nb_kernel234_x86_64_sse2.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,365 行 · 第 1/5 页
S
2,365 行
;# square it mulpd xmm4, xmm4 mulpd xmm5, xmm5 mulpd xmm6, xmm6 addpd xmm4, xmm5 addpd xmm4, xmm6 ;# rsq in xmm4 cvtpd2ps xmm5, xmm4 rsqrtps xmm5, xmm5 cvtps2pd xmm2, xmm5 ;# lu in low xmm2 ;# lookup seed in xmm2 movapd xmm5, xmm2 ;# copy of lu mulpd xmm2, xmm2 ;# lu*lu movapd xmm1, [rsp + nb234_three] mulpd xmm2, xmm4 ;# rsq*lu*lu movapd xmm0, [rsp + nb234_half] subpd xmm1, xmm2 ;# 30-rsq*lu*lu mulpd xmm1, xmm5 mulpd xmm1, xmm0 ;# xmm0=iter1 of rinv (new lu) movapd xmm5, xmm1 ;# copy of lu mulpd xmm1, xmm1 ;# lu*lu movapd xmm2, [rsp + nb234_three] mulpd xmm1, xmm4 ;# rsq*lu*lu movapd xmm0, [rsp + nb234_half] subpd xmm2, xmm1 ;# 30-rsq*lu*lu mulpd xmm2, xmm5 mulpd xmm2, xmm0 ;# xmm0=iter2 of rinv (new lu) mulpd xmm4, xmm2 ;# xmm4=r mulpd xmm4, [rsp + nb234_tsc] cvttpd2pi mm6, xmm4 ;# mm6 = lu idx cvtpi2pd xmm5, mm6 subpd xmm4, xmm5 movapd xmm1, xmm4 ;# xmm1=eps ;# xmm2=rinv movapd xmm3, xmm4 ;# eps pslld mm6, 3 ;# idx *= 8 mov rsi, [rbp + nb234_VFtab] movd r10d, mm6 psrlq mm6, 32 movd r11d, mm6 ;# indices in r10, r11. Load dispersion and repulsion tables in parallel. movapd xmm4, [rsi + r10*8] ;# Y1d F1d movapd xmm0, [rsi + r11*8] ;# Y2d F2d movapd xmm8, [rsi + r10*8 + 32] ;# Y1r F1r movapd xmm3, [rsi + r11*8 + 32] ;# Y2r F2r movapd xmm5, xmm4 movapd xmm9, xmm8 unpcklpd xmm4, xmm0 ;# Y1d Y2d unpckhpd xmm5, xmm0 ;# F1d F2d unpcklpd xmm8, xmm3 ;# Y1r Y2r unpckhpd xmm9, xmm3 ;# F1r F2r movapd xmm6, [rsi + r10*8 + 16] ;# G1d H1d movapd xmm0, [rsi + r11*8 + 16] ;# G2d H2d movapd xmm10, [rsi + r10*8 + 48] ;# G1r H1r movapd xmm3, [rsi + r11*8 + 48] ;# G2r H2r movapd xmm7, xmm6 movapd xmm11, xmm10 unpcklpd xmm6, xmm0 ;# G1d G2d unpckhpd xmm7, xmm0 ;# H1d H2d unpcklpd xmm10, xmm3 ;# G1r G2r unpckhpd xmm11, xmm3 ;# H1r H2r ;# tables ready, in xmm4-xmm7 and xmm8-xmm11 mulpd xmm7, xmm1 ;# Heps mulpd xmm11, xmm1 mulpd xmm6, xmm1 ;# Geps mulpd xmm10, xmm1 mulpd xmm7, xmm1 ;# Heps2 mulpd xmm11, xmm1 addpd xmm5, xmm6 ;# F+Geps addpd xmm9, xmm10 addpd xmm5, xmm7 ;# F+Geps+Heps2 = Fp addpd xmm9, xmm11 addpd xmm7, xmm7 ;# 2*Heps2 addpd xmm11, xmm11 addpd xmm7, xmm6 ;# 