nb_kernel430_ia32_sse2.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 1,716 行 · 第 1/4 页
S
1,716 行
lea eax, [eax + eax*2] ;# replace jnr with j3 lea ebx, [ebx + ebx*2] ;# move two coordinates to xmm0-xmm2 movlpd xmm0, [esi + eax*8] movlpd xmm1, [esi + eax*8 + 8] movlpd xmm2, [esi + eax*8 + 16] movhpd xmm0, [esi + ebx*8] movhpd xmm1, [esi + ebx*8 + 8] movhpd xmm2, [esi + ebx*8 + 16] mov edi, [ebp + nb430nf_faction] ;# move nb430nf_ix-iz to xmm4-xmm6 movapd xmm4, [esp + nb430nf_ix] movapd xmm5, [esp + nb430nf_iy] movapd xmm6, [esp + nb430nf_iz] ;# calc dr subpd xmm4, xmm0 subpd xmm5, xmm1 subpd xmm6, xmm2 ;# square it mulpd xmm4,xmm4 mulpd xmm5,xmm5 mulpd xmm6,xmm6 addpd xmm4, xmm5 addpd xmm4, xmm6 ;# rsq in xmm4 cvtpd2ps xmm5, xmm4 rsqrtps xmm5, xmm5 cvtps2pd xmm2, xmm5 ;# lu in low xmm2 ;# lookup seed in xmm2 movapd xmm5, xmm2 ;# copy of lu mulpd xmm2, xmm2 ;# lu*lu movapd xmm1, [esp + nb430nf_three] mulpd xmm2, xmm4 ;# rsq*lu*lu movapd xmm0, [esp + nb430nf_half] subpd xmm1, xmm2 ;# 30-rsq*lu*lu mulpd xmm1, xmm5 mulpd xmm1, xmm0 ;# xmm0=iter1 of rinv (new lu) movapd xmm5, xmm1 ;# copy of lu mulpd xmm1, xmm1 ;# lu*lu movapd xmm2, [esp + nb430nf_three] mulpd xmm1, xmm4 ;# rsq*lu*lu movapd xmm0, [esp + nb430nf_half] subpd xmm2, xmm1 ;# 30-rsq*lu*lu mulpd xmm2, xmm5 mulpd xmm0, xmm2 ;# xmm0=iter2 of rinv mulpd xmm4, xmm0 ;# xmm4=r movapd [esp + nb430nf_r], xmm4 mulpd xmm4, [esp + nb430nf_gbscale] cvttpd2pi mm6, xmm4 ;# mm6 = lu idx cvtpi2pd xmm5, mm6 subpd xmm4, xmm5 movapd xmm1, xmm4 ;# xmm1=eps movapd xmm2, xmm1 mulpd xmm2, xmm2 ;# xmm2=eps2 pslld mm6, 2 ;# idx *= 4 mov esi, [ebp + nb430nf_GBtab] movd ecx, mm6 psrlq mm6, 32 movd edx, mm6 ;# indices in eax/ebx ;# Coulomb movapd xmm4, [esi + ecx*8] ;# Y1 F1 movapd xmm3, [esi + edx*8] ;# Y2 F2 movapd xmm5, xmm4 unpcklpd xmm4, xmm3 ;# Y1 Y2 unpckhpd xmm5, xmm3 ;# F1 F2 movapd xmm6, [esi + ecx*8 + 16] ;# G1 H1 movapd xmm3, [esi + edx*8 + 16] ;# G2 H2 movapd xmm7, xmm6 unpcklpd xmm6, xmm3 ;# G1 G2 unpckhpd xmm7, xmm3 ;# H1 H2 ;# coulomb table ready, in xmm4-xmm7 mulpd xmm6, xmm1 ;# xmm6=Geps mulpd xmm7, xmm2 ;# xmm7=Heps2 addpd xmm5, xmm6 addpd xmm5, xmm7 ;# xmm5=Fp movapd xmm3, [esp + nb430nf_qq] mulpd xmm5, xmm1 ;# xmm5=eps*Fp addpd xmm5, xmm4 ;# xmm5=VV mulpd xmm5, xmm3 ;# vcoul=qq*VV addpd xmm5, [esp + nb430nf_vctot] movapd [esp + nb430nf_vctot], xmm5 movapd xmm4, [esp + nb430nf_r] mulpd xmm4, [esp + nb430nf_tsc] cvttpd2pi mm6, xmm4 ;# mm6 = lu idx cvtpi2pd xmm5, mm6 subpd xmm4, xmm5 movapd xmm1, xmm4 ;# xmm1=eps movapd xmm2, xmm1 mulpd xmm2, xmm2 ;# xmm2=eps2 pslld mm6, 3 ;# idx *= 8 mov esi, [ebp + nb430nf_VFtab] movd ecx, mm6 psrlq mm6, 32 movd edx, mm6 ;# indices in eax/ebx ;# Dispersion movapd xmm4, [esi + ecx*8] ;# Y1 F1 movapd xmm3, [esi + edx*8] ;# Y2 F2 movapd xmm5, xmm4 unpcklpd xmm4, xmm3 ;# Y1 Y2 unpckhpd xmm5, xmm3 ;# F1 F2 movapd xmm6, [esi + ecx*8 + 16] ;# G1 H1 movapd xmm3, [esi + edx*8 + 16] ;# G2 H2 movapd xmm7, xmm6 unpcklpd xmm6, xmm3 ;# G1 G2 unpckhpd xmm7, xmm3 ;# H1 H2 ;# Dispersion table ready, in xmm4-xmm7 mulpd xmm6, xmm1 ;# xmm6=Geps mulpd xmm7, xmm2 ;# xmm7=Heps2 addpd xmm5, xmm6 addpd xmm5, xmm7 ;# xmm5=Fp mulpd xmm5, xmm1 ;# xmm5=eps*Fp addpd xmm5, xmm4 ;# xmm5=VV mulpd xmm5, [esp + nb430nf_c6] ;# Vvdw6 addpd xmm5, [esp + nb430nf_Vvdwtot] movapd [esp + nb430nf_Vvdwtot], xmm5 ;# Repulsion movapd xmm4, [esi + ecx*8 + 32] ;# Y1 F1 movapd xmm3, [esi + edx*8 + 32] ;# Y2 F2 movapd xmm5, xmm4 unpcklpd xmm4, xmm3 ;# Y1 Y2 unpckhpd xmm5, xmm3 ;# F1 F2 movapd xmm6, [esi + ecx*8 + 48] ;# G1 H1 movapd xmm3, [esi + edx*8 + 48] ;# G2 H2 movapd xmm7, xmm6 unpcklpd xmm6, xmm3 ;# G1 G2 unpckhpd xmm7, xmm3 ;# H1 H2 ;# Dispersion table ready, in xmm4-xmm7 mulpd xmm6, xmm1 ;# xmm6=Geps mulpd xmm7, xmm2 ;# xmm7=Heps2 addpd xmm5, xmm6 addpd xmm5, xmm7 ;# xmm5=Fp mulpd xmm5, xmm1 ;# xmm5=eps*Fp addpd xmm5, xmm4 ;# xmm5=VV mulpd xmm5, [esp + nb430nf_c12] ;# Vvdw12 addpd xmm5, [esp + nb430nf_Vvdwtot] movapd [esp + nb430nf_Vvdwtot], xmm5 xorpd xmm4, xmm4 ;# should we do one more iteration? sub dword ptr [esp + nb430nf_innerk], 2 jl .nb430nf_checksingle jmp .nb430nf_unroll_loop.nb430nf_checksingle: mov edx, [esp + nb430nf_innerk] and edx, 1 jnz .nb430nf_dosingle jmp .nb430nf_updateouterdata.nb430nf_dosingle: mov esi, [ebp + nb430nf_charge] mov edx, [ebp + nb430nf_invsqrta] mov edi, [ebp + nb430nf_pos] mov ecx, [esp + nb430nf_innerjjnr] mov eax, [ecx] xorpd xmm6, xmm6 movapd xmm7, xmm6 movsd xmm7, [edx + eax*8] movlpd xmm6, [esi + eax*8] ;# xmm6(0) has the charge mulsd xmm7, [esp + nb430nf_isai] movapd [esp + nb430nf_isaprod], xmm7 movapd xmm1, xmm7 mulpd xmm1, [esp + nb430nf_gbtsc] movapd [esp + nb430nf_gbscale], xmm1 mulsd xmm7, [esp + nb430nf_iq] mulsd xmm6, xmm7 movapd [esp + nb430nf_qq], xmm6 mov esi, [ebp + nb430nf_type] mov edx, [esi + eax*4] mov esi, [ebp + nb430nf_vdwparam] shl edx, 1 mov edi, [esp + nb430nf_ntia] add