nb_kernel133_x86_64_sse2.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,141 行 · 第 1/5 页
S
2,141 行
## calc dr subsd nb133_ixH2(%rsp),%xmm3 subsd nb133_iyH2(%rsp),%xmm4 subsd nb133_izH2(%rsp),%xmm5 ## store dr movapd %xmm3,nb133_dxH2(%rsp) movapd %xmm4,nb133_dyH2(%rsp) movapd %xmm5,nb133_dzH2(%rsp) ## square it mulsd %xmm3,%xmm3 mulsd %xmm4,%xmm4 mulsd %xmm5,%xmm5 addsd %xmm4,%xmm5 addsd %xmm3,%xmm5 ## move j coords to xmm4-xmm2 movapd %xmm0,%xmm4 movapd %xmm1,%xmm3 ## xmm2 already contains z ## calc dr subsd nb133_ixM(%rsp),%xmm4 subsd nb133_iyM(%rsp),%xmm3 subsd nb133_izM(%rsp),%xmm2 ## store dr movapd %xmm4,nb133_dxM(%rsp) movapd %xmm3,nb133_dyM(%rsp) movapd %xmm2,nb133_dzM(%rsp) ## square it mulpd %xmm2,%xmm2 mulpd %xmm3,%xmm3 mulpd %xmm4,%xmm4 addpd %xmm3,%xmm4 addpd %xmm2,%xmm4 ## rsqM in xmm4, rsqH2 in xmm5, rsqH1 in xmm6, rsqO in xmm7 ## start with rsqH1 - put seed in xmm2 cvtsd2ss %xmm6,%xmm2 rsqrtss %xmm2,%xmm2 cvtss2sd %xmm2,%xmm2 movapd %xmm2,%xmm3 mulsd %xmm2,%xmm2 movapd nb133_three(%rsp),%xmm1 mulsd %xmm6,%xmm2 ## rsq*lu*lu subsd %xmm2,%xmm1 ## 30-rsq*lu*lu mulsd %xmm3,%xmm1 ## lu*(3-rsq*lu*lu) mulsd nb133_half(%rsp),%xmm1 ## iter1 ( new lu) movapd %xmm1,%xmm3 mulsd %xmm1,%xmm1 ## lu*lu mulsd %xmm1,%xmm6 ## rsq*lu*lu movapd nb133_three(%rsp),%xmm1 subsd %xmm6,%xmm1 ## 3-rsq*lu*lu mulsd %xmm3,%xmm1 ## lu*( 3-rsq*lu*lu) mulsd nb133_half(%rsp),%xmm1 ## rinv movapd %xmm1,nb133_rinvH1(%rsp) ## rsqH2 - seed in xmm2 cvtsd2ss %xmm5,%xmm2 rsqrtss %xmm2,%xmm2 cvtss2sd %xmm2,%xmm2 movapd %xmm2,%xmm3 mulsd %xmm2,%xmm2 movapd nb133_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 nb133_half(%rsp),%xmm1 ## iter1 ( new lu) movapd %xmm1,%xmm3 mulsd %xmm1,%xmm1 ## lu*lu mulsd %xmm1,%xmm5 ## rsq*lu*lu movapd nb133_three(%rsp),%xmm1 subsd %xmm5,%xmm1 ## 3-rsq*lu*lu mulsd %xmm3,%xmm1 ## lu*( 3-rsq*lu*lu) mulsd nb133_half(%rsp),%xmm1 ## rinv movapd %xmm1,nb133_rinvH2(%rsp) ## rsqM - seed in xmm2 cvtsd2ss %xmm4,%xmm2 rsqrtss %xmm2,%xmm2 cvtss2sd %xmm2,%xmm2 movapd %xmm2,%xmm3 mulsd %xmm2,%xmm2 movapd nb133_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 nb133_half(%rsp),%xmm1 ## iter1 ( new