nb_kernel332_x86_64_sse2.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,381 行 · 第 1/5 页
S
2,381 行
## move to integer registers pshufd $1,%xmm1,%xmm13 pshufd $1,%xmm4,%xmm14 pshufd $1,%xmm7,%xmm15 movd %xmm1,%r8d movd %xmm4,%r10d movd %xmm7,%r12d movd %xmm13,%r9d movd %xmm14,%r11d movd %xmm15,%r13d movq nb332_VFtab(%rbp),%rsi ## calculate eps subpd %xmm2,%xmm0 subpd %xmm5,%xmm3 subpd %xmm8,%xmm6 movapd %xmm0,nb332_epsO(%rsp) movapd %xmm3,nb332_epsH1(%rsp) movapd %xmm6,nb332_epsH2(%rsp) ## Load LOTS of table data movlpd (%rsi,%r8,8),%xmm0 movlpd 8(%rsi,%r8,8),%xmm1 movlpd 16(%rsi,%r8,8),%xmm2 movlpd 24(%rsi,%r8,8),%xmm3 movlpd (%rsi,%r10,8),%xmm4 movlpd 8(%rsi,%r10,8),%xmm5 movlpd 16(%rsi,%r10,8),%xmm6 movlpd 24(%rsi,%r10,8),%xmm7 movlpd (%rsi,%r12,8),%xmm8 movlpd 8(%rsi,%r12,8),%xmm9 movlpd 16(%rsi,%r12,8),%xmm10 movlpd 24(%rsi,%r12,8),%xmm11 movhpd (%rsi,%r9,8),%xmm0 movhpd 8(%rsi,%r9,8),%xmm1 movhpd 16(%rsi,%r9,8),%xmm2 movhpd 24(%rsi,%r9,8),%xmm3 movhpd (%rsi,%r11,8),%xmm4 movhpd 8(%rsi,%r11,8),%xmm5 movhpd 16(%rsi,%r11,8),%xmm6 movhpd 24(%rsi,%r11,8),%xmm7 movhpd (%rsi,%r13,8),%xmm8 movhpd 8(%rsi,%r13,8),%xmm9 movhpd 16(%rsi,%r13,8),%xmm10 movhpd 24(%rsi,%r13,8),%xmm11 ## table data ready in xmm0-xmm3 , xmm4-xmm7 , and xmm8-xmm11 movapd nb332_epsO(%rsp),%xmm12 movapd nb332_epsH1(%rsp),%xmm13 movapd nb332_epsH2(%rsp),%xmm14 mulpd %xmm12,%xmm3 ## Heps mulpd %xmm13,%xmm7 mulpd %xmm14,%xmm11 mulpd %xmm12,%xmm2 ## Geps mulpd %xmm13,%xmm6 mulpd %xmm14,%xmm10 mulpd %xmm12,%xmm3 ## Heps2 mulpd %xmm13,%xmm7 mulpd %xmm14,%xmm11 addpd %xmm2,%xmm1 ## F+Geps 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 nb332_qqOH(%rsp),%xmm12 movapd nb332_qqHH(%rsp),%xmm13 addpd %xmm0,%xmm1 ## VV addpd %xmm4,%xmm5 addpd %xmm8,%xmm9 mulpd %xmm12,%xmm1 ## VV*qq = vcoul mulpd %xmm13,%xmm5 mulpd %xmm13,%xmm9 mulpd %xmm12,%xmm3 ## FF*qq = fij mulpd %xmm13,%xmm7 mulpd %xmm13,%xmm11 ## accumulate vctot addpd nb332_vctot(%rsp),%xmm1 addpd %xmm9,%xmm5 addpd %xmm5,%xmm1 movapd %xmm1,nb332_vctot(%rsp) movapd nb332_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 nb332_rinvOH1(%rsp),%xmm4 mulpd nb332_rinvH1H1(%rsp),%xmm8 mulpd nb332_rinvH2H1(%rsp),%xmm11 ## move j H1 forces to xmm0-xmm2 movq nb332_faction(%rbp),%rdi movlpd 24(%rdi,%rax,8),%xmm0 movlpd 32(%rdi,%rax,8),%xmm1 movlpd 40(%rdi,%rax,8),%xmm2 movhpd 24(%rdi,%rbx,8),%xmm0 movhpd 