nb_kernel330_x86_64_sse.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,051 行 · 第 1/4 页
S
2,051 行
mulps xmm3, [rsp + nb330_qq] ;# FF*qq = fij mulps xmm7, [rsp + nb330_c6] ;# fijD mulps xmm11, [rsp + nb330_c12] ;#fijR ;# accumulate vctot addps xmm1, [rsp + nb330_vctot] ;# accumulate Vvdwtot addps xmm5, [rsp + nb330_Vvdwtot] addps xmm5, xmm9 xorps xmm9, xmm9 movaps [rsp + nb330_vctot], xmm1 movaps [rsp + nb330_Vvdwtot], xmm5 mov rsi, [rbp + nb330_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 - - addps xmm3, xmm7 addps xmm3, xmm11 mulps xmm3, xmm15 mulps xmm3, [rsp + nb330_tsc] ;# fscal subps xmm9, xmm3 movaps xmm10, xmm9 movaps xmm11, xmm9 movhps xmm0, [rsi + rbx*4] ;# x1 y1 x2 y2 movhps xmm1, [rsi + rdx*4] ;# x3 y3 x4 y4 movaps xmm12, [rsp + nb330_fix] movaps xmm13, [rsp + nb330_fiy] movaps xmm14, [rsp + nb330_fiz] mulps xmm9, [rsp + nb330_dx] mulps xmm10, [rsp + nb330_dy] mulps xmm11, [rsp + nb330_dz] ;# accumulate i forces addps xmm12, xmm9 addps xmm13, xmm10 addps xmm14, xmm11 movaps [rsp + nb330_fix], xmm12 movaps [rsp + nb330_fiy], xmm13 movaps [rsp + nb330_fiz], xmm14 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 + nb330_innerk], 4 jl .nb330_finish_inner jmp .nb330_unroll_loop.nb330_finish_inner: ;# check if at least two particles remain add dword ptr [rsp + nb330_innerk], 4 mov edx, [rsp + nb330_innerk] and edx, 2 jnz .nb330_dopair jmp .nb330_checksingle.nb330_dopair: mov rcx, [rsp + nb330_innerjjnr] mov eax, [rcx] mov ebx, [rcx + 4] add qword ptr [rsp + nb330_innerjjnr], 8 mov rsi, [rbp + nb330_charge] movss xmm0, [rsi + rax*4] movss xmm2, [rsi + rbx*4] unpcklps xmm0, xmm2 ;# jqa jqb mulps xmm0, [rsp + nb330_iq] movaps [rsp + nb330_qq], xmm0 mov rsi, [rbp + nb330_type] ;# vdw parameters mov r12d, [rsi + rax*4] mov r13d, [rsi + rbx*4] shl r12d, 1 shl r13d, 1 mov edi, [rsp + nb330_ntia] add r12d, edi add r13d, edi mov rsi, [rbp + nb330_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 + nb330_c6], xmm0 movaps [rsp + nb330_c12], xmm3 lea rax, [rax + rax*2] ;# replace jnr with j3 lea rbx, [rbx + rbx*2] ;# load coordinates mov rdi, [rbp + nb330_pos] movlps xmm1, [rdi + rax*4] ;# x1 y1 - - movlps xmm2, [rdi + rbx*4] ;# x2 y2 - - movss xmm5, [rdi + rax*4 + 8] ;# z1 - - - movss xmm6, [rdi + rbx*4 + 8] ;# z2 - - - unpcklps xmm1, xmm2 ;# x1 x2 y1 y2 movhlps xmm2, xmm1 ;# y1 y2 - - unpcklps xmm5, xmm6 ;# z1 z2 - - ;# calc dr subps xmm1, [rsp + nb330_ix] subps xmm2, [rsp + nb330_iy] subps