nb_kernel231_x86_64_sse.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,077 行 · 第 1/4 页
S
2,077 行
add ecx, edi add edx, edi movlps xmm6, [rsi + rax*4] movlps xmm7, [rsi + rcx*4] movhps xmm6, [rsi + rbx*4] movhps xmm7, [rsi + rdx*4] movaps xmm4, xmm6 shufps xmm4, xmm7, 136 ;# constant 10001000 shufps xmm6, xmm7, 221 ;# constant 11011101 movd eax, mm0 movd ebx, mm1 movd ecx, mm2 movd edx, mm3 movaps [rsp + nb231nf_c6], xmm4 movaps [rsp + nb231nf_c12], xmm6 mov rsi, [rbp + nb231nf_pos] ;# base of pos[] lea rax, [rax + rax*2] ;# replace jnr with j3 lea rbx, [rbx + rbx*2] lea rcx, [rcx + rcx*2] ;# replace jnr with j3 lea rdx, [rdx + rdx*2] ;# move four coordinates to xmm0-xmm2 movlps xmm4, [rsi + rax*4] movlps xmm5, [rsi + rcx*4] movss xmm2, [rsi + rax*4 + 8] movss xmm6, [rsi + rcx*4 + 8] movhps xmm4, [rsi + rbx*4] movhps xmm5, [rsi + rdx*4] movss xmm0, [rsi + rbx*4 + 8] movss xmm1, [rsi + rdx*4 + 8] shufps xmm2, xmm0, 0 shufps xmm6, xmm1, 0 movaps xmm0, xmm4 movaps xmm1, xmm4 shufps xmm2, xmm6, 136 ;# constant 10001000 shufps xmm0, xmm5, 136 ;# constant 10001000 shufps xmm1, xmm5, 221 ;# constant 11011101 ;# move ixO-izO to xmm4-xmm6 movaps xmm4, [rsp + nb231nf_ixO] movaps xmm5, [rsp + nb231nf_iyO] movaps xmm6, [rsp + nb231nf_izO] ;# calc dr subps xmm4, xmm0 subps xmm5, xmm1 subps xmm6, xmm2 ;# square it mulps xmm4,xmm4 mulps xmm5,xmm5 mulps xmm6,xmm6 addps xmm4, xmm5 addps xmm4, xmm6 movaps xmm7, xmm4 ;# rsqO in xmm7 ;# move ixH1-izH1 to xmm4-xmm6 movaps xmm4, [rsp + nb231nf_ixH1] movaps xmm5, [rsp + nb231nf_iyH1] movaps xmm6, [rsp + nb231nf_izH1] ;# calc dr subps xmm4, xmm0 subps xmm5, xmm1 subps xmm6, xmm2 ;# square it mulps xmm4,xmm4 mulps xmm5,xmm5 mulps xmm6,xmm6 addps xmm6, xmm5 addps xmm6, xmm4 ;# rsqH1 in xmm6 ;# move ixH2-izH2 to xmm3-xmm5 movaps xmm3, [rsp + nb231nf_ixH2] movaps xmm4, [rsp + nb231nf_iyH2] movaps xmm5, [rsp + nb231nf_izH2] ;# calc dr subps xmm3, xmm0 subps xmm4, xmm1 subps xmm5, xmm2 ;# square it mulps xmm3,xmm3 mulps xmm4,xmm4 mulps xmm5,xmm5 addps xmm5, xmm4 addps xmm5, xmm3 ;# rsqH2 in xmm5, rsqH1 in xmm6, rsqO in xmm7 movaps xmm0, xmm5 movaps xmm1, xmm6 movaps xmm2, xmm7 mulps xmm0, [rsp + nb231nf_krf] mulps xmm1, [rsp + nb231nf_krf] mulps xmm2, [rsp + nb231nf_krf] movaps [rsp + nb231nf_krsqH2], xmm0 movaps [rsp + nb231nf_krsqH1], xmm1 movaps [rsp + nb231nf_krsqO], xmm2 ;# start