nb_kernel204_x86_64_sse.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,119 行 · 第 1/5 页
S
2,119 行
mulps %xmm6,%xmm4 mulps %xmm3,%xmm0 ## fscal mulps %xmm4,%xmm7 ## fscal movaps %xmm0,%xmm1 movaps %xmm0,%xmm2 mulps nb204_dxH1M(%rsp),%xmm0 mulps nb204_dyH1M(%rsp),%xmm1 mulps nb204_dzH1M(%rsp),%xmm2 ## update forces H1 - j water movaps nb204_fjxM(%rsp),%xmm3 movaps nb204_fjyM(%rsp),%xmm4 movaps nb204_fjzM(%rsp),%xmm5 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm5 movaps %xmm3,nb204_fjxM(%rsp) movaps %xmm4,nb204_fjyM(%rsp) movaps %xmm5,nb204_fjzM(%rsp) addps nb204_fixH1(%rsp),%xmm0 addps nb204_fiyH1(%rsp),%xmm1 addps nb204_fizH1(%rsp),%xmm2 movaps %xmm0,nb204_fixH1(%rsp) movaps %xmm1,nb204_fiyH1(%rsp) movaps %xmm2,nb204_fizH1(%rsp) ## do forces H2 - j water movaps %xmm7,%xmm0 movaps %xmm7,%xmm1 movaps %xmm7,%xmm2 mulps nb204_dxH2M(%rsp),%xmm0 mulps nb204_dyH2M(%rsp),%xmm1 mulps nb204_dzH2M(%rsp),%xmm2 movaps nb204_fjxM(%rsp),%xmm3 movaps nb204_fjyM(%rsp),%xmm4 movaps nb204_fjzM(%rsp),%xmm5 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm5 movq nb204_faction(%rbp),%rsi movaps %xmm3,nb204_fjxM(%rsp) movaps %xmm4,nb204_fjyM(%rsp) movaps %xmm5,nb204_fjzM(%rsp) addps nb204_fixH2(%rsp),%xmm0 addps nb204_fiyH2(%rsp),%xmm1 addps nb204_fizH2(%rsp),%xmm2 movaps %xmm0,nb204_fixH2(%rsp) movaps %xmm1,nb204_fiyH2(%rsp) movaps %xmm2,nb204_fizH2(%rsp) ## update j water forces from local variables movlps 36(%rsi,%rax,4),%xmm0 movlps 12(%rsi,%rax,4),%xmm1 movhps 24(%rsi,%rax,4),%xmm1 movaps nb204_fjxM(%rsp),%xmm3 movaps nb204_fjyM(%rsp),%xmm4 movaps nb204_fjzM(%rsp),%xmm5 movaps %xmm5,%xmm6 movaps %xmm5,%xmm7 shufps $2,%xmm6,%xmm6 ## 00000010 shufps $3,%xmm7,%xmm7 ## 00000011 addss 44(%rsi,%rax,4),%xmm5 addss 20(%rsi,%rax,4),%xmm6 addss 32(%rsi,%rax,4),%xmm7 movss %xmm5,44(%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 %xmm0,36(%rsi,%rax,4) movlps %xmm1,12(%rsi,%rax,4) movhps %xmm1,24(%rsi,%rax,4) decl nb204_innerk(%rsp) jz _nb_kernel204_x86_64_sse.nb204_updateouterdata jmp _nb_kernel204_x86_64_sse.nb204_single_loop_nb_kernel204_x86_64_sse.nb204_updateouterdata: movl nb204_ii3(%rsp),%ecx movq nb204_faction(%rbp),%rdi movq nb204_fshift(%rbp),%rsi movl nb204_is3(%rsp),%edx ## accumulate H1 i forces in xmm0, xmm1, xmm2 movaps nb204_fixH1(%rsp),%xmm0 movaps nb204_fiyH1(%rsp),%xmm1 movaps nb204_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 movaps %xmm0,%xmm6 movss %xmm2,%xmm7 movlhps %xmm1,%xmm6 shufps $8,%xmm6,%xmm6 ## 00001000 ## accumulate H2i forces in xmm0, xmm1, xmm2 