2*Heps2+Geps addpd xmm11, xmm10 addpd xmm7, xmm5 ;# FF = Fp + 2*Heps2 + Geps addpd xmm11, xmm9 mulpd xmm5, xmm1 ;# eps*Fp mulpd xmm9, xmm1 addpd xmm5, xmm4 ;# VV addpd xmm9, xmm8 mulpd xmm5, [rsp + nb234_c6] ;# VV*c6 = vnb6 mulpd xmm9, [rsp + nb234_c12] ;# VV*c12 = vnb12 addpd xmm5, xmm9 addpd xmm5, [rsp + nb234_Vvdwtot] movapd [rsp + nb234_Vvdwtot], xmm5 mulpd xmm7, [rsp + nb234_c6] ;# FF*c6 = fnb6 mulpd xmm11, [rsp + nb234_c12] ;# FF*c12 = fnb12 addpd xmm7, xmm11 mulpd xmm7, [rsp + nb234_tsc] mulpd xmm7, xmm2 xorpd xmm9, xmm9 subpd xmm9, xmm7 mulpd xmm13, xmm9 mulpd xmm14, xmm9 mulpd xmm15, xmm9 movapd xmm0, [rsp + nb234_fixO] movapd xmm1, [rsp + nb234_fiyO] movapd xmm2, [rsp + nb234_fizO] ;# accumulate i forces addpd xmm0, xmm13 addpd xmm1, xmm14 addpd xmm2, xmm15 movapd [rsp + nb234_fixO], xmm0 movapd [rsp + nb234_fiyO], xmm1 movapd [rsp + nb234_fizO], xmm2 ;# the fj's - start by accumulating forces from memory movlpd xmm3, [rdi + rax*8] movlpd xmm4, [rdi + rax*8 + 8] movlpd xmm5, [rdi + rax*8 + 16] movhpd xmm3, [rdi + rbx*8] movhpd xmm4, [rdi + rbx*8 + 8] movhpd xmm5, [rdi + rbx*8 + 16] addpd xmm3, xmm13 addpd xmm4, xmm14 addpd xmm5, xmm15 movlpd [rdi + rax*8], xmm3 movlpd [rdi + rax*8 + 8], xmm4 movlpd [rdi + rax*8 + 16], xmm5 movhpd [rdi + rbx*8], xmm3 movhpd [rdi + rbx*8 + 8], xmm4 movhpd [rdi + rbx*8 + 16], xmm5 ;# done with OO interaction ;# move j H1 coordinates to local temp variables mov rsi, [rbp + nb234_pos] movlpd xmm0, [rsi + rax*8 + 24] movlpd xmm1, [rsi + rax*8 + 32] movlpd xmm2, [rsi + rax*8 + 40] movhpd xmm0, [rsi + rbx*8 + 24] movhpd xmm1, [rsi + rbx*8 + 32] movhpd xmm2, [rsi + rbx*8 + 40] ;# xmm0 = H1x ;# xmm1 = H1y ;# xmm2 = H1z movapd xmm3, xmm0 movapd xmm4, xmm1 movapd xmm5, xmm2 movapd xmm6, xmm0 movapd xmm7, xmm1 movapd xmm8, xmm2 subpd xmm0, [rsp + nb234_ixH1] subpd xmm1, [rsp + nb234_iyH1] subpd xmm2, [rsp + nb234_izH1] subpd xmm3, [rsp + nb234_ixH2] subpd xmm4, [rsp + nb234_iyH2] subpd xmm5, [rsp + nb234_izH2] subpd xmm6, [rsp + nb234_ixM] subpd xmm7, [rsp + nb234_iyM] subpd xmm8, [rsp + nb234_izM] movapd [rsp + nb234_dxH1H1], xmm0 movapd [rsp + nb234_dyH1H1], xmm1 movapd [rsp + nb234_dzH1H1], xmm2 mulpd xmm0, xmm0 mulpd