nb_kernel214_x86_64_sse.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,140 行 · 第 1/5 页
S
2,140 行
movl (%rdx),%eax addq $4,nb214_innerjjnr(%rsp) movq nb214_pos(%rbp),%rsi lea (%rax,%rax,2),%rax ## fetch j coordinates movlps (%rsi,%rax,4),%xmm3 ## Ox Oy movlps 16(%rsi,%rax,4),%xmm4 ## H1y H1z movlps 32(%rsi,%rax,4),%xmm5 ## H2z Mx movhps 8(%rsi,%rax,4),%xmm3 ## Ox Oy Oz H1x movhps 24(%rsi,%rax,4),%xmm4 ## H1y H1z H2x H2y movhps 40(%rsi,%rax,4),%xmm5 ## H2z Mx My Mz ## transpose movaps %xmm4,%xmm0 movaps %xmm3,%xmm1 movaps %xmm4,%xmm2 movaps %xmm3,%xmm6 shufps $18,%xmm5,%xmm4 ## (00010010) h2x - Mx - shufps $193,%xmm0,%xmm3 ## (11000001) Oy - H1y - shufps $35,%xmm5,%xmm2 ## (00100011) H2y - My - shufps $18,%xmm0,%xmm1 ## (00010010) Oz - H1z - ## xmm6: Ox - - H1x xmm5: H2z - - Mz shufps $140,%xmm4,%xmm6 ## (10001100) Ox H1x H2x Mx shufps $136,%xmm2,%xmm3 ## (10001000) Oy H1y H2y My shufps $200,%xmm5,%xmm1 ## (11001000) Oz H1z H2z Mz ## store all j coordinates in jO movaps %xmm6,nb214_jxO(%rsp) movaps %xmm3,nb214_jyO(%rsp) movaps %xmm1,nb214_jzO(%rsp) movaps %xmm6,%xmm0 ## jxO movaps %xmm1,%xmm2 ## jzO movaps %xmm3,%xmm1 ## jyO movaps %xmm3,%xmm4 ## jyO movaps %xmm6,%xmm3 ## jxO movaps %xmm2,%xmm5 ## jzO ## do O and M in parallel subps nb214_ixO(%rsp),%xmm0 subps nb214_iyO(%rsp),%xmm1 subps nb214_izO(%rsp),%xmm2 subps nb214_ixM(%rsp),%xmm3 subps nb214_iyM(%rsp),%xmm4 subps nb214_izM(%rsp),%xmm5 movaps %xmm0,nb214_dxOO(%rsp) movaps %xmm1,nb214_dyOO(%rsp) movaps %xmm2,nb214_dzOO(%rsp) movaps %xmm3,nb214_dxMM(%rsp) movaps %xmm4,nb214_dyMM(%rsp) movaps %xmm5,nb214_dzMM(%rsp) mulps %xmm0,%xmm0 mulps %xmm1,%xmm1 mulps %xmm2,%xmm2 addps %xmm1,%xmm0 addps %xmm2,%xmm0 ## have rsq in xmm0 mulps %xmm3,%xmm3 mulps %xmm4,%xmm4 mulps %xmm5,%xmm5 addps %xmm3,%xmm4 addps %xmm5,%xmm4 ## have rsq in xmm4 ## Save M data movaps %xmm4,nb214_rsqMM(%rsp) ## do 1/x for O and 1/sqrt(x) for M rcpss %xmm0,%xmm1 rsqrtps %xmm4,%xmm5 movss nb214_two(%rsp),%xmm2 movaps %xmm5,%xmm6 mulss %xmm1,%xmm0 mulps %xmm5,%xmm5 subss %xmm0,%xmm2 movaps nb214_three(%rsp),%xmm7 mulss %xmm1,%xmm2 ## 1/r2 mulps %xmm4,%xmm5 movss %xmm2,%xmm0 subps %xmm5,%xmm7 mulss %xmm2,%xmm2 mulps %xmm6,%xmm7 mulss %xmm0,%xmm2 ## 1/r6 mulps nb214_half(%rsp),%xmm7 ## rinv iH1 - j water movss %xmm2,%xmm1 movaps %xmm7,nb214_rinvMM(%rsp) mulss %xmm2,%xmm2 ## 1/r12 mulss nb214_c6(%rsp),%xmm1 mulss nb214_c12(%rsp),%xmm2 movss %xmm2,%xmm3 subss %xmm1,%xmm3 addss nb214_Vvdwtot(%rsp),%xmm3 movss %xmm3,nb214_Vvdwtot(%rsp) mulss nb214_six(%rsp),%xmm1 mulss nb214_twelve(%rsp),%xmm2 subss %xmm1,%xmm2 mulss %xmm2,%xmm0 ## fscal movss %xmm0,%xmm1 movss %xmm0,%xmm2 mulss nb214_dxOO(%rsp),%xmm0 mulss nb214_dyOO(%rsp),%xmm1 mulss nb214_dzOO(%rsp),%xmm2 xorps %xmm3,%xmm3 xorps %xmm4,%xmm4 xorps %xmm5,%xmm5 addss %xmm0,%xmm3 addss %xmm1,%xmm4 addss %xmm2,%xmm5 movaps %xmm3,nb214_fjxO(%rsp) movaps %xmm4,nb214_fjyO(%rsp) movaps %xmm5,nb214_fjzO(%rsp) addss nb214_fixO(%rsp),%xmm0 addss nb214_fiyO(%rsp),%xmm1 addss nb214_fizO(%rsp),%xmm2 movss %xmm0,nb214_fixO(%rsp) movss %xmm1,nb214_fiyO(%rsp) movss %xmm2,nb214_fizO(%rsp) ## do M coulomb interaction movaps nb214_rinvMM(%rsp),%xmm0 movaps %xmm0,%xmm7 ## xmm7=rinv movaps nb214_krf(%rsp),%xmm5 mulps %xmm0,%xmm0 ## xmm0=rinvsq ## fetch charges to xmm3 (temporary) xorps %xmm3,%xmm3 movss nb214_qqMH(%rsp),%xmm3 movhps nb214_qqMM(%rsp),%xmm3 shufps $193,%xmm3,%xmm3 ## 11000001 mulps nb214_rsqMM(%rsp),%xmm5 ## xmm5=krsq movaps %xmm5,%xmm6 addps %xmm7,%xmm6 ## xmm6=rinv+ krsq subps nb214_crf(%rsp),%xmm6 mulps %xmm3,%xmm6 ## xmm6=voul=qq*(rinv+ krsq-crf) mulps nb214_two(%rsp),%xmm5 subps %xmm5,%xmm7 ## xmm7=rinv-2*krsq mulps %xmm3,%xmm7 ## xmm7 = coul part of fscal addps nb214_vctot(%rsp),%xmm6 movaps %xmm6,nb214_vctot(%rsp) mulps %xmm7,%xmm0 movaps %xmm0,%xmm1 movaps %xmm0,%xmm2 mulps nb214_dxMM(%rsp),%xmm0 mulps nb214_dyMM(%rsp),%xmm1 mulps nb214_dzMM(%rsp),%xmm2 ## update forces M - j water movaps nb214_fjxO(%rsp),%xmm3 movaps nb214_fjyO(%rsp),%xmm4 movaps nb214_fjzO(%rsp),%xmm5 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm5 movaps %xmm3,nb214_fjxO(%rsp) movaps %xmm4,nb214_fjyO(%rsp) movaps %xmm5,nb214_fjzO(%rsp) addps nb214_fixM(%rsp),%xmm0 addps nb214_fiyM(%rsp),%xmm1 addps nb214_fizM(%rsp),%xmm2 movaps %xmm0,nb214_fixM(%rsp) movaps %xmm1,nb214_fiyM(%rsp) movaps %xmm2,nb214_fizM(%rsp) ## i H1 & H2 simultaneously first get i particle coords: movaps