nb_kernel200_ia32_sse.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 1,306 行 · 第 1/3 页
S
1,306 行
.nb200_finish_inner: ;# check if at least two particles remain add dword ptr [esp + nb200_innerk], 4 mov edx, [esp + nb200_innerk] and edx, 2 jnz .nb200_dopair jmp .nb200_checksingle.nb200_dopair: mov esi, [ebp + nb200_charge] mov ecx, [esp + nb200_innerjjnr] mov eax, [ecx] mov ebx, [ecx + 4] add dword ptr [esp + nb200_innerjjnr], 8 xorps xmm3, xmm3 movss xmm3, [esi + eax*4] movss xmm6, [esi + ebx*4] shufps xmm3, xmm6, 12 ;# constant 00001100 shufps xmm3, xmm3, 88 ;# constant 01011000 ;# xmm3(0,1) has the charges mov edi, [ebp + nb200_pos] lea eax, [eax + eax*2] lea ebx, [ebx + ebx*2] ;# move coordinates to xmm0-xmm2 movlps xmm1, [edi + eax*4] movss xmm2, [edi + eax*4 + 8] movhps xmm1, [edi + ebx*4] movss xmm0, [edi + ebx*4 + 8] mulps xmm3, [esp + nb200_iq] xorps xmm7,xmm7 movlhps xmm3, xmm7 shufps xmm2, xmm0, 0 movaps xmm0, xmm1 shufps xmm2, xmm2, 136 ;# constant 10001000 shufps xmm0, xmm0, 136 ;# constant 10001000 shufps xmm1, xmm1, 221 ;# constant 11011101 mov edi, [ebp + nb200_faction] ;# move ix-iz to xmm4-xmm6 movaps xmm4, [esp + nb200_ix] movaps xmm5, [esp + nb200_iy] movaps xmm6, [esp + nb200_iz] ;# calc dr subps xmm4, xmm0 subps xmm5, xmm1 subps xmm6, xmm2 ;# store dr movaps [esp + nb200_dx], xmm4 movaps [esp + nb200_dy], xmm5 movaps [esp + nb200_dz], xmm6 ;# square it mulps xmm4,xmm4 mulps xmm5,xmm5 mulps xmm6,xmm6 addps xmm4, xmm5 addps xmm4, xmm6 ;# rsq in xmm4 movaps xmm7, [esp + nb200_krf] rsqrtps xmm5, xmm4 ;# lookup seed in xmm5 movaps xmm2, xmm5 mulps xmm5, xmm5 movaps xmm1, [esp + nb200_three] mulps xmm5, xmm4 ;# rsq*lu*lu movaps xmm0, [esp + nb200_half] mulps xmm7, xmm4 ;# xmm7=krsq subps xmm1, xmm5 ;# constant 30-rsq*lu*lu mulps xmm1, xmm2 mulps xmm0, xmm1 ;# xmm0=rinv movaps xmm4, xmm0 mulps xmm4, xmm4 ;# xmm4=rinvsq movaps xmm6, xmm0 addps xmm6, xmm7 ;# xmm6=rinv+ krsq subps xmm6, [esp + nb200_crf] ;# xmm6=rinv+ krsq-crf mulps xmm6, xmm3 ;# xmm6=vcoul=qq*(rinv+ krsq-crf) mulps xmm7, [esp + nb200_two] subps xmm0, xmm7 mulps xmm3, xmm0 mulps xmm4, xmm3 ;# xmm4=total fscal addps xmm6, [esp + nb200_vctot] movaps xmm0, [esp + nb200_dx] movaps xmm1, [esp + nb200_dy] movaps xmm2, [esp + nb200_dz] movaps [esp + nb200_vctot], xmm6 mulps xmm0, xmm4 mulps xmm1, xmm4 mulps xmm2, xmm4 ;# xmm0-xmm2 contains tx-tz (partial force) ;# now update f_i movaps xmm3, [esp + nb200_fix] movaps xmm4, [esp + nb200_fiy] movaps xmm5, [esp + nb200_fiz] addps xmm3, xmm0 addps xmm4, xmm1 addps xmm5, xmm2 movaps [esp + nb200_fix], xmm3 movaps [esp + nb200_fiy], xmm4 movaps [esp + nb200_fiz], xmm5 ;# update the fj's movss xmm3, [edi + eax*4] movss xmm4, [edi + eax*4 + 4] movss xmm5, [edi + eax*4 + 8] subss xmm3, xmm0 subss xmm4, xmm1 subss xmm5, xmm2 movss [edi + eax*4], xmm3 movss [edi + eax*4 + 4], xmm4 movss [edi + eax*4 + 8], xmm5 shufps xmm0, xmm0, 225 ;# constant 11100001 shufps