nb_kernel231_x86_64_sse.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,077 行 · 第 1/4 页
S
2,077 行
;# rinv in xmm9, xmm10, xmm11 movaps [rsp + nb231_rsqO], xmm0 movaps [rsp + nb231_rsqH1], xmm3 movaps [rsp + nb231_rsqH2], xmm6 movaps [rsp + nb231_rinvO], xmm9 movaps [rsp + nb231_rinvH1], xmm10 movaps [rsp + nb231_rinvH2], xmm11 ;# table LJ interaction mulps xmm0, xmm9 mulps xmm0, [rsp + nb231_tsc] ;# rtab ;# truncate and convert to integers cvttps2dq xmm1, xmm0 ;# convert back to float cvtdq2ps xmm2, xmm1 ;# multiply by 8 pslld xmm1, 3 ;# move to integer registers movhlps xmm13, xmm1 movd r8d, xmm1 movd r10d, xmm13 shufps xmm1, xmm1, 1 shufps xmm13, xmm13, 1 movd r9d, xmm1 movd r11d, xmm13 ;# calculate eps subps xmm0, xmm2 mov rsi, [rbp + nb231_VFtab] movlps xmm5, [rsi + r8*4] movlps xmm9, [rsi + r8*4 + 16] movlps xmm7, [rsi + r10*4] movlps xmm11, [rsi + r10*4 + 16] movhps xmm5, [rsi + r9*4] movhps xmm9, [rsi + r9*4 + 16] movhps xmm7, [rsi + r11*4] movhps xmm11, [rsi + r11*4 + 16] movaps xmm4, xmm5 movaps xmm8, xmm9 shufps xmm4, xmm7, 136 ;# 10001000 shufps xmm8, xmm11, 136 ;# 10001000 shufps xmm5, xmm7, 221 ;# 11011101 shufps xmm9, xmm11, 221 ;# 11011101 movlps xmm7, [rsi + r8*4 + 8] movlps xmm11, [rsi + r8*4 + 24] movlps xmm13, [rsi + r10*4 + 8] movlps xmm14, [rsi + r10*4 + 24] movhps xmm7, [rsi + r9*4 + 8] movhps xmm11, [rsi + r9*4 + 24] movhps xmm13, [rsi + r11*4 + 8] movhps xmm14, [rsi + r11*4 + 24] movaps xmm6, xmm7 movaps xmm10, xmm11 shufps xmm6, xmm13, 136 ;# 10001000 shufps xmm10, xmm14, 136 ;# 10001000 shufps xmm7, xmm13, 221 ;# 11011101 shufps xmm11, xmm14, 221 ;# 11011101 ;# dispersion table in xmm4-xmm7, repulsion table in xmm8-xmm11 mulps xmm7, xmm0 ;# Heps mulps xmm11, xmm0 mulps xmm6, xmm0 ;# Geps mulps xmm10, xmm0 mulps xmm7, xmm0 ;# Heps2 mulps xmm11, xmm0 addps xmm5, xmm6 ;# F+Geps addps xmm9, xmm10 addps xmm5, xmm7 ;# F+Geps+Heps2 = Fp addps xmm9, xmm11 addps xmm7, xmm7 ;# 2*Heps2 addps xmm11, xmm11 addps xmm7, xmm6 ;# 2*Heps2+Geps addps xmm11, xmm10 addps xmm7, xmm5 ;# FF = Fp + 2*Heps2 + Geps addps xmm11, xmm9 mulps xmm5, xmm0 ;# eps*Fp mulps xmm9, xmm0 movaps xmm12, [rsp + nb231_c6] movaps xmm13, [rsp + nb231_c12] addps xmm5, xmm4 ;# VV addps xmm9, xmm8 mulps xmm5, xmm12 ;# VV*c6 = vnb6 mulps xmm9, xmm13 ;# VV*c12 = vnb12 addps xmm5, xmm9 addps xmm5, [rsp + nb231_Vvdwtot] movaps [rsp + nb231_Vvdwtot], xmm5 mulps xmm7, xmm12 ;# FF*c6 = fnb6 mulps xmm11, xmm13 ;# FF*c12 = fnb12 addps xmm7, xmm11 mulps xmm7, [rsp + nb231_tsc] movaps [rsp + nb231_fstmp], xmm7 ;# Coulomb reaction-field interaction movaps xmm0, [rsp + nb231_rsqO] movaps xmm3, [rsp + nb231_rsqH1] movaps xmm6, [rsp + nb231_rsqH2] movaps xmm9, [rsp + nb231_rinvO] movaps xmm10, [rsp + nb231_rinvH1] movaps xmm11, [rsp + nb231_rinvH2] movaps xmm1, xmm9 ;# copy of rinv movaps xmm4, xmm10 movaps xmm7, xmm11 movaps xmm2, [rsp + nb231_krf] mulps xmm10, xmm10 ;# rinvsq mulps xmm11, xmm11 