nb_kernel131_x86_64_sse2.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,044 行 · 第 1/4 页
S
2,044 行
mulpd %xmm3,%xmm3 mulpd %xmm4,%xmm4 mulpd %xmm5,%xmm5 addpd %xmm4,%xmm5 addpd %xmm3,%xmm5 ## rsqH2 in xmm5, rsqH1 in xmm6, rsqO in xmm7 ## start with rsqO - put seed in xmm2 cvtpd2ps %xmm7,%xmm2 rsqrtps %xmm2,%xmm2 cvtps2pd %xmm2,%xmm2 movapd %xmm2,%xmm3 mulpd %xmm2,%xmm2 movapd nb131nf_three(%rsp),%xmm4 mulpd %xmm7,%xmm2 ## rsq*lu*lu subpd %xmm2,%xmm4 ## 30-rsq*lu*lu mulpd %xmm3,%xmm4 ## lu*(3-rsq*lu*lu) mulpd nb131nf_half(%rsp),%xmm4 ## iter1 ( new lu) movapd %xmm4,%xmm3 mulpd %xmm4,%xmm4 ## lu*lu mulpd %xmm4,%xmm7 ## rsq*lu*lu movapd nb131nf_three(%rsp),%xmm4 subpd %xmm7,%xmm4 ## 3-rsq*lu*lu mulpd %xmm3,%xmm4 ## lu*( 3-rsq*lu*lu) mulpd nb131nf_half(%rsp),%xmm4 ## rinv movapd %xmm4,%xmm7 ## rinvO in xmm7 ## rsqH1 - seed in xmm2 cvtpd2ps %xmm6,%xmm2 rsqrtps %xmm2,%xmm2 cvtps2pd %xmm2,%xmm2 movapd %xmm2,%xmm3 mulpd %xmm2,%xmm2 movapd nb131nf_three(%rsp),%xmm4 mulpd %xmm6,%xmm2 ## rsq*lu*lu subpd %xmm2,%xmm4 ## 30-rsq*lu*lu mulpd %xmm3,%xmm4 ## lu*(3-rsq*lu*lu) mulpd nb131nf_half(%rsp),%xmm4 ## iter1 ( new lu) movapd %xmm4,%xmm3 mulpd %xmm4,%xmm4 ## lu*lu mulpd %xmm4,%xmm6 ## rsq*lu*lu movapd nb131nf_three(%rsp),%xmm4 subpd %xmm6,%xmm4 ## 3-rsq*lu*lu mulpd %xmm3,%xmm4 ## lu*( 3-rsq*lu*lu) mulpd nb131nf_half(%rsp),%xmm4 ## rinv movapd %xmm4,%xmm6 ## rinvH1 in xmm6 movapd %xmm6,nb131nf_rinvH1(%rsp) ## rsqH2 - seed in xmm2 cvtpd2ps %xmm5,%xmm2 rsqrtps %xmm2,%xmm2 cvtps2pd %xmm2,%xmm2 movapd %xmm2,%xmm3 mulpd %xmm2,%xmm2 movapd nb131nf_three(%rsp),%xmm4 mulpd %xmm5,%xmm2 ## rsq*lu*lu subpd %xmm2,%xmm4 ## 30-rsq*lu*lu mulpd %xmm3,%xmm4 ## lu*(3-rsq*lu*lu) mulpd nb131nf_half(%rsp),%xmm4 ## iter1 ( new lu) movapd %xmm4,%xmm3 mulpd %xmm4,%xmm4 ## lu*lu mulpd %xmm4,%xmm5 ## rsq*lu*lu movapd nb131nf_three(%rsp),%xmm4 subpd %xmm5,%xmm4 ## 3-rsq*lu*lu mulpd %xmm3,%xmm4 ## lu*( 3-rsq*lu*lu) mulpd nb131nf_half(%rsp),%xmm4 ## rinv movapd %xmm4,%xmm5 ## rinvH2 in xmm5 movapd %xmm5,nb131nf_rinvH2(%rsp) ## do O interactions movapd %xmm7,%xmm0 mulpd nb131nf_qqO(%rsp),%xmm7 addpd nb131nf_vctot(%rsp),%xmm7 movapd %xmm7,nb131nf_vctot(%rsp) movapd nb131nf_rsqO(%rsp),%xmm4 ## LJ table interaction. xmm0=rinv, xmm4=rsq mulpd %xmm0,%xmm4 ## xmm4=r mulpd nb131nf_tsc(%rsp),%xmm4 cvttpd2pi %xmm4,%mm6 ## mm6 = lu idx cvtpi2pd %mm6,%xmm5 subpd %xmm5,%xmm4 movapd %xmm4,%xmm1 ## xmm1=eps movapd %xmm1,%xmm2 mulpd %xmm2,%xmm2 ## xmm2=eps2 pslld $3,%mm6 ## idx *= 8 movq nb131nf_VFtab(%rbp),%rsi movd %mm6,%eax psrlq $32,%mm6 movd %mm6,%ebx ## dispersion movlpd (%rsi,%rax,8),%xmm4 ## Y1 movlpd (%rsi,%rbx,8),%xmm3 ## Y2 movhpd 8(%rsi,%rax,8),%xmm4 ## Y1 F1 movhpd 8(%rsi,%rbx,8),%xmm3 ## Y2 F2 movapd %xmm4,%xmm5 unpcklpd %xmm3,%xmm4 ## Y1 Y2 unpckhpd %xmm3,%xmm5 ## F1 F2 movlpd 16(%rsi,%rax,8),%xmm6 ## G1 movlpd 16(%rsi,%rbx,8),%xmm3 ## G2 movhpd 24(%rsi,%rax,8),%xmm6 ## G1 H1 movhpd 24(%rsi,%rbx,8),%xmm3 ## G2 H2 movapd %xmm6,%xmm7 unpcklpd %xmm3,%xmm6 ## G1 G2 unpckhpd %xmm3,%xmm7 ## H1 H2 ## dispersion table ready, in xmm4-xmm7 mulpd %xmm1,%xmm6 ## xmm6=Geps mulpd %xmm2,%xmm7 ## xmm7=Heps2 addpd %xmm6,%xmm5 addpd %xmm7,%xmm5 ## xmm5=Fp mulpd %xmm1,%xmm5 ## xmm5=eps*Fp addpd %xmm4,%xmm5 ## xmm5=VV movapd nb131nf_c6(%rsp),%xmm4 mulpd %xmm4,%xmm5 ## Vvdw6 ## Update Vvdwtot directly addpd nb131nf_Vvdwtot(%rsp),%xmm5 movapd %xmm5,nb131nf_Vvdwtot(%rsp) ## repulsion movlpd 32(%rsi,%rax,8),%xmm4 ## Y1 movlpd 32(%rsi,%rbx,8),%xmm3 ## Y2 movhpd 40(%rsi,%rax,8),%xmm4 ## Y1 F1 movhpd 40(%rsi,%rbx,8),%xmm3 ## Y2 F2 movapd %xmm4,%xmm5 unpcklpd %xmm3,%xmm4 ## Y1 Y2 unpckhpd %xmm3,%xmm5 ## F1 F2 movlpd 48(%rsi,%rax,8),%xmm6 ## G1 movlpd 48(%rsi,%rbx,8),%xmm3 ## G2 movhpd 56(%rsi,%rax,8),%xmm6 ## G1 H1 movhpd 56(%rsi,%rbx,8),%xmm3 ## G2 H2 movapd %xmm6,%xmm7 unpcklpd %xmm3,%xmm6 ## G1 G2 unpckhpd %xmm3,%xmm7 ## H1 H2 ## table ready, in xmm4-xmm7 mulpd %xmm1,%xmm6 ## xmm6=Geps mulpd %xmm2,%xmm7 ## xmm7=Heps2 addpd %xmm6,%xmm5 addpd %xmm7,%xmm5 ## xmm5=Fp mulpd %xmm1,%xmm5 ## xmm5=eps*Fp addpd %xmm4,%xmm5 ## xmm5=VV movapd nb131nf_c12(%rsp),%xmm4 mulpd %xmm4,%xmm5 addpd nb131nf_Vvdwtot(%rsp),%xmm5 movapd %xmm5,nb131nf_Vvdwtot(%rsp) ## H1 & H2 interactions movapd nb131nf_rinvH1(%rsp),%xmm6 addpd nb131nf_rinvH2(%rsp),%xmm6 mulpd nb131nf_qqH(%rsp),%xmm6 ## vcoul addpd nb131nf_vctot(%rsp),%xmm6 movapd %xmm6,nb131nf_vctot(%rsp) ## should we do one more iteration? subl $2,nb131nf_innerk(%rsp) jl _nb_kernel131nf_x86_64_sse2.nb131nf_checksingle jmp _nb_kernel131nf_x86_64_sse2.nb131nf_unroll_loop_nb_kernel131nf_x86_64_sse2.nb131nf_checksingle: movl nb131nf_innerk(%rsp),%edx andl $1,%edx jnz _nb_kernel131nf_x86_64_sse2.nb131nf_dosingle jmp _nb_kernel131nf_x86_64_sse2.nb131nf_updateouterdata_nb_kernel131nf_x86_64_sse2.nb131nf_dosingle: movq nb131nf_innerjjnr(%rsp),%rdx ## pointer to jjnr[k] movl (%rdx),%eax addq $4,nb131nf_innerjjnr(%rsp) movq