nb_kernel312_x86_64_sse2.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,364 行 · 第 1/5 页
S
2,364 行
jmp _nb_kernel312_x86_64_sse2.nb312_unroll_loop_nb_kernel312_x86_64_sse2.nb312_checksingle: movl nb312_innerk(%rsp),%edx andl $1,%edx jnz _nb_kernel312_x86_64_sse2.nb312_dosingle jmp _nb_kernel312_x86_64_sse2.nb312_updateouterdata_nb_kernel312_x86_64_sse2.nb312_dosingle: movq nb312_innerjjnr(%rsp),%rdx ## pointer to jjnr[k] movl (%rdx),%eax movq nb312_pos(%rbp),%rsi lea (%rax,%rax,2),%rax ## move j O coordinates to local temp variables movsd (%rsi,%rax,8),%xmm0 movsd 8(%rsi,%rax,8),%xmm1 movsd 16(%rsi,%rax,8),%xmm2 ## xmm0 = Ox ## xmm1 = Oy ## xmm2 = Oz movsd %xmm0,%xmm3 movsd %xmm1,%xmm4 movsd %xmm2,%xmm5 movsd %xmm0,%xmm6 movsd %xmm1,%xmm7 movsd %xmm2,%xmm8 subsd nb312_ixO(%rsp),%xmm0 subsd nb312_iyO(%rsp),%xmm1 subsd nb312_izO(%rsp),%xmm2 subsd nb312_ixH1(%rsp),%xmm3 subsd nb312_iyH1(%rsp),%xmm4 subsd nb312_izH1(%rsp),%xmm5 subsd nb312_ixH2(%rsp),%xmm6 subsd nb312_iyH2(%rsp),%xmm7 subsd nb312_izH2(%rsp),%xmm8 movsd %xmm0,nb312_dxOO(%rsp) movsd %xmm1,nb312_dyOO(%rsp) movsd %xmm2,nb312_dzOO(%rsp) mulsd %xmm0,%xmm0 mulsd %xmm1,%xmm1 mulsd %xmm2,%xmm2 movsd %xmm3,nb312_dxH1O(%rsp) movsd %xmm4,nb312_dyH1O(%rsp) movsd %xmm5,nb312_dzH1O(%rsp) mulsd %xmm3,%xmm3 mulsd %xmm4,%xmm4 mulsd %xmm5,%xmm5 movsd %xmm6,nb312_dxH2O(%rsp) movsd %xmm7,nb312_dyH2O(%rsp) movsd %xmm8,nb312_dzH2O(%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 jO 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 nb312_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 nb312_half(%rsp),%xmm15 mulsd %xmm15,%xmm9 ## first iteration for rinvOO mulsd %xmm15,%xmm10 ## first iteration for rinvH1O mulsd %xmm15,%xmm11 ## first iteration for rinvH2O ## 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 nb312_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 nb312_half(%rsp),%xmm15 mulsd %xmm15,%xmm9 ## rinvOO mulsd %xmm15,%xmm10 ## rinvH1O mulsd %xmm15,%xmm11 ## rinvH2O movsd %xmm9,nb312_rinvOO(%rsp) movsd %xmm10,nb312_rinvH1O(%rsp) movsd %xmm11,nb312_rinvH2O(%rsp) ## O interactions ## rsq in xmm0,xmm3,xmm6 ## rinv in xmm9, xmm10, xmm11 movsd nb312_tsc(%rsp),%xmm1 mulsd %xmm9,%xmm0 ## r mulsd %xmm10,%xmm3 mulsd %xmm11,%xmm6 mulsd %xmm1,%xmm0 ## rtab mulsd %xmm1,%xmm3 mulsd %xmm1,%xmm6 ## truncate and convert to integers cvttsd2si %xmm0,%r8d cvttsd2si %xmm3,%r10d cvttsd2si %xmm6,%r12d ## convert back to float cvtsi2sd %r8d,%xmm2 cvtsi2sd %r10d,%xmm5 cvtsi2sd %r12d,%xmm8 ## multiply by 4 shlq $2,%r8 shll $2,%r10d shll $2,%r12d movq nb312_VFtab(%rbp),%rsi ## calculate eps subsd %xmm2,%xmm0 subsd %xmm5,%xmm3 subsd %xmm8,%xmm6 movsd %xmm0,nb312_epsO(%rsp) movsd %xmm3,nb312_epsH1(%rsp) movsd %xmm6,nb312_epsH2(%rsp) ## Load