nb_kernel132_ia32_sse.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,176 行 · 第 1/5 页
S
2,176 行
subps xmm3, xmm0 subps xmm4, xmm1 subps xmm5, xmm2 addps xmm0, [esp + nb132_fixO] addps xmm1, [esp + nb132_fiyO] addps xmm2, [esp + nb132_fizO] movaps [esp + nb132_fjxH1], xmm3 movaps [esp + nb132_fjyH1], xmm4 movaps [esp + nb132_fjzH1], xmm5 movaps [esp + nb132_fixO], xmm0 movaps [esp + nb132_fiyO], xmm1 movaps [esp + nb132_fizO], xmm2 ;# O-H2 interaction movaps xmm0, [esp + nb132_rinvOH2] movaps xmm4, xmm0 ;# xmm7=rinv mulps xmm0, xmm0 ;# rinvsq mulps xmm4, [esp + nb132_qqOH] ;# xmm4=voul=qq*(rinv+ krsq) mulps xmm0, xmm4 addps xmm6, xmm4 ;# add to local vctot movaps xmm1, xmm0 movaps xmm2, xmm0 xorps xmm3, xmm3 movaps xmm4, xmm3 movaps xmm5, xmm3 mulps xmm0, [esp + nb132_dxOH2] mulps xmm1, [esp + nb132_dyOH2] mulps xmm2, [esp + nb132_dzOH2] subps xmm3, xmm0 subps xmm4, xmm1 subps xmm5, xmm2 addps xmm0, [esp + nb132_fixO] addps xmm1, [esp + nb132_fiyO] addps xmm2, [esp + nb132_fizO] movaps [esp + nb132_fjxH2], xmm3 movaps [esp + nb132_fjyH2], xmm4 movaps [esp + nb132_fjzH2], xmm5 movaps [esp + nb132_fixO], xmm0 movaps [esp + nb132_fiyO], xmm1 movaps [esp + nb132_fizO], xmm2 ;# H1-O interaction movaps xmm0, [esp + nb132_rinvH1O] movaps xmm4, xmm0 ;# xmm7=rinv mulps xmm0, xmm0 ;# rinvsq mulps xmm4, [esp + nb132_qqOH] ;# xmm4=voul=qq*(rinv+ krsq) mulps xmm0, xmm4 addps xmm6, xmm4 ;# add to local vctot movaps xmm1, xmm0 movaps xmm2, xmm0 movaps xmm3, [esp + nb132_fjxO] movaps xmm4, [esp + nb132_fjyO] movaps xmm5, [esp + nb132_fjzO] mulps xmm0, [esp + nb132_dxH1O] mulps xmm1, [esp + nb132_dyH1O] mulps xmm2, [esp + nb132_dzH1O] subps xmm3, xmm0 subps xmm4, xmm1 subps xmm5, xmm2 addps xmm0, [esp + nb132_fixH1] addps xmm1, [esp + nb132_fiyH1] addps xmm2, [esp + nb132_fizH1] movaps [esp + nb132_fjxO], xmm3 movaps [esp + nb132_fjyO], xmm4 movaps [esp + nb132_fjzO], xmm5 movaps [esp + nb132_fixH1], xmm0 movaps [esp + nb132_fiyH1], xmm1 movaps [esp + nb132_fizH1], xmm2 ;# H1-H1 interaction movaps xmm0, [esp + nb132_rinvH1H1] movaps xmm4, xmm0 ;# xmm7=rinv mulps xmm0, xmm0 ;# rinvsq mulps xmm4, [esp + nb132_qqHH] ;# xmm4=voul=qq*(rinv+ krsq) mulps xmm0, xmm4 addps xmm6, xmm4 ;# add to local vctot movaps xmm1, xmm0 movaps xmm2, xmm0 movaps xmm3, [esp + nb132_fjxH1] movaps xmm4, [esp + nb132_fjyH1] movaps xmm5, [esp + nb132_fjzH1] mulps xmm0, [esp + nb132_dxH1H1] mulps xmm1, [esp + nb132_dyH1H1] mulps xmm2, [esp + nb132_dzH1H1] subps xmm3, xmm0 