nb131nf_charge(%rbp),%rsi ## base of charge[] xorpd %xmm3,%xmm3 movlpd (%rsi,%rax,8),%xmm3 movapd %xmm3,%xmm4 mulpd nb131nf_iqO(%rsp),%xmm3 mulpd nb131nf_iqH(%rsp),%xmm4 movd %eax,%mm0 ## use mmx registers as temp storage movapd %xmm3,nb131nf_qqO(%rsp) movapd %xmm4,nb131nf_qqH(%rsp) movq nb131nf_type(%rbp),%rsi movl (%rsi,%rax,4),%eax movq nb131nf_vdwparam(%rbp),%rsi shll %eax movl nb131nf_ntia(%rsp),%edi addl %edi,%eax movlpd (%rsi,%rax,8),%xmm6 ## c6a movhpd 8(%rsi,%rax,8),%xmm6 ## c6a c12a xorpd %xmm7,%xmm7 movapd %xmm6,%xmm4 unpcklpd %xmm7,%xmm4 unpckhpd %xmm7,%xmm6 movd %mm0,%eax movapd %xmm4,nb131nf_c6(%rsp) movapd %xmm6,nb131nf_c12(%rsp) movq nb131nf_pos(%rbp),%rsi ## base of pos[] lea (%rax,%rax,2),%rax ## replace jnr with j3 ## move coordinates to xmm0-xmm2 movlpd (%rsi,%rax,8),%xmm0 movlpd 8(%rsi,%rax,8),%xmm1 movlpd 16(%rsi,%rax,8),%xmm2 ## move ixO-izO to xmm4-xmm6 movapd nb131nf_ixO(%rsp),%xmm4 movapd nb131nf_iyO(%rsp),%xmm5 movapd nb131nf_izO(%rsp),%xmm6 ## calc dr subsd %xmm0,%xmm4 subsd %xmm1,%xmm5 subsd %xmm2,%xmm6 ## square it mulsd %xmm4,%xmm4 mulsd %xmm5,%xmm5 mulsd %xmm6,%xmm6 addsd %xmm5,%xmm4 addsd %xmm6,%xmm4 movapd %xmm4,%xmm7 ## rsqO in xmm7 movapd %xmm7,nb131nf_rsqO(%rsp) ## move ixH1-izH1 to xmm4-xmm6 movapd nb131nf_ixH1(%rsp),%xmm4 movapd nb131nf_iyH1(%rsp),%xmm5 movapd nb131nf_izH1(%rsp),%xmm6 ## calc dr subsd %xmm0,%xmm4 subsd %xmm1,%xmm5 subsd %xmm2,%xmm6 ## square it mulsd %xmm4,%xmm4 mulsd %xmm5,%xmm5 mulsd %xmm6,%xmm6 addsd %xmm5,%xmm6 addsd %xmm4,%xmm6 ## rsqH1 in xmm6 ## move ixH2-izH2 to xmm3-xmm5 movapd nb131nf_ixH2(%rsp),%xmm3 movapd nb131nf_iyH2(%rsp),%xmm4 movapd nb131nf_izH2(%rsp),%xmm5 ## calc dr subsd %xmm0,%xmm3 subsd %xmm1,%xmm4 subsd %xmm2,%xmm5 ## square it mulsd %xmm3,%xmm3 mulsd %xmm4,%xmm4 mulsd %xmm5,%xmm5 addsd %xmm4,%xmm5 addsd %xmm3,%xmm5 ## rsqH2 in xmm5, rsqH1 in xmm6, rsqO in xmm7 ## start with rsqO - put seed in xmm2 cvtsd2ss %xmm7,%xmm2 rsqrtss %xmm2,%xmm2 cvtss2sd %xmm2,%xmm2 movapd %xmm2,%xmm3 mulsd %xmm2,%xmm2 movapd nb131nf_three(%rsp),%xmm4 mulsd %xmm7,%xmm2 ## rsq*lu*lu subsd %xmm2,%xmm4 ## 30-rsq*lu*lu mulsd %xmm3,%xmm4 ## lu*(3-rsq*lu*lu) mulsd nb131nf_half(%rsp),%xmm4 ## iter1 ( new lu) movapd %xmm4,%xmm3 mulsd %xmm4,%xmm4 ## lu*lu mulsd %xmm4,%xmm7 ## rsq*lu*lu movapd nb131nf_three(%rsp),%xmm4 subsd %xmm7,%xmm4 ## 3-rsq*lu*lu mulsd %xmm3,%xmm4 ## lu*( 3-rsq*lu*lu) mulsd nb131nf_half(%rsp),%xmm4 ## rinv movapd %xmm4,%xmm7 ## rinvO