nb_kernel331_ia32_sse2.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,337 行 · 第 1/5 页
S
2,337 行
movapd xmm6, [esp + nb331_izO] ;# calc dr subsd xmm4, xmm0 subsd xmm5, xmm1 subsd xmm6, xmm2 ;# store dr movapd [esp + nb331_dxO], xmm4 movapd [esp + nb331_dyO], xmm5 movapd [esp + nb331_dzO], xmm6 ;# square it mulsd xmm4,xmm4 mulsd xmm5,xmm5 mulsd xmm6,xmm6 addsd xmm4, xmm5 addsd xmm4, xmm6 movapd xmm7, xmm4 ;# rsqO in xmm7 ;# move ixH1-izH1 to xmm4-xmm6 movapd xmm4, [esp + nb331_ixH1] movapd xmm5, [esp + nb331_iyH1] movapd xmm6, [esp + nb331_izH1] ;# calc dr subsd xmm4, xmm0 subsd xmm5, xmm1 subsd xmm6, xmm2 ;# store dr movapd [esp + nb331_dxH1], xmm4 movapd [esp + nb331_dyH1], xmm5 movapd [esp + nb331_dzH1], xmm6 ;# square it mulsd xmm4,xmm4 mulsd xmm5,xmm5 mulsd xmm6,xmm6 addsd xmm6, xmm5 addsd xmm6, xmm4 ;# rsqH1 in xmm6 ;# move ixH2-izH2 to xmm3-xmm5 movapd xmm3, [esp + nb331_ixH2] movapd xmm4, [esp + nb331_iyH2] movapd xmm5, [esp + nb331_izH2] ;# calc dr subsd xmm3, xmm0 subsd xmm4, xmm1 subsd xmm5, xmm2 ;# store dr movapd [esp + nb331_dxH2], xmm3 movapd [esp + nb331_dyH2], xmm4 movapd [esp + nb331_dzH2], xmm5 ;# square it mulsd xmm3,xmm3 mulsd xmm4,xmm4 mulsd xmm5,xmm5 addsd xmm5, xmm4 addsd xmm5, xmm3 ;# rsqH2 in xmm5, rsqH1 in xmm6, rsqO in xmm7 ;# start with rsqO - put seed in xmm2 cvtsd2ss xmm2, xmm7 rsqrtss xmm2, xmm2 cvtss2sd xmm2, xmm2 movapd xmm3, xmm2 mulsd xmm2, xmm2 movapd xmm4, [esp + nb331_three] mulsd xmm2, xmm7 ;# rsq*lu*lu subsd xmm4, xmm2 ;# 30-rsq*lu*lu mulsd xmm4, xmm3 ;# lu*(3-rsq*lu*lu) mulsd xmm4, [esp + nb331_half] ;# iter1 ( new lu) movapd xmm2, xmm7 movapd xmm3, xmm4 mulsd xmm4, xmm4 ;# lu*lu mulsd xmm2, xmm4 ;# rsq*lu*lu movapd xmm4, [esp + nb331_three] subsd xmm4, xmm2 ;# 3-rsq*lu*lu mulsd xmm4, xmm3 ;# lu*( 3-rsq*lu*lu) mulsd xmm4, [esp + nb331_half] ;# rinv movapd [esp + nb331_rinvO], xmm4 ;# rinvO in xmm4 mulsd xmm7, xmm4 movapd [esp + nb331_rO], xmm7 ;# r in xmm7 ;# rsqH1 - seed in xmm2 cvtsd2ss xmm2, xmm6 rsqrtss xmm2, xmm2 cvtss2sd xmm2, xmm2 movapd xmm3, xmm2 mulsd xmm2, xmm2 movapd xmm4, [esp + nb331_three] mulsd xmm2, xmm6 ;# rsq*lu*lu subsd xmm4, xmm2 ;# 30-rsq*lu*lu mulsd xmm4, xmm3 ;# lu*(3-rsq*lu*lu) mulsd xmm4, [esp + nb331_half] ;# iter1 ( new lu) movapd xmm2, xmm6 movapd xmm3, xmm4 mulsd xmm4, xmm4 ;# lu*lu mulsd xmm2, xmm4 ;# rsq*lu*lu movapd xmm4, [esp + nb331_three] subsd xmm4, xmm2 ;# 3-rsq*lu*lu mulsd xmm4, xmm3 ;# lu*( 3-rsq*lu*lu) mulsd xmm4, [esp + nb331_half] ;# rinv movapd [esp + nb331_rinvH1], xmm4 ;# rinvH1 mulsd xmm6, xmm4 movapd [esp + nb331_rH1], xmm6 ;# rH1 ;# rsqH2 - seed in xmm2 cvtsd2ss xmm2, xmm5 rsqrtss xmm2, xmm2 cvtss2sd xmm2, xmm2 movapd xmm3, xmm2 mulsd xmm2, xmm2 movapd xmm4, [esp + nb331_three] mulsd xmm2, xmm5 ;# rsq*lu*lu subsd xmm4, xmm2 ;# 30-rsq*lu*lu mulsd xmm4, xmm3 ;# lu*(3-rsq*lu*lu) mulsd xmm4, [esp + nb331_half] ;# iter1 ( new lu) movapd xmm2, xmm5 movapd xmm3, xmm4 mulsd xmm4, xmm4 ;# lu*lu mulsd xmm2, xmm4 ;# rsq*lu*lu movapd xmm4, [esp + nb331_three] subsd xmm4, xmm2 ;# 3-rsq*lu*lu mulsd xmm4, xmm3 ;# lu*( 3-rsq*lu*lu) mulsd xmm4, [esp + nb331_half] ;# rinv movapd [esp + nb331_rinvH2], xmm4 ;# rinv mulsd xmm5, xmm4 movapd [esp + nb331_rH2], xmm5 ;# r ;# do O interactions movd mm0, eax ;# rO is still in xmm7 mulsd xmm7, [esp + nb331_tsc] cvttsd2si eax, xmm7 ;# lu idx cvtsi2sd xmm6, eax subsd xmm7, xmm6 movapd xmm1, xmm7 ;# xmm1=eps movapd xmm2, xmm7 mulsd xmm2, xmm2 ;# xmm2=eps2 shl eax, 2 mov esi, [ebp + nb331_VFtab] lea eax, [eax + eax*2] ;# idx *= 3 (total *=12 now) movsd xmm4, [esi + eax*8 + 0] ;# Y1 movsd xmm5, [esi + eax*8 + 8] ;# F1 movsd xmm6, [esi + eax*8 + 16] ;# G1 movsd xmm7, [esi + eax*8 + 24] ;# H1 ;# coulomb table ready, in xmm4-xmm7 mulsd xmm6, xmm1 ;# xmm6=Geps mulsd xmm7, xmm2 ;# xmm7=Heps2 addsd xmm5, xmm6 ;# F+Geps addsd xmm5, xmm7 ;# xmm5=Fp=F+Geps+Heps2 mulsd xmm7, [esp + nb331_two] ;# two*Heps2 addsd xmm7, xmm6 ;# Geps + 2*Heps2 addsd xmm7, xmm5 ;# xmm7=FF mulsd xmm5, xmm1 ;# xmm5=eps*Fp addsd xmm5, xmm4 ;# xmm5=VV movapd xmm0, [esp + nb331_qqO] mulsd xmm5, xmm0 ;# vcoul=qq*VV mulsd xmm0, xmm7 ;# fijC=FF*qq ;# at this point xmm5 contains vcoul and xmm0 fijC ;# increment vcoul - then we can get rid of mm5 ;# update vctot addsd xmm5, [esp + nb331_vctot] movlpd [esp + nb331_vctot], xmm5 ;# Dispersion movsd xmm4, [esi + eax*8 + 32] ;# Y1 movsd xmm5, [esi + eax*8 + 40] ;# F1 movsd xmm6, [esi + eax*8 + 48] ;# G1 movsd xmm7, [esi + eax*8 + 56] ;# H1 ;# Dispersion table ready, in xmm4-xmm7 mulsd xmm6, xmm1 ;# xmm6=Geps mulsd xmm7, xmm2 ;# xmm7=Heps2 addsd xmm5, xmm6 addsd xmm5, xmm7 ;# xmm5=Fp mulsd xmm7, [esp + nb331_two] ;# two*Heps2 addsd