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
mov edx, [esp + nb133nf_innerjjnr] ;# pointer to jjnr[k] mov eax, [edx] mov ebx, [edx + 4] mov ecx, [edx + 8] mov edx, [edx + 12] ;# eax-edx=jnr1-4 add dword ptr [esp + nb133nf_innerjjnr], 16 ;# advance pointer (unrolled 4) mov esi, [ebp + nb133nf_charge] ;# base of charge[] movss xmm3, [esi + eax*4] movss xmm4, [esi + ecx*4] movss xmm6, [esi + ebx*4] movss xmm7, [esi + edx*4] shufps xmm3, xmm6, 0 shufps xmm4, xmm7, 0 shufps xmm3, xmm4, 136 ;# constant 10001000 ;# all charges in xmm3 movaps xmm4, xmm3 ;# and in xmm4 mulps xmm3, [esp + nb133nf_iqM] mulps xmm4, [esp + nb133nf_iqH] movd mm0, eax ;# use mmx registers as temp storage movd mm1, ebx movd mm2, ecx movd mm3, edx movaps [esp + nb133nf_qqM], xmm3 movaps [esp + nb133nf_qqH], xmm4 mov esi, [ebp + nb133nf_type] mov eax, [esi + eax*4] mov ebx, [esi + ebx*4] mov ecx, [esi + ecx*4] mov edx, [esi + edx*4] mov esi, [ebp + nb133nf_vdwparam] shl eax, 1 shl ebx, 1 shl ecx, 1 shl edx, 1 mov edi, [esp + nb133nf_ntia] add eax, edi add ebx, edi add ecx, edi add edx, edi movlps xmm6, [esi + eax*4] movlps xmm7, [esi + ecx*4] movhps xmm6, [esi + ebx*4] movhps xmm7, [esi + edx*4] movaps xmm4, xmm6 shufps xmm4, xmm7, 136 ;# constant 10001000 shufps xmm6, xmm7, 221 ;# constant 11011101 movd eax, mm0 movd ebx, mm1 movd ecx, mm2 movd edx, mm3 movaps [esp + nb133nf_c6], xmm4 movaps [esp + nb133nf_c12], xmm6 mov esi, [ebp + nb133nf_pos] ;# base of pos[] lea eax, [eax + eax*2] ;# replace jnr with j3 lea ebx, [ebx + ebx*2] lea ecx, [ecx + ecx*2] ;# replace jnr with j3 lea edx, [edx + edx*2] ;# move four coordinates to xmm0-xmm2 movlps xmm4, [esi + eax*4] movlps xmm5, [esi + ecx*4] movss xmm2, [esi + eax*4 + 8] movss xmm6, [esi + ecx*4 + 8] movhps xmm4, [esi + ebx*4] movhps xmm5, [esi + edx*4] movss xmm0, [esi + ebx*4 + 8] movss xmm1, [esi + edx*4 + 8] shufps xmm2, xmm0, 0 shufps xmm6, xmm1, 0 movaps xmm0, xmm4 movaps xmm1, xmm4 shufps xmm2, xmm6, 136 ;# constant 10001000 shufps xmm0, xmm5, 136 ;# constant 10001000 shufps xmm1, xmm5, 221 ;# constant 11011101 ;# move ixO-izO to xmm4-xmm6 movaps xmm4, [esp + nb133nf_ixO] movaps xmm5, [esp + nb133nf_iyO] movaps xmm6, [esp + nb133nf_izO] ;# calc dr subps xmm4, xmm0 subps xmm5, xmm1 subps xmm6, xmm2 ;# square it mulps xmm4,xmm4 mulps xmm5,xmm5 mulps xmm6,xmm6 addps xmm4, xmm5 addps xmm4, xmm6 movaps xmm7, xmm4 ;# rsqO in xmm7 ;# move ixH1-izH1 to xmm4-xmm6 movaps xmm4, [esp + nb133nf_ixH1] movaps xmm5, [esp + nb133nf_iyH1] movaps xmm6, [esp + nb133nf_izH1] ;# calc dr subps xmm4, xmm0 subps xmm5, xmm1 subps xmm6, xmm2 ;# 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 + nb133nf_ixH2] movaps xmm4, [esp + nb133nf_iyH2] movaps xmm5, [esp + nb133nf_izH2] ;# calc dr subps xmm3, xmm0 subps xmm4, xmm1 subps xmm5, xmm2 ;# 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 + nb133nf_iyM] movaps xmm4, [esp + nb133nf_izM] subps xmm3, xmm1 subps xmm4, xmm2 movaps xmm2, [esp + nb133nf_ixM] subps xmm2, xmm0 ;# 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 ;# rsqH1 - seed in xmm2 rsqrtps xmm2, xmm6 movaps xmm3, xmm2 mulps xmm2, xmm2 movaps xmm0, [esp + nb133nf_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 + nb133nf_half] movaps [esp + nb133nf_rinvH1], xmm0 ;# rinvH1 ;# rsqH2 - seed to xmm2 rsqrtps xmm2, xmm5 movaps xmm3, xmm2 mulps xmm2, xmm2 movaps xmm0, [esp + nb133nf_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 + nb133nf_half] movaps [esp + nb133nf_rinvH2], xmm0 ;# rinvH2 ;# rsqM - seed to xmm2 rsqrtps xmm2, xmm4 movaps xmm3, xmm2 mulps xmm2, xmm2 movaps xmm0, [esp + nb133nf_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 + nb133nf_half] movaps [esp + nb133nf_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 + nb133nf_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 + nb133nf_half] movaps xmm0, xmm4 ;# xmm0=rinvO mulps xmm7, xmm0 mulps xmm7, [esp + nb133nf_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 mov esi, [ebp + nb133nf_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 xmm5, xmm1 ;# xmm5=eps*Fp addps xmm5, xmm4 ;# xmm5=VV movaps xmm4, [esp + nb133nf_c6] mulps xmm5, xmm4 ;# Vvdw6 addps xmm5, [esp + nb133nf_Vvdwtot] movaps [esp + nb133nf_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 xmm5, xmm1 ;# xmm5=eps*Fp addps xmm5, xmm4 ;# xmm5=VV movaps xmm4, [esp + nb133nf_c12] mulps xmm5, xmm4 ;# Vvdw12 addps xmm5, [esp + nb133nf_Vvdwtot] movaps [esp + nb133nf_Vvdwtot], xmm5 ;# Do H1-H2-M interactions movaps xmm7, [esp + nb133nf_rinvH1] addps xmm7, [esp + nb133nf_rinvH2] movaps xmm6, [esp + nb133nf_rinvM] mulps xmm7, [esp + nb133nf_qqH] mulps xmm6, [esp + nb133nf_qqM] addps xmm7, xmm6 addps xmm7, [esp + nb133nf_vctot] movaps [esp + nb133nf_vctot], xmm7 ;# should we do one more iteration? sub dword ptr [esp + nb133nf_innerk], 4 jl .nb133nf_odd_inner jmp .nb133nf_unroll_loop.nb133nf_odd_inner: add dword ptr [esp + nb133nf_innerk], 4 jnz .nb133nf_odd_loop jmp .nb133nf_updateouterdata.nb133nf_odd_loop: mov edx, [esp + nb133nf_innerjjnr] ;# pointer to jjnr[k] mov eax, [edx] add dword ptr [esp + nb133nf_innerjjnr], 4 xorps xmm4, xmm4 ;# clear reg. movss xmm4, [esp + nb133nf_iqM] mov esi, [ebp + nb133nf_charge] movhps xmm4, [esp + nb133nf_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 + nb133nf_qqM], xmm3 ;# use dummy qq for storage xorps xmm6, xmm6 mov esi, [ebp + nb133nf_type] mov ebx, [esi + eax*4] mov esi, [ebp + nb133nf_vdwparam] shl ebx, 1 add ebx, [esp + nb133nf_ntia] movlps xmm6, [esi + ebx*4] movaps xmm7, xmm6 shufps xmm6, xmm6, 252 ;# constant 11111100 shufps xmm7, xmm7, 253 ;# constant 11111101 movaps [esp + nb133nf_c6], xmm6 movaps [esp + nb133nf_c12], xmm7 mov esi, [ebp + nb133nf_pos] lea eax, [eax + eax*2] movss xmm3, [esp + nb133nf_ixO] movss xmm4, [esp + nb133nf_iyO] movss xmm5, [esp + nb133nf_izO] movss xmm0, [esp + nb133nf_ixH1] movss xmm1, [esp + nb133nf_iyH1] movss xmm2, [esp + nb133nf_izH1] unpcklps xmm3, [esp + nb133nf_ixH2] ;# ixO ixH2 - - unpcklps xmm4, [esp + nb133nf_iyH2] ;# iyO iyH2 - - unpcklps xmm5, [esp + nb133nf_izH2] ;# izO izH2 - - unpcklps xmm0, [esp + nb133nf_ixM] ;# ixH1 ixM - - unpcklps xmm1, [esp + nb133nf_iyM] ;# iyH1 iyM - - unpcklps xmm2, [esp + nb133nf_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 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 + nb133nf_three] mulps xmm5, xmm4 ;# rsq*lu*lu movaps xmm0, [esp + nb133nf_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 + nb133nf_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 mov esi, [ebp + nb133nf_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 xmm5, xmm1 ;# xmm5=eps*Fp addss xmm5, xmm4 ;# xmm5=VV movss xmm4, [esp + nb133nf_c6] mulss xmm5, xmm4 ;# Vvdw6 ;# put scalar force on stack Update Vvdwtot directly addss xmm5, [esp + nb133nf_Vvdwtot] movss [esp + nb133nf_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 xmm5, xmm1 ;# xmm5=eps*Fp addss xmm5, xmm4 ;# xmm5=VV movss xmm4, [esp + nb133nf_c12] mulss xmm5, xmm4 ;# Vvdw12 addss xmm5, [esp + nb133nf_Vvdwtot] movss [esp + nb133nf_Vvdwtot], xmm5 mulps xmm0, [esp + nb133nf_qqM] ;# xmm0=vcoul addps xmm0, [esp + nb133nf_vctot] movaps [esp + nb133nf_vctot], xmm0 dec dword ptr [esp + nb133nf_innerk] jz .nb133nf_updateouterdata jmp .nb133nf_odd_loop.nb133nf_updateouterdata: ;# get n from stack mov esi, [esp + nb133nf_n] ;# get group index for i particle mov edx, [ebp + nb133nf_gid] ;# base of gid[] mov edx, [edx + esi*4] ;# ggid=gid[n] ;# accumulate total potential energy and update it movaps xmm7, [esp + nb133nf_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 + nb133nf_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 + nb133nf_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 + nb133nf_Vvdw] addss xmm7, [eax + edx*4] ;# move back to mem movss [eax + edx*4], xmm7 ;# finish if last mov ecx, [esp + nb133nf_nn1] ;# esi already loaded with n inc esi sub ecx, esi jecxz .nb133nf_outerend ;# not last, iterate outer loop once more! mov [esp + nb133nf_n], esi jmp .nb133nf_outer.nb133nf_outerend: ;# check if more outer neighborlists remain mov ecx, [esp + nb133nf_nri] ;# esi already loaded with n above sub ecx, esi jecxz .nb133nf_end ;# non-zero, do one more workunit jmp .nb133nf_threadloop.nb133nf_end: emms mov eax, [esp + nb133nf_nouter] mov ebx, [esp + nb133nf_ninner] mov ecx, [ebp + nb133nf_outeriter] mov edx, [ebp + nb133nf_inneriter] mov [ecx], eax mov [edx], ebx mov eax, [esp + nb133nf_salign] add esp, eax add esp, 540 pop edi pop esi pop edx pop ecx pop ebx pop eax leave ret ⌨️ 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?