nb214_jxO(%rsp),%xmm0 movaps nb214_jyO(%rsp),%xmm1 movaps nb214_jzO(%rsp),%xmm2 movaps %xmm0,%xmm3 movaps %xmm1,%xmm4 movaps %xmm2,%xmm5 subps nb214_ixH1(%rsp),%xmm0 subps nb214_iyH1(%rsp),%xmm1 subps nb214_izH1(%rsp),%xmm2 subps nb214_ixH2(%rsp),%xmm3 subps nb214_iyH2(%rsp),%xmm4 subps nb214_izH2(%rsp),%xmm5 movaps %xmm0,nb214_dxH1H1(%rsp) movaps %xmm1,nb214_dyH1H1(%rsp) movaps %xmm2,nb214_dzH1H1(%rsp) movaps %xmm3,nb214_dxH2H2(%rsp) movaps %xmm4,nb214_dyH2H2(%rsp) movaps %xmm5,nb214_dzH2H2(%rsp) mulps %xmm0,%xmm0 mulps %xmm1,%xmm1 mulps %xmm2,%xmm2 mulps %xmm3,%xmm3 mulps %xmm4,%xmm4 mulps %xmm5,%xmm5 addps %xmm1,%xmm0 addps %xmm3,%xmm4 addps %xmm2,%xmm0 ## have rsqH1 in xmm0 addps %xmm5,%xmm4 ## have rsqH2 in xmm4 movaps %xmm0,nb214_rsqH1H1(%rsp) movaps %xmm4,nb214_rsqH2H2(%rsp) ## start doing invsqrt use rsq values in xmm0, xmm4 rsqrtps %xmm0,%xmm1 rsqrtps %xmm4,%xmm5 movaps %xmm1,%xmm2 movaps %xmm5,%xmm6 mulps %xmm1,%xmm1 mulps %xmm5,%xmm5 movaps nb214_three(%rsp),%xmm3 movaps %xmm3,%xmm7 mulps %xmm0,%xmm1 mulps %xmm4,%xmm5 subps %xmm1,%xmm3 subps %xmm5,%xmm7 mulps %xmm2,%xmm3 mulps %xmm6,%xmm7 mulps nb214_half(%rsp),%xmm3 ## rinvH1H1 mulps nb214_half(%rsp),%xmm7 ## rinvH2H2 movaps %xmm3,nb214_rinvH1H1(%rsp) movaps %xmm7,nb214_rinvH2H2(%rsp) ## Do H1 coulomb interaction movaps nb214_rinvH1H1(%rsp),%xmm0 movaps %xmm0,%xmm7 ## xmm7=rinv movaps nb214_krf(%rsp),%xmm5 mulps %xmm0,%xmm0 ## xmm0=rinvsq ## fetch charges to xmm3 (temporary) xorps %xmm3,%xmm3 movss nb214_qqHH(%rsp),%xmm3 movhps nb214_qqMH(%rsp),%xmm3 shufps $193,%xmm3,%xmm3 ## 11000001 mulps nb214_rsqH1H1(%rsp),%xmm5 ## xmm5=krsq movaps %xmm5,%xmm6 addps %xmm7,%xmm6 ## xmm6=rinv+ krsq subps nb214_crf(%rsp),%xmm6 mulps %xmm3,%xmm6 ## xmm6=voul=qq*(rinv+ krsq-crf) mulps nb214_two(%rsp),%xmm5 subps %xmm5,%xmm7 ## xmm7=rinv-2*krsq mulps %xmm3,%xmm7 ## xmm7 = coul part of fscal addps nb214_vctot(%rsp),%xmm6 movaps %xmm6,nb214_vctot(%rsp) mulps %xmm7,%xmm0 movaps %xmm0,%xmm1 movaps %xmm0,%xmm2 mulps nb214_dxH1H1(%rsp),%xmm0 mulps nb214_dyH1H1(%rsp),%xmm1 mulps nb214_dzH1H1(%rsp),%xmm2 ## update forces H1 - j water movaps nb214_fjxO(%rsp),%xmm3 movaps nb214_fjyO(%rsp),%xmm4 movaps nb214_fjzO(%rsp),%xmm5 