nb_kernel132_x86_64_sse2.intel_syntax.s

来自「最著名最快的分子模拟软件」· S 代码 · 共 2,328 行 · 第 1/5 页

    addsd xmm0, xmm7
    addsd xmm1, xmm8
    addsd xmm2, xmm9
    addsd xmm7, [rsp + nb132_fixO]
    addsd xmm8, [rsp + nb132_fiyO]
    addsd xmm9, [rsp + nb132_fizO]

    addsd xmm0, xmm10
    addsd xmm1, xmm11
    addsd xmm2, xmm12
    addsd xmm10, [rsp + nb132_fixH1]
    addsd xmm11, [rsp + nb132_fiyH1]
    addsd xmm12, [rsp + nb132_fizH1]

    addsd xmm0, xmm13
    addsd xmm1, xmm14
    addsd xmm2, xmm15
    addsd xmm13, [rsp + nb132_fixH2]
    addsd xmm14, [rsp + nb132_fiyH2]
    addsd xmm15, [rsp + nb132_fizH2]

    movsd [rsp + nb132_fixO], xmm7
    movsd [rsp + nb132_fiyO], xmm8
    movsd [rsp + nb132_fizO], xmm9
    movsd [rsp + nb132_fixH1], xmm10
    movsd [rsp + nb132_fiyH1], xmm11
    movsd [rsp + nb132_fizH1], xmm12
    movsd [rsp + nb132_fixH2], xmm13
    movsd [rsp + nb132_fiyH2], xmm14
    movsd [rsp + nb132_fizH2], xmm15
   
    ;# store back j O forces from xmm0-xmm2
	mov   rdi, [rbp + nb132_faction]	
	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 
    mov rsi, [rbp + nb132_pos]
    movsd xmm0, [rsi + rax*8 + 24] 
    movsd xmm1, [rsi + rax*8 + 32] 
    movsd xmm2, [rsi + rax*8 + 40] 
   
    ;# xmm0 = H1x
    ;# xmm1 = H1y
    ;# xmm2 = H1z
        
    movapd xmm3, xmm0
    movapd xmm4, xmm1
    movapd xmm5, xmm2
    movapd xmm6, xmm0
    movapd xmm7, xmm1
    movapd xmm8, xmm2
    
    subsd xmm0, [rsp + nb132_ixO]
    subsd xmm1, [rsp + nb132_iyO]
    subsd xmm2, [rsp + nb132_izO]
    subsd xmm3, [rsp + nb132_ixH1]
    subsd xmm4, [rsp + nb132_iyH1]
    subsd xmm5, [rsp + nb132_izH1]
    subsd xmm6, [rsp + nb132_ixH2]
    subsd xmm7, [rsp + nb132_iyH2]
    subsd xmm8, [rsp + nb132_izH2]
    
	movapd [rsp + nb132_dxOH1], xmm0
	movapd [rsp + nb132_dyOH1], xmm1
	movapd [rsp + nb132_dzOH1], xmm2
	mulsd  xmm0, xmm0
	mulsd  xmm1, xmm1
	mulsd  xmm2, xmm2
	movapd [rsp + nb132_dxH1H1], xmm3
	movapd [rsp + nb132_dyH1H1], xmm4
	movapd [rsp + nb132_dzH1H1], xmm5
	mulsd  xmm3, xmm3
	mulsd  xmm4, xmm4
	mulsd  xmm5, xmm5
	movapd [rsp + nb132_dxH2H1], xmm6
	movapd [rsp + nb132_dyH2H1], xmm7
	movapd [rsp + nb132_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
	
	movapd  xmm2, xmm1
	movapd  xmm5, xmm4
    movapd  xmm8, xmm7
    
	mulsd   xmm1, xmm1 ;# lu*lu
	mulsd   xmm4, xmm4 ;# lu*lu
    mulsd   xmm7, xmm7 ;# lu*lu
		
	movapd  xmm9, [rsp + nb132_three]
	movapd  xmm10, xmm9
    movapd  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)

