nb_kernel310_ia32_sse.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 1,851 行 · 第 1/4 页
S
1,851 行
mulps xmm5, xmm1 ;# xmm5=eps*Fp addps xmm5, xmm4 ;# xmm5=VV mulps xmm5, xmm3 ;# vcoul=qq*VV mulps xmm3, xmm7 ;# fijC=FF*qq ;# L-J movaps xmm4, xmm0 mulps xmm4, xmm0 ;# xmm4=rinvsq ;# at this point mm5 contains vcoul and mm3 fijC ;# increment vcoul - then we can get rid of mm5 ;# update vctot addps xmm5, [esp + nb310_vctot] movaps xmm6, xmm4 mulps xmm6, xmm4 movaps [esp + nb310_vctot], xmm5 mulps xmm6, xmm4 ;# xmm6=rinvsix movaps xmm4, xmm6 mulps xmm4, xmm4 ;# xmm4=rinvtwelve mulps xmm6, [esp + nb310_c6] mulps xmm4, [esp + nb310_c12] movaps xmm7, [esp + nb310_Vvdwtot] addps xmm7, xmm4 mulps xmm4, [esp + nb310_twelve] subps xmm7, xmm6 mulps xmm3, [esp + nb310_tsc] mulps xmm6, [esp + nb310_six] movaps [esp + nb310_Vvdwtot], xmm7 subps xmm4, xmm6 mulps xmm4, xmm0 subps xmm4, xmm3 mulps xmm4, xmm0 movaps xmm0, [esp + nb310_dx] movaps xmm1, [esp + nb310_dy] movaps xmm2, [esp + nb310_dz] movd eax, mm0 movd ebx, mm1 movd ecx, mm2 movd edx, mm3 mov edi, [ebp + nb310_faction] mulps xmm0, xmm4 mulps xmm1, xmm4 mulps xmm2, xmm4 ;# xmm0-xmm2 contains tx-tz (partial force) ;# now update f_i movaps xmm3, [esp + nb310_fix] movaps xmm4, [esp + nb310_fiy] movaps xmm5, [esp + nb310_fiz] addps xmm3, xmm0 addps xmm4, xmm1 addps xmm5, xmm2 movaps [esp + nb310_fix], xmm3 movaps [esp + nb310_fiy], xmm4 movaps [esp + nb310_fiz], xmm5 ;# the fj's - start by accumulating x & y forces from memory movlps xmm4, [edi + eax*4] movlps xmm6, [edi + ecx*4] movhps xmm4, [edi + ebx*4] movhps xmm6, [edi + edx*4] movaps xmm3, xmm4 shufps xmm3, xmm6, 136 ;# constant 10001000 shufps xmm4, xmm6, 221 ;# constant 11011101 ;# now xmm3-xmm5 contains fjx, fjy, fjz subps xmm3, xmm0 subps xmm4, xmm1 ;# unpack them back so we can store them - first x & y in xmm3/xmm4 movaps xmm6, xmm3 unpcklps xmm6, xmm4 unpckhps xmm3, xmm4 ;# xmm6(l)=x & y for j1, (h) for j2 ;# xmm3(l)=x & y for j3, (h) for j4 movlps [edi + eax*4], xmm6 movlps [edi + ecx*4], xmm3 movhps [edi + ebx*4], xmm6 movhps [edi + edx*4], xmm3 ;# and the z forces movss xmm4, [edi + eax*4 + 8] movss xmm5, [edi + ebx*4 + 8] movss xmm6, [edi + ecx*4 + 8] movss xmm7, [edi + edx*4 + 8] subss xmm4, xmm2 shufps xmm2, xmm2, 229 ;# constant 11100101 subss xmm5, xmm2 shufps xmm2, xmm2, 234 ;# constant 11101010 subss xmm6, xmm2 shufps xmm2, xmm2, 255 ;# constant 11111111 subss xmm7, xmm2 movss [edi + eax*4 + 8], xmm4 movss [edi + ebx*4 + 8], xmm5 movss [edi + ecx*4 + 8], xmm6 movss [edi + edx*4 + 8], xmm7 ;# should we do one more iteration? sub dword ptr [esp + nb310_innerk], 