nb_kernel314_x86_64_sse2.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,365 行 · 第 1/5 页
S
2,365 行
addpd %xmm6,%xmm5 addpd %xmm10,%xmm9 addpd %xmm3,%xmm1 ## F+Geps+Heps2 = Fp addpd %xmm7,%xmm5 addpd %xmm11,%xmm9 addpd %xmm3,%xmm3 ## 2*Heps2 addpd %xmm7,%xmm7 addpd %xmm11,%xmm11 addpd %xmm2,%xmm3 ## 2*Heps2+Geps addpd %xmm6,%xmm7 addpd %xmm10,%xmm11 addpd %xmm1,%xmm3 ## FF = Fp + 2*Heps2 + Geps addpd %xmm5,%xmm7 addpd %xmm9,%xmm11 mulpd %xmm12,%xmm1 ## eps*Fp mulpd %xmm13,%xmm5 mulpd %xmm14,%xmm9 movapd nb314_qqMH(%rsp),%xmm12 movapd nb314_qqMM(%rsp),%xmm13 addpd %xmm0,%xmm1 ## VV addpd %xmm4,%xmm5 addpd %xmm8,%xmm9 mulpd %xmm12,%xmm1 ## VV*qq = vcoul mulpd %xmm12,%xmm5 mulpd %xmm13,%xmm9 mulpd %xmm12,%xmm3 ## FF*qq = fij mulpd %xmm12,%xmm7 mulpd %xmm13,%xmm11 ## accumulate vctot addpd nb314_vctot(%rsp),%xmm1 addpd %xmm9,%xmm5 addpd %xmm5,%xmm1 movapd %xmm1,nb314_vctot(%rsp) movapd nb314_tsc(%rsp),%xmm10 mulpd %xmm10,%xmm3 ## fscal mulpd %xmm10,%xmm7 mulpd %xmm11,%xmm10 xorpd %xmm4,%xmm4 xorpd %xmm8,%xmm8 xorpd %xmm11,%xmm11 subpd %xmm3,%xmm4 subpd %xmm7,%xmm8 subpd %xmm10,%xmm11 mulpd nb314_rinvH1M(%rsp),%xmm4 mulpd nb314_rinvH2M(%rsp),%xmm8 mulpd nb314_rinvMM(%rsp),%xmm11 ## move j M forces to xmm0-xmm2 movq nb314_faction(%rbp),%rdi movlpd 72(%rdi,%rax,8),%xmm0 movlpd 80(%rdi,%rax,8),%xmm1 movlpd 88(%rdi,%rax,8),%xmm2 movhpd 72(%rdi,%rbx,8),%xmm0 movhpd 80(%rdi,%rbx,8),%xmm1 movhpd 88(%rdi,%rbx,8),%xmm2 movapd %xmm4,%xmm3 movapd %xmm4,%xmm5 movapd %xmm8,%xmm7 movapd %xmm8,%xmm9 movapd %xmm11,%xmm10 movapd %xmm11,%xmm12 mulpd nb314_dxH1M(%rsp),%xmm3 mulpd nb314_dyH1M(%rsp),%xmm4 mulpd nb314_dzH1M(%rsp),%xmm5 mulpd nb314_dxH2M(%rsp),%xmm7 mulpd nb314_dyH2M(%rsp),%xmm8 mulpd nb314_dzH2M(%rsp),%xmm9 mulpd nb314_dxMM(%rsp),%xmm10 mulpd nb314_dyMM(%rsp),%xmm11 mulpd nb314_dzMM(%rsp),%xmm12 addpd %xmm3,%xmm0 addpd %xmm4,%xmm1 addpd %xmm5,%xmm2 addpd nb314_fixH1(%rsp),%xmm3 addpd nb314_fiyH1(%rsp),%xmm4 addpd nb314_fizH1(%rsp),%xmm5 addpd %xmm7,%xmm0 addpd %xmm8,%xmm1 addpd %xmm9,%xmm2 addpd nb314_fixH2(%rsp),%xmm7 addpd nb314_fiyH2(%rsp),%xmm8 addpd