nb_kernel332_x86_64_sse.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,305 行 · 第 1/5 页
S
2,305 行
mulps xmm11, xmm4 ;# rinvH2H2 movaps [rsp + nb332_rinvOH2], xmm9 movaps [rsp + nb332_rinvH1H2], xmm10 movaps [rsp + nb332_rinvH2H2], xmm11 ;# H2 interactions ;# rsq in xmm0,xmm3,xmm6 ;# rinv in xmm9, xmm10, xmm11 movaps xmm1, [rsp + nb332_tsc] mulps xmm0, xmm9 ;# r mulps xmm3, xmm10 mulps xmm6, xmm11 mulps xmm0, xmm1 ;# rtab mulps xmm3, xmm1 mulps xmm6, xmm1 ;# truncate and convert to integers cvttps2dq xmm1, xmm0 cvttps2dq xmm4, xmm3 cvttps2dq xmm7, xmm6 ;# convert back to float cvtdq2ps xmm2, xmm1 cvtdq2ps xmm5, xmm4 cvtdq2ps xmm8, xmm7 ;# multiply by 4 pslld xmm1, 2 pslld xmm4, 2 pslld xmm7, 2 ;# multiply by three (copy, mult. by two, add back) movaps xmm10, xmm1 movaps xmm11, xmm4 movaps xmm12, xmm7 pslld xmm1, 1 pslld xmm4, 1 pslld xmm7, 1 paddd xmm1, xmm10 paddd xmm4, xmm11 paddd xmm7, xmm12 ;# move to integer registers movhlps xmm13, xmm1 movhlps xmm14, xmm4 movhlps xmm15, xmm7 movd eax, xmm1 movd r8d, xmm4 movd r12d, xmm7 movd ecx, xmm13 movd r10d, xmm14 movd r14d, xmm15 pshufd xmm1, xmm1, 1 pshufd xmm4, xmm4, 1 pshufd xmm7, xmm7, 1 pshufd xmm13, xmm13, 1 pshufd xmm14, xmm14, 1 pshufd xmm15, xmm15, 1 movd ebx, xmm1 movd r9d, xmm4 movd r13d, xmm7 movd edx, xmm13 movd r11d, xmm14 movd r15d, xmm15 mov rsi, [rbp + nb332_VFtab] ;# calculate eps subps xmm0, xmm2 subps xmm3, xmm5 subps xmm6, xmm8 movaps [rsp + nb332_epsO], xmm0 movaps [rsp + nb332_epsH1], xmm3 movaps [rsp + nb332_epsH2], xmm6 ;# Load LOTS of table data movlps xmm1, [rsi + rax*4] movlps xmm5, [rsi + r8*4] movlps xmm9, [rsi + r12*4] movlps xmm3, [rsi + rcx*4] movlps xmm7, [rsi + r10*4] movlps xmm11, [rsi + r14*4] movhps xmm1, [rsi + rbx*4] movhps xmm5, [rsi + r9*4] movhps xmm9, [rsi + r13*4] movhps xmm3, [rsi + rdx*4] movhps xmm7, [rsi + r11*4] movhps xmm11, [rsi + r15*4] movaps xmm0, xmm1 movaps xmm4, xmm5 movaps xmm8, xmm9 shufps xmm0, xmm3, 136 ;# 10001000 shufps xmm4, xmm7, 136 ;# 10001000 shufps xmm8, xmm11, 136 ;# 10001000 shufps xmm1, xmm3, 221 ;# 11011101 shufps xmm5, xmm7, 221 ;# 11011101 shufps xmm9, xmm11, 221 ;# 11011101 movlps xmm3, [rsi + rax*4 + 8] movlps xmm7, [rsi + r8*4 + 8] movlps xmm11, [rsi + r12*4 + 8] movlps xmm12, [rsi + rcx*4 + 8] movlps xmm13, [rsi + r10*4 + 8] movlps xmm14, [rsi + r14*4 + 8] movhps xmm3, [rsi + rbx*4 + 8] movhps xmm7, [rsi + r9*4 + 8] movhps xmm11, [rsi + r13*4 + 8] movhps xmm12, [rsi + rdx*4 + 8] movhps xmm13, [rsi + r11*4 + 8] movhps xmm14, [rsi + r15*4 + 8] movaps xmm2, xmm3 movaps xmm6, xmm7 movaps xmm10, xmm11 shufps xmm2, xmm12, 136 ;# 10001000 shufps xmm6, xmm13, 136 ;# 10001000 shufps xmm10, xmm14, 136 ;# 10001000 shufps