4 jl .nb310_finish_inner jmp .nb310_unroll_loop.nb310_finish_inner: ;# check if at least two particles remain add dword ptr [esp + nb310_innerk], 4 mov edx, [esp + nb310_innerk] and edx, 2 jnz .nb310_dopair jmp .nb310_checksingle.nb310_dopair: mov esi, [ebp + nb310_charge] mov ecx, [esp + nb310_innerjjnr] mov eax, [ecx] mov ebx, [ecx + 4] add dword ptr [esp + nb310_innerjjnr], 8 xorps xmm7, xmm7 movss xmm3, [esi + eax*4] movss xmm6, [esi + ebx*4] shufps xmm3, xmm6, 0 shufps xmm3, xmm3, 8 ;# constant 00001000 ;# xmm3(0,1) has the charges mulps xmm3, [esp + nb310_iq] movlhps xmm3, xmm7 movaps [esp + nb310_qq], xmm3 mov esi, [ebp + nb310_type] mov ecx, eax mov edx, ebx mov ecx, [esi + ecx*4] mov edx, [esi + edx*4] mov esi, [ebp + nb310_vdwparam] shl ecx, 1 shl edx, 1 mov edi, [esp + nb310_ntia] add ecx, edi add edx, edi movlps xmm6, [esi + ecx*4] movhps xmm6, [esi + edx*4] mov edi, [ebp + nb310_pos] movaps xmm4, xmm6 shufps xmm4, xmm4, 8 ;# constant 00001000 shufps xmm6, xmm6, 13 ;# constant 00001101 movlhps xmm4, xmm7 movlhps xmm6, xmm7 movaps [esp + nb310_c6], xmm4 movaps [esp + nb310_c12], xmm6 lea eax, [eax + eax*2] lea ebx, [ebx + ebx*2] ;# move coordinates to xmm0-xmm2 movlps xmm1, [edi + eax*4] movss xmm2, [edi + eax*4 + 8] movhps xmm1, [edi + ebx*4] movss xmm0, [edi + ebx*4 + 8] movlhps xmm3, xmm7 shufps xmm2, xmm0, 0 movaps xmm0, xmm1 shufps xmm2, xmm2, 136 ;# constant 10001000 shufps xmm0, xmm0, 136 ;# constant 10001000 shufps xmm1, xmm1, 221 ;# constant 11011101 mov edi, [ebp + nb310_faction] ;# move ix-iz to xmm4-xmm6 xorps xmm7, xmm7 movaps xmm4, [esp + nb310_ix] movaps xmm5, [esp + nb310_iy] movaps xmm6, [esp + nb310_iz] ;# calc dr subps xmm4, xmm0 subps xmm5, xmm1 subps xmm6, xmm2 ;# store dr movaps [esp + nb310_dx], xmm4 movaps [esp + nb310_dy], xmm5 movaps [esp + nb310_dz], xmm6 ;# square it mulps xmm4,xmm4 mulps xmm5,xmm5 mulps xmm6,xmm6 addps xmm4, xmm5 addps xmm4, xmm6 ;# rsq in xmm4 rsqrtps xmm5, xmm4 ;# lookup seed in xmm5 movaps xmm2, xmm5 mulps xmm5, xmm5 movaps xmm1, [esp + nb310_three] mulps xmm5, xmm4 ;# rsq*lu*lu movaps xmm0, [esp + nb310_half] subps xmm1, xmm5 ;# constant 30-rsq*lu*lu mulps xmm1, xmm2 mulps xmm0, xmm1 ;# xmm0=rinv mulps xmm4, xmm0 ;# xmm4=r mulps xmm4, [esp + nb310_tsc] cvttps2pi mm6, xmm4 ;# mm6 contain lu indices cvtpi2ps xmm6, mm6 subps xmm4, xmm6 movaps xmm1, xmm4 ;# xmm1=eps movaps xmm2, xmm1 mulps xmm2, xmm2 ;# xmm2=eps2 pslld mm6, 2 mov esi, [ebp + nb310_VFtab] movd ecx, mm6 psrlq mm6, 32 movd edx, mm6 movlps xmm5, [esi + ecx*4] movhps xmm5, [esi + edx*4] ;# got half coulomb table movaps xmm4, xmm5 shufps xmm4, xmm4, 