nb_kernel333_ia32_sse2.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,347 行 · 第 1/5 页
S
2,347 行
movsd xmm7, [esi + eax*8 + 24] ;# H1 ;# coulomb table ready, in xmm4-xmm7 mulsd xmm6, xmm1 ;# xmm6=Geps mulsd xmm7, xmm2 ;# xmm7=Heps2 addsd xmm5, xmm6 addsd xmm5, xmm7 ;# xmm5=Fp mulsd xmm7, [esp + nb333_two] ;# two*Heps2 movapd xmm3, [esp + nb333_qqH] addsd xmm7, xmm6 addsd xmm7, xmm5 ;# xmm7=FF mulsd xmm5, xmm1 ;# xmm5=eps*Fp addsd xmm5, xmm4 ;# xmm5=VV mulsd xmm5, xmm3 ;# vcoul=qq*VV mulsd xmm3, xmm7 ;# fijC=FF*qq ;# at this point mm5 contains vcoul and xmm3 fijC ;# increment vcoul xorpd xmm4, xmm4 addsd xmm5, [esp + nb333_vctot] mulsd xmm3, [esp + nb333_rinvH2] movlpd [esp + nb333_vctot], xmm5 mulsd xmm3, [esp + nb333_tsc] subsd xmm4, xmm3 movapd xmm0, [esp + nb333_dxH2] movapd xmm1, [esp + nb333_dyH2] movapd xmm2, [esp + nb333_dzH2] mulsd xmm0, xmm4 mulsd xmm1, xmm4 mulsd xmm2, xmm4 movd eax, mm0 ;# update H2 forces movapd xmm3, [esp + nb333_fixH2] movapd xmm4, [esp + nb333_fiyH2] movapd xmm7, [esp + nb333_fizH2] addsd xmm3, xmm0 addsd xmm4, xmm1 addsd xmm7, xmm2 movlpd [esp + nb333_fixH2], xmm3 movlpd [esp + nb333_fiyH2], xmm4 movlpd [esp + nb333_fizH2], xmm7 ;# update j forces with water H1 addsd xmm0, [esp + nb333_fjx] addsd xmm1, [esp + nb333_fjy] addsd xmm2, [esp + nb333_fjz] movlpd [esp + nb333_fjx], xmm0 movlpd [esp + nb333_fjy], xmm1 movlpd [esp + nb333_fjz], xmm2 ;# M interactions movapd xmm7, [esp + nb333_rM] mulsd xmm7, [esp + nb333_tsc] cvttsd2si eax, xmm7 ;# mm6 = lu idx cvtsi2sd xmm6, eax subsd xmm7, xmm6 movapd xmm1, xmm7 ;# xmm1=eps movapd xmm2, xmm1 mulsd xmm2, xmm2 ;# xmm2=eps2 shl eax, 2 ;# idx *= 4 mov esi, [ebp + nb333_VFtab] lea eax, [eax + eax*2] ;# idx *= 3 (total *=12 now) movsd xmm4, [esi + eax*8] ;# Y1 movsd xmm5, [esi + eax*8 + 8] ;# F1 movsd xmm6, [esi + eax*8 + 16] ;# G1 movsd xmm7, [esi + eax*8 + 24] ;# H1 ;# coulomb table ready, in xmm4-xmm7 mulsd xmm6, xmm1 ;# xmm6=Geps mulsd xmm7, xmm2 ;# xmm7=Heps2 addsd xmm5, xmm6 addsd xmm5, xmm7 ;# xmm5=Fp mulsd xmm7, [esp + nb333_two] ;# two*Heps2 movapd xmm3, [esp + nb333_qqM] addsd xmm7, xmm6 addsd xmm7, xmm5 ;# xmm7=FF mulsd xmm5, xmm1 ;# xmm5=eps*Fp addsd xmm5, xmm4 ;# xmm5=VV mulsd xmm5, xmm3 ;# vcoul=qq*VV mulsd xmm3, xmm7 ;# fijC=FF*qq ;# at this point mm5 contains vcoul and xmm3 fijC ;# increment vcoul xorpd xmm4, xmm4 addsd xmm5, [esp + nb333_vctot] mulsd xmm3, [esp + nb333_rinvM] movlpd [esp + nb333_vctot], xmm5 mulsd xmm3, [esp + nb333_tsc] subsd xmm4, xmm3 movapd xmm0, [esp + nb333_dxM] movapd