xmm3, xmm12, 221 ;# 11011101 shufps xmm7, xmm13, 221 ;# 11011101 shufps xmm11, xmm14, 221 ;# 11011101 ;# table data ready in xmm0-xmm3 , xmm4-xmm7 , and xmm8-xmm11 movaps xmm12, [rsp + nb332_epsO] movaps xmm13, [rsp + nb332_epsH1] movaps xmm14, [rsp + nb332_epsH2] mulps xmm3, xmm12 ;# Heps mulps xmm7, xmm13 mulps xmm11, xmm14 mulps xmm2, xmm12 ;# Geps mulps xmm6, xmm13 mulps xmm10, xmm14 mulps xmm3, xmm12 ;# Heps2 mulps xmm7, xmm13 mulps xmm11, xmm14 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, xmm13 mulps xmm9, xmm14 movaps xmm12, [rsp + nb332_qqOH] movaps xmm13, [rsp + nb332_qqHH] addps xmm1, xmm0 ;# VV addps xmm5, xmm4 addps xmm9, xmm8 mulps xmm1, xmm12 ;# VV*qq = vcoul mulps xmm5, xmm13 mulps xmm9, xmm13 mulps xmm3, xmm12 ;# FF*qq = fij mulps xmm7, xmm13 mulps xmm11, xmm13 ;# accumulate vctot addps xmm1, [rsp + nb332_vctot] addps xmm5, xmm9 addps xmm1, xmm5 movaps [rsp + nb332_vctot], xmm1 movaps xmm10, [rsp + nb332_tsc] mulps xmm3, xmm10 ;# fscal mulps xmm7, xmm10 mulps xmm10, xmm11 movd eax, mm0 ;# restore j3 from mm0-mm3 movd ebx, mm1 movd ecx, mm2 movd edx, mm3 ;# move j H2 forces to local temp variables mov rdi, [rbp + nb332_faction] movlps xmm11, [rdi + rax*4 + 24] ;# jxH2a jyH2a - - movlps xmm12, [rdi + rcx*4 + 24] ;# jxH2c jyH2c - - movhps xmm11, [rdi + rbx*4 + 24] ;# jxH2a jyH2a jxH2b jyH2b movhps xmm12, [rdi + rdx*4 + 24] ;# jxH2c jyH2c jxH2d jyH2d movss xmm13, [rdi + rax*4 + 32] ;# jzH2a - - - movss xmm14, [rdi + rcx*4 + 32] ;# jzH2c - - - movss xmm1, [rdi + rbx*4 + 32] ;# jzH2b - - - movss xmm2, [rdi + rdx*4 + 32] ;# jzH2d - - - movlhps xmm13, xmm1 ;# jzH2a - jzH2b - movlhps xmm14, xmm2 ;# jzH2c - jzH2d - shufps xmm13, xmm14, 136 ;# 10001000 => jzH2a jzH2b jzH2c jzH2d ;# xmm11: jxH2a jyH2a jxH2b jyH2b ;# xmm12: jxH2c jyH2c jxH2d jyH2d ;# xmm13: jzH2a jzH2b jzH2c jzH2d xorps xmm0, xmm0 xorps xmm4, xmm4 xorps xmm8, xmm8 mulps xmm3, [rsp + nb332_rinvOH2] mulps xmm7, [rsp + nb332_rinvH1H2] mulps xmm10, [rsp + nb332_rinvH2H2] subps xmm0, xmm3 subps xmm4, xmm7 subps xmm8, xmm10 movaps xmm1, xmm0 movaps xmm2, xmm0 movaps xmm3, xmm4 movaps xmm5, xmm4 movaps xmm6, xmm8 movaps xmm7, xmm8 mulps xmm0, [rsp + nb332_dxOH2] mulps xmm1, [rsp + nb332_dyOH2] mulps xmm2, [rsp + nb332_dzOH2] mulps xmm3, [rsp + nb332_dxH1H2] mulps xmm4, [rsp + nb332_dyH1H2] mulps xmm5, [rsp + nb332_dzH1H2] mulps xmm6, [rsp + nb332_dxH2H2] mulps xmm7, [rsp + nb332_dyH2H2] mulps xmm8, [rsp + nb332_dzH2H2] movaps xmm14, xmm0 movaps xmm15, xmm1 addps xmm13, xmm2 addps xmm0, [rsp + nb332_fixO] addps xmm1, [rsp + nb332_fiyO] addps xmm2, [rsp + nb332_fizO] addps xmm14, xmm3 addps xmm15, xmm4 addps xmm13, xmm5 addps xmm3, [rsp + nb332_fixH1] addps xmm4, [rsp + nb332_fiyH1] addps xmm5, [rsp + nb332_fizH1] addps xmm14, xmm6 addps xmm15, xmm7 addps xmm13, xmm8 addps xmm6, [rsp + nb332_fixH2] addps xmm7, [rsp + nb332_fiyH2] addps xmm8, [rsp + nb332_fizH2] movaps [rsp + nb332_fixO], xmm0 movaps [rsp + nb332_fiyO], xmm1 movaps [rsp + nb332_fizO], xmm2 movaps [rsp + nb332_fixH1], xmm3 movaps [rsp + nb332_fiyH1], xmm4 movaps [rsp + nb332_fizH1], xmm5 movaps [rsp + nb332_fixH2], xmm6 movaps [rsp + nb332_fiyH2], xmm7 movaps [rsp + nb332_fizH2], xmm8 ;# xmm11 = fH2x ;# xmm12 = fH2y ;# xmm13 = fH2z movaps xmm0, xmm14 unpcklps xmm14, xmm15 unpckhps xmm0, xmm15 addps xmm11, xmm14 addps xmm12, xmm0 movhlps xmm14, xmm13 ;# fH2zc fH2zd movlps [rdi + rax*4 + 24], xmm11 movhps [rdi + rbx*4 + 24], xmm11 movlps [rdi + rcx*4 + 24], xmm12 movhps [rdi + rdx*4 + 24], xmm12 movss [rdi + rax*4 + 32], xmm13 movss [rdi + rcx*4 + 32], xmm14 shufps xmm13, xmm13, 1 shufps xmm14, xmm14, 1 movss [rdi + rbx*4 + 32], xmm13 movss [rdi + rdx*4 + 32], xmm14 ;# should we do one more iteration? sub dword ptr [rsp + nb332_innerk], 4 jl .nb332_single_check jmp .nb332_unroll_loop.nb332_single_check: add dword ptr [rsp + nb332_innerk], 4 jnz .nb332_single_loop jmp .nb332_updateouterdata.nb332_single_loop: mov rdx, [rsp + nb332_innerjjnr] ;# pointer to jjnr[k] mov eax, [rdx] add qword ptr [rsp + nb332_innerjjnr], 4 mov rsi, [rbp + nb332_pos] lea rax, [rax + rax*2] ;# fetch j coordinates xorps xmm0, xmm0 xorps xmm1, xmm1 xorps xmm2, xmm2 movss xmm0, [rsi + rax*4] ;# jxO - - - movss xmm1, [rsi + rax*4 + 4] ;# jyO - - - movss xmm2, [rsi + rax*4 + 8] ;# jzO - - - movlps xmm6, [rsi + rax*4 + 12] ;# xmm6 = jxH1 jyH1 - - movss xmm7, [rsi + rax*4 + 20] ;# xmm7 = jzH1 - - - movhps xmm6, [rsi + rax*4 + 24] ;# xmm6 = jxH1 jyH1 jxH2 jyH2 movss xmm5, [rsi + rax*4 + 32] ;# xmm5 = jzH2 - - - ;# have all coords, time for some shuffling. shufps xmm6, xmm6, 216 ;# 11011000 ;# xmm6 = jxH1 jxH2 jyH1 jyH2 unpcklps xmm7, xmm5 ;# xmm7 = jzH1 jzH2 - - movlhps xmm0, xmm6 ;# xmm0 = jxO 0 jxH1 jxH2 shufps xmm1, xmm6, 228 ;# 11100100 ;# xmm1 = jyO 0 jyH1 jyH2 shufps xmm2, xmm7, 68 ;# 01000100 ;# xmm2 = jzO 0 jzH1 jzH2 ;# store all j coordinates in jO movaps [rsp + nb332_jxO], xmm0 movaps [rsp + nb332_jyO], xmm1 movaps [rsp + nb332_jzO], xmm2 subps xmm0, [rsp + nb332_ixO] subps xmm1, [rsp + nb332_iyO] subps xmm2, [rsp + nb332_izO] movaps [rsp + nb332_dxOO], xmm0 movaps [rsp + nb332_dyOO], xmm1 movaps [rsp + nb332_dzOO], xmm2 mulps xmm0, xmm0 mulps xmm1, xmm1 mulps xmm2, xmm2 addps xmm0, xmm1 addps xmm0, xmm2 ;# have rsq in xmm0 ;# do invsqrt rsqrtps xmm1, xmm0 movaps xmm2, xmm1 mulps xmm1, xmm1 movaps xmm3, [rsp + nb332_three] mulps xmm1, xmm0 subps xmm3, xmm1 mulps xmm3, xmm2 