subps xmm4, xmm1 subps xmm5, xmm2 addps xmm0, [esp + nb132_fixH1] addps xmm1, [esp + nb132_fiyH1] addps xmm2, [esp + nb132_fizH1] movaps [esp + nb132_fjxH1], xmm3 movaps [esp + nb132_fjyH1], xmm4 movaps [esp + nb132_fjzH1], xmm5 movaps [esp + nb132_fixH1], xmm0 movaps [esp + nb132_fiyH1], xmm1 movaps [esp + nb132_fizH1], xmm2 ;# H1-H2 interaction movaps xmm0, [esp + nb132_rinvH1H2] movaps xmm4, xmm0 ;# xmm7=rinv mulps xmm0, xmm0 ;# rinvsq mulps xmm4, [esp + nb132_qqHH] ;# xmm4=voul=qq*(rinv+ krsq) mulps xmm0, xmm4 addps xmm6, xmm4 ;# add to local vctot movaps xmm1, xmm0 movaps xmm2, xmm0 movaps xmm3, [esp + nb132_fjxH2] movaps xmm4, [esp + nb132_fjyH2] movaps xmm5, [esp + nb132_fjzH2] mulps xmm0, [esp + nb132_dxH1H2] mulps xmm1, [esp + nb132_dyH1H2] mulps xmm2, [esp + nb132_dzH1H2] subps xmm3, xmm0 subps xmm4, xmm1 subps xmm5, xmm2 addps xmm0, [esp + nb132_fixH1] addps xmm1, [esp + nb132_fiyH1] addps xmm2, [esp + nb132_fizH1] movaps [esp + nb132_fjxH2], xmm3 movaps [esp + nb132_fjyH2], xmm4 movaps [esp + nb132_fjzH2], xmm5 movaps [esp + nb132_fixH1], xmm0 movaps [esp + nb132_fiyH1], xmm1 movaps [esp + nb132_fizH1], xmm2 ;# H2-O interaction movaps xmm0, [esp + nb132_rinvH2O] movaps xmm4, xmm0 ;# xmm7=rinv mulps xmm0, xmm0 ;# rinvsq mulps xmm4, [esp + nb132_qqOH] ;# xmm4=voul=qq*(rinv+ krsq) mulps xmm0, xmm4 addps xmm6, xmm4 ;# add to local vctot movaps xmm1, xmm0 movaps xmm2, xmm0 movaps xmm3, [esp + nb132_fjxO] movaps xmm4, [esp + nb132_fjyO] movaps xmm5, [esp + nb132_fjzO] mulps xmm0, [esp + nb132_dxH2O] mulps xmm1, [esp + nb132_dyH2O] mulps xmm2, [esp + nb132_dzH2O] subps xmm3, xmm0 subps xmm4, xmm1 subps xmm5, xmm2 addps xmm0, [esp + nb132_fixH2] addps xmm1, [esp + nb132_fiyH2] addps xmm2, [esp + nb132_fizH2] movaps [esp + nb132_fjxO], xmm3 movaps [esp + nb132_fjyO], xmm4 movaps [esp + nb132_fjzO], xmm5 movaps [esp + nb132_fixH2], xmm0 movaps [esp + nb132_fiyH2], xmm1 movaps [esp + nb132_fizH2], xmm2 ;# H2-H1 interaction movaps xmm0, [esp + nb132_rinvH2H1] movaps xmm4, xmm0 ;# xmm7=rinv mulps xmm0, xmm0 ;# rinvsq mulps xmm4, [esp + nb132_qqHH] ;# xmm4=voul=qq*(rinv+ krsq) mulps xmm0, xmm4 addps xmm6, xmm4 ;# add to local vctot movaps xmm1, xmm0 movaps xmm2, xmm0 movaps xmm3, [esp + nb132_fjxH1] movaps xmm4, [esp + nb132_fjyH1] movaps xmm5, [esp + nb132_fjzH1] mulps xmm0, [esp + nb132_dxH2H1] mulps xmm1, [esp + nb132_dyH2H1] mulps xmm2, [esp + nb132_dzH2H1] subps xmm3, xmm0 subps xmm4, xmm1 subps xmm5, xmm2 addps xmm0, [esp + nb132_fixH2] addps xmm1, [esp + nb132_fiyH2] addps xmm2, [esp + nb132_fizH2] movaps [esp + nb132_fjxH1], xmm3 movaps [esp + nb132_fjyH1], xmm4 movaps [esp + nb132_fjzH1], xmm5 movaps [esp + nb132_fixH2], xmm0 movaps [esp + nb132_fiyH2], xmm1 movaps [esp + nb132_fizH2], xmm2 ;# H2-H2 interaction movaps xmm0, [esp + nb132_rinvH2H2] movaps xmm4, xmm0 ;# xmm7=rinv mulps xmm0, xmm0 ;# rinvsq mulps xmm4, [esp + nb132_qqHH] ;# xmm4=voul=qq*(rinv+ krsq) mulps xmm0, xmm4 addps xmm6, xmm4 ;# add to local vctot movaps xmm1, xmm0 movaps xmm2, xmm0 movaps xmm1, xmm0 movaps [esp + nb132_vctot], xmm6 movaps xmm2, xmm0 movaps xmm3, [esp + nb132_fjxH2] movaps xmm4, [esp + nb132_fjyH2] movaps xmm5, [esp + nb132_fjzH2] mulps xmm0, [esp + nb132_dxH2H2] mulps xmm1, [esp + nb132_dyH2H2] mulps xmm2, [esp + nb132_dzH2H2] subps xmm3, xmm0 subps xmm4, xmm1 subps xmm5, xmm2 addps xmm0, [esp + nb132_fixH2] addps xmm1, [esp + nb132_fiyH2] addps xmm2, [esp + nb132_fizH2] movaps [esp + nb132_fjxH2], xmm3 movaps [esp + nb132_fjyH2], xmm4 movaps [esp + nb132_fjzH2], xmm5 movaps [esp + nb132_fixH2], xmm0 movaps [esp + nb132_fiyH2], xmm1 movaps [esp + nb132_fizH2], xmm2 mov edi, [ebp + nb132_faction] ;# Did all interactions - now update j forces ;# At this stage forces are still on the stack, in positions: ;# fjxO, fjyO, fjzO, ... , fjzH2. ;# Each position is a quadruplet of forces for the four ;# corresponding j waters, so we need to transpose them before ;# adding to the memory positions. ;# ;# This _used_ to be a simple transpose, but the resulting high number ;# of unaligned 128-bit load/stores might trigger a possible hardware ;# bug on Athlon and Opteron chips, so I have worked around it ;# to use 64-bit load/stores instead. The performance hit should be ;# very modest, since the 128-bit unaligned memory instructions were ;# slow anyway. ;# 4 j waters with three atoms each - first do Oxygen X & Y forces for 4 j particles movaps xmm0, [esp + nb132_fjxO] ;# xmm0= fjxOa fjxOb fjxOc fjxOd movaps xmm2, [esp + nb132_fjyO] ;# xmm1= fjyOa fjyOb fjyOc fjyOd movlps xmm3, [edi + eax*4] movlps xmm4, [edi + ecx*4] movaps xmm1, xmm0 unpcklps xmm0, xmm2 ;# xmm0= fjxOa fjyOa fjxOb fjyOb unpckhps xmm1, xmm2 ;# xmm1= fjxOc fjyOc fjxOd fjyOd movhps xmm3, [edi + ebx*4] movhps xmm4, [edi + edx*4] addps xmm3, xmm0 addps xmm4, xmm1 movlps [edi + eax*4], xmm3 movlps [edi + ecx*4], xmm4 movhps [edi + ebx*4], xmm3 movhps [edi + edx*4], xmm4 ;# Oxygen Z & first hydrogen X forces