nb_kernel131_ia32_sse.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,031 行 · 第 1/4 页
S
2,031 行
;# rsqH1 - seed in xmm2 rsqrtps xmm2, xmm6 movaps xmm3, xmm2 mulps xmm2, xmm2 movaps xmm4, [esp + nb131_three] mulps xmm2, xmm6 ;# rsq*lu*lu subps xmm4, xmm2 ;# constant 30-rsq*lu*lu mulps xmm4, xmm3 ;# lu*(3-rsq*lu*lu) mulps xmm4, [esp + nb131_half] movaps [esp + nb131_rinvH1], xmm4 ;# rinvH1 in xmm6 ;# rsqH2 - seed in xmm2 rsqrtps xmm2, xmm5 movaps xmm3, xmm2 mulps xmm2, xmm2 movaps xmm4, [esp + nb131_three] mulps xmm2, xmm5 ;# rsq*lu*lu subps xmm4, xmm2 ;# constant 30-rsq*lu*lu mulps xmm4, xmm3 ;# lu*(3-rsq*lu*lu) mulps xmm4, [esp + nb131_half] movaps [esp + nb131_rinvH2], xmm4 ;# rinvH2 in xmm5 ;# do O table interactions - rsqO in xmm7. mulps xmm7, [esp + nb131_rinvO] mulps xmm7, [esp + nb131_tsc] ;# rtab movhlps xmm5, xmm7 cvttps2pi mm6, xmm7 cvttps2pi mm7, xmm5 ;# mm6/mm7 contain lu indices cvtpi2ps xmm6, mm6 cvtpi2ps xmm5, mm7 movlhps xmm6, xmm5 subps xmm7, xmm6 movaps xmm1, xmm7 ;# xmm1=eps movaps xmm2, xmm1 mulps xmm2, xmm2 ;# xmm2=eps2 pslld mm6, 3 pslld mm7, 3 movd mm0, eax movd mm1, ebx movd mm2, ecx movd mm3, edx mov esi, [ebp + nb131_VFtab] movd eax, mm6 psrlq mm6, 32 movd ecx, mm7 psrlq mm7, 32 movd ebx, mm6 movd edx, mm7 ;# dispersion movlps xmm5, [esi + eax*4] movlps xmm7, [esi + ecx*4] movhps xmm5, [esi + ebx*4] movhps xmm7, [esi + edx*4] ;# got half dispersion table movaps xmm4, xmm5 shufps xmm4, xmm7, 136 ;# constant 10001000 shufps xmm5, xmm7, 221 ;# constant 11011101 movlps xmm7, [esi + eax*4 + 8] movlps xmm3, [esi + ecx*4 + 8] movhps xmm7, [esi + ebx*4 + 8] movhps xmm3, [esi + edx*4 + 8] ;# other half of dispersion table movaps xmm6, xmm7 shufps xmm6, xmm3, 136 ;# constant 10001000 shufps xmm7, xmm3, 221 ;# constant 11011101 ;# dispersion table ready, in xmm4-xmm7 mulps xmm6, xmm1 ;# xmm6=Geps mulps xmm7, xmm2 ;# xmm7=Heps2 addps xmm5, xmm6 addps xmm5, xmm7 ;# xmm5=Fp mulps xmm7, [esp + nb131_two] ;# two*Heps2 addps xmm7, xmm6 addps xmm7, xmm5 ;# xmm7=FF mulps xmm5, xmm1 ;# xmm5=eps*Fp addps xmm5, xmm4 ;# xmm5=VV movaps xmm4, [esp + nb131_c6] mulps xmm7, xmm4 ;# fijD mulps xmm5, xmm4 ;# Vvdw6 mulps xmm7, [esp + nb131_tsc] ;# put scalar force on stack Update Vvdwtot directly addps xmm5, [esp + nb131_Vvdwtot] movaps [esp + nb131_fstmp], xmm7 movaps [esp + nb131_Vvdwtot], xmm5 ;# repulsion movlps xmm5, [esi + eax*4 + 16] movlps xmm7, [esi + ecx*4 + 16] movhps xmm5, [esi + ebx*4 + 16] movhps xmm7, [esi + edx*4 + 16] ;# got half repulsion table movaps xmm4, xmm5 shufps xmm4, xmm7, 136 ;# constant 10001000 shufps xmm5, xmm7, 221 ;# constant 11011101 movlps xmm7, [esi + eax*4 + 24] movlps xmm3, [esi + ecx*4 + 24] movhps xmm7, [esi + ebx*4 + 24] movhps xmm3, [esi + edx*4 + 24] ;# other half of repulsion table movaps xmm6, xmm7 shufps xmm6, xmm3, 136 ;# constant 10001000 shufps xmm7, xmm3, 221 ;# constant 11011101 ;# table ready, in xmm4-xmm7 