xmm1, xmm1 mulpd xmm2, xmm2 movapd [rsp + nb234_dxH2H1], xmm3 movapd [rsp + nb234_dyH2H1], xmm4 movapd [rsp + nb234_dzH2H1], xmm5 mulpd xmm3, xmm3 mulpd xmm4, xmm4 mulpd xmm5, xmm5 movapd [rsp + nb234_dxMH1], xmm6 movapd [rsp + nb234_dyMH1], xmm7 movapd [rsp + nb234_dzMH1], xmm8 mulpd xmm6, xmm6 mulpd xmm7, xmm7 mulpd xmm8, xmm8 addpd xmm0, xmm1 addpd xmm0, xmm2 addpd xmm3, xmm4 addpd xmm3, xmm5 addpd xmm6, xmm7 addpd xmm6, xmm8 ;# start doing invsqrt for jH1 atoms cvtpd2ps xmm1, xmm0 cvtpd2ps xmm4, xmm3 cvtpd2ps xmm7, xmm6 rsqrtps xmm1, xmm1 rsqrtps xmm4, xmm4 rsqrtps xmm7, xmm7 cvtps2pd xmm1, xmm1 cvtps2pd xmm4, xmm4 cvtps2pd xmm7, xmm7 movapd xmm2, xmm1 movapd xmm5, xmm4 movapd xmm8, xmm7 mulpd xmm1, xmm1 ;# lu*lu mulpd xmm4, xmm4 ;# lu*lu mulpd xmm7, xmm7 ;# lu*lu movapd xmm9, [rsp + nb234_three] movapd xmm10, xmm9 movapd xmm11, xmm9 mulpd xmm1, xmm0 ;# rsq*lu*lu mulpd xmm4, xmm3 ;# rsq*lu*lu mulpd xmm7, xmm6 ;# rsq*lu*lu subpd xmm9, xmm1 subpd xmm10, xmm4 subpd xmm11, xmm7 ;# 3-rsq*lu*lu mulpd xmm9, xmm2 mulpd xmm10, xmm5 mulpd xmm11, xmm8 ;# lu*(3-rsq*lu*lu) movapd xmm15, [rsp + nb234_half] mulpd xmm9, xmm15 ;# first iteration for rinvH1H1 mulpd xmm10, xmm15 ;# first iteration for rinvH2H1 mulpd xmm11, xmm15 ;# first iteration for rinvMH1 ;# second iteration step movapd xmm2, xmm9 movapd xmm5, xmm10 movapd xmm8, xmm11 mulpd xmm2, xmm2 ;# lu*lu mulpd xmm5, xmm5 ;# lu*lu mulpd xmm8, xmm8 ;# lu*lu movapd xmm1, [rsp + nb234_three] movapd xmm4, xmm1 movapd xmm7, xmm1 mulpd xmm2, xmm0 ;# rsq*lu*lu mulpd xmm5, xmm3 ;# rsq*lu*lu mulpd xmm8, xmm6 ;# rsq*lu*lu subpd xmm1, xmm2 subpd xmm4, xmm5 subpd xmm7, xmm8 ;# 3-rsq*lu*lu mulpd xmm9, xmm1 mulpd xmm10, xmm4 mulpd xmm11, xmm7 ;# lu*(3-rsq*lu*lu) movapd xmm15, [rsp + nb234_half] mulpd xmm9, xmm15 ;# rinvH1H1 mulpd xmm10, xmm15 ;# rinvH2H1 mulpd xmm11, xmm15 ;# rinvMH1 ;# H1 interactions ;# rsq in xmm0,xmm3,xmm6 ;# rinv in xmm9, xmm10, xmm11 movapd xmm1, xmm9 ;# copy of rinv movapd xmm4, xmm10 movapd xmm7, xmm11 movapd xmm2, [rsp + nb234_krf] mulpd xmm9, xmm9 ;# rinvsq mulpd xmm10, xmm10 mulpd xmm11, xmm11 mulpd xmm0, xmm2 ;# k*rsq mulpd xmm3, xmm2 mulpd xmm6, xmm2 movapd xmm2, xmm0 ;# copy of k*rsq movapd xmm5, xmm3 