for 4 j particles movaps xmm0, [esp + nb132_fjzO] ;# xmm0= fjzOa fjzOb fjzOc fjzOd movaps xmm2, [esp + nb132_fjxH1] ;# xmm1= fjxH1a fjxH1b fjxH1c fjxH1d movlps xmm3, [edi + eax*4 + 8] movlps xmm4, [edi + ecx*4 + 8] movaps xmm1, xmm0 unpcklps xmm0, xmm2 ;# xmm0= fjzOa fjxH1a fjzOb fjxH1b unpckhps xmm1, xmm2 ;# xmm1= fjzOc fjxH1c fjzOd fjxH1d movhps xmm3, [edi + ebx*4 + 8] movhps xmm4, [edi + edx*4 + 8] addps xmm3, xmm0 addps xmm4, xmm1 movlps [edi + eax*4 + 8], xmm3 movlps [edi + ecx*4 + 8], xmm4 movhps [edi + ebx*4 + 8], xmm3 movhps [edi + edx*4 + 8], xmm4 ;# First hydrogen Y & Z forces for 4 j particles movaps xmm0, [esp + nb132_fjyH1] ;# xmm0= fjyH1a fjyH1b fjyH1c fjyH1d movaps xmm2, [esp + nb132_fjzH1] ;# xmm1= fjzH1a fjzH1b fjzH1c fjzH1d movlps xmm3, [edi + eax*4 + 16] movlps xmm4, [edi + ecx*4 + 16] movaps xmm1, xmm0 unpcklps xmm0, xmm2 ;# xmm0= fjyH1a fjzH1a fjyH1b fjzH1b unpckhps xmm1, xmm2 ;# xmm1= fjyH1c fjzH1c fjyH1d fjzH1d movhps xmm3, [edi + ebx*4 + 16] movhps xmm4, [edi + edx*4 + 16] addps xmm3, xmm0 addps xmm4, xmm1 movlps [edi + eax*4 + 16], xmm3 movlps [edi + ecx*4 + 16], xmm4 movhps [edi + ebx*4 + 16], xmm3 movhps [edi + edx*4 + 16], xmm4 ;# Second hydrogen X & Y forces for 4 j particles movaps xmm0, [esp + nb132_fjxH2] ;# xmm0= fjxH2a fjxH2b fjxH2c fjxH2d movaps xmm2, [esp + nb132_fjyH2] ;# xmm1= fjyH2a fjyH2b fjyH2c fjyH2d movlps xmm3, [edi + eax*4 + 24] movlps xmm4, [edi + ecx*4 + 24] movaps xmm1, xmm0 unpcklps xmm0, xmm2 ;# xmm0= fjxH2a fjyH2a fjxH2b fjyH2b unpckhps xmm1, xmm2 ;# xmm1= fjxH2c fjyH2c fjxH2d fjyH2d movhps xmm3, [edi + ebx*4 + 24] movhps xmm4, [edi + edx*4 + 24] addps xmm3, xmm0 addps xmm4, xmm1 movlps [edi + eax*4 + 24], xmm3 movlps [edi + ecx*4 + 24], xmm4 movhps [edi + ebx*4 + 24], xmm3 movhps [edi + edx*4 + 24], xmm4 ;# Second hydrogen Z forces for 4 j particles ;# Just load the four Z coords into one reg. each movss xmm4, [edi + eax*4 + 32] movss xmm5, [edi + ebx*4 + 32] movss xmm6, [edi + ecx*4 + 32] movss xmm7, [edi + edx*4 + 32] ;# add what we have on the stack addss xmm4, [esp + nb132_fjzH2] addss xmm5, [esp + nb132_fjzH2b] addss xmm6, [esp + nb132_fjzH2c] addss xmm7, [esp + nb132_fjzH2d] ;# store back movss [edi + eax*4 + 32], xmm4 movss [edi + ebx*4 + 32], xmm5 movss [edi + ecx*4 + 32], xmm6 movss [edi + edx*4 + 32], xmm7 ;# should we do one more iteration? sub dword ptr [esp + nb132_innerk], 4 jl .nb132_single_check jmp .nb132_unroll_loop.nb132_single_check: add dword ptr [esp + nb132_innerk], 4 jnz .nb132_single_loop jmp .nb132_updateouterdata.nb132_single_loop: mov edx, [esp + nb132_innerjjnr] ;# pointer to jjnr[k] mov eax, [edx] add dword ptr [esp + nb132_innerjjnr], 4 mov esi, [ebp + nb132_pos] lea eax, [eax + eax*2] ;# fetch j coordinates xorps xmm3, xmm3 xorps xmm4, xmm4 xorps xmm5, xmm5 movss xmm3, [esi + eax*4] ;# jxO - - - movss xmm4, [esi + eax*4 + 4] ;# jyO - - - movss xmm5, [esi + eax*4 + 8] ;# jzO - - - movlps xmm6, [esi + eax*4 + 12] ;# xmm6 = jxH1 jyH1 - - movss xmm7, [esi + eax*4 + 20] ;# xmm7 = jzH1 - - - movhps xmm6, [esi + eax*4 + 24] ;# xmm6 = jxH1 jyH1 jxH2 jyH2 movss xmm2, [esi + eax*4 + 32] ;# xmm2 = jzH2 - - - ;# have all coords, time for some shuffling. shufps xmm6, xmm6, 216 ;# constant 11011000 ;# xmm6 = jxH1 jxH2 jyH1 jyH2 unpcklps xmm7, xmm2 ;# xmm7 = jzH1 jzH2 - - movaps xmm0, [esp + nb132_ixO] movaps xmm1, [esp + nb132_iyO] movaps xmm2, [esp + nb132_izO] movlhps xmm3, xmm6 ;# xmm3 = jxO 0 jxH1 jxH2 shufps xmm4, xmm6, 228 ;# constant 11100100 ;# xmm4 = jyO 0 jyH1 jyH2 shufps xmm5, xmm7, 68 ;# constant 01000100 ;# xmm5 = jzO 0 jzH1 jzH2 ;# store all j coordinates in jO movaps [esp + nb132_jxO], xmm3 movaps [esp + nb132_jyO], xmm4 movaps [esp + nb132_jzO], xmm5 subps xmm0, xmm3 subps xmm1, xmm4 subps xmm2, xmm5 movaps [esp + nb132_dxOO], xmm0 movaps [esp + nb132_dyOO], xmm1 movaps [esp + nb132_dzOO], xmm2 mulps xmm0, xmm0 mulps xmm1, xmm1 mulps xmm2, xmm2 addps xmm0, xmm1 addps xmm0, xmm2 ;# have rsq in xmm0 movaps [esp + nb132_rsqOO], xmm0 movaps xmm6, xmm0 ;# do invsqrt rsqrtps xmm1, xmm0 movaps xmm2, xmm1 mulps xmm1, xmm1 movaps xmm3, [esp + nb132_three] mulps xmm1, xmm0 subps xmm3, xmm1 mulps xmm3, xmm2 mulps xmm3, [esp + nb132_half] ;# rinv iO - j water in xmm3 movaps [esp + nb132_rinvOO], xmm3 movaps xmm0, xmm3 ;# rinv xorps xmm4, xmm4 ;# fetch charges to xmm4 (temporary) movss xmm4, [esp + nb132_qqOO] movhps xmm4, [esp + nb132_qqOH] mulps xmm3, xmm4 ;# vcoul movaps xmm6, xmm3 mulps xmm3, xmm0 movaps [esp + nb132_fstmp], xmm3 ;# save it addps xmm6, [esp + nb132_vctot] movaps [esp + nb132_vctot], xmm6 movaps xmm0, [esp + nb132_rinvOO] movss xmm1, xmm0 mulss xmm1, [esp + nb132_rsqOO] ;# xmm1=r mulss xmm1, [esp + nb132_tsc] cvttps2pi mm6, xmm1 cvtpi2ps xmm3, mm6 subss xmm1, xmm3 ;# xmm1=eps movss xmm2, xmm1 mulss xmm2, xmm2 ;# xmm2=eps2 pslld mm6, 3 movd mm0, eax mov esi, [ebp + nb132_VFtab] movd eax, mm6 ;# dispersion ⌨️ 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?