lu) movapd %xmm1,%xmm3 mulsd %xmm1,%xmm1 ## lu*lu mulsd %xmm1,%xmm4 ## rsq*lu*lu movapd nb133_three(%rsp),%xmm1 subsd %xmm4,%xmm1 ## 3-rsq*lu*lu mulsd %xmm3,%xmm1 ## lu*( 3-rsq*lu*lu) mulsd nb133_half(%rsp),%xmm1 ## rinv movapd %xmm1,nb133_rinvM(%rsp) ## rsqO - put seed in xmm2 cvtsd2ss %xmm7,%xmm2 rsqrtss %xmm2,%xmm2 cvtss2sd %xmm2,%xmm2 movsd %xmm2,%xmm3 mulsd %xmm2,%xmm2 movsd nb133_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 nb133_half(%rsp),%xmm4 ## iter1 ( new lu) movsd %xmm4,%xmm3 mulsd %xmm4,%xmm4 ## lu*lu mulsd %xmm4,%xmm7 ## rsq*lu*lu movsd nb133_three(%rsp),%xmm4 subsd %xmm7,%xmm4 ## 3-rsq*lu*lu mulsd %xmm3,%xmm4 ## lu*( 3-rsq*lu*lu) mulsd nb133_half(%rsp),%xmm4 ## rinv movsd %xmm4,%xmm7 ## rinvO in xmm7 movsd nb133_rsqO(%rsp),%xmm4 movapd %xmm7,%xmm0 ## LJ table interaction. mulsd %xmm7,%xmm4 ## xmm4=r mulsd nb133_tsc(%rsp),%xmm4 cvttsd2si %xmm4,%ebx ## mm6 = lu idx cvtsi2sd %ebx,%xmm5 subpd %xmm5,%xmm4 movapd %xmm4,%xmm1 ## xmm1=eps movapd %xmm1,%xmm2 mulpd %xmm2,%xmm2 ## xmm2=eps2 shll $3,%ebx movq nb133_VFtab(%rbp),%rsi ## dispersion movlpd (%rsi,%rbx,8),%xmm4 ## Y1 movhpd 8(%rsi,%rbx,8),%xmm4 ## Y1 F1 movapd %xmm4,%xmm5 unpcklpd %xmm3,%xmm4 ## Y1 Y2 unpckhpd %xmm3,%xmm5 ## F1 F2 movlpd 16(%rsi,%rbx,8),%xmm6 ## G1 movhpd 24(%rsi,%rbx,8),%xmm6 ## G1 H1 movapd %xmm6,%xmm7 unpcklpd %xmm3,%xmm6 ## G1 G2 unpckhpd %xmm3,%xmm7 ## H1 H2 ## dispersion table ready, in xmm4-xmm7 mulsd %xmm1,%xmm6 ## xmm6=Geps mulsd %xmm2,%xmm7 ## xmm7=Heps2 addsd %xmm6,%xmm5 addsd %xmm7,%xmm5 ## xmm5=Fp mulsd nb133_two(%rsp),%xmm7 ## two*Heps2 addsd %xmm6,%xmm7 addsd %xmm5,%xmm7 ## xmm7=FF mulsd %xmm1,%xmm5 ## xmm5=eps*Fp addsd %xmm4,%xmm5 ## xmm5=VV movsd nb133_c6(%rsp),%xmm4 mulsd %xmm4,%xmm7 ## fijD mulsd %xmm4,%xmm5 ## Vvdw6 ## put scalar force on stack Update Vvdwtot directly addsd nb133_Vvdwtot(%rsp),%xmm5 xorpd %xmm3,%xmm3 mulsd nb133_tsc(%rsp),%xmm7 subsd %xmm7,%xmm3 movsd %xmm3,nb133_fstmp(%rsp) movsd %xmm5,nb133_Vvdwtot(%rsp) ## repulsion movlpd 32(%rsi,%rbx,8),%xmm4 ## Y1 movhpd 40(%rsi,%rbx,8),%xmm4 ## Y1 F1 movapd %xmm4,%xmm5 unpcklpd %xmm3,%xmm4 ## Y1 Y2 unpckhpd %xmm3,%xmm5 ## F1 F2 movlpd 