nb314_fizH2(%rsp),%xmm9 addpd %xmm10,%xmm0 addpd %xmm11,%xmm1 addpd %xmm12,%xmm2 addpd nb314_fixM(%rsp),%xmm10 addpd nb314_fiyM(%rsp),%xmm11 addpd nb314_fizM(%rsp),%xmm12 movapd %xmm3,nb314_fixH1(%rsp) movapd %xmm4,nb314_fiyH1(%rsp) movapd %xmm5,nb314_fizH1(%rsp) movapd %xmm7,nb314_fixH2(%rsp) movapd %xmm8,nb314_fiyH2(%rsp) movapd %xmm9,nb314_fizH2(%rsp) movapd %xmm10,nb314_fixM(%rsp) movapd %xmm11,nb314_fiyM(%rsp) movapd %xmm12,nb314_fizM(%rsp) ## store back j M forces from xmm0-xmm2 movlpd %xmm0,72(%rdi,%rax,8) movlpd %xmm1,80(%rdi,%rax,8) movlpd %xmm2,88(%rdi,%rax,8) movhpd %xmm0,72(%rdi,%rbx,8) movhpd %xmm1,80(%rdi,%rbx,8) movhpd %xmm2,88(%rdi,%rbx,8) ## should we do one more iteration? subl $2,nb314_innerk(%rsp) jl _nb_kernel314_x86_64_sse2.nb314_checksingle jmp _nb_kernel314_x86_64_sse2.nb314_unroll_loop_nb_kernel314_x86_64_sse2.nb314_checksingle: movl nb314_innerk(%rsp),%edx andl $1,%edx jnz _nb_kernel314_x86_64_sse2.nb314_dosingle jmp _nb_kernel314_x86_64_sse2.nb314_updateouterdata_nb_kernel314_x86_64_sse2.nb314_dosingle: movq nb314_innerjjnr(%rsp),%rdx ## pointer to jjnr[k] movl (%rdx),%eax movq nb314_pos(%rbp),%rsi lea (%rax,%rax,2),%rax ## load j O coordinates movsd (%rsi,%rax,8),%xmm4 movsd 8(%rsi,%rax,8),%xmm5 movsd 16(%rsi,%rax,8),%xmm6 ## xmm4 = Ox ## xmm5 = Oy ## xmm6 = Oz subsd nb314_ixO(%rsp),%xmm4 subsd nb314_iyO(%rsp),%xmm5 subsd nb314_izO(%rsp),%xmm6 ## store dx/dy/dz movapd %xmm4,%xmm9 movapd %xmm5,%xmm10 movapd %xmm6,%xmm11 ## square it mulsd %xmm4,%xmm4 mulsd %xmm5,%xmm5 mulsd %xmm6,%xmm6 addsd %xmm5,%xmm4 addsd %xmm6,%xmm4 ## rsq in xmm4 cvtsd2ss %xmm4,%xmm6 rcpss %xmm6,%xmm6 cvtss2sd %xmm6,%xmm6 ## lu in low xmm6 ## 1/x lookup seed in xmm6 movapd nb314_two(%rsp),%xmm0 movapd %xmm4,%xmm5 mulsd %xmm6,%xmm4 ## lu*rsq subsd %xmm4,%xmm0 ## 2-lu*rsq mulsd %xmm0,%xmm6 ## (new lu) movapd nb314_two(%rsp),%xmm0 mulsd %xmm6,%xmm5 ## lu*rsq subsd %xmm5,%xmm0 ## 2-lu*rsq mulsd %xmm6,%xmm0 ## xmm0=rinvsq movapd %xmm0,%xmm1 mulsd %xmm0,%xmm1 mulsd %xmm0,%xmm1 ## xmm1=rinvsix movapd %xmm1,%xmm2 mulsd %xmm2,%xmm2 ## xmm2=rinvtwelve mulsd nb314_c6(%rsp),%xmm1 ## mult by c6 mulsd nb314_c12(%rsp),%xmm2 ## mult by c12 movapd %xmm2,%xmm5 subsd %xmm1,%xmm5 ## Vvdw=Vvdw12-Vvdw6 mulsd nb314_six(%rsp),%xmm1 mulsd nb314_twelve(%rsp),%xmm2 subsd %xmm1,%xmm2 mulsd %xmm2,%xmm0 ## xmm0=total fscal ## increment potential addsd nb314_Vvdwtot(%rsp),%xmm5 movsd %xmm5,nb314_Vvdwtot(%rsp) movq nb314_faction(%rbp),%rdi mulsd %xmm0,%xmm9 mulsd %xmm0,%xmm10 mulsd %xmm0,%xmm11 movapd nb314_fixO(%rsp),%xmm0 movapd nb314_fiyO(%rsp),%xmm1 movapd nb314_fizO(%rsp),%xmm2 ## accumulate i forces addsd %xmm9,%xmm0 addsd %xmm10,%xmm1 addsd %xmm11,%xmm2 movsd %xmm0,nb314_fixO(%rsp) movsd %xmm1,nb314_fiyO(%rsp) movsd %xmm2,nb314_fizO(%rsp) ## the fj's - start by accumulating forces from memory movq nb314_faction(%rbp),%rdi addsd (%rdi,%rax,8),%xmm9 addsd 8(%rdi,%rax,8),%xmm10 addsd 16(%rdi,%rax,8),%xmm11 movsd %xmm9,(%rdi,%rax,8) movsd %xmm10,8(%rdi,%rax,8) movsd %xmm11,16(%rdi,%rax,8) ## done with OO interaction ## move j H1 coordinates to local temp variables movq nb314_pos(%rbp),%rsi movsd 24(%rsi,%rax,8),%xmm0 movsd 32(%rsi,%rax,8),%xmm1 movsd 40(%rsi,%rax,8),%xmm2 ## xmm0 = H1x ## xmm1 = H1y ## xmm2 = H1z movsd %xmm0,%xmm3 movsd %xmm1,%xmm4 movsd %xmm2,%xmm5 movsd %xmm0,%xmm6 movsd %xmm1,%xmm7 movsd %xmm2,%xmm8 subsd nb314_ixH1(%rsp),%xmm0 subsd nb314_iyH1(%rsp),%xmm1 subsd nb314_izH1(%rsp),%xmm2 subsd nb314_ixH2(%rsp),%xmm3 subsd nb314_iyH2(%rsp),%xmm4 subsd nb314_izH2(%rsp),%xmm5 subsd nb314_ixM(%rsp),%xmm6 subsd nb314_iyM(%rsp),%xmm7 subsd nb314_izM(%rsp),%xmm8 movsd %xmm0,nb314_dxH1H1(%rsp) movsd %xmm1,nb314_dyH1H1(%rsp) movsd %xmm2,nb314_dzH1H1(%rsp) mulsd %xmm0,%xmm0 mulsd %xmm1,%xmm1 mulsd %xmm2,%xmm2 movsd %xmm3,nb314_dxH2H1(%rsp) movsd %xmm4,nb314_dyH2H1(%rsp) movsd %xmm5,nb314_dzH2H1(%rsp) mulsd %xmm3,%xmm3 mulsd %xmm4,%xmm4 mulsd %xmm5,%xmm5 movsd %xmm6,nb314_dxMH1(%rsp) movsd %xmm7,nb314_dyMH1(%rsp) movsd %xmm8,nb314_dzMH1(%rsp) mulsd %xmm6,%xmm6 mulsd %xmm7,%xmm7 mulsd %xmm8,%xmm8 addsd %xmm1,%xmm0 addsd %xmm2,%xmm0 addsd %xmm4,%xmm3 addsd %xmm5,%xmm3 addsd %xmm7,%xmm6 addsd %xmm8,%xmm6 ## start doing invsqrt for jH1 atoms cvtsd2ss %xmm0,%xmm1 cvtsd2ss %xmm3,%xmm4 cvtsd2ss %xmm6,%xmm7 rsqrtss %xmm1,%xmm1 rsqrtss %xmm4,%xmm4 rsqrtss %xmm7,%xmm7 cvtss2sd %xmm1,%xmm1 cvtss2sd %xmm4,%xmm4 cvtss2sd %xmm7,%xmm7 movsd %xmm1,%xmm2 movsd %xmm4,%xmm5 movsd %xmm7,%xmm8 mulsd %xmm1,%xmm1 ## lu*lu mulsd %xmm4,%xmm4 ## lu*lu mulsd %xmm7,%xmm7 ## lu*lu movsd nb314_three(%rsp),%xmm9 movsd %xmm9,%xmm10 movsd %xmm9,%xmm11 mulsd %xmm0,%xmm1 ## rsq*lu*lu mulsd %xmm3,%xmm4 ## rsq*lu*lu mulsd %xmm6,%xmm7 ## rsq*lu*lu subsd %xmm1,%xmm9 subsd %xmm4,%xmm10 subsd %xmm7,%xmm11 ## 3-rsq*lu*lu mulsd %xmm2,%xmm9 mulsd %xmm5,%xmm10 mulsd %xmm8,%xmm11 ## lu*(3-rsq*lu*lu) movsd nb314_half(%rsp),%xmm15 mulsd %xmm15,%xmm9 ## first iteration for rinvH1H1 mulsd %xmm15,%xmm10 ## first iteration for rinvH2H1 mulsd %xmm15,%xmm11 ## first iteration for rinvMH1 ## second iteration step movsd %xmm9,%xmm2 movsd %xmm10,%xmm5 movsd %xmm11,%xmm8 mulsd %xmm2,%xmm2 ## lu*lu mulsd %xmm5,%xmm5 ## lu*lu mulsd %xmm8,%xmm8 ## lu*lu movsd nb314_three(%rsp),%xmm1 movsd %xmm1,%xmm4 movsd %xmm1,%xmm7 mulsd %xmm0,%xmm2 ## rsq*lu*lu mulsd %xmm3,%xmm5 ## rsq*lu*lu mulsd %xmm6,%xmm8 ## rsq*lu*lu subsd %xmm2,%xmm1 subsd %xmm5,%xmm4 subsd %xmm8,%xmm7 ## 3-rsq*lu*lu mulsd %xmm1,%xmm9 mulsd %xmm4,%xmm10 mulsd %xmm7,%xmm11 ## lu*(3-rsq*lu*lu) movsd nb314_half(%rsp),%xmm15 mulsd %xmm15,%xmm9 ## rinvH1H1 mulsd %xmm15,%xmm10 ## rinvH2H1 mulsd %xmm15,%xmm11 ## rinvMH1 movsd %xmm9,nb314_rinvH1H1(%rsp) movsd %xmm10,nb314_rinvH2H1(%rsp) movsd %xmm11,nb314_rinvMH1(%rsp) ## H1 interactions ## rsq in xmm0,xmm3,xmm6 ## rinv in xmm9, xmm10, xmm11 movsd nb314_tsc(%rsp),%xmm1 mulsd %xmm9,%xmm0 ## r mulsd %xmm10,%xmm3 mulsd %xmm11,%xmm6 mulsd %xmm1,%xmm0 ## rtab mulsd %xmm1,%xmm3 mulsd %xmm1,%xmm6 ## truncate and convert to integers cvttsd2si %xmm0,%r8d cvttsd2si %xmm3,%r10d cvttsd2si %xmm6,%r12d ## convert back to float cvtsi2sd %r8d,%xmm2 cvtsi2sd %r10d,%xmm5 cvtsi2sd %r12d,%xmm8 ## multiply by 4 shlq $2,%r8 shll $2,%r10d shll $2,%r12d movq nb314_VFtab(%rbp),%rsi ## calculate eps subsd %xmm2,%xmm0 subsd %xmm5,%xmm3 subsd %xmm8,%xmm6 movsd %xmm0,nb314_epsH1(%rsp) movsd %xmm3,nb314_epsH2(%rsp) movsd %xmm6,nb314_epsM(%rsp) ## Load LOTS of table data movsd (%rsi,%r8,8),%xmm0 movsd 8(%rsi,%r8,8),%xmm1 movsd 16(%rsi,%r8,8),%xmm2 movsd 24(%rsi,%r8,8),%xmm3 movsd (%rsi,%r10,8),%xmm4 movsd 8(%rsi,%r10,8),%xmm5 movsd 16(%rsi,%r10,8),%xmm6 movsd 24(%rsi,%r10,8),%xmm7 movsd (%rsi,%r12,8),%xmm8 movsd 8(%rsi,%r12,8),%xmm9 movsd 16(%rsi,%r12,8),%xmm10 movsd 24(%rsi,%r12,8),%xmm11 ## table data ready in xmm0-xmm3 , xmm4-xmm7 , and xmm8-xmm11 movsd nb314_epsH1(%rsp),%xmm12 movsd nb314_epsH2(%rsp),%xmm13 movsd nb314_epsM(%rsp),%xmm14 mulsd %xmm12,%xmm3 ## Heps mulsd %xmm13,%xmm7 mulsd %xmm14,%xmm11 mulsd %xmm12,%xmm2 ## Geps mulsd %xmm13,%xmm6 mulsd %xmm14,%xmm10 mulsd %xmm12,%xmm3 ## Heps2 mulsd %xmm13,%xmm7 mulsd %xmm14,%xmm11 addsd %xmm2,%xmm1 ## F+Geps addsd %xmm6,%xmm5 addsd %xmm10,%xmm9 addsd %xmm3,%xmm1 ## F+Geps+Heps2 = Fp addsd %xmm7,%xmm5 addsd %xmm11,%xmm9 addsd %xmm3,%xmm3 ## 2*Heps2 addsd %xmm7,%xmm7 addsd %xmm11,%xmm11 addsd %xmm2,%xmm3 ## 2*Heps2+Geps addsd %xmm6,%xmm7 addsd %xmm10,%xmm11 addsd %xmm1,%xmm3 ## FF = Fp + 2*Heps2 + Geps addsd %xmm5,%xmm7 addsd %xmm9,%xmm11 mulsd %xmm12,%xmm1 ## eps*Fp mulsd %xmm13,%xmm5 mulsd %xmm14,%xmm9 movsd nb314_qqHH(%rsp),%xmm12 movsd nb314_qqMH(%rsp),%xmm13 addsd %xmm0,%xmm1 ## VV addsd %xmm4,%xmm5 addsd %xmm8,%xmm9 mulsd %xmm12,%xmm1 ## VV*qq = vcoul mulsd %xmm12,%xmm5 mulsd %xmm13,%xmm9 mulsd %xmm12,%xmm3 ## FF*qq = fij mulsd %xmm12,%xmm7 mulsd %xmm13,%xmm11 ## accumulate vctot addsd nb314_vctot(%rsp),%xmm1 addsd %xmm9,%xmm5 addsd %xmm5,%xmm1 movsd %xmm1,nb314_vctot(%rsp) movsd nb314_tsc(%rsp),%xmm10 mulsd %xmm10,%xmm3 ## fscal mulsd %xmm10,%xmm7 mulsd %xmm11,%xmm10 xorpd %xmm4,%xmm4 xorpd %xmm8,%xmm8 xorpd %xmm11,%xmm11 subsd %xmm3,%xmm4 subsd %xmm7,%xmm8 subsd %xmm10,%xmm11 mulsd nb314_rinvH1H1(%rsp),%xmm4 mulsd nb314_rinvH2H1(%rsp),%xmm8 mulsd nb314_rinvMH1(%rsp),%xmm11 ## move j H1 forces to xmm0-xmm2 movq nb314_faction(%rbp),%rdi movsd 24(%rdi,%rax,8),%xmm0 movsd 32(%rdi,%rax,8),%xmm1 movsd 40(%rdi,%rax,8),%xmm2 movapd %xmm4,%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?