nb_kernel304_x86_64_sse.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,242 行 · 第 1/5 页
S
2,242 行
movaps %xmm1,%xmm0 movaps %xmm1,%xmm2 mulps nb304_dxH1M(%rsp),%xmm0 mulps nb304_dyH1M(%rsp),%xmm1 mulps nb304_dzH1M(%rsp),%xmm2 ## update forces H1 - j water movaps nb304_fjxM(%rsp),%xmm3 movaps nb304_fjyM(%rsp),%xmm4 movaps nb304_fjzM(%rsp),%xmm5 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm5 movaps %xmm3,nb304_fjxM(%rsp) movaps %xmm4,nb304_fjyM(%rsp) movaps %xmm5,nb304_fjzM(%rsp) addps nb304_fixH1(%rsp),%xmm0 addps nb304_fiyH1(%rsp),%xmm1 addps nb304_fizH1(%rsp),%xmm2 movaps %xmm0,nb304_fixH1(%rsp) movaps %xmm1,nb304_fiyH1(%rsp) movaps %xmm2,nb304_fizH1(%rsp) ## do table for H2 - j water interaction movaps nb304_rinvH2M(%rsp),%xmm0 movaps nb304_rsqH2M(%rsp),%xmm1 mulps %xmm0,%xmm1 ## xmm0=rinv, xmm1=r mulps nb304_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 nb304_two(%rsp),%xmm7 ## two*Heps2 xorps %xmm3,%xmm3 ## fetch charges to xmm3 (temporary) movss nb304_qqMH(%rsp),%xmm3 movhps nb304_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 nb304_vctot(%rsp),%xmm5 movaps %xmm5,nb304_vctot(%rsp) xorps %xmm1,%xmm1 mulps nb304_tsc(%rsp),%xmm3 mulps %xmm0,%xmm3 subps %xmm3,%xmm1 movaps %xmm1,%xmm0 movaps %xmm1,%xmm2 mulps nb304_dxH2M(%rsp),%xmm0 mulps nb304_dyH2M(%rsp),%xmm1 mulps nb304_dzH2M(%rsp),%xmm2 movaps nb304_fjxM(%rsp),%xmm3 movaps nb304_fjyM(%rsp),%xmm4 movaps nb304_fjzM(%rsp),%xmm5 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm5 movq nb304_faction(%rbp),%rsi movaps %xmm3,nb304_fjxM(%rsp) movaps %xmm4,nb304_fjyM(%rsp) movaps %xmm5,nb304_fjzM(%rsp) addps nb304_fixH2(%rsp),%xmm0 addps nb304_fiyH2(%rsp),%xmm1 addps nb304_fizH2(%rsp),%xmm2 movaps %xmm0,nb304_fixH2(%rsp) movaps %xmm1,nb304_fiyH2(%rsp) movaps %xmm2,nb304_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 nb304_fjxM(%rsp),%xmm3 movaps nb304_fjyM(%rsp),%xmm4 movaps nb304_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 nb304_innerk(%rsp) jz _nb_kernel304_x86_64_sse.nb304_updateouterdata jmp _nb_kernel304_x86_64_sse.nb304_single_loop_nb_kernel304_x86_64_sse.nb304_updateouterdata: movl nb304_ii3(%rsp),%ecx movq nb304_faction(%rbp),%rdi movq nb304_fshift(%rbp),%rsi movl nb304_is3(%rsp),%edx ## accumulate H1 i forces in xmm0, xmm1, xmm2 movaps nb304_fixH1(%rsp),%xmm0 movaps nb304_fiyH1(%rsp),%xmm1 movaps nb304_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 H2 i forces in xmm0, xmm1, xmm2 movaps nb304_fixH2(%rsp),%xmm0 movaps nb304_fiyH2(%rsp),%xmm1 movaps nb304_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 nb304_fixM(%rsp),%xmm0 movaps nb304_fiyM(%rsp),%xmm1 movaps nb304_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 nb304_n(%rsp),%esi ## get group index for i particle movq nb304_gid(%rbp),%rdx ## base of gid[] movl (%rdx,%rsi,4),%edx ## ggid=gid[n] ## accumulate total potential energy and update it movaps nb304_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 nb304_Vc(%rbp),%rax addss (%rax,%rdx,4),%xmm7 ## move back to mem movss %xmm7,(%rax,%rdx,4) ## finish if last movl nb304_nn1(%rsp),%ecx ## esi already loaded with n incl %esi subl %esi,%ecx jz _nb_kernel304_x86_64_sse.nb304_outerend ## not last, iterate outer loop once more! movl %esi,nb304_n(%rsp) jmp _nb_kernel304_x86_64_sse.nb304_outer_nb_kernel304_x86_64_sse.nb304_outerend: ## check if more outer neighborlists remain movl nb304_nri(%rsp),%ecx ## esi already loaded with n above subl %esi,%ecx jz _nb_kernel304_x86_64_sse.nb304_end ## non-zero, do one more workunit jmp _nb_kernel304_x86_64_sse.nb304_threadloop_nb_kernel304_x86_64_sse.nb304_end: movl nb304_nouter(%rsp),%eax movl nb304_ninner(%rsp),%ebx movq nb304_outeriter(%rbp),%rcx movq nb304_inneriter(%rbp),%rdx movl %eax,(%rcx) movl %ebx,(%rdx) addq $1528,%rsp emms pop %r15 pop %r14 pop %r13 pop %r12 pop %rbx pop %rbp ret.globl nb_kernel304nf_x86_64_sse.globl _nb_kernel304nf_x86_64_ssenb_kernel304nf_x86_64_sse: _nb_kernel304nf_x86_64_sse: ## Room for return address and rbp (16 bytes).set nb304nf_fshift, 16.set nb304nf_gid, 24.set nb304nf_pos, 32.set nb304nf_faction, 40.set nb304nf_charge, 48.set nb304nf_p_facel, 56.set nb304nf_argkrf, 64.set nb304nf_argcrf, 72.set nb304nf_Vc, 80.set nb304nf_type, 88.set nb304nf_p_ntype, 96.set nb304nf_vdwparam, 104.set nb304nf_Vvdw, 112.set nb304nf_p_tabscale, 120.set nb304nf_VFtab, 128.set nb304nf_invsqrta, 136.set nb304nf_dvda, 144.set nb304nf_p_gbtabscale, 152.set nb304nf_GBtab, 160.set nb304nf_p_nthreads, 168.set nb304nf_count, 176.set nb304nf_mtx, 184.set nb304nf_outeriter, 192.set nb304nf_inneriter, 200.set nb304nf_work, 208 ## stack offsets for local variables ## bottom of stack is cache-aligned for sse use .set nb304nf_ixH1, 0.set nb304nf_iyH1, 16.set nb304nf_izH1, 32.set nb304nf_ixH2, 48.set nb304nf_iyH2, 64.set nb304nf_izH2, 80.set nb304nf_ixM, 96.set nb304nf_iyM, 112.set nb304nf_izM, 128.set nb304nf_jxH1, 144.set nb304nf_jyH1, 160.set nb304nf_jzH1, 176.set nb304nf_jxH2, 192.set nb304nf_jyH2, 208.set nb304nf_jzH2, 224.set nb304nf_jxM, 240.set nb304nf_jyM, 256.set nb304nf_jzM, 272.set nb304nf_qqHH, 288.set nb304nf_qqMH, 304.set nb304nf_qqMM, 320.set nb304nf_tsc, 336.set nb304nf_vctot, 352.set nb304nf_half, 368.set nb304nf_three, 384.set nb304nf_rsqH1H1, 400.set nb304nf_rsqH1H2, 416.set nb304nf_rsqH1M, 432.set nb304nf_rsqH2H1, 448.set nb304nf_rsqH2H2, 464.set nb304nf_rsqH2M, 480.set nb304nf_rsqMH1, 496.set nb304nf_rsqMH2, 512.set nb304nf_rsqMM, 528.set nb304nf_rinvH1H1, 544.set nb304nf_rinvH1H2, 560.set nb304nf_rinvH1M, 576.set nb304nf_rinvH2H1, 592.set nb304nf_rinvH2H2, 608.set nb304nf_rinvH2M, 624.set nb304nf_rinvMH1, 640.set nb304nf_rinvMH2, 656.set nb304nf_rinvMM, 672.set nb304nf_is3, 688.set nb304nf_ii3, 692.set nb304nf_nri, 696.set nb304nf_iinr, 704.set nb304nf_jindex, 712.set nb304nf_jjnr, 720.set nb304nf_shift, 728.set nb304nf_shiftvec, 736.set nb304nf_facel, 744.set nb304nf_innerjjnr, 752.set nb304nf_innerk, 760.set nb304nf_n, 764.set nb304nf_nn1, 768.set nb304nf_nouter, 772.set nb304nf_ninner, 776 push %rbp movq %rsp,%rbp push %rbx emms push %r12 push %r13 push %r14 push %r15 subq $792,%rsp ## local variable stack space (n*16+8) ## zero 32-bit iteration counters movl $0,%eax movl %eax,nb304nf_nouter(%rsp) movl %eax,nb304nf_ninner(%rsp) movl (%rdi),%edi movl %edi⌨️ 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?