movaps nb204_fixH2(%rsp),%xmm0 movaps nb204_fiyH2(%rsp),%xmm1 movaps nb204_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 ## accumulate M i forces in xmm0, xmm1, xmm2 movaps nb204_fixM(%rsp),%xmm0 movaps nb204_fiyM(%rsp),%xmm1 movaps nb204_fizM(%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 36(%rdi,%rcx,4),%xmm3 movss 40(%rdi,%rcx,4),%xmm4 movss 44(%rdi,%rcx,4),%xmm5 subss %xmm0,%xmm3 subss %xmm1,%xmm4 subss %xmm2,%xmm5 movss %xmm3,36(%rdi,%rcx,4) movss %xmm4,40(%rdi,%rcx,4) movss %xmm5,44(%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 nb204_n(%rsp),%esi ## get group index for i particle movq nb204_gid(%rbp),%rdx ## base of gid[] movl (%rdx,%rsi,4),%edx ## ggid=gid[n] ## accumulate total potential energy and update it movaps nb204_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 nb204_Vc(%rbp),%rax addss (%rax,%rdx,4),%xmm7 ## move back to mem movss %xmm7,(%rax,%rdx,4) ## finish if last movl nb204_nn1(%rsp),%ecx ## esi already loaded with n incl %esi subl %esi,%ecx jz _nb_kernel204_x86_64_sse.nb204_outerend ## not last, iterate outer loop once more! movl %esi,nb204_n(%rsp) jmp _nb_kernel204_x86_64_sse.nb204_outer_nb_kernel204_x86_64_sse.nb204_outerend: ## check if more outer neighborlists remain movl nb204_nri(%rsp),%ecx ## esi already loaded with n above subl %esi,%ecx jz _nb_kernel204_x86_64_sse.nb204_end ## non-zero, do one more workunit jmp _nb_kernel204_x86_64_sse.nb204_threadloop_nb_kernel204_x86_64_sse.nb204_end: movl nb204_nouter(%rsp),%eax movl nb204_ninner(%rsp),%ebx movq nb204_outeriter(%rbp),%rcx movq nb204_inneriter(%rbp),%rdx movl %eax,(%rcx) movl %ebx,(%rdx) addq $1544,%rsp emms pop %r15 pop %r14 pop %r13 pop %r12 pop %rbx pop %rbp ret.globl nb_kernel204nf_x86_64_sse.globl _nb_kernel204nf_x86_64_ssenb_kernel204nf_x86_64_sse: _nb_kernel204nf_x86_64_sse: ## Room for return address and rbp (16 bytes).set nb204nf_fshift, 16.set nb204nf_gid, 24.set nb204nf_pos, 32.set nb204nf_faction, 40.set nb204nf_charge, 48.set nb204nf_p_facel, 56.set nb204nf_argkrf, 64.set nb204nf_argcrf, 72.set nb204nf_Vc, 80.set nb204nf_type, 88.set nb204nf_p_ntype, 96.set nb204nf_vdwparam, 104.set nb204nf_Vvdw, 112.set nb204nf_p_tabscale, 120.set nb204nf_VFtab, 128.set nb204nf_invsqrta, 136.set nb204nf_dvda, 144.set nb204nf_p_gbtabscale, 152.set nb204nf_GBtab, 160.set nb204nf_p_nthreads, 168.set nb204nf_count, 176.set nb204nf_mtx, 184.set nb204nf_outeriter, 192.set nb204nf_inneriter, 200.set nb204nf_work, 208 ## stack offsets for local variables ## bottom of stack is cache-aligned