nb_kernel113_ia32_sse.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 1,869 行 · 第 1/4 页
S
1,869 行
addps %xmm4,%xmm6 ## rsqH1 in xmm6 ## move ixH2-izH2 to xmm3-xmm5 movaps nb113_ixH2(%esp),%xmm3 movaps nb113_iyH2(%esp),%xmm4 movaps nb113_izH2(%esp),%xmm5 ## calc dr subps %xmm0,%xmm3 subps %xmm1,%xmm4 subps %xmm2,%xmm5 ## store dr movaps %xmm3,nb113_dxH2(%esp) movaps %xmm4,nb113_dyH2(%esp) movaps %xmm5,nb113_dzH2(%esp) ## square it mulps %xmm3,%xmm3 mulps %xmm4,%xmm4 mulps %xmm5,%xmm5 addps %xmm4,%xmm5 addps %xmm3,%xmm5 ## move ixM-izM to xmm2-xmm4 movaps nb113_iyM(%esp),%xmm3 movaps nb113_izM(%esp),%xmm4 subps %xmm1,%xmm3 subps %xmm2,%xmm4 movaps nb113_ixM(%esp),%xmm2 subps %xmm0,%xmm2 ## store dr movaps %xmm2,nb113_dxM(%esp) movaps %xmm3,nb113_dyM(%esp) movaps %xmm4,nb113_dzM(%esp) ## square it mulps %xmm2,%xmm2 mulps %xmm3,%xmm3 mulps %xmm4,%xmm4 addps %xmm3,%xmm4 addps %xmm2,%xmm4 ## rsqM in xmm4, rsqH2 in xmm5, rsqH1 in xmm6, rsqO in xmm7 ## rsqH1 - seed in xmm2 rsqrtps %xmm6,%xmm2 movaps %xmm2,%xmm3 mulps %xmm2,%xmm2 movaps nb113_three(%esp),%xmm0 mulps %xmm6,%xmm2 ## rsq*lu*lu subps %xmm2,%xmm0 ## constant 30-rsq*lu*lu mulps %xmm3,%xmm0 ## lu*(3-rsq*lu*lu) mulps nb113_half(%esp),%xmm0 movaps %xmm0,nb113_rinvH1(%esp) ## rinvH1 ## rsqH2 - seed to xmm2 rsqrtps %xmm5,%xmm2 movaps %xmm2,%xmm3 mulps %xmm2,%xmm2 movaps nb113_three(%esp),%xmm0 mulps %xmm5,%xmm2 ## rsq*lu*lu subps %xmm2,%xmm0 ## constant 30-rsq*lu*lu mulps %xmm3,%xmm0 ## lu*(3-rsq*lu*lu) mulps nb113_half(%esp),%xmm0 movaps %xmm0,nb113_rinvH2(%esp) ## rinvH2 ## rsqM - seed to xmm2 rsqrtps %xmm4,%xmm2 movaps %xmm2,%xmm3 mulps %xmm2,%xmm2 movaps nb113_three(%esp),%xmm0 mulps %xmm4,%xmm2 ## rsq*lu*lu subps %xmm2,%xmm0 ## constant 30-rsq*lu*lu mulps %xmm3,%xmm0 ## lu*(3-rsq*lu*lu) mulps nb113_half(%esp),%xmm0 movaps %xmm0,nb113_rinvM(%esp) ## Do the O LJ-only interaction directly. rcpps %xmm7,%xmm2 movaps nb113_two(%esp),%xmm1 mulps %xmm2,%xmm7 subps %xmm7,%xmm1 mulps %xmm1,%xmm2 ## rinvsq movaps %xmm2,%xmm0 mulps %xmm2,%xmm0 ## r4 mulps %xmm2,%xmm0 ## r6 movaps %xmm0,%xmm1 mulps %xmm1,%xmm1 ## r12 mulps nb113_c6(%esp),%xmm0 mulps nb113_c12(%esp),%xmm1 movaps %xmm1,%xmm3 subps %xmm0,%xmm3 ## Vvdw12-Vvdw6 addps nb113_Vvdwtot(%esp),%xmm3 movaps %xmm3,nb113_Vvdwtot(%esp) mulps nb113_six(%esp),%xmm0 mulps nb113_twelve(%esp),%xmm1 subps %xmm0,%xmm1 