nb_kernel212_x86_64_sse2.intel_syntax.s

来自「最著名最快的分子模拟软件」· S 代码 · 共 2,308 行 · 第 1/5 页

    movapd xmm8, xmm2
    
    subpd xmm0, [rsp + nb212_ixO]
    subpd xmm1, [rsp + nb212_iyO]
    subpd xmm2, [rsp + nb212_izO]
    subpd xmm3, [rsp + nb212_ixH1]
    subpd xmm4, [rsp + nb212_iyH1]
    subpd xmm5, [rsp + nb212_izH1]
    subpd xmm6, [rsp + nb212_ixH2]
    subpd xmm7, [rsp + nb212_iyH2]
    subpd xmm8, [rsp + nb212_izH2]
    
	movapd [rsp + nb212_dxOH2], xmm0
	movapd [rsp + nb212_dyOH2], xmm1
	movapd [rsp + nb212_dzOH2], xmm2
	mulpd  xmm0, xmm0
	mulpd  xmm1, xmm1
	mulpd  xmm2, xmm2
	movapd [rsp + nb212_dxH1H2], xmm3
	movapd [rsp + nb212_dyH1H2], xmm4
	movapd [rsp + nb212_dzH1H2], xmm5
	mulpd  xmm3, xmm3
	mulpd  xmm4, xmm4
	mulpd  xmm5, xmm5
	movapd [rsp + nb212_dxH2H2], xmm6
	movapd [rsp + nb212_dyH2H2], xmm7
	movapd [rsp + nb212_dzH2H2], xmm8
	mulpd  xmm6, xmm6
	mulpd  xmm7, xmm7
	mulpd  xmm8, xmm8
	addpd  xmm0, xmm1
	addpd  xmm0, xmm2
	addpd  xmm3, xmm4
	addpd  xmm3, xmm5
    addpd  xmm6, xmm7
    addpd  xmm6, xmm8

	;# start doing invsqrt for jH2 atoms
    cvtpd2ps xmm1, xmm0
    cvtpd2ps xmm4, xmm3
    cvtpd2ps xmm7, xmm6
	rsqrtps xmm1, xmm1
	rsqrtps xmm4, xmm4
    rsqrtps xmm7, xmm7
    cvtps2pd xmm1, xmm1
    cvtps2pd xmm4, xmm4
    cvtps2pd xmm7, xmm7
	
	movapd  xmm2, xmm1
	movapd  xmm5, xmm4
    movapd  xmm8, xmm7
    
	mulpd   xmm1, xmm1 ;# lu*lu
	mulpd   xmm4, xmm4 ;# lu*lu
    mulpd   xmm7, xmm7 ;# lu*lu
		
	movapd  xmm9, [rsp + nb212_three]
	movapd  xmm10, xmm9
    movapd  xmm11, xmm9

	mulpd   xmm1, xmm0 ;# rsq*lu*lu
	mulpd   xmm4, xmm3 ;# rsq*lu*lu 
    mulpd   xmm7, xmm6 ;# rsq*lu*lu
	
	subpd   xmm9, xmm1
	subpd   xmm10, xmm4
    subpd   xmm11, xmm7 ;# 3-rsq*lu*lu

	mulpd   xmm9, xmm2
	mulpd   xmm10, xmm5
    mulpd   xmm11, xmm8 ;# lu*(3-rsq*lu*lu)

	movapd  xmm15, [rsp + nb212_half]
	mulpd   xmm9, xmm15  ;# first iteration for rinvOH2 
	mulpd   xmm10, xmm15 ;# first iteration for rinvH1H2
    mulpd   xmm11, xmm15 ;# first iteration for rinvH2H2

