nb_kernel313_x86_64_sse2.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,129 行 · 第 1/5 页
S
2,129 行
movsd (%rax,%rbx,8),%xmm0 movsd 8(%rax,%rbx,8),%xmm1 movsd 16(%rax,%rbx,8),%xmm2 movq nb313nf_iinr(%rsp),%rcx ## rcx = pointer into iinr[] movl (%rcx,%rsi,4),%ebx ## ebx =ii movapd %xmm0,%xmm3 movapd %xmm1,%xmm4 movapd %xmm2,%xmm5 movapd %xmm0,%xmm6 movapd %xmm1,%xmm7 lea (%rbx,%rbx,2),%rbx ## rbx = 3*ii=ii3 movq nb313nf_pos(%rbp),%rax ## rax = base of pos[] movl %ebx,nb313nf_ii3(%rsp) addsd (%rax,%rbx,8),%xmm3 ## ox addsd 8(%rax,%rbx,8),%xmm4 ## oy addsd 16(%rax,%rbx,8),%xmm5 ## oz addsd 24(%rax,%rbx,8),%xmm6 ## h1x addsd 32(%rax,%rbx,8),%xmm7 ## h1y shufpd $0,%xmm3,%xmm3 shufpd $0,%xmm4,%xmm4 shufpd $0,%xmm5,%xmm5 shufpd $0,%xmm6,%xmm6 shufpd $0,%xmm7,%xmm7 movapd %xmm3,nb313nf_ixO(%rsp) movapd %xmm4,nb313nf_iyO(%rsp) movapd %xmm5,nb313nf_izO(%rsp) movapd %xmm6,nb313nf_ixH1(%rsp) movapd %xmm7,nb313nf_iyH1(%rsp) movsd %xmm2,%xmm6 movsd %xmm0,%xmm3 movsd %xmm1,%xmm4 movsd %xmm2,%xmm5 addsd 40(%rax,%rbx,8),%xmm6 ## h1z addsd 48(%rax,%rbx,8),%xmm0 ## h2x addsd 56(%rax,%rbx,8),%xmm1 ## h2y addsd 64(%rax,%rbx,8),%xmm2 ## h2z addsd 72(%rax,%rbx,8),%xmm3 ## mx addsd 80(%rax,%rbx,8),%xmm4 ## my addsd 88(%rax,%rbx,8),%xmm5 ## mz shufpd $0,%xmm6,%xmm6 shufpd $0,%xmm0,%xmm0 shufpd $0,%xmm1,%xmm1 shufpd $0,%xmm2,%xmm2 shufpd $0,%xmm3,%xmm3 shufpd $0,%xmm4,%xmm4 shufpd $0,%xmm5,%xmm5 movapd %xmm6,nb313nf_izH1(%rsp) movapd %xmm0,nb313nf_ixH2(%rsp) movapd %xmm1,nb313nf_iyH2(%rsp) movapd %xmm2,nb313nf_izH2(%rsp) movapd %xmm3,nb313nf_ixM(%rsp) movapd %xmm4,nb313nf_iyM(%rsp) movapd %xmm5,nb313nf_izM(%rsp) ## clear vctot xorpd %xmm4,%xmm4 movapd %xmm4,nb313nf_vctot(%rsp) movapd %xmm4,nb313nf_Vvdwtot(%rsp) movq nb313nf_jindex(%rsp),%rax movl (%rax,%rsi,4),%ecx ## jindex[n] movl 4(%rax,%rsi,4),%edx ## jindex[n+1] subl %ecx,%edx ## number of innerloop atoms movq nb313nf_pos(%rbp),%rsi movq nb313nf_faction(%rbp),%rdi movq nb313nf_jjnr(%rsp),%rax shll $2,%ecx addq %rcx,%rax movq %rax,nb313nf_innerjjnr(%rsp) ## pointer to jjnr[nj0] movl %edx,%ecx subl $2,%edx addl nb313nf_ninner(%rsp),%ecx movl %ecx,nb313nf_ninner(%rsp) addl $0,%edx movl %edx,nb313nf_innerk(%rsp) ## number of innerloop atoms jge _nb_kernel313nf_x86_64_sse2.nb313nf_unroll_loop jmp _nb_kernel313nf_x86_64_sse2.nb313nf_checksingle_nb_kernel313nf_x86_64_sse2.nb313nf_unroll_loop: ## twice unrolled innerloop here movq nb313nf_innerjjnr(%rsp),%rdx ## pointer to jjnr[k] movl (%rdx),%eax movl 4(%rdx),%ebx addq $8,nb313nf_innerjjnr(%rsp) ## advance pointer (unrolled 2) movq nb313nf_charge(%rbp),%rsi ## base of charge[] movlpd (%rsi,%rax,8),%xmm3 movhpd (%rsi,%rbx,8),%xmm3 movapd %xmm3,%xmm4 mulpd nb313nf_iqM(%rsp),%xmm3 mulpd nb313nf_iqH(%rsp),%xmm4 movapd %xmm3,nb313nf_qqM(%rsp) movapd %xmm4,nb313nf_qqH(%rsp) movd %eax,%mm0 ## use mmx registers as temp storage movd %ebx,%mm1 movq nb313nf_type(%rbp),%rsi movl (%rsi,%rax,4),%eax movl (%rsi,%rbx,4),%ebx movq nb313nf_vdwparam(%rbp),%rsi shll %eax shll %ebx movl nb313nf_ntia(%rsp),%edi addl %edi,%eax addl %edi,%ebx movlpd (%rsi,%rax,8),%xmm6 ## c6a movlpd (%rsi,%rbx,8),%xmm7 ## c6b movhpd 8(%rsi,%rax,8),%xmm6 ## c6a c12a movhpd 8(%rsi,%rbx,8),%xmm7 ## c6b c12b movapd %xmm6,%xmm4 unpcklpd %xmm7,%xmm4 unpckhpd %xmm7,%xmm6 movd %mm0,%eax movd %mm1,%ebx movapd %xmm4,nb313nf_c6(%rsp) movapd %xmm6,nb313nf_c12(%rsp) movq nb313nf_pos(%rbp),%rsi ## base of pos[] lea (%rax,%rax,2),%rax ## replace jnr with j3 lea (%rbx,%rbx,2),%rbx ## move two coordinates to xmm0-xmm2 movlpd (%rsi,%rax,8),%xmm0 movlpd 8(%rsi,%rax,8),%xmm1 movlpd 16(%rsi,%rax,8),%xmm2 movhpd (%rsi,%rbx,8),%xmm0 movhpd 8(%rsi,%rbx,8),%xmm1 movhpd 16(%rsi,%rbx,8),%xmm2 ## move ixO-izO to xmm4-xmm6 movapd nb313nf_ixO(%rsp),%xmm4 movapd nb313nf_iyO(%rsp),%xmm5 movapd nb313nf_izO(%rsp),%xmm6 ## calc dr subpd %xmm0,%xmm4 subpd %xmm1,%xmm5 subpd %xmm2,%xmm6 ## square it mulpd %xmm4,%xmm4 mulpd %xmm5,%xmm5 mulpd %xmm6,%xmm6 addpd %xmm5,%xmm4 addpd %xmm6,%xmm4 movapd %xmm4,%xmm7 ## rsqO in xmm7 ## move ixH1-izH1 to xmm4-xmm6 movapd nb313nf_ixH1(%rsp),%xmm4 movapd nb313nf_iyH1(%rsp),%xmm5 movapd nb313nf_izH1(%rsp),%xmm6 ## calc dr