nb_kernel231_x86_64_sse.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,053 行 · 第 1/5 页
S
2,053 行
shufps $221,%xmm5,%xmm1 ## 11011101 ## xmm0 = jx ## xmm1 = jy ## xmm2 = jz movaps %xmm0,%xmm3 movaps %xmm1,%xmm4 movaps %xmm2,%xmm5 movaps %xmm0,%xmm6 movaps %xmm1,%xmm7 movaps %xmm2,%xmm8 subps nb231_ixO(%rsp),%xmm0 subps nb231_iyO(%rsp),%xmm1 subps nb231_izO(%rsp),%xmm2 subps nb231_ixH1(%rsp),%xmm3 subps nb231_iyH1(%rsp),%xmm4 subps nb231_izH1(%rsp),%xmm5 subps nb231_ixH2(%rsp),%xmm6 subps nb231_iyH2(%rsp),%xmm7 subps nb231_izH2(%rsp),%xmm8 movaps %xmm0,nb231_dxO(%rsp) movaps %xmm1,nb231_dyO(%rsp) movaps %xmm2,nb231_dzO(%rsp) mulps %xmm0,%xmm0 mulps %xmm1,%xmm1 mulps %xmm2,%xmm2 movaps %xmm3,nb231_dxH1(%rsp) movaps %xmm4,nb231_dyH1(%rsp) movaps %xmm5,nb231_dzH1(%rsp) mulps %xmm3,%xmm3 mulps %xmm4,%xmm4 mulps %xmm5,%xmm5 movaps %xmm6,nb231_dxH2(%rsp) movaps %xmm7,nb231_dyH2(%rsp) movaps %xmm8,nb231_dzH2(%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 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 nb231_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 nb231_half(%rsp),%xmm4 mulps %xmm4,%xmm9 ## rinvO mulps %xmm4,%xmm10 ## rinvH1 mulps %xmm4,%xmm11 ## rinvH2 ## O interactions ## rsq in xmm0,xmm3,xmm6 ## rinv in xmm9, xmm10, xmm11 movaps %xmm0,nb231_rsqO(%rsp) movaps %xmm3,nb231_rsqH1(%rsp) movaps %xmm6,nb231_rsqH2(%rsp) movaps %xmm9,nb231_rinvO(%rsp) movaps %xmm10,nb231_rinvH1(%rsp) movaps %xmm11,nb231_rinvH2(%rsp) ## table LJ interaction mulps %xmm9,%xmm0 mulps nb231_tsc(%rsp),%xmm0 ## rtab ## truncate and convert to integers cvttps2dq %xmm0,%xmm1 ## convert back to float cvtdq2ps %xmm1,%xmm2 ## multiply by 8 pslld $3,%xmm1 ## move to integer registers movhlps %xmm1,%xmm13 movd %xmm1,%r8d movd %xmm13,%r10d shufps $1,%xmm1,%xmm1 shufps $1,%xmm13,%xmm13 movd %xmm1,%r9d movd %xmm13,%r11d ## calculate eps subps %xmm2,%xmm0 movq nb231_VFtab(%rbp),%rsi movlps (%rsi,%r8,4),%xmm5 movlps 16(%rsi,%r8,4),%xmm9 movlps (%rsi,%r10,4),%xmm7 movlps 16(%rsi,%r10,4),%xmm11 movhps (%rsi,%r9,4),%xmm5 movhps 16(%rsi,%r9,4),%xmm9 movhps (%rsi,%r11,4),%xmm7 movhps 16(%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 8(%rsi,%r8,4),%xmm7 movlps 24(%rsi,%r8,4),%xmm11 movlps 8(%rsi,%r10,4),%xmm13 movlps 24(%rsi,%r10,4),%xmm14 movhps 8(%rsi,%r9,4),%xmm7 movhps 24(%rsi,%r9,4),%xmm11 movhps 8(%rsi,%r11,4),%xmm13 movhps 24(%rsi,%r11,4),%xmm14 movaps %xmm7,%xmm6 movaps %xmm11,%xmm10 shufps $136,%xmm13,%xmm6 ## 10001000 shufps $136,%xmm14,%xmm10 ## 10001000 shufps $221,%xmm13,%xmm7 ## 11011101 shufps $221,%xmm14,%xmm11 ## 11011101 ## dispersion table in xmm4-xmm7, repulsion table in xmm8-xmm11 mulps %xmm0,%xmm7 ## Heps mulps %xmm0,%xmm11 mulps %xmm0,%xmm6 ## Geps mulps %xmm0,%xmm10 mulps %xmm0,%xmm7 ## Heps2 mulps %xmm0,%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 %xmm0,%xmm5 ## eps*Fp mulps %xmm0,%xmm9 movaps nb231_c6(%rsp),%xmm12 movaps