nb_kernel334_x86_64_sse.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,248 行 · 第 1/5 页
S
2,248 行
## store dx/dy/dz movaps %xmm0,%xmm13 movaps %xmm1,%xmm14 movaps %xmm2,%xmm15 ## square it mulps %xmm0,%xmm0 mulps %xmm1,%xmm1 mulps %xmm2,%xmm2 addps %xmm0,%xmm1 addps %xmm2,%xmm1 ## rsq in xmm1 ## calculate rinv=1/sqrt(rsq) rsqrtps %xmm1,%xmm5 movaps %xmm5,%xmm2 mulps %xmm5,%xmm5 movaps nb334_three(%rsp),%xmm4 mulps %xmm1,%xmm5 ## rsq*lu*lu subps %xmm5,%xmm4 ## 30-rsq*lu*lu mulps %xmm2,%xmm4 mulps nb334_half(%rsp),%xmm4 movaps %xmm4,%xmm2 mulps %xmm4,%xmm1 ## xmm2=rinv ## xmm1=r mulps nb334_tsc(%rsp),%xmm1 ## rtab ## truncate and convert to integers cvttps2dq %xmm1,%xmm5 ## convert back to float cvtdq2ps %xmm5,%xmm4 ## multiply by 4 pslld $2,%xmm5 ## multiply by three (copy, mult. by two, add back) movaps %xmm5,%xmm6 pslld $1,%xmm5 paddd %xmm6,%xmm5 ## calculate eps subps %xmm4,%xmm1 ## move to integer registers movhlps %xmm5,%xmm6 movd %xmm5,%r8d movd %xmm6,%r10d pshufd $1,%xmm5,%xmm5 pshufd $1,%xmm6,%xmm6 movd %xmm5,%r9d movd %xmm6,%r11d ## table indices in r8-r11 ## xmm1=eps ## xmm2=rinv movq nb334_VFtab(%rbp),%rsi ## load LJ dispersion and repulsion in parallel ## NB: We are using a combined (LJ+coul) table, ## so the LJ table data is offset 4*4 = 16 bytes movlps 16(%rsi,%r8,4),%xmm5 movlps 32(%rsi,%r8,4),%xmm9 movlps 16(%rsi,%r10,4),%xmm7 movlps 32(%rsi,%r10,4),%xmm11 movhps 16(%rsi,%r9,4),%xmm5 movhps 32(%rsi,%r9,4),%xmm9 movhps 16(%rsi,%r11,4),%xmm7 movhps 32(%rsi,%r11,4),%xmm11 movaps %xmm5,%xmm4 movaps %xmm9,%xmm8 shufps $136,%xmm7,%xmm4 ## 10001000 shufps $136,%xmm11,%xmm8 ## 10001000 shufps $221,%xmm7,%xmm5 ## 11011101 shufps $221,%xmm11,%xmm9 ## 11011101 movlps 24(%rsi,%r8,4),%xmm7 movlps 40(%rsi,%r8,4),%xmm11 movlps 24(%rsi,%r10,4),%xmm0 movlps 40(%rsi,%r10,4),%xmm3 movhps 24(%rsi,%r9,4),%xmm7 movhps 40(%rsi,%r9,4),%xmm11 movhps 24(%rsi,%r11,4),%xmm0 movhps 40(%rsi,%r11,4),%xmm3 movaps %xmm7,%xmm6 movaps %xmm11,%xmm10 shufps $136,%xmm0,%xmm6 ## 10001000 shufps $136,%xmm3,%xmm10 ## 10001000 shufps $221,%xmm0,%xmm7 ## 11011101 shufps $221,%xmm3,%xmm11 ## 11011101 ## dispersion table in xmm4-xmm7, repulsion table in xmm8-xmm11 mulps %xmm1,%xmm7 ## Heps mulps %xmm1,%xmm11 mulps %xmm1,%xmm6 ## Geps mulps %xmm1,%xmm10 mulps %xmm1,%xmm7 ## Heps2 mulps %xmm1,%xmm11 addps %xmm6,%xmm5 ## F+Geps addps %xmm10,%xmm9 addps %xmm7,%xmm5 ## F+Geps+Heps2 = Fp addps %xmm11,%xmm9 addps %xmm7,%xmm7 ## 2*Heps2 addps %xmm11,%xmm11 addps %xmm6,%xmm7 ## 2*Heps2+Geps addps %xmm10,%xmm11 addps %xmm5,%xmm7 ## FF = Fp + 2*Heps2 + Geps addps %xmm9,%xmm11 mulps %xmm1,%xmm5 ## eps*Fp mulps %xmm1,%xmm9 addps %xmm4,%xmm5 ## VV addps %xmm8,%xmm9 mulps nb334_c6(%rsp),%xmm5 ## VV*c6 = vnb6 mulps nb334_c12(%rsp),%xmm9 ## VV*c12 = vnb12 addps %xmm9,%xmm5 addps nb334_Vvdwtot(%rsp),%xmm5 movaps %xmm5,nb334_Vvdwtot(%rsp) mulps nb334_c6(%rsp),%xmm7 ## FF*c6 = fnb6 mulps nb334_c12(%rsp),%xmm11 ## FF*c12 = fnb12 addps %xmm11,%xmm7 mulps nb334_tsc(%rsp),%xmm7 mulps %xmm2,%xmm7 xorps %xmm9,%xmm9 subps %xmm7,%xmm9 ## fx/fy/fz mulps %xmm9,%xmm13 mulps %xmm9,%xmm14 mulps %xmm9,%xmm15 ## increment i force movaps nb334_fixO(%rsp),%xmm0 movaps nb334_fiyO(%rsp),%xmm1 movaps nb334_fizO(%rsp),%xmm2 addps %xmm13,%xmm0 addps %xmm14,%xmm1 addps %xmm15,%xmm2 movaps %xmm0,nb334_fixO(%rsp) movaps %xmm1,nb334_fiyO(%rsp) movaps %xmm2,nb334_fizO(%rsp) ## move j O forces to local temp variables movq nb334_faction(%rbp),%rdi movlps (%rdi,%rax,4),%xmm4 ## jxOa jyOa - - movlps (%rdi,%rcx,4),%xmm5 ## jxOc jyOc - - movhps (%rdi,%rbx,4),%xmm4 ## jxOa jyOa jxOb jyOb movhps (%rdi,%rdx,4),%xmm5 ## jxOc jyOc jxOd jyOd movss 8(%rdi,%rax,4),%xmm6 ## jzOa - - - movss 8(%rdi,%rcx,4),%xmm7 ## jzOc - - - movhps 8(%rdi,%rbx,4),%xmm6 ## jzOa - jzOb - movhps 8(%rdi,%rdx,4),%xmm7 ## jzOc - jzOd - shufps $136,%xmm7,%xmm6 ## 10001000 => jzOa jzOb jzOc jzOd ## xmm4: jxOa jyOa jxOb jyOb ## xmm5: jxOc jyOc jxOd jyOd ## xmm6: jzOa jzOb jzOc jzOd ## update O forces movaps %xmm13,%xmm12 unpcklps %xmm14,%xmm13 ## (local) fjx1 fjx1 fjy1 fjy2 unpckhps %xmm14,%xmm12 ## (local) fjx3 fjx4 fjy3 fjy4 addps %xmm13,%xmm4 addps %xmm12,%xmm5 addps %xmm15,%xmm6 movhlps %xmm6,%xmm7 ## fH1zc fH1zd movlps %xmm4,(%rdi,%rax,4) movhps %xmm4,(%rdi,%rbx,4) movlps %xmm5,(%rdi,%rcx,4) movhps %xmm5,(%rdi,%rdx,4) movss %xmm6,8(%rdi,%rax,4) movss %xmm7,8(%rdi,%rcx,4) shufps $1,%xmm6,%xmm6 shufps $1,%xmm7,%xmm7 movss %xmm6,8(%rdi,%rbx,4) movss %xmm7,8(%rdi,%rdx,4) ## done with OO interaction ## move j H1 coordinates to local temp variables movq nb334_pos(%rbp),%rsi movlps 12(%rsi,%rax,4),%xmm0 ## jxH1a jyH1a - - movlps 12(%rsi,%rcx,4),%xmm1 ## jxH1c jyH1c - - movhps 12(%rsi,%rbx,4),%xmm0 ## jxH1a jyH1a jxH1b jyH1b movhps 12(%rsi,%rdx,4),%xmm1 ## jxH1c jyH1c jxH1d jyH1d movss 20(%rsi,%rax,4),%xmm2 ## jzH1a - - - movss 20(%rsi,%rcx,4),%xmm3 ## jzH1c - - - movhps 20(%rsi,%rbx,4),%xmm2 ## jzH1a - jzH1b - movhps 20(%rsi,%rdx,4),%xmm3 ## jzH1c - jzH1d - movd %eax,%mm0 ## save j3 in mm0-mm3 movd %ebx,%mm1 movd %ecx,%mm2 movd %edx,%mm3 movaps %xmm0,%xmm4 unpcklps %xmm1,%xmm0 ## jxH1a jxH1c jyH1a jyH1c unpckhps %xmm1,%xmm4 ## jxH1b jxH1d jyH1b jyH1d movaps %xmm0,%xmm1 unpcklps %xmm4,%xmm0 ## x unpckhps %xmm4,%xmm1 ## y shufps $136,%xmm3,%xmm2 ## 10001000 => jzH1a jzH1b jzH1c jzH1d ## xmm0 = H1x ## xmm1 = H1y ## xmm2 = H1z movaps %xmm0,%xmm3 movaps %xmm1,%xmm4 movaps %xmm2,%xmm5 movaps %xmm0,%xmm6 movaps %xmm1,%xmm7 movaps %xmm2,%xmm8 subps nb334_ixH1(%rsp),%xmm0 subps nb334_iyH1(%rsp),%xmm1 subps nb334_izH1(%rsp),%xmm2 subps nb334_ixH2(%rsp),%xmm3 subps nb334_iyH2(%rsp),%xmm4 subps nb334_izH2(%rsp),%xmm5 subps nb334_ixM(%rsp),%xmm6 subps nb334_iyM(%rsp),%xmm7 subps nb334_izM(%rsp),%xmm8 movaps %xmm0,nb334_dxH1H1(%rsp) movaps %xmm1,nb334_dyH1H1(%rsp) movaps %xmm2,nb334_dzH1H1(%rsp) mulps %xmm0,%xmm0 mulps %xmm1,%xmm1 mulps %xmm2,%xmm2 movaps %xmm3,nb334_dxH2H1(%rsp) movaps %xmm4,nb334_dyH2H1(%rsp) movaps %xmm5,nb334_dzH2H1(%rsp) mulps %xmm3,%xmm3 mulps %xmm4,%xmm4 mulps %xmm5,%xmm5 movaps %xmm6,nb334_dxMH1(%rsp) movaps %xmm7,nb334_dyMH1(%rsp) movaps %xmm8,nb334_dzMH1(%rsp) mulps %xmm6,%xmm6 mulps %xmm7,%xmm7 mulps %xmm8,%xmm8 addps %xmm1,%xmm0 addps %xmm2,%xmm0 addps %xmm4,%xmm3 addps %xmm5,%xmm3 addps %xmm7,%xmm6 addps %xmm8,%xmm6 ## start doing invsqrt for jH1 atoms rsqrtps %xmm0,%xmm1 rsqrtps %xmm3,%xmm4 rsqrtps %xmm6,%xmm7 movaps %xmm1,%xmm2 movaps %xmm4,%xmm5 movaps %xmm7,%xmm8 mulps %xmm1,%xmm1 ## lu*lu mulps %xmm4,%xmm4 ## lu*lu mulps %xmm7,%xmm7 ## lu*lu movaps nb334_three(%rsp),%xmm9 movaps %xmm9,%xmm10 movaps %xmm9,%xmm11 mulps %xmm0,%xmm1 ## rsq*lu*lu mulps %xmm3,%xmm4 ## rsq*lu*lu mulps %xmm6,%xmm7 ## rsq*lu*lu subps %xmm1,%xmm9 subps %xmm4,%xmm10 subps %xmm7,%xmm11 ## 3-rsq*lu*lu mulps %xmm2,%xmm9 mulps %xmm5,%xmm10 mulps %xmm8,%xmm11 ## lu*(3-rsq*lu*lu) movaps nb334_half(%rsp),%xmm4 mulps %xmm4,%xmm9 ## rinvH1H1 mulps %xmm4,%xmm10 ## rinvH2H1 mulps %xmm4,%xmm11 ## rinvMH1 movaps %xmm9,nb334_rinvH1H1(%rsp) movaps %xmm10,nb334_rinvH2H1(%rsp) movaps %xmm11,nb334_rinvMH1(%rsp) ## H1 interactions ## rsq in xmm0,xmm3,xmm6 ## rinv in xmm9, xmm10, xmm11 movaps nb334_tsc(%rsp),%xmm1 mulps %xmm9,%xmm0 ## r mulps %xmm10,%xmm3 mulps %xmm11,%xmm6 mulps %xmm1,%xmm0 ## rtab mulps %xmm1,%xmm3 mulps %xmm1,%xmm6 ## truncate and convert to integers cvttps2dq %xmm0,%xmm1 cvttps2dq %xmm3,%xmm4 cvttps2dq %xmm6,%xmm7 ## convert back to float cvtdq2ps %xmm1,%xmm2 cvtdq2ps %xmm4,%xmm5 cvtdq2ps %xmm7,%xmm8 ## multiply by 4 pslld $2,%xmm1 pslld $2,%xmm4 pslld $2,%xmm7 ## multiply by three (copy, mult. by two, add back) movaps %xmm1,%xmm10 movaps %xmm4,%xmm11 movaps %xmm7,%xmm12 