nb_kernel310_x86_64_sse.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 1,805 行 · 第 1/4 页
S
1,805 行
mulps xmm2, [rsp + nb310_c6] ;# vvdw6=c6*rinv6 mulps xmm12, [rsp + nb310_c12] ;# vvdw12=c12*rinv12 movaps xmm0, xmm12 subps xmm12, xmm2 ;# Vvdw=Vvdw12-Vvdw6 ;# add potential to vvdwtot addps xmm12, [rsp + nb310_Vvdwtot] movaps [rsp + nb310_Vvdwtot], xmm12 shufps xmm6, xmm8, 136 ;# 10001000 shufps xmm7, xmm8, 221 ;# 11011101 ;# table data ready in xmm4-xmm7 mulps xmm7, xmm3 ;# Heps mulps xmm6, xmm3 ;# Geps mulps xmm7, xmm3 ;# Heps2 addps xmm5, xmm6 ;# F+Geps addps xmm5, xmm7 ;# F+Geps+Heps2 = Fp addps xmm7, xmm7 ;# 2*Heps2 addps xmm7, xmm6 ;# 2*Heps2+Geps addps xmm7, xmm5 ;# FF = Fp + 2*Heps2 + Geps mulps xmm5, xmm3 ;# eps*Fp addps xmm5, xmm4 ;# VV mulps xmm5, [rsp + nb310_qq] ;# VV*qq=vcoul mulps xmm7, [rsp + nb310_qq] ;# FF*qq=fijC ;# LJ forces mulps xmm2, [rsp + nb310_six] mulps xmm0, [rsp + nb310_twelve] subps xmm0, xmm2 mulps xmm0, xmm1 ;# (12*vnb12-6*vnb6)*rinv ;# add potential to vctot addps xmm5, [rsp + nb310_vctot] movaps [rsp + nb310_vctot], xmm5 mulps xmm7, [rsp + nb310_tsc] subps xmm0, xmm7 mulps xmm0, xmm1 ;# fscal ;# calculate scalar force by multiplying dx/dy/dz with fscal mulps xmm9, xmm0 mulps xmm10, xmm0 mulps xmm11, xmm0 mov rsi, [rbp + nb310_faction] ;# the fj's - start by accumulating x & y forces from memory movlps xmm0, [rsi + rax*4] ;# x1 y1 - - movlps xmm1, [rsi + rcx*4] ;# x3 y3 - - movhps xmm0, [rsi + rbx*4] ;# x1 y1 x2 y2 movhps xmm1, [rsi + rdx*4] ;# x3 y3 x4 y4 ;# xmm0-xmm2 contains tx-tz (partial force) ;# accumulate i forces addps xmm13, xmm9 addps xmm14, xmm10 addps xmm15, xmm11 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 + rax*4], xmm0 movlps [rsi + rcx*4], xmm1 movhps [rsi + rbx*4], xmm0 movhps [rsi + rdx*4], xmm1 ;# xmm11: fjz1 fjz2 fjz3 fjz4 pshufd xmm10, xmm11, 1 ;# fjz2 - - - movhlps xmm9, xmm11 ;# fjz3 - - - pshufd xmm8, xmm11, 3 ;# fjz4 - - - addss xmm11, [rsi + rax*4 + 8] addss xmm10, [rsi + rbx*4 + 8] addss xmm9, [rsi + rcx*4 + 8] addss xmm8, [rsi + rdx*4 + 8] movss [rsi + rax*4 + 8], xmm11 movss [rsi + rbx*4 + 8], xmm10 movss [rsi + rcx*4 + 8], xmm9 movss [rsi + rdx*4 + 8], xmm8 ;# should we do one more iteration? sub dword ptr [rsp + nb310_innerk], 4 jl .nb310_finish_inner jmp .nb310_unroll_loop.nb310_finish_inner: ;# check if at least two particles remain add dword