nb_kernel131_x86_64_sse2.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,044 行 · 第 1/4 页
S
2,044 行
movd %xmm1,%r8d movd %xmm13,%r10d ## calculate eps subpd %xmm2,%xmm0 movq nb131_VFtab(%rbp),%rsi movlpd (%rsi,%r8,8),%xmm4 movlpd 8(%rsi,%r8,8),%xmm5 movlpd 16(%rsi,%r8,8),%xmm6 movlpd 24(%rsi,%r8,8),%xmm7 movlpd 32(%rsi,%r8,8),%xmm8 movlpd 40(%rsi,%r8,8),%xmm9 movlpd 48(%rsi,%r8,8),%xmm10 movlpd 56(%rsi,%r8,8),%xmm11 movhpd (%rsi,%r10,8),%xmm4 movhpd 8(%rsi,%r10,8),%xmm5 movhpd 16(%rsi,%r10,8),%xmm6 movhpd 24(%rsi,%r10,8),%xmm7 movhpd 32(%rsi,%r10,8),%xmm8 movhpd 40(%rsi,%r10,8),%xmm9 movhpd 48(%rsi,%r10,8),%xmm10 movhpd 56(%rsi,%r10,8),%xmm11 ## dispersion table in xmm4-xmm7, repulsion table in xmm8-xmm11 mulpd %xmm0,%xmm7 ## Heps mulpd %xmm0,%xmm11 mulpd %xmm0,%xmm6 ## Geps mulpd %xmm0,%xmm10 mulpd %xmm0,%xmm7 ## Heps2 mulpd %xmm0,%xmm11 addpd %xmm6,%xmm5 ## F+Geps addpd %xmm10,%xmm9 addpd %xmm7,%xmm5 ## F+Geps+Heps2 = Fp addpd %xmm11,%xmm9 addpd %xmm7,%xmm7 ## 2*Heps2 addpd %xmm11,%xmm11 addpd %xmm6,%xmm7 ## 2*Heps2+Geps addpd %xmm10,%xmm11 addpd %xmm5,%xmm7 ## FF = Fp + 2*Heps2 + Geps addpd %xmm9,%xmm11 mulpd %xmm0,%xmm5 ## eps*Fp mulpd %xmm0,%xmm9 movapd nb131_c6(%rsp),%xmm12 movapd nb131_c12(%rsp),%xmm13 addpd %xmm4,%xmm5 ## VV addpd %xmm8,%xmm9 mulpd %xmm12,%xmm5 ## VV*c6 = vnb6 mulpd %xmm13,%xmm9 ## VV*c12 = vnb12 addpd %xmm9,%xmm5 addpd nb131_Vvdwtot(%rsp),%xmm5 movapd %xmm5,nb131_Vvdwtot(%rsp) mulpd %xmm12,%xmm7 ## FF*c6 = fnb6 mulpd %xmm13,%xmm11 ## FF*c12 = fnb12 addpd %xmm11,%xmm7 mulpd nb131_tsc(%rsp),%xmm7 movapd nb131_rinvO(%rsp),%xmm9 movapd nb131_rinvH1(%rsp),%xmm10 movapd nb131_rinvH2(%rsp),%xmm11 movapd %xmm9,%xmm0 movapd %xmm10,%xmm1 movapd %xmm11,%xmm2 mulpd %xmm10,%xmm10 mulpd %xmm11,%xmm11 mulpd nb131_qqO(%rsp),%xmm0 mulpd nb131_qqH(%rsp),%xmm1 mulpd nb131_qqH(%rsp),%xmm2 mulpd %xmm0,%xmm9 mulpd %xmm1,%xmm10 mulpd %xmm2,%xmm11 subpd %xmm7,%xmm9 mulpd nb131_rinvO(%rsp),%xmm9 addpd nb131_vctot(%rsp),%xmm0 addpd %xmm2,%xmm1 addpd %xmm1,%xmm0 movapd %xmm0,nb131_vctot(%rsp) ## move j forces to xmm0-xmm2 movq nb131_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 nb131_dxO(%rsp),%xmm7 mulpd nb131_dyO(%rsp),%xmm8 mulpd nb131_dzO(%rsp),%xmm9 mulpd nb131_dxH1(%rsp),%xmm10 mulpd nb131_dyH1(%rsp),%xmm11 mulpd nb131_dzH1(%rsp),%xmm12 mulpd nb131_dxH2(%rsp),%xmm13 mulpd nb131_dyH2(%rsp),%xmm14 mulpd nb131_dzH2(%rsp),%xmm15 addpd %xmm7,%xmm0 addpd %xmm8,%xmm1 addpd %xmm9,%xmm2 addpd nb131_fixO(%rsp),%xmm7 addpd nb131_fiyO(%rsp),%xmm8 addpd