with rsqO (still in xmm7) - seed to xmm2 rsqrtps xmm2, xmm7 movaps xmm3, xmm2 mulps xmm2, xmm2 movaps xmm4, [rsp + nb231nf_three] mulps xmm2, xmm7 ;# rsq*lu*lu subps xmm4, xmm2 ;# constant 30-rsq*lu*lu mulps xmm4, xmm3 ;# lu*(3-rsq*lu*lu) mulps xmm4, [rsp + nb231nf_half] movaps [rsp + nb231nf_rinvO], xmm4 ;# rinvO in xmm0, rsqO in xmm7 ;# rsqH1 - seed in xmm2 rsqrtps xmm2, xmm6 movaps xmm3, xmm2 mulps xmm2, xmm2 movaps xmm4, [rsp + nb231nf_three] mulps xmm2, xmm6 ;# rsq*lu*lu subps xmm4, xmm2 ;# constant 30-rsq*lu*lu mulps xmm4, xmm3 ;# lu*(3-rsq*lu*lu) mulps xmm4, [rsp + nb231nf_half] movaps [rsp + nb231nf_rinvH1], xmm4 ;# rinvH1 in xmm6 ;# rsqH2 - seed in xmm2 rsqrtps xmm2, xmm5 movaps xmm3, xmm2 mulps xmm2, xmm2 movaps xmm4, [rsp + nb231nf_three] mulps xmm2, xmm5 ;# rsq*lu*lu subps xmm4, xmm2 ;# constant 30-rsq*lu*lu mulps xmm4, xmm3 ;# lu*(3-rsq*lu*lu) mulps xmm4, [rsp + nb231nf_half] movaps [rsp + nb231nf_rinvH2], xmm4 ;# rinvH2 in xmm5 ;# do O table interactions - rsqO in xmm7. mulps xmm7, [rsp + nb231nf_rinvO] mulps xmm7, [rsp + nb231nf_tsc] ;# rtab movhlps xmm5, xmm7 cvttps2pi mm6, xmm7 cvttps2pi mm7, xmm5 ;# mm6/mm7 contain lu indices cvtpi2ps xmm6, mm6 cvtpi2ps xmm5, mm7 movlhps xmm6, xmm5 subps xmm7, xmm6 movaps xmm1, xmm7 ;# xmm1=eps movaps xmm2, xmm1 mulps xmm2, xmm2 ;# xmm2=eps2 pslld mm6, 3 pslld mm7, 3 mov rsi, [rbp + nb231nf_VFtab] movd eax, mm6 psrlq mm6, 32 movd ecx, mm7 psrlq mm7, 32 movd ebx, mm6 movd edx, mm7 ;# dispersion movlps xmm5, [rsi + rax*4] movlps xmm7, [rsi + rcx*4] movhps xmm5, [rsi + rbx*4] movhps xmm7, [rsi + rdx*4] ;# got half dispersion table movaps xmm4, xmm5 shufps xmm4, xmm7, 136 ;# constant 10001000 shufps xmm5, xmm7, 221 ;# constant 11011101 movlps xmm7, [rsi + rax*4 + 8] movlps xmm3, [rsi + rcx*4 + 8] movhps xmm7, [rsi + rbx*4 + 8] movhps xmm3, [rsi + rdx*4 + 8] ;# other half of dispersion table movaps xmm6, xmm7 shufps xmm6, xmm3, 136 ;# constant 10001000 shufps xmm7, xmm3, 221 ;# constant 11011101 ;# dispersion table ready, in xmm4-xmm7 mulps xmm6, xmm1 ;# xmm6=Geps mulps xmm7, xmm2 ;# xmm7=Heps2 addps xmm5, xmm6 addps xmm5, xmm7 ;# xmm5=Fp mulps xmm5, xmm1 ;# xmm5=eps*Fp addps xmm5, xmm4 ;# xmm5=VV movaps xmm4, [rsp + nb231nf_c6] mulps xmm5, xmm4 ;# Vvdw6 addps xmm5, [rsp + nb231nf_Vvdwtot] movaps [rsp + nb231nf_Vvdwtot], xmm5 ;# repulsion movlps xmm5, [rsi + rax*4 + 16] movlps xmm7, [rsi + rcx*4 + 16] movhps xmm5, [rsi + rbx*4 + 16] movhps xmm7, [rsi + rdx*4 + 16] ;# got half repulsion table movaps xmm4, xmm5 shufps xmm4, xmm7, 136 ;# constant 10001000 shufps xmm5, xmm7, 221 ;# constant 11011101 movlps xmm7, [rsi + rax*4 + 24] movlps xmm3, [rsi + rcx*4 + 24] movhps xmm7, [rsi + rbx*4 + 24] movhps xmm3, [rsi + rdx*4 + 24] ;# other half of repulsion table movaps xmm6, xmm7 shufps xmm6, xmm3, 136 ;# constant 10001000 shufps xmm7, xmm3, 221 ;# constant 11011101 ;# table ready, in xmm4-xmm7 mulps xmm6, xmm1 ;# xmm6=Geps mulps xmm7, xmm2 ;# xmm7=Heps2 addps xmm5, xmm6 addps xmm5, xmm7 ;# xmm5=Fp mulps xmm5, xmm1 ;# xmm5=eps*Fp addps xmm5, xmm4 ;# xmm5=VV movaps xmm4, [rsp + nb231nf_c12] mulps xmm5, xmm4 ;# Vvdw12 addps xmm5, [rsp + nb231nf_Vvdwtot] movaps [rsp + nb231nf_Vvdwtot], xmm5 movaps xmm2, [rsp + nb231nf_rinvO] movaps xmm1, [rsp + nb231nf_krsqO] addps xmm2, xmm1 ;# rinv+krsq subps xmm2, [rsp + nb231nf_crf] ;# rinv+krsq-crf mulps xmm2, [rsp + nb231nf_qqO] addps xmm2, [rsp + nb231nf_vctot] movaps [rsp + nb231nf_vctot], xmm2 ;# H1 interactions movaps xmm2, [rsp + nb231nf_rinvH1] movaps xmm1, [rsp + nb231nf_krsqH1] addps xmm2, xmm1 ;# rinv+krsq subps xmm2, [rsp + nb231nf_crf] ;# rinv+krsq-crf mulps xmm2, [rsp + nb231nf_qqH] addps xmm2, [rsp + nb231nf_vctot] movaps [rsp + nb231nf_vctot], xmm2 ;# H2 interactions movaps xmm2, [rsp + nb231nf_rinvH2] movaps xmm1, [rsp + nb231nf_krsqH2] addps xmm2, xmm1 ;# rinv+krsq subps xmm2, [rsp + nb231nf_crf] ;# rinv+krsq-crf mulps xmm2, [rsp + nb231nf_qqH] addps xmm2, [rsp + nb231nf_vctot] movaps [rsp + nb231nf_vctot], xmm2 ;# should we do one more iteration? sub dword ptr [rsp + nb231nf_innerk], 4 jl .nb231nf_odd_inner jmp .nb231nf_unroll_loop.nb231nf_odd_inner: add dword ptr [rsp + nb231nf_innerk], 4 jnz .nb231nf_odd_loop jmp .nb231nf_updateouterdata.nb231nf_odd_loop: mov rdx, [rsp + nb231nf_innerjjnr] ;# pointer to jjnr[k] mov eax, [rdx] add qword ptr [rsp + nb231nf_innerjjnr], 4 xorps xmm4, xmm4 movss xmm4, [rsp + nb231nf_iqO] mov rsi, [rbp + nb231nf_charge] movhps xmm4, [rsp + nb231nf_iqH] movss xmm3, [rsi + rax*4] ;# charge in xmm3 shufps xmm3, xmm3, 0 mulps xmm3, xmm4 movaps [rsp + nb231nf_qqO], xmm3 ;# use oxygen qq for storage xorps xmm6, xmm6 mov rsi, [rbp + nb231nf_type] mov ebx, [rsi + rax*4] mov rsi, [rbp + nb231nf_vdwparam] shl