    ;# second iteration step    
	movapd  xmm2, xmm9
	movapd  xmm5, xmm10
    movapd  xmm8, xmm11
    
	mulpd   xmm2, xmm2 ;# lu*lu
	mulpd   xmm5, xmm5 ;# lu*lu
    mulpd   xmm8, xmm8 ;# lu*lu
		
	movapd  xmm1, [rsp + nb212_three]
	movapd  xmm4, xmm1
    movapd  xmm7, xmm1

	mulpd   xmm2, xmm0 ;# rsq*lu*lu
	mulpd   xmm5, xmm3 ;# rsq*lu*lu 
    mulpd   xmm8, xmm6 ;# rsq*lu*lu
	
	subpd   xmm1, xmm2
	subpd   xmm4, xmm5
    subpd   xmm7, xmm8 ;# 3-rsq*lu*lu

	mulpd   xmm9, xmm1
	mulpd   xmm10, xmm4
    mulpd   xmm11, xmm7 ;# lu*(3-rsq*lu*lu)

	movapd  xmm15, [rsp + nb212_half]
	mulpd   xmm9, xmm15  ;#  rinvOH2
	mulpd   xmm10, xmm15 ;#   rinvH1H2
    mulpd   xmm11, xmm15 ;#   rinvH2H2
	
	
	;# H2 interactions 
    ;# rsq in xmm0,xmm3,xmm6  
    ;# rinv in xmm9, xmm10, xmm11

    movapd xmm1, xmm9 ;# copy of rinv
    movapd xmm4, xmm10
    movapd xmm7, xmm11
    movapd xmm2, [rsp + nb212_krf]    
    mulpd  xmm9, xmm9   ;# rinvsq
    mulpd  xmm10, xmm10
    mulpd  xmm11, xmm11
    mulpd  xmm0, xmm2  ;# k*rsq
    mulpd  xmm3, xmm2
    mulpd  xmm6, xmm2
    movapd xmm2, xmm0 ;# copy of k*rsq
    movapd xmm5, xmm3
    movapd xmm8, xmm6
    addpd  xmm2, xmm1  ;# rinv+krsq
    addpd  xmm5, xmm4
    addpd  xmm8, xmm7
    movapd xmm14, [rsp + nb212_crf]
    subpd  xmm2, xmm14   ;# rinv+krsq-crf
    subpd  xmm5, xmm14
    subpd  xmm8, xmm14
    movapd xmm12, [rsp + nb212_qqOH]
    movapd xmm13, [rsp + nb212_qqHH]    
    mulpd  xmm2, xmm12 ;# xmm6=voul=qq*(rinv+ krsq-crf)
    mulpd  xmm5, xmm13 ;# xmm6=voul=qq*(rinv+ krsq-crf)
    mulpd  xmm8, xmm13 ;# xmm6=voul=qq*(rinv+ krsq-crf)
    addpd  xmm0, xmm0 ;# 2*krsq
    addpd  xmm3, xmm3 
    addpd  xmm6, xmm6 
    subpd  xmm1, xmm0 ;# rinv-2*krsq
    subpd  xmm4, xmm3
    subpd  xmm7, xmm6
    mulpd  xmm1, xmm12   ;# (rinv-2*krsq)*qq
    mulpd  xmm4, xmm13
    mulpd  xmm7, xmm13
    addpd  xmm2, [rsp + nb212_vctot]
    addpd  xmm5, xmm8
    addpd  xmm2, xmm5
    movapd [rsp + nb212_vctot], xmm2
    
    mulpd  xmm9, xmm1   ;# fscal
    mulpd  xmm10, xmm4
    mulpd  xmm11, xmm7

    ;# move j H2 forces to xmm0-xmm2
	movlpd xmm0, [rdi + rax*8 + 48]
	movlpd xmm1, [rdi + rax*8 + 56]
	movlpd xmm2, [rdi + rax*8 + 64]
	movhpd xmm0, [rdi + rbx*8 + 48]
	movhpd xmm1, [rdi + rbx*8 + 56]
	movhpd xmm2, [rdi + rbx*8 + 64]

    movapd xmm7, xmm9
    movapd xmm8, xmm9
    movapd xmm13, xmm11
    movapd xmm14, xmm11
    movapd xmm15, xmm11
    movapd xmm11, xmm10
    movapd xmm12, xmm10

	mulpd xmm7, [rsp + nb212_dxOH2]
	mulpd xmm8, [rsp + nb212_dyOH2]
	mulpd xmm9, [rsp + nb212_dzOH2]
	mulpd xmm10, [rsp + nb212_dxH1H2]
	mulpd xmm11, [rsp + nb212_dyH1H2]
	mulpd xmm12, [rsp + nb212_dzH1H2]
	mulpd xmm13, [rsp + nb212_dxH2H2]
	mulpd xmm14, [rsp + nb212_dyH2H2]
	mulpd xmm15, [rsp + nb212_dzH2H2]

