nb_kernel333_ia32_sse2.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,347 行 · 第 1/5 页
S
2,347 行
movsd xmm3, [edx + ebx*8 + 8] movsd xmm4, [edx + ebx*8 + 24] mov esi, [ebp + nb333nf_p_facel] movsd xmm5, [esi] mulsd xmm3, xmm5 mulsd xmm4, xmm5 shufpd xmm3, xmm3, 0 shufpd xmm4, xmm4, 0 movapd [esp + nb333nf_iqH], xmm3 movapd [esp + nb333nf_iqM], xmm4 mov edx, [ebp + nb333nf_type] mov ecx, [edx + ebx*4] shl ecx, 1 mov edi, [ebp + nb333nf_p_ntype] imul ecx, [edi] ;# ecx = ntia = 2*ntype*type[ii0] mov [esp + nb333nf_ntia], ecx .nb333nf_threadloop: mov esi, [ebp + nb333nf_count] ;# pointer to sync counter mov eax, [esi].nb333nf_spinlock: mov ebx, eax ;# ebx=*count=nn0 add ebx, 1 ;# ebx=nn1=nn0+10 lock cmpxchg [esi], ebx ;# write nn1 to *counter, ;# if it hasnt changed. ;# or reread *counter to eax. pause ;# -> better p4 performance jnz .nb333nf_spinlock ;# if(nn1>nri) nn1=nri mov ecx, [esp + nb333nf_nri] mov edx, ecx sub ecx, ebx cmovle ebx, edx ;# if(nn1>nri) nn1=nri ;# Cleared the spinlock if we got here. ;# eax contains nn0, ebx contains nn1. mov [esp + nb333nf_n], eax mov [esp + nb333nf_nn1], ebx sub ebx, eax ;# calc number of outer lists mov esi, eax ;# copy n to esi jg .nb333nf_outerstart jmp .nb333nf_end.nb333nf_outerstart: ;# ebx contains number of outer iterations add ebx, [esp + nb333nf_nouter] mov [esp + nb333nf_nouter], ebx.nb333nf_outer: mov eax, [ebp + nb333nf_shift] ;# eax = pointer into shift[] mov ebx, [eax+esi*4] ;# ebx=shift[n] lea ebx, [ebx + ebx*2] ;# ebx=3*is mov [esp + nb333nf_is3],ebx ;# store is3 mov eax, [ebp + nb333nf_shiftvec] ;# eax = base of shiftvec[] movsd xmm0, [eax + ebx*8] movsd xmm1, [eax + ebx*8 + 8] movsd xmm2, [eax + ebx*8 + 16] mov ecx, [ebp + nb333nf_iinr] ;# ecx = pointer into iinr[] mov ebx, [ecx+esi*4] ;# ebx =ii movapd xmm3, xmm0 movapd xmm4, xmm1 movapd xmm5, xmm2 movapd xmm6, xmm0 movapd xmm7, xmm1 lea ebx, [ebx + ebx*2] ;# ebx = 3*ii=ii3 mov eax, [ebp + nb333nf_pos] ;# eax = base of pos[] mov [esp + nb333nf_ii3], ebx addsd xmm3, [eax + ebx*8] ;# ox addsd xmm4, [eax + ebx*8 + 8] ;# oy addsd xmm5, [eax + ebx*8 + 16] ;# oz addsd xmm6, [eax + ebx*8 + 24] ;# h1x addsd xmm7, [eax + ebx*8 + 32] ;# h1y shufpd xmm3, xmm3, 0 shufpd xmm4, xmm4, 0 shufpd xmm5, xmm5, 0 shufpd xmm6, xmm6, 0 shufpd xmm7, xmm7, 0 movapd [esp + nb333nf_ixO], xmm3 movapd [esp + nb333nf_iyO], xmm4 movapd [esp + nb333nf_izO], xmm5 movapd [esp + nb333nf_ixH1], xmm6 movapd [esp + nb333nf_iyH1], xmm7 movsd xmm6, xmm2 movsd xmm3, xmm0 movsd xmm4, xmm1 movsd