nb_kernel204_x86_64_sse.s

来自「最著名最快的分子模拟软件」· S 代码 · 共 2,119 行 · 第 1/5 页

        ## assume we have at least one i particle - start directly 
        movq  nb204nf_iinr(%rsp),%rcx           ## rcx = pointer into iinr[]    
        movl  (%rcx),%ebx               ## ebx =ii 

        movq  nb204nf_charge(%rbp),%rdx
        movss 4(%rdx,%rbx,4),%xmm3
        movss %xmm3,%xmm4
        movss 12(%rdx,%rbx,4),%xmm5
        movq nb204nf_p_facel(%rbp),%rsi
        movss (%rsi),%xmm0
        movss nb204nf_facel(%rsp),%xmm6
        mulss  %xmm3,%xmm3
        mulss  %xmm5,%xmm4
        mulss  %xmm5,%xmm5
        mulss  %xmm6,%xmm3
        mulss  %xmm6,%xmm4
        mulss  %xmm6,%xmm5
        shufps $0,%xmm3,%xmm3
        shufps $0,%xmm4,%xmm4
        shufps $0,%xmm5,%xmm5
        movaps %xmm3,nb204nf_qqHH(%rsp)
        movaps %xmm4,nb204nf_qqMH(%rsp)
        movaps %xmm5,nb204nf_qqMM(%rsp)

_nb_kernel204nf_x86_64_sse.nb204nf_threadloop: 
        movq  nb204nf_count(%rbp),%rsi            ## pointer to sync counter
        movl  (%rsi),%eax
_nb_kernel204nf_x86_64_sse.nb204nf_spinlock: 
        movl  %eax,%ebx                         ## ebx=*count=nn0
        addq  $1,%rbx                          ## rbx=nn1=nn0+10
        lock 
        cmpxchgl %ebx,(%rsi)                    ## write nn1 to *counter,
                                                ## if it hasnt changed.
                                                ## or reread *counter to eax.
        pause                                   ## -> better p4 performance
        jnz _nb_kernel204nf_x86_64_sse.nb204nf_spinlock

        ## if(nn1>nri) nn1=nri
        movl nb204nf_nri(%rsp),%ecx
        movl %ecx,%edx
        subl %ebx,%ecx
        cmovlel %edx,%ebx                       ## if(nn1>nri) nn1=nri
        ## Cleared the spinlock if we got here.
        ## eax contains nn0, ebx contains nn1.
        movl %eax,nb204nf_n(%rsp)
        movl %ebx,nb204nf_nn1(%rsp)
        subl %eax,%ebx                          ## calc number of outer lists
        movl %eax,%esi                          ## copy n to esi
        jg  _nb_kernel204nf_x86_64_sse.nb204nf_outerstart
        jmp _nb_kernel204nf_x86_64_sse.nb204nf_end

_nb_kernel204nf_x86_64_sse.nb204nf_outerstart: 
        ## ebx contains number of outer iterations
        addl nb204nf_nouter(%rsp),%ebx
        movl %ebx,nb204nf_nouter(%rsp)

_nb_kernel204nf_x86_64_sse.nb204nf_outer: 
        movq  nb204nf_shift(%rsp),%rax          ## rax = pointer into shift[] 
        movl  (%rax,%rsi,4),%ebx                ## rbx=shift[n] 

        lea  (%rbx,%rbx,2),%rbx        ## rbx=3*is 
        movl  %ebx,nb204nf_is3(%rsp)            ## store is3 

        movq  nb204nf_shiftvec(%rsp),%rax     ## rax = base of shiftvec[] 

        movss (%rax,%rbx,4),%xmm0
        movss 4(%rax,%rbx,4),%xmm1
        movss 8(%rax,%rbx,4),%xmm2

        movq  nb204nf_iinr(%rsp),%rcx           ## rcx = pointer into iinr[]    
        movl  (%rcx,%rsi,4),%ebx                ## ebx =ii 

        lea  (%rbx,%rbx,2),%rbx        ## rbx = 3*ii=ii3 
        movq  nb204nf_pos(%rbp),%rax    ## rax = base of pos[]  
        movl  %ebx,nb204nf_ii3(%rsp)