32(%rdi,%rbx,8),%xmm1 movhpd 40(%rdi,%rbx,8),%xmm2 movapd %xmm4,%xmm3 movapd %xmm4,%xmm5 movapd %xmm8,%xmm7 movapd %xmm8,%xmm9 movapd %xmm11,%xmm10 movapd %xmm11,%xmm12 mulpd nb332_dxOH1(%rsp),%xmm3 mulpd nb332_dyOH1(%rsp),%xmm4 mulpd nb332_dzOH1(%rsp),%xmm5 mulpd nb332_dxH1H1(%rsp),%xmm7 mulpd nb332_dyH1H1(%rsp),%xmm8 mulpd nb332_dzH1H1(%rsp),%xmm9 mulpd nb332_dxH2H1(%rsp),%xmm10 mulpd nb332_dyH2H1(%rsp),%xmm11 mulpd nb332_dzH2H1(%rsp),%xmm12 addpd %xmm3,%xmm0 addpd %xmm4,%xmm1 addpd %xmm5,%xmm2 addpd nb332_fixO(%rsp),%xmm3 addpd nb332_fiyO(%rsp),%xmm4 addpd nb332_fizO(%rsp),%xmm5 addpd %xmm7,%xmm0 addpd %xmm8,%xmm1 addpd %xmm9,%xmm2 addpd nb332_fixH1(%rsp),%xmm7 addpd nb332_fiyH1(%rsp),%xmm8 addpd nb332_fizH1(%rsp),%xmm9 addpd %xmm10,%xmm0 addpd %xmm11,%xmm1 addpd %xmm12,%xmm2 addpd nb332_fixH2(%rsp),%xmm10 addpd nb332_fiyH2(%rsp),%xmm11 addpd nb332_fizH2(%rsp),%xmm12 movapd %xmm3,nb332_fixO(%rsp) movapd %xmm4,nb332_fiyO(%rsp) movapd %xmm5,nb332_fizO(%rsp) movapd %xmm7,nb332_fixH1(%rsp) movapd %xmm8,nb332_fiyH1(%rsp) movapd %xmm9,nb332_fizH1(%rsp) movapd %xmm10,nb332_fixH2(%rsp) movapd %xmm11,nb332_fiyH2(%rsp) movapd %xmm12,nb332_fizH2(%rsp) ## store back j H1 forces from xmm0-xmm2 movlpd %xmm0,24(%rdi,%rax,8) movlpd %xmm1,32(%rdi,%rax,8) movlpd %xmm2,40(%rdi,%rax,8) movhpd %xmm0,24(%rdi,%rbx,8) movhpd %xmm1,32(%rdi,%rbx,8) movhpd %xmm2,40(%rdi,%rbx,8) ## move j H2 coordinates to local temp variables movq nb332_pos(%rbp),%rsi movlpd 48(%rsi,%rax,8),%xmm0 movlpd 56(%rsi,%rax,8),%xmm1 movlpd 64(%rsi,%rax,8),%xmm2 movhpd 48(%rsi,%rbx,8),%xmm0 movhpd 56(%rsi,%rbx,8),%xmm1 movhpd 64(%rsi,%rbx,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 subpd nb332_ixO(%rsp),%xmm0 subpd nb332_iyO(%rsp),%xmm1 subpd nb332_izO(%rsp),%xmm2 subpd nb332_ixH1(%rsp),%xmm3 subpd nb332_iyH1(%rsp),%xmm4 subpd nb332_izH1(%rsp),%xmm5 subpd nb332_ixH2(%rsp),%xmm6 subpd nb332_iyH2(%rsp),%xmm7 subpd nb332_izH2(%rsp),%xmm8 movapd %xmm0,nb332_dxOH2(%rsp) movapd %xmm1,nb332_dyOH2(%rsp) movapd %xmm2,nb332_dzOH2(%rsp) mulpd %xmm0,%xmm0 mulpd %xmm1,%xmm1 mulpd %xmm2,%xmm2 movapd %xmm3,nb332_dxH1H2(%rsp) movapd %xmm4,nb332_dyH1H2(%rsp) movapd %xmm5,nb332_dzH1H2(%rsp) mulpd %xmm3,%xmm3 mulpd %xmm4,%xmm4 mulpd %xmm5,%xmm5 movapd %xmm6,nb332_dxH2H2(%rsp) movapd %xmm7,nb332_dyH2H2(%rsp) movapd %xmm8,nb332_dzH2H2(%rsp) mulpd %xmm6,%xmm6 mulpd %xmm7,%xmm7 mulpd %xmm8,%xmm8 addpd %xmm1,%xmm0 addpd %xmm2,%xmm0 addpd %xmm4,%xmm3 addpd %xmm5,%xmm3 addpd %xmm7,%xmm6 addpd %xmm8,%xmm6 ## start doing invsqrt for jH2 atoms cvtpd2ps %xmm0,%xmm1 cvtpd2ps %xmm3,%xmm4 cvtpd2ps %xmm6,%xmm7 rsqrtps %xmm1,%xmm1 rsqrtps %xmm4,%xmm4 rsqrtps %xmm7,%xmm7 cvtps2pd %xmm1,%xmm1 cvtps2pd %xmm4,%xmm4 cvtps2pd %xmm7,%xmm7 movapd %xmm1,%xmm2 movapd %xmm4,%xmm5 movapd %xmm7,%xmm8 mulpd %xmm1,%xmm1 ## lu*lu mulpd %xmm4,%xmm4 ## lu*lu mulpd %xmm7,%xmm7 ## lu*lu movapd nb332_three(%rsp),%xmm9 movapd %xmm9,%xmm10 movapd %xmm9,%xmm11 mulpd %xmm0,%xmm1 ## rsq*lu*lu mulpd %xmm3,%xmm4 ## rsq*lu*lu mulpd %xmm6,%xmm7 ## rsq*lu*lu subpd %xmm1,%xmm9 subpd %xmm4,%xmm10 subpd %xmm7,%xmm11 ## 3-rsq*lu*lu mulpd %xmm2,%xmm9 mulpd %xmm5,%xmm10 mulpd %xmm8,%xmm11 ## lu*(3-rsq*lu*lu) movapd nb332_half(%rsp),%xmm15 mulpd %xmm15,%xmm9 ## first iteration for rinvOH2 mulpd %xmm15,%xmm10 ## first iteration for rinvH1H2 mulpd %xmm15,%xmm11 ## first iteration for rinvH2H2 ## second iteration step movapd %xmm9,%xmm2 movapd %xmm10,%xmm5 movapd %xmm11,%xmm8 mulpd %xmm2,%xmm2 ## lu*lu mulpd %xmm5,%xmm5 ## lu*lu mulpd %xmm8,%xmm8 ## lu*lu movapd nb332_three(%rsp),%xmm1 movapd %xmm1,%xmm4 movapd %xmm1,%xmm7 mulpd %xmm0,%xmm2 ## rsq*lu*lu mulpd %xmm3,%xmm5 ## rsq*lu*lu mulpd %xmm6,%xmm8 ## rsq*lu*lu subpd %xmm2,%xmm1 subpd %xmm5,%xmm4 subpd %xmm8,%xmm7 ## 3-rsq*lu*lu mulpd %xmm1,%xmm9 mulpd %xmm4,%xmm10 mulpd %xmm7,%xmm11 ## lu*(3-rsq*lu*lu) movapd nb332_half(%rsp),%xmm15 mulpd %xmm15,%xmm9 ## rinvOH2 mulpd %xmm15,%xmm10 ## rinvH1H2 mulpd %xmm15,%xmm11 ## rinvH2H2 movapd %xmm9,nb332_rinvOH2(%rsp) movapd %xmm10,nb332_rinvH1H2(%rsp) movapd %xmm11,nb332_rinvH2H2(%rsp) ## H2 interactions ## rsq in xmm0,xmm3,xmm6 ## rinv in xmm9, xmm10, xmm11 movapd nb332_tsc(%rsp),%xmm1 mulpd %xmm9,%xmm0 ## r mulpd %xmm10,%xmm3 mulpd %xmm11,%xmm6 mulpd %xmm1,%xmm0 ## rtab mulpd %xmm1,%xmm3 mulpd %xmm1,%xmm6 ## truncate and convert to integers cvttpd2dq %xmm0,%xmm1 cvttpd2dq %xmm3,%xmm4 cvttpd2dq %xmm6,%xmm7 ## convert back to float cvtdq2pd %xmm1,%xmm2 cvtdq2pd %xmm4,%xmm5 cvtdq2pd %xmm7,%xmm8 ## multiply by 4 pslld $2,%xmm1 pslld $2,%xmm4 pslld $2,%xmm7 ## multiply by three (copy, mult. by two, add back) movdqa %xmm1,%xmm10 movdqa %xmm4,%xmm11 movdqa %xmm7,%xmm12 pslld $1,%xmm1 pslld $1,%xmm4 pslld $1,%xmm7 paddd %xmm10,%xmm1 paddd %xmm11,%xmm4 paddd %xmm12,%xmm7 ## move to integer