nb_kernel103_x86_64_sse2.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 1,598 行 · 第 1/4 页
S
1,598 行
subpd %xmm7,%xmm11 ## 3-rsq*lu*lu mulpd %xmm2,%xmm9 mulpd %xmm5,%xmm10 mulpd %xmm8,%xmm11 ## lu*(3-rsq*lu*lu) movapd nb103_half(%rsp),%xmm15 mulpd %xmm15,%xmm9 ## first iteration for rinvH1 mulpd %xmm15,%xmm10 ## first iteration for rinvH2 mulpd %xmm15,%xmm11 ## first iteration for rinvM ## 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 nb103_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 nb103_half(%rsp),%xmm15 mulpd %xmm15,%xmm9 ## rinvH1 mulpd %xmm15,%xmm10 ## rinvH2 mulpd %xmm15,%xmm11 ## rinvM ## interactions movapd %xmm9,%xmm0 movapd %xmm10,%xmm1 movapd %xmm11,%xmm2 mulpd %xmm9,%xmm9 mulpd %xmm10,%xmm10 mulpd %xmm11,%xmm11 mulpd nb103_qqH(%rsp),%xmm0 mulpd nb103_qqH(%rsp),%xmm1 mulpd nb103_qqM(%rsp),%xmm2 mulpd %xmm0,%xmm9 mulpd %xmm1,%xmm10 mulpd %xmm2,%xmm11 addpd nb103_vctot(%rsp),%xmm0 addpd %xmm2,%xmm1 addpd %xmm1,%xmm0 movapd %xmm0,nb103_vctot(%rsp) ## move j forces to xmm0-xmm2 movq nb103_faction(%rbp),%rdi movlpd (%rdi,%rax,8),%xmm0 movlpd 8(%rdi,%rax,8),%xmm1 movlpd 16(%rdi,%rax,8),%xmm2 movhpd (%rdi,%rbx,8),%xmm0 movhpd 8(%rdi,%rbx,8),%xmm1 movhpd 16(%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 nb103_dxH1(%rsp),%xmm7 mulpd nb103_dyH1(%rsp),%xmm8 mulpd nb103_dzH1(%rsp),%xmm9 mulpd nb103_dxH2(%rsp),%xmm10 mulpd nb103_dyH2(%rsp),%xmm11 mulpd nb103_dzH2(%rsp),%xmm12 mulpd nb103_dxM(%rsp),%xmm13 mulpd nb103_dyM(%rsp),%xmm14 mulpd nb103_dzM(%rsp),%xmm15 addpd %xmm7,%xmm0 addpd %xmm8,%xmm1 addpd %xmm9,%xmm2 addpd nb103_fixH1(%rsp),%xmm7 addpd nb103_fiyH1(%rsp),%xmm8 addpd nb103_fizH1(%rsp),%xmm9 addpd %xmm10,%xmm0 addpd %xmm11,%xmm1 addpd %xmm12,%xmm2 addpd nb103_fixH2(%rsp),%xmm10 addpd nb103_fiyH2(%rsp),%xmm11 addpd nb103_fizH2(%rsp),%xmm12 addpd %xmm13,%xmm0 addpd %xmm14,%xmm1 addpd %xmm15,%xmm2 addpd nb103_fixM(%rsp),%xmm13 addpd nb103_fiyM(%rsp),%xmm14 addpd nb103_fizM(%rsp),%xmm15 movapd %xmm7,nb103_fixH1(%rsp) movapd %xmm8,nb103_fiyH1(%rsp) movapd %xmm9,nb103_fizH1(%rsp) movapd %xmm10,nb103_fixH2(%rsp) movapd %xmm11,nb103_fiyH2(%rsp) movapd %xmm12,nb103_fizH2(%rsp) movapd %xmm13,nb103_fixM(%rsp) movapd %xmm14,nb103_fiyM(%rsp) movapd %xmm15,nb103_fizM(%rsp) ## store back j forces from xmm0-xmm2 movlpd %xmm0,(%rdi,%rax,8) movlpd %xmm1,8(%rdi,%rax,8) movlpd %xmm2,16(%rdi,%rax,8) movhpd %xmm0,(%rdi,%rbx,8) movhpd %xmm1,8(%rdi,%rbx,8) movhpd %xmm2,16(%rdi,%rbx,8) ## should we do one more iteration? subl $2,nb103_innerk(%rsp) jl _nb_kernel103_x86_64_sse2.nb103_checksingle jmp _nb_kernel103_x86_64_sse2.nb103_unroll_loop_nb_kernel103_x86_64_sse2.nb103_checksingle: movl nb103_innerk(%rsp),%edx andl $1,%edx jnz _nb_kernel103_x86_64_sse2.nb103_dosingle jmp _nb_kernel103_x86_64_sse2.nb103_updateouterdata_nb_kernel103_x86_64_sse2.nb103_dosingle: movq nb103_innerjjnr(%rsp),%rdx ## pointer to jjnr[k] movl (%rdx),%eax movq nb103_charge(%rbp),%rsi ## base of charge[] xorpd %xmm6,%xmm6 movlpd (%rsi,%rax,8),%xmm6 ## jq A movapd nb103_iqM(%rsp),%xmm3 movapd nb103_iqH(%rsp),%xmm4 mulsd %xmm6,%xmm3 ## qqM mulsd %xmm6,%xmm4 ## qqH movapd %xmm3,nb103_qqM(%rsp) movapd %xmm4,nb103_qqH(%rsp) movq nb103_pos(%rbp),%rsi ## base of pos[] lea (%rax,%rax,2),%rax ## replace jnr with j3 ## move coordinates to xmm4-xmm6 & xmm0-xmm2 movlpd (%rsi,%rax,8),%xmm4 movlpd 8(%rsi,%rax,8),%xmm5 movlpd 16(%rsi,%rax,8),%xmm6 movapd %xmm4,%xmm0 movapd %xmm5,%xmm1 movapd %xmm6,%xmm2 ## calc dr subsd nb103_ixM(%rsp),%xmm4 subsd nb103_iyM(%rsp),%xmm5 subsd nb103_izM(%rsp),%xmm6 ## store dr movapd %xmm4,nb103_dxM(%rsp) movapd %xmm5,nb103_dyM(%rsp) movapd %xmm6,nb103_dzM(%rsp) ## square it mulsd %xmm4,%xmm4 mulsd %xmm5,%xmm5 mulsd %xmm6,%xmm6 addsd %xmm5,%xmm4 addsd %xmm6,%xmm4 movapd %xmm4,%xmm7 ## rsqM in xmm7 ## move j coords to xmm4-xmm6 movapd %xmm0,%xmm4 movapd %xmm1,%xmm5 movapd %xmm2,%xmm6 ## calc dr subsd nb103_ixH1(%rsp),%xmm4 subsd nb103_iyH1(%rsp),%xmm5 subsd nb103_izH1(%rsp),%xmm6 ## store dr movapd %xmm4,nb103_dxH1(%rsp) movapd %xmm5,nb103_dyH1(%rsp) movapd %xmm6,nb103_dzH1(%rsp) ## square it mulsd %xmm4,%xmm4 mulsd %xmm5,%xmm5 mulsd %xmm6,%xmm6 addsd %xmm5,%xmm6 addsd %xmm4,%xmm6 ## rsqH1 in xmm6 ## move j coords to xmm3-xmm5 movapd %xmm0,%xmm3 movapd %xmm1,%xmm4 movapd %xmm2,%xmm5 ## calc dr subsd nb103_ixH2(%rsp),%xmm3 subsd nb103_iyH2(%rsp),%xmm4 subsd nb103_izH2(%rsp),%xmm5 ## store dr movapd %xmm3,nb103_dxH2(%rsp) movapd %xmm4,nb103_dyH2(%rsp) movapd %xmm5,nb103_dzH2(%rsp) ## square it mulsd %xmm3,%xmm3 mulsd %xmm4,%xmm4 mulsd %xmm5,%xmm5 addsd %xmm4,%xmm5 addsd %xmm3,%xmm5 ## rsqH2 in xmm5, rsqH1 in xmm6, rsqO in xmm7 ## start with rsqM - put seed in xmm2 cvtsd2ss %xmm7,%xmm2 rsqrtss %xmm2,%xmm2 cvtss2sd %xmm2,%xmm2 movapd %xmm2,%xmm3 mulsd %xmm2,%xmm2 movapd nb103_three(%rsp),%xmm4 mulsd %xmm7,%xmm2 ## rsq*lu*lu subsd %xmm2,%xmm4 ## 30-rsq*lu*lu mulsd %xmm3,%xmm4 ## lu*(3-rsq*lu*lu) mulsd nb103_half(%rsp),%xmm4 ## iter1 ( new lu) movapd %xmm4,%xmm3 mulsd %xmm4,%xmm4 ## lu*lu mulsd %xmm4,%xmm7 ## rsq*lu*lu movapd nb103_three(%rsp),%xmm4 subsd %xmm7,%xmm4 ## 3-rsq*lu*lu mulsd %xmm3,%xmm4 ## lu*( 3-rsq*lu*lu) mulsd nb103_half(%rsp),%xmm4 ## rinv movapd %xmm4,%xmm7 ## rinvM in xmm7 ## rsqH1 - seed in xmm2 cvtsd2ss %xmm6,%xmm2 rsqrtss %xmm2,%xmm2 cvtss2sd %xmm2,%xmm2 movapd %xmm2,%xmm3 mulsd %xmm2,%xmm2 movapd nb103_three(%rsp),%xmm4 mulsd %xmm6,%xmm2 ## rsq*lu*lu subsd %xmm2,%xmm4 ## 30-rsq*lu*lu mulsd %xmm3,%xmm4 ## lu*(3-rsq*lu*lu) mulsd nb103_half(%rsp),%xmm4 ## iter1 ( new lu) movapd %xmm4,%xmm3 mulsd %xmm4,%xmm4 ## lu*lu mulsd %xmm4,%xmm6 ## rsq*lu*lu movapd nb103_three(%rsp),%xmm4 subsd %xmm6,%xmm4 ## 3-rsq*lu*lu mulsd %xmm3,%xmm4 ## lu*( 3-rsq*lu*lu) mulsd nb103_half(%rsp),%xmm4 ## rinv movapd %xmm4,%xmm6 ## rinvH1 in xmm6 ## rsqH2 - seed in xmm2 cvtsd2ss %xmm5,%xmm2 rsqrtss %xmm2,%xmm2 cvtss2sd %xmm2,%xmm2 movapd %xmm2,%xmm3 mulsd %xmm2,%xmm2 movapd nb103_three(%rsp),%xmm4 mulsd %xmm5,%xmm2 ## rsq*lu*lu subsd %xmm2,%xmm4 ## 30-rsq*lu*lu mulsd %xmm3,%xmm4 ## lu*(3-rsq*lu*lu) mulsd nb103_half(%rsp),%xmm4 ## iter1 ( new lu) movapd %xmm4,%xmm3 mulsd %xmm4,%xmm4 ## lu*lu mulsd %xmm4,%xmm5 ## rsq*lu*lu movapd nb103_three(%rsp),%xmm4 subsd %xmm5,%xmm4 ## 3-rsq*lu*lu mulsd %xmm3,%xmm4 ## lu*( 3-rsq*lu*lu) mulsd nb103_half(%rsp),%xmm4 ## rinv movapd %xmm4,%xmm5 ## rinvH2 in xmm5 ## do M interactions movapd %xmm7,%xmm4 mulsd %xmm4,%xmm4 ## xmm7=rinv, xmm4=rinvsq mulsd nb103_qqM(%rsp),%xmm7 ## xmm7=vcoul mulsd %xmm7,%xmm4 ## total fsM in xmm4 addsd nb103_vctot(%rsp),%xmm7 movlpd %xmm7,nb103_vctot(%rsp) movapd