nb_kernel114_x86_64_sse2.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,294 行 · 第 1/5 页
S
2,294 行
movapd %xmm7,nb114_fixH1(%rsp) movapd %xmm8,nb114_fiyH1(%rsp) movapd %xmm9,nb114_fizH1(%rsp) movapd %xmm10,nb114_fixH2(%rsp) movapd %xmm11,nb114_fiyH2(%rsp) movapd %xmm12,nb114_fizH2(%rsp) movapd %xmm13,nb114_fixM(%rsp) movapd %xmm14,nb114_fiyM(%rsp) movapd %xmm15,nb114_fizM(%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) ## move j M coordinates to local temp variables movq nb114_pos(%rbp),%rsi movlpd 72(%rsi,%rax,8),%xmm0 movlpd 80(%rsi,%rax,8),%xmm1 movlpd 88(%rsi,%rax,8),%xmm2 movhpd 72(%rsi,%rbx,8),%xmm0 movhpd 80(%rsi,%rbx,8),%xmm1 movhpd 88(%rsi,%rbx,8),%xmm2 ## xmm0 = Mx ## xmm1 = My ## xmm2 = Mz movapd %xmm0,%xmm3 movapd %xmm1,%xmm4 movapd %xmm2,%xmm5 movapd %xmm0,%xmm6 movapd %xmm1,%xmm7 movapd %xmm2,%xmm8 subpd nb114_ixH1(%rsp),%xmm0 subpd nb114_iyH1(%rsp),%xmm1 subpd nb114_izH1(%rsp),%xmm2 subpd nb114_ixH2(%rsp),%xmm3 subpd nb114_iyH2(%rsp),%xmm4 subpd nb114_izH2(%rsp),%xmm5 subpd nb114_ixM(%rsp),%xmm6 subpd nb114_iyM(%rsp),%xmm7 subpd nb114_izM(%rsp),%xmm8 movapd %xmm0,nb114_dxH1M(%rsp) movapd %xmm1,nb114_dyH1M(%rsp) movapd %xmm2,nb114_dzH1M(%rsp) mulpd %xmm0,%xmm0 mulpd %xmm1,%xmm1 mulpd %xmm2,%xmm2 movapd %xmm3,nb114_dxH2M(%rsp) movapd %xmm4,nb114_dyH2M(%rsp) movapd %xmm5,nb114_dzH2M(%rsp) mulpd %xmm3,%xmm3 mulpd %xmm4,%xmm4 mulpd %xmm5,%xmm5 movapd %xmm6,nb114_dxMM(%rsp) movapd %xmm7,nb114_dyMM(%rsp) movapd %xmm8,nb114_dzMM(%rsp) mulpd %xmm6,%xmm6 mulpd %xmm7,%xmm7 mulpd %xmm8,%xmm8 addpd %xmm1,%xmm0 addpd %xmm2,%xmm0 addpd %xmm4,%xmm3 addpd %xmm5,%xmm3 addpd %xmm7,%xmm6 addpd %xmm8,%xmm6 ## start doing invsqrt for jM atoms cvtpd2ps %xmm0,%xmm1 cvtpd2ps %xmm3,%xmm4 cvtpd2ps %xmm6,%xmm7 rsqrtps %xmm1,%xmm1 rsqrtps %xmm4,%xmm4 rsqrtps %xmm7,%xmm7 cvtps2pd %xmm1,%xmm1 cvtps2pd %xmm4,%xmm4 cvtps2pd %xmm7,%xmm7 movapd %xmm1,%xmm2 movapd %xmm4,%xmm5 movapd %xmm7,%xmm8 mulpd %xmm1,%xmm1 ## lu*lu mulpd %xmm4,%xmm4 ## lu*lu mulpd %xmm7,%xmm7 ## lu*lu movapd nb114_three(%rsp),%xmm9 movapd %xmm9,%xmm10 movapd %xmm9,%xmm11 mulpd %xmm0,%xmm1 ## rsq*lu*lu mulpd %xmm3,%xmm4 ## rsq*lu*lu mulpd %xmm6,%xmm7 ## rsq*lu*lu subpd %xmm1,%xmm9 subpd %xmm4,%xmm10 subpd %xmm7,%xmm11 ## 3-rsq*lu*lu mulpd %xmm2,%xmm9 mulpd %xmm5,%xmm10 mulpd %xmm8,%xmm11 ## lu*(3-rsq*lu*lu) movapd nb114_half(%rsp),%xmm15 mulpd %xmm15,%xmm9 ## first iteration for rinvH1M mulpd %xmm15,%xmm10 ## first iteration for rinvH2M mulpd %xmm15,%xmm11 ## first iteration for rinvMM ## second iteration step movapd %xmm9,%xmm2 movapd %xmm10,%xmm5 movapd %xmm11,%xmm8 mulpd %xmm2,%xmm2 ## lu*lu mulpd %xmm5,%xmm5 ## lu*lu mulpd %xmm8,%xmm8 ## lu*lu movapd nb114_three(%rsp),%xmm1 movapd %xmm1,%xmm4 movapd %xmm1,%xmm7 mulpd %xmm0,%xmm2 ## rsq*lu*lu mulpd %xmm3,%xmm5 ## rsq*lu*lu mulpd %xmm6,%xmm8 ## rsq*lu*lu subpd %xmm2,%xmm1 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 nb114_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 nb114_qqMH(%rsp),%xmm0 mulpd nb114_qqMH(%rsp),%xmm1 mulpd nb114_qqMM(%rsp),%xmm2 mulpd %xmm0,%xmm9 mulpd %xmm1,%xmm10 mulpd %xmm2,%xmm11 addpd nb114_vctot(%rsp),%xmm0 addpd %xmm2,%xmm1 addpd %xmm1,%xmm0 movapd %xmm0,nb114_vctot(%rsp) ## move j M forces to xmm0-xmm2 movq nb114_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 nb114_dxH1M(%rsp),%xmm7 mulpd nb114_dyH1M(%rsp),%xmm8 mulpd nb114_dzH1M(%rsp),%xmm9 mulpd nb114_dxH2M(%rsp),%xmm10 mulpd nb114_dyH2M(%rsp),%xmm11 mulpd nb114_dzH2M(%rsp),%xmm12 mulpd nb114_dxMM(%rsp),%xmm13 mulpd nb114_dyMM(%rsp),%xmm14 mulpd nb114_dzMM(%rsp),%xmm15 addpd %xmm7,%xmm0 addpd %xmm8,%xmm1 addpd %xmm9,%xmm2 addpd nb114_fixH1(%rsp),%xmm7 addpd nb114_fiyH1(%rsp),%xmm8 addpd nb114_fizH1(%rsp),%xmm9 addpd %xmm10,%xmm0 addpd %xmm11,%xmm1 addpd %xmm12,%xmm2 addpd nb114_fixH2(%rsp),%xmm10 addpd nb114_fiyH2(%rsp),%xmm11 addpd nb114_fizH2(%rsp),%xmm12 addpd %xmm13,%xmm0 addpd %xmm14,%xmm1 addpd %xmm15,%xmm2 addpd nb114_fixM(%rsp),%xmm13 addpd nb114_fiyM(%rsp),%xmm14 addpd nb114_fizM(%rsp),%xmm15 movapd %xmm7,nb114_fixH1(%rsp) movapd %xmm8,nb114_fiyH1(%rsp) movapd %xmm9,nb114_fizH1(%rsp) movapd %xmm10,nb114_fixH2(%rsp) movapd %xmm11,nb114_fiyH2(%rsp) movapd %xmm12,nb114_fizH2(%rsp) movapd %xmm13,nb114_fixM(%rsp) movapd %xmm14,nb114_fiyM(%rsp) movapd %xmm15,nb114_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,nb114_innerk(%rsp) jl _nb_kernel114_x86_64_sse2.nb114_checksingle jmp _nb_kernel114_x86_64_sse2.nb114_unroll_loop_nb_kernel114_x86_64_sse2.nb114_checksingle: movl nb114_innerk(%rsp),%edx andl $1,%edx jnz _nb_kernel114_x86_64_sse2.nb114_dosingle jmp _nb_kernel114_x86_64_sse2.nb114_updateouterdata_nb_kernel114_x86_64_sse2.nb114_dosingle: movq nb114_innerjjnr(%rsp),%rdx ## pointer to jjnr[k] movl (%rdx),%eax movq nb114_pos(%rbp),%rsi lea (%rax,%rax,2),%rax ## load j O coordinates movsd (%rsi,%rax,8),%xmm4 movsd 8(%rsi,%rax,8),%xmm5 movsd 16(%rsi,%rax,8),%xmm6 ## xmm4 = Ox ## xmm5 = Oy ## xmm6 = Oz subsd nb114_ixO(%rsp),%xmm4 subsd nb114_iyO(%rsp),%xmm5 subsd nb114_izO(%rsp),%xmm6 ## store dx/dy/dz movapd %xmm4,%xmm9 movapd %xmm5,%xmm10 movapd %xmm6,%xmm11 ## square it mulsd %xmm4,%xmm4 mulsd %xmm5,%xmm5 mulsd %xmm6,%xmm6 addsd %xmm5,%xmm4 addsd %xmm6,%xmm4 ## rsq in xmm4 cvtsd2ss %xmm4,%xmm6 rcpss %xmm6,%xmm6 cvtss2sd %xmm6,%xmm6 ## lu in low xmm6 ## 1/x lookup seed in xmm6 movapd nb114_two(%rsp),%xmm0 movapd %xmm4,%xmm5 mulsd %xmm6,%xmm4 ## lu*rsq subsd %xmm4,%xmm0 ## 2-lu*rsq mulsd %xmm0,%xmm6 ## (new lu) movapd nb114_two(%rsp),%xmm0 mulsd %xmm6,%xmm5 ## lu*rsq subsd %xmm5,%xmm0 ## 2-lu*rsq mulsd %xmm6,%xmm0 ## xmm0=rinvsq movapd %xmm0,%xmm1 mulsd %xmm0,%xmm1 mulsd %xmm0,%xmm1 ## xmm1=rinvsix movapd %xmm1,%xmm2 mulsd %xmm2,%xmm2 ## xmm2=rinvtwelve mulsd nb114_c6(%rsp),%xmm1 ## mult by c6 mulsd nb114_c12(%rsp),%xmm2 ## mult by c12 movapd %xmm2,%xmm5 subsd %xmm1,%xmm5 ## Vvdw=Vvdw12-Vvdw6 mulsd nb114_six(%rsp),%xmm1 mulsd nb114_twelve(%rsp),%xmm2 subsd %xmm1,%xmm2 mulsd %xmm2,%xmm0 ## xmm0=total fscal ## increment potential addsd nb114_Vvdwtot(%rsp),%xmm5 movsd %xmm5,nb114_Vvdwtot(%rsp) movq nb114_faction(%rbp),%rdi mulsd %xmm0,%xmm9 mulsd %xmm0,%xmm10 mulsd %xmm0,%xmm11 movapd nb114_fixO(%rsp),%xmm0 movapd nb114_fiyO(%rsp),%xmm1 movapd nb114_fizO(%rsp),%xmm2 ## accumulate i forces addsd %xmm9,%xmm0 addsd %xmm10,%xmm1 addsd %xmm11,%xmm2 movsd %xmm0,nb114_fixO(%rsp) movsd %xmm1,nb114_fiyO(%rsp) movsd %xmm2,nb114_fizO(%rsp) addsd (%rdi,%rax,8),%xmm9 addsd 8(%rdi,%rax,8),%xmm10 addsd 16(%rdi,%rax,8),%xmm11 movsd %xmm9,(%rdi,%rax,8) movsd %xmm10,8(%rdi,%rax,8) movsd %xmm11,16(%rdi,%rax,8) ## done with OO interaction ## 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 nb114_ixH1(%rsp),%xmm0 subsd nb114_iyH1(%rsp),%xmm1 subsd nb114_izH1(%rsp),%xmm2 subsd nb114_ixH2(%rsp),%xmm3 subsd nb114_iyH2(%rsp),%xmm4 subsd nb114_izH2(%rsp),%xmm5 subsd nb114_ixM(%rsp),%xmm6 subsd nb114_iyM(%rsp),%xmm7 subsd nb114_izM(%rsp),%xmm8 movsd %xmm0,nb114_dxH1H1(%rsp) movsd %xmm1,nb114_dyH1H1(%rsp) movsd %xmm2,nb114_dzH1H1(%rsp) mulsd %xmm0,%xmm0 mulsd %xmm1,%xmm1 mulsd %xmm2,%xmm2 movsd %xmm3,nb114_dxH2H1(%rsp) movsd %xmm4,nb114_dyH2H1(%rsp) movsd %xmm5,nb114_dzH2H1(%rsp) mulsd %xmm3,%xmm3 mulsd %xmm4,%xmm4 mulsd %xmm5,%xmm5 movsd %xmm6,nb114_dxMH1(%rsp) movsd %xmm7,nb114_dyMH1(%rsp) movsd %xmm8,nb114_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 nb114_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 nb114_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 nb114_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 nb114_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⌨️ 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?