nb_kernel102_x86_64_sse2.intel_syntax.s

来自「最著名最快的分子模拟软件」· S 代码 · 共 2,311 行 · 第 1/5 页

	
	;# H2 interactions 
    movapd xmm0, xmm9
    movapd xmm1, xmm10
    movapd xmm2, xmm11
    mulpd  xmm9, xmm9
    mulpd  xmm10, xmm10
    mulpd  xmm11, xmm11
    mulpd  xmm0, [rsp + nb102_qqOH] 
    mulpd  xmm1, [rsp + nb102_qqHH] 
    mulpd  xmm2, [rsp + nb102_qqHH] 
    mulpd  xmm9, xmm0
    mulpd  xmm10, xmm1
    mulpd  xmm11, xmm2
    
    addpd xmm0, [rsp + nb102_vctot] 
    addpd xmm1, xmm2
    addpd xmm0, xmm1
    movapd [rsp + nb102_vctot], xmm0
    
    ;# move j H2 forces to xmm0-xmm2
	mov   rdi, [rbp + nb102_faction]	
	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 + nb102_dxOH2]
	mulpd xmm8, [rsp + nb102_dyOH2]
	mulpd xmm9, [rsp + nb102_dzOH2]
	mulpd xmm10, [rsp + nb102_dxH1H2]
	mulpd xmm11, [rsp + nb102_dyH1H2]
	mulpd xmm12, [rsp + nb102_dzH1H2]
	mulpd xmm13, [rsp + nb102_dxH2H2]
	mulpd xmm14, [rsp + nb102_dyH2H2]
	mulpd xmm15, [rsp + nb102_dzH2H2]

    addpd xmm0, xmm7
    addpd xmm1, xmm8
    addpd xmm2, xmm9
    addpd xmm7, [rsp + nb102_fixO]
    addpd xmm8, [rsp + nb102_fiyO]
    addpd xmm9, [rsp + nb102_fizO]

    addpd xmm0, xmm10
    addpd xmm1, xmm11
    addpd xmm2, xmm12
    addpd xmm10, [rsp + nb102_fixH1]
    addpd xmm11, [rsp + nb102_fiyH1]
    addpd xmm12, [rsp + nb102_fizH1]

    addpd xmm0, xmm13
    addpd xmm1, xmm14
    addpd xmm2, xmm15
    addpd xmm13, [rsp + nb102_fixH2]
    addpd xmm14, [rsp + nb102_fiyH2]
    addpd xmm15, [rsp + nb102_fizH2]

    movapd [rsp + nb102_fixO], xmm7
    movapd [rsp + nb102_fiyO], xmm8
    movapd [rsp + nb102_fizO], xmm9
    movapd [rsp + nb102_fixH1], xmm10
    movapd [rsp + nb102_fiyH1], xmm11
    movapd [rsp + nb102_fizH1], xmm12
    movapd [rsp + nb102_fixH2], xmm13
    movapd [rsp + nb102_fiyH2], xmm14
    movapd [rsp + nb102_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 + nb102_innerk],  2
	jl    .nb102_checksingle
	jmp   .nb102_unroll_loop
.nb102_checksingle:
	mov   edx, [rsp + nb102_innerk]
	and   edx, 1
	jnz   .nb102_dosingle
	jmp   .nb102_updateouterdata
.nb102_dosingle:
	mov   rdx, [rsp + nb102_innerjjnr]     ;# pointer to jjnr[k] 
	mov   eax, [rdx]	

	mov rsi, [rbp + nb102_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 + nb102_ixO]
    subsd xmm1, [rsp + nb102_iyO]
    subsd xmm2, [rsp + nb102_izO]
    subsd xmm3, [rsp + nb102_ixH1]
    subsd xmm4, [rsp + nb102_iyH1]
    subsd xmm5, [rsp + nb102_izH1]
    subsd xmm6, [rsp + nb102_ixH2]
    subsd xmm7, [rsp + nb102_iyH2]
    subsd xmm8, [rsp + nb102_izH2]
    