        movaps %xmm0,%xmm3
        movaps %xmm1,%xmm4
        movaps %xmm2,%xmm5
        addss 12(%rax,%rbx,4),%xmm3
        addss 16(%rax,%rbx,4),%xmm4
        addss 20(%rax,%rbx,4),%xmm5
        shufps $0,%xmm3,%xmm3
        shufps $0,%xmm4,%xmm4
        shufps $0,%xmm5,%xmm5
        movaps %xmm3,nb204nf_ixH1(%rsp)
        movaps %xmm4,nb204nf_iyH1(%rsp)
        movaps %xmm5,nb204nf_izH1(%rsp)

        movss %xmm0,%xmm3
        movss %xmm1,%xmm4
        movss %xmm2,%xmm5
        addss 24(%rax,%rbx,4),%xmm0
        addss 28(%rax,%rbx,4),%xmm1
        addss 32(%rax,%rbx,4),%xmm2
        addss 36(%rax,%rbx,4),%xmm3
        addss 40(%rax,%rbx,4),%xmm4
        addss 44(%rax,%rbx,4),%xmm5

        shufps $0,%xmm0,%xmm0
        shufps $0,%xmm1,%xmm1
        shufps $0,%xmm2,%xmm2
        shufps $0,%xmm3,%xmm3
        shufps $0,%xmm4,%xmm4
        shufps $0,%xmm5,%xmm5
        movaps %xmm0,nb204nf_ixH2(%rsp)
        movaps %xmm1,nb204nf_iyH2(%rsp)
        movaps %xmm2,nb204nf_izH2(%rsp)
        movaps %xmm3,nb204nf_ixM(%rsp)
        movaps %xmm4,nb204nf_iyM(%rsp)
        movaps %xmm5,nb204nf_izM(%rsp)

        ## clear vctot 
        xorps %xmm4,%xmm4
        movaps %xmm4,nb204nf_vctot(%rsp)

        movq  nb204nf_jindex(%rsp),%rax
        movl  (%rax,%rsi,4),%ecx                ## jindex[n] 
        movl  4(%rax,%rsi,4),%edx               ## jindex[n+1] 
        subl  %ecx,%edx                 ## number of innerloop atoms 

        movq  nb204nf_pos(%rbp),%rsi
        movq  nb204nf_faction(%rbp),%rdi
        movq  nb204nf_jjnr(%rsp),%rax
        shll  $2,%ecx
        addq  %rcx,%rax
        movq  %rax,nb204nf_innerjjnr(%rsp)      ## pointer to jjnr[nj0] 
        movl  %edx,%ecx
        subl  $4,%edx
        addl  nb204nf_ninner(%rsp),%ecx
        movl  %ecx,nb204nf_ninner(%rsp)
        addl  $0,%edx
        movl  %edx,nb204nf_innerk(%rsp)         ## number of innerloop atoms 
        jge   _nb_kernel204nf_x86_64_sse.nb204nf_unroll_loop
        jmp   _nb_kernel204nf_x86_64_sse.nb204nf_single_check
_nb_kernel204nf_x86_64_sse.nb204nf_unroll_loop: 
        ## quad-unroll innerloop here 
        movq  nb204nf_innerjjnr(%rsp),%rdx      ## pointer to jjnr[k] 

        movl  (%rdx),%eax
        movl  4(%rdx),%ebx
        movl  8(%rdx),%ecx
        movl  12(%rdx),%edx             ## eax-edx=jnr1-4 

        addq $16,nb204nf_innerjjnr(%rsp)             ## advance pointer (unrolled 4) 

        movq nb204nf_pos(%rbp),%rsi     ## base of pos[] 

        lea  (%rax,%rax,2),%rax        ## replace jnr with j3 
        lea  (%rbx,%rbx,2),%rbx
        lea  (%rcx,%rcx,2),%rcx        ## replace jnr with j3 
        lea  (%rdx,%rdx,2),%rdx

        ## move j coordinates to local temp variables 
        movlps 12(%rsi,%rax,4),%xmm2
        movlps 24(%rsi,%rax,4),%xmm3
        movlps 36(%rsi,%rax,4),%xmm4

        movlps 12(%rsi,%rbx,4),%xmm5
        movlps 24(%rsi,%rbx,4),%xmm6
        movlps 36(%rsi,%rbx,4),%xmm7

        movhps 12(%rsi,%rcx,4),%xmm2
        movhps 24(%rsi,%rcx,4),%xmm3
        movhps 36(%rsi,%rcx,4),%xmm4

        movhps 12(%rsi,%rdx,4),%xmm5
        movhps 24(%rsi,%rdx,4),%xmm6
        movhps 36(%rsi,%rdx,4),%xmm7

