nb_kernel132_x86_64_sse.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,232 行 · 第 1/5 页
S
2,232 行
addps xmm0, [rsp + nb132_fixO] addps xmm1, [rsp + nb132_fiyO] addps xmm2, [rsp + nb132_fizO] movaps [rsp + nb132_fixO], xmm0 movaps [rsp + nb132_fiyO], xmm1 movaps [rsp + nb132_fizO], xmm2 movaps [rsp + nb132_fjxO], xmm3 movaps [rsp + nb132_fjyO], xmm4 movaps [rsp + nb132_fjzO], xmm5 ;# done with i O Now do i H1 & H2 simultaneously first get i particle coords: movaps xmm0, [rsp + nb132_jxO] movaps xmm1, [rsp + nb132_jyO] movaps xmm2, [rsp + nb132_jzO] movaps xmm3, xmm0 movaps xmm4, xmm1 movaps xmm5, xmm2 subps xmm0, [rsp + nb132_ixH1] subps xmm1, [rsp + nb132_iyH1] subps xmm2, [rsp + nb132_izH1] subps xmm3, [rsp + nb132_ixH2] subps xmm4, [rsp + nb132_iyH2] subps xmm5, [rsp + nb132_izH2] movaps [rsp + nb132_dxH1O], xmm0 movaps [rsp + nb132_dyH1O], xmm1 movaps [rsp + nb132_dzH1O], xmm2 movaps [rsp + nb132_dxH2O], xmm3 movaps [rsp + nb132_dyH2O], xmm4 movaps [rsp + nb132_dzH2O], xmm5 mulps xmm0, xmm0 mulps xmm1, xmm1 mulps xmm2, xmm2 mulps xmm3, xmm3 mulps xmm4, xmm4 mulps xmm5, xmm5 addps xmm0, xmm1 addps xmm4, xmm3 addps xmm0, xmm2 ;# have rsqH1 in xmm0 addps xmm4, xmm5 ;# have rsqH2 in xmm4 ;# do invsqrt rsqrtps xmm1, xmm0 rsqrtps xmm5, xmm4 movaps xmm2, xmm1 movaps xmm6, xmm5 mulps xmm1, xmm1 mulps xmm5, xmm5 movaps xmm3, [rsp + nb132_three] movaps xmm7, xmm3 mulps xmm1, xmm0 mulps xmm5, xmm4 subps xmm3, xmm1 subps xmm7, xmm5 mulps xmm3, xmm2 mulps xmm7, xmm6 mulps xmm3, [rsp + nb132_half] ;# rinv H1 - j water mulps xmm7, [rsp + nb132_half] ;# rinv H2 - j water ;# assemble charges in xmm6 xorps xmm6, xmm6 movss xmm6, [rsp + nb132_qqOH] movhps xmm6, [rsp + nb132_qqHH] movaps xmm0, xmm3 movaps xmm1, xmm7 mulps xmm3, xmm6 ;# vcoul mulps xmm7, xmm6 ;# vcoul mulps xmm0, xmm0 mulps xmm1, xmm1 mulps xmm0, xmm3 addps xmm3, xmm7 mulps xmm7, xmm1 addps xmm3, [rsp + nb132_vctot] movaps [rsp + nb132_vctot], xmm3 movaps xmm1, xmm0 movaps xmm2, xmm0 mulps xmm0, [rsp + nb132_dxH1O] mulps xmm1, [rsp + nb132_dyH1O] mulps xmm2, [rsp + nb132_dzH1O] ;# update forces H1 - j water movaps xmm3, xmm0 movaps xmm4, xmm1 movaps xmm5, xmm2 addps xmm0, [rsp + nb132_fixH1] addps xmm1, [rsp + nb132_fiyH1] addps xmm2, [rsp + nb132_fizH1] addps xmm3, [rsp + nb132_fjxO] addps xmm4, [rsp + nb132_fjyO] addps xmm5, [rsp + nb132_fjzO] movaps [rsp + nb132_fixH1], xmm0 movaps [rsp + nb132_fiyH1], xmm1 movaps [rsp + nb132_fizH1], xmm2 movaps [rsp + nb132_fjxO], xmm3 movaps [rsp + nb132_fjyO], xmm4 movaps [rsp + nb132_fjzO], xmm5 ;# do forces H2 - j water movaps xmm0, xmm7 movaps xmm1, xmm7 movaps xmm2, xmm7 mulps xmm0, [rsp + nb132_dxH2O] mulps xmm1, [rsp + nb132_dyH2O] mulps xmm2, [rsp + nb132_dzH2O] movaps xmm3, xmm0 movaps xmm4, xmm1 movaps xmm5, xmm2 addps xmm0, [rsp + nb132_fixH2] addps xmm1, [rsp + nb132_fiyH2] addps xmm2, [rsp + nb132_fizH2] addps xmm3, [rsp + nb132_fjxO] addps xmm4, [rsp + nb132_fjyO] addps xmm5, [rsp + nb132_fjzO] movaps [rsp + nb132_fixH2], xmm0 movaps [rsp + nb132_fiyH2], xmm1 movaps [rsp + nb132_fizH2], xmm2 ;# fjx/fjy/fjz in xmm3-xmm5 mov rsi, [rbp + nb132_faction] ;# update j water forces from local variables movlps xmm0, [rsi + rax*4] ;# movlps xmm1, [rsi + rax*4 + 12] movhps xmm1, [rsi + rax*4 + 24] movaps xmm6, xmm5 movaps xmm7, xmm5 shufps xmm6, xmm6, 2 ;# constant 00000010 shufps xmm7, xmm7, 3 ;# constant 00000011 addss xmm5, [rsi + rax*4 + 8] addss xmm6, [rsi + rax*4 + 20] addss xmm7, [rsi + rax*4 + 32] movss [rsi + rax*4 + 8], xmm5 movss [rsi + rax*4 + 20], xmm6 movss [rsi + rax*4 + 32], xmm7 movaps xmm5, xmm3 unpcklps xmm3, xmm4 unpckhps xmm5, xmm4 addps xmm0, xmm3 addps xmm1, xmm5 movlps [rsi + rax*4], xmm0 movlps [rsi + rax*4 + 12], xmm1 movhps [rsi + rax*4 + 24], xmm1 dec dword ptr [rsp + nb132_innerk] jz .nb132_updateouterdata jmp .nb132_single_loop.nb132_updateouterdata: mov ecx, [rsp + nb132_ii3] mov rdi, [rbp + nb132_faction] mov rsi, [rbp + nb132_fshift] mov edx, [rsp + nb132_is3] ;# accumulate Oi forces in xmm0, xmm1, xmm2 movaps xmm0, [rsp + nb132_fixO] movaps xmm1, [rsp + nb132_fiyO] movaps xmm2, [rsp + nb132_fizO] movhlps xmm3, xmm0 movhlps xmm4, xmm1 movhlps xmm5, xmm2 addps xmm0, xmm3 addps xmm1, xmm4 addps xmm2, xmm5 ;# sum is in 1/2 in xmm0-xmm2 movaps xmm3, xmm0 movaps xmm4, xmm1 movaps xmm5, xmm2 shufps xmm3, xmm3, 1 shufps xmm4, xmm4, 1 shufps xmm5, xmm5, 1 addss xmm0, xmm3 addss xmm1, xmm4 addss xmm2, xmm5 ;# xmm0-xmm2 has single force in pos0 ;# increment i force movss xmm3, [rdi + rcx*4] movss xmm4, [rdi + rcx*4 + 4] movss xmm5, [rdi + rcx*4 + 8] subss xmm3, xmm0 subss xmm4, xmm1 subss xmm5, xmm2 movss [rdi + rcx*4], xmm3 movss [rdi + rcx*4 + 4], xmm4 movss [rdi + rcx*4 + 8], xmm5 ;# accumulate force in xmm6/xmm7 for fshift movaps xmm6, xmm0 movss xmm7, xmm2 movlhps xmm6, xmm1 shufps xmm6, xmm6, 8 ;# constant 00001000 ;# accumulate H1i forces in xmm0, xmm1, xmm2 movaps xmm0, [rsp + nb132_fixH1] movaps xmm1, [rsp + nb132_fiyH1] movaps xmm2, [rsp + nb132_fizH1] movhlps xmm3, xmm0 movhlps xmm4, xmm1 movhlps xmm5, xmm2 addps xmm0, xmm3 addps xmm1, xmm4 addps xmm2, xmm5 ;# sum is in 1/2 in xmm0-xmm2 movaps xmm3, xmm0 movaps xmm4, xmm1 movaps xmm5, xmm2 shufps xmm3, xmm3, 1 shufps xmm4, xmm4, 1 shufps xmm5, xmm5, 1 addss xmm0, xmm3 addss xmm1, xmm4 addss xmm2, xmm5 ;# xmm0-xmm2 has single force in pos0 ;# increment i force movss xmm3, [rdi + rcx*4 + 12] movss xmm4, [rdi + rcx*4 + 16] movss xmm5, [rdi + rcx*4 + 20] subss xmm3, xmm0 subss xmm4, xmm1 subss xmm5, xmm2 movss [rdi + rcx*4 + 12], xmm3 movss [rdi + rcx*4 + 16], xmm4 movss [rdi + rcx*4 + 20], xmm5 ;# accumulate force in xmm6/xmm7 for fshift addss xmm7, xmm2 movlhps xmm0, xmm1 shufps xmm0, xmm0, 8 ;# constant 00001000 addps xmm6, xmm0 ;# accumulate H2i forces in xmm0, xmm1, xmm2 movaps xmm0, [rsp + nb132_fixH2] movaps xmm1, [rsp + nb132_fiyH2] movaps xmm2, [rsp + nb132_fizH2] movhlps xmm3, xmm0 movhlps xmm4, xmm1 movhlps xmm5, xmm2 addps xmm0, xmm3 addps xmm1, xmm4 addps xmm2, xmm5 ;# sum is in 1/2 in xmm0-xmm2 movaps xmm3, xmm0 movaps xmm4, xmm1 movaps xmm5, xmm2 shufps xmm3, xmm3, 1 shufps xmm4, xmm4, 1 shufps xmm5, xmm5, 1 addss xmm0, xmm3 addss xmm1, xmm4 addss xmm2, xmm5 ;# xmm0-xmm2 has single force in pos0 ;# increment i force movss xmm3, [rdi + rcx*4 + 24] movss xmm4, [rdi + rcx*4 + 28] movss xmm5, [rdi + rcx*4 + 32] subss xmm3, xmm0 subss xmm4, xmm1 subss xmm5, xmm2 movss [rdi + rcx*4 + 24], xmm3 movss [rdi + rcx*4 + 28], xmm4 movss [rdi + rcx*4 + 32], xmm5 ;# accumulate force in xmm6/xmm7 for fshift addss xmm7, xmm2 movlhps xmm0, xmm1 shufps xmm0, xmm0, 8 ;# constant 00001000 addps xmm6, xmm0 ;# increment fshift force movlps xmm3, [rsi + rdx*4] movss xmm4, [rsi + rdx*4 + 8] subps xmm3, xmm6 subss xmm4, xmm7 movlps [rsi + rdx*4], xmm3 movss [rsi + rdx*4 + 8], xmm4 ;# get n from stack mov esi, [rsp + nb132_n] ;# get group index for i particle mov rdx, [rbp + nb132_gid] ;# base of gid[] mov edx, [rdx + rsi*4] ;# ggid=gid[n] ;# accumulate total potential energy and update it movaps xmm7, [rsp + nb132_vctot] ;# accumulate movhlps xmm6, xmm7 addps xmm7, xmm6 ;# pos 0-1 in xmm7 have the sum now movaps xmm6, xmm7 shufps xmm6, xmm6, 1 addss xmm7, xmm6 ;# add earlier value from mem mov rax, [rbp + nb132_Vc] addss xmm7, [rax + rdx*4] ;# move back to mem movss [rax + rdx*4], xmm7 ;# accumulate total lj energy and update it movaps xmm7, [rsp + nb132_Vvdwtot] ;# accumulate movhlps xmm6, xmm7 addps xmm7, xmm6 ;# pos 0-1 in xmm7 have the sum now movaps xmm6, xmm7 shufps xmm6, xmm6, 1 addss xmm7, xmm6 ;# add earlier value from mem mov rax, [rbp + nb132_Vvdw] addss xmm7, [rax + rdx*4] ;# move back to mem movss [rax + rdx*4], xmm7 ;# finish if last mov ecx, [rsp + nb132_nn1] ;# esi already loaded with n inc esi sub ecx, esi jz .nb132_outerend ;# not last, iterate outer loop once more! mov [rsp + nb132_n], esi jmp .nb132_outer.nb132_outerend: ;# check if more outer neighborlists remain mov ecx, [rsp + nb132_nri] ;# esi