nb_kernel130_ia32_sse.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 1,988 行 · 第 1/4 页
S
1,988 行
movaps nb130_dx(%esp),%xmm0 movaps nb130_dy(%esp),%xmm1 movaps nb130_dz(%esp),%xmm2 movl nb130_faction(%ebp),%edi mulps %xmm3,%xmm0 mulps %xmm3,%xmm1 mulps %xmm3,%xmm2 ## xmm0-xmm2 contains tx-tz (partial force) ## now update f_i movaps nb130_fix(%esp),%xmm3 movaps nb130_fiy(%esp),%xmm4 movaps nb130_fiz(%esp),%xmm5 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm5 movaps %xmm3,nb130_fix(%esp) movaps %xmm4,nb130_fiy(%esp) movaps %xmm5,nb130_fiz(%esp) ## the fj's - start by accumulating x & y forces from memory movlps (%edi,%eax,4),%xmm4 movlps (%edi,%ecx,4),%xmm6 movhps (%edi,%ebx,4),%xmm4 movhps (%edi,%edx,4),%xmm6 movaps %xmm4,%xmm3 shufps $136,%xmm6,%xmm3 ## constant 10001000 shufps $221,%xmm6,%xmm4 ## constant 11011101 ## now xmm3-xmm5 contains fjx, fjy, fjz subps %xmm0,%xmm3 subps %xmm1,%xmm4 ## unpack them back so we can store them - first x & y in xmm3/xmm4 movaps %xmm3,%xmm6 unpcklps %xmm4,%xmm6 unpckhps %xmm4,%xmm3 ## xmm6(l)=x & y for j1, (h) for j2 ## xmm3(l)=x & y for j3, (h) for j4 movlps %xmm6,(%edi,%eax,4) movlps %xmm3,(%edi,%ecx,4) movhps %xmm6,(%edi,%ebx,4) movhps %xmm3,(%edi,%edx,4) ## and the z forces movss 8(%edi,%eax,4),%xmm4 movss 8(%edi,%ebx,4),%xmm5 movss 8(%edi,%ecx,4),%xmm6 movss 8(%edi,%edx,4),%xmm7 subss %xmm2,%xmm4 shufps $229,%xmm2,%xmm2 ## constant 11100101 subss %xmm2,%xmm5 shufps $234,%xmm2,%xmm2 ## constant 11101010 subss %xmm2,%xmm6 shufps $255,%xmm2,%xmm2 ## constant 11111111 subss %xmm2,%xmm7 movss %xmm4,8(%edi,%eax,4) movss %xmm5,8(%edi,%ebx,4) movss %xmm6,8(%edi,%ecx,4) movss %xmm7,8(%edi,%edx,4) ## should we do one more iteration? subl $4,nb130_innerk(%esp) jl _nb_kernel130_ia32_sse.nb130_finish_inner jmp _nb_kernel130_ia32_sse.nb130_unroll_loop_nb_kernel130_ia32_sse.nb130_finish_inner: ## check if at least two particles remain addl $4,nb130_innerk(%esp) movl nb130_innerk(%esp),%edx andl $2,%edx jnz _nb_kernel130_ia32_sse.nb130_dopair jmp _nb_kernel130_ia32_sse.nb130_checksingle_nb_kernel130_ia32_sse.nb130_dopair: movl nb130_charge(%ebp),%esi movl nb130_innerjjnr(%esp),%ecx movl (%ecx),%eax movl 4(%ecx),%ebx addl $8,nb130_innerjjnr(%esp) xorps %xmm3,%xmm3 movss (%esi,%eax,4),%xmm3 movss (%esi,%ebx,4),%xmm6 shufps $12,%xmm6,%xmm3 ## constant 00001100 shufps $88,%xmm3,%xmm3 ## constant 01011000 ;# xmm3(0,1) has the charges movl nb130_type(%ebp),%esi movl %eax,%ecx movl %ebx,%edx movl (%esi,%ecx,4),%ecx movl (%esi,%edx,4),%edx movl nb130_vdwparam(%ebp),%esi shll %ecx shll %edx movl nb130_ntia(%esp),%edi addl %edi,%ecx addl %edi,%edx movlps (%esi,%ecx,4),%xmm6 movhps (%esi,%edx,4),%xmm6 movl nb130_pos(%ebp),%edi xorps %xmm7,%xmm7 movaps %xmm6,%xmm4 shufps $8,%xmm4,%xmm4 ## constant 00001000 shufps $13,%xmm6,%xmm6 ## constant 00001101 movlhps %xmm7,%xmm4 movlhps %xmm7,%xmm6 movaps %xmm4,nb130_c6(%esp) movaps %xmm6,nb130_c12(%esp) leal (%eax,%eax,2),%eax leal (%ebx,%ebx,2),%ebx ## move coordinates to xmm0-xmm2 movlps (%edi,%eax,4),%xmm1 movss 8(%edi,%eax,4),%xmm2 movhps (%edi,%ebx,4),%xmm1 movss 8(%edi,%ebx,4),%xmm0 mulps nb130_iq(%esp),%xmm3 movlhps %xmm7,%xmm3 shufps $0,%xmm0,%xmm2 movaps %xmm1,%xmm0 shufps $136,%xmm2,%xmm2 ## constant 10001000 shufps $136,%xmm0,%xmm0 ## constant 10001000 shufps $221,%xmm1,%xmm1 ## constant 11011101 movl nb130_faction(%ebp),%edi ## move ix-iz to xmm4-xmm6 xorps %xmm7,%xmm7 movaps nb130_ix(%esp),%xmm4 movaps nb130_iy(%esp),%xmm5 movaps nb130_iz(%esp),%xmm6 ## calc dr subps %xmm0,%xmm4 subps %xmm1,%xmm5 subps %xmm2,%xmm6 ## store dr movaps %xmm4,nb130_dx(%esp) movaps %xmm5,nb130_dy(%esp) movaps %xmm6,nb130_dz(%esp) ## square it mulps %xmm4,%xmm4 mulps %xmm5,%xmm5 mulps %xmm6,%xmm6 addps %xmm5,%xmm4 addps %xmm6,%xmm4 ## rsq in xmm4 movaps nb130_krf(%esp),%xmm7 rsqrtps %xmm4,%xmm5 ## lookup seed in xmm5 movaps %xmm5,%xmm2 mulps %xmm5,%xmm5 movaps nb130_three(%esp),%xmm1 mulps %xmm4,%xmm5 ## rsq*lu*lu movaps nb130_half(%esp),%xmm0 mulps %xmm4,%xmm7 ## xmm7=krsq subps %xmm5,%xmm1 ## constant 30-rsq*lu*lu mulps %xmm2,%xmm1 mulps %xmm1,%xmm0 ## xmm0=rinv movaps %xmm0,%xmm1 movaps %xmm0,%xmm2 mulps %xmm0,%xmm3 ## vcoul mulps %xmm3,%xmm2 ## vcoul*rinv addps nb130_vctot(%esp),%xmm3 movaps %xmm3,nb130_vctot(%esp) movaps %xmm2,nb130_fstmp(%esp) ## LJ table mulps %xmm1,%xmm4 ## r mulps nb130_tsc(%esp),%xmm4 ## rtab movaps %xmm1,%xmm0 ## copy of rinv cvttps2pi %xmm4,%mm6 cvtpi2ps %mm6,%xmm6 subps %xmm6,%xmm4 movaps %xmm4,%xmm1 ## xmm1=eps movaps %xmm1,%xmm2 mulps %xmm2,%xmm2 ## xmm2=eps2 pslld $3,%mm6 movd %eax,%mm0 movd %ebx,%mm1 movl nb130_VFtab(%ebp),%esi movd %mm6,%eax psrlq $32,%mm6 movd %mm6,%ebx ## dispersion movlps (%esi,%eax,4),%xmm5 movhps (%esi,%ebx,4),%xmm5 movaps %xmm5,%xmm4 shufps $136,%xmm7,%xmm4 ## constant 10001000 shufps $221,%xmm7,%xmm5 ## constant 11011101 movlps 8(%esi,%eax,4),%xmm7 movhps 8(%esi,%ebx,4),%xmm7 movaps %xmm7,%xmm6 shufps $136,%xmm3,%xmm6 ## constant 10001000 shufps $221,%xmm3,%xmm7 ## constant 11011101 ## dispersion table ready, in xmm4-xmm7 mulps %xmm1,%xmm6 ## xmm6=Geps mulps %xmm2,%xmm7 ## xmm7=Heps2 addps %xmm6,%xmm5 addps %xmm7,%xmm5 ## xmm5=Fp mulps nb130_two(%esp),%xmm7 ## two*Heps2 addps %xmm6,%xmm7 addps %xmm5,%xmm7 ## xmm7=FF mulps %xmm1,%xmm5 ## xmm5=eps*Fp addps %xmm4,%xmm5 ## xmm5=VV movaps nb130_c6(%esp),%xmm4 mulps %xmm4,%xmm7 ## fijD mulps %xmm4,%xmm5 ## Vvdw6 movaps nb130_fstmp(%esp),%xmm3 mulps nb130_tsc(%esp),%xmm7 subps %xmm7,%xmm3 ## put scalar force on stack Update Vvdwtot directly addps nb130_Vvdwtot(%esp),%xmm5 movaps %xmm3,nb130_fstmp(%esp) movaps %xmm5,nb130_Vvdwtot(%esp) ## repulsion movlps 16(%esi,%eax,4),%xmm5 movhps 16(%esi,%ebx,4),%xmm5 movaps %xmm5,%xmm4 shufps $136,%xmm7,%xmm4 ## constant 10001000 shufps $221,%xmm7,%xmm5 ## constant 11011101 movlps 24(%esi,%eax,4),%xmm7 movhps 24(%esi,%ebx,4),%xmm7 movaps %xmm7,%xmm6 shufps $136,%xmm3,%xmm6 ## constant 10001000 shufps $221,%xmm3,%xmm7 ## constant 11011101 ## table ready, in xmm4-xmm7 mulps %xmm1,%xmm6 ## xmm6=Geps mulps %xmm2,%xmm7 ## xmm7=Heps2 addps %xmm6,%xmm5 addps %xmm7,%xmm5 ## xmm5=Fp mulps nb130_two(%esp),%xmm7 ## two*Heps2 addps %xmm6,%xmm7 addps %xmm5,%xmm7 ## xmm7=FF mulps %xmm1,%xmm5 ## xmm5=eps*Fp addps %xmm4,%xmm5 ## xmm5=VV movaps nb130_c12(%esp),%xmm4 mulps %xmm4,%xmm7 ## fijR mulps %xmm4,%xmm5 ## Vvdw12 movaps nb130_fstmp(%esp),%xmm3 mulps nb130_tsc(%esp),%xmm7 subps %xmm7,%xmm3 addps nb130_Vvdwtot(%esp),%xmm5 movaps %xmm5,nb130_Vvdwtot(%esp) xorps %xmm4,%xmm4 mulps %xmm0,%xmm3 movd %mm0,%eax movd %mm1,%ebx movaps nb130_dx(%esp),%xmm0 movaps nb130_dy(%esp),%xmm1 movaps nb130_dz(%esp),%xmm2 mulps %xmm3,%xmm0 mulps %xmm3,%xmm1 mulps %xmm3,%xmm2 ## xmm0-xmm2 contains tx-tz (partial force) ## now update f_i movaps nb130_fix(%esp),%xmm3 movaps nb130_fiy(%esp),%xmm4 movaps nb130_fiz(%esp),%xmm5 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm5 movaps %xmm3,nb130_fix(%esp) movaps %xmm4,nb130_fiy(%esp) movaps %xmm5,nb130_fiz(%esp) ## update the fj's movss (%edi,%eax,4),%xmm3 movss 4(%edi,%eax,4),%xmm4 movss 8(%edi,%eax,4),%xmm5 subss %xmm0,%xmm3 subss %xmm1,%xmm4 subss %xmm2,%xmm5 movss %xmm3,(%edi,%eax,4) movss %xmm4,4(%edi,%eax,4) movss %xmm5,8(%edi,%eax,4) shufps $225,%xmm0,%xmm0 ## constant 11100001 shufps $225,%xmm1,%xmm1 ## constant 11100001 shufps $225,%xmm2,%xmm2 ## constant 11100001 movss (%edi,%ebx,4),%xmm3 movss 4(%edi,%ebx,4),%xmm4 movss 8(%edi,%ebx,4),%xmm5 subss %xmm0,%xmm3 subss %xmm1,%xmm4 subss %xmm2,%xmm5 movss %xmm3,(%edi,%ebx,4) movss %xmm4,4(%edi,%ebx,4) movss %xmm5,8(%edi,%ebx,4)_nb_kernel130_ia32_sse.nb130_checksingle: movl nb130_innerk(%esp),%edx andl $1,%edx jnz _nb_kernel130_ia32_sse.nb130_dosingle jmp _nb_kernel130_ia32_sse.nb130_updateouterdata_nb_kernel130_ia32_sse.nb130_dosingle: movl nb130_charge(%ebp),%esi movl nb130_pos(%ebp),%edi movl nb130_innerjjnr(%esp),%ecx xorps %xmm3,%xmm3 movl (%ecx),%eax movss (%esi,%eax,4),%xmm3 ## xmm3(0) has the charge movl nb130_type(%ebp),%esi movl %eax,%ecx movl (%esi,%ecx,4),%ecx movl nb130_vdwparam(%ebp),%esi shll %ecx addl nb130_ntia(%esp),%ecx xorps %xmm6,%xmm6 movlps (%esi,%ecx,4),%xmm6 movaps %xmm6,%xmm4 shufps $252,%xmm4,%xmm4 ## constant 11111100 shufps $253,%xmm6,%xmm6 ## constant 11111101 movaps %xmm4,nb130_c6(%esp) movaps %xmm6,nb130_c12(%esp) leal (%eax,%eax,2),%eax ## move coordinates to xmm0-xmm2 movss (%edi,%eax,4),%xmm0 movss 4(%edi,%eax,4),%xmm1 movss 8(%edi,%eax,4),%xmm2 mulps nb130_iq(%esp),%xmm3 xorps %xmm7,%xmm7 movaps nb130_ix(%esp),%xmm4 movaps nb130_iy(%esp),%xmm5 movaps nb130_iz(%esp),%xmm6 ## calc dr subps %xmm0,%xmm4 subps %xmm1,%xmm5 subps %xmm2,%xmm6 ## store dr movaps %xmm4,nb130_dx(%esp) movaps %xmm5,nb130_dy(%esp) movaps %xmm6,nb130_dz(%esp) ## square it mulps %xmm4,%xmm4 mulps %xmm5,%xmm5 mulps %xmm6,%xmm6 addps %xmm5,%xmm4 addps %xmm6,%xmm4 ## rsq in xmm4 movss nb130_krf(%esp),%xmm7 rsqrtss %xmm4,%xmm5 ## lookup seed in xmm5 movss %xmm5,%xmm2 mulss %xmm5,%xmm5 movss nb130_three(%esp),%xmm1 mulss %xmm4,%xmm5 ## rsq*lu*lu movss nb130_half(%esp),%xmm0 mulss %xmm4,%xmm7 ## xmm7=krsq subss %xmm5,%xmm1 ## constant 30-rsq*lu*lu mulss %xmm2,%xmm1 mulss %xmm1,%xmm0 ## xmm0=rinv movss %xmm0,%xmm1 movss %xmm0,%xmm2 mulss %xmm0,%xmm3 ## vcoul mulss %xmm3,%xmm2 ## vcoul*rinv