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

    movhpd xmm2, [rsi + rbx*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
    
    subpd xmm0, [rsp + nb132_ixO]
    subpd xmm1, [rsp + nb132_iyO]
    subpd xmm2, [rsp + nb132_izO]
    subpd xmm3, [rsp + nb132_ixH1]
    subpd xmm4, [rsp + nb132_iyH1]
    subpd xmm5, [rsp + nb132_izH1]
    subpd xmm6, [rsp + nb132_ixH2]
    subpd xmm7, [rsp + nb132_iyH2]
    subpd xmm8, [rsp + nb132_izH2]
    
	movapd [rsp + nb132_dxOH2], xmm0
	movapd [rsp + nb132_dyOH2], xmm1
	movapd [rsp + nb132_dzOH2], xmm2
	mulpd  xmm0, xmm0
	mulpd  xmm1, xmm1
	mulpd  xmm2, xmm2
	movapd [rsp + nb132_dxH1H2], xmm3
	movapd [rsp + nb132_dyH1H2], xmm4
	movapd [rsp + nb132_dzH1H2], xmm5
	mulpd  xmm3, xmm3
	mulpd  xmm4, xmm4
	mulpd  xmm5, xmm5
	movapd [rsp + nb132_dxH2H2], xmm6
	movapd [rsp + nb132_dyH2H2], xmm7
	movapd [rsp + nb132_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 + nb132_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 + nb132_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 + nb132_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 + nb132_half]
	mulpd   xmm9, xmm15  ;#  rinvOH2
	mulpd   xmm10, xmm15 ;#   rinvH1H2
    mulpd   xmm11, xmm15 ;#   rinvH2H2
	
	;# H2 interactions 
    movapd xmm0, xmm9
    movapd xmm1, xmm10
    movapd xmm2, xmm11
    mulpd  xmm9, xmm9
    mulpd  xmm10, xmm10
    mulpd  xmm11, xmm11
    mulpd  xmm0, [rsp + nb132_qqOH] 
    mulpd  xmm1, [rsp + nb132_qqHH] 
    mulpd  xmm2, [rsp + nb132_qqHH] 
    mulpd  xmm9, xmm0
    mulpd  xmm10, xmm1
    mulpd  xmm11, xmm2
    
    addpd xmm0, [rsp + nb132_vctot] 
    addpd xmm1, xmm2
    addpd xmm0, xmm1
    movapd [rsp + nb132_vctot], xmm0
    
    ;# 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 + nb132_dxOH2]
	mulpd xmm8, [rsp + nb132_dyOH2]
	mulpd xmm9, [rsp + nb132_dzOH2]
	mulpd xmm10, [rsp + nb132_dxH1H2]
	mulpd xmm11, [rsp + nb132_dyH1H2]
	mulpd xmm12, [rsp + nb132_dzH1H2]
	mulpd xmm13, [rsp + nb132_dxH2H2]
	mulpd xmm14, [rsp + nb132_dyH2H2]
	mulpd xmm15, [rsp + nb132_dzH2H2]

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

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

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

    movapd [rsp + nb132_fixO], xmm7
    movapd [rsp + nb132_fiyO], xmm8
    movapd [rsp + nb132_fizO], xmm9
    movapd [rsp + nb132_fixH1], xmm10
    movapd [rsp + nb132_fiyH1], xmm11
    movapd [rsp + nb132_fizH1], xmm12
    movapd [rsp + nb132_fixH2], xmm13
    movapd [rsp + nb132_fiyH2], xmm14
    movapd [rsp + nb132_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 + nb132_innerk],  2
	jl    .nb132_checksingle
	jmp   .nb132_unroll_loop
.nb132_checksingle:
	mov   edx, [rsp + nb132_innerk]
	and   edx, 1
	jnz   .nb132_dosingle
	jmp   .nb132_updateouterdata
.nb132_dosingle:
	mov   rdx, [rsp + nb132_innerjjnr]     ;# pointer to jjnr[k] 
	mov   eax, [rdx]	
	
	add qword ptr [rsp + nb132_innerjjnr],  8	;# advance pointer (unrolled 2) 

	mov rsi, [rbp + nb132_pos]       ;# base of pos[] 

	lea   rax, [rax + rax*2]     ;# replace jnr with j3 
	
	;# 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
        
    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_dxOO], xmm0
	movapd [rsp + nb132_dyOO], xmm1
	movapd [rsp + nb132_dzOO], xmm2
	mulsd  xmm0, xmm0
	mulsd  xmm1, xmm1
	mulsd  xmm2, xmm2
	movapd [rsp + nb132_dxH1O], xmm3
	movapd [rsp + nb132_dyH1O], xmm4
	movapd [rsp + nb132_dzH1O], xmm5
	mulsd  xmm3, xmm3
	mulsd  xmm4, xmm4
	mulsd  xmm5, xmm5
	movapd [rsp + nb132_dxH2O], xmm6
	movapd [rsp + nb132_dyH2O], xmm7
	movapd [rsp + nb132_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
	
	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 rinvOO 
	mulsd   xmm10, xmm15 ;# first iteration for rinvH1O
    mulsd   xmm11, xmm15 ;# first iteration for rinvH2O	