nb231_c12(%rsp),%xmm13 addps %xmm4,%xmm5 ## VV addps %xmm8,%xmm9 mulps %xmm12,%xmm5 ## VV*c6 = vnb6 mulps %xmm13,%xmm9 ## VV*c12 = vnb12 addps %xmm9,%xmm5 addps nb231_Vvdwtot(%rsp),%xmm5 movaps %xmm5,nb231_Vvdwtot(%rsp) mulps %xmm12,%xmm7 ## FF*c6 = fnb6 mulps %xmm13,%xmm11 ## FF*c12 = fnb12 addps %xmm11,%xmm7 mulps nb231_tsc(%rsp),%xmm7 movaps %xmm7,nb231_fstmp(%rsp) ## Coulomb reaction-field interaction movaps nb231_rsqO(%rsp),%xmm0 movaps nb231_rsqH1(%rsp),%xmm3 movaps nb231_rsqH2(%rsp),%xmm6 movaps nb231_rinvO(%rsp),%xmm9 movaps nb231_rinvH1(%rsp),%xmm10 movaps nb231_rinvH2(%rsp),%xmm11 movaps %xmm9,%xmm1 ## copy of rinv movaps %xmm10,%xmm4 movaps %xmm11,%xmm7 movaps nb231_krf(%rsp),%xmm2 mulps %xmm10,%xmm10 ## rinvsq mulps %xmm11,%xmm11 mulps %xmm2,%xmm0 ## k*rsq mulps %xmm2,%xmm3 mulps %xmm2,%xmm6 movaps %xmm0,%xmm2 ## copy of k*rsq movaps %xmm3,%xmm5 movaps %xmm6,%xmm8 addps %xmm1,%xmm2 ## rinv+krsq addps %xmm4,%xmm5 addps %xmm7,%xmm8 subps nb231_crf(%rsp),%xmm2 ## rinv+krsq-crf subps nb231_crf(%rsp),%xmm5 subps nb231_crf(%rsp),%xmm8 mulps nb231_qqO(%rsp),%xmm2 ## voul=qq*(rinv+ krsq-crf) mulps nb231_qqH(%rsp),%xmm5 ## voul=qq*(rinv+ krsq-crf) mulps nb231_qqH(%rsp),%xmm8 ## voul=qq*(rinv+ krsq-crf) addps %xmm0,%xmm0 ## 2*krsq addps %xmm3,%xmm3 addps %xmm6,%xmm6 subps %xmm0,%xmm1 ## rinv-2*krsq subps %xmm3,%xmm4 subps %xmm6,%xmm7 mulps nb231_qqO(%rsp),%xmm1 ## (rinv-2*krsq)*qq mulps nb231_qqH(%rsp),%xmm4 mulps nb231_qqH(%rsp),%xmm7 addps nb231_vctot(%rsp),%xmm2 addps %xmm8,%xmm5 addps %xmm5,%xmm2 movaps %xmm2,%xmm15 mulps %xmm9,%xmm1 ## fijC mulps %xmm10,%xmm4 mulps %xmm11,%xmm7 ## xmm1, xmm4, xmm7 contains fscal coul ## xmm15 contains vctot subps nb231_fstmp(%rsp),%xmm1 mulps %xmm9,%xmm1 movaps %xmm2,nb231_vctot(%rsp) ## move j forces to local temp variables movq nb231_faction(%rbp),%rdi movlps (%rdi,%rax,4),%xmm9 ## jxa jya - - movlps (%rdi,%rcx,4),%xmm10 ## jxc jyc - - movhps (%rdi,%rbx,4),%xmm9 ## jxa jya jxb jyb movhps (%rdi,%rdx,4),%xmm10 ## jxc jyc jxd jyd movss 8(%rdi,%rax,4),%xmm11 ## jza - - - movss 8(%rdi,%rcx,4),%xmm12 ## jzc - - - movss 8(%rdi,%rbx,4),%xmm6 ## jzb movss 8(%rdi,%rdx,4),%xmm8 ## jzd movlhps %xmm6,%xmm11 ## jza - jzb - movlhps %xmm8,%xmm12 ## jzc - jzd - shufps $136,%xmm12,%xmm11 ## 10001000 => jza jzb jzc jzd ## xmm9: jxa jya jxb jyb ## xmm10: jxc jyc jxd jyd ## xmm11: jza jzb jzc jzd movaps %xmm1,%xmm0 movaps %xmm1,%xmm2 movaps %xmm4,%xmm3 movaps %xmm4,%xmm5 movaps %xmm7,%xmm6 movaps %xmm7,%xmm8 mulps nb231_dxO(%rsp),%xmm0 mulps nb231_dyO(%rsp),%xmm1 mulps nb231_dzO(%rsp),%xmm2 mulps nb231_dxH1(%rsp),%xmm3 mulps nb231_dyH1(%rsp),%xmm4 mulps nb231_dzH1(%rsp),%xmm5 mulps nb231_dxH2(%rsp),%xmm6 mulps nb231_dyH2(%rsp),%xmm7 mulps nb231_dzH2(%rsp),%xmm8 movaps %xmm0,%xmm13 movaps %xmm1,%xmm14 addps %xmm2,%xmm11 addps nb231_fixO(%rsp),%xmm0 addps nb231_fiyO(%rsp),%xmm1 addps nb231_fizO(%rsp),%xmm2 addps %xmm3,%xmm13 addps %xmm4,%xmm14 addps %xmm5,%xmm11 addps nb231_fixH1(%rsp),%xmm3 addps nb231_fiyH1(%rsp),%xmm4 addps nb231_fizH1(%rsp),%xmm5 addps %xmm6,%xmm13 addps %xmm7,%xmm14 addps %xmm8,%xmm11 addps nb231_fixH2(%rsp),%xmm6 addps nb231_fiyH2(%rsp),%xmm7 addps nb231_fizH2(%rsp),%xmm8 movaps %xmm0,nb231_fixO(%rsp) movaps %xmm1,nb231_fiyO(%rsp) movaps %xmm2,nb231_fizO(%rsp) movaps %xmm3,nb231_fixH1(%rsp) movaps %xmm4,nb231_fiyH1(%rsp) movaps %xmm5,nb231_fizH1(%rsp) movaps %xmm6,nb231_fixH2(%rsp) movaps %xmm7,nb231_fiyH2(%rsp) movaps %xmm8,nb231_fizH2(%rsp) ## xmm9 = fjx ## xmm10 = fjy ## xmm11 = fjz movaps %xmm13,%xmm15 unpcklps %xmm14,%xmm13 unpckhps %xmm14,%xmm15 addps %xmm13,%xmm9 addps %xmm15,%xmm10 movhlps %xmm11,%xmm12 ## fjzc fjzd movlps %xmm9,(%rdi,%rax,4) movhps %xmm9,(%rdi,%rbx,4) movlps %xmm10,(%rdi,%rcx,4) movhps %xmm10,(%rdi,%rdx,4) movss %xmm11,8(%rdi,%rax,4) movss %xmm12,8(%rdi,%rcx,4) shufps $1,%xmm11,%xmm11 shufps $1,%xmm12,%xmm12 movss %xmm11,8(%rdi,%rbx,4) movss %xmm12,8(%rdi,%rdx,4) ## should we do one more iteration? subl $4,nb231_innerk(%rsp) jl _nb_kernel231_x86_64_sse.nb231_odd_inner jmp _nb_kernel231_x86_64_sse.nb231_unroll_loop_nb_kernel231_x86_64_sse.nb231_odd_inner: addl $4,nb231_innerk(%rsp) jnz _nb_kernel231_x86_64_sse.nb231_odd_loop jmp _nb_kernel231_x86_64_sse.nb231_updateouterdata_nb_kernel231_x86_64_sse.nb231_odd_loop: movq nb231_innerjjnr(%rsp),%rdx ## pointer to jjnr[k] movl (%rdx),%eax addq $4,nb231_innerjjnr(%rsp) xorps %xmm4,%xmm4 movss nb231_iqO(%rsp),%xmm4 movq nb231_charge(%rbp),%rsi movhps nb231_iqH(%rsp),%xmm4 movss (%rsi,%rax,4),%xmm3 ## charge in xmm3 shufps $0,%xmm3,%xmm3 mulps %xmm4,%xmm3 movaps %xmm3,nb231_qqO(%rsp) ## use oxygen qq for storage xorps %xmm6,%xmm6 movq nb231_type(%rbp),%rsi movl (%rsi,%rax,4),%ebx movq nb231_vdwparam(%rbp),%rsi shll %ebx addl nb231_ntia(%rsp),%ebx movlps (%rsi,%rbx,4),%xmm6 movaps %xmm6,%xmm7 shufps $252,%xmm6,%xmm6 ## constant 11111100 shufps $253,%xmm7,%xmm7 ## constant 11111101 movaps %xmm6,nb231_c6(%rsp) movaps %xmm7,nb231_c12(%rsp) movq nb231_pos(%rbp),%rsi lea (%rax,%rax,2),%rax ## move j coords to xmm0-xmm2 movss (%rsi,%rax,4),%xmm3 movss 4(%rsi,%rax,4),%xmm4 movss 8(%rsi,%rax,4),%xmm5 shufps $0,%xmm3,%xmm3 shufps $0,%xmm4,%xmm4 shufps $0,%xmm5,%xmm5 movss nb231_ixO(%rsp),%xmm0 movss nb231_iyO(%rsp),%xmm1 movss nb231_izO(%rsp),%xmm2 movlps nb231_ixH1(%rsp),%xmm6 movlps nb231_ixH2(%rsp),%xmm7 unpcklps %xmm7,%xmm6 movlhps %xmm6,%xmm0 movlps nb231_iyH1(%rsp),%xmm6 movlps nb231_iyH2(%rsp),%xmm7 unpcklps %xmm7,%xmm6 movlhps %xmm6,%xmm1 movlps nb231_izH1(%rsp),%xmm6 movlps nb231_izH2(%rsp),%xmm7 unpcklps %xmm7,%xmm6 movlhps %xmm6,%xmm2 subps %xmm0,%xmm3 subps %xmm1,%xmm4 subps %xmm2,%xmm5 movaps %xmm3,nb231_dxO(%rsp) movaps %xmm4,nb231_dyO(%rsp) movaps %xmm5,nb231_dzO(%rsp) mulps %xmm3,%xmm3 mulps %xmm4,%xmm4 mulps %xmm5,%xmm5⌨️ 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?