nb_kernel102_x86_64_sse2.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,225 行 · 第 1/5 页
S
2,225 行
mulsd %xmm1,%xmm10 mulsd %xmm2,%xmm11 addsd nb102_vctot(%rsp),%xmm0 addsd %xmm2,%xmm1 addsd %xmm1,%xmm0 movsd %xmm0,nb102_vctot(%rsp) ## move j H1 forces to xmm0-xmm2 movsd 24(%rdi,%rax,8),%xmm0 movsd 32(%rdi,%rax,8),%xmm1 movsd 40(%rdi,%rax,8),%xmm2 movsd %xmm9,%xmm7 movsd %xmm9,%xmm8 movsd %xmm11,%xmm13 movsd %xmm11,%xmm14 movsd %xmm11,%xmm15 movsd %xmm10,%xmm11 movsd %xmm10,%xmm12 mulsd nb102_dxOH1(%rsp),%xmm7 mulsd nb102_dyOH1(%rsp),%xmm8 mulsd nb102_dzOH1(%rsp),%xmm9 mulsd nb102_dxH1H1(%rsp),%xmm10 mulsd nb102_dyH1H1(%rsp),%xmm11 mulsd nb102_dzH1H1(%rsp),%xmm12 mulsd nb102_dxH2H1(%rsp),%xmm13 mulsd nb102_dyH2H1(%rsp),%xmm14 mulsd nb102_dzH2H1(%rsp),%xmm15 addsd %xmm7,%xmm0 addsd %xmm8,%xmm1 addsd %xmm9,%xmm2 addsd nb102_fixO(%rsp),%xmm7 addsd nb102_fiyO(%rsp),%xmm8 addsd nb102_fizO(%rsp),%xmm9 addsd %xmm10,%xmm0 addsd %xmm11,%xmm1 addsd %xmm12,%xmm2 addsd nb102_fixH1(%rsp),%xmm10 addsd nb102_fiyH1(%rsp),%xmm11 addsd nb102_fizH1(%rsp),%xmm12 addsd %xmm13,%xmm0 addsd %xmm14,%xmm1 addsd %xmm15,%xmm2 addsd nb102_fixH2(%rsp),%xmm13 addsd nb102_fiyH2(%rsp),%xmm14 addsd nb102_fizH2(%rsp),%xmm15 movsd %xmm7,nb102_fixO(%rsp) movsd %xmm8,nb102_fiyO(%rsp) movsd %xmm9,nb102_fizO(%rsp) movsd %xmm10,nb102_fixH1(%rsp) movsd %xmm11,nb102_fiyH1(%rsp) movsd %xmm12,nb102_fizH1(%rsp) movsd %xmm13,nb102_fixH2(%rsp) movsd %xmm14,nb102_fiyH2(%rsp) movsd %xmm15,nb102_fizH2(%rsp) ## store back j H1 forces from xmm0-xmm2 movsd %xmm0,24(%rdi,%rax,8) movsd %xmm1,32(%rdi,%rax,8) movsd %xmm2,40(%rdi,%rax,8) ## move j H2 coordinates to local temp variables movsd 48(%rsi,%rax,8),%xmm0 movsd 56(%rsi,%rax,8),%xmm1 movsd 64(%rsi,%rax,8),%xmm2 ## xmm0 = H2x ## xmm1 = H2y ## xmm2 = H2z movsd %xmm0,%xmm3 movsd %xmm1,%xmm4 movsd %xmm2,%xmm5 movsd %xmm0,%xmm6 movsd %xmm1,%xmm7 movsd %xmm2,%xmm8 subsd nb102_ixO(%rsp),%xmm0 subsd nb102_iyO(%rsp),%xmm1 subsd nb102_izO(%rsp),%xmm2 subsd nb102_ixH1(%rsp),%xmm3 subsd nb102_iyH1(%rsp),%xmm4 subsd nb102_izH1(%rsp),%xmm5 subsd nb102_ixH2(%rsp),%xmm6 subsd nb102_iyH2(%rsp),%xmm7 subsd nb102_izH2(%rsp),%xmm8 movsd %xmm0,nb102_dxOH2(%rsp) movsd %xmm1,nb102_dyOH2(%rsp) movsd %xmm2,nb102_dzOH2(%rsp) mulsd %xmm0,%xmm0 mulsd %xmm1,%xmm1 mulsd %xmm2,%xmm2 movsd %xmm3,nb102_dxH1H2(%rsp) movsd %xmm4,nb102_dyH1H2(%rsp) movsd %xmm5,nb102_dzH1H2(%rsp) mulsd %xmm3,%xmm3 mulsd %xmm4,%xmm4 mulsd %xmm5,%xmm5 movsd %xmm6,nb102_dxH2H2(%rsp) movsd %xmm7,nb102_dyH2H2(%rsp) movsd %xmm8,nb102_dzH2H2(%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 jH2 