nb_kernel311_ia32_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
;# calc dr subps xmm4, xmm0 subps xmm5, xmm1 subps xmm6, xmm2 ;# store dr movaps [esp + nb311_dxO], xmm4 movaps [esp + nb311_dyO], xmm5 movaps [esp + nb311_dzO], xmm6 ;# square it mulps xmm4,xmm4 mulps xmm5,xmm5 mulps xmm6,xmm6 addps xmm4, xmm5 addps xmm4, xmm6 movaps xmm7, xmm4 ;# rsqO in xmm7 ;# move ixH1-izH1 to xmm4-xmm6 movaps xmm4, [esp + nb311_ixH1] movaps xmm5, [esp + nb311_iyH1] movaps xmm6, [esp + nb311_izH1] ;# calc dr subps xmm4, xmm0 subps xmm5, xmm1 subps xmm6, xmm2 ;# store dr movaps [esp + nb311_dxH1], xmm4 movaps [esp + nb311_dyH1], xmm5 movaps [esp + nb311_dzH1], xmm6 ;# square it mulps xmm4,xmm4 mulps xmm5,xmm5 mulps xmm6,xmm6 addps xmm6, xmm5 addps xmm6, xmm4 ;# rsqH1 in xmm6 ;# move ixH2-izH2 to xmm3-xmm5 movaps xmm3, [esp + nb311_ixH2] movaps xmm4, [esp + nb311_iyH2] movaps xmm5, [esp + nb311_izH2] ;# calc dr subps xmm3, xmm0 subps xmm4, xmm1 subps xmm5, xmm2 ;# store dr movaps [esp + nb311_dxH2], xmm3 movaps [esp + nb311_dyH2], xmm4 movaps [esp + nb311_dzH2], xmm5 ;# square it mulps xmm3,xmm3 mulps xmm4,xmm4 mulps xmm5,xmm5 addps xmm5, xmm4 addps xmm5, xmm3 ;# rsqH2 in xmm5, rsqH1 in xmm6, rsqO in xmm7 ;# start with rsqO - seed to xmm2 rsqrtps xmm2, xmm7 movaps xmm3, xmm2 mulps xmm2, xmm2 movaps xmm4, [esp + nb311_three] mulps xmm2, xmm7 ;# rsq*lu*lu subps xmm4, xmm2 ;# constant 30-rsq*lu*lu mulps xmm4, xmm3 ;# lu*(3-rsq*lu*lu) mulps xmm4, [esp + nb311_half] movaps [esp + nb311_rinvO], xmm4 ;# rinvO in xmm4 mulps xmm7, xmm4 movaps [esp + nb311_rO], xmm7 ;# rsqH1 - seed in xmm2 rsqrtps xmm2, xmm6 movaps xmm3, xmm2 mulps xmm2, xmm2 movaps xmm4, [esp + nb311_three] mulps xmm2, xmm6 ;# rsq*lu*lu subps xmm4, xmm2 ;# constant 30-rsq*lu*lu mulps xmm4, xmm3 ;# lu*(3-rsq*lu*lu) mulps xmm4, [esp + nb311_half] movaps [esp + nb311_rinvH1], xmm4 ;# rinvH1 in xmm4 mulps xmm6, xmm4 movaps [esp + nb311_rH1], xmm6 ;# rsqH2 - seed to xmm2 rsqrtps xmm2, xmm5 movaps xmm3, xmm2 mulps xmm2, xmm2 movaps xmm4, [esp + nb311_three] mulps xmm2, xmm5 ;# rsq*lu*lu subps xmm4, xmm2 ;# constant 30-rsq*lu*lu mulps xmm4, xmm3 ;# lu*(3-rsq*lu*lu) mulps xmm4, [esp + nb311_half] movaps [esp + nb311_rinvH2], xmm4 ;# rinvH2 in xmm4 mulps xmm5, xmm4 movaps [esp + nb311_rH2], xmm5 ;# do O interactions ;# rO is still in xmm7 mulps xmm7, [esp + nb311_tsc] movhlps xmm4, xmm7 cvttps2pi mm6, xmm7 cvttps2pi mm7, xmm4 ;# mm6/mm7 contain lu indices cvtpi2ps xmm3, mm6 cvtpi2ps xmm4, mm7 movlhps xmm3, xmm4 subps xmm7, xmm3 movaps xmm1, xmm7 ;# xmm1=eps movaps xmm2, xmm1 mulps xmm2, xmm2 ;# xmm2=eps2 pslld mm6, 2 pslld mm7, 2 movd mm0, eax movd mm1, ebx movd mm2, ecx