nb_kernel312_x86_64_sse2.intel_syntax.s

来自「最著名最快的分子模拟软件」· S 代码 · 共 2,377 行 · 第 1/5 页

    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 + nb312_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 + nb312_half]
	mulpd   xmm9, xmm15  ;# first iteration for rinvOO 
	mulpd   xmm10, xmm15 ;# first iteration for rinvH1O
    mulpd   xmm11, xmm15 ;# first iteration for rinvH2O	

    ;# 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 + nb312_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 + nb312_half]
	mulpd   xmm9, xmm15  ;#  rinvOO 
	mulpd   xmm10, xmm15 ;#   rinvH1O
    mulpd   xmm11, xmm15 ;#   rinvH2O
	
	movapd  [rsp + nb312_rinvOO], xmm9
	movapd  [rsp + nb312_rinvH1O], xmm10
	movapd  [rsp + nb312_rinvH2O], xmm11
	
	;# O interactions 
    ;# rsq in xmm0,xmm3,xmm6  
    ;# rinv in xmm9, xmm10, xmm11

    movapd xmm1, [rsp + nb312_tsc]
    mulpd  xmm0, xmm9  ;# r
    mulpd  xmm3, xmm10
    mulpd  xmm6, xmm11
    mulpd  xmm0, xmm1 ;# rtab
    mulpd  xmm3, xmm1
    mulpd  xmm6, xmm1
    
    ;# truncate and convert to integers
    cvttpd2dq xmm1, xmm0
    cvttpd2dq xmm4, xmm3
    cvttpd2dq xmm7, xmm6        

    ;# convert back to float
    cvtdq2pd  xmm2, xmm1
    cvtdq2pd  xmm5, xmm4
    cvtdq2pd  xmm8, xmm7
    
    ;# multiply by 4
    pslld   xmm1, 2
    pslld   xmm4, 2
    pslld   xmm7, 2
    
    ;# move to integer registers
    pshufd xmm13, xmm1, 1
    pshufd xmm14, xmm4, 1
    pshufd xmm15, xmm7, 1
    movd    r8d, xmm1
    movd    r10d, xmm4
    movd    r12d, xmm7
    movd    r9d, xmm13
    movd    r11d, xmm14
    movd    r13d, xmm15
        
    mov  rsi, [rbp + nb312_VFtab]

    ;# calculate eps
    subpd     xmm0, xmm2
    subpd     xmm3, xmm5
    subpd     xmm6, xmm8

    movapd    [rsp + nb312_epsO], xmm0
    movapd    [rsp + nb312_epsH1], xmm3
    movapd    [rsp + nb312_epsH2], xmm6

    ;# Load LOTS of table data
    movlpd xmm0,  [rsi + r8*8]
    movlpd xmm1,  [rsi + r8*8 + 8]
    movlpd xmm2,  [rsi + r8*8 + 16]
    movlpd xmm3,  [rsi + r8*8 + 24]
    movlpd xmm4,  [rsi + r10*8]
    movlpd xmm5,  [rsi + r10*8 + 8]
    movlpd xmm6,  [rsi + r10*8 + 16]
    movlpd xmm7,  [rsi + r10*8 + 24]
    movlpd xmm8,  [rsi + r12*8]
    movlpd xmm9,  [rsi + r12*8 + 8]
    movlpd xmm10, [rsi + r12*8 + 16]
    movlpd xmm11, [rsi + r12*8 + 24]
    movhpd xmm0,  [rsi + r9*8]
    movhpd xmm1,  [rsi + r9*8 + 8]
    movhpd xmm2,  [rsi + r9*8 + 16]
    movhpd xmm3,  [rsi + r9*8 + 24]
    movhpd xmm4,  [rsi + r11*8]
    movhpd xmm5,  [rsi + r11*8 + 8]
    movhpd xmm6,  [rsi + r11*8 + 16]
    movhpd xmm7,  [rsi + r11*8 + 24]
    movhpd xmm8,  [rsi + r13*8]
    movhpd xmm9,  [rsi + r13*8 + 8]
    movhpd xmm10, [rsi + r13*8 + 16]
    movhpd xmm11, [rsi + r13*8 + 24]
    ;# table data ready in xmm0-xmm3 , xmm4-xmm7 , and xmm8-xmm11
 