48(%rsi,%rbx,8),%xmm6 ## G1 movhpd 56(%rsi,%rbx,8),%xmm6 ## G1 H1 movapd %xmm6,%xmm7 unpcklpd %xmm3,%xmm6 ## G1 G2 unpckhpd %xmm3,%xmm7 ## H1 H2 ## table ready, in xmm4-xmm7 mulsd %xmm1,%xmm6 ## xmm6=Geps mulsd %xmm2,%xmm7 ## xmm7=Heps2 addsd %xmm6,%xmm5 addsd %xmm7,%xmm5 ## xmm5=Fp mulsd nb133_two(%rsp),%xmm7 ## two*Heps2 addsd %xmm6,%xmm7 addsd %xmm5,%xmm7 ## xmm7=FF mulsd %xmm1,%xmm5 ## xmm5=eps*Fp addsd %xmm4,%xmm5 ## xmm5=VV movsd nb133_c12(%rsp),%xmm4 mulsd %xmm4,%xmm7 mulsd %xmm4,%xmm5 addsd nb133_Vvdwtot(%rsp),%xmm5 movsd nb133_fstmp(%rsp),%xmm3 mulsd nb133_tsc(%rsp),%xmm7 subsd %xmm7,%xmm3 movsd %xmm5,nb133_Vvdwtot(%rsp) mulsd %xmm0,%xmm3 movsd nb133_dxO(%rsp),%xmm0 movsd nb133_dyO(%rsp),%xmm1 movsd nb133_dzO(%rsp),%xmm2 movq nb133_faction(%rbp),%rdi mulsd %xmm3,%xmm0 mulsd %xmm3,%xmm1 mulsd %xmm3,%xmm2 ## update O forces movapd nb133_fixO(%rsp),%xmm3 movapd nb133_fiyO(%rsp),%xmm4 movapd nb133_fizO(%rsp),%xmm7 addsd %xmm0,%xmm3 addsd %xmm1,%xmm4 addsd %xmm2,%xmm7 movsd %xmm3,nb133_fixO(%rsp) movsd %xmm4,nb133_fiyO(%rsp) movsd %xmm7,nb133_fizO(%rsp) ## update j forces with water O movsd %xmm0,nb133_fjx(%rsp) movsd %xmm1,nb133_fjy(%rsp) movsd %xmm2,nb133_fjz(%rsp) ## H1 interactions movsd nb133_rinvH1(%rsp),%xmm6 movsd %xmm6,%xmm4 mulsd %xmm4,%xmm4 ## xmm6=rinv, xmm4=rinvsq mulsd nb133_qqH(%rsp),%xmm6 ## vcoul mulsd %xmm6,%xmm4 ## fscal addsd nb133_vctot(%rsp),%xmm6 movsd %xmm6,nb133_vctot(%rsp) movapd nb133_dxH1(%rsp),%xmm0 movapd nb133_dyH1(%rsp),%xmm1 movapd nb133_dzH1(%rsp),%xmm2 mulsd %xmm4,%xmm0 mulsd %xmm4,%xmm1 mulsd %xmm4,%xmm2 ## update H1 forces movapd nb133_fixH1(%rsp),%xmm3 movapd nb133_fiyH1(%rsp),%xmm4 movapd nb133_fizH1(%rsp),%xmm7 addsd %xmm0,%xmm3 addsd %xmm1,%xmm4 addsd %xmm2,%xmm7 movsd %xmm3,nb133_fixH1(%rsp) movsd %xmm4,nb133_fiyH1(%rsp) movsd %xmm7,nb133_fizH1(%rsp) ## update j forces with water H1 addsd nb133_fjx(%rsp),%xmm0 addsd nb133_fjy(%rsp),%xmm1 addsd nb133_fjz(%rsp),%xmm2 movsd %xmm0,nb133_fjx(%rsp) movsd %xmm1,nb133_fjy(%rsp) movsd %xmm2,nb133_fjz(%rsp) ## H2 interactions movsd nb133_rinvH2(%rsp),%xmm6 movsd %xmm6,%xmm4 mulsd %xmm4,%xmm4 ## xmm6=rinv, xmm4=rinvsq mulsd