nb_kernel200_x86_64_sse2.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 1,003 行 · 第 1/2 页
S
1,003 行
movsd [rdi + rcx*8 + 8], xmm4 movsd [rdi + rcx*8 + 16], xmm5 ;# increment fshift force movsd xmm3, [rsi + rdx*8] movsd xmm4, [rsi + rdx*8 + 8] movsd xmm5, [rsi + rdx*8 + 16] subsd xmm3, xmm13 subsd xmm4, xmm14 subsd xmm5, xmm15 movsd [rsi + rdx*8], xmm3 movsd [rsi + rdx*8 + 8], xmm4 movsd [rsi + rdx*8 + 16], xmm5 ;# get n from stack mov esi, [rsp + nb200_n] ;# get group index for i particle mov rdx, [rbp + nb200_gid] ;# base of gid[] mov edx, [rdx + rsi*4] ;# ggid=gid[n] ;# accumulate total coulomb energy and update it movhlps xmm6, xmm12 addsd xmm12, xmm6 ;# low xmm12 have the sum now ;# add earlier value from mem mov rax, [rbp + nb200_Vc] addsd xmm12, [rax + rdx*8] ;# move back to mem movsd [rax + rdx*8], xmm12 ;# finish if last mov ecx, [rsp + nb200_nn1] ;# esi already loaded with n inc esi sub ecx, esi jz .nb200_outerend ;# not last, iterate outer loop once more! mov [rsp + nb200_n], esi jmp .nb200_outer.nb200_outerend: ;# check if more outer neighborlists remain mov ecx, [rsp + nb200_nri] ;# esi already loaded with n above sub ecx, esi jz .nb200_end ;# non-zero, do one more workunit jmp .nb200_threadloop.nb200_end: mov eax, [rsp + nb200_nouter] mov ebx, [rsp + nb200_ninner] mov rcx, [rbp + nb200_outeriter] mov rdx, [rbp + nb200_inneriter] mov [rcx], eax mov [rdx], ebx add rsp, 360 emms pop r15 pop r14 pop r13 pop r12 pop rbx pop rbp ret.globl nb_kernel200nf_x86_64_sse2.globl _nb_kernel200nf_x86_64_sse2nb_kernel200nf_x86_64_sse2: _nb_kernel200nf_x86_64_sse2: ;# Room for return address and rbp (16 bytes).equiv nb200nf_fshift, 16.equiv nb200nf_gid, 24.equiv nb200nf_pos, 32.equiv nb200nf_faction, 40.equiv nb200nf_charge, 48.equiv nb200nf_p_facel, 56.equiv nb200nf_argkrf, 64.equiv nb200nf_argcrf, 72.equiv nb200nf_Vc, 80.equiv nb200nf_type, 88.equiv nb200nf_p_ntype, 96.equiv nb200nf_vdwparam, 104.equiv nb200nf_Vvdw, 112.equiv nb200nf_p_tabscale, 120.equiv nb200nf_VFtab, 128.equiv nb200nf_invsqrta, 136.equiv nb200nf_dvda, 144.equiv nb200nf_p_gbtabscale, 152.equiv nb200nf_GBtab, 160.equiv nb200nf_p_nthreads, 168.equiv nb200nf_count, 176.equiv nb200nf_mtx, 184.equiv nb200nf_outeriter, 192.equiv nb200nf_inneriter, 200.equiv nb208nf_work, 200 ;# stack offsets for local variables ;# bottom of stack is cache-aligned for sse use .equiv nb200nf_ix, 0.equiv nb200nf_iy, 16.equiv nb200nf_iz, 32.equiv nb200nf_iq, 48.equiv nb200nf_vctot, 64.equiv nb200nf_half, 80.equiv nb200nf_three, 96.equiv nb200nf_krf, 112.equiv nb200nf_crf, 128.equiv nb200nf_is3, 144.equiv nb200nf_ii3, 148.equiv nb200nf_nri, 152.equiv nb200nf_iinr, 160.equiv nb200nf_jindex, 168.equiv nb200nf_jjnr, 176.equiv nb200nf_shift, 184.equiv nb200nf_shiftvec, 192.equiv nb200nf_facel, 200.equiv nb200nf_innerjjnr, 208.equiv nb200nf_innerk, 216.equiv nb200nf_n, 220.equiv nb200nf_nn1, 