ptr [rsp + nb310_innerk], 4 mov edx, [rsp + nb310_innerk] and edx, 2 jnz .nb310_dopair jmp .nb310_checksingle.nb310_dopair: ;# twice-unrolled innerloop here mov rdx, [rsp + nb310_innerjjnr] ;# pointer to jjnr[k] mov eax, [rdx] mov ebx, [rdx + 4] add qword ptr [rsp + nb310_innerjjnr], 8 ;# advance pointer (unrolled 2) mov rsi, [rbp + nb310_charge] movss xmm0, [rsi + rax*4] movss xmm2, [rsi + rbx*4] unpcklps xmm0, xmm2 ;# jqa jqb mulps xmm0, [rsp + nb310_iq] movaps [rsp + nb310_qq], xmm0 mov rsi, [rbp + nb310_type] ;# vdw parameters mov r12d, [rsi + rax*4] mov r13d, [rsi + rbx*4] shl r12d, 1 shl r13d, 1 mov edi, [rsp + nb310_ntia] add r12d, edi add r13d, edi mov rsi, [rbp + nb310_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 + nb310_c6], xmm0 movaps [rsp + nb310_c12], xmm3 lea rax, [rax + rax*2] ;# replace jnr with j3 lea rbx, [rbx + rbx*2] ;# load coordinates mov rdi, [rbp + nb310_pos] movlps xmm4, [rdi + rax*4] ;# x1 y1 - - movlps xmm5, [rdi + rbx*4] ;# x2 y2 - - movss xmm6, [rdi + rax*4 + 8] ;# z1 - - - movss xmm7, [rdi + rbx*4 + 8] ;# z2 - - - unpcklps xmm4, xmm5 ;# x1 x2 y1 y2 movhlps xmm5, xmm4 ;# y1 y2 - - unpcklps xmm6, xmm7 ;# z1 z2 - - ;# calc dr subps xmm4, [rsp + nb310_ix] subps xmm5, [rsp + nb310_iy] subps xmm6, [rsp + nb310_iz] ;# store dr in xmm9-xmm11 movaps xmm9, xmm4 movaps xmm10, xmm5 movaps xmm11, xmm6 ;# square it mulps xmm4,xmm4 mulps xmm5,xmm5 mulps xmm6,xmm6 addps xmm4, xmm5 addps xmm4, xmm6 ;# rsq in xmm4 ;# calculate rinv=1/sqrt(rsq) rsqrtps xmm5, xmm4 movaps xmm2, xmm5 mulps xmm5, xmm5 movaps xmm1, [rsp + nb310_three] mulps xmm5, xmm4 ;# rsq*lu*lu subps xmm1, xmm5 ;# 30-rsq*lu*lu mulps xmm1, xmm2 mulps xmm1, [rsp + nb310_half] ;# xmm1=rinv movaps xmm3, xmm4 ;# xmm3=rsq mulps xmm3, xmm1 ;# r mulps xmm3, [rsp + nb310_tsc] ;# rtab ;# truncate and convert to integers cvttps2dq xmm2, xmm3 ;# convert back to float cvtdq2ps xmm0, xmm2 ;# multiply by 4 pslld xmm2, 2 ;# move to integer registers movd r12d, xmm2 pshufd xmm2, xmm2, 1 movd r13d, xmm2 ;# calculate eps subps xmm3, xmm0 mov rsi, [rbp + nb310_VFtab] ;# load table data movlps xmm4, [rsi + r12*4] movlps xmm5, [rsi + r13*4] unpcklps xmm4, xmm5 movhlps xmm5, xmm4 movaps xmm0, xmm1 ;# rinv mulps xmm0, xmm0 ;# rinvsq movaps xmm2, xmm0 ;# rinvsq mulps xmm2, xmm2 ;# rinv4 mulps xmm2, xmm0 ;# rinv6 movaps xmm12, xmm2 mulps xmm12, xmm12 ;# rinv12 