nb_kernel312_x86_64_sse.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,232 行 · 第 1/5 页
S
2,232 行
mulps %xmm6,%xmm7 mulps nb312_half(%rsp),%xmm3 ## rinv H1 - j water mulps nb312_half(%rsp),%xmm7 ## rinv H2 - j water ## start with H1, save H2 data movaps %xmm7,nb312_rinvH2O(%rsp) movaps %xmm3,%xmm1 mulps %xmm0,%xmm1 ## xmm1=r movaps %xmm3,%xmm0 ## xmm0=rinv mulps nb312_tsc(%rsp),%xmm1 movhlps %xmm1,%xmm2 cvttps2pi %xmm1,%mm6 cvttps2pi %xmm2,%mm7 ## mm6/mm7 contain lu indices cvtpi2ps %mm6,%xmm3 cvtpi2ps %mm7,%xmm2 movlhps %xmm2,%xmm3 subps %xmm3,%xmm1 ## xmm1=eps movaps %xmm1,%xmm2 mulps %xmm2,%xmm2 ## xmm2=eps2 pslld $2,%mm6 pslld $2,%mm7 movd %mm6,%ebx movd %mm7,%ecx psrlq $32,%mm7 movd %mm7,%edx ## table indices in ebx,ecx,edx movlps (%rsi,%rbx,4),%xmm5 movlps (%rsi,%rcx,4),%xmm7 movhps (%rsi,%rdx,4),%xmm7 ## got half coulomb table movaps %xmm5,%xmm4 shufps $136,%xmm7,%xmm4 ## 10001000 shufps $221,%xmm7,%xmm5 ## 11011101 movlps 8(%rsi,%rbx,4),%xmm7 movlps 8(%rsi,%rcx,4),%xmm3 movhps 8(%rsi,%rdx,4),%xmm3 ## other half of coulomb table movaps %xmm7,%xmm6 shufps $136,%xmm3,%xmm6 ## 10001000 shufps $221,%xmm3,%xmm7 ## 11011101 ## coulomb table ready, in xmm4-xmm7 mulps %xmm1,%xmm6 ## xmm6=Geps mulps %xmm2,%xmm7 ## xmm7=Heps2 addps %xmm6,%xmm5 addps %xmm7,%xmm5 ## xmm5=Fp mulps nb312_two(%rsp),%xmm7 ## two*Heps2 xorps %xmm3,%xmm3 ## fetch charges to xmm3 (temporary) movss nb312_qqOH(%rsp),%xmm3 movhps nb312_qqHH(%rsp),%xmm3 addps %xmm6,%xmm7 addps %xmm5,%xmm7 ## xmm7=FF mulps %xmm1,%xmm5 ## xmm5=eps*Fp addps %xmm4,%xmm5 ## xmm5=VV mulps %xmm3,%xmm5 ## vcoul=qq*VV mulps %xmm7,%xmm3 ## fijC=FF*qq ## at this point xmm5 contains vcoul and xmm3 fijC addps nb312_vctot(%rsp),%xmm5 movaps %xmm5,nb312_vctot(%rsp) xorps %xmm1,%xmm1 mulps nb312_tsc(%rsp),%xmm3 mulps %xmm0,%xmm3 subps %xmm3,%xmm1 movaps %xmm1,%xmm0 movaps %xmm1,%xmm2 mulps nb312_dxH1O(%rsp),%xmm0 mulps nb312_dyH1O(%rsp),%xmm1 mulps nb312_dzH1O(%rsp),%xmm2 ## update forces H1 - j water movaps nb312_fjxO(%rsp),%xmm3 movaps nb312_fjyO(%rsp),%xmm4 movaps nb312_fjzO(%rsp),%xmm5 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm5 movaps %xmm3,nb312_fjxO(%rsp) movaps %xmm4,nb312_fjyO(%rsp) movaps %xmm5,nb312_fjzO(%rsp) addps nb312_fixH1(%rsp),%xmm0 addps nb312_fiyH1(%rsp),%xmm1 addps nb312_fizH1(%rsp),%xmm2 