subpd %xmm0,%xmm4 subpd %xmm1,%xmm5 subpd %xmm2,%xmm6 ## square it mulpd %xmm4,%xmm4 mulpd %xmm5,%xmm5 mulpd %xmm6,%xmm6 addpd %xmm5,%xmm6 addpd %xmm4,%xmm6 ## rsqH1 in xmm6 ## move ixH2-izH2 to xmm3-xmm5 movapd nb313nf_ixH2(%rsp),%xmm3 movapd nb313nf_iyH2(%rsp),%xmm4 movapd nb313nf_izH2(%rsp),%xmm5 ## calc dr subpd %xmm0,%xmm3 subpd %xmm1,%xmm4 subpd %xmm2,%xmm5 ## square it mulpd %xmm3,%xmm3 mulpd %xmm4,%xmm4 mulpd %xmm5,%xmm5 addpd %xmm4,%xmm5 addpd %xmm3,%xmm5 ## move ixM-izM to xmm2-xmm4 movapd nb313nf_iyM(%rsp),%xmm3 movapd nb313nf_izM(%rsp),%xmm4 subpd %xmm1,%xmm3 subpd %xmm2,%xmm4 movapd nb313nf_ixM(%rsp),%xmm2 subpd %xmm0,%xmm2 ## square it mulpd %xmm2,%xmm2 mulpd %xmm3,%xmm3 mulpd %xmm4,%xmm4 addpd %xmm3,%xmm4 addpd %xmm2,%xmm4 ## rsqM in xmm4, rsqH2 in xmm5, rsqH1 in xmm6, rsqO in xmm7 ## 1/x for O - rsqO is in xmm7 cvtpd2ps %xmm7,%xmm2 movapd %xmm7,%xmm3 rcpps %xmm2,%xmm2 cvtps2pd %xmm2,%xmm2 movapd nb313nf_two(%rsp),%xmm1 movapd %xmm1,%xmm0 mulpd %xmm2,%xmm7 subpd %xmm7,%xmm1 mulpd %xmm1,%xmm2 ## iter1 mulpd %xmm2,%xmm3 subpd %xmm3,%xmm0 mulpd %xmm2,%xmm0 ## xmm0=rinvsq movapd %xmm0,nb313nf_rinvsqO(%rsp) ## rsqH1 - seed in xmm2 cvtpd2ps %xmm6,%xmm2 rsqrtps %xmm2,%xmm2 cvtps2pd %xmm2,%xmm2 movapd %xmm2,%xmm3 mulpd %xmm2,%xmm2 movapd nb313nf_three(%rsp),%xmm0 mulpd %xmm6,%xmm2 ## rsq*lu*lu subpd %xmm2,%xmm0 ## 30-rsq*lu*lu mulpd %xmm3,%xmm0 ## lu*(3-rsq*lu*lu) mulpd nb313nf_half(%rsp),%xmm0 ## iter1 ( new lu) movapd %xmm6,%xmm2 movapd %xmm0,%xmm3 mulpd %xmm0,%xmm0 ## lu*lu mulpd %xmm0,%xmm2 ## rsq*lu*lu movapd nb313nf_three(%rsp),%xmm0 subpd %xmm2,%xmm0 ## 3-rsq*lu*lu mulpd %xmm3,%xmm0 ## lu*( 3-rsq*lu*lu) mulpd nb313nf_half(%rsp),%xmm0 ## rinv movapd %xmm0,nb313nf_rinvH1(%rsp) ## rinvH1 mulpd %xmm0,%xmm6 movapd %xmm6,nb313nf_rH1(%rsp) ## rH1 ## rsqH2 - seed in xmm2 cvtpd2ps %xmm5,%xmm2 rsqrtps %xmm2,%xmm2 cvtps2pd %xmm2,%xmm2 movapd %xmm2,%xmm3 mulpd %xmm2,%xmm2 movapd nb313nf_three(%rsp),%xmm0 mulpd %xmm5,%xmm2 ## rsq*lu*lu subpd %xmm2,%xmm0 ## 30-rsq*lu*lu mulpd %xmm3,%xmm0 ## lu*(3-rsq*lu*lu) mulpd nb313nf_half(%rsp),%xmm0 ## iter1 ( new lu) movapd %xmm5,%xmm2 movapd %xmm0,%xmm3 mulpd %xmm0,%xmm0 ## lu*lu mulpd %xmm0,%xmm2 ## rsq*lu*lu movapd nb313nf_three(%rsp),%xmm0 subpd %xmm2,%xmm0 ## 3-rsq*lu*lu mulpd %xmm3,%xmm0 ## lu*( 3-rsq*lu*lu) mulpd nb313nf_half(%rsp),%xmm0 ## rinv movapd %xmm0,nb313nf_rinvH2(%rsp) ## rinv mulpd %xmm0,%xmm5 movapd %xmm5,nb313nf_rH2(%rsp) ## r ## rsqM - seed in xmm2 cvtpd2ps %xmm4,%xmm2 rsqrtps %xmm2,%xmm2 cvtps2pd %xmm2,%xmm2 movapd %xmm2,%xmm3 mulpd %xmm2,%xmm2 movapd nb313nf_three(%rsp),%xmm0 mulpd %xmm4,%xmm2 ## rsq*lu*lu subpd %xmm2,%xmm0 ## 30-rsq*lu*lu mulpd %xmm3,%xmm0 ## lu*(3-rsq*lu*lu) mulpd nb313nf_half(%rsp),%xmm0 ## iter1 ( new lu) movapd %xmm4,%xmm2 movapd %xmm0,%xmm3 mulpd %xmm0,%xmm0 ## lu*lu mulpd %xmm0,%xmm2 ## rsq*lu*lu movapd nb313nf_three(%rsp),%xmm0 subpd %xmm2,%xmm0 ## 3-rsq*lu*lu mulpd %xmm3,%xmm0 ## lu*( 3-rsq*lu*lu) mulpd nb313nf_half(%rsp),%xmm0 ## rinv movapd %xmm0,nb313nf_rinvM(%rsp) ## rinv mulpd %xmm0,%xmm4 movapd %xmm4,nb313nf_rM(%rsp) ## r ## do O interactions movapd nb313nf_rinvsqO(%rsp),%xmm0 movapd %xmm0,%xmm1 mulpd %xmm1,%xmm1 ## rinv4 mulpd %xmm0,%xmm1 ##rinvsix movapd %xmm1,%xmm2 mulpd %xmm2,%xmm2 ## rinvtwelve mulpd nb313nf_c6(%rsp),%xmm1 mulpd nb313nf_c12(%rsp),%xmm2 movapd %xmm2,%xmm3 subpd %xmm1,%xmm3 ## Vvdw=Vvdw12-Vvdw6 addpd nb313nf_Vvdwtot(%rsp),%xmm3 movapd %xmm3,nb313nf_Vvdwtot(%rsp) ## Done with O interactions - now H1! movapd nb313nf_rH1(%rsp),%xmm7 mulpd nb313nf_tsc(%rsp),%xmm7 cvttpd2pi %xmm7,%mm6 ## mm6 = lu idx cvtpi2pd %mm6,%xmm6 subpd %xmm6,%xmm7 movapd %xmm7,%xmm1 ## xmm1=eps movapd %xmm1,%xmm2 mulpd %xmm2,%xmm2 ## xmm2=eps2 movd %eax,%mm0 movd %ebx,%mm1 pslld $2,%mm6 ## idx *= 4 movq nb313nf_VFtab(%rbp),%rsi movd %mm6,%eax psrlq $32,%mm6 movd %mm6,%ebx ## indices in eax/ebx movapd (%rsi,%rax,8),%xmm4 ## Y1 F1 movapd (%rsi,%rbx,8),%xmm3 ## Y2 F2 movapd %xmm4,%xmm5 unpcklpd %xmm3,%xmm4 ## Y1 Y2 unpckhpd %xmm3,%xmm5 ## F1 F2 movapd 16(%rsi,%rax,8),%xmm6 ## G1 H1 movapd 16(%rsi,%rbx,8),%xmm3 ## G2 H2 movapd %xmm6,%xmm7 unpcklpd %xmm3,%xmm6 ## G1 G2 unpckhpd %xmm3,%xmm7 ## H1 H2 ## coulomb table ready, in xmm4-xmm7 mulpd %xmm1,%xmm6 ## xmm6=Geps mulpd %xmm2,%xmm7 ## xmm7=Heps2 addpd %xmm6,%xmm5 addpd %xmm7,%xmm5 ## xmm5=Fp movapd nb313nf_qqH(%rsp),%xmm3 mulpd %xmm1,%xmm5 ## xmm5=eps*Fp addpd %xmm4,%xmm5 ## xmm5=VV mulpd %xmm3,%xmm5 ## vcoul=qq*VV ## increment vcoul addpd nb313nf_vctot(%rsp),%xmm5 movapd %xmm5,nb313nf_vctot(%rsp) ## H2 interactions movapd nb313nf_rH2(%rsp),%xmm7 mulpd nb313nf_tsc(%rsp),%xmm7 cvttpd2pi %xmm7,%mm6 ## mm6 = lu idx cvtpi2pd %mm6,%xmm6 subpd %xmm6,%xmm7 movapd %xmm7,%xmm1 ## xmm1=eps movapd %xmm1,%xmm2 mulpd %xmm2,%xmm2 ## xmm2=eps2 pslld $2,%mm6 ## idx *= 4 movq nb313nf_VFtab(%rbp),%rsi movd %mm6,%eax psrlq $32,%mm6 movd %mm6,%ebx ## indices in eax/ebx movapd (%rsi,%rax,8),%xmm4 ## Y1 F1 movapd (%rsi,%rbx,8),%xmm3 ## Y2 F2 movapd %xmm4,%xmm5 unpcklpd %xmm3,%xmm4 ## Y1 Y2 unpckhpd %xmm3,%xmm5 ## F1 F2 movapd 16(%rsi,%rax,8),%xmm6 ## G1 H1 movapd 16(%rsi,%rbx,8),%xmm3 ## G2 H2 movapd %xmm6,%xmm7 unpcklpd %xmm3,%xmm6 ## G1 G2 unpckhpd %xmm3,%xmm7 ## H1 H2 ## coulomb table ready, in xmm4-xmm7 mulpd %xmm1,%xmm6 ## xmm6=Geps mulpd %xmm2,%xmm7 ## xmm7=Heps2 addpd %xmm6,%xmm5 addpd %xmm7,%xmm5 ## xmm5=Fp movapd nb313nf_qqH(%rsp),%xmm3 mulpd %xmm1,%xmm5 ## xmm5=eps*Fp addpd %xmm4,%xmm5 ## xmm5=VV mulpd %xmm3,%xmm5 ## vcoul=qq*VV ## increment vcoul addpd nb313nf_vctot(%rsp),%xmm5 movapd %xmm5,nb313nf_vctot(%rsp) ## M interactions movapd nb313nf_rM(%rsp),%xmm7 mulpd nb313nf_tsc(%rsp),%xmm7 cvttpd2pi %xmm7,%mm6 ## mm6 = lu idx cvtpi2pd %mm6,%xmm6 subpd %xmm6,%xmm7 movapd %xmm7,%xmm1 ## xmm1=eps movapd %xmm1,%xmm2 mulpd %xmm2,%xmm2 ## xmm2=eps2 movd %eax,%mm0 movd %ebx,%mm1 pslld $2,%mm6 ## idx *= 4 movq nb313nf_VFtab(%rbp),%rsi movd %mm6,%eax psrlq $32,%mm6 movd %mm6,%ebx ## indices in eax/ebx movapd (%rsi,%rax,8),%xmm4 ## Y1 F1 movapd (%rsi,%rbx,8),%xmm3 ## Y2 F2 movapd %xmm4,%xmm5 unpcklpd %xmm3,%xmm4 ## Y1 Y2 unpckhpd %xmm3,%xmm5 ## F1 F2 movapd 16(%rsi,%rax,8),%xmm6 ## G1 H1 movapd 16(%rsi,%rbx,8),%xmm3 ## G2 H2 ⌨️ 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?