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 nb102_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 nb102_half(%rsp),%xmm15 mulsd %xmm15,%xmm9 ## first iteration for rinvOH2 mulsd %xmm15,%xmm10 ## first iteration for rinvH1H2 mulsd %xmm15,%xmm11 ## first iteration for rinvH2H2 ## 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 nb102_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 nb102_half(%rsp),%xmm15 mulsd %xmm15,%xmm9 ## rinvOH2 mulsd %xmm15,%xmm10 ## rinvH1H2 mulsd %xmm15,%xmm11 ## rinvH2H2 ## H2 interactions movsd %xmm9,%xmm0 movsd %xmm10,%xmm1 movsd %xmm11,%xmm2 mulsd %xmm9,%xmm9 mulsd %xmm10,%xmm10 mulsd %xmm11,%xmm11 mulsd nb102_qqOH(%rsp),%xmm0 mulsd nb102_qqHH(%rsp),%xmm1 mulsd nb102_qqHH(%rsp),%xmm2 mulsd %xmm0,%xmm9 mulsd %xmm1,%xmm10 mulsd %xmm2,%xmm11 addsd nb102_vctot(%rsp),%xmm0 addsd %xmm2,%xmm1 addsd %xmm1,%xmm0 movsd %xmm0,nb102_vctot(%rsp) ## move j H2 forces to xmm0-xmm2 movsd 48(%rdi,%rax,8),%xmm0 movsd 56(%rdi,%rax,8),%xmm1 movsd 64(%rdi,%rax,8),%xmm2 movsd %xmm9,%xmm7 movsd %xmm9,%xmm8 movsd %xmm11,%xmm13 movsd %xmm11,%xmm14 movsd %xmm11,%xmm15 movsd %xmm10,%xmm11 movsd %xmm10,%xmm12 mulsd nb102_dxOH2(%rsp),%xmm7 mulsd nb102_dyOH2(%rsp),%xmm8 mulsd nb102_dzOH2(%rsp),%xmm9 mulsd nb102_dxH1H2(%rsp),%xmm10 mulsd nb102_dyH1H2(%rsp),%xmm11 mulsd nb102_dzH1H2(%rsp),%xmm12 mulsd nb102_dxH2H2(%rsp),%xmm13 mulsd nb102_dyH2H2(%rsp),%xmm14 mulsd nb102_dzH2H2(%rsp),%xmm15 addsd %xmm7,%xmm0 addsd %xmm8,%xmm1 addsd %xmm9,%xmm2 addsd nb102_fixO(%rsp),%xmm7 addsd nb102_fiyO(%rsp),%xmm8 addsd nb102_fizO(%rsp),%xmm9 addsd %xmm10,%xmm0 addsd %xmm11,%xmm1 addsd %xmm12,%xmm2 addsd nb102_fixH1(%rsp),%xmm10 addsd nb102_fiyH1(%rsp),%xmm11 addsd nb102_fizH1(%rsp),%xmm12 addsd %xmm13,%xmm0 addsd %xmm14,%xmm1 addsd %xmm15,%xmm2 addsd nb102_fixH2(%rsp),%xmm13 addsd nb102_fiyH2(%rsp),%xmm14 addsd nb102_fizH2(%rsp),%xmm15 movsd %xmm7,nb102_fixO(%rsp) movsd %xmm8,nb102_fiyO(%rsp) movsd %xmm9,nb102_fizO(%rsp) movsd %xmm10,nb102_fixH1(%rsp) movsd %xmm11,nb102_fiyH1(%rsp) movsd %xmm12,nb102_fizH1(%rsp) movsd %xmm13,nb102_fixH2(%rsp) movsd %xmm14,nb102_fiyH2(%rsp) movsd %xmm15,nb102_fizH2(%rsp) ## store back j H2 forces from xmm0-xmm2 movsd %xmm0,48(%rdi,%rax,8) movsd %xmm1,56(%rdi,%rax,8) movsd %xmm2,64(%rdi,%rax,8)_nb_kernel102_x86_64_sse2.nb102_updateouterdata: movl nb102_ii3(%rsp),%ecx movq nb102_faction(%rbp),%rdi movq nb102_fshift(%rbp),%rsi movl nb102_is3(%rsp),%edx ## accumulate Oi forces in xmm0, xmm1, xmm2 movapd nb102_fixO(%rsp),%xmm0 movapd nb102_fiyO(%rsp),%xmm1 movapd nb102_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 