        movaps %xmm2,%xmm0
        movaps %xmm3,%xmm1
        unpcklps %xmm5,%xmm0
        unpcklps %xmm6,%xmm1
        unpckhps %xmm5,%xmm2
        unpckhps %xmm6,%xmm3
        movaps %xmm4,%xmm5
        movaps   %xmm0,%xmm6
        unpcklps %xmm7,%xmm4
        unpckhps %xmm7,%xmm5
        movaps   %xmm1,%xmm7
        movlhps  %xmm2,%xmm0
        movaps %xmm0,nb204nf_jxH1(%rsp)
        movhlps  %xmm6,%xmm2
        movaps %xmm2,nb204nf_jyH1(%rsp)
        movlhps  %xmm3,%xmm1
        movaps %xmm1,nb204nf_jxH2(%rsp)
        movhlps  %xmm7,%xmm3
        movaps   %xmm4,%xmm6
        movaps %xmm3,nb204nf_jyH2(%rsp)
        movlhps  %xmm5,%xmm4
        movaps %xmm4,nb204nf_jxM(%rsp)
        movhlps  %xmm6,%xmm5
        movaps %xmm5,nb204nf_jyM(%rsp)

        movss  20(%rsi,%rax,4),%xmm0
        movss  32(%rsi,%rax,4),%xmm1
        movss  44(%rsi,%rax,4),%xmm2

        movss  20(%rsi,%rcx,4),%xmm3
        movss  32(%rsi,%rcx,4),%xmm4
        movss  44(%rsi,%rcx,4),%xmm5

        movhps 16(%rsi,%rbx,4),%xmm0
        movhps 28(%rsi,%rbx,4),%xmm1
        movhps 40(%rsi,%rbx,4),%xmm2

        movhps 16(%rsi,%rdx,4),%xmm3
        movhps 28(%rsi,%rdx,4),%xmm4
        movhps 40(%rsi,%rdx,4),%xmm5

        shufps $204,%xmm3,%xmm0 ## 11001100
        shufps $204,%xmm4,%xmm1 ## 11001100
        shufps $204,%xmm5,%xmm2 ## 11001100
        movaps %xmm0,nb204nf_jzH1(%rsp)
        movaps %xmm1,nb204nf_jzH2(%rsp)
        movaps %xmm2,nb204nf_jzM(%rsp)

        movaps nb204nf_ixH1(%rsp),%xmm0
        movaps nb204nf_iyH1(%rsp),%xmm1
        movaps nb204nf_izH1(%rsp),%xmm2
        movaps nb204nf_ixH1(%rsp),%xmm3
        movaps nb204nf_iyH1(%rsp),%xmm4
        movaps nb204nf_izH1(%rsp),%xmm5
        subps  nb204nf_jxH1(%rsp),%xmm0
        subps  nb204nf_jyH1(%rsp),%xmm1
        subps  nb204nf_jzH1(%rsp),%xmm2
        subps  nb204nf_jxH2(%rsp),%xmm3
        subps  nb204nf_jyH2(%rsp),%xmm4
        subps  nb204nf_jzH2(%rsp),%xmm5
        mulps  %xmm0,%xmm0
        mulps  %xmm1,%xmm1
        mulps  %xmm2,%xmm2
        mulps  %xmm3,%xmm3
        mulps  %xmm4,%xmm4
        mulps  %xmm5,%xmm5
        addps  %xmm1,%xmm0
        addps  %xmm2,%xmm0
        addps  %xmm4,%xmm3
        addps  %xmm5,%xmm3
        movaps %xmm0,nb204nf_rsqH1H1(%rsp)
        movaps %xmm3,nb204nf_rsqH1H2(%rsp)

        movaps nb204nf_ixH1(%rsp),%xmm0
        movaps nb204nf_iyH1(%rsp),%xmm1
        movaps nb204nf_izH1(%rsp),%xmm2
        movaps nb204nf_ixH2(%rsp),%xmm3
        movaps nb204nf_iyH2(%rsp),%xmm4
        movaps nb204nf_izH2(%rsp),%xmm5
        subps  nb204nf_jxM(%rsp),%xmm0
        subps  nb204nf_jyM(%rsp),%xmm1
        subps  nb204nf_jzM(%rsp),%xmm2
        subps  nb204nf_jxH1(%rsp),%xmm3
        subps  nb204nf_jyH1(%rsp),%xmm4
        subps  nb204nf_jzH1(%rsp),%xmm5
        mulps  %xmm0,%xmm0
        mulps  %xmm1,%xmm1
        mulps  %xmm2,%xmm2
        mulps  %xmm3,%xmm3
        mulps  %xmm4,%xmm4
        mulps  %xmm5,%xmm5
        addps  %xmm1,%xmm0
        addps  %xmm2,%xmm0
        addps  %xmm4,%xmm3
        addps  %xmm5,%xmm3
        movaps %xmm0,nb204nf_rsqH1M(%rsp)
        movaps %xmm3,nb204nf_rsqH2H1(%rsp)

