nb_kernel111_ia32_sse.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 1,712 行 · 第 1/4 页
S
1,712 行
;# calc dr subps xmm4, xmm0 subps xmm5, xmm1 subps xmm6, xmm2 ;# store dr movaps [esp + nb111_dxO], xmm4 movaps [esp + nb111_dyO], xmm5 movaps [esp + nb111_dzO], xmm6 ;# 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 + nb111_ixH1] movaps xmm5, [esp + nb111_iyH1] movaps xmm6, [esp + nb111_izH1] ;# calc dr subps xmm4, xmm0 subps xmm5, xmm1 subps xmm6, xmm2 ;# store dr movaps [esp + nb111_dxH1], xmm4 movaps [esp + nb111_dyH1], xmm5 movaps [esp + nb111_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 + nb111_ixH2] movaps xmm4, [esp + nb111_iyH2] movaps xmm5, [esp + nb111_izH2] ;# calc dr subps xmm3, xmm0 subps xmm4, xmm1 subps xmm5, xmm2 ;# store dr movaps [esp + nb111_dxH2], xmm3 movaps [esp + nb111_dyH2], xmm4 movaps [esp + nb111_dzH2], xmm5 ;# square it mulps xmm3,xmm3 mulps xmm4,xmm4 mulps xmm5,xmm5 addps xmm5, xmm4 addps xmm5, xmm3 ;# rsqH2 in xmm5, rsqH1 in xmm6, rsqO in xmm7 ;# start with rsqO - seed in xmm2 rsqrtps xmm2, xmm7 movaps xmm3, xmm2 mulps xmm2, xmm2 movaps xmm4, [esp + nb111_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 + nb111_half] movaps xmm7, xmm4 ;# rinvO in xmm7 ;# rsqH1 - seed in xmm2 rsqrtps xmm2, xmm6 movaps xmm3, xmm2 mulps xmm2, xmm2 movaps xmm4, [esp + nb111_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 + nb111_half] movaps xmm6, xmm4 ;# rinvH1 in xmm6 ;# rsqH2 - seed in xmm2 rsqrtps xmm2, xmm5 movaps xmm3, xmm2 mulps xmm2, xmm2 movaps xmm4, [esp + nb111_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 + nb111_half] movaps xmm5, xmm4 ;# rinvH2 in xmm5 ;# do O interactions movaps xmm4, xmm7 mulps xmm4, xmm4 ;# xmm7=rinv, xmm4=rinvsq movaps xmm1, xmm4 mulps xmm1, xmm4 mulps xmm1, xmm4 ;# xmm1=rinvsix movaps xmm2, xmm1 mulps xmm2, xmm2 ;# xmm2=rinvtwelve mulps xmm7, [esp + nb111_qqO] ;# xmm7=vcoul mulps xmm1, [esp + nb111_c6] mulps xmm2, [esp + nb111_c12] movaps xmm3, xmm2 subps xmm3, xmm1 ;# Vvdw=Vvdw12-Vvdw6 addps xmm3, [esp + nb111_Vvdwtot] mulps xmm1, [esp + nb111_six] mulps xmm2, [esp + nb111_twelve] subps xmm2, xmm1 addps xmm2, xmm7 mulps xmm4, xmm2 ;# total fsO in xmm4 addps xmm7, [esp + nb111_vctot] movaps [esp + nb111_Vvdwtot], xmm3 movaps [esp + nb111_vctot], xmm7 movaps xmm0, [esp + nb111_dxO] movaps xmm1, [esp + nb111_dyO] movaps xmm2, [esp + nb111_dzO] mulps xmm0, xmm4 mulps xmm1, xmm4 mulps xmm2, xmm4 ;# update O forces movaps xmm3, [esp + nb111_fixO] movaps xmm4, [esp + nb111_fiyO] movaps xmm7, [esp + nb111_fizO] addps xmm3, xmm0 addps xmm4, xmm1 addps xmm7, xmm2 movaps [esp + nb111_fixO], xmm3 movaps [esp + nb111_fiyO], xmm4 movaps [esp + nb111_fizO], xmm7 ;# update j forces with water O movaps [esp + nb111_fjx], xmm0 movaps [esp + nb111_fjy], xmm1 movaps [esp + nb111_fjz], xmm2 ;# H1 interactions movaps xmm4, xmm6 mulps xmm4, xmm4 ;# xmm6=rinv, xmm4=rinvsq mulps xmm6, [esp + nb111_qqH] ;# xmm6=vcoul mulps xmm4, xmm6 ;# total fsH1 in xmm4 addps xmm6, [esp + nb111_vctot] movaps xmm0, [esp + nb111_dxH1] movaps xmm1, [esp + nb111_dyH1] movaps xmm2, [esp + nb111_dzH1] movaps [esp + nb111_vctot], xmm6 mulps xmm0, xmm4 mulps xmm1, xmm4 mulps xmm2, xmm4 ;# update H1 forces movaps xmm3, [esp + nb111_fixH1] movaps xmm4, [esp + nb111_fiyH1] movaps xmm7, [esp + nb111_fizH1] addps xmm3, xmm0 addps xmm4, xmm1 addps xmm7, xmm2 movaps [esp + nb111_fixH1], xmm3 movaps [esp + nb111_fiyH1], xmm4 movaps [esp + nb111_fizH1], xmm7 ;# update j forces with water H1 addps xmm0, [esp + nb111_fjx] addps xmm1, [esp + nb111_fjy] addps xmm2, [esp + nb111_fjz] movaps [esp + nb111_fjx], xmm0 movaps [esp + nb111_fjy], xmm1 movaps [esp + nb111_fjz], xmm2 ;# H2 interactions movaps xmm4, xmm5 mulps xmm4, xmm4 ;# xmm5=rinv, xmm4=rinvsq mulps xmm5, [esp + nb111_qqH] ;# xmm5=vcoul mulps xmm4, xmm5 ;# total fsH1 in xmm4 addps xmm5, [esp + nb111_vctot] movaps xmm0, [esp + nb111_dxH2] movaps xmm1, [esp + nb111_dyH2] movaps xmm2, [esp + nb111_dzH2] movaps [esp + nb111_vctot], xmm5 mulps xmm0, xmm4 mulps xmm1, xmm4 mulps xmm2, xmm4 ;# update H2 forces movaps xmm3, [esp + nb111_fixH2] movaps xmm4, [esp + nb111_fiyH2] movaps xmm7, [esp + nb111_fizH2] addps xmm3, xmm0 addps xmm4, xmm1 addps xmm7, xmm2 movaps [esp + nb111_fixH2], xmm3 movaps [esp + nb111_fiyH2], xmm4 movaps [esp + nb111_fizH2], xmm7 mov edi, [ebp + nb111_faction] ;# update j forces addps xmm0, [esp + nb111_fjx] addps xmm1, [esp + nb111_fjy] addps xmm2, [esp + nb111_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 + nb111_innerk], 4 jl .nb111_odd_inner jmp .nb111_unroll_loop.nb111_odd_inner: add dword ptr [esp + nb111_innerk], 4 jnz .nb111_odd_loop jmp .nb111_updateouterdata.nb111_odd_loop: mov edx, [esp + nb111_innerjjnr] ;# pointer to jjnr[k] mov eax, [edx] add dword ptr [esp + nb111_innerjjnr], 4 xorps xmm4, xmm4 movss xmm4, [esp + nb111_iqO] mov esi, [ebp + nb111_charge] movhps xmm4, [esp + nb111_iqH] movss xmm3, [esi + eax*4] ;# charge in xmm3 shufps xmm3, xmm3, 0 mulps xmm3, xmm4 movaps [esp + nb111_qqO], xmm3 ;# use oxygen qq for storage xorps xmm6, xmm6 mov esi, [ebp + nb111_type] mov ebx, [esi + eax*4] mov esi, [ebp + nb111_vdwparam] shl ebx, 1 add ebx, [esp + nb111_ntia] movlps xmm6, [esi + ebx*4] movaps xmm7, xmm6 shufps xmm6, xmm6, 252 ;# constant 11111100 shufps xmm7, xmm7, 253 ;# constant 11111101 movaps [esp + nb111_c6], xmm6 movaps [esp + nb111_c12], xmm7 mov esi, [ebp + nb111_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 + nb111_ixO] movss xmm4, [esp + nb111_iyO] movss xmm5, [esp + nb111_izO] movlps xmm6, [esp + nb111_ixH1] movlps xmm7, [esp + nb111_ixH2] unpcklps