nb_kernel213_ia32_sse.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 1,984 行 · 第 1/4 页
S
1,984 行
;# store dr movaps [esp + nb213_dxH1], xmm4 movaps [esp + nb213_dyH1], xmm5 movaps [esp + nb213_dzH1], xmm6 ;# square it mulps xmm4,xmm4 mulps xmm5,xmm5 mulps xmm6,xmm6 addps xmm6, xmm5 addps xmm6, xmm4 ;# rsqH1 in xmm6 ;# move ixH2-izH2 to xmm3-xmm5 movaps xmm3, [esp + nb213_ixH2] movaps xmm4, [esp + nb213_iyH2] movaps xmm5, [esp + nb213_izH2] ;# calc dr subps xmm3, xmm0 subps xmm4, xmm1 subps xmm5, xmm2 ;# store dr movaps [esp + nb213_dxH2], xmm3 movaps [esp + nb213_dyH2], xmm4 movaps [esp + nb213_dzH2], xmm5 ;# square it mulps xmm3,xmm3 mulps xmm4,xmm4 mulps xmm5,xmm5 addps xmm5, xmm4 addps xmm5, xmm3 ;# move ixM-izM to xmm2-xmm4 movaps xmm3, [esp + nb213_iyM] movaps xmm4, [esp + nb213_izM] subps xmm3, xmm1 subps xmm4, xmm2 movaps xmm2, [esp + nb213_ixM] subps xmm2, xmm0 ;# store dr movaps [esp + nb213_dxM], xmm2 movaps [esp + nb213_dyM], xmm3 movaps [esp + nb213_dzM], xmm4 ;# square it mulps xmm2,xmm2 mulps xmm3,xmm3 mulps xmm4,xmm4 addps xmm4, xmm3 addps xmm4, xmm2 ;# rsqM in xmm4, rsqH2 in xmm5, rsqH1 in xmm6, rsqO in xmm7 movaps xmm0, xmm4 movaps xmm1, xmm5 movaps xmm2, xmm6 mulps xmm0, [esp + nb213_krf] mulps xmm1, [esp + nb213_krf] mulps xmm2, [esp + nb213_krf] movaps [esp + nb213_krsqM], xmm0 movaps [esp + nb213_krsqH2], xmm1 movaps [esp + nb213_krsqH1], xmm2 ;# rsqH1 - seed in xmm2 rsqrtps xmm2, xmm6 movaps xmm3, xmm2 mulps xmm2, xmm2 movaps xmm0, [esp + nb213_three] mulps xmm2, xmm6 ;# rsq*lu*lu subps xmm0, xmm2 ;# constant 30-rsq*lu*lu mulps xmm0, xmm3 ;# lu*(3-rsq*lu*lu) mulps xmm0, [esp + nb213_half] movaps [esp + nb213_rinvH1], xmm0 ;# rinvH1 ;# rsqH2 - seed to xmm2 rsqrtps xmm2, xmm5 movaps xmm3, xmm2 mulps xmm2, xmm2 movaps xmm0, [esp + nb213_three] mulps xmm2, xmm5 ;# rsq*lu*lu subps xmm0, xmm2 ;# constant 30-rsq*lu*lu mulps xmm0, xmm3 ;# lu*(3-rsq*lu*lu) mulps xmm0, [esp + nb213_half] movaps [esp + nb213_rinvH2], xmm0 ;# rinvH2 ;# rsqM - seed to xmm2 rsqrtps xmm2, xmm4 movaps xmm3, xmm2 mulps xmm2, xmm2 movaps xmm0, [esp + nb213_three] mulps xmm2, xmm4 ;# rsq*lu*lu subps xmm0, xmm2 ;# constant 30-rsq*lu*lu mulps xmm0, xmm3 ;# lu*(3-rsq*lu*lu) mulps xmm0, [esp + nb213_half] movaps [esp + nb213_rinvM], xmm0 ;# Do the O LJ-only interaction directly. rcpps xmm2, xmm7 movaps xmm1, [esp + nb213_two] mulps xmm7, xmm2 subps xmm1, xmm7 mulps xmm2, xmm1 ;# rinvsq movaps xmm0, xmm2 mulps xmm0, xmm2 ;# r4 mulps xmm0, xmm2 ;# r6 movaps xmm1, xmm0 mulps