edx, edi movlpd xmm6, [esi + edx*8] ;# c6a movhpd xmm6, [esi + edx*8 + 8] ;# c6a c12a xorpd xmm7, xmm7 movapd xmm4, xmm6 unpcklpd xmm4, xmm7 unpckhpd xmm6, xmm7 movapd [esp + nb430nf_c6], xmm4 movapd [esp + nb430nf_c12], xmm6 mov esi, [ebp + nb430nf_pos] ;# base of pos[] lea eax, [eax + eax*2] ;# replace jnr with j3 ;# move two coordinates to xmm0-xmm2 movlpd xmm0, [esi + eax*8] movlpd xmm1, [esi + eax*8 + 8] movlpd xmm2, [esi + eax*8 + 16] mov edi, [ebp + nb430nf_faction] ;# move nb430nf_ix-iz to xmm4-xmm6 movapd xmm4, [esp + nb430nf_ix] movapd xmm5, [esp + nb430nf_iy] movapd xmm6, [esp + nb430nf_iz] ;# calc dr subsd xmm4, xmm0 subsd xmm5, xmm1 subsd xmm6, xmm2 ;# square it mulsd xmm4,xmm4 mulsd xmm5,xmm5 mulsd xmm6,xmm6 addsd xmm4, xmm5 addsd xmm4, xmm6 ;# rsq in xmm4 cvtsd2ss xmm5, xmm4 rsqrtss xmm5, xmm5 cvtss2sd xmm2, xmm5 ;# lu in low xmm2 ;# lookup seed in xmm2 movapd xmm5, xmm2 ;# copy of lu mulsd xmm2, xmm2 ;# lu*lu movapd xmm1, [esp + nb430nf_three] mulsd xmm2, xmm4 ;# rsq*lu*lu movapd xmm0, [esp + nb430nf_half] subsd xmm1, xmm2 ;# 30-rsq*lu*lu mulsd xmm1, xmm5 mulsd xmm1, xmm0 ;# xmm0=iter1 of rinv (new lu) movapd xmm5, xmm1 ;# copy of lu mulsd xmm1, xmm1 ;# lu*lu movapd xmm2, [esp + nb430nf_three] mulsd xmm1, xmm4 ;# rsq*lu*lu movapd xmm0, [esp + nb430nf_half] subsd xmm2, xmm1 ;# 30-rsq*lu*lu mulsd xmm2, xmm5 mulsd xmm0, xmm2 ;# xmm0=iter2 of rinv (new lu) mulsd xmm4, xmm0 ;# xmm4=r movsd [esp + nb430nf_r], xmm4 mulsd xmm4, [esp + nb430nf_gbscale] cvttsd2si edx, xmm4 ;# mm6 = lu idx cvtsi2sd xmm5, edx subsd xmm4, xmm5 movapd xmm1, xmm4 ;# xmm1=eps movapd xmm2, xmm1 mulsd xmm2, xmm2 ;# xmm2=eps2 shl edx, 2 ;# idx *= 4 mov esi, [ebp + nb430nf_GBtab] ;# Coulomb movapd xmm4, [esi + edx*8] ;# Y1 F1 xorpd xmm3, xmm3 movapd xmm5, xmm4 unpcklpd xmm4, xmm3 ;# Y1 unpckhpd xmm5, xmm3 ;# F1 movapd xmm6, [esi + edx*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 movapd xmm3, [esp + nb430nf_qq] mulsd xmm5, xmm1 ;# xmm5=eps*Fp addsd xmm5, xmm4 ;# xmm5=VV mulsd xmm5, xmm3 ;# vcoul=qq*VV addsd xmm5, [esp + nb430nf_vctot] movsd [esp + nb430nf_vctot], xmm5 movsd xmm4, [esp + nb430nf_r] mulsd xmm4, [esp + nb430nf_tsc] cvttsd2si edx, xmm4 ;# mm6 = lu idx cvtsi2sd xmm5, edx subsd xmm4, xmm5 movsd xmm1, xmm4 ;# xmm1=eps movsd xmm2, xmm1 mulsd xmm2, xmm2 ;# xmm2=eps2 shl edx, 3 mov esi, [ebp + nb430nf_VFtab] ;# Dispersion movapd xmm4, [esi + edx*8] ;# Y1 F1 xorpd xmm3, xmm3 movapd xmm5, xmm4 