xmm5, xmm2 addsd xmm6, [eax + ebx*8 + 40] ;# h1z addsd xmm0, [eax + ebx*8 + 48] ;# h2x addsd xmm1, [eax + ebx*8 + 56] ;# h2y addsd xmm2, [eax + ebx*8 + 64] ;# h2z addsd xmm3, [eax + ebx*8 + 72] ;# mx addsd xmm4, [eax + ebx*8 + 80] ;# my addsd xmm5, [eax + ebx*8 + 88] ;# mz shufpd xmm6, xmm6, 0 shufpd xmm0, xmm0, 0 shufpd xmm1, xmm1, 0 shufpd xmm2, xmm2, 0 shufpd xmm3, xmm3, 0 shufpd xmm4, xmm4, 0 shufpd xmm5, xmm5, 0 movapd [esp + nb333nf_izH1], xmm6 movapd [esp + nb333nf_ixH2], xmm0 movapd [esp + nb333nf_iyH2], xmm1 movapd [esp + nb333nf_izH2], xmm2 movapd [esp + nb333nf_ixM], xmm3 movapd [esp + nb333nf_iyM], xmm4 movapd [esp + nb333nf_izM], xmm5 ;# clear vctot xorpd xmm4, xmm4 movapd [esp + nb333nf_vctot], xmm4 movapd [esp + nb333nf_Vvdwtot], xmm4 mov eax, [ebp + nb333nf_jindex] mov ecx, [eax+esi*4] ;# jindex[n] mov edx, [eax + esi*4 + 4] ;# jindex[n+1] sub edx, ecx ;# number of innerloop atoms mov esi, [ebp + nb333nf_pos] mov edi, [ebp + nb333nf_faction] mov eax, [ebp + nb333nf_jjnr] shl ecx, 2 add eax, ecx mov [esp + nb333nf_innerjjnr], eax ;# pointer to jjnr[nj0] mov ecx, edx sub edx, 2 add ecx, [esp + nb333nf_ninner] mov [esp + nb333nf_ninner], ecx add edx, 0 mov [esp + nb333nf_innerk], edx ;# number of innerloop atoms jge .nb333nf_unroll_loop jmp .nb333nf_checksingle.nb333nf_unroll_loop: ;# twice unrolled innerloop here mov edx, [esp + nb333nf_innerjjnr] ;# pointer to jjnr[k] mov eax, [edx] mov ebx, [edx + 4] add dword ptr [esp + nb333nf_innerjjnr], 8 ;# advance pointer (unrolled 2) mov esi, [ebp + nb333nf_charge] ;# base of charge[] movlpd xmm3, [esi + eax*8] movhpd xmm3, [esi + ebx*8] movapd xmm4, xmm3 mulpd xmm3, [esp + nb333nf_iqM] mulpd xmm4, [esp + nb333nf_iqH] movapd [esp + nb333nf_qqM], xmm3 movapd [esp + nb333nf_qqH], xmm4 movd mm0, eax ;# use mmx registers as temp storage movd mm1, ebx mov esi, [ebp + nb333nf_type] mov eax, [esi + eax*4] mov ebx, [esi + ebx*4] mov esi, [ebp + nb333nf_vdwparam] shl eax, 1 shl ebx, 1 mov edi, [esp + nb333nf_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 + nb333nf_c6], xmm4 movapd [esp + nb333nf_c12], xmm6 mov esi, [ebp + nb333nf_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 + nb333nf_ixO] movapd xmm5, [esp + nb333nf_iyO] movapd xmm6, [esp + nb333nf_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 + nb333nf_ixH1] movapd xmm5, [esp + nb333nf_iyH1] movapd xmm6, [esp + nb333nf_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 + nb333nf_ixH2] movapd xmm4, [esp + nb333nf_iyH2] movapd xmm5, [esp + nb333nf_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 ;# move ixM-izM to xmm2-xmm4 movapd xmm3, [esp + nb333nf_iyM] movapd xmm4, [esp + nb333nf_izM] subpd xmm3, xmm1 subpd xmm4, xmm2 movapd xmm2, [esp + nb333nf_ixM] subpd xmm2, xmm0 ;# square it mulpd xmm2,xmm2 mulpd xmm3,xmm3 mulpd xmm4,xmm4 addpd xmm4, xmm3 addpd xmm4, xmm2 ;# rsqM in xmm4, rsqH2 in xmm5, rsqH1 in xmm6, rsqO in xmm7 ;# rsqO - put seed in xmm2 cvtpd2ps xmm2, xmm7 rsqrtps xmm2, xmm2 cvtps2pd xmm2, xmm2 movapd xmm3, xmm2 mulpd xmm2, xmm2 movapd xmm0, [esp + nb333nf_three] mulpd xmm2, xmm7 ;# rsq*lu*lu subpd xmm0, xmm2 ;# 30-rsq*lu*lu mulpd xmm0, xmm3 ;# lu*(3-rsq*lu*lu) mulpd xmm0, [esp + nb333nf_half] ;# iter1 ( new lu) movapd xmm2, xmm7 movapd xmm3, xmm0 mulpd xmm0, xmm0 ;# lu*lu mulpd xmm2, xmm0 ;# rsq*lu*lu movapd xmm0, [esp + nb333nf_three] subpd xmm0, xmm2 ;# 3-rsq*lu*lu mulpd xmm0, xmm3 ;# lu*( 3-rsq*lu*lu) mulpd xmm0, [esp + nb333nf_half] ;# rinv mulpd xmm7, xmm0 movapd [esp + nb333nf_rO], xmm7 ;# r in xmm7 ;# rsqH1 - seed in xmm2 cvtpd2ps xmm2, xmm6 rsqrtps xmm2, xmm2 cvtps2pd xmm2, xmm2 movapd xmm3, xmm2 mulpd xmm2, xmm2 movapd xmm0, [esp + nb333nf_three] mulpd xmm2, xmm6 ;# rsq*lu*lu subpd xmm0, xmm2 ;# 30-rsq*lu*lu mulpd xmm0, xmm3 ;# lu*(3-rsq*lu*lu) mulpd xmm0, [esp + nb333nf_half] ;# iter1 ( new lu) movapd xmm2, xmm6 movapd xmm3, xmm0 mulpd xmm0, xmm0 ;# lu*lu mulpd xmm2, xmm0 ;# rsq*lu*lu movapd xmm0, [esp + nb333nf_three] subpd xmm0, xmm2 ;# 3-rsq*lu*lu mulpd xmm0, xmm3 ;# lu*( 3-rsq*lu*lu) mulpd xmm0, [esp + nb333nf_half] ;# rinv mulpd xmm6, xmm0 movapd [esp + nb333nf_rH1], xmm6 ;# rH1 ;# rsqH2 - seed in xmm2 cvtpd2ps xmm2, xmm5 rsqrtps xmm2, xmm2 cvtps2pd xmm2, xmm2 movapd xmm3, xmm2 mulpd xmm2, xmm2 movapd xmm0, [esp + nb333nf_three] mulpd xmm2, xmm5 ;# rsq*lu*lu subpd xmm0, xmm2 ;# 30-rsq*lu*lu mulpd xmm0, xmm3 ;# lu*(3-rsq*lu*lu) mulpd xmm0, [esp + nb333nf_half] ;# iter1 ( new lu) movapd xmm2, xmm5 movapd xmm3, xmm0 mulpd xmm0, xmm0 ;# lu*lu mulpd xmm2, xmm0 ;# rsq*lu*lu movapd xmm0, [esp + nb333nf_three] subpd xmm0, xmm2 ;# 3-rsq*lu*lu mulpd xmm0, xmm3 ;# lu*( 3-rsq*lu*lu) mulpd xmm0, [esp + nb333nf_half] ;# rinv mulpd xmm5, xmm0 movapd [esp + nb333nf_rH2], xmm5 ;# r ;# rsqM - seed in xmm2 cvtpd2ps xmm2, xmm4 rsqrtps xmm2, xmm2 cvtps2pd xmm2, xmm2 movapd xmm3, xmm2 mulpd xmm2, xmm2 movapd xmm0, [esp + nb333nf_three] mulpd xmm2, xmm4 ;# rsq*lu*lu subpd xmm0, xmm2 ;# 30-rsq*lu*lu mulpd xmm0, xmm3 ;# lu*(3-rsq*lu*lu) mulpd xmm0, [esp + nb333nf_half] ;# iter1 ( new lu) movapd xmm2, xmm4 