xmm1, xmm1, 225 ;# constant 11100001 shufps xmm2, xmm2, 225 ;# constant 11100001 movss xmm3, [edi + ebx*4] movss xmm4, [edi + ebx*4 + 4] movss xmm5, [edi + ebx*4 + 8] subss xmm3, xmm0 subss xmm4, xmm1 subss xmm5, xmm2 movss [edi + ebx*4], xmm3 movss [edi + ebx*4 + 4], xmm4 movss [edi + ebx*4 + 8], xmm5 .nb200_checksingle: mov edx, [esp + nb200_innerk] and edx, 1 jnz .nb200_dosingle jmp .nb200_updateouterdata.nb200_dosingle: mov esi, [ebp + nb200_charge] mov edi, [ebp + nb200_pos] mov ecx, [esp + nb200_innerjjnr] xorps xmm3, xmm3 mov eax, [ecx] movss xmm3, [esi + eax*4] ;# xmm3(0) has the charge lea eax, [eax + eax*2] ;# move coordinates to xmm0-xmm2 movss xmm0, [edi + eax*4] movss xmm1, [edi + eax*4 + 4] movss xmm2, [edi + eax*4 + 8] mulps xmm3, [esp + nb200_iq] xorps xmm7, xmm7 movaps xmm4, [esp + nb200_ix] movaps xmm5, [esp + nb200_iy] movaps xmm6, [esp + nb200_iz] ;# calc dr subps xmm4, xmm0 subps xmm5, xmm1 subps xmm6, xmm2 ;# store dr movaps [esp + nb200_dx], xmm4 movaps [esp + nb200_dy], xmm5 movaps [esp + nb200_dz], xmm6 ;# square it mulps xmm4,xmm4 mulps xmm5,xmm5 mulps xmm6,xmm6 addps xmm4, xmm5 addps xmm4, xmm6 ;# rsq in xmm4 movaps xmm7, [esp + nb200_krf] rsqrtps xmm5, xmm4 ;# lookup seed in xmm5 movaps xmm2, xmm5 mulps xmm5, xmm5 movaps xmm1, [esp + nb200_three] mulps xmm5, xmm4 ;# rsq*lu*lu movaps xmm0, [esp + nb200_half] mulps xmm7, xmm4 ;# xmm7=krsq subps xmm1, xmm5 ;# constant 30-rsq*lu*lu mulps xmm1, xmm2 mulps xmm0, xmm1 ;# xmm0=rinv movaps xmm4, xmm0 mulps xmm4, xmm4 ;# xmm4=rinvsq movaps xmm6, xmm0 addps xmm6, xmm7 ;# xmm6=rinv+ krsq subps xmm6, [esp + nb200_crf] ;# xmm6=rinv+ krsq-crf mulps xmm6, xmm3 ;# xmm6=vcoul mulps xmm7, [esp + nb200_two] subps xmm0, xmm7 mulps xmm3, xmm0 mulps xmm4, xmm3 ;# xmm4=total fscal addss xmm6, [esp + nb200_vctot] mov edi, [ebp + nb200_faction] movaps xmm0, [esp + nb200_dx] movaps xmm1, [esp + nb200_dy] movaps xmm2, [esp + nb200_dz] movss [esp + nb200_vctot], xmm6 mulps xmm0, xmm4 mulps xmm1, xmm4 mulps xmm2, xmm4 ;# xmm0-xmm2 contains tx-tz (partial force) ;# now update f_i movaps xmm3, [esp + nb200_fix] movaps xmm4, [esp + nb200_fiy] movaps xmm5, [esp + nb200_fiz] addss xmm3, xmm0 addss xmm4, xmm1 addss xmm5, xmm2 movaps [esp + nb200_fix], xmm3 movaps [esp + nb200_fiy], xmm4 movaps [esp + nb200_fiz], xmm5 ;# update fj movss xmm3, [edi + eax*4] movss xmm4, [edi + eax*4 + 4] movss xmm5, [edi + eax*4 + 8] subss xmm3, xmm0 subss xmm4, xmm1 subss xmm5, xmm2 movss [edi + eax*4], xmm3 movss [edi + eax*4 + 4], xmm4 movss [edi + eax*4 + 8], xmm5 .nb200_updateouterdata: mov ecx, [esp + nb200_ii3] mov edi, [ebp + nb200_faction] mov esi, [ebp + nb200_fshift] mov edx, [esp + nb200_is3] ;# accumulate i forces in xmm0, xmm1, xmm2 movaps xmm0, [esp + nb200_fix] movaps xmm1, [esp + nb200_fiy] movaps xmm2, [esp + nb200_fiz] 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 ;# increment fshift force movss xmm3, [esi + edx*4] movss xmm4, [esi + edx*4 + 4] movss xmm5, [esi + edx*4 + 8] addss xmm3, xmm0 addss xmm4, xmm1 addss xmm5, xmm2 movss [esi + edx*4], xmm3 movss [esi + edx*4 + 4], xmm4 movss [esi + edx*4 + 8], xmm5 ;# get n from stack mov esi, [esp + nb200_n] ;# get group index for i particle mov edx, [ebp + nb200_gid] ;# base of gid[] mov edx, [edx + esi*4] ;# ggid=gid[n] ;# accumulate total potential energy and update it movaps xmm7, [esp + nb200_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 + nb200_Vc] addss xmm7, [eax + edx*4] ;# move back to mem movss [eax + edx*4], xmm7 ;# finish if last mov ecx, [esp + nb200_nn1] ;# esi already loaded with n inc esi sub ecx, esi jecxz .nb200_outerend ;# not last, iterate outer loop once more! mov [esp + nb200_n], esi jmp .nb200_outer.nb200_outerend: ;# check if more outer neighborlists remain mov ecx, [esp + nb200_nri] ;# esi already loaded with n above sub ecx, esi jecxz .nb200_end ;# non-zero, do one more workunit jmp .nb200_threadloop.nb200_end: emms mov eax, [esp + nb200_nouter] mov ebx, [esp + nb200_ninner] mov ecx, [ebp + nb200_outeriter] mov edx, [ebp + nb200_inneriter] mov [ecx], eax mov [edx], ebx mov eax, [esp + nb200_salign] add esp, eax add esp, 300 pop edi pop esi pop edx pop ecx pop ebx pop eax leave ret.globl nb_kernel200nf_ia32_sse.globl _nb_kernel200nf_ia32_ssenb_kernel200nf_ia32_sse: _nb_kernel200nf_ia32_sse: .equiv nb200nf_p_nri, 8.equiv nb200nf_iinr, 12.equiv nb200nf_jindex, 16.equiv nb200nf_jjnr, 20.equiv nb200nf_shift, 24.equiv nb200nf_shiftvec, 28.equiv nb200nf_fshift, 32.equiv nb200nf_gid, 36.equiv nb200nf_pos, 40.equiv nb200nf_faction, 44.equiv nb200nf_charge, 48.equiv nb200nf_p_facel, 52.equiv nb200nf_argkrf, 56.equiv nb200nf_argcrf, 60.equiv nb200nf_Vc, 64.equiv nb200nf_type, 68.equiv nb200nf_p_ntype, 72.equiv nb200nf_vdwparam, 76.equiv nb200nf_Vvdw, 80.equiv nb200nf_p_tabscale, 84.equiv nb200nf_VFtab, 88.equiv nb200nf_invsqrta, 92.equiv nb200nf_dvda, 96.equiv nb200nf_p_gbtabscale, 100.equiv nb200nf_GBtab, 104.equiv nb200nf_p_nthreads, 108.equiv nb200nf_count, 112.equiv nb200nf_mtx, 116.equiv nb200nf_outeriter, 120.equiv nb200nf_inneriter, 124.equiv nb200nf_work, 128 ;# stack offsets for local variables ;# bottom of stack is cache-aligned for sse use .equiv nb200nf_ix, 0.equiv nb200nf_iy, 16.equiv nb200nf_iz, 32.equiv nb200nf_iq, 48.equiv nb200nf_vctot, 64.equiv nb200nf_half, 80.equiv nb200nf_three, 96.equiv nb200nf_krf, 112.equiv nb200nf_crf, 128.equiv nb200nf_is3, 144.equiv nb200nf_ii3, 148.equiv nb200nf_innerjjnr, 152.equiv nb200nf_innerk, 156.equiv nb200nf_n, 160.equiv nb200nf_nn1, 164.equiv nb200nf_nri, 168.equiv nb200nf_facel, 172.equiv nb200nf_nouter, 176.equiv nb200nf_ninner, 180.equiv nb200nf_salign, 184 push ebp mov ebp,esp push eax push ebx push ecx push edx push esi push edi sub esp, 188 ;# local stack space mov eax, esp and eax, 0xf sub esp, eax mov [esp + nb200nf_salign], eax⌨️ 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?