224.equiv nb200nf_nouter, 228.equiv nb200nf_ninner, 232 push rbp mov rbp, rsp push rbx emms push r12 push r13 push r14 push r15 sub rsp, 248 ;# local variable stack space (n*16+8) ;# zero 32-bit iteration counters mov eax, 0 mov [rsp + nb200nf_nouter], eax mov [rsp + nb200nf_ninner], eax mov edi, [rdi] mov [rsp + nb200nf_nri], edi mov [rsp + nb200nf_iinr], rsi mov [rsp + nb200nf_jindex], rdx mov [rsp + nb200nf_jjnr], rcx mov [rsp + nb200nf_shift], r8 mov [rsp + nb200nf_shiftvec], r9 mov rsi, [rbp + nb200nf_p_facel] movsd xmm0, [rsi] movsd [rsp + nb200nf_facel], xmm0 mov rsi, [rbp + nb200nf_argkrf] mov rdi, [rbp + nb200nf_argcrf] movsd xmm1, [rsi] movsd xmm2, [rdi] shufpd xmm1, xmm1, 0 shufpd xmm2, xmm2, 0 movapd [rsp + nb200nf_krf], xmm1 movapd [rsp + nb200nf_crf], xmm2 ;# create constant floating-point factors on stack mov eax, 0x00000000 ;# lower half of double half IEEE (hex) mov ebx, 0x3fe00000 mov [rsp + nb200nf_half], eax mov [rsp + nb200nf_half+4], ebx movsd xmm1, [rsp + nb200nf_half] shufpd xmm1, xmm1, 0 ;# splat to all elements movapd xmm3, xmm1 addpd xmm3, xmm3 ;# one movapd xmm2, xmm3 addpd xmm2, xmm2 ;# two addpd xmm3, xmm2 ;# three movapd [rsp + nb200nf_half], xmm1 movapd [rsp + nb200nf_three], xmm3.nb200nf_threadloop: mov rsi, [rbp + nb200nf_count] ;# pointer to sync counter mov eax, [rsi].nb200nf_spinlock: mov ebx, eax ;# ebx=*count=nn0 add ebx, 1 ;# ebx=nn1=nn0+10 lock cmpxchg [esi], ebx ;# write nn1 to *counter, ;# if it hasnt changed. ;# or reread *counter to eax. pause ;# -> better p4 performance jnz .nb200nf_spinlock ;# if(nn1>nri) nn1=nri mov ecx, [rsp + nb200nf_nri] mov edx, ecx sub ecx, ebx cmovle ebx, edx ;# if(nn1>nri) nn1=nri ;# Cleared the spinlock if we got here. ;# eax contains nn0, ebx contains nn1. mov [rsp + nb200nf_n], eax mov [rsp + nb200nf_nn1], ebx sub ebx, eax ;# calc number of outer lists mov esi, eax ;# copy n to esi jg .nb200nf_outerstart jmp .nb200nf_end.nb200nf_outerstart: ;# ebx contains number of outer iterations add ebx, [rsp + nb200nf_nouter] mov [rsp + nb200nf_nouter], ebx.nb200nf_outer: mov rax, [rsp + nb200nf_shift] ;# rax = pointer into shift[] mov ebx, [rax+rsi*4] ;# rbx=shift[n] lea rbx, [rbx + rbx*2] ;# rbx=3*is mov rax, [rsp + nb200nf_shiftvec] ;# rax = base of shiftvec[] movsd xmm0, [rax + rbx*8] movsd xmm1, [rax + rbx*8 + 8] movsd xmm2, [rax + rbx*8 + 16] mov rcx, [rsp + nb200nf_iinr] ;# rcx = pointer into iinr[] mov ebx, [rcx+rsi*4] ;# ebx =ii mov rdx, [rbp + nb200nf_charge] movsd xmm3, [rdx + rbx*8] mulsd xmm3, [rsp + nb200nf_facel] shufpd xmm3, xmm3, 0 lea rbx, [rbx + rbx*2] ;# rbx = 3*ii=ii3 mov rax, [rbp + nb200nf_pos] ;# rax = base of pos[] addsd xmm0, [rax + rbx*8] addsd xmm1, [rax + rbx*8 + 8] addsd xmm2, [rax + rbx*8 + 16] movapd [rsp + nb200nf_iq], xmm3 shufpd xmm0, xmm0, 0 shufpd xmm1, xmm1, 0 shufpd xmm2, xmm2, 0 movapd [rsp + nb200nf_ix], xmm0 movapd [rsp + nb200nf_iy], xmm1 