nb_kernel202_x86_64_sse.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,118 行 · 第 1/5 页
S
2,118 行
movlps 24(%rsi,%rcx,4),%xmm1 ## jxH2c jyH2c - - movhps 24(%rsi,%rbx,4),%xmm0 ## jxH2a jyH2a jxH2b jyH2b movhps 24(%rsi,%rdx,4),%xmm1 ## jxH2c jyH2c jxH2d jyH2d movss 32(%rsi,%rax,4),%xmm2 ## jzH2a - - - movss 32(%rsi,%rcx,4),%xmm3 ## jzH2c - - - movss 32(%rsi,%rbx,4),%xmm5 ## jzOb - - - movss 32(%rsi,%rdx,4),%xmm6 ## jzOd - - - movlhps %xmm5,%xmm2 ## jzOa - jzOb - movlhps %xmm6,%xmm3 ## jzOc - jzOd - movaps %xmm0,%xmm4 unpcklps %xmm1,%xmm0 ## jxH2a jxH2c jyH2a jyH2c unpckhps %xmm1,%xmm4 ## jxH2b jxH2d jyH2b jyH2d movaps %xmm0,%xmm1 unpcklps %xmm4,%xmm0 ## x unpckhps %xmm4,%xmm1 ## y 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 nb202_ixO(%rsp),%xmm0 subps nb202_iyO(%rsp),%xmm1 subps nb202_izO(%rsp),%xmm2 subps nb202_ixH1(%rsp),%xmm3 subps nb202_iyH1(%rsp),%xmm4 subps nb202_izH1(%rsp),%xmm5 subps nb202_ixH2(%rsp),%xmm6 subps nb202_iyH2(%rsp),%xmm7 subps nb202_izH2(%rsp),%xmm8 movaps %xmm0,nb202_dxOH2(%rsp) movaps %xmm1,nb202_dyOH2(%rsp) movaps %xmm2,nb202_dzOH2(%rsp) mulps %xmm0,%xmm0 mulps %xmm1,%xmm1 mulps %xmm2,%xmm2 movaps %xmm3,nb202_dxH1H2(%rsp) movaps %xmm4,nb202_dyH1H2(%rsp) movaps %xmm5,nb202_dzH1H2(%rsp) mulps %xmm3,%xmm3 mulps %xmm4,%xmm4 mulps %xmm5,%xmm5 movaps %xmm6,nb202_dxH2H2(%rsp) movaps %xmm7,nb202_dyH2H2(%rsp) movaps %xmm8,nb202_dzH2H2(%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 nb202_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 nb202_half(%rsp),%xmm4 mulps %xmm4,%xmm9 ## rinvOH2 mulps %xmm4,%xmm10 ## rinvH1H2 mulps %xmm4,%xmm11 ## rinvH2H2 ## H2 interactions ## rsq in xmm0,xmm3,xmm6 ## rinv in xmm9, xmm10, xmm11 movaps %xmm9,%xmm1 ## copy of rinv movaps %xmm10,%xmm4 movaps %xmm11,%xmm7 movaps nb202_krf(%rsp),%xmm2 mulps %xmm9,%xmm9 ## rinvsq mulps %xmm10,%xmm10 mulps %xmm11,%xmm11 mulps %xmm2,%xmm0 ## k*rsq mulps %xmm2,%xmm3 mulps %xmm2,%xmm6 movaps %xmm0,%xmm2 ## copy of k*rsq movaps %xmm3,%xmm5 movaps %xmm6,%xmm8 addps %xmm1,%xmm2 ## rinv+krsq addps %xmm4,%xmm5 addps %xmm7,%xmm8 movaps nb202_crf(%rsp),%xmm14 subps %xmm14,%xmm2 ## rinv+krsq-crf subps %xmm14,%xmm5 subps %xmm14,%xmm8 movaps nb202_qqOH(%rsp),%xmm12 movaps nb202_qqHH(%rsp),%xmm13 mulps %xmm12,%xmm2 ## xmm6=voul=qq*(rinv+ krsq-crf) mulps %xmm13,%xmm5 ## xmm6=voul=qq*(rinv+ krsq-crf) mulps %xmm13,%xmm8 ## xmm6=voul=qq*(rinv+ krsq-crf) addps %xmm0,%xmm0 ## 2*krsq addps %xmm3,%xmm3 addps %xmm6,%xmm6 subps %xmm0,%xmm1 ## rinv-2*krsq subps %xmm3,%xmm4 subps %xmm6,%xmm7 mulps %xmm12,%xmm1 ## (rinv-2*krsq)*qq mulps %xmm13,%xmm4 mulps %xmm13,%xmm7 addps %xmm8,%xmm5 addps %xmm15,%xmm2 addps %xmm5,%xmm2 movaps %xmm2,nb202_vctot(%rsp) mulps %xmm9,%xmm1 ## fscal mulps %xmm10,%xmm4 mulps %xmm11,%xmm7 ## move j H2 forces to local temp variables movlps 24(%rdi,%rax,4),%xmm9 ## jxH2a jyH2a - - movlps 24(%rdi,%rcx,4),%xmm10 ## jxH2c jyH2c - - movhps 24(%rdi,%rbx,4),%xmm9 ## jxH2a jyH2a jxH2b jyH2b movhps 24(%rdi,%rdx,4),%xmm10 ## jxH2c jyH2c jxH2d jyH2d movss 32(%rdi,%rax,4),%xmm11 ## jzH2a - - - movss 32(%rdi,%rcx,4),%xmm12 ## jzH2c - - - movss 32(%rdi,%rbx,4),%xmm2 ## jzH2b - - - movss 32(%rdi,%rdx,4),%xmm3 ## jzH2d - - - movlhps %xmm2,%xmm11 ## jzH2a - jzH2b - movlhps %xmm3,%xmm12 ## jzH2c - jzH2d - shufps $136,%xmm12,%xmm11 ## 10001000 => jzH2a jzH2b jzH2c jzH2d ## xmm9: jxH2a jyH2a jxH2b jyH2b ## xmm10: jxH2c jyH2c jxH2d jyH2d ## xmm11: jzH2a jzH2b jzH2c jzH2d movaps %xmm1,%xmm0 movaps %xmm1,%xmm2 movaps %xmm4,%xmm3 movaps %xmm4,%xmm5 movaps %xmm7,%xmm6 movaps %xmm7,%xmm8 mulps nb202_dxOH2(%rsp),%xmm0 mulps nb202_dyOH2(%rsp),%xmm1 mulps nb202_dzOH2(%rsp),%xmm2 mulps nb202_dxH1H2(%rsp),%xmm3 mulps nb202_dyH1H2(%rsp),%xmm4 mulps nb202_dzH1H2(%rsp),%xmm5 mulps nb202_dxH2H2(%rsp),%xmm6 mulps nb202_dyH2H2(%rsp),%xmm7 mulps nb202_dzH2H2(%rsp),%xmm8 movaps %xmm0,%xmm13 movaps %xmm1,%xmm14 addps %xmm2,%xmm11 addps nb202_fixO(%rsp),%xmm0 addps nb202_fiyO(%rsp),%xmm1 addps nb202_fizO(%rsp),%xmm2 addps %xmm3,%xmm13 addps %xmm4,%xmm14 addps %xmm5,%xmm11 addps nb202_fixH1(%rsp),%xmm3 addps nb202_fiyH1(%rsp),%xmm4 addps nb202_fizH1(%rsp),%xmm5 addps %xmm6,%xmm13 addps %xmm7,%xmm14 addps %xmm8,%xmm11 addps nb202_fixH2(%rsp),%xmm6 addps nb202_fiyH2(%rsp),%xmm7 addps nb202_fizH2(%rsp),%xmm8 movaps %xmm0,nb202_fixO(%rsp) movaps %xmm1,nb202_fiyO(%rsp) movaps %xmm2,nb202_fizO(%rsp) movaps %xmm3,nb202_fixH1(%rsp) movaps %xmm4,nb202_fiyH1(%rsp) movaps %xmm5,nb202_fizH1(%rsp) movaps %xmm6,nb202_fixH2(%rsp) movaps %xmm7,nb202_fiyH2(%rsp) movaps %xmm8,nb202_fizH2(%rsp) ## xmm0 = fH2x ## xmm1 = fH2y ## xmm2 = fH2z movaps %xmm13,%xmm15 unpcklps %xmm14,%xmm13 unpckhps %xmm14,%xmm15 addps %xmm13,%xmm9 addps %xmm15,%xmm10 