mulps xmm6, xmm1 ;# xmm6=Geps mulps xmm7, xmm2 ;# xmm7=Heps2 addps xmm5, xmm6 addps xmm5, xmm7 ;# xmm5=Fp mulps xmm7, [esp + nb131_two] ;# two*Heps2 addps xmm7, xmm6 addps xmm7, xmm5 ;# xmm7=FF mulps xmm5, xmm1 ;# xmm5=eps*Fp addps xmm5, xmm4 ;# xmm5=VV movaps xmm4, [esp + nb131_c12] mulps xmm7, xmm4 ;# fijR mulps xmm5, xmm4 ;# Vvdw12 mulps xmm7, [esp + nb131_tsc] addps xmm7, [esp + nb131_fstmp] movaps [esp + nb131_fstmp], xmm7 addps xmm5, [esp + nb131_Vvdwtot] movaps [esp + nb131_Vvdwtot], xmm5 ;# analytical coulomb movaps xmm0, [esp + nb131_rinvO] movaps xmm2, xmm0 mulps xmm2, [esp + nb131_qqO] movaps xmm3, xmm2 addps xmm2, [esp + nb131_vctot] movaps [esp + nb131_vctot], xmm2 mulps xmm3, xmm0 subps xmm3, [esp + nb131_fstmp] mulps xmm3, xmm0 movaps xmm0, [esp + nb131_dxO] movaps xmm1, [esp + nb131_dyO] movaps xmm2, [esp + nb131_dzO] mulps xmm0, xmm3 mulps xmm1, xmm3 mulps xmm2, xmm3 movd eax, mm0 movd ebx, mm1 movd ecx, mm2 movd edx, mm3 ;# update O forces movaps xmm3, [esp + nb131_fixO] movaps xmm4, [esp + nb131_fiyO] movaps xmm7, [esp + nb131_fizO] addps xmm3, xmm0 addps xmm4, xmm1 addps xmm7, xmm2 movaps [esp + nb131_fixO], xmm3 movaps [esp + nb131_fiyO], xmm4 movaps [esp + nb131_fizO], xmm7 ;# update j forces with water O movaps [esp + nb131_fjx], xmm0 movaps [esp + nb131_fjy], xmm1 movaps [esp + nb131_fjz], xmm2 ;# H1 interactions movaps xmm0, [esp + nb131_rinvH1] movaps xmm2, xmm0 mulps xmm2, [esp + nb131_qqH] movaps xmm3, xmm2 addps xmm2, [esp + nb131_vctot] movaps [esp + nb131_vctot], xmm2 mulps xmm3, xmm0 mulps xmm3, xmm0 movaps xmm0, [esp + nb131_dxH1] movaps xmm1, [esp + nb131_dyH1] movaps xmm2, [esp + nb131_dzH1] mulps xmm0, xmm3 mulps xmm1, xmm3 mulps xmm2, xmm3 ;# update H1 forces movaps xmm3, [esp + nb131_fixH1] movaps xmm4, [esp + nb131_fiyH1] movaps xmm7, [esp + nb131_fizH1] addps xmm3, xmm0 addps xmm4, xmm1 addps xmm7, xmm2 movaps [esp + nb131_fixH1], xmm3 movaps [esp + nb131_fiyH1], xmm4 movaps [esp + nb131_fizH1], xmm7 ;# update j forces with water H1 addps xmm0, [esp + nb131_fjx] addps xmm1, [esp + nb131_fjy] addps xmm2, [esp + nb131_fjz] movaps [esp + nb131_fjx], xmm0 movaps [esp + nb131_fjy], xmm1 movaps [esp + nb131_fjz], xmm2 ;# H2 interactions movaps xmm0, [esp + nb131_rinvH2] movaps xmm2, xmm0 mulps xmm2, [esp + nb131_qqH] movaps xmm3, xmm2 addps xmm2, [esp + nb131_vctot] movaps [esp + nb131_vctot], xmm2 mulps xmm3, xmm0 mulps xmm3, xmm0 movaps xmm0, [esp + nb131_dxH2] movaps xmm1, [esp + nb131_dyH2] movaps xmm2, [esp + nb131_dzH2] mulps xmm0, xmm3 mulps xmm1, xmm3 mulps xmm2, xmm3 ;# update H2 forces movaps xmm3, [esp + nb131_fixH2] movaps xmm4, [esp + nb131_fiyH2] movaps xmm7, [esp + nb131_fizH2] addps xmm3, xmm0 addps xmm4, xmm1 addps xmm7, xmm2 movaps [esp + nb131_fixH2], xmm3 movaps [esp + nb131_fiyH2], xmm4 movaps [esp + nb131_fizH2], xmm7 mov edi, [ebp + nb131_faction] ;# update j forces addps xmm0, [esp + nb131_fjx] addps