nb_kernel301_x86_64_sse2.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,119 行 · 第 1/4 页
S
2,119 行
;# Load LOTS of table data movlpd xmm0, [rsi + r8*8] movlpd xmm1, [rsi + r8*8 + 8] movlpd xmm2, [rsi + r8*8 + 16] movlpd xmm3, [rsi + r8*8 + 24] movlpd xmm4, [rsi + r10*8] movlpd xmm5, [rsi + r10*8 + 8] movlpd xmm6, [rsi + r10*8 + 16] movlpd xmm7, [rsi + r10*8 + 24] movlpd xmm8, [rsi + r12*8] movlpd xmm9, [rsi + r12*8 + 8] movlpd xmm10, [rsi + r12*8 + 16] movlpd xmm11, [rsi + r12*8 + 24] movhpd xmm0, [rsi + r9*8] movhpd xmm1, [rsi + r9*8 + 8] movhpd xmm2, [rsi + r9*8 + 16] movhpd xmm3, [rsi + r9*8 + 24] movhpd xmm4, [rsi + r11*8] movhpd xmm5, [rsi + r11*8 + 8] movhpd xmm6, [rsi + r11*8 + 16] movhpd xmm7, [rsi + r11*8 + 24] movhpd xmm8, [rsi + r13*8] movhpd xmm9, [rsi + r13*8 + 8] movhpd xmm10, [rsi + r13*8 + 16] movhpd xmm11, [rsi + r13*8 + 24] ;# table data ready in xmm0-xmm3 , xmm4-xmm7 , and xmm8-xmm11 movapd xmm12, [rsp + nb301_epsO] movapd xmm13, [rsp + nb301_epsH1] movapd xmm14, [rsp + nb301_epsH2] mulpd xmm3, xmm12 ;# Heps mulpd xmm7, xmm13 mulpd xmm11, xmm14 mulpd xmm2, xmm12 ;# Geps mulpd xmm6, xmm13 mulpd xmm10, xmm14 mulpd xmm3, xmm12 ;# Heps2 mulpd xmm7, xmm13 mulpd xmm11, xmm14 addpd xmm1, xmm2 ;# F+Geps addpd xmm5, xmm6 addpd xmm9, xmm10 addpd xmm1, xmm3 ;# F+Geps+Heps2 = Fp addpd xmm5, xmm7 addpd xmm9, xmm11 addpd xmm3, xmm3 ;# 2*Heps2 addpd xmm7, xmm7 addpd xmm11, xmm11 addpd xmm3, xmm2 ;# 2*Heps2+Geps addpd xmm7, xmm6 addpd xmm11, xmm10 addpd xmm3, xmm1 ;# FF = Fp + 2*Heps2 + Geps addpd xmm7, xmm5 addpd xmm11, xmm9 mulpd xmm1, xmm12 ;# eps*Fp mulpd xmm5, xmm13 mulpd xmm9, xmm14 movapd xmm12, [rsp + nb301_qqO] movapd xmm13, [rsp + nb301_qqH] addpd xmm1, xmm0 ;# VV addpd xmm5, xmm4 addpd xmm9, xmm8 mulpd xmm1, xmm12 ;# VV*qq = vcoul mulpd xmm5, xmm13 mulpd xmm9, xmm13 mulpd xmm3, xmm12 ;# FF*qq = fij mulpd xmm7, xmm13 mulpd xmm11, xmm13 ;# accumulate vctot addpd xmm1, [rsp + nb301_vctot] addpd xmm5, xmm9 addpd xmm1, xmm5 movapd [rsp + nb301_vctot], xmm1 movapd xmm10, [rsp + nb301_tsc] mulpd xmm3, xmm10 ;# fscal mulpd xmm7, xmm10 mulpd xmm10, xmm11 xorpd xmm4, xmm4 xorpd xmm8, xmm8 xorpd xmm11, xmm11 mulpd xmm3, [rsp + nb301_rinvO] mulpd xmm7, [rsp + nb301_rinvH1] mulpd xmm10, [rsp + nb301_rinvH2] subpd xmm4, xmm3 subpd xmm8, xmm7 