nb_kernel102_x86_64_sse.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,099 行 · 第 1/5 页
S
2,099 行
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 nb102_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 nb102_half(%rsp),%xmm0 mulps %xmm0,%xmm9 ## rinvOH2 mulps %xmm0,%xmm10 ## rinvH1H2 mulps %xmm0,%xmm11 ## rinvH2H2 ## H2 interactions movaps %xmm9,%xmm0 movaps %xmm10,%xmm1 movaps %xmm11,%xmm2 mulps %xmm9,%xmm9 mulps %xmm10,%xmm10 mulps %xmm11,%xmm11 mulps nb102_qqOH(%rsp),%xmm0 mulps nb102_qqHH(%rsp),%xmm1 mulps nb102_qqHH(%rsp),%xmm2 mulps %xmm0,%xmm9 mulps %xmm1,%xmm10 mulps %xmm2,%xmm11 addps nb102_vctot(%rsp),%xmm0 addps %xmm2,%xmm1 addps %xmm1,%xmm0 movaps %xmm0,nb102_vctot(%rsp) ## move j H2 forces to local temp variables 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 - - - movss 32(%rdi,%rbx,4),%xmm7 ## jzH2b - - - movss 32(%rdi,%rdx,4),%xmm8 ## jzH2d - - - movlhps %xmm7,%xmm2 ## jzH2a - jzH2b - movlhps %xmm8,%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 nb102_dxOH2(%rsp),%xmm7 mulps nb102_dyOH2(%rsp),%xmm8 mulps nb102_dzOH2(%rsp),%xmm9 mulps nb102_dxH1H2(%rsp),%xmm10 mulps nb102_dyH1H2(%rsp),%xmm11 mulps nb102_dzH1H2(%rsp),%xmm12 mulps nb102_dxH2H2(%rsp),%xmm13 mulps nb102_dyH2H2(%rsp),%xmm14 mulps nb102_dzH2H2(%rsp),%xmm15 movaps %xmm7,%xmm3 movaps %xmm8,%xmm4 addps %xmm9,%xmm2 addps nb102_fixO(%rsp),%xmm7 addps nb102_fiyO(%rsp),%xmm8 addps nb102_fizO(%rsp),%xmm9 addps %xmm10,%xmm3 addps %xmm11,%xmm4 addps %xmm12,%xmm2 addps nb102_fixH1(%rsp),%xmm10 addps nb102_fiyH1(%rsp),%xmm11 addps nb102_fizH1(%rsp),%xmm12 addps %xmm13,%xmm3 addps %xmm14,%xmm4 addps %xmm15,%xmm2 addps nb102_fixH2(%rsp),%xmm13 addps nb102_fiyH2(%rsp),%xmm14 addps nb102_fizH2(%rsp),%xmm15 movaps %xmm7,nb102_fixO(%rsp) movaps %xmm8,nb102_fiyO(%rsp) movaps %xmm9,nb102_fizO(%rsp) movaps %xmm10,nb102_fixH1(%rsp) movaps %xmm11,nb102_fiyH1(%rsp) movaps %xmm12,nb102_fizH1(%rsp) movaps %xmm13,nb102_fixH2(%rsp) movaps %xmm14,nb102_fiyH2(%rsp) movaps %xmm15,nb102_fizH2(%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) ## should we do one more iteration? subl $4,nb102_innerk(%rsp) jl _nb_kernel102_x86_64_sse.nb102_single_check jmp _nb_kernel102_x86_64_sse.nb102_unroll_loop_nb_kernel102_x86_64_sse.nb102_single_check: addl $4,nb102_innerk(%rsp) jnz _nb_kernel102_x86_64_sse.nb102_single_loop jmp _nb_kernel102_x86_64_sse.nb102_updateouterdata_nb_kernel102_x86_64_sse.nb102_single_loop: movq