movapd [rsp + nb200nf_iz], xmm2 mov [rsp + nb200nf_ii3], ebx ;# clear vctot xorpd xmm4, xmm4 movapd [rsp + nb200nf_vctot], xmm4 mov rax, [rsp + nb200nf_jindex] mov ecx, [rax + rsi*4] ;# jindex[n] mov edx, [rax + rsi*4 + 4] ;# jindex[n+1] sub edx, ecx ;# number of innerloop atoms mov rsi, [rbp + nb200nf_pos] mov rax, [rsp + nb200nf_jjnr] shl ecx, 2 add rax, rcx mov [rsp + nb200nf_innerjjnr], rax ;# pointer to jjnr[nj0] mov ecx, edx sub edx, 2 add ecx, [rsp + nb200nf_ninner] mov [rsp + nb200nf_ninner], ecx add edx, 0 mov [rsp + nb200nf_innerk], edx ;# number of innerloop atoms jge .nb200nf_unroll_loop jmp .nb200nf_checksingle.nb200nf_unroll_loop: ;# twice unrolled innerloop here mov rdx, [rsp + nb200nf_innerjjnr] ;# pointer to jjnr[k] mov eax, [rdx] mov ebx, [rdx + 4] add qword ptr [rsp + nb200nf_innerjjnr], 8 ;# advance pointer (unrolled 2) mov rsi, [rbp + nb200nf_charge] ;# base of charge[] movlpd xmm3, [rsi + rax*8] movhpd xmm3, [rsi + rbx*8] movapd xmm5, [rsp + nb200nf_iq] mulpd xmm3, xmm5 ;# qq lea rax, [rax + rax*2] ;# replace jnr with j3 lea rbx, [rbx + rbx*2] mov rsi, [rbp + nb200nf_pos] ;# base of pos[] ;# move two coordinates to xmm0-xmm2 movlpd xmm0, [rsi + rax*8] movlpd xmm1, [rsi + rax*8 + 8] movlpd xmm2, [rsi + rax*8 + 16] movhpd xmm0, [rsi + rbx*8] movhpd xmm1, [rsi + rbx*8 + 8] movhpd xmm2, [rsi + rbx*8 + 16] ;# move ix-iz to xmm4-xmm6 movapd xmm4, [rsp + nb200nf_ix] movapd xmm5, [rsp + nb200nf_iy] movapd xmm6, [rsp + nb200nf_iz] ;# calc dr subpd xmm4, xmm0 subpd xmm5, xmm1 subpd xmm6, xmm2 ;# square it mulpd xmm4,xmm4 mulpd xmm5,xmm5 mulpd xmm6,xmm6 addpd xmm4, xmm5 addpd xmm4, xmm6 ;# rsq in xmm4 cvtpd2ps xmm5, xmm4 rsqrtps xmm5, xmm5 cvtps2pd xmm2, xmm5 ;# lu in low xmm2 movapd xmm7, [rsp + nb200nf_krf] ;# lookup seed in xmm2 movapd xmm5, xmm2 ;# copy of lu mulpd xmm2, xmm2 ;# lu*lu movapd xmm1, [rsp + nb200nf_three] mulpd xmm7, xmm4 ;# krsq mulpd xmm2, xmm4 ;# rsq*lu*lu movapd xmm0, [rsp + nb200nf_half] subpd xmm1, xmm2 ;# 30-rsq*lu*lu mulpd xmm1, xmm5 mulpd xmm1, xmm0 ;# xmm0=iter1 of rinv (new lu) movapd xmm5, xmm1 ;# copy of lu mulpd xmm1, xmm1 ;# lu*lu movapd xmm2, [rsp + nb200nf_three] mulpd xmm1, xmm4 ;# rsq*lu*lu movapd xmm0, [rsp + nb200nf_half] subpd xmm2, xmm1 ;# 30-rsq*lu*lu mulpd xmm2, xmm5 mulpd xmm0, xmm2 ;# xmm0=rinv movapd xmm4, xmm0 mulpd xmm4, xmm4 ;# xmm4=rinvsq movapd xmm6, xmm0 addpd xmm6, xmm7 ;# xmm6=rinv+ krsq movapd xmm1, xmm4 subpd xmm6, [rsp + nb200nf_crf] mulpd xmm6, xmm3 ;# xmm6=vcoul=qq*(rinv+ krsq) addpd xmm6, [rsp + nb200nf_vctot] movapd [rsp + nb200nf_vctot], xmm6 ;# should we do one more iteration? sub dword ptr [rsp + nb200nf_innerk], 2 jl .nb200nf_checksingle jmp .nb200nf_unroll_loop.nb200nf_checksingle: mov edx, [rsp + nb200nf_innerk] and edx, 1 jnz .nb200nf_dosingle jmp .nb200nf_updateouterdata.nb200nf_dosingle: mov rsi, [rbp + nb200nf_charge] mov rdi, [rbp + nb200nf_pos] mov