mulps %xmm2,%xmm1 ## fscal movaps nb113_dxO(%esp),%xmm3 movaps nb113_dyO(%esp),%xmm4 movaps nb113_dzO(%esp),%xmm5 mulps %xmm1,%xmm3 mulps %xmm1,%xmm4 mulps %xmm1,%xmm5 ## tx in xmm3-xmm5 ## update O forces movaps nb113_fixO(%esp),%xmm0 movaps nb113_fiyO(%esp),%xmm1 movaps nb113_fizO(%esp),%xmm2 addps %xmm3,%xmm0 addps %xmm4,%xmm1 addps %xmm5,%xmm2 movaps %xmm0,nb113_fixO(%esp) movaps %xmm1,nb113_fiyO(%esp) movaps %xmm2,nb113_fizO(%esp) ## update j forces with water O movaps %xmm3,nb113_fjx(%esp) movaps %xmm4,nb113_fjy(%esp) movaps %xmm5,nb113_fjz(%esp) ## Do H1 interaction movaps nb113_rinvH1(%esp),%xmm7 movaps %xmm7,%xmm4 mulps %xmm4,%xmm4 ## xmm7=rinv, xmm4=rinvsq mulps nb113_qqH(%esp),%xmm7 ## xmm7=vcoul mulps %xmm7,%xmm4 ## total fsH1 in xmm4 addps nb113_vctot(%esp),%xmm7 movaps %xmm7,nb113_vctot(%esp) movaps nb113_dxH1(%esp),%xmm0 movaps nb113_dyH1(%esp),%xmm1 movaps nb113_dzH1(%esp),%xmm2 mulps %xmm4,%xmm0 mulps %xmm4,%xmm1 mulps %xmm4,%xmm2 ## update H1 forces movaps nb113_fixH1(%esp),%xmm3 movaps nb113_fiyH1(%esp),%xmm4 movaps nb113_fizH1(%esp),%xmm7 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm7 movaps %xmm3,nb113_fixH1(%esp) movaps %xmm4,nb113_fiyH1(%esp) movaps %xmm7,nb113_fizH1(%esp) ## update j forces with water H1 addps nb113_fjx(%esp),%xmm0 addps nb113_fjy(%esp),%xmm1 addps nb113_fjz(%esp),%xmm2 movaps %xmm0,nb113_fjx(%esp) movaps %xmm1,nb113_fjy(%esp) movaps %xmm2,nb113_fjz(%esp) ## Done with H1, do H2 interactions movaps nb113_rinvH2(%esp),%xmm7 movaps %xmm7,%xmm4 mulps %xmm4,%xmm4 ## xmm7=rinv, xmm4=rinvsq mulps nb113_qqH(%esp),%xmm7 ## xmm7=vcoul mulps %xmm7,%xmm4 ## total fsH2 in xmm4 addps nb113_vctot(%esp),%xmm7 movaps %xmm7,nb113_vctot(%esp) movaps nb113_dxH2(%esp),%xmm0 movaps nb113_dyH2(%esp),%xmm1 movaps nb113_dzH2(%esp),%xmm2 mulps %xmm4,%xmm0 mulps %xmm4,%xmm1 mulps %xmm4,%xmm2 ## update H2 forces movaps nb113_fixH2(%esp),%xmm3 movaps nb113_fiyH2(%esp),%xmm4 movaps nb113_fizH2(%esp),%xmm7 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm7 movaps %xmm3,nb113_fixH2(%esp) movaps %xmm4,nb113_fiyH2(%esp) movaps %xmm7,nb113_fizH2(%esp) addps nb113_fjx(%esp),%xmm0 addps nb113_fjy(%esp),%xmm1 addps nb113_fjz(%esp),%xmm2 movaps %xmm0,nb113_fjx(%esp) movaps %xmm1,nb113_fjy(%esp) movaps %xmm2,nb113_fjz(%esp) ## Done with H2, do M interactions movaps nb113_rinvM(%esp),%xmm7 