movaps %xmm0,nb312_fixH1(%rsp) movaps %xmm1,nb312_fiyH1(%rsp) movaps %xmm2,nb312_fizH1(%rsp) ## do table for H2 - j water interaction movaps nb312_rinvH2O(%rsp),%xmm0 movaps nb312_rsqH2O(%rsp),%xmm1 mulps %xmm0,%xmm1 ## xmm0=rinv, xmm1=r mulps nb312_tsc(%rsp),%xmm1 movhlps %xmm1,%xmm2 cvttps2pi %xmm1,%mm6 cvttps2pi %xmm2,%mm7 ## mm6/mm7 contain lu indices cvtpi2ps %mm6,%xmm3 cvtpi2ps %mm7,%xmm2 movlhps %xmm2,%xmm3 subps %xmm3,%xmm1 ## xmm1=eps movaps %xmm1,%xmm2 mulps %xmm2,%xmm2 ## xmm2=eps2 pslld $2,%mm6 pslld $2,%mm7 movd %mm6,%ebx movd %mm7,%ecx psrlq $32,%mm7 movd %mm7,%edx ## table indices in ebx,ecx,edx movlps (%rsi,%rbx,4),%xmm5 movlps (%rsi,%rcx,4),%xmm7 movhps (%rsi,%rdx,4),%xmm7 ## got half coulomb table movaps %xmm5,%xmm4 shufps $136,%xmm7,%xmm4 ## 10001000 shufps $221,%xmm7,%xmm5 ## 11011101 movlps 8(%rsi,%rbx,4),%xmm7 movlps 8(%rsi,%rcx,4),%xmm3 movhps 8(%rsi,%rdx,4),%xmm3 ## other half of coulomb table movaps %xmm7,%xmm6 shufps $136,%xmm3,%xmm6 ## 10001000 shufps $221,%xmm3,%xmm7 ## 11011101 ## coulomb table ready, in xmm4-xmm7 mulps %xmm1,%xmm6 ## xmm6=Geps mulps %xmm2,%xmm7 ## xmm7=Heps2 addps %xmm6,%xmm5 addps %xmm7,%xmm5 ## xmm5=Fp mulps nb312_two(%rsp),%xmm7 ## two*Heps2 xorps %xmm3,%xmm3 ## fetch charges to xmm3 (temporary) movss nb312_qqOH(%rsp),%xmm3 movhps nb312_qqHH(%rsp),%xmm3 addps %xmm6,%xmm7 addps %xmm5,%xmm7 ## xmm7=FF mulps %xmm1,%xmm5 ## xmm5=eps*Fp addps %xmm4,%xmm5 ## xmm5=VV mulps %xmm3,%xmm5 ## vcoul=qq*VV mulps %xmm7,%xmm3 ## fijC=FF*qq ## at this point xmm5 contains vcoul and xmm3 fijC addps nb312_vctot(%rsp),%xmm5 movaps %xmm5,nb312_vctot(%rsp) xorps %xmm1,%xmm1 mulps nb312_tsc(%rsp),%xmm3 mulps %xmm0,%xmm3 subps %xmm3,%xmm1 movaps %xmm1,%xmm0 movaps %xmm1,%xmm2 mulps nb312_dxH2O(%rsp),%xmm0 mulps nb312_dyH2O(%rsp),%xmm1 mulps nb312_dzH2O(%rsp),%xmm2 movaps nb312_fjxO(%rsp),%xmm3 movaps nb312_fjyO(%rsp),%xmm4 movaps nb312_fjzO(%rsp),%xmm5 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm5 movq nb312_faction(%rbp),%rsi movaps %xmm3,nb312_fjxO(%rsp) movaps %xmm4,nb312_fjyO(%rsp) movaps %xmm5,nb312_fjzO(%rsp) addps nb312_fixH2(%rsp),%xmm0 addps nb312_fiyH2(%rsp),%xmm1 addps nb312_fizH2(%rsp),%xmm2 movaps %xmm0,nb312_fixH2(%rsp) movaps %xmm1,nb312_fiyH2(%rsp) movaps %xmm2,nb312_fizH2(%rsp) ## update j water forces from local