nb102_fixH1(%rsp),%xmm0 movapd nb102_fiyH1(%rsp),%xmm1 movapd nb102_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 nb102_fixH2(%rsp),%xmm0 movapd nb102_fiyH2(%rsp),%xmm1 movapd nb102_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 movapd %xmm0,%xmm3 movapd %xmm1,%xmm4 movapd %xmm2,%xmm5 ## increment i force movsd 48(%rdi,%rcx,8),%xmm3 movsd 56(%rdi,%rcx,8),%xmm4 movsd 64(%rdi,%rcx,8),%xmm5 subsd %xmm0,%xmm3 subsd %xmm1,%xmm4 subsd %xmm2,%xmm5 movsd %xmm3,48(%rdi,%rcx,8) movsd %xmm4,56(%rdi,%rcx,8) movsd %xmm5,64(%rdi,%rcx,8) ## accumulate force in xmm6/xmm7 for fshift addsd %xmm2,%xmm7 unpcklpd %xmm1,%xmm0 addpd %xmm0,%xmm6 ## increment fshift force movlpd (%rsi,%rdx,8),%xmm3 movhpd 8(%rsi,%rdx,8),%xmm3 movsd 16(%rsi,%rdx,8),%xmm4 subpd %xmm6,%xmm3 subsd %xmm7,%xmm4 movlpd %xmm3,(%rsi,%rdx,8) movhpd %xmm3,8(%rsi,%rdx,8) movsd %xmm4,16(%rsi,%rdx,8) ## get n from stack movl nb102_n(%rsp),%esi ## get group index for i particle movq nb102_gid(%rbp),%rdx ## base of gid[] movl (%rdx,%rsi,4),%edx ## ggid=gid[n] ## accumulate total potential energy and update it movapd nb102_vctot(%rsp),%xmm7 ## accumulate movhlps %xmm7,%xmm6 addsd %xmm6,%xmm7 ## pos 0-1 in xmm7 have the sum now ## add earlier value from mem movq nb102_Vc(%rbp),%rax addsd (%rax,%rdx,8),%xmm7 ## move back to mem movsd %xmm7,(%rax,%rdx,8) ## finish if last movl nb102_nn1(%rsp),%ecx ## esi already loaded with n incl %esi subl %esi,%ecx jz _nb_kernel102_x86_64_sse2.nb102_outerend ## not last, iterate outer loop once more! movl %esi,nb102_n(%rsp) jmp _nb_kernel102_x86_64_sse2.nb102_outer_nb_kernel102_x86_64_sse2.nb102_outerend: ## check if more outer neighborlists remain movl nb102_nri(%rsp),%ecx ## esi already loaded with n above subl %esi,%ecx jz _nb_kernel102_x86_64_sse2.nb102_end ## non-zero, do one more workunit jmp _nb_kernel102_x86_64_sse2.nb102_threadloop_nb_kernel102_x86_64_sse2.nb102_end: movl nb102_nouter(%rsp),%eax movl nb102_ninner(%rsp),%ebx movq nb102_outeriter(%rbp),%rcx movq nb102_inneriter(%rbp),%rdx movl %eax,(%rcx) movl %ebx,(%rdx) addq $1496,%rsp emms pop %r15 pop %r14 pop %r13 pop %r12 pop %rbx pop %rbp ret.globl nb_kernel102nf_x86_64_sse2.globl _nb_kernel102nf_x86_64_sse2nb_kernel102nf_x86_64_sse2: _nb_kernel102nf_x86_64_sse2: .set nb102nf_fshift, 16.set nb102nf_gid, 24.set nb102nf_pos, 32.set nb102nf_faction, 40.set nb102nf_charge, 48.set nb102nf_p_facel, 56.set nb102nf_argkrf, 64.set nb102nf_argcrf, 72.set nb102nf_Vc, 80.set nb102nf_type, 88.set nb102nf_p_ntype, 96.set nb102nf_vdwparam, 104.set nb102nf_Vvdw, 112⌨️ 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?