nb103_dxM(%rsp),%xmm0 movapd nb103_dyM(%rsp),%xmm1 movapd nb103_dzM(%rsp),%xmm2 mulsd %xmm4,%xmm0 mulsd %xmm4,%xmm1 mulsd %xmm4,%xmm2 ## update M forces movapd nb103_fixM(%rsp),%xmm3 movapd nb103_fiyM(%rsp),%xmm4 movapd nb103_fizM(%rsp),%xmm7 addsd %xmm0,%xmm3 addsd %xmm1,%xmm4 addsd %xmm2,%xmm7 movlpd %xmm3,nb103_fixM(%rsp) movlpd %xmm4,nb103_fiyM(%rsp) movlpd %xmm7,nb103_fizM(%rsp) ## update j forces with water M movlpd %xmm0,nb103_fjx(%rsp) movlpd %xmm1,nb103_fjy(%rsp) movlpd %xmm2,nb103_fjz(%rsp) ## H1 interactions movapd %xmm6,%xmm4 mulsd %xmm4,%xmm4 ## xmm6=rinv, xmm4=rinvsq mulsd nb103_qqH(%rsp),%xmm6 ## xmm6=vcoul mulsd %xmm6,%xmm4 ## total fsH1 in xmm4 addsd nb103_vctot(%rsp),%xmm6 movapd nb103_dxH1(%rsp),%xmm0 movapd nb103_dyH1(%rsp),%xmm1 movapd nb103_dzH1(%rsp),%xmm2 movlpd %xmm6,nb103_vctot(%rsp) mulsd %xmm4,%xmm0 mulsd %xmm4,%xmm1 mulsd %xmm4,%xmm2 ## update H1 forces movapd nb103_fixH1(%rsp),%xmm3 movapd nb103_fiyH1(%rsp),%xmm4 movapd nb103_fizH1(%rsp),%xmm7 addsd %xmm0,%xmm3 addsd %xmm1,%xmm4 addsd %xmm2,%xmm7 movlpd %xmm3,nb103_fixH1(%rsp) movlpd %xmm4,nb103_fiyH1(%rsp) movlpd %xmm7,nb103_fizH1(%rsp) ## update j forces with water H1 addsd nb103_fjx(%rsp),%xmm0 addsd nb103_fjy(%rsp),%xmm1 addsd nb103_fjz(%rsp),%xmm2 movsd %xmm0,nb103_fjx(%rsp) movsd %xmm1,nb103_fjy(%rsp) movsd %xmm2,nb103_fjz(%rsp) ## H2 interactions movapd %xmm5,%xmm4 mulsd %xmm4,%xmm4 ## xmm5=rinv, xmm4=rinvsq mulsd nb103_qqH(%rsp),%xmm5 ## xmm5=vcoul mulsd %xmm5,%xmm4 ## total fsH1 in xmm4 addsd nb103_vctot(%rsp),%xmm5 movapd nb103_dxH2(%rsp),%xmm0 movapd nb103_dyH2(%rsp),%xmm1 movapd nb103_dzH2(%rsp),%xmm2 movlpd %xmm5,nb103_vctot(%rsp) mulsd %xmm4,%xmm0 mulsd %xmm4,%xmm1 mulsd %xmm4,%xmm2 ## update H2 forces movapd nb103_fixH2(%rsp),%xmm3 movapd nb103_fiyH2(%rsp),%xmm4 movapd nb103_fizH2(%rsp),%xmm7 addsd %xmm0,%xmm3 addsd %xmm1,%xmm4 addsd %xmm2,%xmm7 movlpd %xmm3,nb103_fixH2(%rsp) movlpd %xmm4,nb103_fiyH2(%rsp) movlpd %xmm7,nb103_fizH2(%rsp) movq nb103_faction(%rbp),%rdi ## update j forces addsd nb103_fjx(%rsp),%xmm0 addsd nb103_fjy(%rsp),%xmm1 addsd nb103_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)⌨️ 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?