variables movlps (%rsi,%rax,4),%xmm0 movlps 12(%rsi,%rax,4),%xmm1 movhps 24(%rsi,%rax,4),%xmm1 movaps nb312_fjxO(%rsp),%xmm3 movaps nb312_fjyO(%rsp),%xmm4 movaps nb312_fjzO(%rsp),%xmm5 movaps %xmm5,%xmm6 movaps %xmm5,%xmm7 shufps $2,%xmm6,%xmm6 ## 00000010 shufps $3,%xmm7,%xmm7 ## 00000011 addss 8(%rsi,%rax,4),%xmm5 addss 20(%rsi,%rax,4),%xmm6 addss 32(%rsi,%rax,4),%xmm7 movss %xmm5,8(%rsi,%rax,4) movss %xmm6,20(%rsi,%rax,4) movss %xmm7,32(%rsi,%rax,4) movaps %xmm3,%xmm5 unpcklps %xmm4,%xmm3 unpckhps %xmm4,%xmm5 addps %xmm3,%xmm0 addps %xmm5,%xmm1 movlps %xmm0,(%rsi,%rax,4) movlps %xmm1,12(%rsi,%rax,4) movhps %xmm1,24(%rsi,%rax,4) decl nb312_innerk(%rsp) jz _nb_kernel312_x86_64_sse.nb312_updateouterdata jmp _nb_kernel312_x86_64_sse.nb312_single_loop_nb_kernel312_x86_64_sse.nb312_updateouterdata: movl nb312_ii3(%rsp),%ecx movq nb312_faction(%rbp),%rdi movq nb312_fshift(%rbp),%rsi movl nb312_is3(%rsp),%edx ## accumulate Oi forces in xmm0, xmm1, xmm2 movaps nb312_fixO(%rsp),%xmm0 movaps nb312_fiyO(%rsp),%xmm1 movaps nb312_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 nb312_fixH1(%rsp),%xmm0 movaps nb312_fiyH1(%rsp),%xmm1 movaps nb312_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 movss 12(%rdi,%rcx,4),%xmm3 movss 16(%rdi,%rcx,4),%xmm4 movss 20(%rdi,%rcx,4),%xmm5 subss %xmm0,%xmm3 subss %xmm1,%xmm4 subss %xmm2,%xmm5 movss %xmm3,12(%rdi,%rcx,4) movss %xmm4,16(%rdi,%rcx,4) movss %xmm5,20(%rdi,%rcx,4) ## accumulate force in xmm6/xmm7 for fshift addss %xmm2,%xmm7 movlhps %xmm1,%xmm0 shufps $8,%xmm0,%xmm0 ## 00001000 addps %xmm0,%xmm6 ## accumulate H2i forces in xmm0, xmm1, xmm2 movaps nb312_fixH2(%rsp),%xmm0 movaps nb312_fiyH2(%rsp),%xmm1 movaps nb312_fizH2(%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 24(%rdi,%rcx,4),%xmm3 movss 28(%rdi,%rcx,4),%xmm4 movss 32(%rdi,%rcx,4),%xmm5 subss %xmm0,%xmm3 subss %xmm1,%xmm4 subss %xmm2,%xmm5 movss %xmm3,24(%rdi,%rcx,4) movss %xmm4,28(%rdi,%rcx,4) movss %xmm5,32(%rdi,%rcx,4) ## accumulate force in xmm6/xmm7 for fshift addss %xmm2,%xmm7 movlhps %xmm1,%xmm0 shufps $8,%xmm0,%xmm0 ## 00001000 addps %xmm0,%xmm6 ## increment fshift force movlps (%rsi,%rdx,4),%xmm3 movss 8(%rsi,%rdx,4),%xmm4 subps %xmm6,%xmm3 subss %xmm7,%xmm4 movlps %xmm3,(%rsi,%rdx,4) movss %xmm4,8(%rsi,%rdx,4) ## get n from stack movl nb312_n(%rsp),%esi ## get group index for i particle movq nb312_gid(%rbp),%rdx ## base of gid[] movl (%rdx,%rsi,4),%edx ## ggid=gid[n] ## accumulate total potential energy and update it movaps nb312_vctot(%rsp),%xmm7 ## accumulate movhlps %xmm7,%xmm6 addps %xmm6,%xmm7 ## pos 0-1 in xmm7 have the sum now movaps %xmm7,%xmm6 shufps $1,%xmm6,%xmm6 addss %xmm6,%xmm7 ## add earlier value from mem movq nb312_Vc(%rbp),%rax addss (%rax,%rdx,4),%xmm7 ## move back to mem movss %xmm7,(%rax,%rdx,4) ## accumulate total lj energy and update it movaps nb312_Vvdwtot(%rsp),%xmm7 ## accumulate movhlps %xmm7,%xmm6 addps %xmm6,%xmm7 ## pos 0-1 in xmm7 have the sum now movaps %xmm7,%xmm6 shufps $1,%xmm6,%xmm6 addss %xmm6,%xmm7 ## add earlier value from mem movq nb312_Vvdw(%rbp),%rax addss (%rax,%rdx,4),%xmm7 ## move back to mem movss %xmm7,(%rax,%rdx,4) ## finish if last movl nb312_nn1(%rsp),%ecx ## esi already loaded with n incl %esi subl %esi,%ecx jz _nb_kernel312_x86_64_sse.nb312_outerend ## not last, iterate outer loop once more! movl %esi,nb312_n(%rsp) jmp _nb_kernel312_x86_64_sse.nb312_outer_nb_kernel312_x86_64_sse.nb312_outerend: ## check if more outer neighborlists remain movl nb312_nri(%rsp),%ecx ## esi already loaded with n above subl %esi,%ecx jz _nb_kernel312_x86_64_sse.nb312_end ## non-zero, do one more workunit jmp _nb_kernel312_x86_64_sse.nb312_threadloop_nb_kernel312_x86_64_sse.nb312_end: movl nb312_nouter(%rsp),%eax movl nb312_ninner(%rsp),%ebx movq nb312_outeriter(%rbp),%rcx movq nb312_inneriter(%rbp),%rdx movl %eax,(%rcx) movl %ebx,(%rdx) addq $1624,%rsp emms pop %r15 pop %r14 pop %r13 pop %r12 pop %rbx pop %rbp ret.globl nb_kernel312nf_x86_64_sse.globl _nb_kernel312nf_x86_64_ssenb_kernel312nf_x86_64_sse: _nb_kernel312nf_x86_64_sse: ## Room for return address and rbp (16 bytes).set nb312nf_fshift, 16.set nb312nf_gid, 24.set nb312nf_pos, 32.set nb312nf_faction, 40.set nb312nf_charge, 48.set nb312nf_p_facel, 56.set nb312nf_argkrf, 64.set nb312nf_argcrf, 72.set nb312nf_Vc, 80.set nb312nf_type, 88.set nb312nf_p_ntype, 96.set nb312nf_vdwparam, 104.set nb312nf_Vvdw, 112.set nb312nf_p_tabscale, 120.set nb312nf_VFtab, 128.set nb312nf_invsqrta, 136.set nb312nf_dvda, 144.set nb312nf_p_gbtabscale, 152.set nb312nf_GBtab, 160.set nb312nf_p_nthr⌨️ 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?