nb133_qqH(%rsp),%xmm6 ## vcoul mulsd %xmm6,%xmm4 ## fscal addsd nb133_vctot(%rsp),%xmm6 movsd %xmm6,nb133_vctot(%rsp) movapd nb133_dxH2(%rsp),%xmm0 movapd nb133_dyH2(%rsp),%xmm1 movapd nb133_dzH2(%rsp),%xmm2 mulsd %xmm4,%xmm0 mulsd %xmm4,%xmm1 mulsd %xmm4,%xmm2 ## update H2 forces movapd nb133_fixH2(%rsp),%xmm3 movapd nb133_fiyH2(%rsp),%xmm4 movapd nb133_fizH2(%rsp),%xmm7 addsd %xmm0,%xmm3 addsd %xmm1,%xmm4 addsd %xmm2,%xmm7 movsd %xmm3,nb133_fixH2(%rsp) movsd %xmm4,nb133_fiyH2(%rsp) movsd %xmm7,nb133_fizH2(%rsp) ## update j forces with water H2 addsd nb133_fjx(%rsp),%xmm0 addsd nb133_fjy(%rsp),%xmm1 addsd nb133_fjz(%rsp),%xmm2 movsd %xmm0,nb133_fjx(%rsp) movsd %xmm1,nb133_fjy(%rsp) movsd %xmm2,nb133_fjz(%rsp) ## M interactions movsd nb133_rinvM(%rsp),%xmm6 movsd %xmm6,%xmm4 mulsd %xmm4,%xmm4 ## xmm6=rinv, xmm4=rinvsq mulsd nb133_qqM(%rsp),%xmm6 ## vcoul mulsd %xmm6,%xmm4 ## fscal addsd nb133_vctot(%rsp),%xmm6 movsd %xmm6,nb133_vctot(%rsp) movapd nb133_dxM(%rsp),%xmm0 movapd nb133_dyM(%rsp),%xmm1 movapd nb133_dzM(%rsp),%xmm2 mulsd %xmm4,%xmm0 mulsd %xmm4,%xmm1 mulsd %xmm4,%xmm2 ## update M forces movapd nb133_fixM(%rsp),%xmm3 movapd nb133_fiyM(%rsp),%xmm4 movapd nb133_fizM(%rsp),%xmm7 addsd %xmm0,%xmm3 addsd %xmm1,%xmm4 addsd %xmm2,%xmm7 movsd %xmm3,nb133_fixM(%rsp) movsd %xmm4,nb133_fiyM(%rsp) movsd %xmm7,nb133_fizM(%rsp) movq nb133_faction(%rbp),%rdi ## update j forces addsd nb133_fjx(%rsp),%xmm0 addsd nb133_fjy(%rsp),%xmm1 addsd nb133_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_kernel133_x86_64_sse2.nb133_updateouterdata: movl nb133_ii3(%rsp),%ecx movq nb133_faction(%rbp),%rdi movq nb133_fshift(%rbp),%rsi movl nb133_is3(%rsp),%edx ## accumulate Oi forces in xmm0, xmm1, xmm2 movapd nb133_fixO(%rsp),%xmm0 movapd nb133_fiyO(%rsp),%xmm1 movapd nb133_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 nb133_fixH1(%rsp),%xmm0 movapd nb133_fiyH1(%rsp),%xmm1 movapd nb133_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 nb133_fixH2(%rsp),%xmm0 movapd nb133_fiyH2(%rsp),%xmm1 movapd nb133_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⌨️ 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?