nb_kernel303_x86_64_sse2.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,073 行 · 第 1/5 页
S
2,073 行
mulsd nb303_tsc(%rsp),%xmm7 cvttsd2si %xmm7,%r8d ## mm6 = lu idx cvtsi2sd %r8d,%xmm6 subsd %xmm6,%xmm7 movapd %xmm7,%xmm1 ## xmm1=eps movapd %xmm1,%xmm2 mulsd %xmm2,%xmm2 ## xmm2=eps2 shll $2,%r8d ## idx *= 4 movq nb303_VFtab(%rbp),%rsi movapd (%rsi,%r8,8),%xmm4 ## Y1 F1 xorpd %xmm3,%xmm3 movapd %xmm4,%xmm5 unpcklpd %xmm3,%xmm4 ## Y1 unpckhpd %xmm3,%xmm5 ## F1 movapd 16(%rsi,%r8,8),%xmm6 ## G1 H1 xorpd %xmm3,%xmm3 movapd %xmm6,%xmm7 unpcklpd %xmm3,%xmm6 ## G1 unpckhpd %xmm3,%xmm7 ## H1 ## coulomb table ready, in xmm4-xmm7 mulsd %xmm1,%xmm6 ## xmm6=Geps mulsd %xmm2,%xmm7 ## xmm7=Heps2 addsd %xmm6,%xmm5 addsd %xmm7,%xmm5 ## xmm5=Fp mulsd nb303_two(%rsp),%xmm7 ## two*Heps2 movapd nb303_qqM(%rsp),%xmm3 addsd %xmm6,%xmm7 addsd %xmm5,%xmm7 ## xmm7=FF mulsd %xmm1,%xmm5 ## xmm5=eps*Fp addsd %xmm4,%xmm5 ## xmm5=VV mulsd %xmm3,%xmm5 ## vcoul=qq*VV mulsd %xmm7,%xmm3 ## fijC=FF*qq ## at this point mm5 contains vcoul and xmm3 fijC ## increment vcoul - then we can get rid of mm5 addsd nb303_vctot(%rsp),%xmm5 movlpd %xmm5,nb303_vctot(%rsp) xorpd %xmm4,%xmm4 mulsd nb303_tsc(%rsp),%xmm3 mulsd nb303_rinvM(%rsp),%xmm3 subsd %xmm3,%xmm4 movapd nb303_dxM(%rsp),%xmm0 movapd nb303_dyM(%rsp),%xmm1 movapd nb303_dzM(%rsp),%xmm2 mulsd %xmm4,%xmm0 mulsd %xmm4,%xmm1 mulsd %xmm4,%xmm2 ## tx in xmm0-xmm2 ## update M forces movapd nb303_fixM(%rsp),%xmm3 movapd nb303_fiyM(%rsp),%xmm4 movapd nb303_fizM(%rsp),%xmm7 addsd %xmm0,%xmm3 addsd %xmm1,%xmm4 addsd %xmm2,%xmm7 movlpd %xmm3,nb303_fixM(%rsp) movlpd %xmm4,nb303_fiyM(%rsp) movlpd %xmm7,nb303_fizM(%rsp) ## update j forces with water M movlpd %xmm0,nb303_fjx(%rsp) movlpd %xmm1,nb303_fjy(%rsp) movlpd %xmm2,nb303_fjz(%rsp) ## Done with M interactions - now H1! movapd nb303_rH1(%rsp),%xmm7 mulsd nb303_tsc(%rsp),%xmm7 cvttsd2si %xmm7,%r8d ## mm6 = lu idx cvtsi2sd %r8d,%xmm6 subsd %xmm6,%xmm7 movapd %xmm7,%xmm1 ## xmm1=eps movapd %xmm1,%xmm2 mulsd %xmm2,%xmm2 ## xmm2=eps2 shll $2,%r8d ## idx *= 4 movq nb303_VFtab(%rbp),%rsi movapd (%rsi,%r8,8),%xmm4 ## Y1 F1 xorpd %xmm3,%xmm3 movapd %xmm4,%xmm5 unpcklpd %xmm3,%xmm4 ## Y1 unpckhpd %xmm3,%xmm5 ## F1 movapd 16(%rsi,%r8,8),%xmm6 ## G1 H1 xorpd %xmm3,%xmm3 movapd %xmm6,%xmm7 unpcklpd %xmm3,%xmm6 ## G1 unpckhpd %xmm3,%xmm7 ## H1 ## coulomb table ready, in xmm4-xmm7 mulsd %xmm1,%xmm6 ## xmm6=Geps mulsd %xmm2,%xmm7 ## xmm7=Heps2 addsd %xmm6,%xmm5 addsd %xmm7,%xmm5 ## xmm5=Fp mulsd nb303_two(%rsp),%xmm7 ## two*Heps2 movapd nb303_qqH(%rsp),%xmm3 addsd %xmm6,%xmm7 addsd %xmm5,%xmm7 ## xmm7=FF mulsd %xmm1,%xmm5 ## xmm5=eps*Fp addsd %xmm4,%xmm5 ## xmm5=VV mulsd %xmm3,%xmm5 ## vcoul=qq*VV mulsd %xmm7,%xmm3 ## fijC=FF*qq ## at this point mm5 contains vcoul and xmm3 fijC ## increment vcoul xorpd %xmm4,%xmm4 addsd nb303_vctot(%rsp),%xmm5 mulsd nb303_rinvH1(%rsp),%xmm3 movlpd %xmm5,nb303_vctot(%rsp) mulsd nb303_tsc(%rsp),%xmm3 subsd %xmm3,%xmm4 movapd nb303_dxH1(%rsp),%xmm0 movapd nb303_dyH1(%rsp),%xmm1 movapd nb303_dzH1(%rsp),%xmm2 mulsd %xmm4,%xmm0 mulsd %xmm4,%xmm1 mulsd %xmm4,%xmm2 ## update H1 forces movapd nb303_fixH1(%rsp),%xmm3 movapd nb303_fiyH1(%rsp),%xmm4 movapd nb303_fizH1(%rsp),%xmm7 addsd %xmm0,%xmm3 addsd %xmm1,%xmm4 addsd %xmm2,%xmm7 movlpd %xmm3,nb303_fixH1(%rsp) movlpd %xmm4,nb303_fiyH1(%rsp) movlpd %xmm7,nb303_fizH1(%rsp) ## update j forces with water H1 addsd nb303_fjx(%rsp),%xmm0 addsd nb303_fjy(%rsp),%xmm1 addsd nb303_fjz(%rsp),%xmm2 movlpd %xmm0,nb303_fjx(%rsp) movlpd %xmm1,nb303_fjy(%rsp) movlpd %xmm2,nb303_fjz(%rsp) ## Done with H1, finally we do H2 interactions movapd nb303_rH2(%rsp),%xmm7 mulsd nb303_tsc(%rsp),%xmm7 cvttsd2si %xmm7,%r8d ## mm6 = lu idx cvtsi2sd %r8d,%xmm6 subsd %xmm6,%xmm7 movapd %xmm7,%xmm1 ## xmm1=eps movapd %xmm1,%xmm2 mulsd %xmm2,%xmm2 ## xmm2=eps2 shll $2,%r8d ## idx *= 4 movq nb303_VFtab(%rbp),%rsi movapd (%rsi,%r8,8),%xmm4 ## Y1 F1 xorpd %xmm3,%xmm3 movapd %xmm4,%xmm5 unpcklpd %xmm3,%xmm4 ## Y1 unpckhpd %xmm3,%xmm5 ## F1 movapd 16(%rsi,%r8,8),%xmm6 ## G1 H1 xorpd %xmm3,%xmm3 movapd %xmm6,%xmm7 unpcklpd %xmm3,%xmm6 ## G1 unpckhpd %xmm3,%xmm7 ## H1 ## coulomb table ready, in xmm4-xmm7 mulsd %xmm1,%xmm6 ## xmm6=Geps mulsd %xmm2,%xmm7 ## xmm7=Heps2 addsd %xmm6,%xmm5 addsd %xmm7,%xmm5 ## xmm5=Fp mulsd nb303_two(%rsp),%xmm7 ## two*Heps2 movapd nb303_qqH(%rsp),%xmm3 addsd %xmm6,%xmm7 addsd %xmm5,%xmm7 ## xmm7=FF mulsd %xmm1,%xmm5 ## xmm5=eps*Fp addsd %xmm4,%xmm5 ## xmm5=VV mulsd %xmm3,%xmm5 ## vcoul=qq*VV mulsd %xmm7,%xmm3 ## fijC=FF*qq ## at this point mm5 contains vcoul and xmm3 fijC ## increment vcoul xorpd %xmm4,%xmm4 addsd nb303_vctot(%rsp),%xmm5 mulsd nb303_rinvH2(%rsp),%xmm3 movlpd %xmm5,nb303_vctot(%rsp) mulsd nb303_tsc(%rsp),%xmm3 subsd %xmm3,%xmm4 movapd nb303_dxH2(%rsp),%xmm0 movapd nb303_dyH2(%rsp),%xmm1 movapd nb303_dzH2(%rsp),%xmm2 mulsd %xmm4,%xmm0 mulsd %xmm4,%xmm1 mulsd %xmm4,%xmm2 ## update H2 forces movapd nb303_fixH2(%rsp),%xmm3 movapd nb303_fiyH2(%rsp),%xmm4 movapd nb303_fizH2(%rsp),%xmm7 addsd %xmm0,%xmm3 addsd %xmm1,%xmm4 addsd %xmm2,%xmm7 movlpd %xmm3,nb303_fixH2(%rsp) movlpd %xmm4,nb303_fiyH2(%rsp) movlpd %xmm7,nb303_fizH2(%rsp) movq nb303_faction(%rbp),%rdi ## update j forces ## update j forces with water H1 addsd nb303_fjx(%rsp),%xmm0 addsd nb303_fjy(%rsp),%xmm1 addsd nb303_fjz(%rsp),%xmm2 ## the fj's - start by accumulating forces from memory movlpd (%rdi,%rax,8),%xmm3 movlpd 8(%rdi,%rax,8),%xmm4 movlpd 16(%rdi,%rax,8),%xmm5 addsd %xmm0,%xmm3 addsd %xmm1,%xmm4 addsd %xmm2,%xmm5 movlpd %xmm3,(%rdi,%rax,8) movlpd %xmm4,8(%rdi,%rax,8) movlpd %xmm5,16(%rdi,%rax,8)_nb_kernel303_x86_64_sse2.nb303_updateouterdata: movl nb303_ii3(%rsp),%ecx movq nb303_faction(%rbp),%rdi movq nb303_fshift(%rbp),%rsi movl nb303_is3(%rsp),%edx ## accumulate H1i forces in xmm0, xmm1, xmm2 movapd nb303_fixH1(%rsp),%xmm0 movapd nb303_fiyH1(%rsp),%xmm1 movapd nb303_fizH1(%rsp),%xmm2 movhlps %xmm0,%xmm3 movhlps %xmm1,%xmm4 movhlps %xmm2,%xmm5 addsd %xmm3,%xmm0 addsd %xmm4,%xmm1 addsd %xmm5,%xmm2 ## sum is in low xmm0-xmm2 ## increment i force movsd 24(%rdi,%rcx,8),%xmm3 movsd 32(%rdi,%rcx,8),%xmm4 movsd 40(%rdi,%rcx,8),%xmm5 subsd %xmm0,%xmm3 subsd %xmm1,%xmm4 subsd %xmm2,%xmm5 movsd %xmm3,24(%rdi,%rcx,8) movsd %xmm4,32(%rdi,%rcx,8) movsd %xmm5,40(%rdi,%rcx,8) ## accumulate force in xmm6/xmm7 for fshift movapd %xmm0,%xmm6 movsd %xmm2,%xmm7 unpcklpd %xmm1,%xmm6 ## accumulate H2i forces in xmm0, xmm1, xmm2 movapd nb303_fixH2(%rsp),%xmm0 movapd nb303_fiyH2(%rsp),%xmm1 movapd nb303_fizH2(%rsp),%xmm2 movhlps %xmm0,%xmm3 movhlps %xmm1,%xmm4 movhlps %xmm2,%xmm5 addsd %xmm3,%xmm0 addsd %xmm4,%xmm1 addsd %xmm5,%xmm2 ## sum is in low xmm0-xmm2 ## increment i force movsd 48(%rdi,%rcx,8),%xmm3 movsd 56(%rdi,%rcx,8),%xmm4 movsd 64(%rdi,%rcx,8),%xmm5 subsd %xmm0,%xmm3 subsd %xmm1,%xmm4 subsd %xmm2,%xmm5 movsd %xmm3,48(%rdi,%rcx,8) movsd %xmm4,56(%rdi,%rcx,8) movsd %xmm5,64(%rdi,%rcx,8) ## accumulate force in xmm6/xmm7 for fshift addsd %xmm2,%xmm7 unpcklpd %xmm1,%xmm0 addpd %xmm0,%xmm6 ## accumulate H2i forces in xmm0, xmm1, xmm2 movapd nb303_fixM(%rsp),%xmm0 movapd nb303_fiyM(%rsp),%xmm1 movapd nb303_fizM(%rsp),%xmm2 movhlps %xmm0,%xmm3 movhlps %xmm1,%xmm4 movhlps %xmm2,%xmm5 addsd %xmm3,%xmm0 addsd %xmm4,%xmm1 addsd %xmm5,%xmm2 ## sum is in low xmm0-xmm2 movapd %xmm0,%xmm3 movapd %xmm1,%xmm4 movapd %xmm2,%xmm5 ## increment i force movsd 72(%rdi,%rcx,8),%xmm3 movsd 80(%rdi,%rcx,8),%xmm4 movsd 88(%rdi,%rcx,8),%xmm5 subsd %xmm0,%xmm3 subsd %xmm1,%xmm4 subsd %xmm2,%xmm5 movsd %xmm3,72(%rdi,%rcx,8) movsd %xmm4,80(%rdi,%rcx,8) movsd %xmm5,88(%rdi,%rcx,8) ## accumulate force in xmm6/xmm7 for fshift addsd %xmm2,%xmm7 unpcklpd %xmm1,%xmm0 addpd %xmm0,%xmm6 ## increment fshift force movlpd (%rsi,%rdx,8),%xmm3 movhpd 8(%rsi,%rdx,8),%xmm3 movsd 16(%rsi,%rdx,8),%xmm4 subpd %xmm6,%xmm3 subsd %xmm7,%xmm4 movlpd %xmm3,(%rsi,%rdx,8) movhpd %xmm3,8(%rsi,%rdx,8) movsd %xmm4,16(%rsi,%rdx,8) ## get n from stack movl nb303_n(%rsp),%esi ## get group index for i particle movq nb303_gid(%rbp),%rdx ## base of gid[] movl (%rdx,%rsi,4),%edx ## ggid=gid[n] ## accumulate total potential energy and update it movapd nb303_vctot(%rsp),%xmm7 ## accumulate movhlps %xmm7,%xmm6 addsd %xmm6,%xmm7 ## low xmm7 has the sum now ## add earlier value from mem movq nb303_Vc(%rbp),%rax addsd (%rax,%rdx,8),%xmm7 ## move back to mem movsd %xmm7,(%rax,%rdx,8) ## finish if last movl nb303_nn1(%rsp),%ecx ## esi already loaded with n incl %esi subl %esi,%ecx jz _nb_kernel303_x86_64_sse2.nb303_outerend ## not last, iterate outer loop once more! movl %esi,nb303_n(%rsp) jmp _nb_kernel303_x86_64_sse2.nb303_outer_nb_kernel303_x86_64_sse2.nb303_outerend: ## check if more outer neighborlists remain movl nb303_nri(%rsp),%ecx ## esi already loaded with n above subl %esi,%ecx jz _nb_kernel303_x86_64_sse2.nb303_end ## non-zero, do one more workunit jmp _nb_kernel303_x86_64_sse2.nb303_threadloop_nb_kernel303_x86_64_sse2.nb303_end: movl nb303_nouter(%rsp),%eax movl nb303_ninner(%rsp),%ebx movq nb303_outeriter(%rbp),%rcx movq nb303_inneriter(%rbp),%rdx movl %eax,(%rcx) movl %ebx,(%rdx) addq $824,%rsp emms pop %r15 pop %r14 pop %r13 pop %r12 pop %rbx pop %rbp ret.globl nb_kernel303nf_x86_64_sse2.globl _nb_kernel303nf_x86_64_sse2nb_kernel303nf_x86_64_sse2: _nb_kernel303nf_x86_64_sse2: ## Room for return address and rbp (16 bytes).set nb303nf_fshift, 16.set nb303nf_gid, 24.set nb303nf_pos, 32.set nb303nf_faction, 40.set nb303nf_charge, 48.set nb303nf_p_facel, 56.set nb303nf_argkrf, 64.set nb303nf_argcrf, 72.set nb303nf_Vc, 80.set nb303nf_type, 88.set nb303nf_p_ntype, 96.set nb303nf_vdwparam, 104.set nb303nf_Vvdw, 112.set nb303nf_p_tabscale, 120.set nb303nf_VFtab, 128.set nb303nf_invsqrta, 136.set nb303nf_dvda, 144.set nb303nf_p_gbtabscale, 152.set nb303nf_GBtab, 160.set nb303nf_p_nthreads, 168.set nb303nf_count, 176.set nb303nf_mtx, 184.set nb303nf_outeriter, 192.set nb303nf_inneriter, 200.set nb303nf_work, 208 ## stack offsets for local variables ⌨️ 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?