nb_kernel234_x86_64_sse.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,212 行 · 第 1/5 页
S
2,212 行
;# update forces H2 - j water movaps xmm3, [rsp + nb234_fjxO] movaps xmm4, [rsp + nb234_fjyO] movaps xmm5, [rsp + nb234_fjzO] addps xmm3, xmm0 addps xmm4, xmm1 addps xmm5, xmm2 movaps [rsp + nb234_fjxO], xmm3 movaps [rsp + nb234_fjyO], xmm4 movaps [rsp + nb234_fjzO], xmm5 addps xmm0, [rsp + nb234_fixH2] addps xmm1, [rsp + nb234_fiyH2] addps xmm2, [rsp + nb234_fizH2] movaps [rsp + nb234_fixH2], xmm0 movaps [rsp + nb234_fiyH2], xmm1 movaps [rsp + nb234_fizH2], xmm2 mov rsi, [rbp + nb234_faction] ;# update j water forces from local variables. ;# transpose back first movaps xmm0, [rsp + nb234_fjxO] ;# Ox H1x H2x Mx movaps xmm1, [rsp + nb234_fjyO] ;# Oy H1y H2y My movaps xmm2, [rsp + nb234_fjzO] ;# Oz H1z H2z Mz movaps xmm3, xmm0 movaps xmm4, xmm0 unpcklps xmm3, xmm1 ;# Ox Oy - - shufps xmm4, xmm2, 0x1 ;# h1x - Oz - movaps xmm5, xmm1 movaps xmm6, xmm0 unpcklps xmm5, xmm2 ;# - - H1y H1z unpckhps xmm6, xmm1 ;# h2x h2y - - unpckhps xmm1, xmm2 ;# - - My Mz shufps xmm2, xmm0, 50 ;# (00110010) h2z - Mx - shufps xmm3, xmm4, 36 ;# constant 00100100 ;# Ox Oy Oz H1x shufps xmm5, xmm6, 78 ;# constant 01001110 ;# h1y h1z h2x h2y shufps xmm2, xmm1, 232 ;# constant 11101000 ;# h2z mx my mz movlps xmm0, [rsi + rax*4] movlps xmm1, [rsi + rax*4 + 16] movlps xmm4, [rsi + rax*4 + 32] movhps xmm0, [rsi + rax*4 + 8] movhps xmm1, [rsi + rax*4 + 24] movhps xmm4, [rsi + rax*4 + 40] addps xmm0, xmm3 addps xmm1, xmm5 addps xmm4, xmm2 movlps [rsi + rax*4], xmm0 movlps [rsi + rax*4 + 16], xmm1 movlps [rsi + rax*4 + 32], xmm4 movhps [rsi + rax*4 + 8], xmm0 movhps [rsi + rax*4 + 24], xmm1 movhps [rsi + rax*4 + 40], xmm4 dec dword ptr [rsp + nb234_innerk] jz .nb234_updateouterdata jmp .nb234_single_loop.nb234_updateouterdata: mov ecx, [rsp + nb234_ii3] mov rdi, [rbp + nb234_faction] mov rsi, [rbp + nb234_fshift] mov edx, [rsp + nb234_is3] ;# accumulate Oi forces in xmm0, xmm1, xmm2 movaps xmm0, [rsp + nb234_fixO] movaps xmm1, [rsp + nb234_fiyO] movaps xmm2, [rsp + nb234_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 + nb234_fixH1] movaps xmm1, [rsp + nb234_fiyH1] movaps xmm2, [rsp + nb234_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 + nb234_fixH2] movaps xmm1, [rsp + nb234_fiyH2] movaps xmm2, [rsp + nb234_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 ;# accumulate Mi forces in xmm0, xmm1, xmm2 movaps xmm0, [rsp + nb234_fixM] movaps xmm1, [rsp + nb234_fiyM] movaps xmm2, [rsp + nb234_fizM] 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 + 36] movss xmm4, [rdi + rcx*4 + 40] movss xmm5, [rdi + rcx*4 + 44] subss xmm3, xmm0 subss xmm4, xmm1 subss xmm5, xmm2 movss [rdi + rcx*4 + 36], xmm3 movss [rdi + rcx*4 + 40], xmm4 movss [rdi + rcx*4 + 44], 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 + nb234_n] ;# get group index for i particle mov rdx, [rbp + nb234_gid] ;# base of gid[] mov edx, [rdx + rsi*4] ;# ggid=gid[n] ;# accumulate total potential energy and update it movaps xmm7, [rsp + nb234_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 + nb234_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 + nb234_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 + nb234_Vvdw] addss xmm7, [rax + rdx*4] ;# move back to mem movss [rax + rdx*4], xmm7 ;# finish if last mov ecx, [rsp + nb234_nn1] ;# esi already loaded with n inc esi sub ecx, esi jz .nb234_outerend ;# not last, iterate outer loop once more! mov [rsp + nb234_n], esi jmp .nb234_outer.nb234_outerend: ;# check if more outer