unpcklpd xmm4, xmm3 ;# Y1 unpckhpd xmm5, xmm3 ;# F1 movapd xmm6, [esi + edx*8 + 16] ;# G1 H1 xorpd xmm3, xmm3 movapd xmm7, xmm6 unpcklpd xmm6, xmm3 ;# G1 unpckhpd xmm7, xmm3 ;# 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 xmm5, xmm1 ;# xmm5=eps*Fp addsd xmm5, xmm4 ;# xmm5=VV mulsd xmm5, [esp + nb430nf_c6] ;# Vvdw6 addsd xmm5, [esp + nb430nf_Vvdwtot] movlpd [esp + nb430nf_Vvdwtot], xmm5 ;# Repulsion movapd xmm4, [esi + edx*8 + 32] ;# Y1 F1 xorpd xmm3, xmm3 movapd xmm5, xmm4 unpcklpd xmm4, xmm3 ;# Y1 unpckhpd xmm5, xmm3 ;# F1 movapd xmm6, [esi + edx*8 + 48] ;# G1 H1 xorpd xmm3, xmm3 movapd xmm7, xmm6 unpcklpd xmm6, xmm3 ;# G1 unpckhpd xmm7, xmm3 ;# 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 xmm5, xmm1 ;# xmm5=eps*Fp addsd xmm5, xmm4 ;# xmm5=VV mulsd xmm5, [esp + nb430nf_c12] ;# Vvdw12 addsd xmm5, [esp + nb430nf_Vvdwtot] movlpd [esp + nb430nf_Vvdwtot], xmm5.nb430nf_updateouterdata: ;# get n from stack mov esi, [esp + nb430nf_n] ;# get group index for i particle mov edx, [ebp + nb430nf_gid] ;# base of gid[] mov edx, [edx + esi*4] ;# ggid=gid[n] ;# accumulate total potential energy and update it movapd xmm7, [esp + nb430nf_vctot] ;# accumulate movhlps xmm6, xmm7 addsd xmm7, xmm6 ;# low xmm7 has the sum now ;# add earlier value from mem mov eax, [ebp + nb430nf_Vc] addsd xmm7, [eax + edx*8] ;# move back to mem movsd [eax + edx*8], xmm7 ;# accumulate total lj energy and update it movapd xmm7, [esp + nb430nf_Vvdwtot] ;# accumulate movhlps xmm6, xmm7 addsd xmm7, xmm6 ;# low xmm7 has the sum now ;# add earlier value from mem mov eax, [ebp + nb430nf_Vvdw] addsd xmm7, [eax + edx*8] ;# move back to mem movsd [eax + edx*8], xmm7 ;# finish if last mov ecx, [esp + nb430nf_nn1] ;# esi already loaded with n inc esi sub ecx, esi jz .nb430nf_outerend ;# not last, iterate outer loop once more! mov [esp + nb430nf_n], esi jmp .nb430nf_outer.nb430nf_outerend: ;# check if more outer neighborlists remain mov ecx, [esp + nb430nf_nri] ;# esi already loaded with n above sub ecx, esi jz .nb430nf_end ;# non-zero, do one more workunit jmp .nb430nf_threadloop.nb430nf_end: emms mov eax, [esp + nb430nf_nouter] mov ebx, [esp + nb430nf_ninner] mov ecx, [ebp + nb430nf_outeriter] mov edx, [ebp + nb430nf_inneriter] mov [ecx], eax mov [edx], ebx mov eax, [esp + nb430nf_salign] add esp, eax add esp, 328 pop edi pop esi pop edx pop ecx pop ebx pop eax leave ret⌨️ 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?