movaps %xmm7,%xmm4 mulps %xmm4,%xmm4 ## xmm7=rinv, xmm4=rinvsq mulps nb113_qqM(%esp),%xmm7 ## xmm7=vcoul mulps %xmm7,%xmm4 ## total fsM in xmm4 addps nb113_vctot(%esp),%xmm7 movaps %xmm7,nb113_vctot(%esp) movaps nb113_dxM(%esp),%xmm0 movaps nb113_dyM(%esp),%xmm1 movaps nb113_dzM(%esp),%xmm2 mulps %xmm4,%xmm0 mulps %xmm4,%xmm1 mulps %xmm4,%xmm2 ## update M forces movaps nb113_fixM(%esp),%xmm3 movaps nb113_fiyM(%esp),%xmm4 movaps nb113_fizM(%esp),%xmm7 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm7 movaps %xmm3,nb113_fixM(%esp) movaps %xmm4,nb113_fiyM(%esp) movaps %xmm7,nb113_fizM(%esp) movl nb113_faction(%ebp),%edi ## update j forces from stored values addps nb113_fjx(%esp),%xmm0 addps nb113_fjy(%esp),%xmm1 addps nb113_fjz(%esp),%xmm2 movlps (%edi,%eax,4),%xmm4 movlps (%edi,%ecx,4),%xmm7 movhps (%edi,%ebx,4),%xmm4 movhps (%edi,%edx,4),%xmm7 movaps %xmm4,%xmm3 shufps $136,%xmm7,%xmm3 ## constant 10001000 shufps $221,%xmm7,%xmm4 ## constant 11011101 ## xmm3 has fjx, xmm4 has fjy subps %xmm0,%xmm3 subps %xmm1,%xmm4 ## unpack them back for storing movaps %xmm3,%xmm7 unpcklps %xmm4,%xmm7 unpckhps %xmm4,%xmm3 movlps %xmm7,(%edi,%eax,4) movlps %xmm3,(%edi,%ecx,4) movhps %xmm7,(%edi,%ebx,4) movhps %xmm3,(%edi,%edx,4) ## finally z forces movss 8(%edi,%eax,4),%xmm0 movss 8(%edi,%ebx,4),%xmm1 movss 8(%edi,%ecx,4),%xmm3 movss 8(%edi,%edx,4),%xmm4 subss %xmm2,%xmm0 shufps $229,%xmm2,%xmm2 ## constant 11100101 subss %xmm2,%xmm1 shufps $234,%xmm2,%xmm2 ## constant 11101010 subss %xmm2,%xmm3 shufps $255,%xmm2,%xmm2 ## constant 11111111 subss %xmm2,%xmm4 movss %xmm0,8(%edi,%eax,4) movss %xmm1,8(%edi,%ebx,4) movss %xmm3,8(%edi,%ecx,4) movss %xmm4,8(%edi,%edx,4) ## should we do one more iteration? subl $4,nb113_innerk(%esp) jl _nb_kernel113_ia32_sse.nb113_odd_inner jmp _nb_kernel113_ia32_sse.nb113_unroll_loop_nb_kernel113_ia32_sse.nb113_odd_inner: addl $4,nb113_innerk(%esp) jnz _nb_kernel113_ia32_sse.nb113_odd_loop jmp _nb_kernel113_ia32_sse.nb113_updateouterdata_nb_kernel113_ia32_sse.nb113_odd_loop: movl nb113_innerjjnr(%esp),%edx ## pointer to jjnr[k] movl (%edx),%eax addl $4,nb113_innerjjnr(%esp) xorps %xmm4,%xmm4 ## clear reg. movss nb113_iqM(%esp),%xmm4 movl nb113_charge(%ebp),%esi movhps nb113_iqH(%esp),%xmm4 ## [qM 0 qH qH] shufps $41,%xmm4,%xmm4 ## [0 qH qH qM] movss (%esi,%eax,4),%xmm3 ## charge in xmm3 shufps $0,%xmm3,%xmm3 mulps %xmm4,%xmm3 movaps %xmm3,nb113_qqM(%esp) ## use dummy qq for storage xorps %xmm6,%xmm6 movl nb113_type(%ebp),%esi movl (%esi,%eax,4),%ebx movl nb113_vdwparam(%ebp),%esi shll %ebx addl nb113_ntia(%esp),%ebx movlps (%esi,%ebx,4),%xmm6 movaps %xmm6,%xmm7 shufps $252,%xmm6,%xmm6 ## constant 11111100 shufps $253,%xmm7,%xmm7 ## constant 11111101 movaps %xmm6,nb113_c6(%esp) movaps %xmm7,nb113_c12(%esp) movl nb113_pos(%ebp),%esi leal (%eax,%eax,2),%eax movss nb113_ixO(%esp),%xmm3 movss nb113_iyO(%esp),%xmm4 movss nb113_izO(%esp),%xmm5 movss nb113_ixH1(%esp),%xmm0 movss nb113_iyH1(%esp),%xmm1 movss nb113_izH1(%esp),%xmm2 unpcklps nb113_ixH2(%esp),%xmm3 ## ixO ixH2 - - unpcklps nb113_iyH2(%esp),%xmm4 ## iyO iyH2 - - unpcklps nb113_izH2(%esp),%xmm5 ## izO izH2 - - unpcklps nb113_ixM(%esp),%xmm0 ## ixH1 ixM - - unpcklps nb113_iyM(%esp),%xmm1 ## iyH1 iyM - - unpcklps nb113_izM(%esp),%xmm2 ## izH1 izM - - unpcklps %xmm0,%xmm3 ## ixO ixH1 ixH2 ixM unpcklps %xmm1,%xmm4 ## same for y unpcklps %xmm2,%xmm5 ## same for z ## move j coords to xmm0-xmm2 movss (%esi,%eax,4),%xmm0 movss 4(%esi,%eax,4),%xmm1 movss 8(%esi,%eax,4),%xmm2 shufps $0,%xmm0,%xmm0 shufps $0,%xmm1,%xmm1 shufps $0,%xmm2,%xmm2 subps %xmm0,%xmm3 subps %xmm1,%xmm4 subps %xmm2,%xmm5 ## use O distances for storage movaps %xmm3,nb113_dxO(%esp) movaps %xmm4,nb113_dyO(%esp) movaps %xmm5,nb113_dzO(%esp) mulps %xmm3,%xmm3 mulps %xmm4,%xmm4 mulps %xmm5,%xmm5 addps %xmm3,%xmm4 addps %xmm5,%xmm4 ## rsq in xmm4 rsqrtps %xmm4,%xmm5 ## lookup seed in xmm5 movaps %xmm5,%xmm2 mulps %xmm5,%xmm5 movaps nb113_three(%esp),%xmm1 mulps %xmm4,%xmm5 ## rsq*lu*lu movaps nb113_half(%esp),%xmm0 subps %xmm5,%xmm1 ## constant 30-rsq*lu*lu mulps %xmm2,%xmm1 mulps %xmm1,%xmm0 ## xmm0=rinv movaps %xmm0,%xmm1 mulps %xmm1,%xmm1 ## rinvsq movaps %xmm0,%xmm7 mulps nb113_qqM(%esp),%xmm7 ## vcoul movaps %xmm7,%xmm6 addps nb113_vctot(%esp),%xmm7 movaps %xmm7,nb113_vctot(%esp) movaps %xmm1,%xmm2 mulss %xmm1,%xmm1 mulss %xmm2,%xmm1 ## xmm1=rinvsix xorps %xmm4,%xmm4 movss %xmm1,%xmm4 mulss %xmm4,%xmm4 ## xmm4=rinvtwelve mulss nb113_c6(%esp),%xmm1 mulss nb113_c12(%esp),%xmm4 movaps %xmm4,%xmm3 subss %xmm1,%xmm3 ## xmm3=Vvdw12-Vvdw6 mulss nb113_six(%esp),%xmm1 mulss nb113_twelve(%esp),%xmm4 subss %xmm1,%xmm4 addss nb113_Vvdwtot(%esp),%xmm3 movss %xmm3,nb113_Vvdwtot(%esp) addps %xmm6,%xmm4 mulps %xmm0,%xmm4 mulps %xmm0,%xmm4 ## fscal movaps nb113_dxO(%esp),%xmm0 movaps nb113_dyO(%esp),%xmm1 movaps nb113_dzO(%esp),%xmm2 mulps %xmm4,%xmm0 mulps %xmm4,%xmm1 mulps %xmm4,%xmm2 ## xmm0-xmm2 now contains tx-tz (partial force) movss nb113_fixO(%esp),%xmm3 movss nb113_fiyO(%esp),%xmm4 movss nb113_fizO(%esp),%xmm5 addss %xmm0,%xmm3 addss %xmm1,%xmm4 addss %xmm2,%xmm5 movss %xmm3,nb113_fixO(%esp) movss %xmm4,nb113_fiyO(%esp) movss %xmm5,nb113_fizO(%esp) ## updated the O force now do the H's movaps %xmm0,%xmm3 movaps %xmm1,%xmm4 movaps %xmm2,%xmm5 shufps $0x39,%xmm3,%xmm3 ## shift right shufps $0x39,%xmm4,%xmm4 shufps $0x39,%xmm5,%xmm5 addss nb113_fixH1(%esp),%xmm3 addss nb113_fiyH1(%esp),%xmm4 addss nb113_fizH1(%esp),%xmm5 movss %xmm3,nb113_fixH1(%esp) movss %xmm4,nb113_fiyH1(%esp) movss %xmm5,nb113_fizH1(%esp) ## updated the H1 force shufps $0x39,%xmm3,%xmm3 shufps $0x39,%xmm4,%xmm4 shufps $0x39,%xmm5,%xmm5 addss nb113_fixH2(%esp),%xmm3 addss nb113_fiyH2(%esp),%xmm4 addss nb113_fizH2(%esp),%xmm5 movss %xmm3,nb113_fixH2(%esp) movss %xmm4,nb113_fiyH2(%esp) movss %xmm5,nb113_fizH2(%esp) ## updated the H2 force movl nb113_faction(%ebp),%edi shufps $0x39,%xmm3,%xmm3 shufps $0x39,%xmm4,%xmm4 shufps $0x39,%xmm5,%xmm5 addss nb113_fixM(%esp),%xmm3 addss nb113_fiyM(%esp),%xmm4 addss nb113_fizM(%esp),%xmm5 movss %xmm3,nb113_fixM(%esp) movss %xmm4,nb113_fiyM(%esp) movss %xmm5,nb113_fizM(%esp) ## updated the M force ## the fj's - move in from mem start by acc. tx/ty/tz in xmm0, xmm1 movlps (%edi,%eax,4),%xmm6 movss 8(%edi,%eax,4),%xmm7 movhlps %xmm0,%xmm3 movhlps %xmm1,%xmm4 movhlps %xmm2,%xmm5 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm5 movaps %xmm3,%xmm0 movaps %xmm4,%xmm1 movaps %xmm5,%xmm2 shufps $0x39,%xmm3,%xmm3 ## shift right shufps $0x39,%xmm4,%xmm4 shufps $0x39,%xmm5,%xmm5 addss %xmm3,%xmm0 addss %xmm4,%xmm1 addss %xmm5,%xmm2 unpcklps %xmm1,%xmm0 ## x,y sum in xmm0, z sum in xmm2 subps %xmm0,%xmm6 subss %xmm2,%xmm7 movlps %xmm6,(%edi,%eax,4) movss %xmm7,8(%edi,%eax,4) decl nb113_innerk(%esp) jz _nb_kernel113_ia32_sse.nb113_updateouterdata jmp _nb_kernel113_ia32_sse.nb113_odd_loop_nb_kernel113_ia32_sse.nb113_updateouterdata: movl nb113_ii3(%esp),%ecx movl nb113_faction(%ebp),%edi movl nb113_fshift(%ebp),%esi movl nb113_is3(%esp),%edx⌨️ 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?