nb_kernel134_x86_64_sse.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,165 行 · 第 1/5 页
S
2,165 行
shufps $136,%xmm3,%xmm2 ## 10001000 => jzH2a jzH2b jzH2c jzH2d ## xmm0 = H2x ## xmm1 = H2y ## xmm2 = H2z movaps %xmm0,%xmm3 movaps %xmm1,%xmm4 movaps %xmm2,%xmm5 movaps %xmm0,%xmm6 movaps %xmm1,%xmm7 movaps %xmm2,%xmm8 subps nb134_ixH1(%rsp),%xmm0 subps nb134_iyH1(%rsp),%xmm1 subps nb134_izH1(%rsp),%xmm2 subps nb134_ixH2(%rsp),%xmm3 subps nb134_iyH2(%rsp),%xmm4 subps nb134_izH2(%rsp),%xmm5 subps nb134_ixM(%rsp),%xmm6 subps nb134_iyM(%rsp),%xmm7 subps nb134_izM(%rsp),%xmm8 movaps %xmm0,nb134_dxH1H2(%rsp) movaps %xmm1,nb134_dyH1H2(%rsp) movaps %xmm2,nb134_dzH1H2(%rsp) mulps %xmm0,%xmm0 mulps %xmm1,%xmm1 mulps %xmm2,%xmm2 movaps %xmm3,nb134_dxH2H2(%rsp) movaps %xmm4,nb134_dyH2H2(%rsp) movaps %xmm5,nb134_dzH2H2(%rsp) mulps %xmm3,%xmm3 mulps %xmm4,%xmm4 mulps %xmm5,%xmm5 movaps %xmm6,nb134_dxMH2(%rsp) movaps %xmm7,nb134_dyMH2(%rsp) movaps %xmm8,nb134_dzMH2(%rsp) mulps %xmm6,%xmm6 mulps %xmm7,%xmm7 mulps %xmm8,%xmm8 addps %xmm1,%xmm0 addps %xmm2,%xmm0 addps %xmm4,%xmm3 addps %xmm5,%xmm3 addps %xmm7,%xmm6 addps %xmm8,%xmm6 ## start doing invsqrt for jH2 atoms rsqrtps %xmm0,%xmm1 rsqrtps %xmm3,%xmm4 rsqrtps %xmm6,%xmm7 movaps %xmm1,%xmm2 movaps %xmm4,%xmm5 movaps %xmm7,%xmm8 mulps %xmm1,%xmm1 ## lu*lu mulps %xmm4,%xmm4 ## lu*lu mulps %xmm7,%xmm7 ## lu*lu movaps nb134_three(%rsp),%xmm9 movaps %xmm9,%xmm10 movaps %xmm9,%xmm11 mulps %xmm0,%xmm1 ## rsq*lu*lu mulps %xmm3,%xmm4 ## rsq*lu*lu mulps %xmm6,%xmm7 ## rsq*lu*lu subps %xmm1,%xmm9 subps %xmm4,%xmm10 subps %xmm7,%xmm11 ## 3-rsq*lu*lu mulps %xmm2,%xmm9 mulps %xmm5,%xmm10 mulps %xmm8,%xmm11 ## lu*(3-rsq*lu*lu) movaps nb134_half(%rsp),%xmm0 mulps %xmm0,%xmm9 ## rinvH1H2 mulps %xmm0,%xmm10 ## rinvH2H2 mulps %xmm0,%xmm11 ## rinvMH2 ## H2 interactions movaps %xmm9,%xmm0 movaps %xmm10,%xmm1 movaps %xmm11,%xmm2 mulps %xmm9,%xmm9 mulps %xmm10,%xmm10 mulps %xmm11,%xmm11 mulps nb134_qqHH(%rsp),%xmm0 mulps nb134_qqHH(%rsp),%xmm1 mulps nb134_qqMH(%rsp),%xmm2 mulps %xmm0,%xmm9 mulps %xmm1,%xmm10 mulps %xmm2,%xmm11 addps nb134_vctot(%rsp),%xmm0 addps %xmm2,%xmm1 addps %xmm1,%xmm0 movaps %xmm0,nb134_vctot(%rsp) ## move j H2 forces to local temp variables movq nb134_faction(%rbp),%rdi movlps 24(%rdi,%rax,4),%xmm0 ## jxH2a jyH2a - - movlps 24(%rdi,%rcx,4),%xmm1 ## jxH2c jyH2c - - movhps 24(%rdi,%rbx,4),%xmm0 ## jxH2a jyH2a jxH2b jyH2b movhps 24(%rdi,%rdx,4),%xmm1 ## jxH2c jyH2c jxH2d jyH2d movss 32(%rdi,%rax,4),%xmm2 ## jzH2a - - - movss 32(%rdi,%rcx,4),%xmm3 ## jzH2c - - - movhps 32(%rdi,%rbx,4),%xmm2 ## jzH2a - jzH2b - movhps 32(%rdi,%rdx,4),%xmm3 ## jzH2c - jzH2d - shufps $136,%xmm3,%xmm2 ## 10001000 => jzH2a jzH2b jzH2c jzH2d ## xmm0: jxH2a jyH2a jxH2b jyH2b ## xmm1: jxH2c jyH2c jxH2d jyH2d ## xmm2: jzH2a jzH2b jzH2c jzH2d movaps %xmm9,%xmm7 movaps %xmm9,%xmm8 movaps %xmm11,%xmm13 movaps %xmm11,%xmm14 movaps %xmm11,%xmm15 movaps %xmm10,%xmm11 movaps %xmm10,%xmm12 mulps nb134_dxH1H2(%rsp),%xmm7 mulps nb134_dyH1H2(%rsp),%xmm8 mulps nb134_dzH1H2(%rsp),%xmm9 mulps nb134_dxH2H2(%rsp),%xmm10 mulps nb134_dyH2H2(%rsp),%xmm11 mulps nb134_dzH2H2(%rsp),%xmm12 mulps nb134_dxMH2(%rsp),%xmm13 mulps nb134_dyMH2(%rsp),%xmm14 mulps nb134_dzMH2(%rsp),%xmm15 movaps %xmm7,%xmm3 movaps %xmm8,%xmm4 addps %xmm9,%xmm2 addps nb134_fixH1(%rsp),%xmm7 addps nb134_fiyH1(%rsp),%xmm8 addps nb134_fizH1(%rsp),%xmm9 addps %xmm10,%xmm3 addps %xmm11,%xmm4 addps %xmm12,%xmm2 addps nb134_fixH2(%rsp),%xmm10 addps nb134_fiyH2(%rsp),%xmm11 addps nb134_fizH2(%rsp),%xmm12 addps %xmm13,%xmm3 addps %xmm14,%xmm4 addps %xmm15,%xmm2 addps nb134_fixM(%rsp),%xmm13 addps nb134_fiyM(%rsp),%xmm14 addps nb134_fizM(%rsp),%xmm15 movaps %xmm7,nb134_fixH1(%rsp) movaps %xmm8,nb134_fiyH1(%rsp) movaps %xmm9,nb134_fizH1(%rsp) movaps %xmm10,nb134_fixH2(%rsp) movaps %xmm11,nb134_fiyH2(%rsp) movaps %xmm12,nb134_fizH2(%rsp) movaps %xmm13,nb134_fixM(%rsp) movaps %xmm14,nb134_fiyM(%rsp) movaps %xmm15,nb134_fizM(%rsp) ## xmm3 = fH2x , xmm4 = fH2y movaps %xmm3,%xmm5 unpcklps %xmm4,%xmm3 unpckhps %xmm4,%xmm5 addps %xmm3,%xmm0 addps %xmm5,%xmm1 movhlps %xmm2,%xmm3 ## fH2zc fH2zd movlps %xmm0,24(%rdi,%rax,4) movhps %xmm0,24(%rdi,%rbx,4) movlps %xmm1,24(%rdi,%rcx,4) movhps %xmm1,24(%rdi,%rdx,4) movss %xmm2,32(%rdi,%rax,4) movss %xmm3,32(%rdi,%rcx,4) shufps $1,%xmm2,%xmm2 shufps $1,%xmm3,%xmm3 movss %xmm2,32(%rdi,%rbx,4) movss %xmm3,32(%rdi,%rdx,4) ## move j M coordinates to local temp variables movq nb134_pos(%rbp),%rsi ## base of pos[] movlps 36(%rsi,%rax,4),%xmm0 ## jxMa jyMa - - movlps 36(%rsi,%rcx,4),%xmm1 ## jxMc jyMc - - movhps 36(%rsi,%rbx,4),%xmm0 ## jxMa jyMa jxMb jyMb movhps 36(%rsi,%rdx,4),%xmm1 ## jxMc jyMc jxMd jyMd movss 44(%rsi,%rax,4),%xmm2 ## jzMa - - - movss 44(%rsi,%rcx,4),%xmm3 ## jzMc - - - movss 44(%rsi,%rbx,4),%xmm5 ## jzMb - - - movss 44(%rsi,%rdx,4),%xmm6 ## jzMd - - - movlhps %xmm5,%xmm2 ## jzMa - jzMb - movlhps %xmm6,%xmm3 ## jzMc - jzMd - movaps %xmm0,%xmm4 unpcklps %xmm1,%xmm0 ## jxMa jxMc jyMa jyMc unpckhps %xmm1,%xmm4 ## jxMb jxMd jyMb jyMd movaps %xmm0,%xmm1 unpcklps %xmm4,%xmm0 ## x unpckhps %xmm4,%xmm1 ## y shufps $136,%xmm3,%xmm2 ## 10001000 => jzMa jzMb jzMc jzMd ## xmm0 = Mx ## xmm1 = My ## xmm2 = Mz movaps %xmm0,%xmm3 movaps %xmm1,%xmm4 movaps %xmm2,%xmm5 movaps %xmm0,%xmm6 movaps %xmm1,%xmm7 movaps %xmm2,%xmm8 subps nb134_ixH1(%rsp),%xmm0 subps nb134_iyH1(%rsp),%xmm1 subps nb134_izH1(%rsp),%xmm2 subps nb134_ixH2(%rsp),%xmm3 subps nb134_iyH2(%rsp),%xmm4 subps nb134_izH2(%rsp),%xmm5 subps nb134_ixM(%rsp),%xmm6 subps nb134_iyM(%rsp),%xmm7 subps nb134_izM(%rsp),%xmm8 movaps %xmm0,nb134_dxH1M(%rsp) movaps %xmm1,nb134_dyH1M(%rsp) movaps %xmm2,nb134_dzH1M(%rsp) mulps %xmm0,%xmm0 mulps %xmm1,%xmm1 mulps %xmm2,%xmm2 movaps %xmm3,nb134_dxH2M(%rsp) movaps %xmm4,nb134_dyH2M(%rsp) movaps %xmm5,nb134_dzH2M(%rsp) mulps %xmm3,%xmm3 mulps %xmm4,%xmm4 mulps %xmm5,%xmm5 movaps %xmm6,nb134_dxMM(%rsp) movaps %xmm7,nb134_dyMM(%rsp) movaps %xmm8,nb134_dzMM(%rsp) mulps %xmm6,%xmm6 mulps %xmm7,%xmm7 mulps %xmm8,%xmm8 addps %xmm1,%xmm0 addps %xmm2,%xmm0 addps %xmm4,%xmm3 addps %xmm5,%xmm3 addps %xmm7,%xmm6 addps %xmm8,%xmm6 ## start doing invsqrt for jM atoms rsqrtps %xmm0,%xmm1 rsqrtps %xmm3,%xmm4 rsqrtps %xmm6,%xmm7 movaps %xmm1,%xmm2 movaps %xmm4,%xmm5 movaps %xmm7,%xmm8 mulps %xmm1,%xmm1 ## lu*lu mulps %xmm4,%xmm4 ## lu*lu mulps %xmm7,%xmm7 ## lu*lu movaps nb134_three(%rsp),%xmm9 movaps %xmm9,%xmm10 movaps %xmm9,%xmm11 mulps %xmm0,%xmm1 ## rsq*lu*lu mulps %xmm3,%xmm4 ## rsq*lu*lu mulps %xmm6,%xmm7 ## rsq*lu*lu subps %xmm1,%xmm9 subps %xmm4,%xmm10 subps %xmm7,%xmm11 ## 3-rsq*lu*lu mulps %xmm2,%xmm9 mulps %xmm5,%xmm10 mulps %xmm8,%xmm11 ## lu*(3-rsq*lu*lu) movaps nb134_half(%rsp),%xmm0 mulps %xmm0,%xmm9 ## rinvH1M mulps %xmm0,%xmm10 ## rinvH2M mulps %xmm0,%xmm11 ## rinvMM ## M interactions movaps %xmm9,%xmm0 movaps %xmm10,%xmm1 movaps %xmm11,%xmm2 mulps %xmm9,%xmm9 mulps %xmm10,%xmm10 mulps %xmm11,%xmm11 mulps nb134_qqMH(%rsp),%xmm0 mulps nb134_qqMH(%rsp),%xmm1 mulps nb134_qqMM(%rsp),%xmm2 mulps %xmm0,%xmm9 mulps %xmm1,%xmm10 mulps %xmm2,%xmm11 addps nb134_vctot(%rsp),%xmm0 addps %xmm2,%xmm1 addps %xmm1,%xmm0 movaps %xmm0,nb134_vctot(%rsp) ## move j M forces to local temp variables movq nb134_faction(%rbp),%rdi movlps 36(%rdi,%rax,4),%xmm0 ## jxMa jyMa - - movlps 36(%rdi,%rcx,4),%xmm1 ## jxMc jyMc - - movhps 36(%rdi,%rbx,4),%xmm0 ## jxMa jyMa jxMb jyMb movhps 36(%rdi,%rdx,4),%xmm1 ## jxMc jyMc jxMd jyMd movss 44(%rdi,%rax,4),%xmm2 ## jzMa - - - movss 44(%rdi,%rcx,4),%xmm3 ## jzMc - - - movss 44(%rdi,%rbx,4),%xmm7 ## jzMb - - - movss 44(%rdi,%rdx,4),%xmm8 ## jzMd - - - movlhps %xmm7,%xmm2 ## jzMa - jzMb - movlhps %xmm8,%xmm3 ## jzMc - jzMd - shufps $136,%xmm3,%xmm2 ## 10001000 => jzMa jzMb jzMc jzMd ## xmm0: jxMa jyMa jxMb jyMb ## xmm1: jxMc jyMc jxMd jyMd ## xmm2: jzMa jzMb jzMc jzMd movaps %xmm9,%xmm7 movaps %xmm9,%xmm8 movaps %xmm11,%xmm13 movaps %xmm11,%xmm14 movaps %xmm11,%xmm15 movaps %xmm10,%xmm11 movaps %xmm10,%xmm12 mulps nb134_dxH1M(%rsp),%xmm7 mulps nb134_dyH1M(%rsp),%xmm8 mulps nb134_dzH1M(%rsp),%xmm9 mulps nb134_dxH2M(%rsp),%xmm10 mulps nb134_dyH2M(%rsp),%xmm11 mulps nb134_dzH2M(%rsp),%xmm12 mulps nb134_dxMM(%rsp),%xmm13 mulps nb134_dyMM(%rsp),%xmm14 mulps nb134_dzMM(%rsp),%xmm15 movaps %xmm7,%xmm3 movaps %xmm8,%xmm4 addps %xmm9,%xmm2 addps nb134_fixH1(%rsp),%xmm7 addps nb134_fiyH1(%rsp),%xmm8 addps nb134_fizH1(%rsp),%xmm9 addps %xmm10,%xmm3 addps %xmm11,%xmm4 addps %xmm12,%xmm2 addps nb134_fixH2(%rsp),%xmm10 addps nb134_fiyH2(%rsp),%xmm11 addps nb134_fizH2(%rsp),%xmm12 addps %xmm13,%xmm3 addps %xmm14,%xmm4 addps %xmm15,%xmm2 addps nb134_fixM(%rsp),%xmm13 addps nb134_fiyM(%rsp),%xmm14 addps nb134_fizM(%rsp),%xmm15 movaps %xmm7,nb134_fixH1(%rsp) movaps %xmm8,nb134_fiyH1(%rsp) movaps %xmm9,nb134_fizH1(%rsp) movaps %xmm10,nb134_fixH2(%rsp) movaps %xmm11,nb134_fiyH2(%rsp) movaps %xmm12,nb134_fizH2(%rsp) movaps %xmm13,nb134_fixM(%rsp) movaps %xmm14,nb134_fiyM(%rsp) movaps %xmm15,nb134_fizM(%rsp) ## xmm3 = fMx , xmm4 = fMy movaps %xmm3,%xmm5 unpcklps %xmm4,%xmm3 unpckhps %xmm4,%xmm5 addps %xmm3,%xmm0 addps %xmm5,%xmm1 movhlps %xmm2,%xmm3 ## fMzc fMzd movlps %xmm0,36(%rdi,%rax,4) movhps %xmm0,36(%rdi,%rbx,4) movlps %xmm1,36(%rdi,%rcx,4) movhps %xmm1,36(%rdi,%rdx,4) movss %xmm2,44(%rdi,%rax,4) movss %xmm3,44(%rdi,%rcx,4) shufps $1,%xmm2,%xmm2 shufps $1,%xmm3,%xmm3 movss %xmm2,44(%rdi,%rbx,4) movss %xmm3,44(%rdi,%rdx,4) ## should we do one more iteration? subl $4,nb134_innerk(%rsp) jl _nb_kernel134_x86_64_sse.nb134_single_check jmp _nb_kernel134_x86_64_sse.nb134_unroll_loop_nb_kernel134_x86_64_sse.nb134_single_check: addl $4,nb134_innerk(%rsp) jnz _nb_kernel134_x86_64_sse.nb134_single_loop jmp _nb_kernel134_x86_64_sse.nb134_updateouterdata_nb_kernel134_x86_64_sse.nb134_single_loop: movq nb134_innerjjnr(%rsp),%rdx ## pointer to jjnr[k] movl (%rdx),%eax addq $4,nb134_innerjjnr(%rsp) movq nb134_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 - ⌨️ 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?