mulps xmm0, xmm2 ;# k*rsq mulps xmm3, xmm2 mulps xmm6, xmm2 movaps xmm2, xmm0 ;# copy of k*rsq movaps xmm5, xmm3 movaps xmm8, xmm6 addps xmm2, xmm1 ;# rinv+krsq addps xmm5, xmm4 addps xmm8, xmm7 subps xmm2, [rsp + nb231_crf] ;# rinv+krsq-crf subps xmm5, [rsp + nb231_crf] subps xmm8, [rsp + nb231_crf] mulps xmm2, [rsp + nb231_qqO] ;# voul=qq*(rinv+ krsq-crf) mulps xmm5, [rsp + nb231_qqH] ;# voul=qq*(rinv+ krsq-crf) mulps xmm8, [rsp + nb231_qqH] ;# voul=qq*(rinv+ krsq-crf) addps xmm0, xmm0 ;# 2*krsq addps xmm3, xmm3 addps xmm6, xmm6 subps xmm1, xmm0 ;# rinv-2*krsq subps xmm4, xmm3 subps xmm7, xmm6 mulps xmm1, [rsp + nb231_qqO] ;# (rinv-2*krsq)*qq mulps xmm4, [rsp + nb231_qqH] mulps xmm7, [rsp + nb231_qqH] addps xmm2, [rsp + nb231_vctot] addps xmm5, xmm8 addps xmm2, xmm5 movaps xmm15, xmm2 mulps xmm1, xmm9 ;# fijC mulps xmm4, xmm10 mulps xmm7, xmm11 ;# xmm1, xmm4, xmm7 contains fscal coul ;# xmm15 contains vctot subps xmm1, [rsp + nb231_fstmp] mulps xmm1, xmm9 movaps [rsp + nb231_vctot], xmm2 ;# move j forces to local temp variables mov rdi, [rbp + nb231_faction] movlps xmm9, [rdi + rax*4] ;# jxa jya - - movlps xmm10, [rdi + rcx*4] ;# jxc jyc - - movhps xmm9, [rdi + rbx*4] ;# jxa jya jxb jyb movhps xmm10, [rdi + rdx*4] ;# jxc jyc jxd jyd movss xmm11, [rdi + rax*4 + 8] ;# jza - - - movss xmm12, [rdi + rcx*4 + 8] ;# jzc - - - movss xmm6, [rdi + rbx*4 + 8] ;# jzb movss xmm8, [rdi + rdx*4 + 8] ;# jzd movlhps xmm11, xmm6 ;# jza - jzb - movlhps xmm12, xmm8 ;# jzc - jzd - shufps xmm11, xmm12, 136 ;# 10001000 => jza jzb jzc jzd ;# xmm9: jxa jya jxb jyb ;# xmm10: jxc jyc jxd jyd ;# xmm11: jza jzb jzc jzd movaps xmm0, xmm1 movaps xmm2, xmm1 movaps xmm3, xmm4 movaps xmm5, xmm4 movaps xmm6, xmm7 movaps xmm8, xmm7 mulps xmm0, [rsp + nb231_dxO] mulps xmm1, [rsp + nb231_dyO] mulps xmm2, [rsp + nb231_dzO] mulps xmm3, [rsp + nb231_dxH1] mulps xmm4, [rsp + nb231_dyH1] mulps xmm5, [rsp + nb231_dzH1] mulps xmm6, [rsp + nb231_dxH2] mulps xmm7, [rsp + nb231_dyH2] mulps xmm8, [rsp + nb231_dzH2] movaps xmm13, xmm0 movaps xmm14, xmm1 addps xmm11, xmm2 addps xmm0, [rsp + nb231_fixO] addps xmm1, [rsp + nb231_fiyO] addps xmm2, [rsp + nb231_fizO] addps xmm13, xmm3 addps xmm14, xmm4 addps xmm11, xmm5 addps xmm3, [rsp + nb231_fixH1] addps xmm4, [rsp + nb231_fiyH1] addps xmm5, [rsp + nb231_fizH1] addps xmm13, xmm6 addps xmm14, xmm7 addps xmm11, xmm8 addps xmm6, [rsp + nb231_fixH2] addps xmm7, [rsp + nb231_fiyH2] addps xmm8, [rsp + nb231_fizH2] movaps [rsp + nb231_fixO], xmm0 movaps [rsp + nb231_fiyO], xmm1 movaps [rsp + nb231_fizO], xmm2 movaps [rsp + nb231_fixH1], xmm3 movaps [rsp + nb231_fiyH1], xmm4 movaps [rsp + nb231_fizH1], xmm5 movaps [rsp + nb231_fixH2], xmm6 movaps [rsp + nb231_fiyH2], xmm7 movaps [rsp + nb231_fizH2], xmm8 ;# xmm9 = fjx ;# xmm10 = fjy ;# xmm11 = fjz movaps xmm15, xmm13 unpcklps xmm13, xmm14 unpckhps xmm15, xmm14 addps xmm9, xmm13 addps xmm10, xmm15 movhlps xmm12, xmm11 ;# fjzc