in xmm7 ## rsqH1 - seed in xmm2 cvtsd2ss %xmm6,%xmm2 rsqrtss %xmm2,%xmm2 cvtss2sd %xmm2,%xmm2 movapd %xmm2,%xmm3 mulsd %xmm2,%xmm2 movapd nb131nf_three(%rsp),%xmm4 mulsd %xmm6,%xmm2 ## rsq*lu*lu subsd %xmm2,%xmm4 ## 30-rsq*lu*lu mulsd %xmm3,%xmm4 ## lu*(3-rsq*lu*lu) mulsd nb131nf_half(%rsp),%xmm4 ## iter1 ( new lu) movapd %xmm4,%xmm3 mulsd %xmm4,%xmm4 ## lu*lu mulsd %xmm4,%xmm6 ## rsq*lu*lu movapd nb131nf_three(%rsp),%xmm4 subsd %xmm6,%xmm4 ## 3-rsq*lu*lu mulsd %xmm3,%xmm4 ## lu*( 3-rsq*lu*lu) mulsd nb131nf_half(%rsp),%xmm4 ## rinv movapd %xmm4,%xmm6 ## rinvH1 in xmm6 movsd %xmm6,nb131nf_rinvH1(%rsp) ## rsqH2 - seed in xmm2 cvtsd2ss %xmm5,%xmm2 rsqrtss %xmm2,%xmm2 cvtss2sd %xmm2,%xmm2 movapd %xmm2,%xmm3 mulsd %xmm2,%xmm2 movapd nb131nf_three(%rsp),%xmm4 mulsd %xmm5,%xmm2 ## rsq*lu*lu subsd %xmm2,%xmm4 ## 30-rsq*lu*lu mulsd %xmm3,%xmm4 ## lu*(3-rsq*lu*lu) mulsd nb131nf_half(%rsp),%xmm4 ## iter1 ( new lu) movapd %xmm4,%xmm3 mulsd %xmm4,%xmm4 ## lu*lu mulsd %xmm4,%xmm5 ## rsq*lu*lu movapd nb131nf_three(%rsp),%xmm4 subsd %xmm5,%xmm4 ## 3-rsq*lu*lu mulsd %xmm3,%xmm4 ## lu*( 3-rsq*lu*lu) mulsd nb131nf_half(%rsp),%xmm4 ## rinv movapd %xmm4,%xmm5 ## rinvH2 in xmm5 movsd %xmm5,nb131nf_rinvH2(%rsp) ## do O interactions movsd %xmm7,%xmm0 mulsd nb131nf_qqO(%rsp),%xmm7 addsd nb131nf_vctot(%rsp),%xmm7 movsd %xmm7,nb131nf_vctot(%rsp) movsd nb131nf_rsqO(%rsp),%xmm4 ## LJ table interaction. xmm0=rinv, xmm4=rsq mulsd %xmm0,%xmm4 ## xmm4=r mulsd nb131nf_tsc(%rsp),%xmm4 cvttsd2si %xmm4,%ebx ## mm6 = lu idx cvtsi2sd %ebx,%xmm5 subpd %xmm5,%xmm4 movapd %xmm4,%xmm1 ## xmm1=eps movapd %xmm1,%xmm2 mulpd %xmm2,%xmm2 ## xmm2=eps2 shll $3,%ebx movq nb131nf_VFtab(%rbp),%rsi ## dispersion movlpd (%rsi,%rbx,8),%xmm4 ## Y1 movhpd 8(%rsi,%rbx,8),%xmm4 ## Y1 F1 movapd %xmm4,%xmm5 unpcklpd %xmm3,%xmm4 ## Y1 Y2 unpckhpd %xmm3,%xmm5 ## F1 F2 movlpd 16(%rsi,%rbx,8),%xmm6 ## G1 movhpd 24(%rsi,%rbx,8),%xmm6 ## G1 H1 movapd %xmm6,%xmm7 unpcklpd %xmm3,%xmm6 ## G1 G2 unpckhpd %xmm3,%xmm7 ## H1 H2 ## dispersion table ready, in xmm4-xmm7 mulsd %xmm1,%xmm6 ## xmm6=Geps mulsd %xmm2,%xmm7 ## xmm7=Heps2 addsd %xmm6,%xmm5 addsd %xmm7,%xmm5 ## xmm5=Fp mulsd %xmm1,%xmm5 ## xmm5=eps*Fp addsd %xmm4,%xmm5 ## xmm5=VV movsd nb131nf_c6(%rsp),%xmm4 mulsd %xmm4,%xmm5 ## Vvdw6 ## Update Vvdwtot directly addsd nb131nf_Vvdwtot(%rsp),%xmm5 movsd %xmm5,nb131nf_Vvdwtot(%rsp) ## repulsion movlpd 32(%rsi,%rbx,8),%xmm4 ## Y1 movhpd 40(%rsi,%rbx,8),%xmm4 ## Y1 F1 