nb102_innerjjnr(%rsp),%rdx ## pointer to jjnr[k] movl (%rdx),%eax ## index is 32 bits addq $4,nb102_innerjjnr(%rsp) movq nb102_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 ## xmm3 = jxO 0 jxH1 jxH2 shufps $228,%xmm6,%xmm1 ## 11100100 ;# xmm4 = jyO 0 jyH1 jyH2 shufps $68,%xmm7,%xmm2 ## 01000100 ;# xmm5 = jzO 0 jzH1 jzH2 ## store all j coordinates in jO movaps %xmm0,nb102_jxO(%rsp) movaps %xmm1,nb102_jyO(%rsp) movaps %xmm2,nb102_jzO(%rsp) subps nb102_ixO(%rsp),%xmm0 subps nb102_iyO(%rsp),%xmm1 subps nb102_izO(%rsp),%xmm2 movaps %xmm0,nb102_dxOO(%rsp) movaps %xmm1,nb102_dyOO(%rsp) movaps %xmm2,nb102_dzOO(%rsp) mulps %xmm0,%xmm0 mulps %xmm1,%xmm1 mulps %xmm2,%xmm2 addps %xmm1,%xmm0 addps %xmm2,%xmm0 ## have rsq in xmm0 ## do invsqrt rsqrtps %xmm0,%xmm1 movaps %xmm1,%xmm2 mulps %xmm1,%xmm1 movaps nb102_three(%rsp),%xmm3 mulps %xmm0,%xmm1 subps %xmm1,%xmm3 mulps %xmm2,%xmm3 mulps nb102_half(%rsp),%xmm3 ## rinv iO - j water xorps %xmm1,%xmm1 movaps %xmm3,%xmm0 xorps %xmm4,%xmm4 mulps %xmm0,%xmm0 ## xmm0=rinvsq ## fetch charges to xmm4 (temporary) movss nb102_qqOO(%rsp),%xmm4 movhps nb102_qqOH(%rsp),%xmm4 mulps %xmm4,%xmm3 ## xmm3=vcoul mulps %xmm3,%xmm0 ## total fscal addps nb102_vctot(%rsp),%xmm3 movaps %xmm3,nb102_vctot(%rsp) movaps %xmm0,%xmm1 movaps %xmm0,%xmm2 mulps nb102_dxOO(%rsp),%xmm0 mulps nb102_dyOO(%rsp),%xmm1 mulps nb102_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,nb102_fjxO(%rsp) movaps %xmm4,nb102_fjyO(%rsp) movaps %xmm5,nb102_fjzO(%rsp) addps nb102_fixO(%rsp),%xmm0 addps nb102_fiyO(%rsp),%xmm1 addps nb102_fizO(%rsp),%xmm2 movaps %xmm0,nb102_fixO(%rsp) movaps %xmm1,nb102_fiyO(%rsp) movaps %xmm2,nb102_fizO(%rsp) ## done with i O Now do i H1 & H2 simultaneously first get i particle coords: movaps nb102_jxO(%rsp),%xmm0 movaps nb102_jyO(%rsp),%xmm1 movaps nb102_jzO(%rsp),%xmm2 movaps %xmm0,%xmm3 movaps %xmm1,%xmm4 movaps %xmm2,%xmm5 subps nb102_ixH1(%rsp),%xmm0 subps nb102_iyH1(%rsp),%xmm1 subps nb102_izH1(%rsp),%xmm2 subps nb102_ixH2(%rsp),%xmm3 subps nb102_iyH2(%rsp),%xmm4 subps nb102_izH2(%rsp),%xmm5 movaps %xmm0,nb102_dxH1O(%rsp) movaps %xmm1,nb102_dyH1O(%rsp) movaps %xmm2,nb102_dzH1O(%rsp) movaps %xmm3,nb102_dxH2O(%rsp) movaps %xmm4,nb102_dyH2O(%rsp) movaps %xmm5,nb102_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 nb102_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 nb102_half(%rsp),%xmm3 ## rinv H1 - j water mulps nb102_half(%rsp),%xmm7 ## rinv H2 - j water ## assemble charges in xmm6 xorps %xmm6,%xmm6 ## do coulomb interaction movaps %xmm3,%xmm0 movss nb102_qqOH(%rsp),%xmm6 movaps %xmm7,%xmm4 movhps nb102_qqHH(%rsp),%xmm6 mulps %xmm0,%xmm0 ## rinvsq mulps %xmm4,%xmm4 ## rinvsq mulps %xmm6,%xmm3 ## vcoul mulps %xmm6,%xmm7 ## vcoul movaps %xmm3,%xmm2 addps %xmm7,%xmm2 ## total vcoul mulps %xmm3,%xmm0 ## fscal addps nb102_vctot(%rsp),%xmm2 mulps %xmm4,%xmm7 ## fscal movaps %xmm2,nb102_vctot(%rsp) movaps %xmm0,%xmm1 movaps %xmm0,%xmm2 mulps nb102_dxH1O(%rsp),%xmm0 mulps nb102_dyH1O(%rsp),%xmm1 mulps nb102_dzH1O(%rsp),%xmm2 ## update forces H1 - j water movaps nb102_fjxO(%rsp),%xmm3 movaps nb102_fjyO(%rsp),%xmm4 movaps nb102_fjzO(%rsp),%xmm5 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm5 movaps %xmm3,nb102_fjxO(%rsp) movaps %xmm4,nb102_fjyO(%rsp) movaps %xmm5,nb102_fjzO(%rsp) addps nb102_fixH1(%rsp),%xmm0 addps nb102_fiyH1(%rsp),%xmm1 addps nb102_fizH1(%rsp),%xmm2 movaps %xmm0,nb102_fixH1(%rsp) movaps %xmm1,nb102_fiyH1(%rsp) movaps %xmm2,nb102_fizH1(%rsp) ## do forces H2 - j water movaps %xmm7,%xmm0 movaps %xmm7,%xmm1 movaps %xmm7,%xmm2 mulps nb102_dxH2O(%rsp),%xmm0 mulps nb102_dyH2O(%rsp),%xmm1 mulps nb102_dzH2O(%rsp),%xmm2 movaps nb102_fjxO(%rsp),%xmm3 movaps nb102_fjyO(%rsp),%xmm4 movaps nb102_fjzO(%rsp),%xmm5 addps %xmm0,%xmm3 addps %xmm1,%xmm4 addps %xmm2,%xmm5 movq nb102_faction(%rbp),%rsi movaps %xmm3,nb102_fjxO(%rsp) movaps %xmm4,nb102_fjyO(%rsp) movaps %xmm5,nb102_fjzO(%rsp) addps nb102_fixH2(%rsp),%xmm0 addps nb102_fiyH2(%rsp),%xmm1 addps nb102_fizH2(%rsp),%xmm2 movaps %xmm0,nb102_fixH2(%rsp) movaps %xmm1,nb102_fiyH2(%rsp) movaps %xmm2,nb102_fizH2(%rsp) ## 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 nb102_fjxO(%rsp),%xmm3 movaps nb102_fjyO(%rsp),%xmm4 movaps nb102_fjzO(%rsp),%xmm5 movaps %xmm5,%xmm6 movaps %xmm5,%xmm7 shufps $2,%xmm6,%xmm6 ## 00000010 shufps $3,%xmm7,%xmm7 ## 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 nb102_innerk(%rsp) jz _nb_kernel102_x86_64_sse.nb102_updateouterdata jmp _nb_kernel102_x86_64_sse.nb102_single_loop_nb_kernel102_x86_64_sse.nb102_updateouterdata: movl nb102_ii3(%rsp),%ecx movq nb102_faction(%rbp),%rdi movq nb102_fshift(%rbp),%rsi movl nb102_is3(%rsp),%edx ## accumulate Oi forces in xmm0, xmm1, xmm2 movaps nb102_fixO(%rsp),%xmm0 movaps nb102_fiyO(%rsp),%xmm1 movaps nb102_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⌨️ 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?