movlps xmm6, [rsi + r12*4 + 8] movlps xmm7, [rsi + r13*4 + 8] unpcklps xmm6, xmm7 movhlps xmm7, xmm6 ;# table data ready in xmm4-xmm7 mulps xmm2, [rsp + nb310_c6] ;# vvdw6=c6*rinv6 mulps xmm12, [rsp + nb310_c12] ;# vvdw12=c12*rinv12 movaps xmm0, xmm12 subps xmm12, xmm2 ;# Vvdw=Vvdw12-Vvdw6 ;# add potential to vvdwtot addps xmm12, [rsp + nb310_Vvdwtot] movlps [rsp + nb310_Vvdwtot], xmm12 mulps xmm7, xmm3 ;# Heps mulps xmm6, xmm3 ;# Geps mulps xmm7, xmm3 ;# Heps2 addps xmm5, xmm6 ;# F+Geps addps xmm5, xmm7 ;# F+Geps+Heps2 = Fp addps xmm7, xmm7 ;# 2*Heps2 addps xmm7, xmm6 ;# 2*Heps2+Geps addps xmm7, xmm5 ;# FF = Fp + 2*Heps2 + Geps mulps xmm5, xmm3 ;# eps*Fp addps xmm5, xmm4 ;# VV mulps xmm5, [rsp + nb310_qq] ;# VV*qq=vcoul mulps xmm7, [rsp + nb310_qq] ;# FF*qq=fijC ;# LJ forces mulps xmm2, [rsp + nb310_six] mulps xmm0, [rsp + nb310_twelve] subps xmm0, xmm2 mulps xmm0, xmm1 ;# (12*vnb12-6*vnb6)*rinv ;# add potential to vctot addps xmm5, [rsp + nb310_vctot] movlps [rsp + nb310_vctot], xmm5 xorps xmm8, xmm8 mulps xmm7, [rsp + nb310_tsc] subps xmm0, xmm7 mulps xmm0, xmm1 ;# fscal ;# calculate scalar force by multiplying dx/dy/dz with fscal mulps xmm9, xmm0 mulps xmm10, xmm0 mulps xmm11, xmm0 movlhps xmm9, xmm8 movlhps xmm10, xmm8 movlhps xmm11, xmm8 ;# accumulate i forces addps xmm13, xmm9 addps xmm14, xmm10 addps xmm15, xmm11 mov rsi, [rbp + nb310_faction] ;# the fj's - start by accumulating x & y forces from memory 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.nb310_checksingle: mov edx, [rsp + nb310_innerk] and edx, 1 jnz .nb310_dosingle jmp .nb310_updateouterdata.nb310_dosingle: mov rcx, [rsp + nb310_innerjjnr] mov eax, [rcx] mov rsi, [rbp + nb310_charge] movss xmm0, [rsi + rax*4] mulss xmm0, [rsp + nb310_iq] movaps [rsp + nb310_qq], xmm0 mov rsi, [rbp + nb310_type] ;# vdw parameters mov r12d, [rsi + rax*4] shl r12d, 1 mov edi, [rsp + nb310_ntia] add r12d, edi mov rsi, [rbp + nb310_vdwparam] movss xmm0, [rsi + r12*4] movss xmm3, [rsi + r12*4 + 4] movaps [rsp + nb310_c6], xmm0 movaps [rsp + nb310_c12], xmm3 lea rax, [rax + rax*2] ;# replace jnr with j3 mov rdi, [rbp + nb310_pos] movss xmm4, [rdi + rax*4] ;# x1 - - - movss xmm5, [rdi + rax*4 + 4] ;# y2 - - - movss xmm6, [rdi + rax*4 + 8] ;# 13 - - - ;# calc dr subss xmm4, [rsp + nb310_ix] subss xmm5, [rsp + nb310_iy] subss xmm6, [rsp + nb310_iz] ;# store dr in xmm9-xmm11 