ebx, 1 add ebx, [rsp + nb231nf_ntia] movlps xmm6, [rsi + rbx*4] movaps xmm7, xmm6 shufps xmm6, xmm6, 252 ;# constant 11111100 shufps xmm7, xmm7, 253 ;# constant 11111101 movaps [rsp + nb231nf_c6], xmm6 movaps [rsp + nb231nf_c12], xmm7 mov rsi, [rbp + nb231nf_pos] lea rax, [rax + rax*2] ;# move j coords to xmm0-xmm2 movss xmm0, [rsi + rax*4] movss xmm1, [rsi + rax*4 + 4] movss xmm2, [rsi + rax*4 + 8] shufps xmm0, xmm0, 0 shufps xmm1, xmm1, 0 shufps xmm2, xmm2, 0 movss xmm3, [rsp + nb231nf_ixO] movss xmm4, [rsp + nb231nf_iyO] movss xmm5, [rsp + nb231nf_izO] movlps xmm6, [rsp + nb231nf_ixH1] movlps xmm7, [rsp + nb231nf_ixH2] unpcklps xmm6, xmm7 movlhps xmm3, xmm6 movlps xmm6, [rsp + nb231nf_iyH1] movlps xmm7, [rsp + nb231nf_iyH2] unpcklps xmm6, xmm7 movlhps xmm4, xmm6 movlps xmm6, [rsp + nb231nf_izH1] movlps xmm7, [rsp + nb231nf_izH2] unpcklps xmm6, xmm7 movlhps xmm5, xmm6 subps xmm3, xmm0 subps xmm4, xmm1 subps xmm5, xmm2 mulps xmm3, xmm3 mulps xmm4, xmm4 mulps xmm5, xmm5 addps xmm4, xmm3 addps xmm4, xmm5 ;# rsq in xmm4 movaps xmm0, xmm4 mulps xmm0, [rsp + nb231nf_krf] movaps [rsp + nb231nf_krsqO], xmm0 rsqrtps xmm5, xmm4 ;# lookup seed in xmm5 movaps xmm2, xmm5 mulps xmm5, xmm5 movaps xmm1, [rsp + nb231nf_three] mulps xmm5, xmm4 ;# rsq*lu*lu movaps xmm0, [rsp + nb231nf_half] subps xmm1, xmm5 ;# constant 30-rsq*lu*lu mulps xmm1, xmm2 mulps xmm0, xmm1 ;# xmm0=rinv ;# a little trick to avoid NaNs: ;# positions 0,2,and 3 are valid, but not 1. ;# If it contains NaN it doesnt help to mult by 0, ;# So we shuffle it and copy pos 0 to pos1! shufps xmm0, xmm0, 224 ;# constant 11100000 ;# rsq still in xmm4, rinv in xmm0. mulps xmm4, xmm0 mulps xmm4, [rsp + nb231nf_tsc] ;# rtab cvttps2pi mm6, xmm4 cvtpi2ps xmm6, mm6 subss xmm4, xmm6 movss xmm1, xmm4 ;# xmm1=eps movss xmm2, xmm1 mulss xmm2, xmm2 ;# xmm2=eps2 pslld mm6, 3 mov rsi, [rbp + nb231nf_VFtab] movd eax, mm6 ;# dispersion movlps xmm5, [rsi + rax*4] movaps xmm4, xmm5 shufps xmm4, xmm7, 136 ;# constant 10001000 shufps xmm5, xmm7, 221 ;# constant 11011101 movlps xmm7, [rsi + rax*4 + 8] movaps xmm6, xmm7 shufps xmm6, xmm3, 136 ;# constant 10001000 shufps xmm7, xmm3, 221 ;# constant 11011101 ;# dispersion table ready, in xmm4-xmm7 mulss xmm6, xmm1 ;# xmm6=Geps mulss xmm7, xmm2 ;# xmm7=Heps2 addss xmm5, xmm6 addss xmm5, xmm7 ;# xmm5=Fp mulss xmm5, xmm1 ;# xmm5=eps*Fp