nb_kernel104_x86_64_sse2.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,227 行 · 第 1/5 页
S
2,227 行
mulpd %xmm1,%xmm9 mulpd %xmm4,%xmm10 mulpd %xmm7,%xmm11 ## lu*(3-rsq*lu*lu) movapd nb104_half(%rsp),%xmm15 mulpd %xmm15,%xmm9 ## rinvH1M mulpd %xmm15,%xmm10 ## rinvH2M mulpd %xmm15,%xmm11 ## rinvMM ## M interactions movapd %xmm9,%xmm0 movapd %xmm10,%xmm1 movapd %xmm11,%xmm2 mulpd %xmm9,%xmm9 mulpd %xmm10,%xmm10 mulpd %xmm11,%xmm11 mulpd nb104_qqMH(%rsp),%xmm0 mulpd nb104_qqMH(%rsp),%xmm1 mulpd nb104_qqMM(%rsp),%xmm2 mulpd %xmm0,%xmm9 mulpd %xmm1,%xmm10 mulpd %xmm2,%xmm11 addpd nb104_vctot(%rsp),%xmm0 addpd %xmm2,%xmm1 addpd %xmm1,%xmm0 movapd %xmm0,nb104_vctot(%rsp) ## move j M forces to xmm0-xmm2 movq nb104_faction(%rbp),%rdi movlpd 72(%rdi,%rax,8),%xmm0 movlpd 80(%rdi,%rax,8),%xmm1 movlpd 88(%rdi,%rax,8),%xmm2 movhpd 72(%rdi,%rbx,8),%xmm0 movhpd 80(%rdi,%rbx,8),%xmm1 movhpd 88(%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 nb104_dxH1M(%rsp),%xmm7 mulpd nb104_dyH1M(%rsp),%xmm8 mulpd nb104_dzH1M(%rsp),%xmm9 mulpd nb104_dxH2M(%rsp),%xmm10 mulpd nb104_dyH2M(%rsp),%xmm11 mulpd nb104_dzH2M(%rsp),%xmm12 mulpd nb104_dxMM(%rsp),%xmm13 mulpd nb104_dyMM(%rsp),%xmm14 mulpd nb104_dzMM(%rsp),%xmm15 addpd %xmm7,%xmm0 addpd %xmm8,%xmm1 addpd %xmm9,%xmm2 addpd nb104_fixH1(%rsp),%xmm7 addpd nb104_fiyH1(%rsp),%xmm8 addpd nb104_fizH1(%rsp),%xmm9 addpd %xmm10,%xmm0 addpd %xmm11,%xmm1 addpd %xmm12,%xmm2 addpd nb104_fixH2(%rsp),%xmm10 addpd nb104_fiyH2(%rsp),%xmm11 addpd nb104_fizH2(%rsp),%xmm12 addpd %xmm13,%xmm0 addpd %xmm14,%xmm1 addpd %xmm15,%xmm2 addpd nb104_fixM(%rsp),%xmm13 addpd nb104_fiyM(%rsp),%xmm14 addpd nb104_fizM(%rsp),%xmm15 movapd %xmm7,nb104_fixH1(%rsp) movapd %xmm8,nb104_fiyH1(%rsp) movapd %xmm9,nb104_fizH1(%rsp) movapd %xmm10,nb104_fixH2(%rsp) movapd %xmm11,nb104_fiyH2(%rsp) movapd %xmm12,nb104_fizH2(%rsp) movapd %xmm13,nb104_fixM(%rsp) movapd %xmm14,nb104_fiyM(%rsp) movapd %xmm15,nb104_fizM(%rsp) ## store back j M forces from xmm0-xmm2 movlpd %xmm0,72(%rdi,%rax,8) movlpd %xmm1,80(%rdi,%rax,8) movlpd %xmm2,88(%rdi,%rax,8) movhpd %xmm0,72(%rdi,%rbx,8) movhpd %xmm1,80(%rdi,%rbx,8) movhpd %xmm2,88(%rdi,%rbx,8) ## should we do one more iteration? subl $2,nb104_innerk(%rsp) jl _nb_kernel104_x86_64_sse2.nb104_checksingle jmp _nb_kernel104_x86_64_sse2.nb104_unroll_loop_nb_kernel104_x86_64_sse2.nb104_checksingle: movl nb104_innerk(%rsp),%edx andl $1,%edx jnz _nb_kernel104_x86_64_sse2.nb104_dosingle jmp _nb_kernel104_x86_64_sse2.nb104_updateouterdata_nb_kernel104_x86_64_sse2.nb104_dosingle: movq nb104_innerjjnr(%rsp),%rdx ## pointer to jjnr[k] movl (%rdx),%eax movq nb104_pos(%rbp),%rsi lea (%rax,%rax,2),%rax ## 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 nb104_ixH1(%rsp),%xmm0 subsd nb104_iyH1(%rsp),%xmm1 subsd nb104_izH1(%rsp),%xmm2 subsd nb104_ixH2(%rsp),%xmm3 subsd nb104_iyH2(%rsp),%xmm4 subsd nb104_izH2(%rsp),%xmm5 subsd nb104_ixM(%rsp),%xmm6 subsd nb104_iyM(%rsp),%xmm7 subsd nb104_izM(%rsp),%xmm8 movsd %xmm0,nb104_dxH1H1(%rsp) movsd %xmm1,nb104_dyH1H1(%rsp) movsd %xmm2,nb104_dzH1H1(%rsp) mulsd %xmm0,%xmm0 mulsd %xmm1,%xmm1 mulsd %xmm2,%xmm2 movsd %xmm3,nb104_dxH2H1(%rsp) movsd %xmm4,nb104_dyH2H1(%rsp) movsd %xmm5,nb104_dzH2H1(%rsp) mulsd %xmm3,%xmm3 mulsd %xmm4,%xmm4 mulsd %xmm5,%xmm5 movsd %xmm6,nb104_dxMH1(%rsp) movsd %xmm7,nb104_dyMH1(%rsp) movsd %xmm8,nb104_dzMH1(%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 nb104_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 nb104_half(%rsp),%xmm15 mulsd %xmm15,%xmm9 ## first iteration for rinvH1H1 mulsd %xmm15,%xmm10 ## first iteration for rinvH2H1 mulsd %xmm15,%xmm11 ## first iteration for rinvMH1 ## 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 nb104_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 nb104_half(%rsp),%xmm15 mulsd %xmm15,%xmm9 ## rinvH1H1 mulsd %xmm15,%xmm10 ## rinvH2H1 mulsd %xmm15,%xmm11 ## rinvMH1 ## H1 interactions movsd %xmm9,%xmm0 movsd %xmm10,%xmm1 movsd %xmm11,%xmm2 mulsd %xmm9,%xmm9 mulsd %xmm10,%xmm10 mulsd %xmm11,%xmm11 mulsd nb104_qqHH(%rsp),%xmm0 mulsd nb104_qqHH(%rsp),%xmm1 mulsd nb104_qqMH(%rsp),%xmm2 mulsd %xmm0,%xmm9 mulsd %xmm1,%xmm10 mulsd %xmm2,%xmm11 addsd nb104_vctot(%rsp),%xmm0 addsd %xmm2,%xmm1 addsd %xmm1,%xmm0 movsd %xmm0,nb104_vctot(%rsp) ## move j H1 forces to xmm0-xmm2 movsd 24(%rdi,%rax,8),%xmm0 movsd 32(%rdi,%rax,8),%xmm1 movsd 40(%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 nb104_dxH1H1(%rsp),%xmm7 mulsd nb104_dyH1H1(%rsp),%xmm8 mulsd nb104_dzH1H1(%rsp),%xmm9 mulsd nb104_dxH2H1(%rsp),%xmm10 mulsd nb104_dyH2H1(%rsp),%xmm11 mulsd nb104_dzH2H1(%rsp),%xmm12 mulsd nb104_dxMH1(%rsp),%xmm13 mulsd nb104_dyMH1(%rsp),%xmm14 mulsd nb104_dzMH1(%rsp),%xmm15 addsd %xmm7,%xmm0 addsd %xmm8,%xmm1 addsd %xmm9,%xmm2 addsd nb104_fixH1(%rsp),%xmm7 addsd nb104_fiyH1(%rsp),%xmm8 addsd nb104_fizH1(%rsp),%xmm9 addsd %xmm10,%xmm0 addsd %xmm11,%xmm1 addsd %xmm12,%xmm2 addsd nb104_fixH2(%rsp),%xmm10 