nb_kernel111_ia32_sse2.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 1,864 行 · 第 1/4 页
S
1,864 行
mov esi, [ebp + nb111nf_type] mov eax, [esi + eax*4] mov ebx, [esi + ebx*4] mov esi, [ebp + nb111nf_vdwparam] shl eax, 1 shl ebx, 1 mov edi, [esp + nb111nf_ntia] add eax, edi add ebx, edi movlpd xmm6, [esi + eax*8] ;# c6a movlpd xmm7, [esi + ebx*8] ;# c6b movhpd xmm6, [esi + eax*8 + 8] ;# c6a c12a movhpd xmm7, [esi + ebx*8 + 8] ;# c6b c12b movapd xmm4, xmm6 unpcklpd xmm4, xmm7 unpckhpd xmm6, xmm7 movd eax, mm0 movd ebx, mm1 movapd [esp + nb111nf_c6], xmm4 movapd [esp + nb111nf_c12], xmm6 mov esi, [ebp + nb111nf_pos] ;# base of pos[] 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] ;# move ixO-izO to xmm4-xmm6 movapd xmm4, [esp + nb111nf_ixO] movapd xmm5, [esp + nb111nf_iyO] movapd xmm6, [esp + nb111nf_izO] ;# 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 movapd xmm7, xmm4 ;# rsqO in xmm7 ;# move ixH1-izH1 to xmm4-xmm6 movapd xmm4, [esp + nb111nf_ixH1] movapd xmm5, [esp + nb111nf_iyH1] movapd xmm6, [esp + nb111nf_izH1] ;# calc dr subpd xmm4, xmm0 subpd xmm5, xmm1 subpd xmm6, xmm2 ;# square it mulpd xmm4,xmm4 mulpd xmm5,xmm5 mulpd xmm6,xmm6 addpd xmm6, xmm5 addpd xmm6, xmm4 ;# rsqH1 in xmm6 ;# move ixH2-izH2 to xmm3-xmm5 movapd xmm3, [esp + nb111nf_ixH2] movapd xmm4, [esp + nb111nf_iyH2] movapd xmm5, [esp + nb111nf_izH2] ;# calc dr subpd xmm3, xmm0 subpd xmm4, xmm1 subpd xmm5, xmm2 ;# square it mulpd xmm3,xmm3 mulpd xmm4,xmm4 mulpd xmm5,xmm5 addpd xmm5, xmm4 addpd xmm5, xmm3 ;# rsqH2 in xmm5, rsqH1 in xmm6, rsqO in xmm7 ;# start with rsqO - put seed in xmm2 cvtpd2ps xmm2, xmm7 rsqrtps xmm2, xmm2 cvtps2pd xmm2, xmm2 movapd xmm3, xmm2 mulpd xmm2, xmm2 movapd xmm4, [esp + nb111nf_three] mulpd xmm2, xmm7 ;# rsq*lu*lu subpd xmm4, xmm2 ;# 30-rsq*lu*lu mulpd xmm4, xmm3 ;# lu*(3-rsq*lu*lu) mulpd xmm4, [esp + nb111nf_half] ;# iter1 ( new lu) movapd xmm3, xmm4 mulpd xmm4, xmm4 ;# lu*lu mulpd xmm7, xmm4 ;# rsq*lu*lu movapd xmm4, [esp + nb111nf_three] subpd xmm4, xmm7 ;# 3-rsq*lu*lu mulpd xmm4, xmm3 ;# lu*( 3-rsq*lu*lu) mulpd xmm4, [esp + nb111nf_half] ;# rinv movapd xmm7, xmm4 ;# rinvO in xmm7 ;# rsqH1 - seed in xmm2 cvtpd2ps xmm2, xmm6 rsqrtps xmm2, xmm2 cvtps2pd xmm2, xmm2 movapd xmm3, xmm2 mulpd xmm2, xmm2 movapd xmm4, [esp + nb111nf_three] mulpd xmm2, xmm6 ;# rsq*lu*lu subpd xmm4, xmm2 ;# 30-rsq*lu*lu mulpd xmm4, xmm3 ;# lu*(3-rsq*lu*lu) mulpd xmm4, [esp + nb111nf_half] ;# iter1 ( new lu) movapd xmm3, xmm4 