addss xmm5, xmm4 ;# xmm5=VV movss xmm4, [rsp + nb231nf_c6] mulss xmm5, xmm4 ;# Vvdw6 addss xmm5, [rsp + nb231nf_Vvdwtot] movss [rsp + nb231nf_Vvdwtot], xmm5 ;# repulsion movlps xmm5, [rsi + rax*4 + 16] movaps xmm4, xmm5 shufps xmm4, xmm7, 136 ;# constant 10001000 shufps xmm5, xmm7, 221 ;# constant 11011101 movlps xmm7, [rsi + rax*4 + 24] movaps xmm6, xmm7 shufps xmm6, xmm3, 136 ;# constant 10001000 shufps xmm7, xmm3, 221 ;# constant 11011101 ;# table ready, in xmm4-xmm7 mulss xmm6, xmm1 ;# xmm6=Geps mulss xmm7, xmm2 ;# xmm7=Heps2 addss xmm5, xmm6 addss xmm5, xmm7 ;# xmm5=Fp mulss xmm5, xmm1 ;# xmm5=eps*Fp addss xmm5, xmm4 ;# xmm5=VV movss xmm4, [rsp + nb231nf_c12] mulss xmm5, xmm4 ;# Vvdw12 addss xmm5, [rsp + nb231nf_Vvdwtot] movss [rsp + nb231nf_Vvdwtot], xmm5 movaps xmm1, xmm0 ;# xmm1=rinv movaps xmm4, xmm0 movaps xmm3, [rsp + nb231nf_krsqO] addps xmm0, xmm3 ;# xmm0=rinv+ krsq subps xmm0, [rsp + nb231nf_crf] ;# xmm0=rinv+ krsq-crf mulps xmm0, [rsp + nb231nf_qqO] ;# xmm0=vcoul addps xmm0, [rsp + nb231nf_vctot] movaps [rsp + nb231nf_vctot], xmm0 dec dword ptr [rsp + nb231nf_innerk] jz .nb231nf_updateouterdata jmp .nb231nf_odd_loop.nb231nf_updateouterdata: ;# get n from stack mov esi, [rsp + nb231nf_n] ;# get group index for i particle mov rdx, [rbp + nb231nf_gid] ;# base of gid[] mov edx, [rdx + rsi*4] ;# ggid=gid[n] ;# accumulate total potential energy and update it movaps xmm7, [rsp + nb231nf_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 + nb231nf_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 + nb231nf_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 + nb231nf_Vvdw] addss xmm7, [rax + rdx*4] ;# move back to mem movss [rax + rdx*4], xmm7 ;# finish if last mov ecx, [rsp + nb231nf_nn1] ;# esi already loaded with n inc esi sub ecx, esi jz .nb231nf_outerend ;# not last, iterate outer loop once more! mov [rsp + nb231nf_n], esi jmp .nb231nf_outer.nb231nf_outerend: ;# check if more outer neighborlists remain mov ecx, [rsp + nb231nf_nri] ;# esi already loaded with n above sub ecx, esi jz .nb231nf_end ;# non-zero, do one more workunit jmp .nb231nf_threadloop.nb231nf_end: mov eax, [rsp + nb231nf_nouter] mov ebx, [rsp + nb231nf_ninner] mov rcx, [rbp + nb231nf_outeriter] mov rdx, [rbp + nb231nf_inneriter] mov [rcx], eax mov [rdx], ebx add rsp, 552 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?