LOTS of table data movsd (%rsi,%r8,8),%xmm0 movsd 8(%rsi,%r8,8),%xmm1 movsd 16(%rsi,%r8,8),%xmm2 movsd 24(%rsi,%r8,8),%xmm3 movsd (%rsi,%r10,8),%xmm4 movsd 8(%rsi,%r10,8),%xmm5 movsd 16(%rsi,%r10,8),%xmm6 movsd 24(%rsi,%r10,8),%xmm7 movsd (%rsi,%r12,8),%xmm8 movsd 8(%rsi,%r12,8),%xmm9 movsd 16(%rsi,%r12,8),%xmm10 movsd 24(%rsi,%r12,8),%xmm11 ## table data ready in xmm0-xmm3 , xmm4-xmm7 , and xmm8-xmm11 mulsd nb312_epsO(%rsp),%xmm3 ## Heps mulsd nb312_epsH1(%rsp),%xmm7 mulsd nb312_epsH2(%rsp),%xmm11 mulsd nb312_epsO(%rsp),%xmm2 ## Geps mulsd nb312_epsH1(%rsp),%xmm6 mulsd nb312_epsH2(%rsp),%xmm10 mulsd nb312_epsO(%rsp),%xmm3 ## Heps2 mulsd nb312_epsH1(%rsp),%xmm7 mulsd nb312_epsH2(%rsp),%xmm11 addsd %xmm2,%xmm1 ## F+Geps addsd %xmm6,%xmm5 addsd %xmm10,%xmm9 addsd %xmm3,%xmm1 ## F+Geps+Heps2 = Fp addsd %xmm7,%xmm5 addsd %xmm11,%xmm9 addsd %xmm3,%xmm3 ## 2*Heps2 addsd %xmm7,%xmm7 addsd %xmm11,%xmm11 addsd %xmm2,%xmm3 ## 2*Heps2+Geps addsd %xmm6,%xmm7 addsd %xmm10,%xmm11 addsd %xmm1,%xmm3 ## FF = Fp + 2*Heps2 + Geps addsd %xmm5,%xmm7 addsd %xmm9,%xmm11 mulsd nb312_epsO(%rsp),%xmm1 ## eps*Fp mulsd nb312_epsH1(%rsp),%xmm5 mulsd nb312_epsH2(%rsp),%xmm9 addsd %xmm0,%xmm1 ## VV addsd %xmm4,%xmm5 addsd %xmm8,%xmm9 mulsd nb312_qqOO(%rsp),%xmm1 ## VV*qq = vcoul mulsd nb312_qqOH(%rsp),%xmm5 mulsd nb312_qqOH(%rsp),%xmm9 mulsd nb312_qqOO(%rsp),%xmm3 ## FF*qq = fij mulsd nb312_qqOH(%rsp),%xmm7 mulsd nb312_qqOH(%rsp),%xmm11 ## calculate LJ movsd nb312_rinvOO(%rsp),%xmm12 mulsd %xmm12,%xmm12 ## rinvsq movsd %xmm12,%xmm13 ## rinvsq mulsd %xmm12,%xmm12 ## rinv4 mulsd %xmm13,%xmm12 ## rinv6 movsd %xmm12,%xmm13 ## rinv6 mulsd %xmm12,%xmm12 ## rinv12 mulsd nb312_c6(%rsp),%xmm13 mulsd nb312_c12(%rsp),%xmm12 movsd %xmm12,%xmm14 subsd %xmm13,%xmm14 addsd nb312_Vvdwtot(%rsp),%xmm14 mulsd nb312_six(%rsp),%xmm13 mulsd nb312_twelve(%rsp),%xmm12 movsd %xmm14,nb312_Vvdwtot(%rsp) subsd %xmm13,%xmm12 ## LJ fscal mulsd nb312_rinvOO(%rsp),%xmm12 movapd %xmm12,%xmm4 ## accumulate vctot addsd nb312_vctot(%rsp),%xmm1 addsd %xmm9,%xmm5 addsd %xmm5,%xmm1 movsd %xmm1,nb312_vctot(%rsp) xorpd %xmm8,%xmm8 xorpd %xmm12,%xmm12 movsd nb312_tsc(%rsp),%xmm5 mulsd %xmm5,%xmm3 mulsd %xmm5,%xmm7 mulsd %xmm5,%xmm11 subpd %xmm3,%xmm4 subpd %xmm7,%xmm8 subpd %xmm11,%xmm12 mulsd nb312_rinvOO(%rsp),%xmm4 mulsd nb312_rinvH1O(%rsp),%xmm8 mulsd nb312_rinvH2O(%rsp),%xmm12 ## move j O forces to xmm0-xmm2 movq nb312_faction(%rbp),%rdi movsd (%rdi,%rax,8),%xmm0 movsd 8(%rdi,%rax,8),%xmm1 movsd 16(%rdi,%rax,8),%xmm2 movapd %xmm4,%xmm3 movapd %xmm4,%xmm5 movapd %xmm8,%xmm7 movapd %xmm8,%xmm9 movapd %xmm12,%xmm10 movapd %xmm12,%xmm11 mulsd nb312_dxOO(%rsp),%xmm3 mulsd nb312_dyOO(%rsp),%xmm4 mulsd nb312_dzOO(%rsp),%xmm5 mulsd nb312_dxH1O(%rsp),%xmm7 