movaps xmm9, xmm4 movaps xmm10, xmm5 movaps xmm11, xmm6 ;# square it mulss xmm4,xmm4 mulss xmm5,xmm5 mulss xmm6,xmm6 addss xmm4, xmm5 addss xmm4, xmm6 ;# rsq in xmm4 ;# calculate rinv=1/sqrt(rsq) rsqrtss xmm5, xmm4 movaps xmm2, xmm5 mulss xmm5, xmm5 movaps xmm1, [rsp + nb310_three] mulss xmm5, xmm4 ;# rsq*lu*lu subss xmm1, xmm5 ;# 30-rsq*lu*lu mulss xmm1, xmm2 mulss xmm1, [rsp + nb310_half] ;# xmm1=rinv movaps xmm3, xmm4 ;# xmm3=rsq mulss xmm3, xmm1 ;# r mulss xmm3, [rsp + nb310_tsc] ;# rtab ;# truncate and convert to integers cvttss2si r12d, xmm3 ;# convert back to float cvtsi2ss xmm0, r12d ;# multiply by 4 shl r12d, 2 ;# calculate eps subss xmm3, xmm0 mov rsi, [rbp + nb310_VFtab] movaps xmm0, xmm1 ;# rinv mulss xmm0, xmm0 ;# rinvsq movaps xmm2, xmm0 ;# rinvsq mulss xmm2, xmm2 ;# rinv4 mulss xmm2, xmm0 ;# rinv6 movaps xmm12, xmm2 mulss xmm12, xmm12 ;# rinv12 ;# load table data movss xmm4, [rsi + r12*4] movss xmm5, [rsi + r12*4 + 4] movss xmm6, [rsi + r12*4 + 8] movss xmm7, [rsi + r12*4 + 12] ;# table data ready in xmm4-xmm7 mulss xmm2, [rsp + nb310_c6] ;# vvdw6=c6*rinv6 mulss xmm12, [rsp + nb310_c12] ;# vvdw12=c12*rinv12 movaps xmm0, xmm12 subss xmm12, xmm2 ;# Vvdw=Vvdw12-Vvdw6 ;# add potential to vvdwtot addss xmm12, [rsp + nb310_Vvdwtot] movss [rsp + nb310_Vvdwtot], xmm12 mulss xmm7, xmm3 ;# Heps mulss xmm6, xmm3 ;# Geps mulss xmm7, xmm3 ;# Heps2 addss xmm5, xmm6 ;# F+Geps addss xmm5, xmm7 ;# F+Geps+Heps2 = Fp addss xmm7, xmm7 ;# 2*Heps2 addss xmm7, xmm6 ;# 2*Heps2+Geps addss xmm7, xmm5 ;# FF = Fp + 2*Heps2 + Geps mulss xmm5, xmm3 ;# eps*Fp addss xmm5, xmm4 ;# VV mulss xmm5, [rsp + nb310_qq] ;# VV*qq=vcoul mulss xmm7, [rsp + nb310_qq] ;# FF*qq=fijC ;# LJ forces mulss xmm2, [rsp + nb310_six] mulss xmm0, [rsp + nb310_twelve] subss xmm0, xmm2 mulss xmm0, xmm1 ;# (12*vnb12-6*vnb6)*rinv ;# add potential to vctot addss xmm5, [rsp + nb310_vctot] movss [rsp + nb310_vctot], xmm5 mulss xmm7, [rsp + nb310_tsc] subss xmm0, xmm7 mulss xmm0, xmm1 ;# fscal ;# calculate scalar force by multiplying dx/dy/dz with fscal mulss xmm9, xmm0 mulss xmm10, xmm0 mulss xmm11, xmm0 ;# accumulate i forces addss xmm13, xmm9 addss xmm14, xmm10 addss xmm15, xmm11 mov rsi, [rbp + nb310_faction] ;# add to j forces addss xmm9, [rsi + rax*4] addss xmm10, [rsi + rax*4 + 4] addss xmm11, [rsi + rax*4 + 8]⌨️ 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?