nb_kernel130_x86_64_sse.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 1,919 行 · 第 1/4 页
S
1,919 行
movhlps xmm5, xmm4 cvttps2pi mm6, xmm4 cvttps2pi mm7, xmm5 ;# mm6/mm7 contain lu indices cvtpi2ps xmm6, mm6 cvtpi2ps xmm5, mm7 movlhps xmm6, xmm5 subps xmm4, xmm6 movaps xmm1, xmm4 ;# xmm1=eps movaps xmm2, xmm1 mulps xmm2, xmm2 ;# xmm2=eps2 pslld mm6, 3 pslld mm7, 3 mov rsi, [rbp + nb130nf_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 + nb130nf_c6] mulps xmm5, xmm4 ;# Vvdw6 ;# Update Vvdwtot directly addps xmm5, [rsp + nb130nf_Vvdwtot] movaps [rsp + nb130nf_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 + nb130nf_c12] mulps xmm5, xmm4 ;# Vvdw12 addps xmm5, [rsp + nb130nf_Vvdwtot] movaps [rsp + nb130nf_Vvdwtot], xmm5 ;# should we do one more iteration? sub dword ptr [rsp + nb130nf_innerk], 4 jl .nb130nf_finish_inner jmp .nb130nf_unroll_loop.nb130nf_finish_inner: ;# check if at least two particles remain add dword ptr [rsp + nb130nf_innerk], 4 mov edx, [rsp + nb130nf_innerk] and edx, 2 jnz .nb130nf_dopair jmp .nb130nf_checksingle.nb130nf_dopair: mov rsi, [rbp + nb130nf_charge] mov rcx, [rsp + nb130nf_innerjjnr] mov eax, [rcx] mov ebx, [rcx + 4] add qword ptr [rsp + nb130nf_innerjjnr], 8 xorps xmm3, xmm3 movss xmm3, [rsi + rax*4] movss xmm6, [rsi + rbx*4] shufps xmm3, xmm6, 12 ;# constant 00001100 shufps xmm3, xmm3, 88 ;# constant 01011000 ;# xmm3(0,1) has the charges mov rsi, [rbp + nb130nf_type] mov ecx, eax mov edx, ebx mov ecx, [rsi + rcx*4] mov edx, [rsi + rdx*4] mov rsi, [rbp + nb130nf_vdwparam] shl ecx, 1 shl edx, 1 mov edi, [rsp + nb130nf_ntia] add ecx, edi add edx, edi movlps xmm6, [rsi + rcx*4] movhps xmm6, [rsi + rdx*4] mov rdi, [rbp + nb130nf_pos] xorps xmm7,xmm7 movaps xmm4, xmm6 shufps xmm4, xmm4, 8 ;# constant 00001000 shufps xmm6, xmm6, 13 ;# constant 00001101 movlhps xmm4, xmm7 movlhps xmm6, xmm7 movaps [rsp + nb130nf_c6], xmm4 movaps [rsp + nb130nf_c12], xmm6 lea rax, [rax + rax*2] lea rbx, [rbx + rbx*2] ;# move coordinates to xmm0-xmm2 movlps xmm1, [rdi + rax*4] movss xmm2, [rdi + rax*4 + 8] movhps xmm1, [rdi + rbx*4] movss xmm0, [rdi + rbx*4 + 8] mulps xmm3, [rsp + nb130nf_iq] movlhps xmm3, xmm7 shufps xmm2, xmm0, 0 movaps xmm0, xmm1 shufps xmm2, xmm2, 136 ;# constant 10001000 shufps xmm0, xmm0, 136 ;# constant 10001000 shufps xmm1, xmm1, 221 ;# constant 11011101 ;# move ix-iz to xmm4-xmm6 xorps xmm7, xmm7 movaps xmm4, [rsp + nb130nf_ix] movaps xmm5, [rsp + nb130nf_iy] movaps xmm6, [rsp + nb130nf_iz] ;# 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 ;# rsq in xmm4 rsqrtps xmm5, xmm4 ;# lookup seed in xmm5 movaps xmm2, xmm5 mulps xmm5, xmm5 movaps xmm1, [rsp + nb130nf_three] mulps xmm5, xmm4 ;# rsq*lu*lu movaps xmm0, [rsp + nb130nf_half] subps xmm1, xmm5 ;# constant 30-rsq*lu*lu mulps xmm1, xmm2 mulps xmm0, xmm1 ;# xmm0=rinv movaps xmm1, xmm0 mulps xmm3, xmm0 addps xmm3, [rsp + nb130nf_vctot] movaps [rsp + nb130nf_vctot], xmm3 ;# LJ table mulps xmm4, xmm1 ;# r mulps xmm4, [rsp + nb130nf_tsc] ;# rtab movaps xmm0, xmm1 ;# copy of rinv cvttps2pi mm6, xmm4 cvtpi2ps xmm6, mm6 subps xmm4, xmm6 movaps xmm1, xmm4 ;# xmm1=eps movaps xmm2, xmm1 mulps xmm2, xmm2 ;# xmm2=eps2 pslld mm6, 3 mov rsi, [rbp + nb130nf_VFtab] movd eax, mm6 psrlq mm6, 32 movd ebx, mm6 ;# dispersion movlps xmm5, [rsi + rax*4] movhps xmm5, [rsi + rbx*4] movaps xmm4, xmm5 shufps xmm4, xmm7, 136 ;# constant 10001000 shufps xmm5, xmm7, 221 ;# constant 11011101 movlps xmm7, [rsi + rax*4 + 8] movhps xmm7, [rsi + rbx*4 + 8] 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 + nb130nf_c6] mulps xmm5, xmm4 ;# Vvdw6 ;# Update Vvdwtot directly addps xmm5, [rsp + nb130nf_Vvdwtot] movaps [rsp + nb130nf_Vvdwtot], xmm5 ;# repulsion movlps xmm5, [rsi + rax*4 + 16] movhps xmm5, [rsi + rbx*4 + 16] movaps xmm4, xmm5 shufps xmm4, xmm7, 136 ;# constant 10001000 shufps xmm5, xmm7, 221 ;# constant 11011101 movlps xmm7, [rsi + rax*4 + 24] movhps xmm7, [rsi + rbx*4 + 24] 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 + nb130nf_c12] mulps xmm5, xmm4 ;# Vvdw12 addps xmm5, [rsp + nb130nf_Vvdwtot] movaps [rsp + nb130nf_Vvdwtot], xmm5.nb130nf_checksingle: mov edx, [rsp + nb130nf_innerk] and edx, 1 jnz .nb130nf_dosingle jmp .nb130nf_updateouterdata.nb130nf_dosingle: mov rsi, [rbp + nb130nf_charge] mov rdi, [rbp + nb130nf_pos] mov rcx, [rsp + nb130nf_innerjjnr] xorps xmm3, xmm3 mov eax, [rcx] movss xmm3, [rsi + rax*4] ;# xmm3(0) has the charge mov rsi, [rbp + nb130nf_type] mov ecx, eax mov ecx, [rsi + rcx*4] mov rsi, [rbp + nb130nf_vdwparam] shl ecx, 1 add ecx, [rsp + nb130nf_ntia] xorps xmm6, xmm6 movlps xmm6, [rsi + rcx*4] movaps xmm4, xmm6 shufps xmm4, xmm4, 252 ;# constant 11111100 shufps xmm6, xmm6, 253 ;# constant 11111101 movaps [rsp + nb130nf_c6], xmm4 movaps [rsp + nb130nf_c12], xmm6 lea rax, [rax + rax*2] ;# move coordinates to xmm0-xmm2 movss xmm0, [rdi + rax*4] movss xmm1, [rdi + rax*4 + 4] movss xmm2, [rdi + rax*4 + 8] mulps xmm3, [rsp + nb130nf_iq] xorps xmm7, xmm7 movaps xmm4, [rsp + nb130nf_ix] movaps xmm5, [rsp + nb130nf_iy] movaps xmm6, [rsp + nb130nf_iz] ;# 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 ;# rsq in xmm4 rsqrtss xmm5, xmm4 ;# lookup seed in xmm5 movss xmm2, xmm5 mulss xmm5, xmm5 movss xmm1, [rsp + nb130nf_three] mulss xmm5, xmm4 ;# rsq*lu*lu movss xmm0, [rsp + nb130nf_half] subss xmm1, xmm5 ;# constant 30-rsq*lu*lu mulss xmm1, xmm2 mulss xmm0, xmm1 ;# xmm0=rinv movaps xmm1, xmm0 mulss xmm3, xmm0 addss xmm3, [rsp + nb130nf_vctot] movss [rsp + nb130nf_vctot], xmm3 ;# LJ table mulss xmm4, xmm1 ;# r mulss xmm4, [rsp + nb130nf_tsc] ;# rtab movaps xmm0, xmm1 ;# copy of rinv 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 movd mm0, eax mov rsi, [rbp + nb130nf_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 + nb130nf_c6] mulss xmm5, xmm4 ;# Vvdw6 ;# Update Vvdwtot directly addss xmm5, [rsp + nb130nf_Vvdwtot] movss [rsp + nb130nf_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 + nb130nf_c12] mulss xmm5, xmm4 ;# Vvdw12 addss xmm5, [rsp + nb130nf_Vvdwtot] movss [rsp + nb130nf_Vvdwtot], xmm5.nb130nf_updateouterdata: ;# get n from stack mov esi, [rsp + nb130nf_n] ;# get group index for i particle mov rdx, [rbp + nb130nf_gid] ;# base of gid[] mov edx, [rdx + rsi*4] ;# ggid=gid[n] ;# accumulate total potential energy and update it movaps xmm7, [rsp + nb130nf_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 + nb130nf_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 + nb130nf_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 + nb130nf_Vvdw] addss xmm7, [rax + rdx*4] ;# move back to mem movss [rax + rdx*4], xmm7 ;# finish if last mov ecx, [rsp + nb130nf_nn1] ;# esi already loaded with n inc esi sub ecx, esi jz .nb130nf_outerend ;# not last, iterate outer loop once more! mov [rsp + nb130nf_n], esi jmp .nb130nf_outer.nb130nf_outerend: ;# check if more outer neighborlists remain mov ecx, [rsp + nb130nf_nri] ;# esi already loaded with n above sub ecx, esi jz .nb130nf_end ;# non-zero, do one more workunit jmp .nb130nf_threadloop.nb130nf_end: mov eax, [rsp + nb130nf_nouter] mov ebx, [rsp + nb130nf_ninner] mov rcx, [rbp + nb130nf_outeriter] mov rdx, [rbp + nb130nf_inneriter] mov [rcx], eax mov [rdx], ebx add rsp, 328 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?