	movapd  xmm15, [rsp + nb132_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    
	movapd  xmm2, xmm9
	movapd  xmm5, xmm10
    movapd  xmm8, xmm11
    
	mulsd   xmm2, xmm2 ;# lu*lu
	mulsd   xmm5, xmm5 ;# lu*lu
    mulsd   xmm8, xmm8 ;# lu*lu
		
	movapd  xmm1, [rsp + nb132_three]
	movapd  xmm4, xmm1
    movapd  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)

	movapd  xmm15, [rsp + nb132_half]
	mulsd   xmm9, xmm15  ;#  rinvOH1
	mulsd   xmm10, xmm15 ;#   rinvH1H1
    mulsd   xmm11, xmm15 ;#   rinvH2H1
	
	;# H1 interactions 
    movapd xmm0, xmm9
    movapd xmm1, xmm10
    movapd xmm2, xmm11
    mulsd  xmm9, xmm9
    mulsd  xmm10, xmm10
    mulsd  xmm11, xmm11
    mulsd  xmm0, [rsp + nb132_qqOH] 
    mulsd  xmm1, [rsp + nb132_qqHH] 
    mulsd  xmm2, [rsp + nb132_qqHH] 
    mulsd  xmm9, xmm0
    mulsd  xmm10, xmm1
    mulsd  xmm11, xmm2
    
    addsd xmm0, [rsp + nb132_vctot] 
    addsd xmm1, xmm2
    addsd xmm0, xmm1
    movsd [rsp + nb132_vctot], xmm0
    
    ;# move j H1 forces to xmm0-xmm2
	mov   rdi, [rbp + nb132_faction]	
	movsd xmm0, [rdi + rax*8 + 24]
	movsd xmm1, [rdi + rax*8 + 32]
	movsd xmm2, [rdi + rax*8 + 40]

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

	mulsd xmm7, [rsp + nb132_dxOH1]
	mulsd xmm8, [rsp + nb132_dyOH1]
	mulsd xmm9, [rsp + nb132_dzOH1]
	mulsd xmm10, [rsp + nb132_dxH1H1]
	mulsd xmm11, [rsp + nb132_dyH1H1]
	mulsd xmm12, [rsp + nb132_dzH1H1]
	mulsd xmm13, [rsp + nb132_dxH2H1]
	mulsd xmm14, [rsp + nb132_dyH2H1]
	mulsd xmm15, [rsp + nb132_dzH2H1]

    addsd xmm0, xmm7
    addsd xmm1, xmm8
    addsd xmm2, xmm9
    addsd xmm7, [rsp + nb132_fixO]
    addsd xmm8, [rsp + nb132_fiyO]
    addsd xmm9, [rsp + nb132_fizO]

    addsd xmm0, xmm10
    addsd xmm1, xmm11
    addsd xmm2, xmm12
    addsd xmm10, [rsp + nb132_fixH1]
    addsd xmm11, [rsp + nb132_fiyH1]
    addsd xmm12, [rsp + nb132_fizH1]

    addsd xmm0, xmm13
    addsd xmm1, xmm14
    addsd xmm2, xmm15
    addsd xmm13, [rsp + nb132_fixH2]
    addsd xmm14, [rsp + nb132_fiyH2]
    addsd xmm15, [rsp + nb132_fizH2]

    movsd [rsp + nb132_fixO], xmm7
    movsd [rsp + nb132_fiyO], xmm8
    movsd [rsp + nb132_fizO], xmm9
    movsd [rsp + nb132_fixH1], xmm10
    movsd [rsp + nb132_fiyH1], xmm11
    movsd [rsp + nb132_fizH1], xmm12
    movsd [rsp + nb132_fixH2], xmm13
    movsd [rsp + nb132_fiyH2], xmm14
    movsd [rsp + nb132_fizH2], xmm15
   