    addpd xmm0, xmm7
    addpd xmm1, xmm8
    addpd xmm2, xmm9
    addpd xmm7, [rsp + nb212_fixO]
    addpd xmm8, [rsp + nb212_fiyO]
    addpd xmm9, [rsp + nb212_fizO]

    addpd xmm0, xmm10
    addpd xmm1, xmm11
    addpd xmm2, xmm12
    addpd xmm10, [rsp + nb212_fixH1]
    addpd xmm11, [rsp + nb212_fiyH1]
    addpd xmm12, [rsp + nb212_fizH1]

    addpd xmm0, xmm13
    addpd xmm1, xmm14
    addpd xmm2, xmm15
    addpd xmm13, [rsp + nb212_fixH2]
    addpd xmm14, [rsp + nb212_fiyH2]
    addpd xmm15, [rsp + nb212_fizH2]

    movapd [rsp + nb212_fixO], xmm7
    movapd [rsp + nb212_fiyO], xmm8
    movapd [rsp + nb212_fizO], xmm9
    movapd [rsp + nb212_fixH1], xmm10
    movapd [rsp + nb212_fiyH1], xmm11
    movapd [rsp + nb212_fizH1], xmm12
    movapd [rsp + nb212_fixH2], xmm13
    movapd [rsp + nb212_fiyH2], xmm14
    movapd [rsp + nb212_fizH2], xmm15
   
    ;# store back j H2 forces from xmm0-xmm2
	movlpd [rdi + rax*8 + 48], xmm0
	movlpd [rdi + rax*8 + 56], xmm1
	movlpd [rdi + rax*8 + 64], xmm2
	movhpd [rdi + rbx*8 + 48], xmm0
	movhpd [rdi + rbx*8 + 56], xmm1
	movhpd [rdi + rbx*8 + 64], xmm2
	
	;# should we do one more iteration? 
	sub dword ptr [rsp + nb212_innerk],  2
	jl    .nb212_checksingle
	jmp   .nb212_unroll_loop
.nb212_checksingle:
	mov   edx, [rsp + nb212_innerk]
	and   edx, 1
	jnz   .nb212_dosingle
	jmp   .nb212_updateouterdata
.nb212_dosingle:
	mov   rdx, [rsp + nb212_innerjjnr]     ;# pointer to jjnr[k] 
	mov   eax, [rdx]
	
	mov rsi, [rbp + nb212_pos]
	lea   rax, [rax + rax*2]  

	
	;# move j O coordinates to local temp variables 
    movsd xmm0, [rsi + rax*8] 
    movsd xmm1, [rsi + rax*8 + 8] 
    movsd xmm2, [rsi + rax*8 + 16] 

    ;# xmm0 = Ox
    ;# xmm1 = Oy
    ;# xmm2 = Oz
        
    movsd xmm3, xmm0
    movsd xmm4, xmm1
    movsd xmm5, xmm2
    movsd xmm6, xmm0
    movsd xmm7, xmm1
    movsd xmm8, xmm2
    
    subsd xmm0, [rsp + nb212_ixO]
    subsd xmm1, [rsp + nb212_iyO]
    subsd xmm2, [rsp + nb212_izO]
    subsd xmm3, [rsp + nb212_ixH1]
    subsd xmm4, [rsp + nb212_iyH1]
    subsd xmm5, [rsp + nb212_izH1]
    subsd xmm6, [rsp + nb212_ixH2]
    subsd xmm7, [rsp + nb212_iyH2]
    subsd xmm8, [rsp + nb212_izH2]
    
	movsd [rsp + nb212_dxOO], xmm0
	movsd [rsp + nb212_dyOO], xmm1
	movsd [rsp + nb212_dzOO], xmm2
	mulsd  xmm0, xmm0
	mulsd  xmm1, xmm1
	mulsd  xmm2, xmm2
	movsd [rsp + nb212_dxH1O], xmm3
	movsd [rsp + nb212_dyH1O], xmm4
	movsd [rsp + nb212_dzH1O], xmm5
	mulsd  xmm3, xmm3
	mulsd  xmm4, xmm4
	mulsd  xmm5, xmm5
	movsd [rsp + nb212_dxH2O], xmm6
	movsd [rsp + nb212_dyH2O], xmm7
	movsd [rsp + nb212_dzH2O], xmm8
	mulsd  xmm6, xmm6
	mulsd  xmm7, xmm7
	mulsd  xmm8, xmm8
	addsd  xmm0, xmm1
	addsd  xmm0, xmm2
	addsd  xmm3, xmm4
	addsd  xmm3, xmm5
    addsd  xmm6, xmm7
    addsd  xmm6, xmm8