movhlps %xmm11,%xmm12 ## fH2zc fH2zd movlps %xmm9,24(%rdi,%rax,4) movhps %xmm9,24(%rdi,%rbx,4) movlps %xmm10,24(%rdi,%rcx,4) movhps %xmm10,24(%rdi,%rdx,4) movss %xmm11,32(%rdi,%rax,4) movss %xmm12,32(%rdi,%rcx,4) shufps $1,%xmm11,%xmm11 shufps $1,%xmm12,%xmm12 movss %xmm11,32(%rdi,%rbx,4) movss %xmm12,32(%rdi,%rdx,4) ## should we do one more iteration? subl $4,nb202_innerk(%rsp) jl _nb_kernel202_x86_64_sse.nb202_single_check jmp _nb_kernel202_x86_64_sse.nb202_unroll_loop_nb_kernel202_x86_64_sse.nb202_single_check: addl $4,nb202_innerk(%rsp) jnz _nb_kernel202_x86_64_sse.nb202_single_loop jmp _nb_kernel202_x86_64_sse.nb202_updateouterdata_nb_kernel202_x86_64_sse.nb202_single_loop: movq nb202_innerjjnr(%rsp),%rdx ## pointer to jjnr[k] movl (%rdx),%eax addq $4,nb202_innerjjnr(%rsp) movq nb202_pos(%rbp),%rsi lea (%rax,%rax,2),%rax ## fetch j coordinates xorps %xmm0,%xmm0 xorps %xmm1,%xmm1 xorps %xmm2,%xmm2 movss (%rsi,%rax,4),%xmm0 ## jxO - - - movss 4(%rsi,%rax,4),%xmm1 ## jyO - - - movss 8(%rsi,%rax,4),%xmm2 ## jzO - - - movlps 12(%rsi,%rax,4),%xmm6 ## xmm6 = jxH1 jyH1 - - movss 20(%rsi,%rax,4),%xmm7 ## xmm7 = jzH1 - - - movhps 24(%rsi,%rax,4),%xmm6 ## xmm6 = jxH1 jyH1 jxH2 jyH2 movss 32(%rsi,%rax,4),%xmm5 ## xmm5 = jzH2 - - - ## have all coords, time for some shuffling. shufps $216,%xmm6,%xmm6 ## 11011000 ;# xmm6 = jxH1 jxH2 jyH1 jyH2 unpcklps %xmm5,%xmm7 ## xmm7 = jzH1 jzH2 - - movlhps %xmm6,%xmm0 ## xmm0 = jxO 0 jxH1 jxH2 shufps $228,%xmm6,%xmm1 ## 11100100 ;# xmm1 = jyO 0 jyH1 jyH2 shufps $68,%xmm7,%xmm2 ## 01000100 ;# xmm2 = jzO 0 jzH1 jzH2 movaps %xmm0,nb202_jxO(%rsp) movaps %xmm1,nb202_jyO(%rsp) movaps %xmm2,nb202_jzO(%rsp) subps nb202_ixO(%rsp),%xmm0 subps nb202_iyO(%rsp),%xmm1 subps nb202_izO(%rsp),%xmm2 movaps %xmm0,nb202_dxOO(%rsp) movaps %xmm1,nb202_dyOO(%rsp) movaps %xmm2,nb202_dzOO(%rsp) mulps %xmm0,%xmm0 mulps %xmm1,%xmm1 mulps %xmm2,%xmm2 addps %xmm1,%xmm0 addps %xmm2,%xmm0 ## have rsq in xmm0 movaps %xmm0,%xmm6 ## do invsqrt rsqrtps %xmm0,%xmm1 mulps nb202_krf(%rsp),%xmm6 ## xmm6=krsq movaps %xmm1,%xmm2 movaps %xmm6,%xmm7 ## xmm7=krsq mulps %xmm1,%xmm1 movaps nb202_three(%rsp),%xmm3 mulps %xmm0,%xmm1 subps %xmm1,%xmm3 mulps %xmm2,%xmm3 mulps nb202_half(%rsp),%xmm3 ## rinv iO - j water addps %xmm3,%xmm6 ## xmm6=rinv+ krsq mulps nb202_two(%rsp),%xmm7 subps nb202_crf(%rsp),%xmm6 ## xmm6=rinv+ krsq-crf xorps %xmm1,%xmm1 movaps %xmm3,%xmm0 subps %xmm7,%xmm3 ## xmm3=rinv-2*krsq xorps %xmm4,%xmm4 mulps %xmm0,%xmm0 ## xmm0=rinvsq ## fetch charges to xmm4 (temporary) movss nb202_qqOO(%rsp),%xmm4 movhps nb202_qqOH(%rsp),%xmm4 mulps %xmm4,%xmm6 ## vcoul mulps %xmm4,%xmm3 ## coul part of fs addps nb202_vctot(%rsp),%xmm6 mulps %xmm3,%xmm0 ## total fscal movaps %xmm6,nb202_vctot(%rsp) movaps %xmm0,%xmm1 movaps %xmm0,%xmm2 mulps nb202_dxOO(%rsp),%xmm0 mulps nb202_dyOO(%rsp),%xmm1 mulps nb202_dzOO(%rsp),%xmm2 ## initial update for j forces xorps %xmm3,%xmm3 xorps %xmm4,%xmm4 xorps %xmm5,%xmm5 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm5 movaps %xmm3,nb202_fjxO(%rsp) movaps %xmm4,nb202_fjyO(%rsp) movaps %xmm5,nb202_fjzO(%rsp) addps nb202_fixO(%rsp),%xmm0 addps nb202_fiyO(%rsp),%xmm1 addps nb202_fizO(%rsp),%xmm2 movaps %xmm0,nb202_fixO(%rsp) movaps %xmm1,nb202_fiyO(%rsp) movaps %xmm2,nb202_fizO(%rsp) ## done with i O Now do i H1 & H2 simultaneously first get i particle coords: movaps nb202_jxO(%rsp),%xmm0 movaps nb202_jyO(%rsp),%xmm1 movaps nb202_jzO(%rsp),%xmm2 movaps %xmm0,%xmm3 movaps %xmm1,%xmm4 movaps %xmm2,%xmm5 subps nb202_ixH1(%rsp),%xmm0 subps nb202_iyH1(%rsp),%xmm1 subps nb202_izH1(%rsp),%xmm2 subps nb202_ixH2(%rsp),%xmm3 subps nb202_iyH2(%rsp),%xmm4 subps nb202_izH2(%rsp),%xmm5 movaps %xmm0,nb202_dxH1O(%rsp) movaps %xmm1,nb202_dyH1O(%rsp) movaps %xmm2,nb202_dzH1O(%rsp) movaps %xmm3,nb202_dxH2O(%rsp) movaps %xmm4,nb202_dyH2O(%rsp) movaps %xmm5,nb202_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 nb202_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 nb202_half(%rsp),%xmm3 ## rinv H1 - j water mulps nb202_half(%rsp),%xmm7 ## rinv H2 - j water mulps nb202_krf(%rsp),%xmm0 ## krsq mulps nb202_krf(%rsp),%xmm4 ## krsq ## assemble charges in xmm6 xorps %xmm6,%xmm6 movss nb202_qqOH(%rsp),%xmm6 movhps nb202_qqHH(%rsp),%xmm6 movaps %xmm0,%xmm1 movaps %xmm4,%xmm5 addps %xmm3,%xmm0 ## krsq+ rinv addps %xmm7,%xmm4 ## krsq+ rinv subps nb202_crf(%rsp),%xmm0 subps nb202_crf(%rsp),%xmm4 mulps nb202_two(%rsp),%xmm1 mulps nb202_two(%rsp),%xmm5 mulps %xmm6,%xmm0 ## vcoul mulps %xmm6,%xmm4 ## vcoul addps %xmm0,%xmm4 addps nb202_vctot(%rsp),%xmm4 movaps %xmm4,nb202_vctot(%rsp) movaps %xmm3,%xmm0 movaps %xmm7,%xmm4 mulps %xmm3,%xmm3⌨️ 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?