mulpd xmm4, xmm4 ;# lu*lu mulpd xmm6, xmm4 ;# rsq*lu*lu movapd xmm4, [esp + nb111nf_three] subpd xmm4, xmm6 ;# 3-rsq*lu*lu mulpd xmm4, xmm3 ;# lu*( 3-rsq*lu*lu) mulpd xmm4, [esp + nb111nf_half] ;# rinv movapd xmm6, xmm4 ;# rinvH1 in xmm6 ;# rsqH2 - seed in xmm2 cvtpd2ps xmm2, xmm5 rsqrtps xmm2, xmm2 cvtps2pd xmm2, xmm2 movapd xmm3, xmm2 mulpd xmm2, xmm2 movapd xmm4, [esp + nb111nf_three] mulpd xmm2, xmm5 ;# rsq*lu*lu subpd xmm4, xmm2 ;# 30-rsq*lu*lu mulpd xmm4, xmm3 ;# lu*(3-rsq*lu*lu) mulpd xmm4, [esp + nb111nf_half] ;# iter1 ( new lu) movapd xmm3, xmm4 mulpd xmm4, xmm4 ;# lu*lu mulpd xmm5, xmm4 ;# rsq*lu*lu movapd xmm4, [esp + nb111nf_three] subpd xmm4, xmm5 ;# 3-rsq*lu*lu mulpd xmm4, xmm3 ;# lu*( 3-rsq*lu*lu) mulpd xmm4, [esp + nb111nf_half] ;# rinv movapd xmm5, xmm4 ;# rinvH2 in xmm5 ;# do O interactions movapd xmm4, xmm7 mulpd xmm4, xmm4 ;# xmm7=rinv, xmm4=rinvsq movapd xmm1, xmm4 mulpd xmm1, xmm4 mulpd xmm1, xmm4 ;# xmm1=rinvsix movapd xmm2, xmm1 mulpd xmm2, xmm2 ;# xmm2=rinvtwelve mulpd xmm7, [esp + nb111nf_qqO] ;# xmm7=vcoul mulpd xmm1, [esp + nb111nf_c6] mulpd xmm2, [esp + nb111nf_c12] movapd xmm3, xmm2 subpd xmm3, xmm1 ;# Vvdw=Vvdw12-Vvdw6 addpd xmm3, [esp + nb111nf_Vvdwtot] addpd xmm7, [esp + nb111nf_vctot] movapd [esp + nb111nf_Vvdwtot], xmm3 movapd [esp + nb111nf_vctot], xmm7 ;# H1 interactions mulpd xmm6, [esp + nb111nf_qqH] ;# xmm6=vcoul addpd xmm6, [esp + nb111nf_vctot] movapd [esp + nb111nf_vctot], xmm6 ;# H2 interactions mulpd xmm5, [esp + nb111nf_qqH] ;# xmm5=vcoul addpd xmm5, [esp + nb111nf_vctot] movapd [esp + nb111nf_vctot], xmm5 ;# should we do one more iteration? sub dword ptr [esp + nb111nf_innerk], 2 jl .nb111nf_checksingle jmp .nb111nf_unroll_loop.nb111nf_checksingle: mov edx, [esp + nb111nf_innerk] and edx, 1 jnz .nb111nf_dosingle jmp .nb111nf_updateouterdata.nb111nf_dosingle: mov edx, [esp + nb111nf_innerjjnr] ;# pointer to jjnr[k] mov eax, [edx] add dword ptr [esp + nb111nf_innerjjnr], 4 mov esi, [ebp + nb111nf_charge] ;# base of charge[] xorpd xmm3, xmm3 movlpd xmm3, [esi + eax*8] movapd xmm4, xmm3 mulpd xmm3, [esp + nb111nf_iqO] mulpd xmm4, [esp + nb111nf_iqH] movd mm0, eax ;# use mmx registers as temp storage movapd [esp + nb111nf_qqO], xmm3 movapd [esp + nb111nf_qqH], xmm4 mov esi, [ebp + nb111nf_type] mov eax, [esi + eax*4] mov esi, [ebp + nb111nf_vdwparam] shl eax, 1 mov edi, [esp + nb111nf_ntia] add eax, edi movlpd xmm6, [esi + eax*8] ;# c6a movhpd xmm6, [esi + eax*8 + 