fjzd movlps [rdi + rax*4], xmm9 movhps [rdi + rbx*4], xmm9 movlps [rdi + rcx*4], xmm10 movhps [rdi + rdx*4], xmm10 movss [rdi + rax*4 + 8], xmm11 movss [rdi + rcx*4 + 8], xmm12 shufps xmm11, xmm11, 1 shufps xmm12, xmm12, 1 movss [rdi + rbx*4 + 8], xmm11 movss [rdi + rdx*4 + 8], xmm12 ;# should we do one more iteration? sub dword ptr [rsp + nb231_innerk], 4 jl .nb231_odd_inner jmp .nb231_unroll_loop.nb231_odd_inner: add dword ptr [rsp + nb231_innerk], 4 jnz .nb231_odd_loop jmp .nb231_updateouterdata.nb231_odd_loop: mov rdx, [rsp + nb231_innerjjnr] ;# pointer to jjnr[k] mov eax, [rdx] add qword ptr [rsp + nb231_innerjjnr], 4 xorps xmm4, xmm4 movss xmm4, [rsp + nb231_iqO] mov rsi, [rbp + nb231_charge] movhps xmm4, [rsp + nb231_iqH] movss xmm3, [rsi + rax*4] ;# charge in xmm3 shufps xmm3, xmm3, 0 mulps xmm3, xmm4 movaps [rsp + nb231_qqO], xmm3 ;# use oxygen qq for storage xorps xmm6, xmm6 mov rsi, [rbp + nb231_type] mov ebx, [rsi + rax*4] mov rsi, [rbp + nb231_vdwparam] shl ebx, 1 add ebx, [rsp + nb231_ntia] movlps xmm6, [rsi + rbx*4] movaps xmm7, xmm6 shufps xmm6, xmm6, 252 ;# constant 11111100 shufps xmm7, xmm7, 253 ;# constant 11111101 movaps [rsp + nb231_c6], xmm6 movaps [rsp + nb231_c12], xmm7 mov rsi, [rbp + nb231_pos] lea rax, [rax + rax*2] ;# move j coords to xmm0-xmm2 movss xmm3, [rsi + rax*4] movss xmm4, [rsi + rax*4 + 4] movss xmm5, [rsi + rax*4 + 8] shufps xmm3, xmm3, 0 shufps xmm4, xmm4, 0 shufps xmm5, xmm5, 0 movss xmm0, [rsp + nb231_ixO] movss xmm1, [rsp + nb231_iyO] movss xmm2, [rsp + nb231_izO] movlps xmm6, [rsp + nb231_ixH1] movlps xmm7, [rsp + nb231_ixH2] unpcklps xmm6, xmm7 movlhps xmm0, xmm6 movlps xmm6, [rsp + nb231_iyH1] movlps xmm7, [rsp + nb231_iyH2] unpcklps xmm6, xmm7 movlhps xmm1, xmm6 movlps xmm6, [rsp + nb231_izH1] movlps xmm7, [rsp + nb231_izH2] unpcklps xmm6, xmm7 movlhps xmm2, xmm6 subps xmm3, xmm0 subps xmm4, xmm1 subps xmm5, xmm2 movaps [rsp + nb231_dxO], xmm3 movaps [rsp + nb231_dyO], xmm4 movaps [rsp + nb231_dzO], xmm5 mulps xmm3, xmm3 mulps xmm4, xmm4 mulps xmm5, xmm5 addps xmm4, xmm3 addps xmm4, xmm5 ;# rsq in xmm4 movaps xmm0, xmm4 movaps [rsp + nb231_rsqO], xmm0 rsqrtps xmm5, xmm4 ;# lookup seed in xmm5 movaps xmm2, xmm5 mulps xmm5, xmm5 movaps xmm1, [rsp + nb231_three] mulps xmm5, xmm4 ;# rsq*lu*lu movaps xmm0, [rsp + nb231_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 ;# rsq still in xmm4, rinv in xmm0. mulps xmm4, xmm0 mulps xmm4, [rsp + nb231_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 rsi, [rbp + nb231_VFtab] movd r8d, mm6 ;# dispersion movlps xmm5, [rsi + r8*4] movaps xmm4, xmm5 shufps xmm4, xmm7, 136 ;# constant 10001000 shufps xmm5, xmm7, 221 ;# constant 11011101 movlps xmm7, [rsi + r8*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, [rsp + nb231_two] ;# two*Heps2 addss xmm7, xmm6 addss xmm7, xmm5 ;# xmm7=FF mulss xmm5, xmm1 ;# xmm5=eps*Fp addss xmm5, xmm4 ;# xmm5=VV movss xmm4, [rsp + nb231_c6] mulss