136 ;# constant 10001000 shufps xmm5, xmm7, 221 ;# constant 11011101 movlps xmm7, [esi + ecx*4 + 8] movhps xmm7, [esi + edx*4 + 8] movaps xmm6, xmm7 shufps xmm6, xmm6, 136 ;# constant 10001000 shufps xmm7, xmm7, 221 ;# constant 11011101 ;# 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 + nb310_two] ;# two*Heps2 movaps xmm3, [esp + nb310_qq] addps xmm7, xmm6 addps xmm7, xmm5 ;# xmm7=FF mulps xmm5, xmm1 ;# xmm5=eps*Fp addps xmm5, xmm4 ;# xmm5=VV mulps xmm5, xmm3 ;# vcoul=qq*VV mulps xmm3, xmm7 ;# fijC=FF*qq ;# L-J movaps xmm4, xmm0 mulps xmm4, xmm0 ;# xmm4=rinvsq ;# at this point mm5 contains vcoul and mm3 fijC ;# increment vcoul - then we can get rid of mm5 ;# update vctot addps xmm5, [esp + nb310_vctot] movaps xmm6, xmm4 mulps xmm6, xmm4 movaps [esp + nb310_vctot], xmm5 mulps xmm6, xmm4 ;# xmm6=rinvsix movaps xmm4, xmm6 mulps xmm4, xmm4 ;# xmm4=rinvtwelve mulps xmm6, [esp + nb310_c6] mulps xmm4, [esp + nb310_c12] movaps xmm7, [esp + nb310_Vvdwtot] addps xmm7, xmm4 mulps xmm4, [esp + nb310_twelve] subps xmm7, xmm6 mulps xmm3, [esp + nb310_tsc] mulps xmm6, [esp + nb310_six] movaps [esp + nb310_Vvdwtot], xmm7 subps xmm4, xmm6 mulps xmm4, xmm0 subps xmm4, xmm3 mulps xmm4, xmm0 movaps xmm0, [esp + nb310_dx] movaps xmm1, [esp + nb310_dy] movaps xmm2, [esp + nb310_dz] mulps xmm0, xmm4 mulps xmm1, xmm4 mulps xmm2, xmm4 ;# xmm0-xmm2 contains tx-tz (partial force) ;# now update f_i movaps xmm3, [esp + nb310_fix] movaps xmm4, [esp + nb310_fiy] movaps xmm5, [esp + nb310_fiz] addps xmm3, xmm0 addps xmm4, xmm1 addps xmm5, xmm2 movaps [esp + nb310_fix], xmm3 movaps [esp + nb310_fiy], xmm4 movaps [esp + nb310_fiz], xmm5 ;# update the fj's movss xmm3, [edi + eax*4] movss xmm4, [edi + eax*4 + 4] movss xmm5, [edi + eax*4 + 8] subss xmm3, xmm0 subss xmm4, xmm1 subss xmm5, xmm2 movss [edi + eax*4], xmm3 movss [edi + eax*4 + 4], xmm4 movss [edi + eax*4 + 8], xmm5 shufps xmm0, xmm0, 225 ;# constant 11100001 shufps xmm1, xmm1, 225 ;# constant 11100001 shufps xmm2, xmm2, 225 ;# constant 11100001 movss xmm3, [edi + ebx*4] movss xmm4, [edi + ebx*4 + 4] movss xmm5, [edi + ebx*4 + 8] subss xmm3, xmm0 subss xmm4, xmm1 subss xmm5, xmm2 movss [edi + ebx*4], xmm3 movss [edi + ebx*4 + 4], xmm4 movss [edi + ebx*4 + 8], xmm5 .nb310_checksingle: mov edx, [esp + nb310_innerk] and edx, 1 jnz .nb310_dosingle jmp .nb310_updateouterdata.nb310_dosingle: mov esi, [ebp + nb310_charge] mov edi, [ebp + nb310_pos] mov ecx, [esp + nb310_innerjjnr] mov eax, [ecx] xorps xmm6, xmm6 movss xmm6, [esi + eax*4] ;# xmm6(0) has the charge mulps xmm6, [esp + nb310_iq] movaps [esp + nb310_qq], xmm6 mov esi, [ebp + nb310_type] mov