xmm1, [esp + nb131_fjy] addps xmm2, [esp + nb131_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 + nb131_innerk], 4 jl .nb131_odd_inner jmp .nb131_unroll_loop.nb131_odd_inner: add dword ptr [esp + nb131_innerk], 4 jnz .nb131_odd_loop jmp .nb131_updateouterdata.nb131_odd_loop: mov edx, [esp + nb131_innerjjnr] ;# pointer to jjnr[k] mov eax, [edx] add dword ptr [esp + nb131_innerjjnr], 4 xorps xmm4, xmm4 movss xmm4, [esp + nb131_iqO] mov esi, [ebp + nb131_charge] movhps xmm4, [esp + nb131_iqH] movss xmm3, [esi + eax*4] ;# charge in xmm3 shufps xmm3, xmm3, 0 mulps xmm3, xmm4 movaps [esp + nb131_qqO], xmm3 ;# use oxygen qq for storage xorps xmm6, xmm6 mov esi, [ebp + nb131_type] mov ebx, [esi + eax*4] mov esi, [ebp + nb131_vdwparam] shl ebx, 1 add ebx, [esp + nb131_ntia] movlps xmm6, [esi + ebx*4] movaps xmm7, xmm6 shufps xmm6, xmm6, 252 ;# constant 11111100 shufps xmm7, xmm7, 253 ;# constant 11111101 movaps [esp + nb131_c6], xmm6 movaps [esp + nb131_c12], xmm7 mov esi, [ebp + nb131_pos] lea eax, [eax + eax*2] ;# 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 movss xmm3, [esp + nb131_ixO] movss xmm4, [esp + nb131_iyO] movss xmm5, [esp + nb131_izO] movlps xmm6, [esp + nb131_ixH1] movlps xmm7, [esp + nb131_ixH2] unpcklps xmm6, xmm7 movlhps xmm3, xmm6 movlps xmm6, [esp + nb131_iyH1] movlps xmm7, [esp + nb131_iyH2] unpcklps xmm6, xmm7 movlhps xmm4, xmm6 movlps xmm6, [esp + nb131_izH1] movlps xmm7, [esp + nb131_izH2] unpcklps xmm6, xmm7 movlhps xmm5, xmm6 subps xmm3, xmm0 subps xmm4, xmm1 subps xmm5, xmm2 movaps [esp + nb131_dxO], xmm3 movaps [esp + nb131_dyO], xmm4 movaps [esp + nb131_dzO], xmm5 mulps xmm3, xmm3 mulps xmm4, xmm4 mulps xmm5, xmm5 addps xmm4, xmm3 addps xmm4, xmm5 ;# rsq in xmm4 rsqrtps xmm5, xmm4 ;# lookup seed in xmm5 movaps xmm2, xmm5 mulps xmm5, xmm5 movaps xmm1, [esp + nb131_three] mulps xmm5, xmm4 ;# rsq*lu*lu movaps xmm0, [esp + nb131_half] subps xmm1, xmm5 ;# constant 30-rsq*lu*lu mulps xmm1, xmm2 mulps xmm0, xmm1 ;# xmm0=rinv ;# a little trick to avoid NaNs: ;# positions 0,2,and 3 are valid, but not 1. ;# If it contains NaN it doesnt help to mult by 0, ;# So we shuffle it and copy pos 0 to pos1! shufps xmm0, xmm0, 224 ;# constant 11100000 ;# rsq still in xmm4, rinv in xmm0. mulps xmm4, xmm0 mulps xmm4, [esp + nb131_tsc] ;# rtab cvttps2pi mm6, xmm4 cvtpi2ps xmm6, mm6 subss xmm4, xmm6 movss xmm1, xmm4 ;# xmm1=eps movss xmm2, xmm1 mulss xmm2, xmm2 ;# xmm2=eps2 pslld mm6, 3 movd mm0, eax mov esi, [ebp + nb131_VFtab] movd eax, mm6 ;# dispersion movlps xmm5, [esi + eax*4] movaps xmm4, xmm5 shufps xmm4, xmm7, 136 ;# constant 10001000 shufps xmm5, xmm7, 221 ;# constant 11011101 movlps xmm7, [esi + eax*4 + 8] movaps xmm6, xmm7 shufps xmm6, xmm3, 136 ;# constant 10001000 shufps xmm7, xmm3, 221 ;# constant 11011101 ;# dispersion table ready, in xmm4-xmm7 mulss xmm6, xmm1 ;# xmm6=Geps mulss xmm7, xmm2 ;# xmm7=Heps2 addss xmm5, xmm6 