xmm5, [rsp + nb330_iz] ;# store dr movaps [rsp + nb330_dx], xmm1 movaps [rsp + nb330_dy], xmm2 movaps [rsp + nb330_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 + nb330_three] mulps xmm5, xmm1 ;# rsq*lu*lu subps xmm4, xmm5 ;# 30-rsq*lu*lu mulps xmm4, xmm2 mulps xmm4, [rsp + nb330_half] movaps xmm15, xmm4 mulps xmm1, xmm4 ;# xmm15=rinv ;# xmm1=r mulps xmm1, [rsp + nb330_tsc] ;# rtab ;# truncate and convert to integers cvttps2dq xmm5, xmm1 ;# convert back to float cvtdq2ps xmm4, xmm5 ;# multiply by 4 pslld xmm5, 2 ;# multiply by three (copy, mult. by two, add back) movaps xmm6, xmm5 pslld xmm5, 1 paddd xmm5, xmm6 ;# calculate eps subps xmm1, xmm4 ;# move to integer registers movd r8d, xmm5 pshufd xmm5, xmm5, 1 movd r9d, xmm5 movaps [rsp + nb330_eps], xmm1 ;# xmm15=rinv mov rsi, [rbp + nb330_VFtab] ;# load Coulomb and LJ table data in parallel movlps xmm0, [rsi + r8*4] ;# Y1c F1c movlps xmm4, [rsi + r8*4 + 16] ;# Y1d F1d movlps xmm8, [rsi + r8*4 + 32] ;# Y1r F1r movlps xmm1, [rsi + r9*4] ;# Y2c F2c movlps xmm5, [rsi + r9*4 + 16] ;# Y2d F2d movlps xmm9, [rsi + r9*4 + 32] ;# Y2r F2r unpcklps xmm0, xmm1 ;# Y1c Y2c F1c F2c unpcklps xmm4, xmm5 ;# Y1d Y2d F1d F2d unpcklps xmm8, xmm9 ;# Y1r Y2r F1r F2r movhlps xmm1, xmm0 ;# F1c F2c movhlps xmm5, xmm4 ;# F1d F2d movhlps xmm9, xmm8 ;# F1r F2r movlps xmm2, [rsi + r8*4 + 8] ;# G1c H1c movlps xmm6, [rsi + r8*4 + 24] ;# G1d H1d movlps xmm10, [rsi + r8*4 + 40] ;# G1r H1r movlps xmm3, [rsi + r9*4 + 8] ;# G2c H2c movlps xmm7, [rsi + r9*4 + 24] ;# G2d H2d movlps xmm11, [rsi + r9*4 + 40] ;# G2r H2r unpcklps xmm2, xmm3 ;# G1c G2c H1c H2c unpcklps xmm6, xmm7 ;# G1d G2d H1d H2d unpcklps xmm10, xmm11 ;# G1r G2r H1r H2r movhlps xmm3, xmm2 ;# H1c H2c movhlps xmm7, xmm6 ;# H1d H2d movhlps xmm11, xmm10 ;# H1r H2r ;# table data ready. Coul in xmm0-xmm3 , disp in xmm4-xmm7 , rep. in xmm8-xmm11 movaps xmm12, [rsp + nb330_eps] mulps xmm3, xmm12 ;# Heps mulps xmm7, xmm12 mulps xmm11, xmm12 mulps xmm2, xmm12 ;# Geps mulps xmm6, xmm12 mulps xmm10, xmm12 mulps xmm3, xmm12 ;# Heps2 mulps xmm7, xmm12 mulps xmm11, xmm12 addps xmm1, xmm2 ;# F+Geps addps xmm5, xmm6 addps xmm9, xmm10 addps xmm1, xmm3 ;# F+Geps+Heps2 = Fp addps xmm5, xmm7 addps xmm9, xmm11 addps xmm3, xmm3 ;# 2*Heps2 addps xmm7, xmm7 addps xmm11, xmm11 addps xmm3, xmm2 ;# 2*Heps2+Geps addps xmm7, xmm6 addps xmm11, xmm10 addps xmm3, xmm1 ;# FF = Fp + 2*Heps2 + Geps addps xmm7, xmm5 addps xmm11, xmm9 mulps xmm1, xmm12 ;# eps*Fp mulps xmm5, xmm12 mulps xmm9, xmm12 addps xmm1, xmm0 ;# VV addps xmm5, xmm4 addps xmm9, xmm8 mulps xmm1, [rsp + nb330_qq] ;# VV*qq = vcoul mulps xmm5, [rsp + nb330_c6] ;# vnb6 mulps xmm9, [rsp + nb330_c12] ;# vnb12 mulps xmm3, [rsp + nb330_qq] ;# FF*qq = fij mulps xmm7, [rsp + nb330_c6] ;# fijD mulps xmm11, [rsp + nb330_c12] ;#fijR ;# accumulate vctot addps xmm1, [rsp + nb330_vctot] movlps [rsp + nb330_vctot], xmm1 ;# accumulate Vvdwtot addps xmm5, [rsp + nb330_Vvdwtot] addps xmm5, xmm9 movlps [rsp + nb330_Vvdwtot], xmm5 xorps xmm9, xmm9 addps xmm3, xmm7 addps xmm3, xmm11 mulps xmm3, xmm15 mulps xmm3, [rsp + nb330_tsc] ;# fscal subps xmm9, xmm3 movaps xmm10, xmm9 movaps xmm11, xmm9 movaps xmm12, [rsp + nb330_fix] movaps xmm13, [rsp + nb330_fiy] movaps xmm14, [rsp + nb330_fiz] mulps xmm9, [rsp + nb330_dx] mulps xmm10, [rsp + nb330_dy] mulps xmm11, [rsp + nb330_dz] ;# accumulate i forces addps xmm12, xmm9 addps xmm13, xmm10 addps xmm14, xmm11 movlps [rsp + nb330_fix], xmm12 movlps [rsp + nb330_fiy], xmm13 movlps [rsp + nb330_fiz], xmm14 mov rsi, [rbp + nb330_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 .nb330_checksingle: mov edx, [rsp + nb330_innerk] and edx, 1 jnz .nb330_dosingle jmp .nb330_updateouterdata.nb330_dosingle: mov rdi, [rbp + nb330_pos] mov rcx, [rsp + nb330_innerjjnr] mov eax, [rcx] mov rsi, [rbp + nb330_charge] movss xmm0, [rsi + rax*4] mulss xmm0, [rsp + nb330_iq] movaps [rsp + nb330_qq], xmm0 ;# vdw parameters mov rsi, [rbp + nb330_type] mov r12d, [rsi + rax*4] shl r12d, 1 mov edi, [rsp + nb330_ntia] add r12d, edi mov rsi, [rbp + nb330_vdwparam] movss xmm0, [rsi + r12*4] movss xmm3, [rsi + r12*4 + 4] movaps [rsp + nb330_c6], xmm0 movaps [rsp + nb330_c12], xmm3 lea rax, [rax + rax*2] ;# replace jnr with j3 mov rdi, [rbp + nb330_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 + nb330_ix] subss xmm2, [rsp + nb330_iy] subss xmm5, [rsp + nb330_iz] ;# store dr movaps [rsp + nb330_dx], xmm1 movaps [rsp + nb330_dy], xmm2 movaps [rsp + nb330_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 + nb330_three] mulss xmm5, xmm1 ;# rsq*lu*lu subss xmm4, xmm5 ;# 30-rsq*lu*lu mulss xmm4, xmm2 mulss xmm4, [rsp + nb330_half] movaps xmm15, xmm4 mulss xmm1, xmm4 ;# xmm15=rinv ;# xmm1=r mulss xmm1, [rsp + nb330_tsc] ;# rtab ;# truncate and convert to integers cvttss2si r8d, xmm1 ;# convert back to float cvtsi2ss xmm4, r8d ;# multiply by 4 shl r8d, 2 ;# calculate eps subss