ecx, eax mov ecx, [esi + ecx*4] mov esi, [ebp + nb310_vdwparam] shl ecx, 1 add ecx, [esp + nb310_ntia] movlps xmm6, [esi + ecx*4] movaps xmm4, xmm6 shufps xmm4, xmm4, 252 ;# constant 11111100 shufps xmm6, xmm6, 253 ;# constant 11111101 movaps [esp + nb310_c6], xmm4 movaps [esp + nb310_c12], xmm6 lea eax, [eax + eax*2] ;# move coordinates to xmm0-xmm2 movss xmm0, [edi + eax*4] movss xmm1, [edi + eax*4 + 4] movss xmm2, [edi + eax*4 + 8] movaps xmm4, [esp + nb310_ix] movaps xmm5, [esp + nb310_iy] movaps xmm6, [esp + nb310_iz] ;# calc dr subps xmm4, xmm0 subps xmm5, xmm1 subps xmm6, xmm2 ;# store dr movaps [esp + nb310_dx], xmm4 movaps [esp + nb310_dy], xmm5 movaps [esp + nb310_dz], xmm6 ;# square it mulps xmm4,xmm4 mulps xmm5,xmm5 mulps xmm6,xmm6 addps xmm4, xmm5 addps xmm4, xmm6 ;# rsq in xmm4 rsqrtps xmm5, xmm4 ;# lookup seed in xmm5 movaps xmm2, xmm5 mulps xmm5, xmm5 movaps xmm1, [esp + nb310_three] mulps xmm5, xmm4 ;# rsq*lu*lu movaps xmm0, [esp + nb310_half] subps xmm1, xmm5 ;# constant 30-rsq*lu*lu mulps xmm1, xmm2 mulps xmm0, xmm1 ;# xmm0=rinv mulps xmm4, xmm0 ;# xmm4=r mulps xmm4, [esp + nb310_tsc] cvttps2pi mm6, xmm4 ;# mm6 contain lu indices cvtpi2ps xmm6, mm6 subps xmm4, xmm6 movaps xmm1, xmm4 ;# xmm1=eps movaps xmm2, xmm1 mulps xmm2, xmm2 ;# xmm2=eps2 pslld mm6, 2 mov esi, [ebp + nb310_VFtab] movd ebx, mm6 movlps xmm4, [esi + ebx*4] movlps xmm6, [esi + ebx*4 + 8] movaps xmm5, xmm4 movaps xmm7, xmm6 shufps xmm5, xmm5, 1 shufps xmm7, xmm7, 1 ;# 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 + nb310_two] ;# two*Heps2 movaps xmm3, [esp + nb310_qq] addps xmm7, xmm6 addps xmm7, xmm5 ;# xmm7=FF mulps xmm5, xmm1 ;# xmm5=eps*Fp addps xmm5, xmm4 ;# xmm5=VV mulps xmm5, xmm3 ;# vcoul=qq*VV mulps xmm3, xmm7 ;# fijC=FF*qq ;# L-J movaps xmm4, xmm0 mulps xmm4, xmm0 ;# xmm4=rinvsq ;# at this point mm5 contains vcoul and mm3 fijC ;# increment vcoul - then we can get rid of mm5 ;# update vctot addss xmm5, [esp + nb310_vctot] movaps xmm6, xmm4 mulps xmm6, xmm4 movss [esp + nb310_vctot], xmm5 mulps xmm6, xmm4 ;# xmm6=rinvsix movaps xmm4, xmm6 mulps xmm4, xmm4 ;# xmm4=rinvtwelve mulps xmm6, [esp + nb310_c6] mulps xmm4, [esp + nb310_c12] movss xmm7, [esp + nb310_Vvdwtot] addps xmm7, xmm4 mulps xmm4, [esp + nb310_twelve] subps xmm7, xmm6 mulps xmm3, [esp + nb310_tsc] mulps xmm6, [esp + nb310_six] movss [esp + nb310_Vvdwtot], xmm7 subps xmm4, xmm6 mulps xmm4, xmm0 subps xmm4, xmm3 mulps xmm4, xmm0 movaps xmm0, [esp + nb310_dx] movaps xmm1, [esp + nb310_dy] movaps xmm2, [esp + nb310_dz] mov edi, [ebp + nb310_faction]⌨️ 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?