nb_kernel030_x86_64_sse.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 1,748 行 · 第 1/4 页
S
1,748 行
movhps xmm13, [rsi + rbx*4] movhps xmm14, [rsi + rdx*4] addps xmm7, xmm7 ;# 2*Heps2 addps xmm11, xmm11 addps xmm7, xmm6 ;# 2*Heps2+Geps addps xmm11, xmm10 ;# shuffle c6/c12 movaps xmm12, xmm13 shufps xmm12, xmm14, 136 ;# 10001000 => c6 shufps xmm13, xmm14, 221 ;# 11011101 => c12 addps xmm7, xmm5 ;# FF = Fp + 2*Heps2 + Geps addps xmm11, xmm9 mulps xmm5, xmm1 ;# eps*Fp mulps xmm9, xmm1 addps xmm5, xmm4 ;# VV addps xmm9, xmm8 mov rsi, [rbp + nb030_faction] mulps xmm5, xmm12 ;# VV*c6 = vnb6 mulps xmm9, xmm13 ;# VV*c12 = vnb12 addps xmm5, xmm9 addps xmm5, [rsp + nb030_Vvdwtot] movaps [rsp + nb030_Vvdwtot], xmm5 mulps xmm7, xmm12 ;# FF*c6 = fnb6 mulps xmm11, xmm13 ;# FF*c12 = fnb12 addps xmm7, xmm11 mulps xmm7, [rsp + nb030_tsc] mulps xmm7, xmm2 xorps xmm9, xmm9 subps xmm9, xmm7 movaps xmm10, xmm9 movaps xmm11, xmm9 movaps xmm12, [rsp + nb030_fix] movaps xmm13, [rsp + nb030_fiy] movaps xmm14, [rsp + nb030_fiz] mulps xmm9, [rsp + nb030_dx] mulps xmm10, [rsp + nb030_dy] mulps xmm11, [rsp + nb030_dz] ;# accumulate i forces addps xmm12, xmm9 addps xmm13, xmm10 addps xmm14, xmm11 movaps [rsp + nb030_fix], xmm12 movaps [rsp + nb030_fiy], xmm13 movaps [rsp + nb030_fiz], xmm14 ;# the fj's - start by combining x & y forces from memory movlps xmm0, [rsi + r12*4] ;# x1 y1 - - movlps xmm1, [rsi + r14*4] ;# x3 y3 - - movhps xmm0, [rsi + r13*4] ;# x1 y1 x2 y2 movhps xmm1, [rsi + r15*4] ;# x3 y3 x4 y4 movaps xmm8, xmm9 unpcklps xmm9, xmm10 ;# x1 y1 x2 y2 unpckhps xmm8, xmm10 ;# x3 y3 x4 y4 ;# update fjx and fjy addps xmm0, xmm9 addps xmm1, xmm8 movlps [rsi + r12*4], xmm0 movlps [rsi + r14*4], xmm1 movhps [rsi + r13*4], xmm0 movhps [rsi + r15*4], xmm1 ;# xmm11: fjz1 fjz2 fjz3 fjz4 pshufd xmm5, xmm11, 1 ;# fjz2 - - - movhlps xmm4, xmm11 ;# fjz3 - - - pshufd xmm3, xmm11, 3 ;# fjz4 - - - addss xmm11, [rsi + r12*4 + 8] addss xmm5, [rsi + r13*4 + 8] addss xmm4, [rsi + r14*4 + 8] addss xmm3, [rsi + r15*4 + 8] movss [rsi + r12*4 + 8], xmm11 movss [rsi + r13*4 + 8], xmm5 movss [rsi + r14*4 + 8], xmm4 movss [rsi + r15*4 + 8], xmm3 ;# should we do one more iteration? sub dword ptr [rsp + nb030_innerk], 4 jl .nb030_finish_inner jmp .nb030_unroll_loop.nb030_finish_inner: ;# check if at least