addsd nb104_fiyH2(%rsp),%xmm11 addsd nb104_fizH2(%rsp),%xmm12 addsd %xmm13,%xmm0 addsd %xmm14,%xmm1 addsd %xmm15,%xmm2 addsd nb104_fixM(%rsp),%xmm13 addsd nb104_fiyM(%rsp),%xmm14 addsd nb104_fizM(%rsp),%xmm15 movsd %xmm7,nb104_fixH1(%rsp) movsd %xmm8,nb104_fiyH1(%rsp) movsd %xmm9,nb104_fizH1(%rsp) movsd %xmm10,nb104_fixH2(%rsp) movsd %xmm11,nb104_fiyH2(%rsp) movsd %xmm12,nb104_fizH2(%rsp) movsd %xmm13,nb104_fixM(%rsp) movsd %xmm14,nb104_fiyM(%rsp) movsd %xmm15,nb104_fizM(%rsp) ## store back j H1 forces from xmm0-xmm2 movsd %xmm0,24(%rdi,%rax,8) movsd %xmm1,32(%rdi,%rax,8) movsd %xmm2,40(%rdi,%rax,8) ## move j H2 coordinates to local temp variables movsd 48(%rsi,%rax,8),%xmm0 movsd 56(%rsi,%rax,8),%xmm1 movsd 64(%rsi,%rax,8),%xmm2 ## xmm0 = H2x ## xmm1 = H2y ## xmm2 = H2z movsd %xmm0,%xmm3 movsd %xmm1,%xmm4 movsd %xmm2,%xmm5 movsd %xmm0,%xmm6 movsd %xmm1,%xmm7 movsd %xmm2,%xmm8 subsd nb104_ixH1(%rsp),%xmm0 subsd nb104_iyH1(%rsp),%xmm1 subsd nb104_izH1(%rsp),%xmm2 subsd nb104_ixH2(%rsp),%xmm3 subsd nb104_iyH2(%rsp),%xmm4 subsd nb104_izH2(%rsp),%xmm5 subsd nb104_ixM(%rsp),%xmm6 subsd nb104_iyM(%rsp),%xmm7 subsd nb104_izM(%rsp),%xmm8 movsd %xmm0,nb104_dxH1H2(%rsp) movsd %xmm1,nb104_dyH1H2(%rsp) movsd %xmm2,nb104_dzH1H2(%rsp) mulsd %xmm0,%xmm0 mulsd %xmm1,%xmm1 mulsd %xmm2,%xmm2 movsd %xmm3,nb104_dxH2H2(%rsp) movsd %xmm4,nb104_dyH2H2(%rsp) movsd %xmm5,nb104_dzH2H2(%rsp) mulsd %xmm3,%xmm3 mulsd %xmm4,%xmm4 mulsd %xmm5,%xmm5 movsd %xmm6,nb104_dxMH2(%rsp) movsd %xmm7,nb104_dyMH2(%rsp) movsd %xmm8,nb104_dzMH2(%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 jH2 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 nb104_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 nb104_half(%rsp),%xmm15 mulsd %xmm15,%xmm9 ## first iteration for rinvH1H2 mulsd %xmm15,%xmm10 ## first iteration for rinvH2H2 mulsd %xmm15,%xmm11 ## first iteration for rinvMH2 ## 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 nb104_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 nb104_half(%rsp),%xmm15 mulsd %xmm15,%xmm9 ## rinvH1H2 mulsd %xmm15,%xmm10 ## rinvH2H2 mulsd %xmm15,%xmm11 ## rinvMH2 ## H2 interactions movsd %xmm9,%xmm0 movsd %xmm10,%xmm1 movsd %xmm11,%xmm2 mulsd %xmm9,%xmm9 mulsd %xmm10,%xmm10 mulsd %xmm11,%xmm11 mulsd nb104_qqHH(%rsp),%xmm0 mulsd nb104_qqHH(%rsp),%xmm1 mulsd nb104_qqMH(%rsp),%xmm2 mulsd %xmm0,%xmm9 mulsd %xmm1,%xmm10 mulsd %xmm2,%xmm11⌨️ 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?