for sse use .set nb204nf_ixH1, 0.set nb204nf_iyH1, 16.set nb204nf_izH1, 32.set nb204nf_ixH2, 48.set nb204nf_iyH2, 64.set nb204nf_izH2, 80.set nb204nf_ixM, 96.set nb204nf_iyM, 112.set nb204nf_izM, 128.set nb204nf_jxH1, 144.set nb204nf_jyH1, 160.set nb204nf_jzH1, 176.set nb204nf_jxH2, 192.set nb204nf_jyH2, 208.set nb204nf_jzH2, 224.set nb204nf_jxM, 240.set nb204nf_jyM, 256.set nb204nf_jzM, 272.set nb204nf_qqHH, 288.set nb204nf_qqMH, 304.set nb204nf_qqMM, 320.set nb204nf_vctot, 336.set nb204nf_half, 352.set nb204nf_three, 368.set nb204nf_rsqH1H1, 384.set nb204nf_rsqH1H2, 400.set nb204nf_rsqH1M, 416.set nb204nf_rsqH2H1, 432.set nb204nf_rsqH2H2, 448.set nb204nf_rsqH2M, 464.set nb204nf_rsqMH1, 480.set nb204nf_rsqMH2, 496.set nb204nf_rsqMM, 512.set nb204nf_rinvH1H1, 528.set nb204nf_rinvH1H2, 544.set nb204nf_rinvH1M, 560.set nb204nf_rinvH2H1, 576.set nb204nf_rinvH2H2, 592.set nb204nf_rinvH2M, 608.set nb204nf_rinvMH1, 624.set nb204nf_rinvMH2, 640.set nb204nf_rinvMM, 656.set nb204nf_krf, 672.set nb204nf_crf, 688.set nb204nf_is3, 704.set nb204nf_ii3, 708.set nb204nf_innerjjnr, 712.set nb204nf_nri, 720.set nb204nf_iinr, 728.set nb204nf_jindex, 736.set nb204nf_jjnr, 744.set nb204nf_shift, 752.set nb204nf_shiftvec, 760.set nb204nf_facel, 768.set nb204nf_innerk, 776.set nb204nf_n, 780.set nb204nf_nn1, 784.set nb204nf_nouter, 788.set nb204nf_ninner, 792 push %rbp movq %rsp,%rbp push %rbx emms push %r12 push %r13 push %r14 push %r15 subq $808,%rsp ## local variable stack space (n*16+8) ## zero 32-bit iteration counters movl $0,%eax movl %eax,nb204nf_nouter(%rsp) movl %eax,nb204nf_ninner(%rsp) movl (%rdi),%edi movl %edi,nb204nf_nri(%rsp) movq %rsi,nb204nf_iinr(%rsp) movq %rdx,nb204nf_jindex(%rsp) movq %rcx,nb204nf_jjnr(%rsp) movq %r8,nb204nf_shift(%rsp) movq %r9,nb204nf_shiftvec(%rsp) movq nb204nf_p_facel(%rbp),%rsi movss (%rsi),%xmm0 movss %xmm0,nb204nf_facel(%rsp) movq nb204nf_argkrf(%rbp),%rsi movq nb204nf_argcrf(%rbp),%rdi movss (%rsi),%xmm1 movss (%rdi),%xmm2 shufps $0,%xmm1,%xmm1 shufps $0,%xmm2,%xmm2 movaps %xmm1,nb204nf_krf(%rsp) movaps %xmm2,nb204nf_crf(%rsp) ## create constant floating-point factors on stack movl $0x3f000000,%eax ## half in IEEE (hex) movl %eax,nb204nf_half(%rsp) movss nb204nf_half(%rsp),%xmm1 shufps $0,%xmm1,%xmm1 ## splat to all elements movaps %xmm1,%xmm2 addps %xmm2,%xmm2 ## one movaps %xmm2,%xmm3 addps %xmm2,%xmm2 ## two addps %xmm2,%xmm3 ## three movaps %xmm1,nb204nf_half(%rsp) movaps %xmm3,nb204nf_three(%rsp)⌨️ 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?