %xmm10,nb102_fixH1(%rsp) movsd %xmm11,nb102_fiyH1(%rsp) movsd %xmm12,nb102_fizH1(%rsp) movsd %xmm13,nb102_fixH2(%rsp) movsd %xmm14,nb102_fiyH2(%rsp) movsd %xmm15,nb102_fizH2(%rsp) ## store back j O forces from xmm0-xmm2 movsd %xmm0,(%rdi,%rax,8) movsd %xmm1,8(%rdi,%rax,8) movsd %xmm2,16(%rdi,%rax,8) ## move j H1 coordinates to local temp variables movsd 24(%rsi,%rax,8),%xmm0 movsd 32(%rsi,%rax,8),%xmm1 movsd 40(%rsi,%rax,8),%xmm2 ## xmm0 = H1x ## xmm1 = H1y ## xmm2 = H1z movsd %xmm0,%xmm3 movsd %xmm1,%xmm4 movsd %xmm2,%xmm5 movsd %xmm0,%xmm6 movsd %xmm1,%xmm7 movsd %xmm2,%xmm8 subsd nb102_ixO(%rsp),%xmm0 subsd nb102_iyO(%rsp),%xmm1 subsd nb102_izO(%rsp),%xmm2 subsd nb102_ixH1(%rsp),%xmm3 subsd nb102_iyH1(%rsp),%xmm4 subsd nb102_izH1(%rsp),%xmm5 subsd nb102_ixH2(%rsp),%xmm6 subsd nb102_iyH2(%rsp),%xmm7 subsd nb102_izH2(%rsp),%xmm8 movsd %xmm0,nb102_dxOH1(%rsp) movsd %xmm1,nb102_dyOH1(%rsp) movsd %xmm2,nb102_dzOH1(%rsp) mulsd %xmm0,%xmm0 mulsd %xmm1,%xmm1 mulsd %xmm2,%xmm2 movsd %xmm3,nb102_dxH1H1(%rsp) movsd %xmm4,nb102_dyH1H1(%rsp) movsd %xmm5,nb102_dzH1H1(%rsp) mulsd %xmm3,%xmm3 mulsd %xmm4,%xmm4 mulsd %xmm5,%xmm5 movsd %xmm6,nb102_dxH2H1(%rsp) movsd %xmm7,nb102_dyH2H1(%rsp) movsd %xmm8,nb102_dzH2H1(%rsp) mulsd %xmm6,%xmm6 mulsd %xmm7,%xmm7 mulsd %xmm8,%xmm8 addsd %xmm1,%xmm0 addsd %xmm2,%xmm0 addsd %xmm4,%xmm3 addsd %xmm5,%xmm3 addsd %xmm7,%xmm6 addsd %xmm8,%xmm6 ## start doing invsqrt for jH1 atoms cvtsd2ss %xmm0,%xmm1 cvtsd2ss %xmm3,%xmm4 cvtsd2ss %xmm6,%xmm7 rsqrtss %xmm1,%xmm1 rsqrtss %xmm4,%xmm4 rsqrtss %xmm7,%xmm7 cvtss2sd %xmm1,%xmm1 cvtss2sd %xmm4,%xmm4 cvtss2sd %xmm7,%xmm7 movsd %xmm1,%xmm2 movsd %xmm4,%xmm5 movsd %xmm7,%xmm8 mulsd %xmm1,%xmm1 ## lu*lu mulsd %xmm4,%xmm4 ## lu*lu mulsd %xmm7,%xmm7 ## lu*lu movsd nb102_three(%rsp),%xmm9 movsd %xmm9,%xmm10 movsd %xmm9,%xmm11 mulsd %xmm0,%xmm1 ## rsq*lu*lu mulsd %xmm3,%xmm4 ## rsq*lu*lu mulsd %xmm6,%xmm7 ## rsq*lu*lu subsd %xmm1,%xmm9 subsd %xmm4,%xmm10 subsd %xmm7,%xmm11 ## 3-rsq*lu*lu mulsd %xmm2,%xmm9 mulsd %xmm5,%xmm10 mulsd %xmm8,%xmm11 ## lu*(3-rsq*lu*lu) movsd nb102_half(%rsp),%xmm15 mulsd %xmm15,%xmm9 ## first iteration for rinvOH1 mulsd %xmm15,%xmm10 ## first iteration for rinvH1H1 mulsd %xmm15,%xmm11 ## first iteration for rinvH2OH1 ## second iteration step movsd %xmm9,%xmm2 movsd %xmm10,%xmm5 movsd %xmm11,%xmm8 mulsd %xmm2,%xmm2 ## lu*lu mulsd %xmm5,%xmm5 ## lu*lu mulsd %xmm8,%xmm8 ## lu*lu movsd nb102_three(%rsp),%xmm1 movsd %xmm1,%xmm4 movsd %xmm1,%xmm7 mulsd %xmm0,%xmm2 ## rsq*lu*lu mulsd %xmm3,%xmm5 ## rsq*lu*lu mulsd %xmm6,%xmm8 ## rsq*lu*lu subsd %xmm2,%xmm1 subsd %xmm5,%xmm4 subsd %xmm8,%xmm7 ## 3-rsq*lu*lu mulsd %xmm1,%xmm9 mulsd %xmm4,%xmm10 mulsd %xmm7,%xmm11 ## lu*(3-rsq*lu*lu) movsd nb102_half(%rsp),%xmm15 mulsd %xmm15,%xmm9 ## rinvOH1 mulsd %xmm15,%xmm10 ## rinvH1H1 mulsd %xmm15,%xmm11 ## rinvH2H1 ## H1 interactions movsd %xmm9,%xmm0 movsd %xmm10,%xmm1 movsd %xmm11,%xmm2 mulsd %xmm9,%xmm9 mulsd %xmm10,%xmm10 mulsd %xmm11,%xmm11 mulsd nb102_qqOH(%rsp),%xmm0 mulsd nb102_qqHH(%rsp),%xmm1 mulsd nb102_qqHH(%rsp),%xmm2 mulsd %xmm0,%xmm9⌨️ 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?