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm5 movaps %xmm3,nb214_fjxO(%rsp) movaps %xmm4,nb214_fjyO(%rsp) movaps %xmm5,nb214_fjzO(%rsp) addps nb214_fixH1(%rsp),%xmm0 addps nb214_fiyH1(%rsp),%xmm1 addps nb214_fizH1(%rsp),%xmm2 movaps %xmm0,nb214_fixH1(%rsp) movaps %xmm1,nb214_fiyH1(%rsp) movaps %xmm2,nb214_fizH1(%rsp) ## H2 Coulomb movaps nb214_rinvH2H2(%rsp),%xmm0 movaps %xmm0,%xmm7 ## xmm7=rinv movaps nb214_krf(%rsp),%xmm5 mulps %xmm0,%xmm0 ## xmm0=rinvsq ## fetch charges to xmm3 (temporary) xorps %xmm3,%xmm3 movss nb214_qqHH(%rsp),%xmm3 movhps nb214_qqMH(%rsp),%xmm3 shufps $193,%xmm3,%xmm3 ## 11000001 mulps nb214_rsqH2H2(%rsp),%xmm5 ## xmm5=krsq movaps %xmm5,%xmm6 addps %xmm7,%xmm6 ## xmm6=rinv+ krsq subps nb214_crf(%rsp),%xmm6 mulps %xmm3,%xmm6 ## xmm6=voul=qq*(rinv+ krsq-crf) mulps nb214_two(%rsp),%xmm5 subps %xmm5,%xmm7 ## xmm7=rinv-2*krsq mulps %xmm3,%xmm7 ## xmm7 = coul part of fscal addps nb214_vctot(%rsp),%xmm6 ## local vctot summation variable movaps %xmm6,nb214_vctot(%rsp) mulps %xmm7,%xmm0 movaps %xmm0,%xmm1 movaps %xmm0,%xmm2 mulps nb214_dxH2H2(%rsp),%xmm0 mulps nb214_dyH2H2(%rsp),%xmm1 mulps nb214_dzH2H2(%rsp),%xmm2 ## update forces H2 - j water movaps nb214_fjxO(%rsp),%xmm3 movaps nb214_fjyO(%rsp),%xmm4 movaps nb214_fjzO(%rsp),%xmm5 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm5 movaps %xmm3,nb214_fjxO(%rsp) movaps %xmm4,nb214_fjyO(%rsp) movaps %xmm5,nb214_fjzO(%rsp) addps nb214_fixH2(%rsp),%xmm0 addps nb214_fiyH2(%rsp),%xmm1 addps nb214_fizH2(%rsp),%xmm2 movaps %xmm0,nb214_fixH2(%rsp) movaps %xmm1,nb214_fiyH2(%rsp) movaps %xmm2,nb214_fizH2(%rsp) movq nb214_faction(%rbp),%rsi ## update j water forces from local variables. ## transpose back first movaps nb214_fjxO(%rsp),%xmm0 ## Ox H1x H2x Mx movaps nb214_fjyO(%rsp),%xmm1 ## Oy H1y H2y My movaps nb214_fjzO(%rsp),%xmm2 ## Oz H1z H2z Mz movaps %xmm0,%xmm3 movaps %xmm0,%xmm4 unpcklps %xmm1,%xmm3 ## Ox Oy - - shufps $0x1,%xmm2,%xmm4 ## h1x - Oz - movaps %xmm1,%xmm5 movaps %xmm0,%xmm6 unpcklps %xmm2,%xmm5 ## - - H1y H1z unpckhps %xmm1,%xmm6 ## h2x h2y - - unpckhps %xmm2,%xmm1 ## - - My Mz shufps $0x32,%xmm0,%xmm2 ## (00110010) h2z - Mx - shufps $36,%xmm4,%xmm3 ## 00100100 ;# Ox Oy Oz H1x shufps $78,%xmm6,%xmm5 ## 01001110 ;# h1y h1z h2x h2y shufps $232,%xmm1,%xmm2 ## 11101000 ;# h2z mx my mz movlps (%rsi,%rax,4),%xmm0 movlps 16(%rsi,%rax,4),%xmm1 movlps 32(%rsi,%rax,4),%xmm4 movhps 8(%rsi,%rax,4),%xmm0 movhps 24(%rsi,%rax,4),%xmm1 movhps 40(%rsi,%rax,4),%xmm4 addps %xmm3,%xmm0 addps %xmm5,%xmm1 addps %xmm2,%xmm4 movlps %xmm0,(%rsi,%rax,4) movlps %xmm1,16(%rsi,%rax,4) movlps %xmm4,32(%rsi,%rax,4) movhps %xmm0,8(%rsi,%rax,4) movhps %xmm1,24(%rsi,%rax,4) movhps %xmm4,40(%rsi,%rax,4) decl nb214_innerk(%rsp) jz _nb_kernel214_x86_64_sse.nb214_updateouterdata jmp _nb_kernel214_x86_64_sse.nb214_single_loop_nb_kernel214_x86_64_sse.nb214_updateouterdata: movl nb214_ii3(%rsp),%ecx movq nb214_faction(%rbp),%rdi movq nb214_fshift(%rbp),%rsi movl nb214_is3(%rsp),%edx ## accumulate Oi forces in xmm0, xmm1, xmm2 movaps nb214_fixO(%rsp),%xmm0 movaps nb214_fiyO(%rsp),%xmm1 movaps nb214_fizO(%rsp),%xmm2 movhlps %xmm0,%xmm3 movhlps %xmm1,%xmm4 movhlps %xmm2,%xmm5 addps %xmm3,%xmm0 addps %xmm4,%xmm1 addps %xmm5,%xmm2 ## sum is in 1/2 in xmm0-xmm2 movaps %xmm0,%xmm3 movaps %xmm1,%xmm4 movaps %xmm2,%xmm5 shufps $1,%xmm3,%xmm3 shufps $1,%xmm4,%xmm4 shufps $1,%xmm5,%xmm5 addss %xmm3,%xmm0 addss %xmm4,%xmm1 addss %xmm5,%xmm2 ## xmm0-xmm2 has single force in pos0 ## increment i force movss (%rdi,%rcx,4),%xmm3 movss 4(%rdi,%rcx,4),%xmm4 movss 8(%rdi,%rcx,4),%xmm5 subss %xmm0,%xmm3 subss %xmm1,%xmm4 subss %xmm2,%xmm5 movss %xmm3,(%rdi,%rcx,4) movss %xmm4,4(%rdi,%rcx,4) movss %xmm5,8(%rdi,%rcx,4) ## accumulate force in xmm6/xmm7 for fshift movaps %xmm0,%xmm6 movss %xmm2,%xmm7 movlhps %xmm1,%xmm6 shufps $8,%xmm6,%xmm6 ## 00001000 ## accumulate H1i forces in xmm0, xmm1, xmm2 movaps nb214_fixH1(%rsp),%xmm0 movaps nb214_fiyH1(%rsp),%xmm1 movaps nb214_fizH1(%rsp),%xmm2 movhlps %xmm0,%xmm3 movhlps %xmm1,%xmm4 movhlps %xmm2,%xmm5 addps %xmm3,%xmm0 addps %xmm4,%xmm1 addps %xmm5,%xmm2 ## sum is in 1/2 in xmm0-xmm2 movaps %xmm0,%xmm3 movaps %xmm1,%xmm4 movaps %xmm2,%xmm5 shufps $1,%xmm3,%xmm3 shufps $1,%xmm4,%xmm4 shufps $1,%xmm5,%xmm5 addss %xmm3,%xmm0 addss %xmm4,%xmm1 addss %xmm5,%xmm2 ## xmm0-xmm2 has single force in pos0 ## increment i force ⌨️ 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?