    ;# store back j H1 forces from xmm0-xmm2
	mov   rdi, [rbp + nb132_faction]	
	movsd [rdi + rax*8 + 24], xmm0
	movsd [rdi + rax*8 + 32], xmm1
	movsd [rdi + rax*8 + 40], xmm2
       
	;# move j H2 coordinates to local temp variables 
    mov rsi, [rbp + nb132_pos]
    movsd xmm0, [rsi + rax*8 + 48] 
    movsd xmm1, [rsi + rax*8 + 56] 
    movsd xmm2, [rsi + rax*8 + 64] 

    ;# xmm0 = H2x
    ;# xmm1 = H2y
    ;# xmm2 = H2z
        
    movapd xmm3, xmm0
    movapd xmm4, xmm1
    movapd xmm5, xmm2
    movapd xmm6, xmm0
    movapd xmm7, xmm1
    movapd xmm8, xmm2
    
    subsd xmm0, [rsp + nb132_ixO]
    subsd xmm1, [rsp + nb132_iyO]
    subsd xmm2, [rsp + nb132_izO]
    subsd xmm3, [rsp + nb132_ixH1]
    subsd xmm4, [rsp + nb132_iyH1]
    subsd xmm5, [rsp + nb132_izH1]
    subsd xmm6, [rsp + nb132_ixH2]
    subsd xmm7, [rsp + nb132_iyH2]
    subsd xmm8, [rsp + nb132_izH2]
    
	movapd [rsp + nb132_dxOH2], xmm0
	movapd [rsp + nb132_dyOH2], xmm1
	movapd [rsp + nb132_dzOH2], xmm2
	mulsd  xmm0, xmm0
	mulsd  xmm1, xmm1
	mulsd  xmm2, xmm2
	movapd [rsp + nb132_dxH1H2], xmm3
	movapd [rsp + nb132_dyH1H2], xmm4
	movapd [rsp + nb132_dzH1H2], xmm5
	mulsd  xmm3, xmm3
	mulsd  xmm4, xmm4
	mulsd  xmm5, xmm5
	movapd [rsp + nb132_dxH2H2], xmm6
	movapd [rsp + nb132_dyH2H2], xmm7
	movapd [rsp + nb132_dzH2H2], 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 jH2 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
	
	movapd  xmm2, xmm1
	movapd  xmm5, xmm4
    movapd  xmm8, xmm7
    
	mulsd   xmm1, xmm1 ;# lu*lu
	mulsd   xmm4, xmm4 ;# lu*lu
    mulsd   xmm7, xmm7 ;# lu*lu
		
	movapd  xmm9, [rsp + nb132_three]
	movapd  xmm10, xmm9
    movapd  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)

	movapd  xmm15, [rsp + nb132_half]
	mulsd   xmm9, xmm15  ;# first iteration for rinvOH2 
	mulsd   xmm10, xmm15 ;# first iteration for rinvH1H2
    mulsd   xmm11, xmm15 ;# first iteration for rinvH2H2

    ;# second iteration step    
	movapd  xmm2, xmm9
	movapd  xmm5, xmm10
    movapd  xmm8, xmm11
    
	mulsd   xmm2, xmm2 ;# lu*lu
	mulsd   xmm5, xmm5 ;# lu*lu
    mulsd   xmm8, xmm8 ;# lu*lu
		
	movapd  xmm1, [rsp + nb132_three]
	movapd  xmm4, xmm1
    movapd  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)

	movapd  xmm15, [rsp + nb132_half]
	mulsd   xmm9, xmm15  ;#  rinvOH2
	mulsd   xmm10, xmm15 ;#   rinvH1H2
    mulsd   xmm11, xmm15 ;#   rinvH2H2
	
	;# H2 interactions 
    movapd xmm0, xmm9
    movapd xmm1, xmm10
    movapd xmm2, xmm11
    mulsd  xmm9, xmm9
    mulsd  xmm10, xmm10
    mulsd  xmm11, xmm11
    mulsd  xmm0, [rsp + nb132_qqOH] 
    mulsd  xmm1, [rsp + nb132_qqHH] 
    mulsd  xmm2, [rsp + nb132_qqHH] 
    mulsd  xmm9, xmm0
    mulsd  xmm10, xmm1
    mulsd  xmm11, xmm2
    
    addsd xmm0, [rsp + nb132_vctot] 
    addsd xmm1, xmm2
    addsd xmm0, xmm1
    movsd [rsp + nb132_vctot], xmm0
    