	;# start doing invsqrt for jO atoms
    cvtsd2ss xmm1, xmm0
    cvtsd2ss xmm4, xmm3
    cvtsd2ss xmm7, xmm6
	rsqrtss xmm1, xmm1
	rsqrtss xmm4, xmm4
    rsqrtss xmm7, xmm7
    cvtss2sd xmm1, xmm1
    cvtss2sd xmm4, xmm4
    cvtss2sd xmm7, xmm7
	
	movsd  xmm2, xmm1
	movsd  xmm5, xmm4
    movsd  xmm8, xmm7
    
	mulsd   xmm1, xmm1 ;# lu*lu
	mulsd   xmm4, xmm4 ;# lu*lu
    mulsd   xmm7, xmm7 ;# lu*lu
		
	movsd  xmm9, [rsp + nb212_three]
	movsd  xmm10, xmm9
    movsd  xmm11, xmm9

	mulsd   xmm1, xmm0 ;# rsq*lu*lu
	mulsd   xmm4, xmm3 ;# rsq*lu*lu 
    mulsd   xmm7, xmm6 ;# rsq*lu*lu
	
	subsd   xmm9, xmm1
	subsd   xmm10, xmm4
    subsd   xmm11, xmm7 ;# 3-rsq*lu*lu

	mulsd   xmm9, xmm2
	mulsd   xmm10, xmm5
    mulsd   xmm11, xmm8 ;# lu*(3-rsq*lu*lu)

	movsd  xmm15, [rsp + nb212_half]
	mulsd   xmm9, xmm15  ;# first iteration for rinvOO 
	mulsd   xmm10, xmm15 ;# first iteration for rinvH1O
    mulsd   xmm11, xmm15 ;# first iteration for rinvH2O	

    ;# second iteration step    
	movsd  xmm2, xmm9
	movsd  xmm5, xmm10
    movsd  xmm8, xmm11
    
	mulsd   xmm2, xmm2 ;# lu*lu
	mulsd   xmm5, xmm5 ;# lu*lu
    mulsd   xmm8, xmm8 ;# lu*lu
		
	movsd  xmm1, [rsp + nb212_three]
	movsd  xmm4, xmm1
    movsd  xmm7, xmm1

	mulsd   xmm2, xmm0 ;# rsq*lu*lu
	mulsd   xmm5, xmm3 ;# rsq*lu*lu 
    mulsd   xmm8, xmm6 ;# rsq*lu*lu
	
	subsd   xmm1, xmm2
	subsd   xmm4, xmm5
    subsd   xmm7, xmm8 ;# 3-rsq*lu*lu

	mulsd   xmm9, xmm1
	mulsd   xmm10, xmm4
    mulsd   xmm11, xmm7 ;# lu*(3-rsq*lu*lu)

	movsd  xmm15, [rsp + nb212_half]
	mulsd   xmm9, xmm15  ;#  rinvOO 
	mulsd   xmm10, xmm15 ;#   rinvH1O
    mulsd   xmm11, xmm15 ;#   rinvH2O
	