movd mm3, edx mov esi, [ebp + nb311_VFtab] movd eax, mm6 psrlq mm6, 32 movd ecx, mm7 psrlq mm7, 32 movd ebx, mm6 movd edx, mm7 movlps xmm5, [esi + eax*4] movlps xmm7, [esi + ecx*4] movhps xmm5, [esi + ebx*4] movhps xmm7, [esi + edx*4] ;# got half coulomb table movaps xmm4, xmm5 shufps xmm4, xmm7, 136 ;# constant 10001000 shufps xmm5, xmm7, 221 ;# constant 11011101 movlps xmm7, [esi + eax*4 + 8] movlps xmm3, [esi + ecx*4 + 8] movhps xmm7, [esi + ebx*4 + 8] movhps xmm3, [esi + edx*4 + 8] ;# other half of coulomb table movaps xmm6, xmm7 shufps xmm6, xmm3, 136 ;# constant 10001000 shufps xmm7, xmm3, 221 ;# constant 11011101 ;# coulomb table ready, in xmm4-xmm7 mulps xmm6, xmm1 ;# xmm6=Geps mulps xmm7, xmm2 ;# xmm7=Heps2 addps xmm5, xmm6 addps xmm5, xmm7 ;# xmm5=Fp mulps xmm7, [esp + nb311_two] ;# two*Heps2 movaps xmm0, [esp + nb311_qqO] addps xmm7, xmm6 addps xmm7, xmm5 ;# xmm7=FF mulps xmm5, xmm1 ;# xmm5=eps*Fp addps xmm5, xmm4 ;# xmm5=VV mulps xmm5, xmm0 ;# vcoul=qq*VV mulps xmm0, xmm7 ;# fijC=FF*qq ;# do nontable L-J movaps xmm2, [esp + nb311_rinvO] mulps xmm2, xmm2 ;# at this point mm5 contains vcoul and xmm0 fijC ;# increment vcoul - then we can get rid of mm5 addps xmm5, [esp + nb311_vctot] movaps [esp + nb311_vctot], xmm5 movaps xmm1, xmm2 mulps xmm1, xmm1 mulps xmm1, xmm2 ;# xmm1=rinvsix movaps xmm4, xmm1 mulps xmm4, xmm4 ;# xmm4=rinvtwelve mulps xmm1, [esp + nb311_c6] mulps xmm4, [esp + nb311_c12] movaps xmm3, xmm4 subps xmm3, xmm1 ;# xmm3=Vvdw12-Vvdw6 mulps xmm1, [esp + nb311_six] mulps xmm4, [esp + nb311_twelve] subps xmm4, xmm1 addps xmm3, [esp + nb311_Vvdwtot] mulps xmm4, [esp + nb311_rinvO] mulps xmm0, [esp + nb311_tsc] subps xmm4, xmm0 movaps [esp + nb311_Vvdwtot], xmm3 mulps xmm4, [esp + nb311_rinvO] movaps xmm0, [esp + nb311_dxO] movaps xmm1, [esp + nb311_dyO] movaps xmm2, [esp + nb311_dzO] mulps xmm0, xmm4 mulps xmm1, xmm4 mulps xmm2, xmm4 ;# tx in xmm0-xmm2 ;# update O forces movaps xmm3, [esp + nb311_fixO] movaps xmm4, [esp + nb311_fiyO] movaps xmm7, [esp + nb311_fizO] addps xmm3, xmm0 addps xmm4, xmm1 addps xmm7, xmm2 movaps [esp + nb311_fixO], xmm3 movaps [esp + nb311_fiyO], xmm4 movaps [esp + nb311_fizO], xmm7 ;# update j forces with water O movaps [esp + nb311_fjx], xmm0 movaps [esp + nb311_fjy], xmm1 movaps [esp + nb311_fjz], xmm2 ;# Done with O interactions - now H1! movaps xmm7, [esp + nb311_rH1] mulps xmm7, [esp + nb311_tsc] movhlps xmm4, xmm7 cvttps2pi mm6, xmm7 cvttps2pi mm7, xmm4 ;# mm6/mm7 contain lu indices cvtpi2ps xmm3, mm6 cvtpi2ps xmm4, mm7 movlhps xmm3, xmm4 subps xmm7, xmm3 movaps xmm1, xmm7 ;# xmm1=eps movaps xmm2, xmm1 mulps xmm2, xmm2 ;# xmm2=eps2 pslld mm6, 2 pslld