nb_kernel102_x86_64_sse2.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,311 行 · 第 1/5 页
S
2,311 行
mulsd xmm15, [rsp + nb102_dzH2H1] addsd xmm0, xmm7 addsd xmm1, xmm8 addsd xmm2, xmm9 addsd xmm7, [rsp + nb102_fixO] addsd xmm8, [rsp + nb102_fiyO] addsd xmm9, [rsp + nb102_fizO] addsd xmm0, xmm10 addsd xmm1, xmm11 addsd xmm2, xmm12 addsd xmm10, [rsp + nb102_fixH1] addsd xmm11, [rsp + nb102_fiyH1] addsd xmm12, [rsp + nb102_fizH1] addsd xmm0, xmm13 addsd xmm1, xmm14 addsd xmm2, xmm15 addsd xmm13, [rsp + nb102_fixH2] addsd xmm14, [rsp + nb102_fiyH2] addsd xmm15, [rsp + nb102_fizH2] movsd [rsp + nb102_fixO], xmm7 movsd [rsp + nb102_fiyO], xmm8 movsd [rsp + nb102_fizO], xmm9 movsd [rsp + nb102_fixH1], xmm10 movsd [rsp + nb102_fiyH1], xmm11 movsd [rsp + nb102_fizH1], xmm12 movsd [rsp + nb102_fixH2], xmm13 movsd [rsp + nb102_fiyH2], xmm14 movsd [rsp + nb102_fizH2], xmm15 ;# store back j H1 forces from xmm0-xmm2 movsd [rdi + rax*8 + 24], xmm0 movsd [rdi + rax*8 + 32], xmm1 movsd [rdi + rax*8 + 40], xmm2 ;# move j H2 coordinates to local temp variables movsd xmm0, [rsi + rax*8 + 48] movsd xmm1, [rsi + rax*8 + 56] movsd xmm2, [rsi + rax*8 + 64] ;# xmm0 = H2x ;# xmm1 = H2y ;# xmm2 = H2z movsd xmm3, xmm0 movsd xmm4, xmm1 movsd xmm5, xmm2 movsd xmm6, xmm0 movsd xmm7, xmm1 movsd xmm8, xmm2 subsd xmm0, [rsp + nb102_ixO] subsd xmm1, [rsp + nb102_iyO] subsd xmm2, [rsp + nb102_izO] subsd xmm3, [rsp + nb102_ixH1] subsd xmm4, [rsp + nb102_iyH1] subsd xmm5, [rsp + nb102_izH1] subsd xmm6, [rsp + nb102_ixH2] subsd xmm7, [rsp + nb102_iyH2] subsd xmm8, [rsp + nb102_izH2] movsd [rsp + nb102_dxOH2], xmm0 movsd [rsp + nb102_dyOH2], xmm1 movsd [rsp + nb102_dzOH2], xmm2 mulsd xmm0, xmm0 mulsd xmm1, xmm1 mulsd xmm2, xmm2 movsd [rsp + nb102_dxH1H2], xmm3 movsd [rsp + nb102_dyH1H2], xmm4 movsd [rsp + nb102_dzH1H2], xmm5 mulsd xmm3, xmm3 mulsd xmm4, xmm4 mulsd xmm5, xmm5 movsd [rsp + nb102_dxH2H2], xmm6 movsd [rsp + nb102_dyH2H2], xmm7 movsd [rsp + nb102_dzH2H2], xmm8 mulsd xmm6, xmm6 mulsd xmm7, xmm7 mulsd xmm8, xmm8 addsd xmm0, xmm1 addsd xmm0, xmm2 addsd xmm3, xmm4 addsd xmm3, xmm5 addsd xmm6, xmm7 addsd xmm6, xmm8 ;# start doing invsqrt for jH2 atoms cvtsd2ss xmm1, xmm0 cvtsd2ss xmm4, xmm3 cvtsd2ss xmm7, xmm6 rsqrtss xmm1, xmm1 rsqrtss xmm4, xmm4 rsqrtss xmm7, xmm7 cvtss2sd xmm1, xmm1 cvtss2sd xmm4, xmm4 cvtss2sd xmm7, xmm7 movsd xmm2, xmm1 movsd xmm5, xmm4 movsd xmm8, xmm7 mulsd xmm1, xmm1 ;# lu*lu mulsd xmm4, xmm4 ;# lu*lu mulsd xmm7, xmm7 ;# lu*lu movsd xmm9, [rsp + nb102_three] movsd xmm10, xmm9 movsd xmm11, xmm9 mulsd xmm1, xmm0 ;# rsq*lu*lu mulsd xmm4, xmm3 ;# rsq*lu*lu mulsd xmm7, xmm6 ;# rsq*lu*lu subsd xmm9, xmm1 subsd xmm10, xmm4 subsd xmm11, xmm7 ;# 3-rsq*lu*lu mulsd xmm9, xmm2 mulsd xmm10, xmm5 mulsd xmm11, xmm8 ;# lu*(3-rsq*lu*lu) movsd xmm15, [rsp + nb102_half] mulsd xmm9, xmm15 ;# first iteration for rinvOH2 mulsd xmm10, xmm15 ;# first iteration for rinvH1H2 mulsd xmm11, xmm15 ;# first iteration for rinvH2H2 ;# second iteration step movsd xmm2, xmm9 movsd xmm5, xmm10 movsd xmm8, xmm11 mulsd xmm2, xmm2 ;# lu*lu mulsd xmm5, xmm5 ;# lu*lu mulsd xmm8, xmm8 ;# lu*lu movsd xmm1, [rsp + nb102_three] movsd xmm4, xmm1 movsd xmm7, xmm1 mulsd xmm2, xmm0 ;# rsq*lu*lu mulsd xmm5, xmm3 ;# rsq*lu*lu mulsd xmm8, xmm6 ;# rsq*lu*lu subsd xmm1, xmm2 subsd xmm4, xmm5 subsd xmm7, xmm8 ;# 3-rsq*lu*lu mulsd xmm9, xmm1 mulsd xmm10, xmm4 mulsd xmm11, xmm7 ;# lu*(3-rsq*lu*lu) movsd xmm15, [rsp + nb102_half] mulsd xmm9, xmm15 ;# rinvOH2 mulsd xmm10, xmm15 ;# rinvH1H2 mulsd xmm11, xmm15 ;# rinvH2H2 ;# H2 interactions movsd xmm0, xmm9 movsd xmm1, xmm10 movsd xmm2, xmm11 mulsd xmm9, xmm9 mulsd xmm10, xmm10 mulsd xmm11, xmm11 mulsd xmm0, [rsp + nb102_qqOH] mulsd xmm1, [rsp + nb102_qqHH] mulsd xmm2, [rsp + nb102_qqHH] mulsd xmm9, xmm0 mulsd xmm10, xmm1 mulsd xmm11, xmm2 addsd xmm0, [rsp + nb102_vctot] addsd xmm1, xmm2 addsd xmm0, xmm1 movsd [rsp + nb102_vctot], xmm0 ;# move j H2 forces to xmm0-xmm2 movsd xmm0, [rdi + rax*8 + 48] movsd xmm1, [rdi + rax*8 + 56] movsd xmm2, [rdi + rax*8 + 64] movsd xmm7, xmm9 movsd xmm8, xmm9 movsd xmm13, xmm11 movsd xmm14, xmm11 movsd xmm15, xmm11 movsd xmm11, xmm10 movsd xmm12, xmm10 mulsd xmm7, [rsp + nb102_dxOH2] mulsd xmm8, [rsp + nb102_dyOH2] mulsd xmm9, [rsp + nb102_dzOH2] mulsd xmm10, [rsp + nb102_dxH1H2] mulsd xmm11, [rsp + nb102_dyH1H2] mulsd xmm12, [rsp + nb102_dzH1H2] mulsd xmm13, [rsp + nb102_dxH2H2] mulsd xmm14, [rsp + nb102_dyH2H2] mulsd xmm15, [rsp + nb102_dzH2H2] addsd xmm0, xmm7 addsd xmm1, xmm8 addsd xmm2, xmm9 addsd xmm7, [rsp + nb102_fixO] addsd xmm8, [rsp + nb102_fiyO] addsd xmm9, [rsp + nb102_fizO] addsd xmm0, xmm10 addsd xmm1, xmm11 addsd xmm2, xmm12 addsd xmm10, [rsp + nb102_fixH1] addsd xmm11, [rsp + nb102_fiyH1] addsd xmm12, [rsp + nb102_fizH1] addsd xmm0, xmm13 addsd xmm1, xmm14 addsd xmm2, xmm15 addsd xmm13, [rsp + nb102_fixH2] addsd xmm14, [rsp + nb102_fiyH2] addsd xmm15, [rsp + nb102_fizH2] movsd [rsp + nb102_fixO], xmm7 movsd [rsp + nb102_fiyO], xmm8 movsd [rsp + nb102_fizO], xmm9 movsd [rsp + nb102_fixH1], xmm10 movsd [rsp + nb102_fiyH1], xmm11 movsd [rsp + nb102_fizH1], xmm12 movsd [rsp + nb102_fixH2], xmm13 movsd [rsp + nb102_fiyH2], xmm14 movsd [rsp + nb102_fizH2], xmm15 ;# store back j H2 forces from xmm0-xmm2 movsd [rdi + rax*8 + 48], xmm0 movsd [rdi + rax*8 + 56], xmm1 movsd [rdi + rax*8 + 64], xmm2 .nb102_updateouterdata: mov ecx, [rsp + nb102_ii3] mov rdi, [rbp + nb102_faction] mov rsi, [rbp + nb102_fshift] mov edx, [rsp + nb102_is3] ;# accumulate Oi forces in xmm0, xmm1, xmm2 movapd xmm0, [rsp + nb102_fixO] movapd xmm1, [rsp + nb102_fiyO] movapd xmm2, [rsp + nb102_fizO] movhlps xmm3, xmm0 movhlps xmm4, xmm1 movhlps xmm5, xmm2 addsd xmm0, xmm3 addsd xmm1, xmm4 addsd xmm2, xmm5 ;# sum is in low xmm0-xmm2 ;# increment i force movsd xmm3, [rdi + rcx*8] movsd xmm4, [rdi + rcx*8 + 8] movsd xmm5, [rdi + rcx*8 + 16] subsd xmm3, xmm0 subsd xmm4, xmm1 subsd xmm5, xmm2 movsd [rdi + rcx*8], xmm3 movsd [rdi + rcx*8 + 8], xmm4 movsd [rdi + rcx*8 + 16], xmm5 ;# accumulate force in xmm6/xmm7 for fshift movapd xmm6, xmm0 movsd xmm7, xmm2 unpcklpd xmm6, xmm1 ;# accumulate H1i forces in xmm0, xmm1, xmm2 movapd xmm0, [rsp + nb102_fixH1] movapd xmm1, [rsp + nb102_fiyH1] movapd xmm2, [rsp + nb102_fizH1] movhlps xmm3, xmm0 movhlps xmm4, xmm1 movhlps xmm5, xmm2 addsd xmm0, xmm3 addsd xmm1, xmm4 addsd xmm2, xmm5 ;# sum is in low xmm0-xmm2 ;# increment i force movsd xmm3, [rdi + rcx*8 + 24] movsd xmm4, [rdi + rcx*8 + 32] movsd xmm5, [rdi + rcx*8 + 40] subsd xmm3, xmm0 subsd xmm4, xmm1 subsd xmm5, xmm2 movsd [rdi + rcx*8 + 24], xmm3 movsd [rdi + rcx*8 + 32], xmm4 movsd [rdi + rcx*8 + 40], xmm5 ;# accumulate force in xmm6/xmm7 for fshift addsd xmm7, xmm2 unpcklpd xmm0, xmm1 addpd xmm6, xmm0 ;# accumulate H2i forces in xmm0, xmm1, xmm2 movapd xmm0, [rsp + nb102_fixH2] movapd xmm1, [rsp + nb102_fiyH2] movapd xmm2, [rsp + nb102_fizH2] movhlps xmm3, xmm0 movhlps xmm4, xmm1 movhlps xmm5, xmm2 addsd xmm0, xmm3 addsd xmm1, xmm4 addsd xmm2, xmm5 ;# sum is in low xmm0-xmm2 movapd xmm3, xmm0 movapd xmm4, xmm1 movapd xmm5, xmm2 ;# increment i force movsd xmm3, [rdi + rcx*8 + 48] movsd xmm4, [rdi + rcx*8 + 56] movsd xmm5, [rdi + rcx*8 + 64] subsd xmm3, xmm0 subsd xmm4, xmm1 subsd xmm5, xmm2 movsd [rdi + rcx*8 + 48], xmm3 movsd [rdi + rcx*8 + 56], xmm4 movsd [rdi + rcx*8 + 64], xmm5 ;# accumulate force in xmm6/xmm7 for fshift addsd xmm7, xmm2 unpcklpd xmm0, xmm1 addpd xmm6, xmm0 ;# increment fshift force movlpd xmm3, [rsi + rdx*8] movhpd xmm3, [rsi + rdx*8 + 8] movsd xmm4, [rsi + rdx*8 + 16] subpd xmm3, xmm6 subsd xmm4, xmm7 movlpd [rsi + rdx*8], xmm3 movhpd [rsi + rdx*8 + 8], xmm3 movsd [rsi + rdx*8 + 16], xmm4 ;# get n from stack mov esi, [rsp + nb102_n] ;# get group index for i particle mov rdx, [rbp + nb102_gid] ;# base of gid[] mov edx, [rdx + rsi*4] ;# ggid=gid[n] ;# accumulate total potential energy and update it movapd xmm7, [rsp + nb102_vctot] ;# accumulate movhlps xmm6, xmm7 addsd xmm7, xmm6 ;# pos 0-1 in xmm7 have the sum now ;# add earlier value from mem mov rax, [rbp + nb102_Vc] addsd xmm7, [rax + rdx*8] ;# move back to mem movsd [rax + rdx*8], xmm7 ;# finish if last mov ecx, [rsp + nb102_nn1] ;# esi already loaded with n inc esi sub ecx, esi jz .nb102_outerend ;# not last, iterate outer loop once more! mov [rsp + nb102_n], esi jmp .nb102_outer.nb102_outerend: ;# check if more outer neighborlists remain mov ecx, [rsp + nb102_nri] ;# esi already loaded with n above sub ecx, esi jz .nb102_end ;# non-zero, do one more workunit jmp .nb102_threadloop.nb102_end: mov eax, [rsp + nb102_nouter] mov ebx, [rsp + nb102_ninner] mov rcx, [rbp + nb102_outeriter] mov rdx, [rbp + nb102_inneriter] mov [rcx], eax mov [rdx], ebx add rsp, 1496 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: .equiv nb102nf_fshift, 16.equiv nb102nf_gid, 24.equiv nb102nf_pos, 32.equiv nb102nf_faction, 40.equiv nb102nf_charge, 48.equiv nb102nf_p_facel, 56.equiv nb102nf_argkrf, 64.equiv nb102nf_argcrf, 72.equiv nb102nf_Vc, 80.equiv nb102nf_type, 88.equiv nb102nf_p_ntype, 96.equiv nb102nf_vdwparam, 104.equiv nb102nf_Vvdw, 112.equiv nb102nf_p_tabscale, 120.equiv nb102nf_VFtab, 128.equiv nb102nf_invsqrta, 136.equiv nb102nf_dvda, 144.equiv nb102nf_p_gbtabscale, 152.equiv nb102nf_GBtab, 160.equiv nb102nf_p_nthreads, 168.equiv nb102nf_count, 176.equiv nb102nf_mtx, 184.equiv nb102nf_outeriter, 192.equiv nb102nf_inneriter, 200.equiv nb102nf_work, 208 ;# stack offsets for local variables ;# bottom of stack is cache-aligned for sse2 use .equiv nb102nf_ixO, 0.equiv nb102nf_iyO, 16.equiv nb102nf_izO, 32.equiv nb102nf_ixH1, 48.equiv nb102nf_iyH1, 64.equiv nb102nf_izH1, 80.equiv nb102nf_ixH2, 96.equiv nb102nf_iyH2, 112.equiv nb102nf_izH2, 128.equiv nb102nf_jxO, 144.equiv nb102nf_jyO, 160.equiv nb102nf_jzO, 176.equiv nb102nf_jxH1, 192.equiv nb102nf_jyH1, 208.equiv nb102nf_jzH1, 224.equiv nb102nf_jxH2, 240.equiv nb102nf_jyH2, 256.equiv nb102nf_jzH2, 272.equiv nb102nf_qqOO, 288.equiv nb102nf_qqOH, 304.equiv nb102nf_qqHH, 320.equiv nb102nf_vctot, 336.equiv nb102nf_half, 352.equiv nb102nf_three, 368.equiv nb102nf_rsqOO, 384.equiv nb102nf_rsqOH1, 400.equiv nb102nf_rsqOH2, 416.equiv nb102nf_rsqH1O, 432.equiv nb102nf_rsqH1H1, 448.equiv nb102nf_rsqH1H2, 464.equiv nb102nf_rsqH2O, 480.equiv nb102nf_rsqH2H1, 496.equiv nb102nf_rsqH2H2, 512⌨️ 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?