two particles remain add dword ptr [rsp + nb030_innerk], 4 mov edx, [rsp + nb030_innerk] and edx, 2 jnz .nb030_dopair jmp .nb030_checksingle.nb030_dopair: mov rcx, [rsp + nb030_innerjjnr] mov eax, [rcx] mov ebx, [rcx + 4] add qword ptr [rsp + nb030_innerjjnr], 8 mov rsi, [rbp + nb030_type] mov r12d, [rsi + rax*4] mov r13d, [rsi + rbx*4] shl r12d, 1 shl r13d, 1 mov edi, [rsp + nb030_ntia] add r12d, edi add r13d, edi mov rsi, [rbp + nb030_vdwparam] movlps xmm3, [rsi + r12*4] movhps xmm3, [rsi + r13*4] xorps xmm7, xmm7 movaps xmm0, xmm3 shufps xmm0, xmm7, 136 ;# 10001000 shufps xmm3, xmm7, 221 ;# 11011101 movaps [rsp + nb030_c6], xmm0 movaps [rsp + nb030_c12], xmm3 lea rax, [rax + rax*2] ;# replace jnr with j3 lea rbx, [rbx + rbx*2] mov rdi, [rbp + nb030_pos] ;# load coordinates movlps xmm1, [rdi + rax*4] ;# x1 y1 - - movlps xmm4, [rdi + rbx*4] ;# x2 y2 - - movss xmm5, [rdi + rax*4 + 8] ;# z1 - - - movss xmm7, [rdi + rbx*4 + 8] ;# z2 - - - unpcklps xmm1, xmm4 ;# x1 x2 y1 y2 movhlps xmm2, xmm1 ;# y1 y2 - - unpcklps xmm5, xmm7 ;# z1 z2 - - ;# calc dr subps xmm1, [rsp + nb030_ix] subps xmm2, [rsp + nb030_iy] subps xmm5, [rsp + nb030_iz] ;# store dr movaps [rsp + nb030_dx], xmm1 movaps [rsp + nb030_dy], xmm2 movaps [rsp + nb030_dz], xmm5 ;# square it mulps xmm1,xmm1 mulps xmm2,xmm2 mulps xmm5,xmm5 addps xmm1, xmm2 addps xmm1, xmm5 ;# rsq in xmm1 ;# calculate rinv=1/sqrt(rsq) rsqrtps xmm5, xmm1 movaps xmm2, xmm5 mulps xmm5, xmm5 movaps xmm4, [rsp + nb030_three] mulps xmm5, xmm1 ;# rsq*lu*lu subps xmm4, xmm5 ;# 30-rsq*lu*lu mulps xmm4, xmm2 mulps xmm4, [rsp + nb030_half] movaps xmm2, xmm4 mulps xmm1, xmm4 ;# xmm2=rinv ;# xmm1=r mulps xmm1, [rsp + nb030_tsc] ;# rtab ;# truncate and convert to integers cvttps2dq xmm5, xmm1 ;# convert back to float cvtdq2ps xmm4, xmm5 ;# multiply by 8 pslld xmm5, 3 ;# calculate eps subps xmm1, xmm4 ;# move to integer registers movd r8d, xmm5 pshufd xmm5, xmm5, 1 movd r9d, xmm5 ;# xmm1=eps ;# xmm2=rinv mov rsi, [rbp + nb030_VFtab] ;# calculate LJ table movlps xmm4, [rsi + r8*4] movlps xmm5, [rsi + r9*4] unpcklps xmm4, xmm5 movhlps xmm5, xmm4 movlps xmm6, [rsi + r8*4 + 8] movlps xmm7, [rsi + r9*4 + 8] unpcklps xmm6, xmm7 movhlps xmm7, xmm6 movlps xmm8, [rsi + r8*4 + 16] movlps xmm9, [rsi + r9*4 + 16] unpcklps xmm8, xmm9 movhlps xmm9, xmm8 movlps