nb_kernel234_ia32_sse.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,176 行 · 第 1/5 页
S
2,176 行
addps xmm1, [esp + nb234_fiyH1] addps xmm2, [esp + nb234_fizH1] movaps [esp + nb234_fixH1], xmm0 movaps [esp + nb234_fiyH1], xmm1 movaps [esp + nb234_fizH1], xmm2 ;# H2 Coulomb movaps xmm0, [esp + nb234_rinvH2H2] movaps xmm7, xmm0 ;# xmm7=rinv movaps xmm5, [esp + nb234_krf] mulps xmm0, xmm0 ;# xmm0=rinvsq ;# fetch charges to xmm3 (temporary) xorps xmm3, xmm3 movss xmm3, [esp + nb234_qqHH] movhps xmm3, [esp + nb234_qqMH] shufps xmm3, xmm3, 193 ;# constant 11000001 mulps xmm5, [esp + nb234_rsqH2H2] ;# xmm5=krsq movaps xmm6, xmm5 addps xmm6, xmm7 ;# xmm6=rinv+ krsq subps xmm6, [esp + nb234_crf] mulps xmm6, xmm3 ;# xmm6=voul=qq*(rinv+ krsq-crf) mulps xmm5, [esp + nb234_two] subps xmm7, xmm5 ;# xmm7=rinv-2*krsq mulps xmm7, xmm3 ;# xmm7 = coul part of fscal addps xmm6, [esp + nb234_vctot] ;# local vctot summation variable movaps [esp + nb234_vctot], xmm6 mulps xmm0, xmm7 movaps xmm1, xmm0 movaps xmm2, xmm0 mulps xmm0, [esp + nb234_dxH2H2] mulps xmm1, [esp + nb234_dyH2H2] mulps xmm2, [esp + nb234_dzH2H2] ;# update forces H2 - j water movaps xmm3, [esp + nb234_fjxO] movaps xmm4, [esp + nb234_fjyO] movaps xmm5, [esp + nb234_fjzO] subps xmm3, xmm0 subps xmm4, xmm1 subps xmm5, xmm2 movaps [esp + nb234_fjxO], xmm3 movaps [esp + nb234_fjyO], xmm4 movaps [esp + nb234_fjzO], xmm5 addps xmm0, [esp + nb234_fixH2] addps xmm1, [esp + nb234_fiyH2] addps xmm2, [esp + nb234_fizH2] movaps [esp + nb234_fixH2], xmm0 movaps [esp + nb234_fiyH2], xmm1 movaps [esp + nb234_fizH2], xmm2 mov esi, [ebp + nb234_faction] ;# update j water forces from local variables. ;# transpose back first movaps xmm0, [esp + nb234_fjxO] ;# Ox H1x H2x Mx movaps xmm1, [esp + nb234_fjyO] ;# Oy H1y H2y My movaps xmm2, [esp + nb234_fjzO] ;# Oz H1z H2z Mz movaps xmm3, xmm0 movaps xmm4, xmm0 unpcklps xmm3, xmm1 ;# Ox Oy - - shufps xmm4, xmm2, 0x1 ;# h1x - Oz - movaps xmm5, xmm1 movaps xmm6, xmm0 unpcklps xmm5, xmm2 ;# - - H1y H1z unpckhps xmm6, xmm1 ;# h2x h2y - - unpckhps xmm1, xmm2 ;# - - My Mz shufps xmm2, xmm0, 50 ;# (00110010) h2z - Mx - shufps xmm3, xmm4, 36 ;# constant 00100100 ;# Ox Oy Oz H1x shufps xmm5, xmm6, 78 ;# constant 01001110 ;# h1y h1z h2x h2y shufps xmm2, xmm1, 232 ;# constant 11101000 ;# h2z mx my mz movlps xmm0, [esi + eax*4] movlps xmm1, [esi + eax*4 + 16] movlps xmm4, [esi + eax*4 + 32] movhps xmm0, [esi + eax*4 + 8] movhps xmm1, [esi + eax*4 + 24] movhps xmm4, [esi + eax*4 + 40] addps xmm0, xmm3 addps xmm1, xmm5 addps xmm4, xmm2 movlps [esi + eax*4], xmm0 movlps [esi + eax*4 + 16], xmm1 