xmm7, xmm4 ;# fijD mulss xmm5, xmm4 ;# Vvdw6 mulss xmm7, [rsp + nb231_tsc] ;# put scalar force on stack Update Vvdwtot directly addss xmm5, [rsp + nb231_Vvdwtot] movss [rsp + nb231_fstmp], xmm7 movss [rsp + nb231_Vvdwtot], xmm5 ;# repulsion movlps xmm5, [rsi + r8*4 + 16] movaps xmm4, xmm5 shufps xmm4, xmm7, 136 ;# constant 10001000 shufps xmm5, xmm7, 221 ;# constant 11011101 movlps xmm7, [rsi + r8*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, [rsp + nb231_two] ;# two*Heps2 addss xmm7, xmm6 addss xmm7, xmm5 ;# xmm7=FF mulss xmm5, xmm1 ;# xmm5=eps*Fp addss xmm5, xmm4 ;# xmm5=VV movss xmm4, [rsp + nb231_c12] mulss xmm7, xmm4 ;# fijR mulss xmm5, xmm4 ;# Vvdw12 mulss xmm7, [rsp + nb231_tsc] addss xmm7, [rsp + nb231_fstmp] movss [rsp + nb231_fstmp], xmm7 addss xmm5, [rsp + nb231_Vvdwtot] movss [rsp + nb231_Vvdwtot], xmm5 movaps xmm3, [rsp + nb231_rsqO] movaps xmm1, xmm0 ;# xmm1=rinv movaps xmm4, xmm0 mulps xmm3, [rsp + nb231_krf] addps xmm0, xmm3 ;# xmm0=rinv+ krsq mulps xmm3, [rsp + nb231_two] subps xmm0, [rsp + nb231_crf] ;# xmm0=rinv+ krsq-crf subps xmm1, xmm3 ;# xmm1=rinv-2*krsq mulps xmm0, [rsp + nb231_qqO] ;# xmm0=vcoul mulps xmm1, [rsp + nb231_qqO] ;# xmm1=coul part of fs mulps xmm1, xmm4 subss xmm1, [rsp + nb231_fstmp] mulps xmm4, xmm1 addps xmm0, [rsp + nb231_vctot] movaps [rsp + nb231_vctot], xmm0 movaps xmm0, [rsp + nb231_dxO] movaps xmm1, [rsp + nb231_dyO] movaps xmm2, [rsp + nb231_dzO] mulps xmm0, xmm4 mulps xmm1, xmm4 mulps xmm2, xmm4 ;# xmm0-xmm2 contains tx-tz (partial force) movss xmm3, [rsp + nb231_fixO] movss xmm4, [rsp + nb231_fiyO] movss xmm5, [rsp + nb231_fizO] addss xmm3, xmm0 addss xmm4, xmm1 addss xmm5, xmm2 movss [rsp + nb231_fixO], xmm3 movss [rsp + nb231_fiyO], xmm4 movss [rsp + nb231_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, [rsp + nb231_fixH1] addss xmm4, [rsp + nb231_fiyH1] addss xmm5, [rsp + nb231_fizH1] movss [rsp + nb231_fixH1], xmm3 movss [rsp + nb231_fiyH1], xmm4 movss [rsp + nb231_fizH1], xmm5 ;# updated the H1 force mov rdi, [rbp + nb231_faction] shufps xmm3, xmm3, 231 ;# constant 11100111 ;# shift right shufps xmm4, xmm4, 231 ;# constant 11100111 shufps xmm5, xmm5, 231 ;# constant 11100111 addss xmm3, [rsp + nb231_fixH2] addss xmm4, [rsp + nb231_fiyH2] addss xmm5, [rsp + nb231_fizH2] movss [rsp + nb231_fixH2], xmm3 movss [rsp + nb231_fiyH2], xmm4 movss [rsp + nb231_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, [rdi + rax*4] movss xmm7, [rdi + rax*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 addps xmm6, xmm0 addss xmm7, xmm2 movlps [rdi + rax*4], xmm6 movss [rdi + rax*4 + 8], xmm7 dec dword ptr [rsp + nb231_innerk] jz .nb231_updateouterdata jmp .nb231_odd_loop.nb231_updateouterdata: mov ecx, [rsp + nb231_ii3] mov rdi, [rbp + nb231_faction] mov rsi, [rbp + nb231_fshift] mov edx, [rsp + nb231_is3] ;# accumulate Oi forces in xmm0, xmm1, xmm2 ⌨️ 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?