addss xmm5, xmm7 ;# xmm5=Fp mulss xmm7, [esp + nb131_two] ;# two*Heps2 addss xmm7, xmm6 addss xmm7, xmm5 ;# xmm7=FF mulss xmm5, xmm1 ;# xmm5=eps*Fp addss xmm5, xmm4 ;# xmm5=VV movss xmm4, [esp + nb131_c6] mulss xmm7, xmm4 ;# fijD mulss xmm5, xmm4 ;# Vvdw6 mulss xmm7, [esp + nb131_tsc] ;# put scalar force on stack Update Vvdwtot directly addss xmm5, [esp + nb131_Vvdwtot] movss [esp + nb131_fstmp], xmm7 movss [esp + nb131_Vvdwtot], xmm5 ;# repulsion movlps xmm5, [esi + eax*4 + 16] movaps xmm4, xmm5 shufps xmm4, xmm7, 136 ;# constant 10001000 shufps xmm5, xmm7, 221 ;# constant 11011101 movlps xmm7, [esi + eax*4 + 24] movaps xmm6, xmm7 shufps xmm6, xmm3, 136 ;# constant 10001000 shufps xmm7, xmm3, 221 ;# constant 11011101 ;# table ready, in xmm4-xmm7 mulss xmm6, xmm1 ;# xmm6=Geps mulss xmm7, xmm2 ;# xmm7=Heps2 addss xmm5, xmm6 addss xmm5, xmm7 ;# xmm5=Fp mulss xmm7, [esp + nb131_two] ;# two*Heps2 addss xmm7, xmm6 addss xmm7, xmm5 ;# xmm7=FF mulss xmm5, xmm1 ;# xmm5=eps*Fp addss xmm5, xmm4 ;# xmm5=VV movss xmm4, [esp + nb131_c12] mulss xmm7, xmm4 ;# fijR mulss xmm5, xmm4 ;# Vvdw12 mulss xmm7, [esp + nb131_tsc] addss xmm7, [esp + nb131_fstmp] movss [esp + nb131_fstmp], xmm7 addss xmm5, [esp + nb131_Vvdwtot] movss [esp + nb131_Vvdwtot], xmm5 ;# coulomb in analytical form. xmm0=rinv movaps xmm2, xmm0 mulps xmm2, [esp + nb131_qqO] movaps xmm3, xmm2 addps xmm2, [esp + nb131_vctot] movaps [esp + nb131_vctot], xmm2 mulps xmm3, xmm0 subss xmm3, [esp + nb131_fstmp] mulps xmm3, xmm0 movaps xmm0, [esp + nb131_dxO] movaps xmm1, [esp + nb131_dyO] movaps xmm2, [esp + nb131_dzO] movd eax, mm0 mulps xmm0, xmm3 mulps xmm1, xmm3 mulps xmm2, xmm3 ;# xmm0-xmm2 contains tx-tz (partial force) movss xmm3, [esp + nb131_fixO] movss xmm4, [esp + nb131_fiyO] movss xmm5, [esp + nb131_fizO] addss xmm3, xmm0 addss xmm4, xmm1 addss xmm5, xmm2 movss [esp + nb131_fixO], xmm3 movss [esp + nb131_fiyO], xmm4 movss [esp + nb131_fizO], xmm5 ;# updated the O force now do the H's movaps xmm3, xmm0 movaps xmm4, xmm1 movaps xmm5, xmm2 shufps xmm3, xmm3, 230 ;# constant 11100110 ;# shift right shufps xmm4, xmm4, 230 ;# constant 11100110 shufps xmm5, xmm5, 230 ;# constant 11100110 addss xmm3, [esp + nb131_fixH1] addss xmm4, [esp + nb131_fiyH1] addss xmm5, [esp + nb131_fizH1] movss [esp + nb131_fixH1], xmm3 movss [esp + nb131_fiyH1], xmm4 movss [esp + nb131_fizH1], xmm5 ;# updated the H1 force mov edi, [ebp + nb131_faction] shufps xmm3, xmm3, 231 ;# constant 11100111 ;# shift right shufps xmm4, xmm4, 231 ;# constant 11100111 shufps xmm5, xmm5, 231 ;# constant 11100111 addss xmm3, [esp + nb131_fixH2] addss xmm4, [esp + nb131_fiyH2] addss xmm5, [esp + nb131_fizH2] movss [esp + nb131_fixH2], xmm3 movss [esp + nb131_fiyH2], xmm4 movss [esp + nb131_fizH2], xmm5 ;# updated the H2 force ;# the fj's - start by accumulating the tx/ty/tz force in xmm0, xmm1 ⌨️ 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?