subpd xmm11, xmm10 ;# move j forces to xmm0-xmm2 movlpd xmm0, [rdi + rax*8] movlpd xmm1, [rdi + rax*8 + 8] movlpd xmm2, [rdi + rax*8 + 16] movhpd xmm0, [rdi + rbx*8] movhpd xmm1, [rdi + rbx*8 + 8] movhpd xmm2, [rdi + rbx*8 + 16] movapd xmm3, xmm4 movapd xmm5, xmm4 movapd xmm7, xmm8 movapd xmm9, xmm8 movapd xmm10, xmm11 movapd xmm12, xmm11 mulpd xmm3, [rsp + nb301_dxO] mulpd xmm4, [rsp + nb301_dyO] mulpd xmm5, [rsp + nb301_dzO] mulpd xmm7, [rsp + nb301_dxH1] mulpd xmm8, [rsp + nb301_dyH1] mulpd xmm9, [rsp + nb301_dzH1] mulpd xmm10, [rsp + nb301_dxH2] mulpd xmm11, [rsp + nb301_dyH2] mulpd xmm12, [rsp + nb301_dzH2] addpd xmm0, xmm3 addpd xmm1, xmm4 addpd xmm2, xmm5 addpd xmm3, [rsp + nb301_fixO] addpd xmm4, [rsp + nb301_fiyO] addpd xmm5, [rsp + nb301_fizO] addpd xmm0, xmm7 addpd xmm1, xmm8 addpd xmm2, xmm9 addpd xmm7, [rsp + nb301_fixH1] addpd xmm8, [rsp + nb301_fiyH1] addpd xmm9, [rsp + nb301_fizH1] addpd xmm0, xmm10 addpd xmm1, xmm11 addpd xmm2, xmm12 addpd xmm10, [rsp + nb301_fixH2] addpd xmm11, [rsp + nb301_fiyH2] addpd xmm12, [rsp + nb301_fizH2] movapd [rsp + nb301_fixO], xmm3 movapd [rsp + nb301_fiyO], xmm4 movapd [rsp + nb301_fizO], xmm5 movapd [rsp + nb301_fixH1], xmm7 movapd [rsp + nb301_fiyH1], xmm8 movapd [rsp + nb301_fizH1], xmm9 movapd [rsp + nb301_fixH2], xmm10 movapd [rsp + nb301_fiyH2], xmm11 movapd [rsp + nb301_fizH2], xmm12 ;# store back j forces from xmm0-xmm2 movlpd [rdi + rax*8], xmm0 movlpd [rdi + rax*8 + 8], xmm1 movlpd [rdi + rax*8 + 16], xmm2 movhpd [rdi + rbx*8], xmm0 movhpd [rdi + rbx*8 + 8], xmm1 movhpd [rdi + rbx*8 + 16], xmm2 ;# should we do one more iteration? sub dword ptr [rsp + nb301_innerk], 2 jl .nb301_checksingle jmp .nb301_unroll_loop.nb301_checksingle: mov edx, [rsp + nb301_innerk] and edx, 1 jnz .nb301_dosingle jmp .nb301_updateouterdata.nb301_dosingle: mov rdx, [rsp + nb301_innerjjnr] ;# pointer to jjnr[k] mov eax, [rdx] mov rsi, [rbp + nb301_charge] ;# base of charge[] xorpd xmm3, xmm3 movlpd xmm3, [rsi + rax*8] movapd xmm4, xmm3 mulpd xmm3, [rsp + nb301_iqO] mulpd xmm4, [rsp + nb301_iqH] movapd [rsp + nb301_qqO], xmm3 movapd [rsp + nb301_qqH], xmm4 mov rsi, [rbp + nb301_pos] ;# base of pos[] lea rax, [rax + rax*2] ;# replace jnr with j3 ;# move coordinates to xmm0-xmm2 movlpd xmm4, [rsi + rax*8] movlpd xmm5, [rsi + rax*8 + 8] movlpd xmm6, [rsi + rax*8 + 16] movapd xmm0, xmm4 movapd xmm1, xmm5 movapd xmm2, xmm6 ;# calc dr subsd xmm4, [rsp + nb301_ixO] subsd xmm5, [rsp + nb301_iyO] subsd xmm6, [rsp + nb301_izO] ;# store dr movapd [rsp + nb301_dxO], xmm4 movapd [rsp + nb301_dyO], xmm5 movapd [rsp + nb301_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 j coords to xmm4-xmm6 movapd xmm4, xmm0 movapd xmm5, xmm1 movapd xmm6, xmm2 ;# calc dr subsd xmm4, [rsp + nb301_ixH1] subsd xmm5, [rsp + nb301_iyH1] subsd xmm6, [rsp + nb301_izH1] ;# store dr movapd [rsp + nb301_dxH1], xmm4 movapd [rsp + nb301_dyH1], xmm5 movapd [rsp + nb301_dzH1], xmm6 ;# square it mulsd xmm4,xmm4 mulsd xmm5,xmm5 mulsd xmm6,xmm6 addsd xmm6, xmm5 addsd xmm6, xmm4 ;# rsqH1 in xmm6 ;# move j coords to xmm3-xmm5 movapd xmm3, xmm0 movapd xmm4, xmm1 movapd xmm5, xmm2 ;# calc dr subsd xmm3, [rsp + nb301_ixH2] subsd xmm4, [rsp + nb301_iyH2] subsd xmm5, [rsp + nb301_izH2] ;# store dr movapd [rsp + nb301_dxH2], xmm3 movapd [rsp + nb301_dyH2], xmm4 movapd [rsp + nb301_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, [rsp + nb301_three] mulsd xmm2, xmm7 ;# rsq*lu*lu subsd xmm4, xmm2 ;# 30-rsq*lu*lu mulsd xmm4, xmm3 ;# lu*(3-rsq*lu*lu) mulsd xmm4, [rsp + nb301_half] ;# iter1 ( new lu) movapd xmm2, xmm7 movapd xmm3, xmm4 mulsd xmm4, xmm4 ;# lu*lu mulsd xmm2, xmm4 ;# rsq*lu*lu movapd xmm4, [rsp + nb301_three] subsd xmm4, xmm2 ;# 3-rsq*lu*lu mulsd xmm4, xmm3 ;# lu*( 3-rsq*lu*lu) mulsd xmm4, [rsp + nb301_half] ;# rinv movapd [rsp + nb301_rinvO], xmm4 ;# rinvO in xmm4 mulsd xmm7, xmm4 movapd [rsp + nb301_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, [rsp + nb301_three] mulsd xmm2, xmm6 ;# rsq*lu*lu subsd xmm4, xmm2 ;# 30-rsq*lu*lu mulsd xmm4, xmm3 ;# lu*(3-rsq*lu*lu) mulsd xmm4, [rsp + nb301_half] ;# iter1 ( new lu) movapd xmm2, xmm6 movapd xmm3, xmm4 mulsd xmm4, xmm4 ;# lu*lu mulsd xmm2, xmm4 ;# rsq*lu*lu movapd xmm4, [rsp + nb301_three] subsd xmm4, xmm2 ;# 3-rsq*lu*lu mulsd xmm4, xmm3 ;# lu*( 3-rsq*lu*lu) mulsd xmm4, [rsp + nb301_half] ;# rinv movapd [rsp + nb301_rinvH1], xmm4 ;# rinvH1 mulsd xmm6, xmm4 movapd [rsp + nb301_rH1], xmm6 ;# rH1 ;# rsqH2 - seed in xmm2 cvtsd2ss xmm2, xmm5 rsqrtss xmm2, xmm2 cvtss2sd xmm2, xmm2 movapd xmm3, xmm2 mulsd xmm2, xmm2 movapd xmm4, [rsp + nb301_three] mulsd