xmm10, [rsi + r8*4 + 24] movlps xmm11, [rsi + r9*4 + 24] unpcklps xmm10, xmm11 movhlps xmm11, xmm10 ;# dispersion table in xmm4-xmm7, repulsion table in xmm8-xmm11 mulps xmm7, xmm1 ;# Heps mulps xmm11, xmm1 mulps xmm6, xmm1 ;# Geps mulps xmm10, xmm1 mulps xmm7, xmm1 ;# Heps2 mulps xmm11, xmm1 addps xmm5, xmm6 ;# F+Geps addps xmm9, xmm10 addps xmm5, xmm7 ;# F+Geps+Heps2 = Fp addps xmm9, xmm11 addps xmm7, xmm7 ;# 2*Heps2 addps xmm11, xmm11 addps xmm7, xmm6 ;# 2*Heps2+Geps addps xmm11, xmm10 addps xmm7, xmm5 ;# FF = Fp + 2*Heps2 + Geps addps xmm11, xmm9 mulps xmm5, xmm1 ;# eps*Fp mulps xmm9, xmm1 movaps xmm12, [rsp + nb030_c6] movaps xmm13, [rsp + nb030_c12] addps xmm5, xmm4 ;# VV addps xmm9, xmm8 mulps xmm5, xmm12 ;# VV*c6 = vnb6 mulps xmm9, xmm13 ;# VV*c12 = vnb12 addps xmm5, xmm9 xorps xmm8, xmm8 movlhps xmm5, xmm8 addps xmm5, [rsp + nb030_Vvdwtot] movaps [rsp + nb030_Vvdwtot], xmm5 mulps xmm7, xmm12 ;# FF*c6 = fnb6 mulps xmm11, xmm13 ;# FF*c12 = fnb12 addps xmm7, xmm11 mulps xmm7, [rsp + nb030_tsc] mulps xmm7, xmm2 xorps xmm9, xmm9 subps xmm9, xmm7 movaps xmm10, xmm9 movaps xmm11, xmm9 movaps xmm12, [rsp + nb030_fix] movaps xmm13, [rsp + nb030_fiy] movaps xmm14, [rsp + nb030_fiz] mulps xmm9, [rsp + nb030_dx] mulps xmm10, [rsp + nb030_dy] mulps xmm11, [rsp + nb030_dz] movlhps xmm9, xmm8 movlhps xmm10, xmm8 movlhps xmm11, xmm8 ;# accumulate i forces addps xmm12, xmm9 addps xmm13, xmm10 addps xmm14, xmm11 movaps [rsp + nb030_fix], xmm12 movaps [rsp + nb030_fiy], xmm13 movaps [rsp + nb030_fiz], xmm14 ;# the fj's - start by accumulating x & y forces from memory mov rsi, [rbp + nb030_faction] movlps xmm0, [rsi + rax*4] ;# x1 y1 - - movhps xmm0, [rsi + rbx*4] ;# x1 y1 x2 y2 unpcklps xmm9, xmm10 ;# x1 y1 x2 y2 addps xmm0, xmm9 movlps [rsi + rax*4], xmm0 movhps [rsi + rbx*4], xmm0 ;# z forces pshufd xmm8, xmm11, 1 addss xmm11, [rsi + rax*4 + 8] addss xmm8, [rsi + rbx*4 + 8] movss [rsi + rax*4 + 8], xmm11 movss [rsi + rbx*4 + 8], xmm8 .nb030_checksingle: mov edx, [rsp + nb030_innerk] and edx, 1 jnz .nb030_dosingle jmp .nb030_updateouterdata.nb030_dosingle: mov rdi, [rbp + nb030_pos] mov rcx, [rsp + nb030_innerjjnr] mov eax, [rcx] mov rsi, [rbp + nb030_type] mov r12d, [rsi + rax*4] shl r12d, 1 mov edi, [rsp + nb030_ntia] add r12d, edi mov rsi, [rbp + nb030_vdwparam] movss