movlps [esi + eax*4 + 32], xmm4 movhps [esi + eax*4 + 8], xmm0 movhps [esi + eax*4 + 24], xmm1 movhps [esi + eax*4 + 40], xmm4 dec dword ptr [esp + nb234_innerk] jz .nb234_updateouterdata jmp .nb234_single_loop.nb234_updateouterdata: mov ecx, [esp + nb234_ii3] mov edi, [ebp + nb234_faction] mov esi, [ebp + nb234_fshift] mov edx, [esp + nb234_is3] ;# accumulate Oi forces in xmm0, xmm1, xmm2 movaps xmm0, [esp + nb234_fixO] movaps xmm1, [esp + nb234_fiyO] movaps xmm2, [esp + nb234_fizO] movhlps xmm3, xmm0 movhlps xmm4, xmm1 movhlps xmm5, xmm2 addps xmm0, xmm3 addps xmm1, xmm4 addps xmm2, xmm5 ;# sum is in 1/2 in xmm0-xmm2 movaps xmm3, xmm0 movaps xmm4, xmm1 movaps xmm5, xmm2 shufps xmm3, xmm3, 1 shufps xmm4, xmm4, 1 shufps xmm5, xmm5, 1 addss xmm0, xmm3 addss xmm1, xmm4 addss xmm2, xmm5 ;# xmm0-xmm2 has single force in pos0 ;# increment i force movss xmm3, [edi + ecx*4] movss xmm4, [edi + ecx*4 + 4] movss xmm5, [edi + ecx*4 + 8] addss xmm3, xmm0 addss xmm4, xmm1 addss xmm5, xmm2 movss [edi + ecx*4], xmm3 movss [edi + ecx*4 + 4], xmm4 movss [edi + ecx*4 + 8], xmm5 ;# accumulate force in xmm6/xmm7 for fshift movaps xmm6, xmm0 movss xmm7, xmm2 movlhps xmm6, xmm1 shufps xmm6, xmm6, 8 ;# constant 00001000 ;# accumulate H1i forces in xmm0, xmm1, xmm2 movaps xmm0, [esp + nb234_fixH1] movaps xmm1, [esp + nb234_fiyH1] movaps xmm2, [esp + nb234_fizH1] movhlps xmm3, xmm0 movhlps xmm4, xmm1 movhlps xmm5, xmm2 addps xmm0, xmm3 addps xmm1, xmm4 addps xmm2, xmm5 ;# sum is in 1/2 in xmm0-xmm2 movaps xmm3, xmm0 movaps xmm4, xmm1 movaps xmm5, xmm2 shufps xmm3, xmm3, 1 shufps xmm4, xmm4, 1 shufps xmm5, xmm5, 1 addss xmm0, xmm3 addss xmm1, xmm4 addss xmm2, xmm5 ;# xmm0-xmm2 has single force in pos0 ;# increment i force movss xmm3, [edi + ecx*4 + 12] movss xmm4, [edi + ecx*4 + 16] movss xmm5, [edi + ecx*4 + 20] addss xmm3, xmm0 addss xmm4, xmm1 addss xmm5, xmm2 movss [edi + ecx*4 + 12], xmm3 movss [edi + ecx*4 + 16], xmm4 movss [edi + ecx*4 + 20], xmm5 ;# accumulate force in xmm6/xmm7 for fshift addss xmm7, xmm2 movlhps xmm0, xmm1 shufps xmm0, xmm0, 8 ;# constant 00001000 addps xmm6, xmm0 ;# accumulate H2i forces in xmm0, xmm1, xmm2 movaps xmm0, [esp + nb234_fixH2] movaps xmm1, [esp + nb234_fiyH2] movaps xmm2, [esp + nb234_fizH2] movhlps xmm3, xmm0 movhlps xmm4, xmm1 movhlps xmm5, xmm2 addps xmm0, xmm3 addps xmm1, xmm4 addps xmm2, xmm5 ;# sum is in 1/2 in xmm0-xmm2 movaps xmm3, xmm0 movaps xmm4, xmm1 movaps xmm5, xmm2 shufps xmm3, xmm3, 1 shufps xmm4, xmm4, 1 shufps xmm5, xmm5, 1 addss xmm0, xmm3 addss xmm1, xmm4 addss xmm2, xmm5 ;# xmm0-xmm2 has single force in pos0 ;# increment i force movss xmm3, [edi + ecx*4 + 24] movss xmm4, [edi + ecx*4 + 28] movss xmm5, [edi + ecx*4 + 32] addss xmm3, xmm0 addss xmm4, xmm1 addss xmm5, xmm2 movss [edi + ecx*4 + 24], xmm3 movss [edi + ecx*4 + 28], xmm4 movss [edi + ecx*4 + 32], xmm5 ;# accumulate force in xmm6/xmm7 for fshift addss xmm7, xmm2 movlhps xmm0, xmm1 shufps xmm0, xmm0, 8 ;# constant 00001000 addps xmm6, xmm0 ;# accumulate Mi forces in xmm0, xmm1, xmm2 movaps xmm0, [esp + nb234_fixM] movaps xmm1, [esp + nb234_fiyM] movaps xmm2, [esp + nb234_fizM] movhlps xmm3, xmm0 movhlps xmm4, xmm1 movhlps xmm5, xmm2 addps xmm0, xmm3 addps xmm1, xmm4 addps xmm2, xmm5 ;# sum is in 1/2 in xmm0-xmm2 movaps xmm3, xmm0 movaps xmm4, xmm1 movaps xmm5, xmm2 shufps xmm3, xmm3, 1 shufps xmm4, xmm4, 1 shufps xmm5, xmm5, 1 addss xmm0, xmm3 addss xmm1, xmm4 addss xmm2, xmm5 ;# xmm0-xmm2 has single force in pos0 ;# increment i force movss xmm3, [edi + ecx*4 + 36] movss xmm4, [edi + ecx*4 + 40] movss xmm5, [edi + ecx*4 + 44] addss xmm3, xmm0 addss xmm4, xmm1 addss xmm5, xmm2 movss [edi + ecx*4 + 36], xmm3 movss [edi + ecx*4 + 40], xmm4 movss [edi + ecx*4 + 44], xmm5 ;# accumulate force in xmm6/xmm7 for fshift addss xmm7, xmm2 movlhps xmm0, xmm1 shufps xmm0, xmm0, 8 ;# constant 00001000 addps xmm6, xmm0 ;# increment fshift force movlps xmm3, [esi + edx*4] movss xmm4, [esi + edx*4 + 8] addps xmm3, xmm6 addss xmm4, xmm7 movlps [esi + edx*4], xmm3 movss [esi + edx*4 + 8], xmm4 ;# get n from stack mov esi, [esp + nb234_n] ;# get group index for i particle mov edx, [ebp + nb234_gid] ;# base of gid[] mov edx, [edx + esi*4] ;# ggid=gid[n] ;# accumulate total potential energy and update it movaps xmm7, [esp + nb234_vctot] ;# accumulate movhlps xmm6, xmm7 addps xmm7, xmm6 ;# pos 0-1 in xmm7 have the sum now movaps xmm6, xmm7 shufps xmm6, xmm6, 1 addss xmm7, xmm6 ;# add earlier value from mem mov eax, [ebp + nb234_Vc] addss xmm7, [eax + edx*4] ;# move back to mem movss [eax + edx*4], xmm7 ;# accumulate total lj energy and update it movaps xmm7, [esp + nb234_Vvdwtot] ;# accumulate movhlps xmm6, xmm7 addps xmm7, xmm6 ;# pos 0-1 in xmm7 have the sum now movaps xmm6, xmm7 shufps xmm6, xmm6, 1 addss xmm7, xmm6 ;# add earlier value from mem mov eax, [ebp + nb234_Vvdw] addss xmm7, [eax + edx*4] ;# move back to mem movss [eax + edx*4], xmm7 ;# finish if last mov ecx, [esp + nb234_nn1] ;# esi already loaded with n inc esi sub ecx, esi jecxz .nb234_outerend ;# not last, iterate outer loop once more! mov [esp + nb234_n], esi jmp .nb234_outer.nb234_outerend: ;# check if more outer neighborlists remain mov ecx, [esp + nb234_nri] ;# esi already loaded with n above sub ecx, esi jecxz .nb234_end ;# non-zero, do one more workunit jmp .nb234_threadloop.nb234_end: emms mov eax, [esp + nb234_nouter] mov ebx, [esp + nb234_ninner] mov ecx, [ebp + nb234_outeriter] mov edx, [ebp + nb234_inneriter] mov [ecx], eax mov [edx], ebx mov eax, [esp + nb234_salign] add esp, eax add esp, 1848 pop edi pop esi pop edx pop ecx pop ebx pop eax leave ret .globl nb_kernel234nf_ia32_sse.globl _nb_kernel234nf_ia32_ssenb_kernel234nf_ia32_sse: _nb_kernel234nf_ia32_sse: .equiv nb234nf_p_nri, 8.equiv nb234nf_iinr, 12.equiv nb234nf_jindex, 16.equiv nb234nf_jjnr, 20.equiv nb234nf_shift, 24.equiv nb234nf_shiftvec, 28.equiv nb234nf_fshift, 32.equiv nb234nf_gid, 36.equiv nb234nf_pos, 40.equiv nb234nf_faction, 44.equiv nb234nf_charge, 48.equiv nb234nf_p_facel, 52.equiv nb234nf_argkrf, 56.equiv nb234nf_argcrf, 60.equiv nb234nf_Vc, 64.equiv nb234nf_type, 68.equiv nb234nf_p_ntype, 72.equiv nb234nf_vdwparam, 76.equiv nb234nf_Vvdw, 80.equiv nb234nf_p_tabscale, 84.equiv nb234nf_VFtab, 88.equiv nb234nf_invsqrta, 92.equiv nb234nf_dvda, 96.equiv nb234nf_p_gbtabscale, 100.equiv nb234nf_GBtab, 104.equiv nb234nf_p_nthreads, 108.equiv nb234nf_count, 112.equiv nb234nf_mtx, 116.equiv nb234nf_outeriter, 120.equiv nb234nf_inneriter, 124.equiv nb234nf_work, 128 ;# stack offsets for local variables ;# bottom of stack is cache-aligned for sse use .equiv nb234nf_ixO, 0.equiv nb234nf_iyO, 16.equiv nb234nf_izO, 32.equiv nb234nf_ixH1, 48.equiv nb234nf_iyH1, 64.equiv nb234nf_izH1, 80.equiv nb234nf_ixH2, 96.equiv nb234nf_iyH2, 112.equiv nb234nf_izH2, 128.equiv nb234nf_ixM, 144.equiv nb234nf_iyM, 160.equiv nb234nf_izM, 176.equiv nb234nf_jxO, 192.equiv nb234nf_jyO, 208.equiv nb234nf_jzO, 224.equiv nb234nf_jxH1, 240.equiv nb234nf_jyH1, 256.equiv nb234nf_jzH1, 272.equiv nb234nf_jxH2, 288.equiv nb234nf_jyH2, 304.equiv nb234nf_jzH2, 320.equiv nb234nf_jxM, 336.equiv nb234nf_jyM, 352.equiv nb234nf_jzM, 368.equiv nb234nf_qqMM, 384.equiv nb234nf_qqMH, 400.equiv nb234nf_qqHH, 416.equiv nb234nf_c6, 432.equiv nb234nf_c12, 448.equiv nb234nf_tsc, 464.equiv nb234nf_vctot, 480.equiv nb234nf_Vvdwtot, 496.equiv nb234nf_three, 512.equiv nb234nf_rsqOO, 528.equiv nb234nf_rsqH1H1, 544.equiv nb234nf_rsqH1H2, 560.equiv nb234nf_rsqH1M, 576.equiv nb234nf_rsqH2H1, 592.equiv nb234nf_rsqH2H2, 608.equiv nb234nf_rsqH2M, 624.equiv nb234nf_rsqMH1, 640.equiv nb234nf_rsqMH2, 656.equiv nb234nf_rsqMM, 672.equiv nb234nf_rinvOO, 688.equiv nb234nf_rinvH1H1, 704.equiv nb234nf_rinvH1H2, 720.equiv nb234nf_rinvH1M, 736.equiv ⌨️ 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?