	movsd [rsp + nb102_dxOO], xmm0
	movsd [rsp + nb102_dyOO], xmm1
	movsd [rsp + nb102_dzOO], xmm2
	mulsd  xmm0, xmm0
	mulsd  xmm1, xmm1
	mulsd  xmm2, xmm2
	movsd [rsp + nb102_dxH1O], xmm3
	movsd [rsp + nb102_dyH1O], xmm4
	movsd [rsp + nb102_dzH1O], xmm5
	mulsd  xmm3, xmm3
	mulsd  xmm4, xmm4
	mulsd  xmm5, xmm5
	movsd [rsp + nb102_dxH2O], xmm6
	movsd [rsp + nb102_dyH2O], xmm7
	movsd [rsp + nb102_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 + nb102_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 + nb102_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 + nb102_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 + nb102_half]
	mulsd   xmm9, xmm15  ;#  rinvOO 
	mulsd   xmm10, xmm15 ;#   rinvH1O
    mulsd   xmm11, xmm15 ;#   rinvH2O
	
	;# O interactions 
    movsd xmm0, xmm9
    movsd xmm1, xmm10
    movsd xmm2, xmm11
    mulsd  xmm9, xmm9
    mulsd  xmm10, xmm10
    mulsd  xmm11, xmm11
    mulsd  xmm0, [rsp + nb102_qqOO] 
    mulsd  xmm1, [rsp + nb102_qqOH] 
    mulsd  xmm2, [rsp + nb102_qqOH] 
    mulsd  xmm9, xmm0
    mulsd  xmm10, xmm1
    mulsd  xmm11, xmm2
    
    addsd xmm0, [rsp + nb102_vctot] 
    addsd xmm1, xmm2
    addsd xmm0, xmm1
    movsd [rsp + nb102_vctot], xmm0
    
    ;# 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 + nb102_dxOO]
	mulsd xmm8, [rsp + nb102_dyOO]
	mulsd xmm9, [rsp + nb102_dzOO]
	mulsd xmm10, [rsp + nb102_dxH1O]
	mulsd xmm11, [rsp + nb102_dyH1O]
	mulsd xmm12, [rsp + nb102_dzH1O]
	mulsd xmm13, [rsp + nb102_dxH2O]
	mulsd xmm14, [rsp + nb102_dyH2O]
	mulsd xmm15, [rsp + nb102_dzH2O]

    addsd xmm0, xmm7
    addsd xmm1, xmm8
    addsd xmm2, xmm9
    addsd xmm7, [rsp + nb102_fixO]
    addsd xmm8, [rsp + nb102_fiyO]
    addsd xmm9, [rsp + nb102_fizO]

    addsd xmm0, xmm10
    addsd xmm1, xmm11
    addsd xmm2, xmm12
    addsd xmm10, [rsp + nb102_fixH1]
    addsd xmm11, [rsp + nb102_fiyH1]
    addsd xmm12, [rsp + nb102_fizH1]

    addsd xmm0, xmm13
    addsd xmm1, xmm14
    addsd xmm2, xmm15
    addsd xmm13, [rsp + nb102_fixH2]
    addsd xmm14, [rsp + nb102_fiyH2]
    addsd xmm15, [rsp + nb102_fizH2]

    movsd [rsp + nb102_fixO], xmm7
    movsd [rsp + nb102_fiyO], xmm8
    movsd [rsp + nb102_fizO], xmm9
    movsd [rsp + nb102_fixH1], xmm10
    movsd [rsp + nb102_fiyH1], xmm11
    movsd [rsp + nb102_fizH1], xmm12
    movsd [rsp + nb102_fixH2], xmm13
    movsd [rsp + nb102_fiyH2], xmm14
    movsd [rsp + nb102_fizH2], xmm15
   
    ;# store back j O forces from xmm0-xmm2
	movsd [rdi + rax*8],      xmm0
	movsd [rdi + rax*8 + 8],  xmm1
	movsd [rdi + rax*8 + 16], xmm2