xmm0, [rsi + r12*4] movss xmm3, [rsi + r12*4 + 4] movaps [rsp + nb030_c6], xmm0 movaps [rsp + nb030_c12], xmm3 lea rax, [rax + rax*2] ;# replace jnr with j3 mov rdi, [rbp + nb030_pos] ;# load coordinates movss xmm1, [rdi + rax*4] movss xmm2, [rdi + rax*4 + 4] movss xmm5, [rdi + rax*4 + 8] ;# calc dr subss xmm1, [rsp + nb030_ix] subss xmm2, [rsp + nb030_iy] subss xmm5, [rsp + nb030_iz] ;# store dr movaps [rsp + nb030_dx], xmm1 movaps [rsp + nb030_dy], xmm2 movaps [rsp + nb030_dz], xmm5 ;# square it mulss xmm1,xmm1 mulss xmm2,xmm2 mulss xmm5,xmm5 addss xmm1, xmm2 addss xmm1, xmm5 ;# rsq in xmm1 ;# calculate rinv=1/sqrt(rsq) rsqrtss xmm5, xmm1 movaps xmm2, xmm5 mulss xmm5, xmm5 movaps xmm4, [rsp + nb030_three] mulss xmm5, xmm1 ;# rsq*lu*lu subss xmm4, xmm5 ;# 30-rsq*lu*lu mulss xmm4, xmm2 mulss xmm4, [rsp + nb030_half] movaps xmm2, xmm4 mulss xmm1, xmm4 ;# xmm2=rinv ;# xmm1=r mulss xmm1, [rsp + nb030_tsc] ;# rtab ;# truncate and convert to integers cvttss2si r8d, xmm1 ;# convert back to float cvtsi2ss xmm4, r8d ;# multiply by 8 shl r8d, 3 ;# calculate eps subss xmm1, xmm4 ;# xmm1=eps ;# xmm2=rinv mov rsi, [rbp + nb030_VFtab] ;# calculate LJ table movss xmm4, [rsi + r8*4] movss xmm5, [rsi + r8*4 + 4] movss xmm6, [rsi + r8*4 + 8] movss xmm7, [rsi + r8*4 + 12] movss xmm8, [rsi + r8*4 + 16] movss xmm9, [rsi + r8*4 + 20] movss xmm10, [rsi + r8*4 + 24] movss xmm11, [rsi + r8*4 + 28] ;# dispersion table in xmm4-xmm7, repulsion table in xmm8-xmm11 mulss xmm7, xmm1 ;# Heps mulss xmm11, xmm1 mulss xmm6, xmm1 ;# Geps mulss xmm10, xmm1 mulss xmm7, xmm1 ;# Heps2 mulss xmm11, xmm1 addss xmm5, xmm6 ;# F+Geps addss xmm9, xmm10 addss xmm5, xmm7 ;# F+Geps+Heps2 = Fp addss xmm9, xmm11 addss xmm7, xmm7 ;# 2*Heps2 addss xmm11, xmm11 addss xmm7, xmm6 ;# 2*Heps2+Geps addss xmm11, xmm10 addss xmm7, xmm5 ;# FF = Fp + 2*Heps2 + Geps addss xmm11, xmm9 mulss xmm5, xmm1 ;# eps*Fp mulss xmm9, xmm1 movaps xmm12, [rsp + nb030_c6] movaps xmm13, [rsp + nb030_c12] addss xmm5, xmm4 ;# VV addss xmm9, xmm8 mulss xmm5, xmm12 ;# VV*c6 = vnb6 mulss xmm9, xmm13 ;# VV*c12 = vnb12 addss xmm5, xmm9 addss xmm5, [rsp + nb030_Vvdwtot] movss [rsp + nb030_Vvdwtot], xmm5 mulss xmm7, xmm12 ;# FF*c6 = fnb6 mulss xmm11, xmm13 ;# FF*c12 = fnb12 addss xmm7, xmm11 ⌨️ 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?