nb_kernel133_ia32_sse.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,185 行 · 第 1/4 页
S
2,185 行
;# rsqH2 - seed to xmm2 rsqrtps xmm2, xmm5 movaps xmm3, xmm2 mulps xmm2, xmm2 movaps xmm0, [esp + nb133_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 + nb133_half] movaps [esp + nb133_rinvH2], xmm0 ;# rinvH2 ;# rsqM - seed to xmm2 rsqrtps xmm2, xmm4 movaps xmm3, xmm2 mulps xmm2, xmm2 movaps xmm0, [esp + nb133_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 + nb133_half] movaps [esp + nb133_rinvM], xmm0 ;# Do the O LJ-only interaction directly. ;# rsqO is in xmm7 rsqrtps xmm2, xmm7 movaps xmm3, xmm2 mulps xmm2, xmm2 movaps xmm4, [esp + nb133_three] mulps xmm2, xmm7 ;# rsq*lu*lu subps xmm4, xmm2 ;# constant 30-rsq*lu*lu mulps xmm4, xmm3 ;# lu*(3-rsq*lu*lu) mulps xmm4, [esp + nb133_half] movaps xmm0, xmm4 ;# xmm0=rinvO mulps xmm7, xmm0 mulps xmm7, [esp + nb133_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 + nb133_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 + nb133_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 + nb133_c6] mulps xmm7, xmm4 ;# fijD mulps xmm5, xmm4 ;# Vvdw6 mulps xmm7, [esp + nb133_tsc] ;# put scalar force on stack Update Vvdwtot directly addps xmm5, [esp + nb133_Vvdwtot] movaps [esp + nb133_fstmp], xmm7 movaps [esp + nb133_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 + nb133_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 + nb133_c12] mulps xmm7, xmm4 ;# fijR mulps xmm5, xmm4 ;# Vvdw12 mulps xmm7, [esp + nb133_tsc] addps xmm7, [esp + nb133_fstmp] addps xmm5, [esp + nb133_Vvdwtot] movaps [esp + nb133_Vvdwtot], xmm5 movd eax, mm0 movd ebx, mm1 movd ecx, mm2 movd edx, mm3 xorps xmm1, xmm1 mulps xmm7, xmm0 subps xmm1, xmm7 movaps xmm3, [esp + nb133_dxO] movaps xmm4, [esp + nb133_dyO] movaps xmm5, [esp + nb133_dzO] mulps xmm3, xmm1 mulps xmm4, xmm1 mulps xmm5, xmm1 ;# tx in xmm3-xmm5 ;# update O forces movaps xmm0, [esp + nb133_fixO] movaps xmm1, [esp + nb133_fiyO] movaps xmm2, [esp + nb133_fizO] addps xmm0, xmm3 addps xmm1, xmm4 addps xmm2, xmm5 movaps [esp + nb133_fixO], xmm0 movaps [esp + nb133_fiyO], xmm1 movaps [esp + nb133_fizO], xmm2 ;# update j forces with water O movaps [esp + nb133_fjx], xmm3 movaps [esp + nb133_fjy], xmm4 movaps [esp + nb133_fjz], xmm5 ;# Do H1 interaction movaps xmm7, [esp + nb133_rinvH1] movaps xmm4, xmm7 mulps xmm4, xmm4 ;# xmm7=rinv, xmm4=rinvsq mulps xmm7, [esp + nb133_qqH] mulps xmm4, xmm7 ;# total fs H1 in xmm4 addps xmm7, [esp + nb133_vctot] movaps [esp + nb133_vctot], xmm7 movaps xmm0, [esp + nb133_dxH1] movaps xmm1, [esp + nb133_dyH1] movaps xmm2, [esp + nb133_dzH1] mulps xmm0, xmm4 mulps xmm1, xmm4 mulps xmm2, xmm4 ;# update H1 forces movaps xmm3, [esp + nb133_fixH1] movaps xmm4, [esp + nb133_fiyH1] movaps xmm7, [esp + nb133_fizH1] addps xmm3, xmm0 addps xmm4, xmm1 addps xmm7, xmm2 movaps [esp + nb133_fixH1], xmm3 movaps [esp + nb133_fiyH1], xmm4 movaps [esp + nb133_fizH1], xmm7 ;# update j forces with water H1 addps xmm0, [esp + nb133_fjx] addps xmm1, [esp + nb133_fjy] addps xmm2, [esp + nb133_fjz] movaps [esp + nb133_fjx], xmm0 movaps [esp + nb133_fjy], xmm1 movaps [esp + nb133_fjz], xmm2 ;# Done with H1, do H2 interactions movaps xmm7, [esp + nb133_rinvH2] movaps xmm4, xmm7 mulps xmm4, xmm4 ;# xmm7=rinv, xmm4=rinvsq mulps xmm7, [esp + nb133_qqH] mulps xmm4, xmm7 ;# total fs H2 in xmm4 addps xmm7, [esp + nb133_vctot] movaps [esp + nb133_vctot], xmm7 movaps xmm0, [esp + nb133_dxH2] movaps xmm1, [esp + nb133_dyH2] movaps xmm2, [esp + nb133_dzH2] mulps xmm0, xmm4 mulps xmm1, xmm4 mulps xmm2, xmm4 ;# update H2 forces movaps xmm3, [esp + nb133_fixH2] movaps xmm4, [esp + nb133_fiyH2] movaps xmm7, [esp + nb133_fizH2] addps xmm3, xmm0 addps xmm4, xmm1 addps xmm7, xmm2 movaps [esp + nb133_fixH2], xmm3 movaps [esp + nb133_fiyH2], xmm4 movaps [esp + nb133_fizH2], xmm7 addps xmm0, [esp + nb133_fjx] addps xmm1, [esp + nb133_fjy] addps xmm2, [esp + nb133_fjz] movaps [esp + nb133_fjx], xmm0 movaps [esp + nb133_fjy], xmm1 movaps [esp + nb133_fjz], xmm2 ;# Done with H2, do M interactions movaps xmm7, [esp + nb133_rinvM] movaps xmm4, xmm7 mulps xmm4, xmm4 ;# xmm7=rinv, xmm4=rinvsq mulps xmm7, [esp + nb133_qqM] mulps xmm4, xmm7 ;# total fs M in xmm4 addps xmm7, [esp + nb133_vctot] movaps [esp + nb133_vctot], xmm7 movaps xmm0, [esp + nb133_dxM] movaps xmm1, [esp + nb133_dyM] movaps xmm2, [esp + nb133_dzM] mulps xmm0, xmm4 mulps xmm1, xmm4 mulps xmm2, xmm4 ;# update M forces movaps xmm3, [esp + nb133_fixM] movaps xmm4, [esp + nb133_fiyM] movaps xmm7, [esp + nb133_fizM] addps xmm3, xmm0 addps xmm4, xmm1 addps xmm7, xmm2 movaps [esp + nb133_fixM], xmm3 movaps [esp + nb133_fiyM], xmm4 movaps [esp + nb133_fizM], xmm7 mov edi, [ebp + nb133_faction] ;# update j forces from stored values addps xmm0, [esp + nb133_fjx] addps xmm1, [esp + nb133_fjy] addps xmm2, [esp + nb133_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 + nb133_innerk], 4 jl .nb133_odd_inner jmp .nb133_unroll_loop.nb133_odd_inner: add dword ptr [esp + nb133_innerk], 4 jnz .nb133_odd_loop jmp .nb133_updateouterdata.nb133_odd_loop: mov edx, [esp + nb133_innerjjnr] ;# pointer to jjnr[k] mov eax, [edx] add dword ptr [esp + nb133_innerjjnr], 4 xorps xmm4, xmm4 ;# clear reg. movss xmm4, [esp + nb133_iqM] mov esi, [ebp + nb133_charge] movhps xmm4, [esp + nb133_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 + nb133_qqM], xmm3 ;# use dummy qq for storage xorps xmm6, xmm6 mov esi, [ebp + nb133_type] mov ebx, [esi + eax*4] mov esi, [ebp + nb133_vdwparam] shl ebx, 1 add ebx, [esp + nb133_ntia] movlps xmm6, [esi + ebx*4] movaps xmm7, xmm6 shufps xmm6, xmm6, 252 ;# constant 11111100 shufps xmm7, xmm7, 253 ;# constant 11111101 movaps [esp + nb133_c6], xmm6 movaps [esp + nb133_c12], xmm7 mov esi, [ebp + nb133_pos] lea eax, [eax + eax*2] movss xmm3, [esp + nb133_ixO] movss xmm4, [esp + nb133_iyO] movss xmm5, [esp + nb133_izO] movss xmm0, [esp + nb133_ixH1] movss xmm1, [esp + nb133_iyH1] movss xmm2, [esp + nb133_izH1] unpcklps xmm3, [esp + nb133_ixH2] ;# ixO ixH2 - - unpcklps xmm4, [esp + nb133_iyH2] ;# iyO iyH2 - - unpcklps xmm5, [esp + nb133_izH2] ;# izO izH2 - - unpcklps xmm0, [esp + nb133_ixM] ;# ixH1 ixM - - unpcklps xmm1, [esp + nb133_iyM] ;# iyH1 iyM - - unpcklps xmm2, [esp + nb133_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 + nb133_dxO], xmm3 movaps [esp + nb133_dyO], xmm4 movaps [esp + nb133_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 + nb133_three] mulps xmm5, xmm4 ;# rsq*lu*lu movaps xmm0, [esp + nb133_half] subps xmm1, xmm5 ;# constant 30-rsq*lu*lu mulps xmm1, xmm2 mulps xmm0, xmm1 ;# xmm0=rinv, xmm4=rsq #; LJ table interaction mulps xmm4, xmm0 mulps xmm4, [esp + nb133_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 + nb133_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 + nb133_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 + nb133_c6] mulss xmm7, xmm4 ;# fijD mulss xmm5, xmm4 ;# Vvdw6 mulss xmm7, [esp + nb133_tsc] ;# put scalar force on stack Update Vvdwtot directly addss xmm5, [esp + nb133_Vvdwtot] movss [esp + nb133_fstmp], xmm7 movss [esp + nb133_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 + nb133_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 + nb133_c12] mulss xmm7, xmm4 ;# fijR mulss xmm5, xmm4 ;# Vvdw12 mulss xmm7, [esp + nb133_tsc] addss xmm7, [esp + nb133_fstmp] movss [esp + nb133_fstmp], xmm7 addss xmm5, [esp + nb133_Vvdwtot] movss [esp + nb133_Vvdwtot], xmm5 movd eax, mm0 movaps xmm4, xmm0 mulps xmm4, [esp + nb133_qqM] movaps xmm2, xmm4 mulps xmm4, xmm0 subss xmm4, [esp + nb133_fstmp] mulps xmm4, xmm0 addps xmm2, [esp + nb133_vctot] movaps [esp + nb133_vctot], xmm2 movaps xmm0, [esp + nb133_dxO] movaps xmm1, [esp + nb133_dyO] movaps xmm2, [esp + nb133_dzO] mulps xmm0, xmm4 mulps xmm1, xmm4 mulps xmm2, xmm4 ;# xmm0-xmm2 now contains tx-tz (partial force) movss xmm3, [esp + nb133_fixO] movss xmm4, [esp + nb133_fiyO] movss xmm5, [esp + nb133_fizO] addss xmm3, xmm0 addss xmm4, xmm1 addss xmm5, xmm2 movss [esp + nb133_fixO], xmm3 movss [esp + nb133_fiyO], xmm4 movss [esp + nb133_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 + nb133_fixH1] addss xmm4, [esp + nb133_fiyH1] addss xmm5, [esp + nb133_fizH1] movss [esp + nb133_fixH1], xmm3 movss [esp + nb133_fiyH1], xmm4 movss [esp + nb133_fizH1], xmm5 ;# updated the H1 force shufps xmm3, xmm3, 0x39 shufps xmm4, xmm4, 0x39 shufps xmm5, xmm5, 0x39 addss xmm3, [esp + nb133_fixH2] addss xmm4, [esp + nb133_fiyH2] addss xmm5, [esp + nb133_fizH2] movss [esp + nb133_fixH2], xmm3 movss [esp + nb133_fiyH2], xmm4 movss [esp + nb133_fizH2], xmm5 ;# updated the H2 force mov edi, [ebp + nb133_faction] shufps xmm3, xmm3, 0x39 shufps xmm4, xmm4, 0x39 shufps xmm5, xmm5, 0x39 addss xmm3, [esp + nb133_fixM] addss xmm4, [esp + nb133_fiyM] addss xmm5, [esp + nb133_fizM] movss [esp + nb133_fixM], xmm3 movss [esp + nb133_fiyM], xmm4 movss [esp + nb133_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]⌨️ 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?