mulsd nb312_dyH1O(%rsp),%xmm8 mulsd nb312_dzH1O(%rsp),%xmm9 mulsd nb312_dxH2O(%rsp),%xmm10 mulsd nb312_dyH2O(%rsp),%xmm11 mulsd nb312_dzH2O(%rsp),%xmm12 addsd %xmm3,%xmm0 addsd %xmm4,%xmm1 addsd %xmm5,%xmm2 addsd nb312_fixO(%rsp),%xmm3 addsd nb312_fiyO(%rsp),%xmm4 addsd nb312_fizO(%rsp),%xmm5 addsd %xmm7,%xmm0 addsd %xmm8,%xmm1 addsd %xmm9,%xmm2 addsd nb312_fixH1(%rsp),%xmm7 addsd nb312_fiyH1(%rsp),%xmm8 addsd nb312_fizH1(%rsp),%xmm9 addsd %xmm10,%xmm0 addsd %xmm11,%xmm1 addsd %xmm12,%xmm2 addsd nb312_fixH2(%rsp),%xmm10 addsd nb312_fiyH2(%rsp),%xmm11 addsd nb312_fizH2(%rsp),%xmm12 movsd %xmm3,nb312_fixO(%rsp) movsd %xmm4,nb312_fiyO(%rsp) movsd %xmm5,nb312_fizO(%rsp) movsd %xmm7,nb312_fixH1(%rsp) movsd %xmm8,nb312_fiyH1(%rsp) movsd %xmm9,nb312_fizH1(%rsp) movsd %xmm10,nb312_fixH2(%rsp) movsd %xmm11,nb312_fiyH2(%rsp) movsd %xmm12,nb312_fizH2(%rsp) ## store back j O forces from xmm0-xmm2 movsd %xmm0,(%rdi,%rax,8) movsd %xmm1,8(%rdi,%rax,8) movsd %xmm2,16(%rdi,%rax,8) ## move j H1 coordinates to local temp variables movq nb312_pos(%rbp),%rsi 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 nb312_ixO(%rsp),%xmm0 subsd nb312_iyO(%rsp),%xmm1 subsd nb312_izO(%rsp),%xmm2 subsd nb312_ixH1(%rsp),%xmm3 subsd nb312_iyH1(%rsp),%xmm4 subsd nb312_izH1(%rsp),%xmm5 subsd nb312_ixH2(%rsp),%xmm6 subsd nb312_iyH2(%rsp),%xmm7 subsd nb312_izH2(%rsp),%xmm8 movsd %xmm0,nb312_dxOH1(%rsp) movsd %xmm1,nb312_dyOH1(%rsp) movsd %xmm2,nb312_dzOH1(%rsp) mulsd %xmm0,%xmm0 mulsd %xmm1,%xmm1 mulsd %xmm2,%xmm2 movsd %xmm3,nb312_dxH1H1(%rsp) movsd %xmm4,nb312_dyH1H1(%rsp) movsd %xmm5,nb312_dzH1H1(%rsp) mulsd %xmm3,%xmm3 mulsd %xmm4,%xmm4 mulsd %xmm5,%xmm5 movsd %xmm6,nb312_dxH2H1(%rsp) movsd %xmm7,nb312_dyH2H1(%rsp) movsd %xmm8,nb312_dzH2H1(%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 nb312_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 nb312_half(%rsp),%xmm15 mulsd %xmm15,%xmm9 ## first iteration for rinvOH1 mulsd %xmm15,%xmm10 ## first iteration for rinvH1H1 mulsd %xmm15,%xmm11 ## first iteration for rinvH2OH1 ## 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 nb312_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 nb312_half(%rsp),%xmm15 mulsd %xmm15,%xmm9 ## rinvOH1 mulsd %xmm15,%xmm10 ## rinvH1H1 mulsd %xmm15,%xmm11 ## rinvH2H1 movsd %xmm9,nb312_rinvOH1(%rsp) movsd %xmm10,nb312_rinvH1H1(%rsp) movsd %xmm11,nb312_rinvH2H1(%rsp) ## H1 interactions ## rsq in xmm0,xmm3,xmm6 ## rinv in xmm9, xmm10, xmm11 movsd nb312_tsc(%rsp),%xmm1 mulsd %xmm9,%xmm0 ## r mulsd %xmm10,%xmm3 mulsd %xmm11,%xmm6 mulsd %xmm1,%xmm0 ## rtab mulsd %xmm1,%xmm3 mulsd %xmm1,%xmm6 ## truncate and convert to integers cvttsd2si %xmm0,%r8d cvttsd2si %xmm3,%r10d⌨️ 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?