	;# move j H1 coordinates to local temp variables 
    movsd xmm0, [rsi + rax*8 + 24] 
    movsd xmm1, [rsi + rax*8 + 32] 
    movsd xmm2, [rsi + rax*8 + 40] 

    ;# xmm0 = H1x
    ;# xmm1 = H1y
    ;# xmm2 = H1z
        
    movsd xmm3, xmm0
    movsd xmm4, xmm1
    movsd xmm5, xmm2
    movsd xmm6, xmm0
    movsd xmm7, xmm1
    movsd xmm8, xmm2
    
    subsd xmm0, [rsp + nb102_ixO]
    subsd xmm1, [rsp + nb102_iyO]
    subsd xmm2, [rsp + nb102_izO]
    subsd xmm3, [rsp + nb102_ixH1]
    subsd xmm4, [rsp + nb102_iyH1]
    subsd xmm5, [rsp + nb102_izH1]
    subsd xmm6, [rsp + nb102_ixH2]
    subsd xmm7, [rsp + nb102_iyH2]
    subsd xmm8, [rsp + nb102_izH2]
    
	movsd [rsp + nb102_dxOH1], xmm0
	movsd [rsp + nb102_dyOH1], xmm1
	movsd [rsp + nb102_dzOH1], xmm2
	mulsd  xmm0, xmm0
	mulsd  xmm1, xmm1
	mulsd  xmm2, xmm2
	movsd [rsp + nb102_dxH1H1], xmm3
	movsd [rsp + nb102_dyH1H1], xmm4
	movsd [rsp + nb102_dzH1H1], xmm5
	mulsd  xmm3, xmm3
	mulsd  xmm4, xmm4
	mulsd  xmm5, xmm5
	movsd [rsp + nb102_dxH2H1], xmm6
	movsd [rsp + nb102_dyH2H1], xmm7
	movsd [rsp + nb102_dzH2H1], 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 jH1 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 + nb102_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 + nb102_half]
	mulsd   xmm9, xmm15  ;# first iteration for rinvOH1 
	mulsd   xmm10, xmm15 ;# first iteration for rinvH1H1
    mulsd   xmm11, xmm15 ;# first iteration for rinvH2OH1

    ;# 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 + nb102_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 + nb102_half]
	mulsd   xmm9, xmm15  ;#  rinvOH1
	mulsd   xmm10, xmm15 ;#   rinvH1H1
    mulsd   xmm11, xmm15 ;#   rinvH2H1
	
	;# H1 interactions 
    movsd xmm0, xmm9
    movsd xmm1, xmm10
    movsd xmm2, xmm11
    mulsd  xmm9, xmm9
    mulsd  xmm10, xmm10
    mulsd  xmm11, xmm11
    mulsd  xmm0, [rsp + nb102_qqOH] 
    mulsd  xmm1, [rsp + nb102_qqHH] 
    mulsd  xmm2, [rsp + nb102_qqHH] 
    mulsd  xmm9, xmm0
    mulsd  xmm10, xmm1
    mulsd  xmm11, xmm2
    
    addsd xmm0, [rsp + nb102_vctot] 
    addsd xmm1, xmm2
    addsd xmm0, xmm1
    movsd [rsp + nb102_vctot], xmm0
    
    ;# move j H1 forces to xmm0-xmm2
	movsd xmm0, [rdi + rax*8 + 24]
	movsd xmm1, [rdi + rax*8 + 32]
	movsd xmm2, [rdi + rax*8 + 40]

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

	mulsd xmm7, [rsp + nb102_dxOH1]
	mulsd xmm8, [rsp + nb102_dyOH1]
	mulsd xmm9, [rsp + nb102_dzOH1]
	mulsd xmm10, [rsp + nb102_dxH1H1]
	mulsd xmm11, [rsp + nb102_dyH1H1]
	mulsd xmm12, [rsp + nb102_dzH1H1]
	mulsd xmm13, [rsp + nb102_dxH2H1]
	mulsd xmm14, [rsp + nb102_dyH2H1]