movapd xmm8, xmm6 addpd xmm2, xmm1 ;# rinv+krsq addpd xmm5, xmm4 addpd xmm8, xmm7 movapd xmm14, [rsp + nb234_crf] subpd xmm2, xmm14 ;# rinv+krsq-crf subpd xmm5, xmm14 subpd xmm8, xmm14 movapd xmm12, [rsp + nb234_qqHH] movapd xmm13, [rsp + nb234_qqMH] mulpd xmm2, xmm12 ;# voul=qq*(rinv+ krsq-crf) mulpd xmm5, xmm12 ;# voul=qq*(rinv+ krsq-crf) mulpd xmm8, xmm13 ;# voul=qq*(rinv+ krsq-crf) addpd xmm0, xmm0 ;# 2*krsq addpd xmm3, xmm3 addpd xmm6, xmm6 subpd xmm1, xmm0 ;# rinv-2*krsq subpd xmm4, xmm3 subpd xmm7, xmm6 mulpd xmm1, xmm12 ;# (rinv-2*krsq)*qq mulpd xmm4, xmm12 mulpd xmm7, xmm13 addpd xmm2, [rsp + nb234_vctot] addpd xmm5, xmm8 addpd xmm2, xmm5 movapd [rsp + nb234_vctot], xmm2 mulpd xmm9, xmm1 ;# fscal mulpd xmm10, xmm4 mulpd xmm11, xmm7 ;# move j H1 forces to xmm0-xmm2 mov rdi, [rbp + nb234_faction] movlpd xmm0, [rdi + rax*8 + 24] movlpd xmm1, [rdi + rax*8 + 32] movlpd xmm2, [rdi + rax*8 + 40] movhpd xmm0, [rdi + rbx*8 + 24] movhpd xmm1, [rdi + rbx*8 + 32] movhpd xmm2, [rdi + rbx*8 + 40] movapd xmm7, xmm9 movapd xmm8, xmm9 movapd xmm13, xmm11 movapd xmm14, xmm11 movapd xmm15, xmm11 movapd xmm11, xmm10 movapd xmm12, xmm10 mulpd xmm7, [rsp + nb234_dxH1H1] mulpd xmm8, [rsp + nb234_dyH1H1] mulpd xmm9, [rsp + nb234_dzH1H1] mulpd xmm10, [rsp + nb234_dxH2H1] mulpd xmm11, [rsp + nb234_dyH2H1] mulpd xmm12, [rsp + nb234_dzH2H1] mulpd xmm13, [rsp + nb234_dxMH1] mulpd xmm14, [rsp + nb234_dyMH1] mulpd xmm15, [rsp + nb234_dzMH1] addpd xmm0, xmm7 addpd xmm1, xmm8 addpd xmm2, xmm9 addpd xmm7, [rsp + nb234_fixH1] addpd xmm8, [rsp + nb234_fiyH1] addpd xmm9, [rsp + nb234_fizH1] addpd xmm0, xmm10 addpd xmm1, xmm11 addpd xmm2, xmm12 addpd xmm10, [rsp + nb234_fixH2] addpd xmm11, [rsp + nb234_fiyH2] addpd xmm12, [rsp + nb234_fizH2] addpd xmm0, xmm13 addpd xmm1, xmm14 addpd xmm2, xmm15 addpd xmm13, [rsp + nb234_fixM] addpd xmm14, [rsp + nb234_fiyM] addpd xmm15, [rsp + nb234_fizM] movapd [rsp + nb234_fixH1], xmm7 movapd [rsp + nb234_fiyH1], xmm8 movapd [rsp + nb234_fizH1], xmm9 movapd [rsp + nb234_fixH2], xmm10 movapd [rsp + nb234_fiyH2], xmm11 movapd [rsp + nb234_fizH2], xmm12 movapd [rsp + nb234_fixM], xmm13 movapd [rsp + nb234_fiyM], xmm14 movapd [rsp + nb234_fizM], xmm15 ;# store back j H1 forces from xmm0-xmm2 movlpd [rdi + rax*8 + 24], xmm0 movlpd [rdi + rax*8 + 32], xmm1 movlpd [rdi + rax*8 + 40], xmm2 movhpd [rdi + rbx*8 + 24], xmm0 movhpd [rdi + rbx*8 + 32], xmm1 movhpd [rdi + rbx*8 + 40], xmm2 ;# move j H2 coordinates to local temp variables mov rsi, [rbp + nb234_pos] movlpd xmm0, [rsi + rax*8 + 48] movlpd xmm1, [rsi + rax*8 + 56] movlpd xmm2, [rsi + rax*8 + 64] movhpd xmm0, [rsi + rbx*8 + 48] movhpd xmm1, [rsi + rbx*8 + 56] movhpd xmm2, [rsi + rbx*8 + 64] ;# xmm0 = H2x ;# xmm1 = H2y ;# xmm2 = H2z movapd xmm3, xmm0 movapd xmm4, xmm1 movapd xmm5, xmm2 movapd xmm6, xmm0 movapd xmm7, xmm1 movapd xmm8, xmm2 subpd xmm0, [rsp + nb234_ixH1] subpd xmm1, [rsp + nb234_iyH1] subpd xmm2, [rsp + nb234_izH1] subpd xmm3, [rsp + nb234_ixH2] subpd xmm4, [rsp + nb234_iyH2] subpd xmm5, [rsp + nb234_izH2] subpd xmm6, [rsp + nb234_ixM] subpd xmm7, [rsp + nb234_iyM] subpd xmm8, [rsp + nb234_izM] movapd [rsp + nb234_dxH1H2], xmm0 movapd [rsp + nb234_dyH1H2], xmm1 movapd [rsp + nb234_dzH1H2], xmm2 mulpd xmm0, xmm0 mulpd xmm1, xmm1 mulpd xmm2, xmm2 movapd [rsp + nb234_dxH2H2], xmm3 movapd [rsp + nb234_dyH2H2], xmm4 movapd [rsp + nb234_dzH2H2], xmm5 mulpd xmm3, xmm3 mulpd xmm4, xmm4 mulpd xmm5, xmm5 movapd [rsp + nb234_dxMH2], xmm6 movapd [rsp + nb234_dyMH2], xmm7 movapd [rsp + nb234_dzMH2], xmm8 mulpd xmm6, xmm6 mulpd xmm7, xmm7 mulpd xmm8, xmm8 addpd xmm0, xmm1 addpd xmm0, xmm2 addpd xmm3, xmm4 addpd xmm3, xmm5 addpd xmm6, xmm7 addpd xmm6, xmm8 ;# start doing invsqrt for jH2 atoms cvtpd2ps xmm1, xmm0 cvtpd2ps xmm4, xmm3 cvtpd2ps xmm7, xmm6 rsqrtps xmm1, xmm1 rsqrtps xmm4, xmm4 rsqrtps xmm7, xmm7 cvtps2pd xmm1, xmm1 cvtps2pd xmm4, xmm4 cvtps2pd xmm7, xmm7 movapd xmm2, xmm1 movapd xmm5, xmm4 movapd xmm8, xmm7 mulpd xmm1, xmm1 ;# lu*lu mulpd xmm4, xmm4 ;# lu*lu mulpd xmm7, xmm7 ;# lu*lu movapd xmm9, [rsp + nb234_three] movapd xmm10, xmm9 movapd xmm11, xmm9 mulpd xmm1, xmm0 ;# rsq*lu*lu mulpd xmm4, xmm3 ;# rsq*lu*lu mulpd xmm7, xmm6 ;# rsq*lu*lu subpd xmm9, xmm1 subpd xmm10, xmm4 subpd xmm11, xmm7 ;# 3-rsq*lu*lu mulpd xmm9, xmm2 mulpd xmm10, xmm5 mulpd xmm11, xmm8 ;# lu*(3-rsq*lu*lu)⌨️ 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?