xmm1, xmm1 ;# r12 mulps xmm0, [esp + nb213_c6] mulps xmm1, [esp + nb213_c12] movaps xmm3, xmm1 subps xmm3, xmm0 ;# Vvdw12-Vvdw6 addps xmm3, [esp + nb213_Vvdwtot] movaps [esp + nb213_Vvdwtot], xmm3 mulps xmm0, [esp + nb213_six] mulps xmm1, [esp + nb213_twelve] subps xmm1, xmm0 mulps xmm1, xmm2 ;# fscal movaps xmm3, [esp + nb213_dxO] movaps xmm4, [esp + nb213_dyO] movaps xmm5, [esp + nb213_dzO] mulps xmm3, xmm1 mulps xmm4, xmm1 mulps xmm5, xmm1 ;# tx in xmm3-xmm5 ;# update O forces movaps xmm0, [esp + nb213_fixO] movaps xmm1, [esp + nb213_fiyO] movaps xmm2, [esp + nb213_fizO] addps xmm0, xmm3 addps xmm1, xmm4 addps xmm2, xmm5 movaps [esp + nb213_fixO], xmm0 movaps [esp + nb213_fiyO], xmm1 movaps [esp + nb213_fizO], xmm2 ;# update j forces with water O movaps [esp + nb213_fjx], xmm3 movaps [esp + nb213_fjy], xmm4 movaps [esp + nb213_fjz], xmm5 ;# Do H1 interaction movaps xmm7, [esp + nb213_rinvH1] movaps xmm4, xmm7 mulps xmm4, xmm4 ;# xmm7=rinv, xmm4=rinvsq movaps xmm0, xmm7 movaps xmm1, [esp + nb213_krsqH1] addps xmm0, xmm1 subps xmm0, [esp + nb213_crf] ;# xmm0=rinv+ krsq-crf mulps xmm1, [esp + nb213_two] subps xmm7, xmm1 mulps xmm0, [esp + nb213_qqH] mulps xmm7, [esp + nb213_qqH] mulps xmm4, xmm7 ;# total fs H1 in xmm4 addps xmm0, [esp + nb213_vctot] movaps [esp + nb213_vctot], xmm0 movaps xmm0, [esp + nb213_dxH1] movaps xmm1, [esp + nb213_dyH1] movaps xmm2, [esp + nb213_dzH1] mulps xmm0, xmm4 mulps xmm1, xmm4 mulps xmm2, xmm4 ;# update H1 forces movaps xmm3, [esp + nb213_fixH1] movaps xmm4, [esp + nb213_fiyH1] movaps xmm7, [esp + nb213_fizH1] addps xmm3, xmm0 addps xmm4, xmm1 addps xmm7, xmm2 movaps [esp + nb213_fixH1], xmm3 movaps [esp + nb213_fiyH1], xmm4 movaps [esp + nb213_fizH1], xmm7 ;# update j forces with water H1 addps xmm0, [esp + nb213_fjx] addps xmm1, [esp + nb213_fjy] addps xmm2, [esp + nb213_fjz] movaps [esp + nb213_fjx], xmm0 movaps [esp + nb213_fjy], xmm1 movaps [esp + nb213_fjz], xmm2 ;# Done with H1, do H2 interactions movaps xmm7, [esp + nb213_rinvH2] movaps xmm4, xmm7 mulps xmm4, xmm4 ;# xmm7=rinv, xmm4=rinvsq movaps xmm0, xmm7 movaps xmm1, [esp + nb213_krsqH2] addps xmm0, xmm1 subps xmm0, [esp + nb213_crf] ;# xmm0=rinv+ krsq-crf mulps xmm1, [esp + nb213_two] subps xmm7, xmm1 mulps xmm0, [esp + nb213_qqH] mulps xmm7, [esp + nb213_qqH] mulps xmm4, xmm7 ;# total fs H2 in xmm4 addps xmm0, [esp + nb213_vctot] movaps [esp + nb213_vctot], xmm0 movaps xmm0, [esp + nb213_dxH2] movaps xmm1, [esp + nb213_dyH2] movaps xmm2, [esp + nb213_dzH2] mulps xmm0, xmm4 