movapd %xmm4,%xmm5 unpcklpd %xmm3,%xmm4 ## Y1 Y2 unpckhpd %xmm3,%xmm5 ## F1 F2 movlpd 48(%rsi,%rbx,8),%xmm6 ## G1 movhpd 56(%rsi,%rbx,8),%xmm6 ## G1 H1 movapd %xmm6,%xmm7 unpcklpd %xmm3,%xmm6 ## G1 G2 unpckhpd %xmm3,%xmm7 ## H1 H2 ## table ready, in xmm4-xmm7 mulsd %xmm1,%xmm6 ## xmm6=Geps mulsd %xmm2,%xmm7 ## xmm7=Heps2 addsd %xmm6,%xmm5 addsd %xmm7,%xmm5 ## xmm5=Fp mulsd %xmm1,%xmm5 ## xmm5=eps*Fp addsd %xmm4,%xmm5 ## xmm5=VV movsd nb131nf_c12(%rsp),%xmm4 mulsd %xmm4,%xmm5 addsd nb131nf_Vvdwtot(%rsp),%xmm5 movsd %xmm5,nb131nf_Vvdwtot(%rsp) ## H1 & H2 interactions movsd nb131nf_rinvH1(%rsp),%xmm6 addsd nb131nf_rinvH2(%rsp),%xmm6 mulsd nb131nf_qqH(%rsp),%xmm6 ## vcoul addsd nb131nf_vctot(%rsp),%xmm6 movsd %xmm6,nb131nf_vctot(%rsp)_nb_kernel131nf_x86_64_sse2.nb131nf_updateouterdata: ## get n from stack movl nb131nf_n(%rsp),%esi ## get group index for i particle movq nb131nf_gid(%rbp),%rdx ## base of gid[] movl (%rdx,%rsi,4),%edx ## ggid=gid[n] ## accumulate total potential energy and update it movapd nb131nf_vctot(%rsp),%xmm7 ## accumulate movhlps %xmm7,%xmm6 addsd %xmm6,%xmm7 ## low xmm7 has the sum now ## add earlier value from mem movq nb131nf_Vc(%rbp),%rax addsd (%rax,%rdx,8),%xmm7 ## move back to mem movsd %xmm7,(%rax,%rdx,8) ## accumulate total lj energy and update it movapd nb131nf_Vvdwtot(%rsp),%xmm7 ## accumulate movhlps %xmm7,%xmm6 addsd %xmm6,%xmm7 ## low xmm7 has the sum now ## add earlier value from mem movq nb131nf_Vvdw(%rbp),%rax addsd (%rax,%rdx,8),%xmm7 ## move back to mem movsd %xmm7,(%rax,%rdx,8) ## finish if last movl nb131nf_nn1(%rsp),%ecx ## esi already loaded with n incl %esi subl %esi,%ecx jz _nb_kernel131nf_x86_64_sse2.nb131nf_outerend ## not last, iterate outer loop once more! movl %esi,nb131nf_n(%rsp) jmp _nb_kernel131nf_x86_64_sse2.nb131nf_outer_nb_kernel131nf_x86_64_sse2.nb131nf_outerend: ## check if more outer neighborlists remain movl nb131nf_nri(%rsp),%ecx ## esi already loaded with n above subl %esi,%ecx jz _nb_kernel131nf_x86_64_sse2.nb131nf_end ## non-zero, do one more workunit jmp _nb_kernel131nf_x86_64_sse2.nb131nf_threadloop_nb_kernel131nf_x86_64_sse2.nb131nf_end: movl nb131nf_nouter(%rsp),%eax movl nb131nf_ninner(%rsp),%ebx movq nb131nf_outeriter(%rbp),%rcx movq nb131nf_inneriter(%rbp),%rdx movl %eax,(%rcx) movl %ebx,(%rdx) addq $520,%rsp emms pop %r15 pop %r14 pop %r13 pop %r12 pop %rbx pop %rbp ret⌨️ 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?