mulps xmm3, [rsp + nb332_half] ;# rinv iO - j water movaps xmm1, xmm3 mulps xmm1, xmm0 ;# xmm1=r movaps xmm0, xmm3 ;# xmm0=rinv mulps xmm1, [rsp + nb332_tsc] movhlps xmm2, xmm1 cvttps2pi mm6, xmm1 cvttps2pi mm7, xmm2 ;# mm6/mm7 contain lu indices cvtpi2ps xmm3, mm6 cvtpi2ps xmm2, mm7 movlhps xmm3, xmm2 subps xmm1, xmm3 ;# xmm1=eps movaps xmm2, xmm1 mulps xmm2, xmm2 ;# xmm2=eps2 pslld mm6, 2 pslld mm7, 2 movd ebx, mm6 movd ecx, mm7 psrlq mm7, 32 movd edx, mm7 ;# table indices in ebx,ecx,edx mov rsi, [rbp + nb332_VFtab] lea rbx, [rbx + rbx*2] lea rcx, [rcx + rcx*2] lea rdx, [rdx + rdx*2] movlps xmm5, [rsi + rbx*4] movlps xmm7, [rsi + rcx*4] movhps xmm7, [rsi + rdx*4] ;# got half coulomb table movaps xmm4, xmm5 shufps xmm4, xmm7, 136 ;# 10001000 shufps xmm5, xmm7, 221 ;# 11011101 movlps xmm7, [rsi + rbx*4 + 8] movlps xmm3, [rsi + rcx*4 + 8] movhps xmm3, [rsi + rdx*4 + 8] ;# other half of coulomb table movaps xmm6, xmm7 shufps xmm6, xmm3, 136 ;# 10001000 shufps xmm7, xmm3, 221 ;# 11011101 ;# coulomb table ready, in xmm4-xmm7 mulps xmm6, xmm1 ;# xmm6=Geps mulps xmm7, xmm2 ;# xmm7=Heps2 addps xmm5, xmm6 addps xmm5, xmm7 ;# xmm5=Fp mulps xmm7, [rsp + nb332_two] ;# two*Heps2 xorps xmm3, xmm3 ;# fetch charges to xmm3 (temporary) movss xmm3, [rsp + nb332_qqOO] movhps xmm3, [rsp + nb332_qqOH] addps xmm7, xmm6 addps xmm7, xmm5 ;# xmm7=FF mulps xmm5, xmm1 ;# xmm5=eps*Fp addps xmm5, xmm4 ;# xmm5=VV mulps xmm5, xmm3 ;# vcoul=qq*VV mulps xmm3, xmm7 ;# fijC=FF*qq ;# at this point xmm5 contains vcoul and xmm3 fijC addps xmm5, [rsp + nb332_vctot] movaps [rsp + nb332_vctot], xmm5 ;# put scalar force on stack temporarily movaps [rsp + nb332_fstmp], xmm3 ;# dispersion movss xmm4, [rsi + rbx*4 + 16] movss xmm5, [rsi + rbx*4 + 20] movss xmm6, [rsi + rbx*4 + 24] movss xmm7, [rsi + rbx*4 + 28] ;# dispersion table ready, in xmm4-xmm7 mulss xmm6, xmm1 ;# xmm6=Geps mulss xmm7, xmm2 ;# xmm7=Heps2 addss xmm5, xmm6 addss xmm5, xmm7 ;# xmm5=Fp mulss xmm7, [rsp + nb332_two] ;# two*Heps2 addss xmm7, xmm6 addss xmm7, xmm5 ;# xmm7=FF mulss xmm5, xmm1 ;# xmm5=eps*Fp addss xmm5, xmm4 ;# xmm5=VV xorps xmm4, xmm4 movss xmm4, [rsp + nb332_c6] mulps xmm7, xmm4 ;# fijD mulps xmm5, xmm4 ;# Vvdw6 addps xmm7, [rsp + nb332_fstmp] ;# add to fscal ;# put scalar force on stack Update Vvdwtot directly addps xmm5, [rsp + nb332_Vvdwtot] movaps [rsp + nb332_fstmp], xmm7 movaps [rsp + nb332_Vvdwtot], xmm5 ;# repulsion movss xmm4, [rsi + rbx*4 + 32] movss xmm5, [rsi + rbx*4 + 36] movss xmm6, [rsi + rbx*4 + 40] movss xmm7, [rsi + rbx*4 + 44] ;# table ready, in xmm4-xmm7 mulss xmm6, xmm1 ;# xmm6=Geps mulss xmm7, xmm2 ;# xmm7=Heps2 addss xmm5, xmm6⌨️ 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?