xmm2, xmm5 ;# rsq*lu*lu subsd xmm4, xmm2 ;# 30-rsq*lu*lu mulsd xmm4, xmm3 ;# lu*(3-rsq*lu*lu) mulsd xmm4, [rsp + nb301_half] ;# iter1 ( new lu) movapd xmm2, xmm5 movapd xmm3, xmm4 mulsd xmm4, xmm4 ;# lu*lu mulsd xmm2, xmm4 ;# rsq*lu*lu movapd xmm4, [rsp + nb301_three] subsd xmm4, xmm2 ;# 3-rsq*lu*lu mulsd xmm4, xmm3 ;# lu*( 3-rsq*lu*lu) mulsd xmm4, [rsp + nb301_half] ;# rinv movapd [rsp + nb301_rinvH2], xmm4 ;# rinv mulsd xmm5, xmm4 movapd [rsp + nb301_rH2], xmm5 ;# r ;# do O interactions ;# rO is still in xmm7 mulsd xmm7, [rsp + nb301_tsc] cvttsd2si r8d, xmm7 ;# mm6 = lu idx cvtsi2sd xmm6, r8d subsd xmm7, xmm6 movapd xmm1, xmm7 ;# xmm1=eps movapd xmm2, xmm1 mulsd xmm2, xmm2 ;# xmm2=eps2 shl r8d, 2 ;# idx *= 4 mov rsi, [rbp + nb301_VFtab] movapd xmm4, [rsi + r8*8] ;# Y1 F1 xorpd xmm3, xmm3 movapd xmm5, xmm4 unpcklpd xmm4, xmm3 ;# Y1 unpckhpd xmm5, xmm3 ;# F1 movapd xmm6, [rsi + r8*8 + 16] ;# G1 H1 xorpd xmm3, xmm3 movapd xmm7, xmm6 unpcklpd xmm6, xmm3 ;# G1 unpckhpd xmm7, xmm3 ;# 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, [rsp + nb301_two] ;# two*Heps2 movapd xmm3, [rsp + nb301_qqO] 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 - then we can get rid of mm5 addsd xmm5, [rsp + nb301_vctot] movlpd [rsp + nb301_vctot], xmm5 xorpd xmm4, xmm4 mulsd xmm3, [rsp + nb301_tsc] mulsd xmm3, [rsp + nb301_rinvO] subsd xmm4, xmm3 movapd xmm0, [rsp + nb301_dxO] movapd xmm1, [rsp + nb301_dyO] movapd xmm2, [rsp + nb301_dzO] mulsd xmm0, xmm4 mulsd xmm1, xmm4 mulsd xmm2, xmm4 ;# tx in xmm0-xmm2 ;# update O forces movapd xmm3, [rsp + nb301_fixO] movapd xmm4, [rsp + nb301_fiyO] movapd xmm7, [rsp + nb301_fizO] addsd xmm3, xmm0 addsd xmm4, xmm1 addsd xmm7, xmm2 movlpd [rsp + nb301_fixO], xmm3 movlpd [rsp + nb301_fiyO], xmm4 movlpd [rsp + nb301_fizO], xmm7 ;# update j forces with water O movlpd [rsp + nb301_fjx], xmm0 movlpd [rsp + nb301_fjy], xmm1 movlpd [rsp + nb301_fjz], xmm2 ;# Done with O interactions - now H1! movapd xmm7, [rsp + nb301_rH1] mulsd xmm7, [rsp + nb301_tsc] cvttsd2si r8d, xmm7 ;# mm6 = lu idx cvtsi2sd xmm6, r8d subsd xmm7, xmm6 movapd xmm1, xmm7 ;# xmm1=eps movapd xmm2, xmm1 mulsd xmm2, xmm2 ;# xmm2=eps2 shl r8d, 2 ;# idx *= 4 mov rsi, [rbp + nb301_VFtab] movapd xmm4, [rsi + r8*8] ;# Y1 F1 xorpd xmm3, xmm3 movapd xmm5, xmm4 