	;# O interactions 
    ;# rsq in xmm0,xmm3,xmm6  
    ;# rinv in xmm9, xmm10, xmm11
    movsd xmm1, xmm9 ;# copy of rinv
    movsd xmm4, xmm10
    movsd xmm7, xmm11
    movsd xmm2, [rsp + nb212_krf]    
    mulsd  xmm9, xmm9   ;# rinvsq
    mulsd  xmm10, xmm10
    mulsd  xmm11, xmm11
    mulsd  xmm0, xmm2  ;# k*rsq
    mulsd  xmm3, xmm2
    mulsd  xmm6, xmm2
    movsd xmm2, xmm0 ;# copy of k*rsq
    movsd xmm5, xmm3
    movsd xmm8, xmm6
    addsd  xmm2, xmm1  ;# rinv+krsq
    addsd  xmm5, xmm4
    addsd  xmm8, xmm7
    movsd xmm12, xmm9
    mulsd  xmm12, xmm12 ;# rinv4
    mulsd  xmm12, xmm9  ;# rinv6
    subsd  xmm2, [rsp + nb212_crf]   ;# rinv+krsq-crf
    subsd  xmm5, [rsp + nb212_crf]
    subsd  xmm8, [rsp + nb212_crf]   
    mulsd  xmm2, [rsp + nb212_qqOO] ;# voul=qq*(rinv+ krsq-crf)
    mulsd  xmm5, [rsp + nb212_qqOH] ;# voul=qq*(rinv+ krsq-crf)
    mulsd  xmm8, [rsp + nb212_qqOH] ;# voul=qq*(rinv+ krsq-crf)
    addsd  xmm0, xmm0 ;# 2*krsq
    addsd  xmm3, xmm3 
    addsd  xmm6, xmm6 
    subsd  xmm1, xmm0 ;# rinv-2*krsq
    subsd  xmm4, xmm3
    subsd  xmm7, xmm6
    movsd xmm13, xmm12 ;# rinv6
    mulsd xmm12, xmm12 ;# rinv12
	mulsd  xmm13, [rsp + nb212_c6]
	mulsd  xmm12, [rsp + nb212_c12]
    movsd xmm14, xmm12
    subsd  xmm14, xmm13
    mulsd  xmm1, [rsp + nb212_qqOO]   ;# (rinv-2*krsq)*qq
    mulsd  xmm4, [rsp + nb212_qqOH] 
    mulsd  xmm7, [rsp + nb212_qqOH] 
    addsd  xmm2, [rsp + nb212_vctot]
    addsd  xmm5, xmm8
    addsd  xmm2, xmm5
    movsd  [rsp + nb212_vctot], xmm2
 
   	addsd  xmm14, [rsp + nb212_Vvdwtot]
	mulsd  xmm13, [rsp + nb212_six]
	mulsd  xmm12, [rsp + nb212_twelve]
	movsd [rsp + nb212_Vvdwtot], xmm14
    subsd  xmm12, xmm13 ;# LJ fscal        

    addsd xmm1, xmm12
    
    mulsd  xmm9, xmm1   ;# fscal
    mulsd  xmm10, xmm4
    mulsd  xmm11, xmm7

    ;# move j O forces to xmm0-xmm2
	movsd xmm0, [rdi + rax*8]
	movsd xmm1, [rdi + rax*8 + 8]
	movsd xmm2, [rdi + rax*8 + 16]

    movsd xmm7, xmm9
    movsd xmm8, xmm9
    movsd xmm13, xmm11
    movsd xmm14, xmm11
    movsd xmm15, xmm11
    movsd xmm11, xmm10
    movsd xmm12, xmm10

	mulsd xmm7, [rsp + nb212_dxOO]
	mulsd xmm8, [rsp + nb212_dyOO]
	mulsd xmm9, [rsp + nb212_dzOO]
	mulsd xmm10, [rsp + nb212_dxH1O]
	mulsd xmm11, [rsp + nb212_dyH1O]
	mulsd xmm12, [rsp + nb212_dzH1O]
	mulsd xmm13, [rsp + nb212_dxH2O]
	mulsd xmm14, [rsp + nb212_dyH2O]
	mulsd xmm15, [rsp + nb212_dzH2O]

    addsd xmm0, xmm7
    addsd xmm1, xmm8
    addsd xmm2, xmm9
    addsd xmm7, [rsp + nb212_fixO]
    addsd xmm8, [rsp + nb212_fiyO]
    addsd xmm9, [rsp + nb212_fizO]

    addsd xmm0, xmm10
    addsd xmm1, xmm11
    addsd xmm2, xmm12
    addsd xmm10, [rsp + nb212_fixH1]
    addsd xmm11, [rsp + nb212_fiyH1]
    addsd xmm12, [rsp + nb212_fizH1]

    addsd xmm0, xmm13
    addsd xmm1, xmm14
    addsd xmm2, xmm15
    addsd xmm13, [rsp + nb212_fixH2]
    addsd xmm14, [rsp + nb212_fiyH2]
    addsd xmm15, [rsp + nb212_fizH2]