nb_kernel234_x86_64_sse.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,212 行 · 第 1/5 页
S
2,212 行
;# quad-unroll innerloop here mov rdx, [rsp + nb234_innerjjnr] ;# pointer to jjnr[k] mov eax, [rdx] mov ebx, [rdx + 4] mov ecx, [rdx + 8] mov edx, [rdx + 12] ;# eax-edx=jnr1-4 add qword ptr [rsp + nb234_innerjjnr], 16 ;# advance pointer (unroll 4) mov rsi, [rbp + nb234_pos] ;# base of pos[] lea rax, [rax + rax*2] ;# replace jnr with j3 lea rbx, [rbx + rbx*2] lea rcx, [rcx + rcx*2] ;# replace jnr with j3 lea rdx, [rdx + rdx*2] ;# load j O coordinates movlps xmm0, [rsi + rax*4] ;# jxOa jyOa - - movlps xmm1, [rsi + rcx*4] ;# jxOc jyOc - - movhps xmm0, [rsi + rbx*4] ;# jxOa jyOa jxOb jyOb movhps xmm1, [rsi + rdx*4] ;# jxOc jyOc jxOd jyOd movss xmm2, [rsi + rax*4 + 8] ;# jzOa - - - movss xmm3, [rsi + rcx*4 + 8] ;# jzOc - - - movhps xmm2, [rsi + rbx*4 + 8] ;# jzOa - jzOb - movhps xmm3, [rsi + rdx*4 + 8] ;# jzOc - jzOd - movaps xmm4, xmm0 unpcklps xmm0, xmm1 ;# jxOa jxOc jyOa jyOc unpckhps xmm4, xmm1 ;# jxOb jxOd jyOb jyOd movaps xmm1, xmm0 unpcklps xmm0, xmm4 ;# x unpckhps xmm1, xmm4 ;# y shufps xmm2, xmm3, 136 ;# 10001000 => jzOa jzOb jzOc jzOd ;# xmm0 = Ox ;# xmm1 = Oy ;# xmm2 = Oz subps xmm0, [rsp + nb234_ixO] subps xmm1, [rsp + nb234_iyO] subps xmm2, [rsp + nb234_izO] ;# store dx/dy/dz movaps xmm13, xmm0 movaps xmm14, xmm1 movaps xmm15, xmm2 ;# square it mulps xmm0, xmm0 mulps xmm1, xmm1 mulps xmm2, xmm2 addps xmm1, xmm0 addps xmm1, xmm2 ;# rsq in xmm1 ;# calculate rinv=1/sqrt(rsq) rsqrtps xmm5, xmm1 movaps xmm2, xmm5 mulps xmm5, xmm5 movaps xmm4, [rsp + nb234_three] mulps xmm5, xmm1 ;# rsq*lu*lu subps xmm4, xmm5 ;# 30-rsq*lu*lu mulps xmm4, xmm2 mulps xmm4, [rsp + nb234_half] movaps xmm2, xmm4 mulps xmm1, xmm4 ;# xmm2=rinv ;# xmm1=r mulps xmm1, [rsp + nb234_tsc] ;# rtab ;# truncate and convert to integers cvttps2dq xmm5, xmm1 ;# convert back to float cvtdq2ps xmm4, xmm5 ;# multiply by 8 pslld xmm5, 3 ;# calculate eps subps xmm1, xmm4 ;# move to integer registers movhlps xmm6, xmm5 movd r8d, xmm5 movd r10d, xmm6 pshufd xmm5, xmm5, 1 pshufd xmm6, xmm6, 1 movd r9d, xmm5 movd r11d, xmm6 ;# table indices in r8-r11 ;# xmm1=eps ;# xmm2=rinv mov rsi, [rbp + nb234_VFtab] ;# load LJ dispersion and repulsion in parallel 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 xmm0, [rsi + r10*4 + 8] movlps xmm3, [rsi + r10*4 + 24] movhps xmm7, [rsi + r9*4 + 8] movhps xmm11, [rsi + r9*4 + 24] movhps xmm0, [rsi + r11*4 + 8] movhps