nb_kernel334_x86_64_sse2.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,355 行 · 第 1/5 页
S
2,355 行
movapd %xmm10,%xmm5 movapd %xmm11,%xmm8 mulsd %xmm2,%xmm2 ## lu*lu mulsd %xmm5,%xmm5 ## lu*lu mulsd %xmm8,%xmm8 ## lu*lu movapd nb334_three(%rsp),%xmm1 movapd %xmm1,%xmm4 movapd %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) movapd nb334_half(%rsp),%xmm15 mulsd %xmm15,%xmm9 ## rinvH1H1 mulsd %xmm15,%xmm10 ## rinvH2H1 mulsd %xmm15,%xmm11 ## rinvMH1 movapd %xmm9,nb334_rinvH1H1(%rsp) movapd %xmm10,nb334_rinvH2H1(%rsp) movapd %xmm11,nb334_rinvMH1(%rsp) ## H1 interactions ## rsq in xmm0,xmm3,xmm6 ## rinv in xmm9, xmm10, xmm11 movapd nb334_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 shll $2,%r8d shll $2,%r10d shll $2,%r12d lea (%r8,%r8,2),%r8 lea (%r10,%r10,2),%r10 lea (%r12,%r12,2),%r12 movq nb334_VFtab(%rbp),%rsi ## calculate eps subsd %xmm2,%xmm0 subsd %xmm5,%xmm3 subsd %xmm8,%xmm6 movapd %xmm0,%xmm12 ## epsH1 movapd %xmm3,%xmm13 ## epsH2 movapd %xmm6,%xmm14 ## epsM ## 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 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 movapd nb334_qqHH(%rsp),%xmm12 movapd nb334_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 nb334_vctot(%rsp),%xmm1 addsd %xmm9,%xmm5 addsd %xmm5,%xmm1 movsd %xmm1,nb334_vctot(%rsp) movapd nb334_tsc(%rsp),%xmm10 mulsd %xmm10,%xmm3 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 nb334_rinvH1H1(%rsp),%xmm4 mulsd nb334_rinvH2H1(%rsp),%xmm8 mulsd nb334_rinvMH1(%rsp),%xmm11 ## move j H1 forces to xmm0-xmm2 movq nb334_faction(%rbp),%rdi movsd 24(%rdi,%rax,8),%xmm0 movsd 32(%rdi,%rax,8),%xmm1 movsd 40(%rdi,%rax,8),%xmm2 movapd %xmm4,%xmm3 movapd %xmm4,%xmm5 movapd %xmm8,%xmm7 movapd %xmm8,%xmm9 movapd %xmm11,%xmm10 movapd %xmm11,%xmm12 mulpd nb334_dxH1H1(%rsp),%xmm3 mulsd nb334_dyH1H1(%rsp),%xmm4 mulsd nb334_dzH1H1(%rsp),%xmm5 mulsd nb334_dxH2H1(%rsp),%xmm7 mulsd nb334_dyH2H1(%rsp),%xmm8 mulsd nb334_dzH2H1(%rsp),%xmm9 mulsd nb334_dxMH1(%rsp),%xmm10 mulsd nb334_dyMH1(%rsp),%xmm11 mulsd nb334_dzMH1(%rsp),%xmm12 addsd %xmm3,%xmm0 addsd %xmm4,%xmm1 addsd %xmm5,%xmm2 addsd nb334_fixH1(%rsp),%xmm3 addsd nb334_fiyH1(%rsp),%xmm4 addsd nb334_fizH1(%rsp),%xmm5 addsd %xmm7,%xmm0 addsd %xmm8,%xmm1 addsd %xmm9,%xmm2 addsd nb334_fixH2(%rsp),%xmm7 addsd nb334_fiyH2(%rsp),%xmm8 addsd nb334_fizH2(%rsp),%xmm9 addsd %xmm10,%xmm0 addsd %xmm11,%xmm1 addsd %xmm12,%xmm2 addsd nb334_fixM(%rsp),%xmm10 addsd nb334_fiyM(%rsp),%xmm11 addsd nb334_fizM(%rsp),%xmm12 movsd %xmm3,nb334_fixH1(%rsp) movsd %xmm4,nb334_fiyH1(%rsp) movsd %xmm5,nb334_fizH1(%rsp) movsd %xmm7,nb334_fixH2(%rsp) movsd %xmm8,nb334_fiyH2(%rsp) movsd %xmm9,nb334_fizH2(%rsp) movsd %xmm10,nb334_fixM(%rsp) movsd %xmm11,nb334_fiyM(%rsp) movsd %xmm12,nb334_fizM(%rsp) ## store back j H1 forces from xmm0-xmm2 movsd %xmm0,24(%rdi,%rax,8) movsd %xmm1,32(%rdi,%rax,8) movsd %xmm2,40(%rdi,%rax,8) ## move j H2 coordinates to local temp variables movq nb334_pos(%rbp),%rsi movsd 48(%rsi,%rax,8),%xmm0 movsd 56(%rsi,%rax,8),%xmm1 movsd 64(%rsi,%rax,8),%xmm2 ## xmm0 = H2x ## xmm1 = H2y ## xmm2 = H2z movapd %xmm0,%xmm3 movapd %xmm1,%xmm4 movapd %xmm2,%xmm5 movapd %xmm0,%xmm6 movapd %xmm1,%xmm7 movapd %xmm2,%xmm8 subsd nb334_ixH1(%rsp),%xmm0 subsd nb334_iyH1(%rsp),%xmm1 subsd nb334_izH1(%rsp),%xmm2 subsd nb334_ixH2(%rsp),%xmm3 subsd nb334_iyH2(%rsp),%xmm4 subsd nb334_izH2(%rsp),%xmm5 subsd nb334_ixM(%rsp),%xmm6 subsd nb334_iyM(%rsp),%xmm7 subsd nb334_izM(%rsp),%xmm8 movapd %xmm0,nb334_dxH1H2(%rsp) movapd %xmm1,nb334_dyH1H2(%rsp) movapd %xmm2,nb334_dzH1H2(%rsp) mulsd %xmm0,%xmm0 mulsd %xmm1,%xmm1 mulsd %xmm2,%xmm2 movapd %xmm3,nb334_dxH2H2(%rsp) movapd %xmm4,nb334_dyH2H2(%rsp) movapd %xmm5,nb334_dzH2H2(%rsp) mulsd %xmm3,%xmm3 mulsd %xmm4,%xmm4 mulsd %xmm5,%xmm5 movapd %xmm6,nb334_dxMH2(%rsp) movapd %xmm7,nb334_dyMH2(%rsp) movapd %xmm8,nb334_dzMH2(%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 jH2 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 movapd %xmm1,%xmm2 movapd %xmm4,%xmm5 movapd %xmm7,%xmm8 mulsd %xmm1,%xmm1 ## lu*lu mulsd %xmm4,%xmm4 ## lu*lu mulsd %xmm7,%xmm7 ## lu*lu movapd nb334_three(%rsp),%xmm9 movapd %xmm9,%xmm10 movapd %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) movapd nb334_half(%rsp),%xmm15 mulsd %xmm15,%xmm9 ## first iteration for rinvH1H2 mulsd %xmm15,%xmm10 ## first iteration for rinvH2H2 mulsd %xmm15,%xmm11 ## first iteration for rinvMH2 ## second iteration step movapd %xmm9,%xmm2 movapd %xmm10,%xmm5 movapd %xmm11,%xmm8 mulsd %xmm2,%xmm2 ## lu*lu mulsd %xmm5,%xmm5 ## lu*lu mulsd %xmm8,%xmm8 ## lu*lu movapd nb334_three(%rsp),%xmm1 movapd %xmm1,%xmm4 movapd %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) movapd nb334_half(%rsp),%xmm15 mulsd %xmm15,%xmm9 ## rinvH1H2 mulsd %xmm15,%xmm10 ## rinvH2H2 mulsd %xmm15,%xmm11 ## rinvMH2 movapd %xmm9,nb334_rinvH1H2(%rsp) movapd %xmm10,nb334_rinvH2H2(%rsp) movapd %xmm11,nb334_rinvMH2(%rsp) ## H2 interactions ## rsq in xmm0,xmm3,xmm6 ## rinv in xmm9, xmm10, xmm11 movapd nb334_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 shll $2,%r8d shll $2,%r10d shll $2,%r12d lea (%r8,%r8,2),%r8 lea (%r10,%r10,2),%r10 lea (%r12,%r12,2),%r12 movq nb334_VFtab(%rbp),%rsi ## calculate eps subpd %xmm2,%xmm0 subpd %xmm5,%xmm3 subpd %xmm8,%xmm6 movapd %xmm0,%xmm12 ## epsH1 movapd %xmm3,%xmm13 ## epsH2 movapd %xmm6,%xmm14 ## epsM ## 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 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 movapd nb334_qqHH(%rsp),%xmm12 movapd nb334_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 nb334_vctot(%rsp),%xmm1 addsd %xmm9,%xmm5 addsd %xmm5,%xmm1 movsd %xmm1,nb334_vctot(%rsp) movapd nb334_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 nb334_rinvH1H2(%rsp),%xmm4 mulsd nb334_rinvH2H2(%rsp),%xmm8 mulsd nb334_rinvMH2(%rsp),%xmm11 ## move j H2 forces to xmm0-xmm2 movq nb334_faction(%rbp),%rdi movsd 48(%rdi,%rax,8),%xmm0 movsd 56(%rdi,%rax,8),%xmm1 movsd 64(%rdi,%rax,8),%xmm2 movapd %xmm4,%xmm3 movapd %xmm4,%xmm5 movapd %xmm8,%xmm7 movapd %xmm8,%xmm9 movapd %xmm11,%xmm10 movapd %xmm11,%xmm12 ⌨️ 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?