xmm7, xmm6 addsd xmm7, xmm5 ;# xmm7=FF mulsd xmm5, xmm1 ;# xmm5=eps*Fp addsd xmm5, xmm4 ;# xmm5=VV movapd xmm4, [esp + nb331_c6] mulsd xmm7, xmm4 ;# fijD mulsd xmm5, xmm4 ;# Vvdw6 addsd xmm0, xmm7 ;# fscal summation addsd xmm5, [esp + nb331_Vvdwtot] movsd [esp + nb331_Vvdwtot], xmm5 ;# Repulsion movsd xmm4, [esi + eax*8 + 64] ;# Y1 movsd xmm5, [esi + eax*8 + 72] ;# F1 movsd xmm6, [esi + eax*8 + 80] ;# G1 movsd xmm7, [esi + eax*8 + 88] ;# H1 ;# Repulsion table ready, in xmm4-xmm7 mulsd xmm6, xmm1 ;# xmm6=Geps mulsd xmm7, xmm2 ;# xmm7=Heps2 addsd xmm5, xmm6 addsd xmm5, xmm7 ;# xmm5=Fp mulsd xmm7, [esp + nb331_two] ;# two*Heps2 addsd xmm7, xmm6 addsd xmm7, xmm5 ;# xmm7=FF mulsd xmm5, xmm1 ;# xmm5=eps*Fp addsd xmm5, xmm4 ;# xmm5=VV movapd xmm4, [esp + nb331_c12] mulsd xmm7, xmm4 ;# fijR mulsd xmm5, xmm4 ;# Vvdw12 addsd xmm7, xmm0 ;# fscal summation mulsd xmm7, [esp + nb331_tsc] mulsd xmm7, [esp + nb331_rinvO] addsd xmm5, [esp + nb331_Vvdwtot] movsd [esp + nb331_Vvdwtot], xmm5 xorpd xmm4, xmm4 subsd xmm4, xmm7 movapd xmm0, [esp + nb331_dxO] movapd xmm1, [esp + nb331_dyO] movapd xmm2, [esp + nb331_dzO] mulsd xmm0, xmm4 mulsd xmm1, xmm4 mulsd xmm2, xmm4 ;# tx in xmm0-xmm2 ;# update O forces movapd xmm3, [esp + nb331_fixO] movapd xmm4, [esp + nb331_fiyO] movapd xmm7, [esp + nb331_fizO] addsd xmm3, xmm0 addsd xmm4, xmm1 addsd xmm7, xmm2 movlpd [esp + nb331_fixO], xmm3 movlpd [esp + nb331_fiyO], xmm4 movlpd [esp + nb331_fizO], xmm7 ;# update j forces with water O movlpd [esp + nb331_fjx], xmm0 movlpd [esp + nb331_fjy], xmm1 movlpd [esp + nb331_fjz], xmm2 ;# Done with O interactions - now H1! movapd xmm7, [esp + nb331_rH1] mulpd xmm7, [esp + nb331_tsc] cvttsd2si eax, xmm7 ;# mm6 = lu idx cvtsi2sd xmm6, eax subpd xmm7, xmm6 movapd xmm1, xmm7 ;# xmm1=eps movapd xmm2, xmm1 mulpd xmm2, xmm2 ;# xmm2=eps2 shl eax, 2 ;# idx *= 4 mov esi, [ebp + nb331_VFtab] lea eax, [eax + eax*2] ;# idx *= 3 (total *=12 now) movsd xmm4, [esi + eax*8] ;# Y1 movsd xmm5, [esi + eax*8 + 8] ;# F1 movsd xmm6, [esi + eax*8 + 16] ;# G1 movsd xmm7, [esi + eax*8 + 24] ;# H1 ;# coulomb table ready, in xmm4-xmm7 mulsd xmm6, xmm1 ;# xmm6=Geps mulsd xmm7, xmm2 ;# xmm7=Heps2 addsd xmm5, xmm6 addsd xmm5, xmm7 ;# xmm5=Fp mulsd xmm7, [esp + nb331_two] ;# two*Heps2 movapd xmm3, [esp + nb331_qqH] addsd xmm7, xmm6 addsd xmm7, xmm5 ;# xmm7=FF mulsd xmm5, xmm1 ;# xmm5=eps*Fp addsd xmm5, xmm4 ;# xmm5=VV mulsd xmm5, xmm3 ;# vcoul=qq*VV mulsd xmm3, xmm7 ;# fijC=FF*qq ;# at this point mm5 contains vcoul and xmm3 fijC ;# increment vcoul xorpd xmm4, xmm4 addsd xmm5, [esp + nb331_vctot] mulsd xmm3, [esp + nb331_rinvH1] movlpd [esp + nb331_vctot], xmm5 mulsd xmm3, [esp + nb331_tsc] subsd xmm4, xmm3 movapd xmm0, [esp + nb331_dxH1] movapd xmm1, [esp + nb331_dyH1] movapd xmm2, [esp + nb331_dzH1] mulsd xmm0, xmm4 mulsd xmm1, xmm4 mulsd xmm2, xmm4 ;# update H1 forces movapd xmm3, [esp + nb331_fixH1] movapd xmm4, [esp + nb331_fiyH1] movapd xmm7, [esp + nb331_fizH1] addsd xmm3, xmm0 addsd xmm4, xmm1 addsd xmm7, xmm2 movlpd [esp + nb331_fixH1], xmm3 movlpd [esp + nb331_fiyH1], xmm4 movlpd [esp + nb331_fizH1], xmm7 ;# update j forces with water H1 addsd xmm0, [esp + nb331_fjx] addsd xmm1, [esp + nb331_fjy] addsd xmm2, [esp + nb331_fjz] movlpd [esp + nb331_fjx], xmm0 movlpd [esp + nb331_fjy], xmm1 movlpd [esp + nb331_fjz], xmm2 ;# Done with H1, finally we do H2 interactions movapd xmm7, [esp + nb331_rH2] mulsd xmm7, [esp + nb331_tsc] cvttsd2si eax, xmm7 ;# mm6 = lu idx cvtsi2sd xmm6, eax subsd xmm7, xmm6 movapd xmm1, xmm7 ;# xmm1=eps movapd xmm2, xmm1 mulsd xmm2, xmm2 ;# xmm2=eps2 shl eax, 2 ;# idx *= 4 mov esi, [ebp + nb331_VFtab] lea eax, [eax + eax*2] ;# idx *= 3 (total *=12 now) movsd xmm4, [esi + eax*8] ;# Y1 movsd xmm5, [esi + eax*8 + 8] ;# F1 movsd xmm6, [esi + eax*8 + 16] ;# G1 movsd xmm7, [esi + eax*8 + 24] ;# H1 ;# coulomb table ready, in xmm4-xmm7 mulsd xmm6, xmm1 ;# xmm6=Geps mulsd xmm7, xmm2 ;# xmm7=Heps2 addsd xmm5, xmm6 addsd xmm5, xmm7 ;# xmm5=Fp mulsd xmm7, [esp + nb331_two] ;# two*Heps2 movapd xmm3, [esp + nb331_qqH] addsd xmm7, xmm6 addsd xmm7, xmm5 ;# xmm7=FF mulsd xmm5, xmm1 ;# xmm5=eps*Fp addsd xmm5, xmm4 ;# xmm5=VV mulsd xmm5, xmm3 ;# vcoul=qq*VV mulsd xmm3, xmm7 ;# fijC=FF*qq ;# at this point mm5 contains vcoul and xmm3 fijC ;# increment vcoul xorpd xmm4, xmm4 addsd xmm5, [esp + nb331_vctot] mulsd xmm3, [esp + nb331_rinvH2] movlpd [esp + nb331_vctot], xmm5 mulsd xmm3, [esp + nb331_tsc] subsd xmm4, xmm3 movapd xmm0, [esp + nb331_dxH2] movapd xmm1, [esp + nb331_dyH2] movapd xmm2, [esp + nb331_dzH2] mulsd xmm0, xmm4 mulsd xmm1, xmm4 mulsd xmm2, xmm4 movd eax, mm0 ;# update H2 forces movapd xmm3, [esp + nb331_fixH2] movapd xmm4, [esp + nb331_fiyH2] movapd xmm7, [esp + nb331_fizH2] addsd xmm3, xmm0 addsd xmm4, xmm1 addsd xmm7, xmm2 movlpd [esp + nb331_fixH2], xmm3 movlpd [esp + nb331_fiyH2], xmm4 movlpd [esp + nb331_fizH2], xmm7 mov edi, [ebp + nb331_faction] ;# update j forces addsd xmm0, [esp + nb331_fjx] addsd xmm1, [esp + nb331_fjy] addsd xmm2, [esp + nb331_fjz] ;# the fj's - start by accumulating forces from memory movlpd xmm3, [edi + eax*8] movlpd xmm4, [edi + eax*8 + 8] movlpd xmm5, [edi + eax*8 + 16] subsd xmm3, xmm0 subsd xmm4, xmm1 subsd xmm5, xmm2 movlpd [edi + eax*8], xmm3 movlpd [edi + eax*8 + 8], xmm4 movlpd [edi + eax*8 + 16], xmm5.nb331_updateouterdata: mov ecx, [esp + nb331_ii3] mov edi, [ebp + nb331_faction] mov esi, [ebp + nb331_fshift] mov edx, [esp + nb331_is3] ;# accumulate Oi forces in xmm0, xmm1, xmm2 movapd xmm0, [esp + nb331_fixO] movapd xmm1, [esp + nb331_fiyO] movapd xmm2, [esp + nb331_fizO] movhlps xmm3, xmm0 movhlps xmm4, xmm1 movhlps xmm5, xmm2 addsd xmm0, xmm3 addsd xmm1, xmm4 addsd xmm2, xmm5 ;# sum is in low xmm0-xmm2 movapd xmm3, xmm0 movapd xmm4, xmm1 movapd xmm5, xmm2 ;# increment i force movsd xmm3, [edi + ecx*8] movsd xmm4, [edi + ecx*8 + 8] movsd xmm5, [edi + ecx*8 + 16] addsd xmm3, xmm0 addsd xmm4, xmm1 addsd xmm5, xmm2 movsd [edi + ecx*8], xmm3 movsd [edi + ecx*8 + 8], xmm4 movsd [edi + ecx*8 + 16], xmm5 ;# accumulate force in xmm6/xmm7 for fshift movapd xmm6, xmm0 movsd xmm7, xmm2 unpcklpd xmm6, xmm1 ;# accumulate H1i forces in xmm0, xmm1, xmm2 movapd xmm0, [esp + nb331_fixH1] movapd xmm1, [esp + nb331_fiyH1] movapd xmm2, [esp + nb331_fizH1] movhlps xmm3, xmm0 movhlps xmm4, xmm1 movhlps xmm5, xmm2 addsd xmm0, xmm3 addsd xmm1, xmm4 addsd xmm2, xmm5 ;# sum is in low xmm0-xmm2 ;# increment i force movsd xmm3, [edi + ecx*8 + 24] movsd xmm4, [edi + ecx*8 + 32] movsd xmm5, [edi + ecx*8 + 40] addsd xmm3, xmm0 addsd xmm4, xmm1 addsd xmm5, xmm2 movsd [edi + ecx*8 + 24], xmm3 movsd [edi + ecx*8 + 32], xmm4 movsd [edi + ecx*8 + 40], xmm5 ;# accumulate force in xmm6/xmm7 for fshift addsd xmm7, xmm2 unpcklpd xmm0, xmm1 addpd xmm6, xmm0 ;# accumulate H2i forces in xmm0, xmm1, xmm2 movapd xmm0, [esp + nb331_fixH2] movapd xmm1, [esp + nb331_fiyH2] movapd xmm2, [esp + nb331_fizH2] movhlps xmm3, xmm0⌨️ 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?