    ;# move j H2 forces to xmm0-xmm2
	movsd xmm0, [rdi + rax*8 + 48]
	movsd xmm1, [rdi + rax*8 + 56]
	movsd xmm2, [rdi + rax*8 + 64]

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

	mulsd xmm7, [rsp + nb132_dxOH2]
	mulsd xmm8, [rsp + nb132_dyOH2]
	mulsd xmm9, [rsp + nb132_dzOH2]
	mulsd xmm10, [rsp + nb132_dxH1H2]
	mulsd xmm11, [rsp + nb132_dyH1H2]
	mulsd xmm12, [rsp + nb132_dzH1H2]
	mulsd xmm13, [rsp + nb132_dxH2H2]
	mulsd xmm14, [rsp + nb132_dyH2H2]
	mulsd xmm15, [rsp + nb132_dzH2H2]

    addsd xmm0, xmm7
    addsd xmm1, xmm8
    addsd xmm2, xmm9
    addsd xmm7, [rsp + nb132_fixO]
    addsd xmm8, [rsp + nb132_fiyO]
    addsd xmm9, [rsp + nb132_fizO]

    addsd xmm0, xmm10
    addsd xmm1, xmm11
    addsd xmm2, xmm12
    addsd xmm10, [rsp + nb132_fixH1]
    addsd xmm11, [rsp + nb132_fiyH1]
    addsd xmm12, [rsp + nb132_fizH1]

    addsd xmm0, xmm13
    addsd xmm1, xmm14
    addsd xmm2, xmm15
    addsd xmm13, [rsp + nb132_fixH2]
    addsd xmm14, [rsp + nb132_fiyH2]
    addsd xmm15, [rsp + nb132_fizH2]

    movsd [rsp + nb132_fixO], xmm7
    movsd [rsp + nb132_fiyO], xmm8
    movsd [rsp + nb132_fizO], xmm9
    movsd [rsp + nb132_fixH1], xmm10
    movsd [rsp + nb132_fiyH1], xmm11
    movsd [rsp + nb132_fizH1], xmm12
    movsd [rsp + nb132_fixH2], xmm13
    movsd [rsp + nb132_fiyH2], xmm14
    movsd [rsp + nb132_fizH2], xmm15
   
    ;# store back j H2 forces from xmm0-xmm2
	movsd [rdi + rax*8 + 48], xmm0
	movsd [rdi + rax*8 + 56], xmm1
	movsd [rdi + rax*8 + 64], xmm2

.nb132_updateouterdata:
	mov   ecx, [rsp + nb132_ii3]
	mov   rdi, [rbp + nb132_faction]
	mov   rsi, [rbp + nb132_fshift]
	mov   edx, [rsp + nb132_is3]

	;# accumulate  Oi forces in xmm0, xmm1, xmm2 
	movapd xmm0, [rsp + nb132_fixO]
	movapd xmm1, [rsp + nb132_fiyO]
	movapd xmm2, [rsp + nb132_fizO]

	movhlps xmm3, xmm0
	movhlps xmm4, xmm1
	movhlps xmm5, xmm2
	addsd  xmm0, xmm3
	addsd  xmm1, xmm4
	addsd  xmm2, xmm5 ;# sum is in low xmm0-xmm2 

	movapd xmm3, xmm0	
	movapd xmm4, xmm1	
	movapd xmm5, xmm2	

	;# increment i force 
	movsd  xmm3, [rdi + rcx*8]
	movsd  xmm4, [rdi + rcx*8 + 8]
	movsd  xmm5, [rdi + rcx*8 + 16]
	subsd  xmm3, xmm0