neighborlists remain mov ecx, [rsp + nb234_nri] ;# esi already loaded with n above sub ecx, esi jz .nb234_end ;# non-zero, do one more workunit jmp .nb234_threadloop.nb234_end: mov eax, [rsp + nb234_nouter] mov ebx, [rsp + nb234_ninner] mov rcx, [rbp + nb234_outeriter] mov rdx, [rbp + nb234_inneriter] mov [rcx], eax mov [rdx], ebx add rsp, 1912 emms pop r15 pop r14 pop r13 pop r12 pop rbx pop rbp ret .globl nb_kernel234nf_x86_64_sse.globl _nb_kernel234nf_x86_64_ssenb_kernel234nf_x86_64_sse: _nb_kernel234nf_x86_64_sse: ;# Room for return address and rbp (16 bytes).equiv nb234nf_fshift, 16.equiv nb234nf_gid, 24.equiv nb234nf_pos, 32.equiv nb234nf_faction, 40.equiv nb234nf_charge, 48.equiv nb234nf_p_facel, 56.equiv nb234nf_argkrf, 64.equiv nb234nf_argcrf, 72.equiv nb234nf_Vc, 80.equiv nb234nf_type, 88.equiv nb234nf_p_ntype, 96.equiv nb234nf_vdwparam, 104.equiv nb234nf_Vvdw, 112.equiv nb234nf_p_tabscale, 120.equiv nb234nf_VFtab, 128.equiv nb234nf_invsqrta, 136.equiv nb234nf_dvda, 144.equiv nb234nf_p_gbtabscale, 152.equiv nb234nf_GBtab, 160.equiv nb234nf_p_nthreads, 168.equiv nb234nf_count, 176.equiv nb234nf_mtx, 184.equiv nb234nf_outeriter, 192.equiv nb234nf_inneriter, 200.equiv nb234nf_work, 208 ;# stack offsets for local variables ;# bottom of stack is cache-aligned for sse use .equiv nb234nf_ixO, 0.equiv nb234nf_iyO, 16.equiv nb234nf_izO, 32.equiv nb234nf_ixH1, 48.equiv nb234nf_iyH1, 64.equiv nb234nf_izH1, 80.equiv nb234nf_ixH2, 96.equiv nb234nf_iyH2, 112.equiv nb234nf_izH2, 128.equiv nb234nf_ixM, 144.equiv nb234nf_iyM, 160.equiv nb234nf_izM, 176.equiv nb234nf_jxO, 192.equiv nb234nf_jyO, 208.equiv nb234nf_jzO, 224.equiv nb234nf_jxH1, 240.equiv nb234nf_jyH1, 256.equiv nb234nf_jzH1, 272.equiv nb234nf_jxH2, 288.equiv nb234nf_jyH2, 304.equiv nb234nf_jzH2, 320.equiv nb234nf_jxM, 336.equiv nb234nf_jyM, 352.equiv nb234nf_jzM, 368.equiv nb234nf_qqMM, 384.equiv nb234nf_qqMH, 400.equiv nb234nf_qqHH, 416.equiv nb234nf_tsc, 432.equiv nb234nf_c6, 448.equiv nb234nf_c12, 464.equiv nb234nf_vctot, 480.equiv nb234nf_Vvdwtot, 496.equiv nb234nf_half, 512.equiv nb234nf_three, 528.equiv nb234nf_rsqOO, 544.equiv nb234nf_rsqH1H1, 560.equiv nb234nf_rsqH1H2, 576.equiv nb234nf_rsqH1M, 592.equiv nb234nf_rsqH2H1, 608.equiv nb234nf_rsqH2H2, 624.equiv nb234nf_rsqH2M, 640.equiv nb234nf_rsqMH1, 656.equiv nb234nf_rsqMH2, 672.equiv nb234nf_rsqMM, 688.equiv nb234nf_rinvOO, 704.equiv nb234nf_rinvH1H1, 720.equiv nb234nf_rinvH1H2, 736.equiv nb234nf_rinvH1M, 752.equiv nb234nf_rinvH2H1, 768.equiv nb234nf_rinvH2H2, 784.equiv nb234nf_rinvH2M, 800.equiv nb234nf_rinvMH1, 816.equiv nb234nf_rinvMH2, 832.equiv nb234nf_rinvMM, 848.equiv nb234nf_krf, 864.equiv nb234nf_crf, 880.equiv nb234nf_is3, 896.equiv nb234nf_ii3, 900.equiv nb234nf_innerjjnr, 904.equiv nb234nf_nri, 912.equiv nb234nf_iinr, 920.equiv nb234nf_jindex, 928.equiv nb234nf_jjnr, 936.equiv nb234nf_shift, 944.equiv nb234nf_shiftvec, 952.equiv nb234nf_facel, 960.equiv nb234nf_innerk, 968.equiv nb234nf_n, 972.equiv nb234nf_nn1, 976.equiv nb234nf_nouter, 980.equiv nb234nf_ninner, 984 push rbp mov rbp, rsp push rbx emms push r12 push r13 push r14 push r15 sub rsp, 1000 ;# local variable stack space (n*16+8) ;# zero 32-bit iteration counters mov eax, 0 mov [rsp + nb234nf_nouter], eax mov [rsp + nb234nf_ninner], eax mov edi, [rdi] mov [rsp + nb234nf_nri], edi mov [rsp + nb234nf_iinr], rsi mov [rsp + nb234nf_jindex], rdx mov [rsp + nb234nf_jjnr], rcx mov [rsp + nb234nf_shift], r8 mov [rsp + nb234nf_shiftvec], r9 mov rsi, [rbp + nb⌨️ 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?