mm7, 2 movd eax, mm6 psrlq mm6, 32 movd ecx, mm7 psrlq mm7, 32 movd ebx, mm6 movd edx, mm7 movlps xmm5, [esi + eax*4] movlps xmm7, [esi + ecx*4] movhps xmm5, [esi + ebx*4] movhps xmm7, [esi + edx*4] ;# got half coulomb table movaps xmm4, xmm5 shufps xmm4, xmm7, 136 ;# constant 10001000 shufps xmm5, xmm7, 221 ;# constant 11011101 movlps xmm7, [esi + eax*4 + 8] movlps xmm3, [esi + ecx*4 + 8] movhps xmm7, [esi + ebx*4 + 8] movhps xmm3, [esi + edx*4 + 8] ;# other half of coulomb table movaps xmm6, xmm7 shufps xmm6, xmm3, 136 ;# constant 10001000 shufps xmm7, xmm3, 221 ;# constant 11011101 ;# coulomb table ready, in xmm4-xmm7 mulps xmm6, xmm1 ;# xmm6=Geps mulps xmm7, xmm2 ;# xmm7=Heps2 addps xmm5, xmm6 addps xmm5, xmm7 ;# xmm5=Fp mulps xmm7, [esp + nb311_two] ;# two*Heps2 movaps xmm0, [esp + nb311_qqH] addps xmm7, xmm6 addps xmm7, xmm5 ;# xmm7=FF mulps xmm5, xmm1 ;# xmm5=eps*Fp addps xmm5, xmm4 ;# xmm5=VV mulps xmm5, xmm0 ;# vcoul=qq*VV mulps xmm7, xmm0 ;# fijC=FF*qq ;# at this point mm5 contains vcoul and xmm7 fijC ;# increment vcoul xorps xmm4, xmm4 addps xmm5, [esp + nb311_vctot] mulps xmm7, [esp + nb311_rinvH1] movaps [esp + nb311_vctot], xmm5 mulps xmm7, [esp + nb311_tsc] subps xmm4, xmm7 movaps xmm0, [esp + nb311_dxH1] movaps xmm1, [esp + nb311_dyH1] movaps xmm2, [esp + nb311_dzH1] mulps xmm0, xmm4 mulps xmm1, xmm4 mulps xmm2, xmm4 ;# update H1 forces movaps xmm3, [esp + nb311_fixH1] movaps xmm4, [esp + nb311_fiyH1] movaps xmm7, [esp + nb311_fizH1] addps xmm3, xmm0 addps xmm4, xmm1 addps xmm7, xmm2 movaps [esp + nb311_fixH1], xmm3 movaps [esp + nb311_fiyH1], xmm4 movaps [esp + nb311_fizH1], xmm7 ;# update j forces with water H1 addps xmm0, [esp + nb311_fjx] addps xmm1, [esp + nb311_fjy] addps xmm2, [esp + nb311_fjz] movaps [esp + nb311_fjx], xmm0 movaps [esp + nb311_fjy], xmm1 movaps [esp + nb311_fjz], xmm2 ;# Done with H1, finally we do H2 interactions movaps xmm7, [esp + nb311_rH2] mulps xmm7, [esp + nb311_tsc] movhlps xmm4, xmm7 cvttps2pi mm6, xmm7 cvttps2pi mm7, xmm4 ;# mm6/mm7 contain lu indices cvtpi2ps xmm3, mm6 cvtpi2ps xmm4, mm7 movlhps xmm3, xmm4 subps xmm7, xmm3 movaps xmm1, xmm7 ;# xmm1=eps movaps xmm2, xmm1 mulps xmm2, xmm2 ;# xmm2=eps2 pslld mm6, 2 pslld mm7, 2 movd eax, mm6 psrlq mm6, 32 movd ecx, mm7 psrlq mm7, 32 movd ebx, mm6 movd edx, mm7 movlps xmm5, [esi + eax*4] movlps xmm7, [esi + ecx*4] movhps xmm5, [esi + ebx*4] movhps xmm7, [esi + edx*4] ;# got half coulomb table movaps xmm4, xmm5 shufps xmm4, xmm7, 136 ;# constant 10001000 shufps xmm5, xmm7, 221 ;# constant 11011101 movlps xmm7, [esi + eax*4 + 8] movlps xmm3, [esi + ecx*4 + 8] movhps xmm7, [esi + ebx*4 + 8] movhps xmm3, [esi + edx*4 + 8] ;# other half of coulomb table movaps xmm6, xmm7 shufps xmm6, xmm3, 136 ;# constant 10001000 shufps xmm7, xmm3, 221 ;# constant 11011101 ;# coulomb table ready, in xmm4-xmm7 mulps xmm6, xmm1 ;# xmm6=Geps mulps xmm7, xmm2 ;# xmm7=Heps2 addps xmm5, xmm6 addps xmm5, xmm7 ;# xmm5=Fp mulps xmm7, [esp + nb311_two] ;# two*Heps2 movaps xmm0, [esp + nb311_qqH] addps xmm7, xmm6 addps xmm7, xmm5 ;# xmm7=FF mulps xmm5, xmm1 ;# xmm5=eps*Fp addps xmm5, xmm4 ;# xmm5=VV mulps xmm5, xmm0 ;# vcoul=qq*VV mulps xmm7, xmm0 ;# fijC=FF*qq ;# at this point mm5 contains vcoul and xmm0 fijC ;# increment vcoul xorps xmm4, xmm4 addps xmm5, [esp + nb311_vctot] mulps xmm7, [esp + nb311_rinvH2] movaps [esp + nb311_vctot], xmm5 mulps xmm7, [esp + nb311_tsc] subps xmm4, xmm7 movaps xmm0, [esp + nb311_dxH2] movaps xmm1, [esp + nb311_dyH2] movaps xmm2, [esp + nb311_dzH2] mulps xmm0, xmm4 mulps xmm1, xmm4 mulps xmm2, xmm4 movd eax, mm0 movd ebx, mm1 movd ecx, mm2 movd edx, mm3 ;# update H2 forces movaps xmm3, [esp + nb311_fixH2] movaps xmm4, [esp + nb311_fiyH2] movaps xmm7, [esp + nb311_fizH2] addps xmm3, xmm0 addps xmm4, xmm1 addps xmm7, xmm2 movaps [esp + nb311_fixH2], xmm3 movaps [esp + nb311_fiyH2], xmm4 movaps [esp + nb311_fizH2], xmm7 mov edi, [ebp + nb311_faction] ;# update j forces addps xmm0, [esp + nb311_fjx] addps xmm1, [esp + nb311_fjy] addps xmm2, [esp + nb311_fjz] movlps xmm4, [edi + eax*4] movlps xmm7, [edi + ecx*4] movhps xmm4, [edi + ebx*4] movhps xmm7, [edi + edx*4] movaps xmm3, xmm4 shufps xmm3, xmm7, 136 ;# constant 10001000 shufps xmm4, xmm7, 221 ;# constant 11011101 ;# xmm3 has fjx, xmm4 has fjy subps xmm3, xmm0 subps xmm4, xmm1 ;# unpack them back for storing movaps xmm7, xmm3 unpcklps xmm7, xmm4 unpckhps xmm3, xmm4 movlps [edi + eax*4], xmm7 movlps [edi + ecx*4], xmm3 movhps [edi + ebx*4], xmm7 movhps [edi + edx*4], xmm3 ;# finally z forces movss xmm0, [edi + eax*4 + 8] movss xmm1, [edi + ebx*4 + 8] movss xmm3, [edi + ecx*4 + 8] movss xmm4, [edi + edx*4 + 8] subss xmm0, xmm2 shufps xmm2, xmm2, 229 ;# constant 11100101 subss xmm1, xmm2 shufps xmm2, xmm2, 234 ;# constant 11101010 subss xmm3, xmm2 shufps xmm2, xmm2, 255 ;# constant 11111111 subss xmm4, xmm2 movss [edi + eax*4 + 8], xmm0 movss [edi + ebx*4 + 8], xmm1 movss [edi + ecx*4 + 8], xmm3 movss [edi + edx*4 + 8], xmm4 ;# should we do one more iteration? sub dword ptr [esp + nb311_innerk], 4 jl .nb311_odd_inner jmp .nb311_unroll_loop.nb311_odd_inner: add dword ptr [esp + nb311_innerk], 4 jnz .nb311_odd_loop jmp .nb311_updateouterdata.nb311_odd_loop: mov edx, [esp + nb311_innerjjnr] ;# pointer to jjnr[k] ⌨️ 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?