nb_kernel130_x86_64_sse.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 1,895 行 · 第 1/4 页
S
1,895 行
movaps %xmm3,%xmm11 movaps nb130_fix(%rsp),%xmm12 movaps nb130_fiy(%rsp),%xmm13 movaps nb130_fiz(%rsp),%xmm14 mulps nb130_dx(%rsp),%xmm9 mulps nb130_dy(%rsp),%xmm10 mulps nb130_dz(%rsp),%xmm11 ## accumulate i forces addps %xmm9,%xmm12 addps %xmm10,%xmm13 addps %xmm11,%xmm14 movaps %xmm12,nb130_fix(%rsp) movaps %xmm13,nb130_fiy(%rsp) movaps %xmm14,nb130_fiz(%rsp) movq nb130_faction(%rbp),%rsi ## the fj's - start by accumulating x & y forces from memory movlps (%rsi,%rax,4),%xmm0 ## x1 y1 - - movlps (%rsi,%rcx,4),%xmm1 ## x3 y3 - - movhps (%rsi,%rbx,4),%xmm0 ## x1 y1 x2 y2 movhps (%rsi,%rdx,4),%xmm1 ## x3 y3 x4 y4 movaps %xmm9,%xmm8 unpcklps %xmm10,%xmm9 ## x1 y1 x2 y2 unpckhps %xmm10,%xmm8 ## x3 y3 x4 y4 ## update fjx and fjy addps %xmm9,%xmm0 addps %xmm8,%xmm1 movlps %xmm0,(%rsi,%rax,4) movlps %xmm1,(%rsi,%rcx,4) movhps %xmm0,(%rsi,%rbx,4) movhps %xmm1,(%rsi,%rdx,4) ## xmm11: fjz1 fjz2 fjz3 fjz4 pshufd $1,%xmm11,%xmm10 ## fjz2 - - - movhlps %xmm11,%xmm9 ## fjz3 - - - pshufd $3,%xmm11,%xmm8 ## fjz4 - - - addss 8(%rsi,%rax,4),%xmm11 addss 8(%rsi,%rbx,4),%xmm10 addss 8(%rsi,%rcx,4),%xmm9 addss 8(%rsi,%rdx,4),%xmm8 movss %xmm11,8(%rsi,%rax,4) movss %xmm10,8(%rsi,%rbx,4) movss %xmm9,8(%rsi,%rcx,4) movss %xmm8,8(%rsi,%rdx,4) ## should we do one more iteration? subl $4,nb130_innerk(%rsp) jl _nb_kernel130_x86_64_sse.nb130_finish_inner jmp _nb_kernel130_x86_64_sse.nb130_unroll_loop_nb_kernel130_x86_64_sse.nb130_finish_inner: ## check if at least two particles remain addl $4,nb130_innerk(%rsp) movl nb130_innerk(%rsp),%edx andl $2,%edx jnz _nb_kernel130_x86_64_sse.nb130_dopair jmp _nb_kernel130_x86_64_sse.nb130_checksingle_nb_kernel130_x86_64_sse.nb130_dopair: movq nb130_innerjjnr(%rsp),%rcx movl (%rcx),%eax movl 4(%rcx),%ebx addq $8,nb130_innerjjnr(%rsp) movq nb130_charge(%rbp),%rsi movss (%rsi,%rax,4),%xmm0 movss (%rsi,%rbx,4),%xmm2 unpcklps %xmm2,%xmm0 ## jqa jqb mulps nb130_iq(%rsp),%xmm0 movaps %xmm0,nb130_qq(%rsp) movq nb130_type(%rbp),%rsi ## vdw parameters movl (%rsi,%rax,4),%r12d movl (%rsi,%rbx,4),%r13d shll %r12d shll %r13d movl nb130_ntia(%rsp),%edi addl %edi,%r12d addl %edi,%r13d movq nb130_vdwparam(%rbp),%rsi movlps (%rsi,%r12,4),%xmm3 movhps (%rsi,%r13,4),%xmm3 xorps %xmm7,%xmm7 movaps %xmm3,%xmm0 shufps $136,%xmm7,%xmm0 ## 10001000 shufps $221,%xmm7,%xmm3 ## 11011101 movaps %xmm0,nb130_c6(%rsp) movaps %xmm3,nb130_c12(%rsp) lea (%rax,%rax,2),%rax ## replace jnr with j3 lea (%rbx,%rbx,2),%rbx ## load coordinates movq nb130_pos(%rbp),%rdi movlps (%rdi,%rax,4),%xmm1 ## x1 y1 - - movlps (%rdi,%rbx,4),%xmm2 ## x2 y2 - - movss 8(%rdi,%rax,4),%xmm5 ## z1 - - - movss 8(%rdi,%rbx,4),%xmm6 ## z2 - - - unpcklps %xmm2,%xmm1 ## x1 x2 y1 y2 movhlps %xmm1,%xmm2 ## y1 y2 - - unpcklps %xmm6,%xmm5 ## z1 z2 - - ## calc dr subps nb130_ix(%rsp),%xmm1 subps nb130_iy(%rsp),%xmm2 subps nb130_iz(%rsp),%xmm5 ## store dr movaps %xmm1,nb130_dx(%rsp) movaps %xmm2,nb130_dy(%rsp) movaps %xmm5,nb130_dz(%rsp) ## square it mulps %xmm1,%xmm1 mulps %xmm2,%xmm2 mulps %xmm5,%xmm5 addps %xmm2,%xmm1 addps %xmm5,%xmm1 ## rsq in xmm1 ## calculate rinv=1/sqrt(rsq) rsqrtps %xmm1,%xmm5 movaps %xmm5,%xmm2 mulps %xmm5,%xmm5 movaps nb130_three(%rsp),%xmm4 mulps %xmm1,%xmm5 ## rsq*lu*lu subps %xmm5,%xmm4 ## 30-rsq*lu*lu mulps %xmm2,%xmm4 mulps nb130_half(%rsp),%xmm4 movaps %xmm4,%xmm2 mulps %xmm4,%xmm1 ## xmm2=rinv ## xmm1=r mulps nb130_tsc(%rsp),%xmm1 ## rtab ## truncate and convert to integers cvttps2dq %xmm1,%xmm5 ## convert back to float cvtdq2ps %xmm5,%xmm4 ## multiply by 8 pslld $3,%xmm5 ## calculate eps subps %xmm4,%xmm1 ## move to integer registers movd %xmm5,%r8d pshufd $1,%xmm5,%xmm5 movd %xmm5,%r9d ## xmm1=eps ## xmm2=rinv movq nb130_VFtab(%rbp),%rsi ## calculate LJ table movlps (%rsi,%r8,4),%xmm4 movlps (%rsi,%r9,4),%xmm5 unpcklps %xmm5,%xmm4 movhlps %xmm4,%xmm5 movlps 8(%rsi,%r8,4),%xmm6 movlps 8(%rsi,%r9,4),%xmm7 movaps %xmm2,%xmm0 ## rinv mulps nb130_qq(%rsp),%xmm2 ## vcoul=rinv*qq movaps %xmm2,%xmm3 ## copy of vcoul (to calc fscal) mulps %xmm0,%xmm3 ## vcoul*rinv unpcklps %xmm7,%xmm6 movhlps %xmm6,%xmm7 movlps 16(%rsi,%r8,4),%xmm8 movlps 16(%rsi,%r9,4),%xmm9 unpcklps %xmm9,%xmm8 movhlps %xmm8,%xmm9 addps nb130_vctot(%rsp),%xmm2 movlps %xmm2,nb130_vctot(%rsp) movlps 24(%rsi,%r8,4),%xmm10 movlps 24(%rsi,%r9,4),%xmm11 unpcklps %xmm11,%xmm10 movhlps %xmm10,%xmm11 ## 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 movaps nb130_c6(%rsp),%xmm12 movaps nb130_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 nb130_Vvdwtot(%rsp),%xmm5 movlps %xmm5,nb130_Vvdwtot(%rsp) mulps %xmm12,%xmm7 ## FF*c6 = fnb6 mulps %xmm13,%xmm11 ## FF*c12 = fnb12 addps %xmm11,%xmm7 mulps nb130_tsc(%rsp),%xmm7 subps %xmm7,%xmm3 mulps %xmm0,%xmm3 ## fscal movaps %xmm3,%xmm9 movaps %xmm3,%xmm10 movaps %xmm3,%xmm11 xorps %xmm8,%xmm8 movaps