8] ;# c6a c12a xorpd xmm7, xmm7 movapd xmm4, xmm6 unpcklpd xmm4, xmm7 unpckhpd xmm6, xmm7 movd eax, mm0 movd ebx, mm1 movapd [esp + nb111nf_c6], xmm4 movapd [esp + nb111nf_c12], xmm6 mov esi, [ebp + nb111nf_pos] ;# base of pos[] lea eax, [eax + eax*2] ;# replace jnr with j3 ;# move coordinates to xmm0-xmm2 movlpd xmm0, [esi + eax*8] movlpd xmm1, [esi + eax*8 + 8] movlpd xmm2, [esi + eax*8 + 16] ;# move ixO-izO to xmm4-xmm6 movapd xmm4, [esp + nb111nf_ixO] movapd xmm5, [esp + nb111nf_iyO] movapd xmm6, [esp + nb111nf_izO] ;# 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 movapd xmm7, xmm4 ;# rsqO in xmm7 ;# move ixH1-izH1 to xmm4-xmm6 movapd xmm4, [esp + nb111nf_ixH1] movapd xmm5, [esp + nb111nf_iyH1] movapd xmm6, [esp + nb111nf_izH1] ;# calc dr subsd xmm4, xmm0 subsd xmm5, xmm1 subsd xmm6, xmm2 ;# square it mulsd xmm4,xmm4 mulsd xmm5,xmm5 mulsd xmm6,xmm6 addsd xmm6, xmm5 addsd xmm6, xmm4 ;# rsqH1 in xmm6 ;# move ixH2-izH2 to xmm3-xmm5 movapd xmm3, [esp + nb111nf_ixH2] movapd xmm4, [esp + nb111nf_iyH2] movapd xmm5, [esp + nb111nf_izH2] ;# calc dr subsd xmm3, xmm0 subsd xmm4, xmm1 subsd xmm5, xmm2 ;# 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, [esp + nb111nf_three] mulsd xmm2, xmm7 ;# rsq*lu*lu subsd xmm4, xmm2 ;# 30-rsq*lu*lu mulsd xmm4, xmm3 ;# lu*(3-rsq*lu*lu) mulsd xmm4, [esp + nb111nf_half] ;# iter1 ( new lu) movapd xmm3, xmm4 mulsd xmm4, xmm4 ;# lu*lu mulsd xmm7, xmm4 ;# rsq*lu*lu movapd xmm4, [esp + nb111nf_three] subsd xmm4, xmm7 ;# 3-rsq*lu*lu mulsd xmm4, xmm3 ;# lu*( 3-rsq*lu*lu) mulsd xmm4, [esp + nb111nf_half] ;# rinv movapd xmm7, xmm4 ;# rinvO in xmm7 ;# rsqH1 - seed in xmm2 cvtsd2ss xmm2, xmm6 rsqrtss xmm2, xmm2 cvtss2sd xmm2, xmm2 movapd xmm3, xmm2 mulsd xmm2, xmm2 movapd xmm4, [esp + nb111nf_three] mulsd xmm2, xmm6 ;# rsq*lu*lu subsd xmm4, xmm2 ;# 30-rsq*lu*lu mulsd xmm4, xmm3 ;# lu*(3-rsq*lu*lu) mulsd xmm4, [esp + nb111nf_half] ;# iter1 ( new lu) movapd xmm3, xmm4 mulsd xmm4, xmm4 ;# lu*lu mulsd xmm6, xmm4 ;# rsq*lu*lu movapd xmm4, [esp + nb111nf_three] subsd xmm4, xmm6 ;# 3-rsq*lu*lu mulsd xmm4, xmm3 ;# lu*( 3-rsq*lu*lu) mulsd xmm4, [esp + nb111nf_half] ;# rinv movapd xmm6, xmm4 ;# rinvH1 in xmm6 ;# rsqH2 - seed in xmm2 cvtsd2ss xmm2, xmm5 rsqrtss xmm2, xmm2 cvtss2sd xmm2, xmm2 movapd xmm3, xmm2 mulsd xmm2, xmm2 movapd xmm4, [esp + nb111nf_three] mulsd xmm2, xmm5 ;# rsq*lu*lu subsd xmm4, xmm2 ;# 30-rsq*lu*lu mulsd xmm4, xmm3 ;# lu*(3-rsq*lu*lu) mulsd xmm4, [esp + nb111nf_half] ;# iter1 ( new lu) movapd xmm3, xmm4 mulsd xmm4, xmm4 ;# lu*lu mulsd xmm5, xmm4 ;# rsq*lu*lu movapd xmm4, [esp + nb111nf_three] subsd xmm4, xmm5 ;# 3-rsq*lu*lu mulsd xmm4, xmm3 ;# lu*( 3-rsq*lu*lu) mulsd xmm4, [esp + nb111nf_half] ;# rinv movapd xmm5, xmm4 ;# rinvH2 in xmm5 ;# do O interactions movapd xmm4, xmm7 mulsd xmm4, xmm4 ;# xmm7=rinv, xmm4=rinvsq movapd xmm1, xmm4 mulsd xmm1, xmm4 mulsd xmm1, xmm4 ;# xmm1=rinvsix movapd xmm2, xmm1 mulsd xmm2, xmm2 ;# xmm2=rinvtwelve mulsd xmm7, [esp + nb111nf_qqO] ;# xmm7=vcoul mulsd xmm1, [esp + nb111nf_c6] mulsd xmm2, [esp + nb111nf_c12] movapd xmm3, xmm2 subsd xmm3, xmm1 ;# Vvdw=Vvdw12-Vvdw6 addsd xmm3, [esp + nb111nf_Vvdwtot] addsd xmm7, [esp + nb111nf_vctot] movsd [esp + nb111nf_Vvdwtot], xmm3 movsd [esp + nb111nf_vctot], xmm7 ;# H1 interactions mulsd xmm6, [esp + nb111nf_qqH] ;# xmm6=vcoul addsd xmm6, [esp + nb111nf_vctot] movsd [esp + nb111nf_vctot], xmm6 ;# H2 interactions mulsd xmm5, [esp + nb111nf_qqH] ;# xmm5=vcoul addsd xmm5, [esp + nb111nf_vctot] movsd [esp + nb111nf_vctot], xmm5 .nb111nf_updateouterdata: ;# get n from stack mov esi, [esp + nb111nf_n] ;# get group index for i particle mov edx, [ebp + nb111nf_gid] ;# base of gid[] mov edx, [edx + esi*4] ;# ggid=gid[n] ;# accumulate total potential energy and update it movapd xmm7, [esp + nb111nf_vctot] ;# accumulate movhlps xmm6, xmm7 addsd xmm7, xmm6 ;# low xmm7 has the sum now ;# add earlier value from mem mov eax, [ebp + nb111nf_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 + nb111nf_Vvdwtot] ;# accumulate movhlps xmm6, xmm7 addsd xmm7, xmm6 ;# low xmm7 has the sum now ;# add earlier value from mem mov eax, [ebp + nb111nf_Vvdw] addsd xmm7, [eax + edx*8] ;# move back to mem movsd [eax + edx*8], xmm7 ;# finish if last mov ecx, [esp + nb111nf_nn1] ;# esi already loaded with n inc esi sub ecx, esi jz .nb111nf_outerend ;# not last, iterate outer loop once more! mov [esp + nb111nf_n], esi jmp .nb111nf_outer.nb111nf_outerend: ;# check if more outer neighborlists remain mov ecx, [esp + nb111nf_nri] ;# esi already loaded with n above sub ecx, esi jz .nb111nf_end ;# non-zero, do one more workunit jmp .nb111nf_threadloop.nb111nf_end: emms mov eax, [esp + nb111nf_nouter] mov ebx, [esp + nb111nf_ninner] mov ecx, [ebp + nb111nf_outeriter] mov edx, [ebp + nb111nf_inneriter] mov [ecx], eax mov [edx], ebx mov eax, [esp + nb111nf_salign] add esp, eax add esp, 348 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?