    ;# 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  ;#  rinvOO 
	mulsd   xmm10, xmm15 ;#   rinvH1O
    mulsd   xmm11, xmm15 ;#   rinvH2O
	
	;# O interactions 
    ;# rsq in xmm0,xmm3,xmm6  
    ;# rinv in xmm9, xmm10, xmm11
    movapd [rsp + nb132_rsqOO], xmm0
    movapd [rsp + nb132_rsqOH1], xmm3
    movapd [rsp + nb132_rsqOH2], xmm6
    movapd [rsp + nb132_rinvOO], xmm9
    movapd [rsp + nb132_rinvOH1], xmm10
    movapd [rsp + nb132_rinvOH2], xmm11

    ;# table LJ interaction
    mulsd  xmm0, xmm9
    mulsd  xmm0, [rsp + nb132_tsc] ;# rtab

    ;# truncate and convert to integers
    cvttsd2si r8d, xmm0

    ;# convert back to float
    cvtsi2sd  xmm2, r8d
         
    ;# multiply by 8
    shl   r8d, 3

    ;# calculate eps
    subsd     xmm0, xmm2
    mov  rsi, [rbp + nb132_VFtab]
            
    movsd xmm4, [rsi + r8*8]
   	movsd xmm5, [rsi + r8*8 + 8]
	movsd xmm6, [rsi + r8*8 + 16]
	movsd xmm7, [rsi + r8*8 + 24]
    movsd xmm8, [rsi + r8*8 + 32]
   	movsd xmm9, [rsi + r8*8 + 40]
	movsd xmm10, [rsi + r8*8 + 48]
	movsd xmm11, [rsi + r8*8 + 56]
    
    ;# dispersion table in xmm4-xmm7, repulsion table in xmm8-xmm11

    mulsd  xmm7, xmm0    ;# Heps
    mulsd  xmm11, xmm0 
    mulsd  xmm6, xmm0   ;# Geps
    mulsd  xmm10, xmm0 
    mulsd  xmm7, xmm0   ;# Heps2
    mulsd  xmm11, xmm0 
    addsd  xmm5, xmm6  ;# F+Geps
    addsd  xmm9, xmm10 
    addsd  xmm5, xmm7   ;# F+Geps+Heps2 = Fp
    addsd  xmm9, xmm11 
    addsd  xmm7, xmm7    ;# 2*Heps2
    addsd  xmm11, xmm11
    addsd  xmm7, xmm6   ;# 2*Heps2+Geps
    addsd  xmm11, xmm10
    
    addsd  xmm7, xmm5  ;# FF = Fp + 2*Heps2 + Geps
    addsd  xmm11, xmm9
    mulsd  xmm5, xmm0  ;# eps*Fp
    mulsd  xmm9, xmm0
    movapd xmm12, [rsp + nb132_c6]
    movapd xmm13, [rsp + nb132_c12]
    addsd  xmm5, xmm4 ;# VV
    addsd  xmm9, xmm8

    mulsd  xmm5, xmm12  ;# VV*c6 = vnb6
    mulsd  xmm9, xmm13  ;# VV*c12 = vnb12
    addsd  xmm5, xmm9
    addsd  xmm5, [rsp + nb132_Vvdwtot]
    movsd [rsp + nb132_Vvdwtot], xmm5
        
    mulsd  xmm7, xmm12   ;# FF*c6 = fnb6
    mulsd  xmm11, xmm13   ;# FF*c12  = fnb12
    addsd  xmm7, xmm11
    mulsd  xmm7, [rsp + nb132_tsc]

    movapd xmm9, [rsp + nb132_rinvOO]
    movapd xmm10, [rsp + nb132_rinvOH1]
    movapd xmm11, [rsp + nb132_rinvOH2]

    movapd xmm0, xmm9
    movapd xmm1, xmm10
    movapd xmm2, xmm11
    
    mulsd  xmm10, xmm10
    mulsd  xmm11, xmm11
    
    mulsd  xmm0, [rsp + nb132_qqOO] 
    mulsd  xmm1, [rsp + nb132_qqOH] 
    mulsd  xmm2, [rsp + nb132_qqOH] 
    
    mulsd  xmm9, xmm0
    mulsd  xmm10, xmm1
    mulsd  xmm11, xmm2
    
    subsd  xmm9, xmm7
    mulsd  xmm9, [rsp + nb132_rinvOO]
    
    addsd xmm0, [rsp + nb132_vctot] 
    addsd xmm1, xmm2
    addsd xmm0, xmm1
    movsd [rsp + nb132_vctot], xmm0
    
    ;# move j O forces to xmm0-xmm2
	mov   rdi, [rbp + nb132_faction]	
	movsd xmm0, [rdi + rax*8]
	movsd xmm1, [rdi + rax*8 + 8]
	movsd xmm2, [rdi + rax*8 + 16]

    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_dxOO]
	mulsd xmm8, [rsp + nb132_dyOO]
	mulsd xmm9, [rsp + nb132_dzOO]
	mulsd xmm10, [rsp + nb132_dxH1O]
	mulsd xmm11, [rsp + nb132_dyH1O]
	mulsd xmm12, [rsp + nb132_dzH1O]
	mulsd xmm13, [rsp + nb132_dxH2O]
	mulsd xmm14, [rsp + nb132_dyH2O]
	mulsd xmm15, [rsp + nb132_dzH2O]