addss nb130_vctot(%esp),%xmm3 movss %xmm3,nb130_vctot(%esp) movss %xmm2,nb130_fstmp(%esp) ## LJ table mulss %xmm1,%xmm4 ## r mulss nb130_tsc(%esp),%xmm4 ## rtab movaps %xmm1,%xmm0 ## copy of rinv cvttps2pi %xmm4,%mm6 cvtpi2ps %mm6,%xmm6 subss %xmm6,%xmm4 movss %xmm4,%xmm1 ## xmm1=eps movss %xmm1,%xmm2 mulss %xmm2,%xmm2 ## xmm2=eps2 pslld $3,%mm6 movd %eax,%mm0 movl nb130_VFtab(%ebp),%esi movd %mm6,%eax ## dispersion movlps (%esi,%eax,4),%xmm5 movaps %xmm5,%xmm4 shufps $136,%xmm7,%xmm4 ## constant 10001000 shufps $221,%xmm7,%xmm5 ## constant 11011101 movlps 8(%esi,%eax,4),%xmm7 movaps %xmm7,%xmm6 shufps $136,%xmm3,%xmm6 ## constant 10001000 shufps $221,%xmm3,%xmm7 ## constant 11011101 ## dispersion table ready, in xmm4-xmm7 mulss %xmm1,%xmm6 ## xmm6=Geps mulss %xmm2,%xmm7 ## xmm7=Heps2 addss %xmm6,%xmm5 addss %xmm7,%xmm5 ## xmm5=Fp mulss nb130_two(%esp),%xmm7 ## two*Heps2 addss %xmm6,%xmm7 addss %xmm5,%xmm7 ## xmm7=FF mulss %xmm1,%xmm5 ## xmm5=eps*Fp addss %xmm4,%xmm5 ## xmm5=VV movss nb130_c6(%esp),%xmm4 mulss %xmm4,%xmm7 ## fijD mulss %xmm4,%xmm5 ## Vvdw6 movss nb130_fstmp(%esp),%xmm3 mulss nb130_tsc(%esp),%xmm7 subss %xmm7,%xmm3 ## put scalar force on stack Update Vvdwtot directly addss nb130_Vvdwtot(%esp),%xmm5 movss %xmm3,nb130_fstmp(%esp) movss %xmm5,nb130_Vvdwtot(%esp) ## repulsion movlps 16(%esi,%eax,4),%xmm5 movaps %xmm5,%xmm4 shufps $136,%xmm7,%xmm4 ## constant 10001000 shufps $221,%xmm7,%xmm5 ## constant 11011101 movlps 24(%esi,%eax,4),%xmm7 movaps %xmm7,%xmm6 shufps $136,%xmm3,%xmm6 ## constant 10001000 shufps $221,%xmm3,%xmm7 ## constant 11011101 ## table ready, in xmm4-xmm7 mulss %xmm1,%xmm6 ## xmm6=Geps mulss %xmm2,%xmm7 ## xmm7=Heps2 addss %xmm6,%xmm5 addss %xmm7,%xmm5 ## xmm5=Fp mulss nb130_two(%esp),%xmm7 ## two*Heps2 addss %xmm6,%xmm7 addss %xmm5,%xmm7 ## xmm7=FF mulss %xmm1,%xmm5 ## xmm5=eps*Fp addss %xmm4,%xmm5 ## xmm5=VV movss nb130_c12(%esp),%xmm4 mulss %xmm4,%xmm7 ## fijR mulss %xmm4,%xmm5 ## Vvdw12 movss nb130_fstmp(%esp),%xmm3 mulss nb130_tsc(%esp),%xmm7 subss %xmm7,%xmm3 addss nb130_Vvdwtot(%esp),%xmm5 movss %xmm5,nb130_Vvdwtot(%esp) mulss %xmm0,%xmm3 movd %mm0,%eax movaps nb130_dx(%esp),%xmm0 movaps nb130_dy(%esp),%xmm1 movaps nb130_dz(%esp),%xmm2 mulss %xmm3,%xmm0 mulss %xmm3,%xmm1⌨️ 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?