pslld $1,%xmm1 pslld $1,%xmm4 pslld $1,%xmm7 paddd %xmm10,%xmm1 paddd %xmm11,%xmm4 paddd %xmm12,%xmm7 ## move to integer registers movhlps %xmm1,%xmm13 movhlps %xmm4,%xmm14 movhlps %xmm7,%xmm15 movd %xmm1,%eax movd %xmm4,%r8d movd %xmm7,%r12d movd %xmm13,%ecx movd %xmm14,%r10d movd %xmm15,%r14d pshufd $1,%xmm1,%xmm1 pshufd $1,%xmm4,%xmm4 pshufd $1,%xmm7,%xmm7 pshufd $1,%xmm13,%xmm13 pshufd $1,%xmm14,%xmm14 pshufd $1,%xmm15,%xmm15 movd %xmm1,%ebx movd %xmm4,%r9d movd %xmm7,%r13d movd %xmm13,%edx movd %xmm14,%r11d movd %xmm15,%r15d movq nb334_VFtab(%rbp),%rsi ## calculate eps subps %xmm2,%xmm0 subps %xmm5,%xmm3 subps %xmm8,%xmm6 movaps %xmm0,nb334_epsH1(%rsp) movaps %xmm3,nb334_epsH2(%rsp) movaps %xmm6,nb334_epsM(%rsp) ## Load LOTS of table data movlps (%rsi,%rax,4),%xmm1 movlps (%rsi,%r8,4),%xmm5 movlps (%rsi,%r12,4),%xmm9 movlps (%rsi,%rcx,4),%xmm3 movlps (%rsi,%r10,4),%xmm7 movlps (%rsi,%r14,4),%xmm11 movhps (%rsi,%rbx,4),%xmm1 movhps (%rsi,%r9,4),%xmm5 movhps (%rsi,%r13,4),%xmm9 movhps (%rsi,%rdx,4),%xmm3 movhps (%rsi,%r11,4),%xmm7 movhps (%rsi,%r15,4),%xmm11 movaps %xmm1,%xmm0 movaps %xmm5,%xmm4 movaps %xmm9,%xmm8 shufps $136,%xmm3,%xmm0 ## 10001000 shufps $136,%xmm7,%xmm4 ## 10001000 shufps $136,%xmm11,%xmm8 ## 10001000 shufps $221,%xmm3,%xmm1 ## 11011101 shufps $221,%xmm7,%xmm5 ## 11011101 shufps $221,%xmm11,%xmm9 ## 11011101 movlps 8(%rsi,%rax,4),%xmm3 movlps 8(%rsi,%r8,4),%xmm7 movlps 8(%rsi,%r12,4),%xmm11 movlps 8(%rsi,%rcx,4),%xmm12 movlps 8(%rsi,%r10,4),%xmm13 movlps 8(%rsi,%r14,4),%xmm14 movhps 8(%rsi,%rbx,4),%xmm3 movhps 8(%rsi,%r9,4),%xmm7 movhps 8(%rsi,%r13,4),%xmm11 movhps 8(%rsi,%rdx,4),%xmm12 movhps 8(%rsi,%r11,4),%xmm13 movhps 8(%rsi,%r15,4),%xmm14 movaps %xmm3,%xmm2 movaps %xmm7,%xmm6 movaps %xmm11,%xmm10 shufps $136,%xmm12,%xmm2 ## 10001000 shufps $136,%xmm13,%xmm6 ## 10001000 shufps $136,%xmm14,%xmm10 ## 10001000 shufps $221,%xmm12,%xmm3 ## 11011101 shufps $221,%xmm13,%xmm7 ## 11011101 shufps $221,%xmm14,%xmm11 ## 11011101 ## table data ready in xmm0-xmm3 , xmm4-xmm7 , and xmm8-xmm11 movaps nb334_epsH1(%rsp),%xmm12 movaps nb334_epsH2(%rsp),%xmm13 movaps nb334_epsM(%rsp),%xmm14 mulps %xmm12,%xmm3 ## Heps mulps %xmm13,%xmm7 mulps %xmm14,%xmm11 mulps %xmm12,%xmm2 ## Geps mulps %xmm13,%xmm6 mulps %xmm14,%xmm10 mulps %xmm12,%xmm3 ## Heps2 mulps %xmm13,%xmm7 mulps %xmm14,%xmm11 addps %xmm2,%xmm1 ## F+Geps addps %xmm6,%xmm5 addps %xmm10,%xmm9 addps %xmm3,%xmm1 ## F+Geps+Heps2 = Fp addps %xmm7,%xmm5 addps %xmm11,%xmm9⌨️ 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?