mulps xmm1, xmm4 mulps xmm2, xmm4 ;# update H2 forces movaps xmm3, [esp + nb213_fixH2] movaps xmm4, [esp + nb213_fiyH2] movaps xmm7, [esp + nb213_fizH2] addps xmm3, xmm0 addps xmm4, xmm1 addps xmm7, xmm2 movaps [esp + nb213_fixH2], xmm3 movaps [esp + nb213_fiyH2], xmm4 movaps [esp + nb213_fizH2], xmm7 addps xmm0, [esp + nb213_fjx] addps xmm1, [esp + nb213_fjy] addps xmm2, [esp + nb213_fjz] movaps [esp + nb213_fjx], xmm0 movaps [esp + nb213_fjy], xmm1 movaps [esp + nb213_fjz], xmm2 ;# Done with H2, do M interactions movaps xmm7, [esp + nb213_rinvM] movaps xmm4, xmm7 mulps xmm4, xmm4 ;# xmm7=rinv, xmm4=rinvsq movaps xmm0, xmm7 movaps xmm1, [esp + nb213_krsqM] addps xmm0, xmm1 subps xmm0, [esp + nb213_crf] ;# xmm0=rinv+ krsq-crf mulps xmm1, [esp + nb213_two] subps xmm7, xmm1 mulps xmm0, [esp + nb213_qqM] mulps xmm7, [esp + nb213_qqM] mulps xmm4, xmm7 ;# total fs M in xmm4 addps xmm0, [esp + nb213_vctot] movaps [esp + nb213_vctot], xmm0 movaps xmm0, [esp + nb213_dxM] movaps xmm1, [esp + nb213_dyM] movaps xmm2, [esp + nb213_dzM] mulps xmm0, xmm4 mulps xmm1, xmm4 mulps xmm2, xmm4 ;# update M forces movaps xmm3, [esp + nb213_fixM] movaps xmm4, [esp + nb213_fiyM] movaps xmm7, [esp + nb213_fizM] addps xmm3, xmm0 addps xmm4, xmm1 addps xmm7, xmm2 movaps [esp + nb213_fixM], xmm3 movaps [esp + nb213_fiyM], xmm4 movaps [esp + nb213_fizM], xmm7 mov edi, [ebp + nb213_faction] ;# update j forces from stored values addps xmm0, [esp + nb213_fjx] addps xmm1, [esp + nb213_fjy] addps xmm2, [esp + nb213_fjz] movlps xmm4, [edi + eax*4] movlps xmm7, [edi + ecx*4] movhps xmm4, [edi + ebx*4] movhps xmm7, [edi + edx*4] movaps xmm3, xmm4 shufps xmm3, xmm7, 136 ;# constant 10001000 shufps xmm4, xmm7, 221 ;# constant 11011101 ;# xmm3 has fjx, xmm4 has fjy subps xmm3, xmm0 subps xmm4, xmm1 ;# unpack them back for storing movaps xmm7, xmm3 unpcklps xmm7, xmm4 unpckhps xmm3, xmm4 movlps [edi + eax*4], xmm7 movlps [edi + ecx*4], xmm3 movhps [edi + ebx*4], xmm7 movhps [edi + edx*4], xmm3 ;# finally z forces movss xmm0, [edi + eax*4 + 8] movss xmm1, [edi + ebx*4 + 8] movss xmm3, [edi + ecx*4 + 8] movss xmm4, [edi + edx*4 + 8] subss xmm0, xmm2 shufps xmm2, xmm2, 229 ;# constant 11100101 subss xmm1, xmm2 shufps xmm2, xmm2, 234 ;# constant 11101010 subss xmm3, xmm2 shufps xmm2, xmm2, 255 ;# constant 11111111 subss xmm4, xmm2 movss [edi + eax*4 + 8], xmm0 movss [edi + ebx*4 + 8], xmm1 movss [edi + ecx*4 + 8], xmm3 movss [edi + edx*4 + 8], xmm4 ;# should we do one more iteration? sub dword ptr [esp + nb213_innerk], 4 jl .nb213_odd_inner jmp .nb213_unroll_loop.nb213_odd_inner: add dword ptr [esp + nb213_innerk], 4 jnz .nb213_odd_loop jmp .nb213_updateouterdata.nb213_odd_loop: mov edx, [esp + nb213_innerjjnr] ;# pointer to jjnr[k] mov eax, [edx] add dword ptr [esp + nb213_innerjjnr], 4 xorps xmm4, xmm4 ;# clear reg. movss xmm4, [esp + nb213_iqM] mov esi, [ebp + nb213_charge] movhps xmm4, [esp + nb213_iqH] ;# [qM 0 qH qH] shufps xmm4, xmm4, 41 ;# [0 qH qH qM] movss xmm3, [esi + eax*4] ;# charge in xmm3 shufps xmm3, xmm3, 0 mulps xmm3, xmm4 movaps [esp + nb213_qqM], xmm3 ;# use dummy qq for storage xorps xmm6, xmm6 mov esi, [ebp + nb213_type] mov ebx, [esi + eax*4] mov esi, [ebp + nb213_vdwparam] shl ebx, 1 add ebx, [esp + nb213_ntia] movlps xmm6, [esi + ebx*4] movaps xmm7, xmm6 shufps xmm6, xmm6, 252 ;# constant 11111100 shufps xmm7, xmm7, 253 ;# constant 11111101 movaps [esp + nb213_c6], xmm6 movaps [esp + nb213_c12], xmm7 mov esi, [ebp + nb213_pos] lea eax, [eax + eax*2] movss xmm3, [esp + nb213_ixO] movss xmm4, [esp + nb213_iyO] movss xmm5, [esp + nb213_izO] movss xmm0, [esp + nb213_ixH1] movss xmm1, [esp + nb213_iyH1] movss xmm2, [esp + nb213_izH1] unpcklps xmm3, [esp + nb213_ixH2] ;# ixO ixH2 - - unpcklps xmm4, [esp + nb213_iyH2] ;# iyO iyH2 - - unpcklps xmm5, [esp + nb213_izH2] ;# izO izH2 - - unpcklps xmm0, [esp + nb213_ixM] ;# ixH1 ixM - - unpcklps xmm1, [esp + nb213_iyM] ;# iyH1 iyM - - unpcklps xmm2, [esp + nb213_izM] ;# izH1 izM - - unpcklps xmm3, xmm0 ;# ixO ixH1 ixH2 ixM unpcklps xmm4, xmm1 ;# same for y unpcklps xmm5, xmm2 ;# same for z ;# move j coords to xmm0-xmm2 movss xmm0, [esi + eax*4] movss xmm1, [esi + eax*4 + 4] movss xmm2, [esi + eax*4 + 8] shufps xmm0, xmm0, 0 shufps xmm1, xmm1, 0 shufps xmm2, xmm2, 0 subps xmm3, xmm0 subps xmm4, xmm1 subps xmm5, xmm2 ;# use O distances for storage movaps [esp + nb213_dxO], xmm3 movaps [esp + nb213_dyO], xmm4 movaps [esp + nb213_dzO], xmm5 mulps xmm3, xmm3 mulps xmm4, xmm4 mulps xmm5, xmm5 addps xmm4, xmm3 addps xmm4, xmm5 ;# rsq in xmm4 movaps xmm0, xmm4 mulps xmm0, [esp + nb213_krf] movaps [esp + nb213_krsqM], xmm0 rsqrtps xmm5, xmm4 ;# lookup seed in xmm5 movaps xmm2, xmm5 mulps xmm5, xmm5 movaps xmm1, [esp + nb213_three] mulps xmm5, xmm4 ;# rsq*lu*lu movaps xmm0, [esp + nb213_half] subps xmm1, xmm5 ;# constant 30-rsq*lu*lu mulps xmm1, xmm2 mulps xmm0, xmm1 ;# xmm0=rinv movaps xmm4, xmm0 movaps xmm1, xmm0 movaps xmm5, xmm0 mulps xmm4, xmm4 ;# xmm1=rinv, xmm4=rinvsq movaps xmm3, [esp + nb213_krsqM] addps