xmm1, xmm4 ;# mult. by 3 lea r8, [r8 + r8*2] movaps [rsp + nb330_eps], xmm1 ;# xmm15=rinv mov rsi, [rbp + nb330_VFtab] ;# load Coulomb and LJ table data in parallel movss xmm0, [rsi + r8*4] movss xmm1, [rsi + r8*4 + 4] movss xmm2, [rsi + r8*4 + 8] movss xmm3, [rsi + r8*4 + 12] movss xmm4, [rsi + r8*4 + 16] movss xmm5, [rsi + r8*4 + 20] movss xmm6, [rsi + r8*4 + 24] movss xmm7, [rsi + r8*4 + 28] movss xmm8, [rsi + r8*4 + 32] movss xmm9, [rsi + r8*4 + 36] movss xmm10, [rsi + r8*4 + 40] movss xmm11, [rsi + r8*4 + 44] ;# table data ready. Coul in xmm0-xmm3 , disp in xmm4-xmm7 , rep. in xmm8-xmm11 movaps xmm12, [rsp + nb330_eps] mulps xmm3, xmm12 ;# Heps mulps xmm7, xmm12 mulps xmm11, xmm12 mulps xmm2, xmm12 ;# Geps mulps xmm6, xmm12 mulps xmm10, xmm12 mulps xmm3, xmm12 ;# Heps2 mulps xmm7, xmm12 mulps xmm11, xmm12 addps xmm1, xmm2 ;# F+Geps addps xmm5, xmm6 addps xmm9, xmm10 addps xmm1, xmm3 ;# F+Geps+Heps2 = Fp addps xmm5, xmm7 addps xmm9, xmm11 addps xmm3, xmm3 ;# 2*Heps2 addps xmm7, xmm7 addps xmm11, xmm11 addps xmm3, xmm2 ;# 2*Heps2+Geps addps xmm7, xmm6 addps xmm11, xmm10 addps xmm3, xmm1 ;# FF = Fp + 2*Heps2 + Geps addps xmm7, xmm5 addps xmm11, xmm9 mulps xmm1, [rsp + nb330_eps] ;# eps*Fp mulps xmm5, [rsp + nb330_eps] mulps xmm9, [rsp + nb330_eps] addps xmm1, xmm0 ;# VV addps xmm5, xmm4 addps xmm9, xmm8 mulps xmm1, [rsp + nb330_qq] ;# VV*qq = vcoul mulps xmm5, [rsp + nb330_c6] ;# vnb6 mulps xmm9, [rsp + nb330_c12] ;# vnb12 mulps xmm3, [rsp + nb330_qq] ;# FF*qq = fij mulps xmm7, [rsp + nb330_c6] ;# fijD mulps xmm11, [rsp + nb330_c12] ;#fijR ;# accumulate vctot addps xmm1, [rsp + nb330_vctot] movaps [rsp + nb330_vctot], xmm1 ;# accumulate Vvdwtot addps xmm5, [rsp + nb330_Vvdwtot] addps xmm5, xmm9 movaps [rsp + nb330_Vvdwtot], xmm5 xorps xmm9, xmm9 addps xmm3, xmm7 addps xmm3, xmm11 mulss xmm3, xmm15 mulps xmm3, [rsp + nb330_tsc] ;# fscal subps xmm9, xmm3 movaps xmm10, xmm9 movaps xmm11, xmm9 movaps xmm12, [rsp + nb330_fix] movaps xmm13, [rsp + nb330_fiy] movaps xmm14, [rsp + nb330_fiz] mulss xmm9, [rsp + nb330_dx] mulss xmm10, [rsp + nb330_dy] mulss xmm11, [rsp + nb330_dz] ;# accumulate i forces addss xmm12, xmm9 addss xmm13, xmm10 addss xmm14, xmm11 movaps [rsp + nb330_fix], xmm12 movaps [rsp + nb330_fiy], xmm13 movaps [rsp + nb330_fiz], xmm14 mov rsi, [rbp + nb330_faction] ;# add to j forces addss xmm9, [rsi + rax*4] addss xmm10, [rsi + rax*4 + 4]⌨️ 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?