nb130_fix(%rsp),%xmm12 movaps nb130_fiy(%rsp),%xmm13 movaps nb130_fiz(%rsp),%xmm14 mulps nb130_dx(%rsp),%xmm9 mulps nb130_dy(%rsp),%xmm10 mulps nb130_dz(%rsp),%xmm11 movlhps %xmm8,%xmm9 movlhps %xmm8,%xmm10 movlhps %xmm8,%xmm11 ## accumulate i forces addps %xmm9,%xmm12 addps %xmm10,%xmm13 addps %xmm11,%xmm14 movaps %xmm12,nb130_fix(%rsp) movaps %xmm13,nb130_fiy(%rsp) movaps %xmm14,nb130_fiz(%rsp) movq nb130_faction(%rbp),%rsi ## the fj's - start by accumulating x & y forces from memory movlps (%rsi,%rax,4),%xmm0 ## x1 y1 - - movhps (%rsi,%rbx,4),%xmm0 ## x1 y1 x2 y2 unpcklps %xmm10,%xmm9 ## x1 y1 x2 y2 addps %xmm9,%xmm0 movlps %xmm0,(%rsi,%rax,4) movhps %xmm0,(%rsi,%rbx,4) ## z forces pshufd $1,%xmm11,%xmm8 addss 8(%rsi,%rax,4),%xmm11 addss 8(%rsi,%rbx,4),%xmm8 movss %xmm11,8(%rsi,%rax,4) movss %xmm8,8(%rsi,%rbx,4)_nb_kernel130_x86_64_sse.nb130_checksingle: movl nb130_innerk(%rsp),%edx andl $1,%edx jnz _nb_kernel130_x86_64_sse.nb130_dosingle jmp _nb_kernel130_x86_64_sse.nb130_updateouterdata_nb_kernel130_x86_64_sse.nb130_dosingle: movq nb130_pos(%rbp),%rdi movq nb130_innerjjnr(%rsp),%rcx movl (%rcx),%eax movq nb130_charge(%rbp),%rsi movss (%rsi,%rax,4),%xmm0 mulss nb130_iq(%rsp),%xmm0 movaps %xmm0,nb130_qq(%rsp) movq nb130_type(%rbp),%rsi ## vdw parameters movl (%rsi,%rax,4),%r12d shll %r12d movl nb130_ntia(%rsp),%edi addl %edi,%r12d movq nb130_vdwparam(%rbp),%rsi movss (%rsi,%r12,4),%xmm0 movss 4(%rsi,%r12,4),%xmm3 movaps %xmm0,nb130_c6(%rsp) movaps %xmm3,nb130_c12(%rsp) lea (%rax,%rax,2),%rax ## replace jnr with j3 movq nb130_pos(%rbp),%rdi ## load coordinates movss (%rdi,%rax,4),%xmm1 movss 4(%rdi,%rax,4),%xmm2 movss 8(%rdi,%rax,4),%xmm5 ## calc dr subss nb130_ix(%rsp),%xmm1 subss nb130_iy(%rsp),%xmm2 subss nb130_iz(%rsp),%xmm5 ## store dr movaps %xmm1,nb130_dx(%rsp) movaps %xmm2,nb130_dy(%rsp) movaps %xmm5,nb130_dz(%rsp) ## square it mulss %xmm1,%xmm1 mulss %xmm2,%xmm2 mulss %xmm5,%xmm5 addss %xmm2,%xmm1 addss %xmm5,%xmm1 ## rsq in xmm1 ## calculate rinv=1/sqrt(rsq) rsqrtss %xmm1,%xmm5 movaps %xmm5,%xmm2 mulss %xmm5,%xmm5 movaps nb130_three(%rsp),%xmm4 mulss %xmm1,%xmm5 ## rsq*lu*lu subss %xmm5,%xmm4 ## 30-rsq*lu*lu mulss %xmm2,%xmm4 mulss nb130_half(%rsp),%xmm4 movaps %xmm4,%xmm2 mulss %xmm4,%xmm1 ## xmm2=rinv ## xmm1=r mulss nb130_tsc(%rsp),%xmm1 ## rtab ## truncate and convert to integers cvttss2si %xmm1,%r8d ## convert back to float cvtsi2ss %r8d,%xmm4 ## multiply by 8 shll $3,%r8d ## calculate eps subss %xmm4,%xmm1 ## xmm1=eps ## xmm2=rinv movq nb130_VFtab(%rbp),%rsi ## calculate LJ table movss (%rsi,%r8,4),%xmm4 movss 4(%rsi,%r8,4),%xmm5 movss 8(%rsi,%r8,4),%xmm6 movss 12(%rsi,%r8,4),%xmm7 movss 16(%rsi,%r8,4),%xmm8 movss 20(%rsi,%r8,4),%xmm9 movss 24(%rsi,%r8,4),%xmm10 movss 28(%rsi,%r8,4),%xmm11 ## dispersion table in xmm4-xmm7, repulsion table in xmm8-xmm11 ## coulomb interaction movaps %xmm2,%xmm0 ## rinv mulss nb130_qq(%rsp),%xmm2 ## vcoul=rinv*qq movaps %xmm2,%xmm3 ## copy of vcoul (to calc fscal) mulss %xmm0,%xmm3 ## vcoul*rinv addss nb130_vctot(%rsp),%xmm2 movss %xmm2,nb130_vctot(%rsp) ## calculate table interaction mulss %xmm1,%xmm7 ## Heps mulss %xmm1,%xmm11 mulss %xmm1,%xmm6 ## Geps mulss %xmm1,%xmm10 mulss %xmm1,%xmm7 ## Heps2 mulss %xmm1,%xmm11 addss %xmm6,%xmm5 ## F+Geps addss %xmm10,%xmm9 addss %xmm7,%xmm5 ## F+Geps+Heps2 = Fp addss %xmm11,%xmm9 addss %xmm7,%xmm7 ## 2*Heps2 addss %xmm11,%xmm11 addss %xmm6,%xmm7 ## 2*Heps2+Geps addss %xmm10,%xmm11 addss %xmm5,%xmm7 ## FF = Fp + 2*Heps2 + Geps addss %xmm9,%xmm11 mulss %xmm1,%xmm5 ## eps*Fp mulss %xmm1,%xmm9 movaps nb130_c6(%rsp),%xmm12 movaps nb130_c12(%rsp),%xmm13 addss %xmm4,%xmm5 ## VV addss %xmm8,%xmm9 mulss %xmm12,%xmm5 ## VV*c6 = vnb6 mulss %xmm13,%xmm9 ## VV*c12 = vnb12 addss %xmm9,%xmm5 addss nb130_Vvdwtot(%rsp),%xmm5 movss %xmm5,nb130_Vvdwtot(%rsp) mulss %xmm12,%xmm7 ## FF*c6 = fnb6 mulss %xmm13,%xmm11 ## FF*c12 = fnb12 addss %xmm11,%xmm7 mulss nb130_tsc(%rsp),%xmm7 subss %xmm7,%xmm3 mulss %xmm0,%xmm3 ## fscal movaps %xmm3,%xmm9 movaps %xmm3,%xmm10 movaps %xmm3,%xmm11 movaps nb130_fix(%rsp),%xmm12 movaps nb130_fiy(%rsp),%xmm13 movaps nb130_fiz(%rsp),%xmm14 mulss nb130_dx(%rsp),%xmm9 mulss nb130_dy(%rsp),%xmm10 mulss nb130_dz(%rsp),%xmm11 ## accumulate i forces addss %xmm9,%xmm12 addss %xmm10,%xmm13 addss %xmm11,%xmm14 movaps %xmm12,nb130_fix(%rsp) movaps %xmm13,nb130_fiy(%rsp) movaps %xmm14,nb130_fiz(%rsp) movq nb130_faction(%rbp),%rsi ## add to j forces addss (%rsi,%rax,4),%xmm9 addss 4(%rsi,%rax,4),%xmm10 addss 8(%rsi,%rax,4),%xmm11 movss %xmm9,(%rsi,%rax,4) movss %xmm10,4(%rsi,%rax,4) movss %xmm11,8(%rsi,%rax,4)_nb_kernel130_x86_64_sse.nb130_updateouterdata: movl nb130_ii3(%rsp),%ecx movq nb130_faction(%rbp),%rdi movq nb130_fshift(%rbp),%rsi movl nb130_is3(%rsp),%edx ## accumulate i forces in xmm0, xmm1, xmm2 movaps nb130_fix(%rsp),%xmm0 movaps nb130_fiy(%rsp),%xmm1⌨️ 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?