xmm1, [esp + nb333_dyM] movapd xmm2, [esp + nb333_dzM] mulsd xmm0, xmm4 mulsd xmm1, xmm4 mulsd xmm2, xmm4 movd eax, mm0 ;# update M forces movapd xmm3, [esp + nb333_fixM] movapd xmm4, [esp + nb333_fiyM] movapd xmm7, [esp + nb333_fizM] addsd xmm3, xmm0 addsd xmm4, xmm1 addsd xmm7, xmm2 movlpd [esp + nb333_fixM], xmm3 movlpd [esp + nb333_fiyM], xmm4 movlpd [esp + nb333_fizM], xmm7 mov edi, [ebp + nb333_faction] ;# update j forces addsd xmm0, [esp + nb333_fjx] addsd xmm1, [esp + nb333_fjy] addsd xmm2, [esp + nb333_fjz] ;# the fj's - start by accumulating forces from memory movlpd xmm3, [edi + eax*8] movlpd xmm4, [edi + eax*8 + 8] movlpd xmm5, [edi + eax*8 + 16] subsd xmm3, xmm0 subsd xmm4, xmm1 subsd xmm5, xmm2 movlpd [edi + eax*8], xmm3 movlpd [edi + eax*8 + 8], xmm4 movlpd [edi + eax*8 + 16], xmm5.nb333_updateouterdata: mov ecx, [esp + nb333_ii3] mov edi, [ebp + nb333_faction] mov esi, [ebp + nb333_fshift] mov edx, [esp + nb333_is3] ;# accumulate Oi forces in xmm0, xmm1, xmm2 movapd xmm0, [esp + nb333_fixO] movapd xmm1, [esp + nb333_fiyO] movapd xmm2, [esp + nb333_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 movapd xmm3, xmm0 movapd xmm4, xmm1 movapd xmm5, xmm2 ;# increment i force movsd xmm3, [edi + ecx*8] movsd xmm4, [edi + ecx*8 + 8] movsd xmm5, [edi + ecx*8 + 16] addsd xmm3, xmm0 addsd xmm4, xmm1 addsd xmm5, xmm2 movsd [edi + ecx*8], xmm3 movsd [edi + ecx*8 + 8], xmm4 movsd [edi + ecx*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, [esp + nb333_fixH1] movapd xmm1, [esp + nb333_fiyH1] movapd xmm2, [esp + nb333_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, [edi + ecx*8 + 24] movsd xmm4, [edi + ecx*8 + 32] movsd xmm5, [edi + ecx*8 + 40] addsd xmm3, xmm0 addsd xmm4, xmm1 addsd xmm5, xmm2 movsd [edi + ecx*8 + 24], xmm3 movsd [edi + ecx*8 + 32], xmm4 movsd [edi + ecx*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, [esp + nb333_fixH2] movapd xmm1, [esp + nb333_fiyH2] movapd xmm2, [esp + nb333_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, [edi + ecx*8 + 48] movsd xmm4, [edi + ecx*8 + 56] movsd xmm5, [edi + ecx*8 + 64] addsd xmm3, xmm0 addsd xmm4, xmm1 addsd xmm5, xmm2 movsd [edi + ecx*8 + 48], xmm3 movsd [edi + ecx*8 + 56], xmm4 movsd [edi + ecx*8 + 64], xmm5 ;# accumulate force in xmm6/xmm7 for fshift addsd xmm7, xmm2 unpcklpd xmm0, xmm1 addpd xmm6, xmm0 ;# accumulate Mi forces in xmm0, xmm1, xmm2 movapd xmm0, [esp + nb333_fixM] movapd xmm1, [esp + nb333_fiyM] movapd xmm2, [esp + nb333_fizM] 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, [edi + ecx*8 + 72] movsd xmm4, [edi + ecx*8 + 80] movsd xmm5, [edi + ecx*8 + 88] addsd xmm3, xmm0 addsd xmm4, xmm1 addsd xmm5, xmm2 movsd [edi + ecx*8 + 72], xmm3 movsd [edi + ecx*8 + 80], xmm4 movsd [edi + ecx*8 + 88], xmm5 ;# accumulate force in xmm6/xmm7 for fshift addsd xmm7, xmm2 unpcklpd xmm0, xmm1 addpd xmm6, xmm0 ;# increment fshift force movlpd xmm3, [esi + edx*8] movhpd xmm3, [esi + edx*8 + 8] movsd xmm4, [esi + edx*8 + 16] addpd xmm3, xmm6 addsd xmm4, xmm7 movlpd [esi + edx*8], xmm3 movhpd [esi + edx*8 + 8], xmm3 movsd [esi + edx*8 + 16], xmm4 ;# get n from stack mov esi, [esp + nb333_n] ;# get group index for i particle mov edx, [ebp + nb333_gid] ;# base of gid[] mov edx, [edx + esi*4] ;# ggid=gid[n] ;# accumulate total potential energy and update it movapd xmm7, [esp + nb333_vctot] ;# accumulate movhlps xmm6, xmm7 addsd xmm7, xmm6 ;# low xmm7 has the sum now ;# add earlier value from mem mov eax, [ebp + nb333_Vc] addsd xmm7, [eax + edx*8] ;# move back to mem movsd [eax + edx*8], xmm7 ;# accumulate total lj energy and update it movapd xmm7, [esp + nb333_Vvdwtot] ;# accumulate movhlps xmm6, xmm7 addsd xmm7, xmm6 ;# low xmm7 has the sum now ;# add earlier value from mem mov eax, [ebp + nb333_Vvdw] addsd xmm7, [eax + edx*8] ;# move back to mem movsd [eax + edx*8], xmm7 ;# finish if last mov ecx, [esp + nb333_nn1] ;# esi already loaded with n inc esi sub ecx, esi jz .nb333_outerend ;# not last, iterate outer loop once more! mov [esp + nb333_n], esi jmp .nb333_outer.nb333_outerend: ;# check if more outer neighborlists remain mov ecx, [esp + nb333_nri] ;# esi already loaded with n above sub ecx, esi jz .nb333_end ;# non-zero, do one more workunit jmp .nb333_threadloop.nb333_end: emms mov eax, [esp + nb333_nouter] mov ebx, [esp + nb333_ninner] mov ecx, [ebp + nb333_outeriter] mov edx, [ebp + nb333_inneriter] mov [ecx], eax mov [edx], ebx mov eax, [esp + nb333_salign] add esp, eax add esp, 988 pop edi pop esi pop edx pop ecx pop ebx pop eax leave ret .globl nb_kernel333nf_ia32_sse2.globl _nb_kernel333nf_ia32_sse2nb_kernel333nf_ia32_sse2: _nb_kernel333nf_ia32_sse2: .equiv nb333nf_p_nri, 8.equiv nb333nf_iinr, 12.equiv nb333nf_jindex, 16.equiv nb333nf_jjnr, 20.equiv nb333nf_shift, 24.equiv nb333nf_shiftvec, 28.equiv nb333nf_fshift, 32.equiv nb333nf_gid, 36.equiv nb333nf_pos, 40.equiv nb333nf_faction, 44.equiv nb333nf_charge, 48.equiv nb333nf_p_facel, 52.equiv nb333nf_argkrf, 56.equiv nb333nf_argcrf, 60.equiv nb333nf_Vc, 64.equiv nb333nf_type, 68.equiv nb333nf_p_ntype, 72.equiv