registers pshufd $1,%xmm1,%xmm13 pshufd $1,%xmm4,%xmm14 pshufd $1,%xmm7,%xmm15 movd %xmm1,%r8d movd %xmm4,%r10d movd %xmm7,%r12d movd %xmm13,%r9d movd %xmm14,%r11d movd %xmm15,%r13d movq nb332_VFtab(%rbp),%rsi ## calculate eps subpd %xmm2,%xmm0 subpd %xmm5,%xmm3 subpd %xmm8,%xmm6 movapd %xmm0,nb332_epsO(%rsp) movapd %xmm3,nb332_epsH1(%rsp) movapd %xmm6,nb332_epsH2(%rsp) ## Load LOTS of table data movlpd (%rsi,%r8,8),%xmm0 movlpd 8(%rsi,%r8,8),%xmm1 movlpd 16(%rsi,%r8,8),%xmm2 movlpd 24(%rsi,%r8,8),%xmm3 movlpd (%rsi,%r10,8),%xmm4 movlpd 8(%rsi,%r10,8),%xmm5 movlpd 16(%rsi,%r10,8),%xmm6 movlpd 24(%rsi,%r10,8),%xmm7 movlpd (%rsi,%r12,8),%xmm8 movlpd 8(%rsi,%r12,8),%xmm9 movlpd 16(%rsi,%r12,8),%xmm10 movlpd 24(%rsi,%r12,8),%xmm11 movhpd (%rsi,%r9,8),%xmm0 movhpd 8(%rsi,%r9,8),%xmm1 movhpd 16(%rsi,%r9,8),%xmm2 movhpd 24(%rsi,%r9,8),%xmm3 movhpd (%rsi,%r11,8),%xmm4 movhpd 8(%rsi,%r11,8),%xmm5 movhpd 16(%rsi,%r11,8),%xmm6 movhpd 24(%rsi,%r11,8),%xmm7 movhpd (%rsi,%r13,8),%xmm8 movhpd 8(%rsi,%r13,8),%xmm9 movhpd 16(%rsi,%r13,8),%xmm10 movhpd 24(%rsi,%r13,8),%xmm11 ## table data ready in xmm0-xmm3 , xmm4-xmm7 , and xmm8-xmm11 movapd nb332_epsO(%rsp),%xmm12 movapd nb332_epsH1(%rsp),%xmm13 movapd nb332_epsH2(%rsp),%xmm14 mulpd %xmm12,%xmm3 ## Heps mulpd %xmm13,%xmm7 mulpd %xmm14,%xmm11 mulpd %xmm12,%xmm2 ## Geps mulpd %xmm13,%xmm6 mulpd %xmm14,%xmm10 mulpd %xmm12,%xmm3 ## Heps2 mulpd %xmm13,%xmm7 mulpd %xmm14,%xmm11 addpd %xmm2,%xmm1 ## F+Geps 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 nb332_qqOH(%rsp),%xmm12 movapd nb332_qqHH(%rsp),%xmm13 addpd %xmm0,%xmm1 ## VV addpd %xmm4,%xmm5 addpd %xmm8,%xmm9 mulpd %xmm12,%xmm1 ## VV*qq = vcoul mulpd %xmm13,%xmm5 mulpd %xmm13,%xmm9 mulpd %xmm12,%xmm3 ## FF*qq = fij mulpd %xmm13,%xmm7 mulpd %xmm13,%xmm11 ## accumulate vctot addpd nb332_vctot(%rsp),%xmm1 addpd %xmm9,%xmm5 addpd %xmm5,%xmm1 movapd %xmm1,nb332_vctot(%rsp) movapd nb332_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 nb332_rinvOH2(%rsp),%xmm4 mulpd nb332_rinvH1H2(%rsp),%xmm8 mulpd nb332_rinvH2H2(%rsp),%xmm11 ## move j H2 forces to xmm0-xmm2 movq nb332_faction(%rbp),%rdi movlpd 48(%rdi,%rax,8),%xmm0 movlpd 56(%rdi,%rax,8),%xmm1 movlpd 64(%rdi,%rax,8),%xmm2 movhpd 48(%rdi,%rbx,8),%xmm0 movhpd 56(%rdi,%rbx,8),%xmm1 movhpd 64(%rdi,%rbx,8),%xmm2 movapd %xmm4,%xmm3 movapd %xmm4,%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?