xmm3, [rsi + r11*4 + 24] movaps xmm6, xmm7 movaps xmm10, xmm11 shufps xmm6, xmm0, 136 ;# 10001000 shufps xmm10, xmm3, 136 ;# 10001000 shufps xmm7, xmm0, 221 ;# 11011101 shufps xmm11, xmm3, 221 ;# 11011101 ;# dispersion table in xmm4-xmm7, repulsion table in xmm8-xmm11 mulps xmm7, xmm1 ;# Heps mulps xmm11, xmm1 mulps xmm6, xmm1 ;# Geps mulps xmm10, xmm1 mulps xmm7, xmm1 ;# Heps2 mulps xmm11, xmm1 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, xmm1 ;# eps*Fp mulps xmm9, xmm1 addps xmm5, xmm4 ;# VV addps xmm9, xmm8 mulps xmm5, [rsp + nb234_c6] ;# VV*c6 = vnb6 mulps xmm9, [rsp + nb234_c12] ;# VV*c12 = vnb12 addps xmm5, xmm9 addps xmm5, [rsp + nb234_Vvdwtot] movaps [rsp + nb234_Vvdwtot], xmm5 mulps xmm7, [rsp + nb234_c6] ;# FF*c6 = fnb6 mulps xmm11, [rsp + nb234_c12] ;# FF*c12 = fnb12 addps xmm7, xmm11 mulps xmm7, [rsp + nb234_tsc] mulps xmm7, xmm2 xorps xmm9, xmm9 subps xmm9, xmm7 ;# fx/fy/fz mulps xmm13, xmm9 mulps xmm14, xmm9 mulps xmm15, xmm9 ;# increment i force movaps xmm0, [rsp + nb234_fixO] movaps xmm1, [rsp + nb234_fiyO] movaps xmm2, [rsp + nb234_fizO] addps xmm0, xmm13 addps xmm1, xmm14 addps xmm2, xmm15 movaps [rsp + nb234_fixO], xmm0 movaps [rsp + nb234_fiyO], xmm1 movaps [rsp + nb234_fizO], xmm2 ;# move j O forces to local temp variables mov rdi, [rbp + nb234_faction] movlps xmm4, [rdi + rax*4] ;# jxOa jyOa - - movlps xmm5, [rdi + rcx*4] ;# jxOc jyOc - - movhps xmm4, [rdi + rbx*4] ;# jxOa jyOa jxOb jyOb movhps xmm5, [rdi + rdx*4] ;# jxOc jyOc jxOd jyOd movss xmm6, [rdi + rax*4 + 8] ;# jzOa - - - movss xmm7, [rdi + rcx*4 + 8] ;# jzOc - - - movhps xmm6, [rdi + rbx*4 + 8] ;# jzOa - jzOb - movhps xmm7, [rdi + rdx*4 + 8] ;# jzOc - jzOd - shufps xmm6, xmm7, 136 ;# 10001000 => jzOa jzOb jzOc jzOd ;# xmm4: jxOa jyOa jxOb jyOb ;# xmm5: jxOc jyOc jxOd jyOd ;# xmm6: jzOa jzOb jzOc jzOd ;# update O forces movaps xmm12, xmm13 unpcklps xmm13, xmm14 ;# (local) fjx1 fjx1 fjy1 fjy2 unpckhps xmm12, xmm14 ;# (local) fjx3 fjx4 fjy3 fjy4 addps xmm4, xmm13 addps xmm5, xmm12 addps xmm6, xmm15 movhlps xmm7, xmm6 ;# fH1zc fH1zd movlps [rdi + rax*4], xmm4 movhps [rdi + rbx*4], xmm4 movlps [rdi + rcx*4], xmm5 movhps [rdi + rdx*4], xmm5 movss [rdi + rax*4 + 8], xmm6 movss [rdi + rcx*4 + 8], xmm7 shufps xmm6, xmm6, 1 shufps xmm7, xmm7, 1 movss [rdi + rbx*4 + 8], xmm6 movss [rdi + rdx*4 + 8], xmm7 ;# done with OO interaction ;# move j H1 coordinates to local temp variables mov rsi, [rbp + nb234_pos] ;# base of pos[] movlps xmm0, [rsi + rax*4 + 12] ;# jxH1a jyH1a - - movlps xmm1, [rsi + rcx*4 + 12] ;# jxH1c jyH1c - - movhps xmm0, [rsi + rbx*4 + 12] ;# jxH1a jyH1a jxH1b jyH1b movhps xmm1, [rsi + rdx*4 + 12] ;# jxH1c jyH1c jxH1d jyH1d movss xmm2, [rsi + rax*4 + 20] ;# jzH1a - - - movss xmm3, [rsi + rcx*4 + 20] ;# jzH1c - - - movhps xmm2, [rsi + rbx*4 + 20] ;# jzH1a - jzH1b - movhps xmm3, [rsi + rdx*4 + 20] ;# jzH1c - jzH1d - movaps xmm4, xmm0 unpcklps xmm0, xmm1 ;# jxH1a jxH1c jyH1a jyH1c unpckhps xmm4, xmm1 ;# jxH1b jxH1d jyH1b jyH1d movaps xmm1, xmm0 unpcklps xmm0, xmm4 ;# x unpckhps xmm1, xmm4 ;# y shufps xmm2, xmm3, 136 ;# 10001000 => jzH1a jzH1b jzH1c jzH1d ;# xmm0 = H1x ;# xmm1 = H1y ;# xmm2 = H1z movaps xmm3, xmm0 movaps xmm4, xmm1 movaps xmm5, xmm2 movaps xmm6, xmm0 movaps xmm7, xmm1 movaps xmm8, xmm2 subps xmm0, [rsp + nb234_ixH1] subps xmm1, [rsp + nb234_iyH1] subps xmm2, [rsp + nb234_izH1] subps xmm3, [rsp + nb234_ixH2] subps xmm4, [rsp + nb234_iyH2] subps xmm5, [rsp + nb234_izH2] subps xmm6, [rsp + nb234_ixM] subps xmm7, [rsp + nb234_iyM] subps xmm8, [rsp + nb234_izM] movaps [rsp + nb234_dxH1H1], xmm0 movaps [rsp + nb234_dyH1H1], xmm1 movaps [rsp + nb234_dzH1H1], xmm2 mulps xmm0, xmm0 mulps xmm1, xmm1 mulps xmm2, xmm2 movaps [rsp + nb234_dxH2H1], xmm3 movaps [rsp + nb234_dyH2H1], xmm4 movaps [rsp + nb234_dzH2H1], xmm5 mulps xmm3, xmm3 mulps xmm4, xmm4 mulps xmm5, xmm5 movaps [rsp + nb234_dxMH1], xmm6 movaps [rsp + nb234_dyMH1], xmm7 movaps [rsp + nb234_dzMH1], xmm8 mulps xmm6, xmm6 mulps xmm7, xmm7 mulps xmm8, xmm8 addps xmm0, xmm1 addps xmm0, xmm2 addps xmm3, xmm4 addps xmm3, xmm5 addps xmm6, xmm7 addps xmm6, xmm8 ;# start doing invsqrt for jH1 atoms rsqrtps xmm1, xmm0 rsqrtps xmm4, xmm3 rsqrtps xmm7, xmm6 movaps xmm2, xmm1 movaps xmm5, xmm4 movaps xmm8, xmm7 mulps xmm1, xmm1 ;# lu*lu mulps xmm4, xmm4 ;# lu*lu mulps xmm7, xmm7 ;# lu*lu movaps xmm9, [rsp + nb234_three] movaps xmm10, xmm9 movaps xmm11, xmm9 mulps xmm1, xmm0 ;# rsq*lu*lu mulps xmm4, xmm3 ;# rsq*lu*lu mulps xmm7, xmm6 ;# rsq*lu*lu subps xmm9, xmm1 subps xmm10, xmm4 subps xmm11, xmm7 ;# 3-rsq*lu*lu mulps xmm9, xmm2 mulps xmm10, xmm5 mulps xmm11, xmm8 ;# lu*(3-rsq*lu*lu) movaps xmm4, [rsp + nb234_half] mulps xmm9, xmm4 ;# rinvH1H1 mulps xmm10, xmm4 ;# rinvH2H1 mulps xmm11, xmm4 ;# rinvMH1 ;# H1 interactions ;# rsq in xmm0,xmm3,xmm6 ;# rinv in xmm9, xmm10, xmm11 movaps xmm1, xmm9 ;# copy of rinv movaps xmm4, xmm10 movaps xmm7, xmm11 movaps xmm2, [rsp + nb234_krf] mulps xmm9, xmm9 ;# rinvsq mulps xmm10, xmm10 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 movaps xmm14, [rsp + nb234_crf] subps xmm2, xmm14 ;# rinv+krsq-crf subps xmm5, xmm14 subps xmm8, xmm14 movaps xmm12, [rsp + nb234_qqHH] movaps xmm13, [rsp + nb234_qqMH] mulps xmm2, xmm12 ;# voul=qq*(rinv+ krsq-crf) mulps xmm5, xmm12 ;# voul=qq*(rinv+ krsq-crf) mulps xmm8, xmm13 ;# 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, xmm12 ;# (rinv-2*krsq)*qq mulps xmm4, xmm12 mulps xmm7, xmm13 addps xmm2, [rsp + nb234_vctot] addps xmm5, xmm8 addps xmm2, xmm5 movaps xmm15, xmm2 mulps xmm1, xmm9 ;# fscal mulps xmm4, xmm10 mulps xmm7, xmm11 ;# move j H1 forces to local temp variables mov rdi, [rbp + nb234_faction] movlps xmm9, [rdi + rax*4 + 12] ;# jxH1a jyH1a - - movlps xmm10, [rdi + rcx*4 + 12] ;# jxH1c jyH1c - - movhps xmm9, [rdi + rbx*4 + 12] ;# jxH1a jyH1a jxH1b jyH1b movhps xmm10, [rdi + rdx*4 + 12] ;# jxH1c jyH1c jxH1d jyH1d movss xmm11, [rdi + rax*4 + 20] ;# jzH1a - - - movss xmm12, [rdi + rcx*4 + 20] ;# jzH1c - - - movhps xmm11, [rdi + rbx*4 + 20] ;# jzH1a - jzH1b - movhps xmm12, [rdi + rdx*4 + 20] ;# jzH1c - jzH1d - shufps xmm11, xmm12, 136 ;# 10001000 => jzH1a jzH1b jzH1c jzH1d ;# xmm9: jxH1a jyH1a jxH1b jyH1b ;# xmm10: jxH1c jyH1c jxH1d jyH1d ;# xmm11: jzH1a jzH1b jzH1c jzH1d movaps xmm0, xmm1 movaps xmm2, xmm1 movaps xmm3, xmm4 movaps xmm5, xmm4 movaps xmm6, xmm7 movaps xmm8, xmm7 mulps xmm0, [rsp + nb234_dxH1H1] mulps xmm1, [rsp + nb234_dyH1H1] mulps xmm2, [rsp + nb234_dzH1H1] mulps xmm3, [rsp + nb234_dxH2H1] mulps xmm4, [rsp + nb234_dyH2H1] mulps xmm5, [rsp + nb234_dzH2H1] mulps xmm6, [rsp + nb234_dxMH1] mulps xmm7, [rsp + nb234_dyMH1] mulps xmm8, [rsp + nb234_dzMH1] movaps xmm13, xmm0 movaps xmm14, xmm1 addps xmm11, xmm2 addps xmm0, [rsp + nb234_fixH1] addps xmm1, [rsp + nb234_fiyH1] addps xmm2, [rsp + nb234_fizH1] addps xmm13, xmm3 addps xmm14, xmm4 addps xmm11, xmm5 addps xmm3, [rsp + nb234_fixH2] addps xmm4, [rsp + nb234_fiyH2] addps xmm5, [rsp + nb234_fizH2] addps xmm13, xmm6 addps xmm14, xmm7 addps xmm11, xmm8 addps xmm6, [rsp + nb234_fixM] addps xmm7, [rsp + nb234_fiyM] addps xmm8, [rsp + nb234_fizM] movaps [rsp + nb234_fixH1], xmm0 movaps [rsp + nb234_fiyH1], xmm1 movaps [rsp + nb234_fizH1], xmm2 movaps [rsp + nb234_fixH2], xmm3 movaps [rsp + nb234_fiyH2], xmm4 movaps [rsp + nb234_fizH2], xmm5 movaps [rsp + nb234_fixM], xmm6 movaps [rsp + nb234_fiyM], xmm7⌨️ 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?