xmm6, xmm7 movlhps xmm3, xmm6 movlps xmm6, [esp + nb111_iyH1] movlps xmm7, [esp + nb111_iyH2] unpcklps xmm6, xmm7 movlhps xmm4, xmm6 movlps xmm6, [esp + nb111_izH1] movlps xmm7, [esp + nb111_izH2] unpcklps xmm6, xmm7 movlhps xmm5, xmm6 subps xmm3, xmm0 subps xmm4, xmm1 subps xmm5, xmm2 movaps [esp + nb111_dxO], xmm3 movaps [esp + nb111_dyO], xmm4 movaps [esp + nb111_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 + nb111_three] mulps xmm5, xmm4 ;# rsq*lu*lu movaps xmm0, [esp + nb111_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 movaps xmm4, xmm0 mulps xmm4, xmm4 ;# xmm4=rinvsq movaps xmm1, xmm4 mulss xmm1, xmm4 movaps xmm3, [esp + nb111_qqO] mulss xmm1, xmm4 ;# xmm1=rinvsix movaps xmm2, xmm1 mulss xmm2, xmm2 ;# xmm2=rinvtwelve mulps xmm3, xmm0 ;# xmm3=vcoul mulps xmm1, [esp + nb111_c6] mulps xmm2, [esp + nb111_c12] movaps xmm5, xmm2 subss xmm5, xmm1 ;# Vvdw=Vvdw12-Vvdw6 addps xmm5, [esp + nb111_Vvdwtot] mulss xmm1, [esp + nb111_six] mulss xmm2, [esp + nb111_twelve] subss xmm2, xmm1 addps xmm2, xmm3 mulps xmm4, xmm2 ;# xmm4=total fscal addps xmm3, [esp + nb111_vctot] movaps xmm0, [esp + nb111_dxO] movaps xmm1, [esp + nb111_dyO] movaps xmm2, [esp + nb111_dzO] movaps [esp + nb111_vctot], xmm3 movaps [esp + nb111_Vvdwtot], xmm5 mulps xmm0, xmm4 mulps xmm1, xmm4 mulps xmm2, xmm4 ;# xmm0-xmm2 contains tx-tz (partial force) movss xmm3, [esp + nb111_fixO] movss xmm4, [esp + nb111_fiyO] movss xmm5, [esp + nb111_fizO] addss xmm3, xmm0 addss xmm4, xmm1 addss xmm5, xmm2 movss [esp + nb111_fixO], xmm3 movss [esp + nb111_fiyO], xmm4 movss [esp + nb111_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 + nb111_fixH1] addss xmm4, [esp + nb111_fiyH1] addss xmm5, [esp + nb111_fizH1] movss [esp + nb111_fixH1], xmm3 movss [esp + nb111_fiyH1], xmm4 movss [esp + nb111_fizH1], xmm5 ;# updated the H1 force mov edi, [ebp + nb111_faction] shufps xmm3, xmm3, 231 ;# constant 11100111 ;# shift right shufps xmm4, xmm4, 231 ;# constant 11100111 shufps xmm5, xmm5, 231 ;# constant 11100111 addss xmm3, [esp + nb111_fixH2] addss xmm4, [esp + nb111_fiyH2] addss xmm5, [esp + nb111_fizH2] movss [esp + nb111_fixH2], xmm3 movss [esp + nb111_fiyH2], xmm4 movss [esp + nb111_fizH2], xmm5 ;# updated the H2 force ;# the fj's - start by accumulating the tx/ty/tz force in xmm0, xmm1 xorps xmm5, xmm5 movaps xmm3, xmm0 movlps xmm6, [edi + eax*4] movss xmm7, [edi + eax*4 + 8] unpcklps xmm3, xmm1 movlhps xmm3, xmm5 unpckhps xmm0, xmm1 addps xmm0, xmm3 movhlps xmm3, xmm0 addps xmm0, xmm3 ;# x,y sum in xmm0 movhlps xmm1, xmm2 addss xmm2, xmm1 shufps xmm1, xmm1, 1 addss xmm2, xmm1 ;# z sum in xmm2 subps xmm6, xmm0 subss xmm7, xmm2 movlps [edi + eax*4], xmm6 movss [edi + eax*4 + 8], xmm7 dec dword ptr [esp + nb111_innerk]⌨️ 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?