xmm5, xmm3 ;# xmm0=rinv+ krsq subps xmm5, [esp + nb213_crf] ;# xmm0=rinv+ krsq-crf mulps xmm3, [esp + nb213_two] subps xmm1, xmm3 ;# xmm1=rinv-2*krsq movaps xmm7, xmm5 mulps xmm7, [esp + nb213_qqM] ;# xmm0=vcoul mulps xmm1, [esp + nb213_qqM] ;# xmm1=coul part of fs movaps xmm6, xmm1 addps xmm7, [esp + nb213_vctot] movaps [esp + nb213_vctot], xmm7 movaps xmm1, xmm0 mulps xmm1, xmm1 movaps xmm2, xmm1 mulss xmm1, xmm1 mulss xmm1, xmm2 ;# xmm1=rinvsix xorps xmm4, xmm4 movss xmm4, xmm1 mulss xmm4, xmm4 ;# xmm4=rinvtwelve mulss xmm1, [esp + nb213_c6] mulss xmm4, [esp + nb213_c12] movaps xmm3, xmm4 subss xmm3, xmm1 ;# xmm3=Vvdw12-Vvdw6 mulss xmm1, [esp + nb213_six] mulss xmm4, [esp + nb213_twelve] subss xmm4, xmm1 addss xmm3, [esp + nb213_Vvdwtot] movss [esp + nb213_Vvdwtot], xmm3 addps xmm4, xmm6 mulps xmm4, xmm0 mulps xmm4, xmm0 ;# fscal movaps xmm0, [esp + nb213_dxO] movaps xmm1, [esp + nb213_dyO] movaps xmm2, [esp + nb213_dzO] mulps xmm0, xmm4 mulps xmm1, xmm4 mulps xmm2, xmm4 ;# xmm0-xmm2 now contains tx-tz (partial force) movss xmm3, [esp + nb213_fixO] movss xmm4, [esp + nb213_fiyO] movss xmm5, [esp + nb213_fizO] addss xmm3, xmm0 addss xmm4, xmm1 addss xmm5, xmm2 movss [esp + nb213_fixO], xmm3 movss [esp + nb213_fiyO], xmm4 movss [esp + nb213_fizO], xmm5 ;# updated the O force now do the H's movaps xmm3, xmm0 movaps xmm4, xmm1 movaps xmm5, xmm2 shufps xmm3, xmm3, 0x39 ;# shift right shufps xmm4, xmm4, 0x39 shufps xmm5, xmm5, 0x39 addss xmm3, [esp + nb213_fixH1] addss xmm4, [esp + nb213_fiyH1] addss xmm5, [esp + nb213_fizH1] movss [esp + nb213_fixH1], xmm3 movss [esp + nb213_fiyH1], xmm4 movss [esp + nb213_fizH1], xmm5 ;# updated the H1 force shufps xmm3, xmm3, 0x39 shufps xmm4, xmm4, 0x39 shufps xmm5, xmm5, 0x39 addss xmm3, [esp + nb213_fixH2] addss xmm4, [esp + nb213_fiyH2] addss xmm5, [esp + nb213_fizH2] movss [esp + nb213_fixH2], xmm3 movss [esp + nb213_fiyH2], xmm4 movss [esp + nb213_fizH2], xmm5 ;# updated the H2 force mov edi, [ebp + nb213_faction] shufps xmm3, xmm3, 0x39 shufps xmm4, xmm4, 0x39 shufps xmm5, xmm5, 0x39 addss xmm3, [esp + nb213_fixM] addss xmm4, [esp + nb213_fiyM] addss xmm5, [esp + nb213_fizM] movss [esp + nb213_fixM], xmm3 movss [esp + nb213_fiyM], xmm4 movss [esp + nb213_fizM], xmm5 ;# updated the M force ;# the fj's - move in from mem start by acc. tx/ty/tz in xmm0, xmm1 movlps xmm6, [edi + eax*4] movss xmm7, [edi + eax*4 + 8] movhlps xmm3, xmm0 movhlps xmm4, xmm1 movhlps xmm5, xmm2⌨️ 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?