movapd xmm3, xmm0 mulpd xmm0, xmm0 ;# lu*lu mulpd xmm2, xmm0 ;# rsq*lu*lu movapd xmm0, [esp + nb333nf_three] subpd xmm0, xmm2 ;# 3-rsq*lu*lu mulpd xmm0, xmm3 ;# lu*( 3-rsq*lu*lu) mulpd xmm0, [esp + nb333nf_half] ;# rinv mulpd xmm4, xmm0 movapd [esp + nb333nf_rM], xmm4 ;# r ;# do O interactions ;# rO is still in xmm7 mulpd xmm7, [esp + nb333nf_tsc] cvttpd2pi mm6, xmm7 ;# mm6 = lu idx cvtpi2pd xmm6, mm6 subpd xmm7, xmm6 movapd xmm1, xmm7 ;# xmm1=eps movapd xmm2, xmm1 mulpd xmm2, xmm2 ;# xmm2=eps2 pslld mm6, 2 ;# idx *= 4 movd mm0, eax movd mm1, ebx mov esi, [ebp + nb333nf_VFtab] movd eax, mm6 psrlq mm6, 32 movd ebx, mm6 ;# indices in eax/ebx lea eax, [eax + eax*2] ;# idx *= 3 (total *=12 now) lea ebx, [ebx + ebx*2] ;# Dispersion movlpd xmm4, [esi + eax*8 + 32] ;# Y1 movlpd xmm3, [esi + ebx*8 + 32] ;# Y2 movhpd xmm4, [esi + eax*8 + 40] ;# Y1 F1 movhpd xmm3, [esi + ebx*8 + 40] ;# Y2 F2 movapd xmm5, xmm4 unpcklpd xmm4, xmm3 ;# Y1 Y2 unpckhpd xmm5, xmm3 ;# F1 F2 movlpd xmm6, [esi + eax*8 + 48] ;# G1 movlpd xmm3, [esi + ebx*8 + 48] ;# G2 movhpd xmm6, [esi + eax*8 + 56] ;# G1 H1 movhpd xmm3, [esi + ebx*8 + 56] ;# 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 movapd xmm4, [esp + nb333nf_c6] mulpd xmm5, xmm4 ;# Vvdw6 addpd xmm5, [esp + nb333nf_Vvdwtot] movapd [esp + nb333nf_Vvdwtot], xmm5 ;# Repulsion movlpd xmm4, [esi + eax*8 + 64] ;# Y1 movlpd xmm3, [esi + ebx*8 + 64] ;# Y2 movhpd xmm4, [esi + eax*8 + 72] ;# Y1 F1 movhpd xmm3, [esi + ebx*8 + 72] ;# Y2 F2 movapd xmm5, xmm4 unpcklpd xmm4, xmm3 ;# Y1 Y2 unpckhpd xmm5, xmm3 ;# F1 F2 movlpd xmm6, [esi + eax*8 + 80] ;# G1 movlpd xmm3, [esi + ebx*8 + 80] ;# G2 movhpd xmm6, [esi + eax*8 + 88] ;# G1 H1 movhpd xmm3, [esi + ebx*8 + 88] ;# 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 movapd xmm4, [esp + nb333nf_c12] mulpd xmm5, xmm4 ;# Vvdw12 addpd xmm5, [esp + nb333nf_Vvdwtot] movapd [esp + nb333nf_Vvdwtot], xmm5 ;# Done with O interactions - now H1! movapd xmm7, [esp + nb333nf_rH1] mulpd xmm7, [esp + nb333nf_tsc] cvttpd2pi mm6, xmm7 ;# mm6 = lu idx cvtpi2pd xmm6, mm6 subpd xmm7, xmm6 movapd xmm1, xmm7 ;# xmm1=eps movapd xmm2, xmm1 mulpd xmm2, xmm2 ;# xmm2=eps2 pslld mm6, 2 ;# idx *= 4 mov esi, [ebp + nb333nf_VFtab] movd eax, mm6 psrlq mm6, 32 movd ebx, mm6 ;# indices in eax/ebx lea eax, [eax + eax*2] ;# idx *= 3 (total *=12 now) lea ebx, [ebx + ebx*2] movlpd xmm4, [esi + eax*8] ;# Y1 movlpd xmm3, [esi + ebx*8] ;# Y2 movhpd xmm4, [esi + eax*8 + 8] ;# Y1 F1 mo⌨️ 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?