nb_kernel334_x86_64_sse2.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,376 行 · 第 1/5 页
S
2,376 行
movapd [rsp + nb334_rinvH1M], xmm9 movapd [rsp + nb334_rinvH2M], xmm10 movapd [rsp + nb334_rinvMM], xmm11 ;# M interactions ;# rsq in xmm0,xmm3,xmm6 ;# rinv in xmm9, xmm10, xmm11 movapd xmm1, [rsp + nb334_tsc] mulpd xmm0, xmm9 ;# r mulpd xmm3, xmm10 mulpd xmm6, xmm11 mulpd xmm0, xmm1 ;# rtab mulpd xmm3, xmm1 mulpd xmm6, xmm1 ;# truncate and convert to integers cvttpd2dq xmm1, xmm0 cvttpd2dq xmm4, xmm3 cvttpd2dq xmm7, xmm6 ;# convert back to float cvtdq2pd xmm2, xmm1 cvtdq2pd xmm5, xmm4 cvtdq2pd xmm8, xmm7 ;# multiply by 4 pslld xmm1, 2 pslld xmm4, 2 pslld xmm7, 2 ;# multiply by three (copy, mult. by two, add back) movdqa xmm10, xmm1 movdqa xmm11, xmm4 movdqa xmm12, xmm7 pslld xmm1, 1 pslld xmm4, 1 pslld xmm7, 1 paddd xmm1, xmm10 paddd xmm4, xmm11 paddd xmm7, xmm12 ;# move to integer registers pshufd xmm13, xmm1, 1 pshufd xmm14, xmm4, 1 pshufd xmm15, xmm7, 1 movd r8d, xmm1 movd r10d, xmm4 movd r12d, xmm7 movd r9d, xmm13 movd r11d, xmm14 movd r13d, xmm15 mov rsi, [rbp + nb334_VFtab] ;# calculate eps subpd xmm0, xmm2 subpd xmm3, xmm5 subpd xmm6, xmm8 movapd xmm12, xmm0 ;# epsH1 movapd xmm13, xmm3 ;# epsH2 movapd xmm14, xmm6 ;# epsM ;# Load LOTS of table data movlpd xmm0, [rsi + r8*8] movlpd xmm1, [rsi + r8*8 + 8] movlpd xmm2, [rsi + r8*8 + 16] movlpd xmm3, [rsi + r8*8 + 24] movlpd xmm4, [rsi + r10*8] movlpd xmm5, [rsi + r10*8 + 8] movlpd xmm6, [rsi + r10*8 + 16] movlpd xmm7, [rsi + r10*8 + 24] movlpd xmm8, [rsi + r12*8] movlpd xmm9, [rsi + r12*8 + 8] movlpd xmm10, [rsi + r12*8 + 16] movlpd xmm11, [rsi + r12*8 + 24] movhpd xmm0, [rsi + r9*8] movhpd xmm1, [rsi + r9*8 + 8] movhpd xmm2, [rsi + r9*8 + 16] movhpd xmm3, [rsi + r9*8 + 24] movhpd xmm4, [rsi + r11*8] movhpd xmm5, [rsi + r11*8 + 8] movhpd xmm6, [rsi + r11*8 + 16] movhpd xmm7, [rsi + r11*8 + 24] movhpd xmm8, [rsi + r13*8] movhpd xmm9, [rsi + r13*8 + 8] movhpd xmm10, [rsi + r13*8 + 16] movhpd xmm11, [rsi + r13*8 + 24] ;# table data ready in xmm0-xmm3 , xmm4-xmm7 , and xmm8-xmm11 mulpd xmm3, xmm12 ;# Heps mulpd xmm7, xmm13 mulpd xmm11, xmm14 mulpd xmm2, xmm12 ;# Geps mulpd xmm6, xmm13 mulpd xmm10, xmm14 mulpd xmm3, xmm12 ;# Heps2 mulpd xmm7, xmm13 mulpd xmm11, xmm14 addpd xmm1, xmm2 ;# F+Geps addpd xmm5, xmm6 addpd xmm9, xmm10 addpd xmm1, xmm3 ;# F+Geps+Heps2 = Fp addpd xmm5, xmm7 addpd xmm9, xmm11 addpd xmm3, xmm3 ;# 2*Heps2 addpd xmm7, xmm7 addpd xmm11, xmm11 addpd xmm3, xmm2 ;# 2*Heps2+Geps addpd xmm7, xmm6 addpd xmm11, xmm10 addpd xmm3, xmm1 ;# FF = Fp + 2*Heps2 + Geps addpd xmm7, xmm5 addpd xmm11, xmm9 mulpd xmm1, xmm12 ;# eps*Fp mulpd xmm5, xmm13 mulpd xmm9, xmm14 movapd xmm12, [rsp + nb334_qqMH] movapd xmm13, [rsp + nb334_qqMM] addpd xmm1, xmm0 ;# VV addpd xmm5, xmm4 addpd xmm9, xmm8 mulpd xmm1, xmm12 ;# VV*qq = vcoul mulpd xmm5, xmm12 mulpd xmm9, xmm13 mulpd xmm3, xmm12 ;# FF*qq = fij mulpd xmm7, xmm12 mulpd xmm11, xmm13 ;# accumulate vctot addpd xmm1, [rsp + nb334_vctot] addpd xmm5, xmm9 addpd xmm1, xmm5 movapd [rsp + nb334_vctot], xmm1 movapd xmm10, [rsp + nb334_tsc] mulpd xmm3, xmm10 ;# fscal mulpd xmm7, xmm10 mulpd xmm10, xmm11 xorpd xmm4, xmm4 xorpd xmm8, xmm8 xorpd xmm11, xmm11 subpd xmm4, xmm3 subpd xmm8, xmm7 subpd xmm11, xmm10 mulpd xmm4, [rsp + nb334_rinvH1M] mulpd xmm8, [rsp + nb334_rinvH2M] mulpd xmm11, [rsp + nb334_rinvMM] ;# move j M forces to xmm0-xmm2 mov rdi, [rbp + nb334_faction] movlpd xmm0, [rdi + rax*8 + 72] movlpd xmm1, [rdi + rax*8 + 80] movlpd xmm2, [rdi + rax*8 + 88] movhpd xmm0, [rdi + rbx*8 + 72] movhpd xmm1, [rdi + rbx*8 + 80] movhpd xmm2, [rdi + rbx*8 + 88] movapd xmm3, xmm4 movapd xmm5, xmm4 movapd xmm7, xmm8 movapd xmm9, xmm8 movapd xmm10, xmm11 movapd xmm12, xmm11 mulpd xmm3, [rsp + nb334_dxH1M] mulpd xmm4, [rsp + nb334_dyH1M] mulpd xmm5, [rsp + nb334_dzH1M] mulpd xmm7, [rsp + nb334_dxH2M] mulpd xmm8, [rsp + nb334_dyH2M] mulpd xmm9, [rsp + nb334_dzH2M] mulpd xmm10, [rsp + nb334_dxMM] mulpd xmm11, [rsp + nb334_dyMM] mulpd xmm12, [rsp + nb334_dzMM] addpd xmm0, xmm3 addpd xmm1, xmm4 addpd xmm2, xmm5 addpd xmm3, [rsp + nb334_fixH1] addpd xmm4, [rsp + nb334_fiyH1] addpd xmm5, [rsp + nb334_fizH1] addpd xmm0, xmm7 addpd xmm1, xmm8 addpd xmm2, xmm9 addpd xmm7, [rsp + nb334_fixH2] addpd xmm8, [rsp + nb334_fiyH2] addpd xmm9, [rsp + nb334_fizH2] addpd xmm0, xmm10 addpd xmm1, xmm11 addpd xmm2, xmm12 addpd xmm10, [rsp + nb334_fixM] addpd xmm11, [rsp + nb334_fiyM] addpd xmm12, [rsp + nb334_fizM] movapd [rsp + nb334_fixH1], xmm3 movapd [rsp + nb334_fiyH1], xmm4 movapd [rsp + nb334_fizH1], xmm5 movapd [rsp + nb334_fixH2], xmm7 movapd [rsp + nb334_fiyH2], xmm8 movapd [rsp + nb334_fizH2], xmm9 movapd [rsp + nb334_fixM], xmm10 movapd [rsp + nb334_fiyM], xmm11 movapd [rsp + nb334_fizM], xmm12 ;# store back j M forces from xmm0-xmm2 movlpd [rdi + rax*8 + 72], xmm0 movlpd [rdi + rax*8 + 80], xmm1 movlpd [rdi + rax*8 + 88], xmm2 movhpd [rdi + rbx*8 + 72], xmm0 movhpd [rdi + rbx*8 + 80], xmm1 movhpd [rdi + rbx*8 + 88], xmm2 ;# should we do one more iteration? sub dword ptr [rsp + nb334_innerk], 2 jl .nb334_checksingle jmp .nb334_unroll_loop.nb334_checksingle: mov edx, [rsp + nb334_innerk] and edx, 1 jnz .nb334_dosingle jmp .nb334_updateouterdata.nb334_dosingle: mov rdx, [rsp + nb334_innerjjnr] ;# pointer to jjnr[k] mov eax, [rdx] mov rsi, [rbp + nb334_pos] ;# base of pos[] lea rax, [rax + rax*2] ;# replace jnr with j3 ;# load j O coordinates movsd xmm4, [rsi + rax*8] movsd xmm5, [rsi + rax*8 + 8] movsd xmm6, [rsi + rax*8 + 16] ;# xmm4 = Ox ;# xmm5 = Oy ;# xmm6 = Oz subsd xmm4, [rsp + nb334_ixO] subsd xmm5, [rsp + nb334_iyO] subsd xmm6, [rsp + nb334_izO] ;# store dx/dy/dz movapd xmm13, xmm4 movapd xmm14, xmm5 movapd xmm15, xmm6 ;# square it mulsd xmm4, xmm4 mulsd xmm5, xmm5 mulsd xmm6, xmm6 addsd xmm4, xmm5 addsd xmm4, xmm6 ;# rsq in xmm4 cvtsd2ss xmm5, xmm4 rsqrtss xmm5, xmm5 cvtss2sd xmm2, xmm5 ;# lu in low xmm2 ;# lookup seed in xmm2 movapd xmm5, xmm2 ;# copy of lu mulsd xmm2, xmm2 ;# lu*lu movapd xmm1, [rsp + nb334_three] mulsd xmm2, xmm4 ;# rsq*lu*lu movapd xmm0, [rsp + nb334_half] subsd xmm1, xmm2 ;# 30-rsq*lu*lu mulsd xmm1, xmm5 mulsd xmm1, xmm0 ;# xmm0=iter1 of rinv (new lu) movapd xmm5, xmm1 ;# copy of lu mulsd xmm1, xmm1 ;# lu*lu movapd xmm2, [rsp + nb334_three] mulsd xmm1, xmm4 ;# rsq*lu*lu movapd xmm0, [rsp + nb334_half] subsd xmm2, xmm1 ;# 30-rsq*lu*lu mulsd xmm2, xmm5 mulsd xmm2, xmm0 ;# xmm0=iter2 of rinv (new lu) mulsd xmm4, xmm2 ;# xmm4=r mulsd xmm4, [rsp + nb334_tsc] cvttsd2si r10d, xmm4 ;# mm6 = lu idx cvtsi2sd xmm5, r10d subsd xmm4, xmm5 movapd xmm1, xmm4 ;# xmm1=eps ;# xmm2=rinv movapd xmm3, xmm4 ;# eps shl r10d, 2 ;# idx *= 4 mov rsi, [rbp + nb334_VFtab] ;# multiply by 3 lea r10, [r10 + r10*2] ;# indices in r10, r11. Load dispersion and repulsion tables in parallel. ;# NB: We are using a combined (coul+lj) table, so LJ data is offest ;# 4*8=32 bytes. movsd xmm4, [rsi + r10*8 + 32] movsd xmm5, [rsi + r10*8 + 40] movsd xmm6, [rsi + r10*8 + 48] movsd xmm7, [rsi + r10*8 + 56] movsd xmm8, [rsi + r10*8 + 64] movsd xmm9, [rsi + r10*8 + 72] movsd xmm10, [rsi + r10*8 + 80] movsd xmm11, [rsi + r10*8 + 88] ;# tables ready, in xmm4-xmm7 and xmm8-xmm11 mulsd xmm7, xmm1 ;# Heps mulsd xmm11, xmm1 mulsd xmm6, xmm1 ;# Geps mulsd xmm10, xmm1 mulsd xmm7, xmm1 ;# Heps2 mulsd xmm11, xmm1 addsd xmm5, xmm6 ;# F+Geps addsd xmm9, xmm10 addsd xmm5, xmm7 ;# F+Geps+Heps2 = Fp addsd xmm9, xmm11 addsd xmm7, xmm7 ;# 2*Heps2 addsd xmm11, xmm11 addsd xmm7, xmm6 ;# 2*Heps2+Geps addsd xmm11, xmm10 addsd xmm7, xmm5 ;# FF = Fp + 2*Heps2 + Geps addsd xmm11, xmm9 mulsd xmm5, xmm1 ;# eps*Fp mulsd xmm9, xmm1 addsd xmm5, xmm4 ;# VV addsd xmm9, xmm8 mulsd xmm5, [rsp + nb334_c6] ;# VV*c6 = vnb6 mulsd xmm9, [rsp + nb334_c12] ;# VV*c12 = vnb12 addsd xmm5, xmm9 addsd xmm5, [rsp + nb334_Vvdwtot] movsd [rsp + nb334_Vvdwtot], xmm5 mulsd xmm7, [rsp + nb334_c6] ;# FF*c6 = fnb6 mulsd xmm11, [rsp + nb334_c12] ;# FF*c12 = fnb12 addsd xmm7, xmm11 mulsd xmm7, [rsp + nb334_tsc] mulsd xmm7, xmm2 xorpd xmm9, xmm9 subsd xmm9, xmm7 mulsd xmm13, xmm9 mulsd xmm14, xmm9 mulsd xmm15, xmm9 movapd xmm0, [rsp + nb334_fixO] movapd xmm1, [rsp + nb334_fiyO] movapd xmm2, [rsp + nb334_fizO] ;# accumulate i forces addsd xmm0, xmm13 addsd xmm1, xmm14 addsd xmm2, xmm15 movsd [rsp + nb334_fixO], xmm0 movsd [rsp + nb334_fiyO], xmm1 movsd [rsp + nb334_fizO], xmm2 ;# the fj's - start by accumulating forces from memory mov rdi, [rbp + nb334_faction] movsd xmm3, [rdi + rax*8] movsd xmm4, [rdi + rax*8 + 8] movsd xmm5, [rdi + rax*8 + 16] addsd xmm3, xmm13 addsd xmm4, xmm14 addsd xmm5, xmm15 movsd [rdi + rax*8], xmm3 movsd [rdi + rax*8 + 8], xmm4 movsd [rdi + rax*8 + 16], xmm5 ;# done with OO interaction ;# move j H1 coordinates to local temp variables mov rsi, [rbp + nb334_pos] movsd xmm0, [rsi + rax*8 + 24] movsd xmm1, [rsi + rax*8 + 32] movsd xmm2, [rsi + rax*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 subsd xmm0, [rsp + nb334_ixH1] subsd xmm1, [rsp + nb334_iyH1] subsd xmm2, [rsp + nb334_izH1] subsd xmm3, [rsp + nb334_ixH2] subsd xmm4, [rsp + nb334_iyH2] subsd xmm5, [rsp + nb334_izH2] subsd xmm6, [rsp + nb334_ixM] subsd xmm7, [rsp + nb334_iyM] subsd xmm8, [rsp + nb334_izM] movapd [rsp + nb334_dxH1H1], xmm0 movapd [rsp + nb334_dyH1H1], xmm1 movapd [rsp + nb334_dzH1H1], xmm2 mulsd xmm0, xmm0 mulsd xmm1, xmm1 mulsd xmm2, xmm2 movapd [rsp + nb334_dxH2H1], xmm3 movapd [rsp + nb334_dyH2H1], xmm4 movapd [rsp + nb334_dzH2H1], xmm5 mulsd xmm3, xmm3 mulsd xmm4, xmm4 mulsd xmm5, xmm5 movapd [rsp + nb334_dxMH1], xmm6 movapd [rsp + nb334_dyMH1], xmm7 movapd [rsp + nb334_dzMH1], xmm8 mulsd xmm6, xmm6 mulsd xmm7, xmm7 mulsd xmm8, xmm8 addsd xmm0, xmm1 addsd xmm0, xmm2 addsd xmm3, xmm4 addsd xmm3, xmm5 addsd xmm6, xmm7 addsd xmm6, xmm8 ;# start doing invsqrt for jH1 atoms cvtsd2ss xmm1, xmm0 cvtsd2ss xmm4, xmm3 cvtsd2ss xmm7, xmm6 rsqrtss xmm1, xmm1 rsqrtss xmm4, xmm4 rsqrtss xmm7, xmm7 cvtss2sd xmm1, xmm1 cvtss2sd xmm4, xmm4 cvtss2sd xmm7, xmm7 movapd xmm2, xmm1 movapd xmm5, xmm4 movapd xmm8, xmm7 mulsd xmm1, xmm1 ;# lu*lu mulsd xmm4, xmm4 ;# lu*lu mulsd xmm7, xmm7 ;# lu*lu movapd xmm9, [rsp + nb334_three] movapd xmm10, xmm9 movapd xmm11, xmm9 mulsd xmm1, xmm0 ;# rsq*lu*lu mulsd xmm4, xmm3 ;# rsq*lu*lu mulsd xmm7, xmm6 ;# rsq*lu*lu subsd xmm9, xmm1 subsd xmm10, xmm4 subsd xmm11, xmm7 ;# 3-rsq*lu*lu mulsd xmm9, xmm2 mulsd xmm10, xmm5 mulsd xmm11, xmm8 ;# lu*(3-rsq*lu*lu) movapd xmm15, [rsp + nb334_half] mulsd xmm9, xmm15 ;# first iteration for rinvH1H1 ⌨️ 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?