nb131_fizO(%rsp),%xmm9 addpd %xmm10,%xmm0 addpd %xmm11,%xmm1 addpd %xmm12,%xmm2 addpd nb131_fixH1(%rsp),%xmm10 addpd nb131_fiyH1(%rsp),%xmm11 addpd nb131_fizH1(%rsp),%xmm12 addpd %xmm13,%xmm0 addpd %xmm14,%xmm1 addpd %xmm15,%xmm2 addpd nb131_fixH2(%rsp),%xmm13 addpd nb131_fiyH2(%rsp),%xmm14 addpd nb131_fizH2(%rsp),%xmm15 movapd %xmm7,nb131_fixO(%rsp) movapd %xmm8,nb131_fiyO(%rsp) movapd %xmm9,nb131_fizO(%rsp) movapd %xmm10,nb131_fixH1(%rsp) movapd %xmm11,nb131_fiyH1(%rsp) movapd %xmm12,nb131_fizH1(%rsp) movapd %xmm13,nb131_fixH2(%rsp) movapd %xmm14,nb131_fiyH2(%rsp) movapd %xmm15,nb131_fizH2(%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,nb131_innerk(%rsp) jl _nb_kernel131_x86_64_sse2.nb131_checksingle jmp _nb_kernel131_x86_64_sse2.nb131_unroll_loop_nb_kernel131_x86_64_sse2.nb131_checksingle: movl nb131_innerk(%rsp),%edx andl $1,%edx jnz _nb_kernel131_x86_64_sse2.nb131_dosingle jmp _nb_kernel131_x86_64_sse2.nb131_updateouterdata_nb_kernel131_x86_64_sse2.nb131_dosingle: movq nb131_innerjjnr(%rsp),%rdx ## pointer to jjnr[k] movl (%rdx),%eax addq $8,nb131_innerjjnr(%rsp) ## advance pointer (unrolled 2) movq nb131_charge(%rbp),%rsi ## base of charge[] movsd (%rsi,%rax,8),%xmm3 movapd %xmm3,%xmm4 mulsd nb131_iqO(%rsp),%xmm3 mulsd nb131_iqH(%rsp),%xmm4 movapd %xmm3,nb131_qqO(%rsp) movapd %xmm4,nb131_qqH(%rsp) movq nb131_type(%rbp),%rsi movl (%rsi,%rax,4),%r8d movq nb131_vdwparam(%rbp),%rsi shll %r8d movl nb131_ntia(%rsp),%edi addl %edi,%r8d movsd (%rsi,%r8,8),%xmm6 ## c6a movsd 8(%rsi,%r8,8),%xmm7 ## c12a movapd %xmm6,nb131_c6(%rsp) movapd %xmm7,nb131_c12(%rsp) movq nb131_pos(%rbp),%rsi ## base of pos[] lea (%rax,%rax,2),%rax ## replace jnr with j3 ## move j coordinates to local temp variables movsd (%rsi,%rax,8),%xmm0 movsd 8(%rsi,%rax,8),%xmm1 movsd 16(%rsi,%rax,8),%xmm2 ## xmm0 = jx ## xmm1 = jy ## xmm2 = jz movapd %xmm0,%xmm3 movapd %xmm1,%xmm4 movapd %xmm2,%xmm5 movapd %xmm0,%xmm6 movapd %xmm1,%xmm7 movapd %xmm2,%xmm8 subsd nb131_ixO(%rsp),%xmm0 subsd nb131_iyO(%rsp),%xmm1 subsd nb131_izO(%rsp),%xmm2 subsd nb131_ixH1(%rsp),%xmm3 subsd nb131_iyH1(%rsp),%xmm4 subsd nb131_izH1(%rsp),%xmm5 subsd nb131_ixH2(%rsp),%xmm6 subsd nb131_iyH2(%rsp),%xmm7 subsd nb131_izH2(%rsp),%xmm8 movapd %xmm0,nb131_dxO(%rsp) movapd %xmm1,nb131_dyO(%rsp) movapd %xmm2,nb131_dzO(%rsp) mulsd %xmm0,%xmm0 mulsd %xmm1,%xmm1 mulsd %xmm2,%xmm2 movapd %xmm3,nb131_dxH1(%rsp) movapd %xmm4,nb131_dyH1(%rsp) movapd %xmm5,nb131_dzH1(%rsp) mulsd %xmm3,%xmm3 mulsd %xmm4,%xmm4 mulsd %xmm5,%xmm5 movapd %xmm6,nb131_dxH2(%rsp) movapd %xmm7,nb131_dyH2(%rsp) movapd %xmm8,nb131_dzH2(%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 j 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 movapd %xmm1,%xmm2 movapd %xmm4,%xmm5 movapd %xmm7,%xmm8 mulsd %xmm1,%xmm1 ## lu*lu mulsd %xmm4,%xmm4 ## lu*lu mulsd %xmm7,%xmm7 ## lu*lu movapd nb131_three(%rsp),%xmm9 movapd %xmm9,%xmm10 movapd %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) movapd nb131_half(%rsp),%xmm15 mulsd %xmm15,%xmm9 ## first iteration for rinvO mulsd %xmm15,%xmm10 ## first iteration for rinvH1 mulsd %xmm15,%xmm11 ## first iteration for rinvH2 ## second iteration step movapd %xmm9,%xmm2 movapd %xmm10,%xmm5 movapd %xmm11,%xmm8 mulsd %xmm2,%xmm2 ## lu*lu mulsd %xmm5,%xmm5 ## lu*lu mulsd %xmm8,%xmm8 ## lu*lu movapd nb131_three(%rsp),%xmm1 movapd %xmm1,%xmm4 movapd %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) movapd nb131_half(%rsp),%xmm15 mulsd %xmm15,%xmm9 ## rinvO mulsd %xmm15,%xmm10 ## rinvH1 mulsd %xmm15,%xmm11 ## rinvH2 ## O interactions ## rsq in xmm0,xmm3,xmm6 ## rinv in xmm9, xmm10, xmm11 movapd %xmm0,nb131_rsqO(%rsp) movapd %xmm3,nb131_rsqH1(%rsp) movapd %xmm6,nb131_rsqH2(%rsp) movapd %xmm9,nb131_rinvO(%rsp) movapd %xmm10,nb131_rinvH1(%rsp) movapd %xmm11,nb131_rinvH2(%rsp) ## table LJ interaction mulsd %xmm9,%xmm0 mulsd nb131_tsc(%rsp),%xmm0 ## rtab ## truncate and convert to integers cvttsd2si %xmm0,%r8d ## convert back to float cvtsi2sd %r8d,%xmm2 ## mult. by 8 shll $3,%r8d ## calculate eps subsd %xmm2,%xmm0 movq nb131_VFtab(%rbp),%rsi movsd (%rsi,%r8,8),%xmm4 movsd 8(%rsi,%r8,8),%xmm5 movsd 16(%rsi,%r8,8),%xmm6 movsd 24(%rsi,%r8,8),%xmm7 movsd 32(%rsi,%r8,8),%xmm8 movsd 40(%rsi,%r8,8),%xmm9 movsd 48(%rsi,%r8,8),%xmm10 movsd 56(%rsi,%r8,8),%xmm11 ## dispersion table in xmm4-xmm7, repulsion table in xmm8-xmm11 mulsd %xmm0,%xmm7 ## Heps mulsd %xmm0,%xmm11 mulsd %xmm0,%xmm6 ## Geps mulsd %xmm0,%xmm10 mulsd %xmm0,%xmm7 ## Heps2 mulsd %xmm0,%xmm11 addsd %xmm6,%xmm5 ## F+Geps addsd %xmm10,%xmm9 addsd %xmm7,%xmm5 ## F+Geps+Heps2 = Fp addsd %xmm11,%xmm9 addsd %xmm7,%xmm7 ## 2*Heps2 addsd %xmm11,%xmm11 addsd %xmm6,%xmm7 ## 2*Heps2+Geps addsd %xmm10,%xmm11 addsd %xmm5,%xmm7 ## FF = Fp + 2*Heps2 + Geps addsd %xmm9,%xmm11 mulsd %xmm0,%xmm5 ## eps*Fp mulsd %xmm0,%xmm9 movapd nb131_c6(%rsp),%xmm12 movapd nb131_c12(%rsp),%xmm13 addsd %xmm4,%xmm5 ## VV addsd %xmm8,%xmm9 mulsd %xmm12,%xmm5 ## VV*c6 = vnb6 mulsd %xmm13,%xmm9 ## VV*c12 = vnb12 addsd %xmm9,%xmm5 addsd nb131_Vvdwtot(%rsp),%xmm5 movsd %xmm5,nb131_Vvdwtot(%rsp) mulsd %xmm12,%xmm7 ## FF*c6 = fnb6 mulsd %xmm13,%xmm11 ## FF*c12 = fnb12 addsd %xmm11,%xmm7 mulsd nb131_tsc(%rsp),%xmm7 movapd nb131_rinvO(%rsp),%xmm9 movapd nb131_rinvH1(%rsp),%xmm10 movapd nb131_rinvH2(%rsp),%xmm11 movapd %xmm9,%xmm0 movapd %xmm10,%xmm1 movapd %xmm11,%xmm2 mulsd %xmm10,%xmm10 mulsd %xmm11,%xmm11 mulsd nb131_qqO(%rsp),%xmm0 mulsd nb131_qqH(%rsp),%xmm1 mulsd nb131_qqH(%rsp),%xmm2 mulsd %xmm0,%xmm9 mulsd %xmm1,%xmm10 mulsd %xmm2,%xmm11 subsd %xmm7,%xmm9 mulsd nb131_rinvO(%rsp),%xmm9 addsd nb131_vctot(%rsp),%xmm0 addsd %xmm2,%xmm1 addsd %xmm1,%xmm0 movsd %xmm0,nb131_vctot(%rsp) ## move j forces to xmm0-xmm2 movq nb131_faction(%rbp),%rdi movsd (%rdi,%rax,8),%xmm0 movsd 8(%rdi,%rax,8),%xmm1 movsd 16(%rdi,%rax,8),%xmm2 movapd %xmm9,%xmm7 movapd %xmm9,%xmm8 movapd %xmm11,%xmm13 movapd %xmm11,%xmm14 movapd %xmm11,%xmm15 movapd %xmm10,%xmm11 movapd %xmm10,%xmm12 mulsd nb131_dxO(%rsp),%xmm7 mulsd nb131_dyO(%rsp),%xmm8 mulsd nb131_dzO(%rsp),%xmm9 mulsd nb131_dxH1(%rsp),%xmm10 mulsd nb131_dyH1(%rsp),%xmm11 mulsd nb131_dzH1(%rsp),%xmm12 mulsd nb131_dxH2(%rsp),%xmm13 mulsd nb131_dyH2(%rsp),%xmm14 mulsd nb131_dzH2(%rsp),%xmm15 addsd %xmm7,%xmm0 addsd %xmm8,%xmm1 addsd %xmm9,%xmm2 addsd nb131_fixO(%rsp),%xmm7 addsd nb131_fiyO(%rsp),%xmm8 addsd nb131_fizO(%rsp),%xmm9 addsd %xmm10,%xmm0 addsd %xmm11,%xmm1 addsd %xmm12,%xmm2 addsd nb131_fixH1(%rsp),%xmm10 addsd nb131_fiyH1(%rsp),%xmm11 addsd nb131_fizH1(%rsp),%xmm12 addsd %xmm13,%xmm0 addsd %xmm14,%xmm1 addsd %xmm15,%xmm2 addsd nb131_fixH2(%rsp),%xmm13 addsd nb131_fiyH2(%rsp),%xmm14 addsd nb131_fizH2(%rsp),%xmm15 movsd %xmm7,nb131_fixO(%rsp) movsd %xmm8,nb131_fiyO(%rsp) movsd %xmm9,nb131_fizO(%rsp) movsd %xmm10,nb131_fixH1(%rsp) movsd %xmm11,nb131_fiyH1(%rsp) movsd %xmm12,nb131_fizH1(%rsp) movsd %xmm13,nb131_fixH2(%rsp) movsd %xmm14,nb131_fiyH2(%rsp) movsd %xmm15,nb131_fizH2(%rsp) ## store back j forces from xmm0-xmm2 movsd %xmm0,(%rdi,%rax,8) movsd %xmm1,8(%rdi,%rax,8) movsd %xmm2,16(%rdi,%rax,8)_nb_kernel131_x86_64_sse2.nb131_updateouterdata: movl nb131_ii3(%rsp),%ecx movq nb131_faction(%rbp),%rdi movq nb131_fshift(%rbp),%rsi movl nb131_is3(%rsp),%edx ## accumulate Oi forces in xmm0, xmm1, xmm2 movapd nb131_fixO(%rsp),%xmm0 movapd nb131_fiyO(%rsp),%xmm1 movapd nb131_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 movapd %xmm0,%xmm3 movapd %xmm1,%xmm4 movapd %xmm2,%xmm5 ## 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⌨️ 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?