nb_kernel132_x86_64_sse.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,146 行 · 第 1/5 页
S
2,146 行
addss %xmm5,%xmm7 ## xmm7=FF mulss %xmm1,%xmm5 ## xmm5=eps*Fp addss %xmm4,%xmm5 ## xmm5=VV movss nb132_c12(%rsp),%xmm4 mulss %xmm4,%xmm7 ## fijR mulss %xmm4,%xmm5 ## Vvdw12 movaps nb132_fstmp(%rsp),%xmm3 mulss nb132_tsc(%rsp),%xmm7 subss %xmm7,%xmm3 addss nb132_Vvdwtot(%rsp),%xmm5 movss %xmm5,nb132_Vvdwtot(%rsp) mulps %xmm3,%xmm0 movaps %xmm0,%xmm1 movaps %xmm0,%xmm2 mulps nb132_dxOO(%rsp),%xmm0 mulps nb132_dyOO(%rsp),%xmm1 mulps nb132_dzOO(%rsp),%xmm2 ## initial update for j forces movaps %xmm0,%xmm3 movaps %xmm1,%xmm4 movaps %xmm2,%xmm5 addps nb132_fixO(%rsp),%xmm0 addps nb132_fiyO(%rsp),%xmm1 addps nb132_fizO(%rsp),%xmm2 movaps %xmm0,nb132_fixO(%rsp) movaps %xmm1,nb132_fiyO(%rsp) movaps %xmm2,nb132_fizO(%rsp) movaps %xmm3,nb132_fjxO(%rsp) movaps %xmm4,nb132_fjyO(%rsp) movaps %xmm5,nb132_fjzO(%rsp) ## done with i O Now do i H1 & H2 simultaneously first get i particle coords: movaps nb132_jxO(%rsp),%xmm0 movaps nb132_jyO(%rsp),%xmm1 movaps nb132_jzO(%rsp),%xmm2 movaps %xmm0,%xmm3 movaps %xmm1,%xmm4 movaps %xmm2,%xmm5 subps nb132_ixH1(%rsp),%xmm0 subps nb132_iyH1(%rsp),%xmm1 subps nb132_izH1(%rsp),%xmm2 subps nb132_ixH2(%rsp),%xmm3 subps nb132_iyH2(%rsp),%xmm4 subps nb132_izH2(%rsp),%xmm5 movaps %xmm0,nb132_dxH1O(%rsp) movaps %xmm1,nb132_dyH1O(%rsp) movaps %xmm2,nb132_dzH1O(%rsp) movaps %xmm3,nb132_dxH2O(%rsp) movaps %xmm4,nb132_dyH2O(%rsp) movaps %xmm5,nb132_dzH2O(%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 ## do invsqrt rsqrtps %xmm0,%xmm1 rsqrtps %xmm4,%xmm5 movaps %xmm1,%xmm2 movaps %xmm5,%xmm6 mulps %xmm1,%xmm1 mulps %xmm5,%xmm5 movaps nb132_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 nb132_half(%rsp),%xmm3 ## rinv H1 - j water mulps nb132_half(%rsp),%xmm7 ## rinv H2 - j water ## assemble charges in xmm6 xorps %xmm6,%xmm6 movss nb132_qqOH(%rsp),%xmm6 movhps nb132_qqHH(%rsp),%xmm6 movaps %xmm3,%xmm0 movaps %xmm7,%xmm1 mulps %xmm6,%xmm3 ## vcoul mulps %xmm6,%xmm7 ## vcoul mulps %xmm0,%xmm0 mulps %xmm1,%xmm1 mulps %xmm3,%xmm0 addps %xmm7,%xmm3 mulps %xmm1,%xmm7 addps nb132_vctot(%rsp),%xmm3 movaps %xmm3,nb132_vctot(%rsp) movaps %xmm0,%xmm1 movaps %xmm0,%xmm2 mulps nb132_dxH1O(%rsp),%xmm0 mulps nb132_dyH1O(%rsp),%xmm1 mulps nb132_dzH1O(%rsp),%xmm2 ## update forces H1 - j water movaps %xmm0,%xmm3 movaps %xmm1,%xmm4 movaps %xmm2,%xmm5 addps nb132_fixH1(%rsp),%xmm0 addps nb132_fiyH1(%rsp),%xmm1 addps nb132_fizH1(%rsp),%xmm2 addps nb132_fjxO(%rsp),%xmm3 addps nb132_fjyO(%rsp),%xmm4 addps nb132_fjzO(%rsp),%xmm5 movaps %xmm0,nb132_fixH1(%rsp) movaps %xmm1,nb132_fiyH1(%rsp) movaps %xmm2,nb132_fizH1(%rsp) movaps %xmm3,nb132_fjxO(%rsp) movaps %xmm4,nb132_fjyO(%rsp) movaps %xmm5,nb132_fjzO(%rsp) ## do forces H2 - j water movaps %xmm7,%xmm0 movaps %xmm7,%xmm1 movaps %xmm7,%xmm2 mulps nb132_dxH2O(%rsp),%xmm0 mulps nb132_dyH2O(%rsp),%xmm1 mulps nb132_dzH2O(%rsp),%xmm2 movaps %xmm0,%xmm3 movaps %xmm1,%xmm4 movaps %xmm2,%xmm5 addps nb132_fixH2(%rsp),%xmm0 addps nb132_fiyH2(%rsp),%xmm1 addps nb132_fizH2(%rsp),%xmm2 addps nb132_fjxO(%rsp),%xmm3 addps nb132_fjyO(%rsp),%xmm4 addps nb132_fjzO(%rsp),%xmm5 movaps %xmm0,nb132_fixH2(%rsp) movaps %xmm1,nb132_fiyH2(%rsp) movaps %xmm2,nb132_fizH2(%rsp) ## fjx/fjy/fjz in xmm3-xmm5 movq nb132_faction(%rbp),%rsi ## 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 %xmm5,%xmm6 movaps %xmm5,%xmm7 shufps $2,%xmm6,%xmm6 ## constant 00000010 shufps $3,%xmm7,%xmm7 ## constant 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 nb132_innerk(%rsp) jz _nb_kernel132_x86_64_sse.nb132_updateouterdata jmp _nb_kernel132_x86_64_sse.nb132_single_loop_nb_kernel132_x86_64_sse.nb132_updateouterdata: movl nb132_ii3(%rsp),%ecx movq nb132_faction(%rbp),%rdi movq nb132_fshift(%rbp),%rsi movl nb132_is3(%rsp),%edx ## accumulate Oi forces in xmm0, xmm1, xmm2 movaps nb132_fixO(%rsp),%xmm0 movaps nb132_fiyO(%rsp),%xmm1 movaps nb132_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 ## constant 00001000 ## accumulate H1i forces in xmm0, xmm1, xmm2 movaps nb132_fixH1(%rsp),%xmm0 movaps nb132_fiyH1(%rsp),%xmm1 movaps nb132_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 ## constant 00001000 addps %xmm0,%xmm6 ## accumulate H2i forces in xmm0, xmm1, xmm2 movaps nb132_fixH2(%rsp),%xmm0 movaps nb132_fiyH2(%rsp),%xmm1 movaps nb132_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 ## constant 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 nb132_n(%rsp),%esi ## get group index for i particle movq nb132_gid(%rbp),%rdx ## base of gid[] movl (%rdx,%rsi,4),%edx ## ggid=gid[n] ## accumulate total potential energy and update it movaps nb132_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 nb132_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 nb132_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 nb132_Vvdw(%rbp),%rax addss (%rax,%rdx,4),%xmm7 ## move back to mem movss %xmm7,(%rax,%rdx,4) ## finish if last movl nb132_nn1(%rsp),%ecx ## esi already loaded with n incl %esi subl %esi,%ecx jz _nb_kernel132_x86_64_sse.nb132_outerend ## not last, iterate outer loop once more! movl %esi,nb132_n(%rsp) jmp _nb_kernel132_x86_64_sse.nb132_outer_nb_kernel132_x86_64_sse.nb132_outerend: ## check if more outer neighborlists remain movl nb132_nri(%rsp),%ecx ## esi already loaded with n above subl %esi,%ecx jz _nb_kernel132_x86_64_sse.nb132_end ## non-zero, do one more workunit jmp _nb_kernel132_x86_64_sse.nb132_threadloop_nb_kernel132_x86_64_sse.nb132_end: movl nb132_nouter(%rsp),%eax movl nb132_ninner(%rsp),%ebx movq nb132_outeriter(%rbp),%rcx movq nb132_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_kernel132nf_x86_64_sse.globl _nb_kernel132nf_x86_64_ssenb_kernel132nf_x86_64_sse: _nb_kernel132nf_x86_64_sse: ## Room for return address and rbp (16 bytes).set nb132nf_fshift, 16.set nb132nf_gid, 24.set nb132nf_pos, 32.set nb132nf_faction, 40.set nb132nf_charge, 48.set nb132nf_p_facel, 56.set nb132nf_argkrf, 64.set nb132nf_argcrf, 72.set nb132nf_Vc, 80.set nb132nf_type, 88.set nb132nf_p_ntype, 96.set nb132nf_vdwparam, 104.set nb132nf_Vvdw, 112.set nb132nf_p_tabscale, 120.set nb132nf_VFtab, 128.set nb132nf_invsqrta, 136.set nb132nf_dvda, 144.set nb132nf_p_gbtabscale, 152.set nb132nf_GBtab, 160.set nb132nf_p_nthreads, 168.set nb132nf_count, 176.set nb132nf_mtx, 184.set nb132nf_outeriter, 192.set nb132nf_inneriter, 200.set nb132nf_work, 208 ## stack offsets for local variables ## bottom of stack is cache-aligned for sse use .set nb132nf_ixO, 0.set nb132nf_iyO, 16.set nb132nf_izO, 32.set nb132nf_ixH1, 48.set nb132nf_iyH1, 64.set nb132nf_izH1, 80.set nb132nf_ixH2, 96.set nb132nf_iyH2, 112.set nb132nf_izH2, 128.set nb132nf_jxO, 144.set nb132nf_jyO, 160⌨️ 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?