        movaps nb204nf_ixH2(%rsp),%xmm0
        movaps nb204nf_iyH2(%rsp),%xmm1
        movaps nb204nf_izH2(%rsp),%xmm2
        movaps nb204nf_ixH2(%rsp),%xmm3
        movaps nb204nf_iyH2(%rsp),%xmm4
        movaps nb204nf_izH2(%rsp),%xmm5
        subps  nb204nf_jxH2(%rsp),%xmm0
        subps  nb204nf_jyH2(%rsp),%xmm1
        subps  nb204nf_jzH2(%rsp),%xmm2
        subps  nb204nf_jxM(%rsp),%xmm3
        subps  nb204nf_jyM(%rsp),%xmm4
        subps  nb204nf_jzM(%rsp),%xmm5
        mulps  %xmm0,%xmm0
        mulps  %xmm1,%xmm1
        mulps  %xmm2,%xmm2
        mulps  %xmm3,%xmm3
        mulps  %xmm4,%xmm4
        mulps  %xmm5,%xmm5
        addps  %xmm1,%xmm0
        addps  %xmm2,%xmm0
        addps  %xmm4,%xmm3
        addps  %xmm5,%xmm3
        movaps %xmm0,nb204nf_rsqH2H2(%rsp)
        movaps %xmm3,nb204nf_rsqH2M(%rsp)

        movaps nb204nf_ixM(%rsp),%xmm0
        movaps nb204nf_iyM(%rsp),%xmm1
        movaps nb204nf_izM(%rsp),%xmm2
        movaps nb204nf_ixM(%rsp),%xmm3
        movaps nb204nf_iyM(%rsp),%xmm4
        movaps nb204nf_izM(%rsp),%xmm5
        subps  nb204nf_jxH1(%rsp),%xmm0
        subps  nb204nf_jyH1(%rsp),%xmm1
        subps  nb204nf_jzH1(%rsp),%xmm2
        subps  nb204nf_jxH2(%rsp),%xmm3
        subps  nb204nf_jyH2(%rsp),%xmm4
        subps  nb204nf_jzH2(%rsp),%xmm5
        mulps  %xmm0,%xmm0
        mulps  %xmm1,%xmm1
        mulps  %xmm2,%xmm2
        mulps  %xmm3,%xmm3
        mulps  %xmm4,%xmm4
        mulps  %xmm5,%xmm5
        addps  %xmm1,%xmm0
        addps  %xmm2,%xmm0
        addps  %xmm3,%xmm4
        addps  %xmm5,%xmm4
        movaps %xmm0,nb204nf_rsqMH1(%rsp)
        movaps %xmm4,nb204nf_rsqMH2(%rsp)

        movaps nb204nf_ixM(%rsp),%xmm0
        movaps nb204nf_iyM(%rsp),%xmm1
        movaps nb204nf_izM(%rsp),%xmm2
        subps  nb204nf_jxM(%rsp),%xmm0
        subps  nb204nf_jyM(%rsp),%xmm1
        subps  nb204nf_jzM(%rsp),%xmm2
        mulps %xmm0,%xmm0
        mulps %xmm1,%xmm1
        mulps %xmm2,%xmm2
        addps %xmm1,%xmm0
        addps %xmm2,%xmm0
        movaps %xmm0,nb204nf_rsqMM(%rsp)