rcx, [rsp + nb200nf_innerjjnr] xorpd xmm3, xmm3 mov eax, [rcx] movlpd xmm3, [rsi + rax*8] movapd xmm5, [rsp + nb200nf_iq] mulpd xmm3, xmm5 ;# qq mov rsi, [rbp + nb200nf_pos] ;# base of pos[] lea rax, [rax + rax*2] ;# replace jnr with j3 ;# move two coordinates to xmm0-xmm2 movlpd xmm0, [rsi + rax*8] movlpd xmm1, [rsi + rax*8 + 8] movlpd xmm2, [rsi + rax*8 + 16] ;# move ix-iz to xmm4-xmm6 movapd xmm4, [rsp + nb200nf_ix] movapd xmm5, [rsp + nb200nf_iy] movapd xmm6, [rsp + nb200nf_iz] ;# calc dr subsd xmm4, xmm0 subsd xmm5, xmm1 subsd xmm6, xmm2 ;# square it mulsd xmm4,xmm4 mulsd xmm5,xmm5 mulsd xmm6,xmm6 addsd xmm4, xmm5 addsd xmm4, xmm6 ;# rsq in xmm4 cvtsd2ss xmm5, xmm4 rsqrtss xmm5, xmm5 cvtss2sd xmm2, xmm5 ;# lu in low xmm2 movapd xmm7, [rsp + nb200nf_krf] ;# lookup seed in xmm2 movapd xmm5, xmm2 ;# copy of lu mulsd xmm2, xmm2 ;# lu*lu movapd xmm1, [rsp + nb200nf_three] mulsd xmm7, xmm4 ;# krsq mulsd xmm2, xmm4 ;# rsq*lu*lu movapd xmm0, [rsp + nb200nf_half] subsd xmm1, xmm2 ;# 30-rsq*lu*lu mulsd xmm1, xmm5 mulsd xmm1, xmm0 ;# xmm0=iter1 of rinv (new lu) movapd xmm5, xmm1 ;# copy of lu mulsd xmm1, xmm1 ;# lu*lu movapd xmm2, [rsp + nb200nf_three] mulsd xmm1, xmm4 ;# rsq*lu*lu movapd xmm0, [rsp + nb200nf_half] subsd xmm2, xmm1 ;# 30-rsq*lu*lu mulsd xmm2, xmm5 mulsd xmm0, xmm2 ;# xmm0=rinv movapd xmm4, xmm0 mulsd xmm4, xmm4 ;# xmm4=rinvsq movapd xmm6, xmm0 addsd xmm6, xmm7 ;# xmm6=rinv+ krsq movapd xmm1, xmm4 subsd xmm6, [rsp + nb200nf_crf] mulsd xmm1, xmm4 mulsd xmm1, xmm4 ;# xmm1=rinvsix movapd xmm2, xmm1 mulsd xmm2, xmm2 ;# xmm2=rinvtwelve mulsd xmm6, xmm3 ;# xmm6=vcoul=qq*(rinv+ krsq) addsd xmm6, [rsp + nb200nf_vctot] movlpd [rsp + nb200nf_vctot], xmm6.nb200nf_updateouterdata: ;# get n from stack mov esi, [rsp + nb200nf_n] ;# get group index for i particle mov rdx, [rbp + nb200nf_gid] ;# base of gid[] mov edx, [rdx + rsi*4] ;# ggid=gid[n] ;# accumulate total potential energy and update it movapd xmm7, [rsp + nb200nf_vctot] ;# accumulate movhlps xmm6, xmm7 addsd xmm7, xmm6 ;# low xmm7 has the sum now ;# add earlier value from mem mov rax, [rbp + nb200nf_Vc] addsd xmm7, [rax + rdx*8] ;# move back to mem movsd [rax + rdx*8], xmm7 ;# finish if last mov ecx, [rsp + nb200nf_nn1] ;# esi already loaded with n inc esi sub ecx, esi jz .nb200nf_outerend ;# not last, iterate outer loop once more! mov [rsp + nb200nf_n], esi jmp .nb200nf_outer.nb200nf_outerend: ;# check if more outer neighborlists remain mov ecx, [rsp + nb200nf_nri] ;# esi already loaded with n above sub ecx, esi jz .nb200nf_end ;# non-zero, do one more workunit jmp .nb200nf_threadloop.nb200nf_end: mov eax, [rsp + nb200nf_nouter] mov ebx, [rsp + nb200nf_ninner] mov rcx, [rbp + nb200nf_outeriter] mov rdx, [rbp + nb200nf_inneriter] mov [rcx], eax mov [rdx], ebx add rsp, 248 emms pop r15 pop r14 pop r13 pop r12 pop rbx pop rbp ret⌨️ 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?