nb333nf_vdwparam, 76.equiv nb333nf_Vvdw, 80.equiv nb333nf_p_tabscale, 84.equiv nb333nf_VFtab, 88.equiv nb333nf_invsqrta, 92.equiv nb333nf_dvda, 96.equiv nb333nf_p_gbtabscale, 100.equiv nb333nf_GBtab, 104.equiv nb333nf_p_nthreads, 108.equiv nb333nf_count, 112.equiv nb333nf_mtx, 116.equiv nb333nf_outeriter, 120.equiv nb333nf_inneriter, 124.equiv nb333nf_work, 128 ;# stack offsets for local variables ;# bottom of stack is cache-aligned for sse2 use .equiv nb333nf_ixO, 0.equiv nb333nf_iyO, 16.equiv nb333nf_izO, 32.equiv nb333nf_ixH1, 48.equiv nb333nf_iyH1, 64.equiv nb333nf_izH1, 80.equiv nb333nf_ixH2, 96.equiv nb333nf_iyH2, 112.equiv nb333nf_izH2, 128.equiv nb333nf_ixM, 144.equiv nb333nf_iyM, 160.equiv nb333nf_izM, 176.equiv nb333nf_iqM, 192.equiv nb333nf_iqH, 208.equiv nb333nf_qqM, 224.equiv nb333nf_qqH, 240.equiv nb333nf_rO, 256.equiv nb333nf_rH1, 272.equiv nb333nf_rH2, 288.equiv nb333nf_rM, 304.equiv nb333nf_tsc, 320.equiv nb333nf_c6, 336.equiv nb333nf_c12, 352.equiv nb333nf_vctot, 368.equiv nb333nf_Vvdwtot, 384.equiv nb333nf_half, 400.equiv nb333nf_three, 416.equiv nb333nf_is3, 432.equiv nb333nf_ii3, 436.equiv nb333nf_ntia, 440.equiv nb333nf_innerjjnr, 444.equiv nb333nf_innerk, 448.equiv nb333nf_n, 452.equiv nb333nf_nn1, 456.equiv nb333nf_nri, 460.equiv nb333nf_nouter, 464.equiv nb333nf_ninner, 468.equiv nb333nf_salign, 472 push ebp mov ebp,esp push eax push ebx push ecx push edx push esi push edi sub esp, 476 ;# local stack space mov eax, esp and eax, 0xf sub esp, eax mov [esp + nb333nf_salign], eax emms ;# Move args passed by reference to stack mov ecx, [ebp + nb333nf_p_nri] mov ecx, [ecx] mov [esp + nb333nf_nri], ecx ;# zero iteration counters mov eax, 0 mov [esp + nb333nf_nouter], eax mov [esp + nb333nf_ninner], eax ;# create constant floating-point factors on stack mov eax, 0x00000000 ;# lower half of double 0.5 IEEE (hex) mov ebx, 0x3fe00000 mov [esp + nb333nf_half], eax mov [esp + nb333nf_half+4], ebx movsd xmm1, [esp + nb333nf_half] shufpd xmm1, xmm1, 0 ;# splat to all elements movapd xmm3, xmm1 addpd xmm3, xmm3 ;# 1.0 movapd xmm2, xmm3 addpd xmm2, xmm2 ;# 2.0 addpd xmm3, xmm2 ;# 3.0 movapd [esp + nb333nf_half], xmm1 movapd [esp + nb333nf_three], xmm3 mov eax, [ebp + nb333nf_p_tabscale] movsd xmm5, [eax] shufpd xmm5, xmm5, 0 movapd [esp + nb333nf_tsc], xmm5 ;# assume we have at least one i particle - start directly mov ecx, [ebp + nb333nf_iinr] ;# ecx = pointer into iinr[] mov ebx, [ecx] ;# ebx =ii mov edx, [ebp + nb333nf_charge]⌨️ 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?