        ## start doing invsqrt use rsq values in xmm0, xmm4 
        rsqrtps %xmm0,%xmm1
        rsqrtps %xmm4,%xmm5
        movaps  %xmm1,%xmm2
        movaps  %xmm5,%xmm6
        mulps   %xmm1,%xmm1
        mulps   %xmm5,%xmm5
        movaps  nb204nf_three(%rsp),%xmm3
        movaps  %xmm3,%xmm7
        mulps   %xmm0,%xmm1
        mulps   %xmm4,%xmm5
        subps   %xmm1,%xmm3
        subps   %xmm5,%xmm7
        mulps   %xmm2,%xmm3
        mulps   %xmm6,%xmm7
        mulps   nb204nf_half(%rsp),%xmm3   ## rinvH2H2 
        mulps   nb204nf_half(%rsp),%xmm7   ## rinvH2H1 
        movaps  %xmm3,nb204nf_rinvMM(%rsp)
        movaps  %xmm7,nb204nf_rinvMH2(%rsp)

        rsqrtps nb204nf_rsqH1H1(%rsp),%xmm1
        rsqrtps nb204nf_rsqH1H2(%rsp),%xmm5
        movaps  %xmm1,%xmm2
        movaps  %xmm5,%xmm6
        mulps   %xmm1,%xmm1
        mulps   %xmm5,%xmm5
        movaps  nb204nf_three(%rsp),%xmm3
        movaps  %xmm3,%xmm7
        mulps   nb204nf_rsqH1H1(%rsp),%xmm1
        mulps   nb204nf_rsqH1H2(%rsp),%xmm5
        subps   %xmm1,%xmm3
        subps   %xmm5,%xmm7
        mulps   %xmm2,%xmm3
        mulps   %xmm6,%xmm7
        mulps   nb204nf_half(%rsp),%xmm3
        mulps   nb204nf_half(%rsp),%xmm7
        movaps  %xmm3,nb204nf_rinvH1H1(%rsp)
        movaps  %xmm7,nb204nf_rinvH1H2(%rsp)

        rsqrtps nb204nf_rsqH1M(%rsp),%xmm1
        rsqrtps nb204nf_rsqH2H1(%rsp),%xmm5
        movaps  %xmm1,%xmm2
        movaps  %xmm5,%xmm6
        mulps   %xmm1,%xmm1
        mulps   %xmm5,%xmm5
        movaps  nb204nf_three(%rsp),%xmm3
        movaps  %xmm3,%xmm7
        mulps   nb204nf_rsqH1M(%rsp),%xmm1
        mulps   nb204nf_rsqH2H1(%rsp),%xmm5
        subps   %xmm1,%xmm3
        subps   %xmm5,%xmm7
        mulps   %xmm2,%xmm3
        mulps   %xmm6,%xmm7
        mulps   nb204nf_half(%rsp),%xmm3
        mulps   nb204nf_half(%rsp),%xmm7
        movaps  %xmm3,nb204nf_rinvH1M(%rsp)
        movaps  %xmm7,nb204nf_rinvH2H1(%rsp)

        rsqrtps nb204nf_rsqH2H2(%rsp),%xmm1
        rsqrtps nb204nf_rsqH2M(%rsp),%xmm5
        movaps  %xmm1,%xmm2
        movaps  %xmm5,%xmm6
        mulps   %xmm1,%xmm1
        mulps   %xmm5,%xmm5
        movaps  nb204nf_three(%rsp),%xmm3
        movaps  %xmm3,%xmm7
        mulps   nb204nf_rsqH2H2(%rsp),%xmm1
        mulps   nb204nf_rsqH2M(%rsp),%xmm5
        subps   %xmm1,%xmm3
        subps   %xmm5,%xmm7
        mulps   %xmm2,%xmm3
        mulps   %xmm6,%xmm7
        mulps   nb204nf_half(%rsp),%xmm3
        mulps   nb204nf_half(%rsp),%xmm7
        movaps  %xmm3,nb204nf_rinvH2H2(%rsp)
        movaps  %xmm7,nb204nf_rinvH2M(%rsp)

        rsqrtps nb204nf_rsqMH1(%rsp),%xmm1
        movaps  %xmm1,%xmm2
        mulps   %xmm1,%xmm1
        movaps  nb204nf_three(%rsp),%xmm3
        mulps   nb204nf_rsqMH1(%rsp),%xmm1
        subps   %xmm1,%xmm3
        mulps   %xmm2,%xmm3
        mulps   nb204nf_half(%rsp),%xmm3
        movaps  %xmm3,nb204nf_rinvMH1(%rsp)

        ## Coulomb interactions 
        ## add all H-H rsq in xmm2, H-M rsq xmm4
        ## H-H rinv in xmm0, H-M in xmm1
        movaps nb204nf_rinvH1H1(%rsp),%xmm0
        movaps nb204nf