already loaded with n above sub ecx, esi jz .nb132_end ;# non-zero, do one more workunit jmp .nb132_threadloop.nb132_end: mov eax, [rsp + nb132_nouter] mov ebx, [rsp + nb132_ninner] mov rcx, [rbp + nb132_outeriter] mov rdx, [rbp + nb132_inneriter] mov [rcx], eax mov [rdx], ebx add rsp, 1624 emms pop r15 pop r14 pop r13 pop r12 pop rbx pop rbp ret.globl nb_kernel132nf_x86_64_sse.globl _nb_kernel132nf_x86_64_ssenb_kernel132nf_x86_64_sse: _nb_kernel132nf_x86_64_sse: ;# Room for return address and rbp (16 bytes).equiv nb132nf_fshift, 16.equiv nb132nf_gid, 24.equiv nb132nf_pos, 32.equiv nb132nf_faction, 40.equiv nb132nf_charge, 48.equiv nb132nf_p_facel, 56.equiv nb132nf_argkrf, 64.equiv nb132nf_argcrf, 72.equiv nb132nf_Vc, 80.equiv nb132nf_type, 88.equiv nb132nf_p_ntype, 96.equiv nb132nf_vdwparam, 104.equiv nb132nf_Vvdw, 112.equiv nb132nf_p_tabscale, 120.equiv nb132nf_VFtab, 128.equiv nb132nf_invsqrta, 136.equiv nb132nf_dvda, 144.equiv nb132nf_p_gbtabscale, 152.equiv nb132nf_GBtab, 160.equiv nb132nf_p_nthreads, 168.equiv nb132nf_count, 176.equiv nb132nf_mtx, 184.equiv nb132nf_outeriter, 192.equiv nb132nf_inneriter, 200.equiv nb132nf_work, 208 ;# stack offsets for local variables ;# bottom of stack is cache-aligned for sse use .equiv nb132nf_ixO, 0.equiv nb132nf_iyO, 16.equiv nb132nf_izO, 32.equiv nb132nf_ixH1, 48.equiv nb132nf_iyH1, 64.equiv nb132nf_izH1, 80.equiv nb132nf_ixH2, 96.equiv nb132nf_iyH2, 112.equiv nb132nf_izH2, 128.equiv nb132nf_jxO, 144.equiv nb132nf_jyO, 160.equiv nb132nf_jzO, 176.equiv nb132nf_jxH1, 192.equiv nb132nf_jyH1, 208.equiv nb132nf_jzH1, 224.equiv nb132nf_jxH2, 240.equiv nb132nf_jyH2, 256.equiv nb132nf_jzH2, 272.equiv nb132nf_qqOO, 288.equiv nb132nf_qqOH, 304.equiv nb132nf_qqHH, 320.equiv nb132nf_c6, 336.equiv nb132nf_c12, 352.equiv nb132nf_vctot, 368.equiv nb132nf_Vvdwtot, 384.equiv nb132nf_half, 400.equiv nb132nf_three, 416.equiv nb132nf_rsqOO, 432.equiv nb132nf_rsqOH1, 448.equiv nb132nf_rsqOH2, 464.equiv nb132nf_rsqH1O, 480.equiv nb132nf_rsqH1H1, 496.equiv nb132nf_rsqH1H2, 512.equiv nb132nf_rsqH2O, 528.equiv nb132nf_rsqH2H1, 544.equiv nb132nf_rsqH2H2, 560.equiv nb132nf_rinvOO, 576.equiv nb132nf_rinvOH1, 592.equiv nb132nf_rinvOH2, 608.equiv nb132nf_rinvH1O, 624.equiv nb132nf_rinvH1H1, 640.equiv nb132nf_rinvH1H2, 656.equiv nb132nf_rinvH2O, 672.equiv nb132nf_rinvH2H1, 688.equiv nb132nf_rinvH2H2, 704.equiv nb132nf_krf, 720.equiv nb132nf_crf, 736.equiv nb132nf_tsc, 752.equiv nb132nf_nri, 768.equiv nb132nf_iinr, 776.equiv nb132nf_jindex, 784.equiv nb132nf_jjnr, 792.equiv nb132nf_shift, 800.equiv nb132nf_shiftvec, 808⌨️ 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?