unpcklpd xmm4, xmm3 ;# Y1 unpckhpd xmm5, xmm3 ;# F1 movapd xmm6, [rsi + r8*8 + 16] ;# G1 H1 xorpd xmm3, xmm3 movapd xmm7, xmm6 unpcklpd xmm6, xmm3 ;# G1 unpckhpd xmm7, xmm3 ;# 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, [rsp + nb301_two] ;# two*Heps2 movapd xmm3, [rsp + nb301_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, [rsp + nb301_vctot] mulsd xmm3, [rsp + nb301_rinvH1] movlpd [rsp + nb301_vctot], xmm5 mulsd xmm3, [rsp + nb301_tsc] subsd xmm4, xmm3 movapd xmm0, [rsp + nb301_dxH1] movapd xmm1, [rsp + nb301_dyH1] movapd xmm2, [rsp + nb301_dzH1] mulsd xmm0, xmm4 mulsd xmm1, xmm4 mulsd xmm2, xmm4 ;# update H1 forces movapd xmm3, [rsp + nb301_fixH1] movapd xmm4, [rsp + nb301_fiyH1] movapd xmm7, [rsp + nb301_fizH1] addsd xmm3, xmm0 addsd xmm4, xmm1 addsd xmm7, xmm2 movlpd [rsp + nb301_fixH1], xmm3 movlpd [rsp + nb301_fiyH1], xmm4 movlpd [rsp + nb301_fizH1], xmm7 ;# update j forces with water H1 addsd xmm0, [rsp + nb301_fjx] addsd xmm1, [rsp + nb301_fjy] addsd xmm2, [rsp + nb301_fjz] movlpd [rsp + nb301_fjx], xmm0 movlpd [rsp + nb301_fjy], xmm1 movlpd [rsp + nb301_fjz], xmm2 ;# Done with H1, finally we do H2 interactions movapd xmm7, [rsp + nb301_rH2] mulsd xmm7, [rsp + nb301_tsc] cvttsd2si r8d, xmm7 ;# mm6 = lu idx cvtsi2sd xmm6, r8d subsd xmm7, xmm6 movapd xmm1, xmm7 ;# xmm1=eps movapd xmm2, xmm1 mulsd xmm2, xmm2 ;# xmm2=eps2 shl r8d, 2 ;# idx *= 4 mov rsi, [rbp + nb301_VFtab] movapd xmm4, [rsi + r8*8] ;# Y1 F1 xorpd xmm3, xmm3 movapd xmm5, xmm4 unpcklpd xmm4, xmm3 ;# Y1 unpckhpd xmm5, xmm3 ;# F1 movapd xmm6, [rsi + r8*8 + 16] ;# G1 H1 xorpd xmm3, xmm3 movapd xmm7, xmm6 unpcklpd xmm6, xmm3 ;# G1 unpckhpd xmm7, xmm3 ;# 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, [rsp + nb301_two] ;# two*Heps2 movapd xmm3, [rsp + nb301_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, [rsp + nb301_vctot] mulsd xmm3, [rsp + nb301_rinvH2] movlpd [rsp + nb301_vctot], xmm5 mulsd xmm3, [rsp + nb301_tsc] subsd xmm4, xmm3 movapd xmm0, [rsp + nb301_dxH2] movapd xmm1, [rsp + nb301_dyH2] movapd xmm2, [rsp + nb301_dzH2] mulsd xmm0, xmm4 mulsd xmm1, xmm4 mulsd xmm2, xmm4 ;# update H2 forces movapd xmm3, [rsp + nb301_fixH2] movapd xmm4, [rsp + nb301_fiyH2] movapd xmm7, [rsp + nb301_fizH2] addsd 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?