    mulpd  xmm3, [rsp + nb312_epsO]   ;# Heps
    mulpd  xmm7, [rsp + nb312_epsH1]
    mulpd  xmm11, [rsp + nb312_epsH2]
    mulpd  xmm2, [rsp + nb312_epsO]   ;# Geps
    mulpd  xmm6, [rsp + nb312_epsH1]
    mulpd  xmm10, [rsp + nb312_epsH2]
    mulpd  xmm3, [rsp + nb312_epsO]   ;# Heps2
    mulpd  xmm7, [rsp + nb312_epsH1]
    mulpd  xmm11, [rsp + nb312_epsH2]

    addpd  xmm1, xmm2   ;# F+Geps
    addpd  xmm5, xmm6
    addpd  xmm9, xmm10 
    addpd  xmm1, xmm3   ;# F+Geps+Heps2 = Fp
    addpd  xmm5, xmm7
    addpd  xmm9, xmm11 
    addpd  xmm3, xmm3    ;# 2*Heps2
    addpd  xmm7, xmm7
    addpd  xmm11, xmm11
    addpd  xmm3, xmm2    ;# 2*Heps2+Geps
    addpd  xmm7, xmm6  
    addpd  xmm11, xmm10
    addpd  xmm3, xmm1   ;# FF = Fp + 2*Heps2 + Geps
    addpd  xmm7, xmm5
    addpd  xmm11, xmm9
    mulpd  xmm1, [rsp + nb312_epsO]   ;# eps*Fp
    mulpd  xmm5, [rsp + nb312_epsH1]
    mulpd  xmm9, [rsp + nb312_epsH2]
    addpd  xmm1, xmm0     ;# VV
    addpd  xmm5, xmm4
    addpd  xmm9, xmm8
    mulpd  xmm1, [rsp + nb312_qqOO]   ;# VV*qq = vcoul
    mulpd  xmm5, [rsp + nb312_qqOH]
    mulpd  xmm9, [rsp + nb312_qqOH]
    mulpd  xmm3, [rsp + nb312_qqOO]    ;# FF*qq = fij
    mulpd  xmm7, [rsp + nb312_qqOH]
    mulpd  xmm11, [rsp + nb312_qqOH] 
    
    ;# calculate LJ
    movapd xmm12, [rsp + nb312_rinvOO]
    mulpd  xmm12, xmm12 ;# rinvsq
    movapd xmm13, xmm12 ;# rinvsq
    mulpd  xmm12, xmm12 ;# rinv4
    mulpd  xmm12, xmm13 ;# rinv6
    movapd xmm13, xmm12 ;# rinv6
    mulpd  xmm12, xmm12 ;# rinv12
	mulpd  xmm13, [rsp + nb312_c6]
	mulpd  xmm12, [rsp + nb312_c12]
    movapd xmm14, xmm12
    subpd  xmm14, xmm13
    
	addpd  xmm14, [rsp + nb312_Vvdwtot]
	mulpd  xmm13, [rsp + nb312_six]
	mulpd  xmm12, [rsp + nb312_twelve]
	movapd [rsp + nb312_Vvdwtot], xmm14
    subpd  xmm12, xmm13 ;# LJ fscal    
    mulpd  xmm12, [rsp + nb312_rinvOO]
    movapd xmm4, xmm12
    
    ;# accumulate vctot
    addpd  xmm1, [rsp + nb312_vctot]
    addpd  xmm5, xmm9
    addpd  xmm1, xmm5
    movapd [rsp + nb312_vctot], xmm1
    
    xorpd xmm8, xmm8
    xorpd xmm12, xmm12
    
    movapd xmm5, [rsp + nb312_tsc]
    mulpd  xmm3, xmm5  ;# fscal
    mulpd  xmm7, xmm5
    mulpd  xmm11, xmm5

    subpd xmm4, xmm3
    subpd xmm8, xmm7
    subpd xmm12, xmm11
    
    mulpd xmm4, [rsp + nb312_rinvOO]
    mulpd xmm8, [rsp + nb312_rinvH1O]
    mulpd xmm12, [rsp + nb312_rinvH2O]
    
    ;# move j O forces to xmm0-xmm2
    mov rdi, [rbp + nb312_faction]
	movlpd xmm0, [rdi + rax*8]
	movlpd xmm1, [rdi + rax*8 + 8]
	movlpd xmm2, [rdi + rax*8 + 16]
	movhpd xmm0, [rdi + rbx*8]
	movhpd xmm1, [rdi + rbx*8 + 8]
	movhpd xmm2, [rdi + rbx*8 + 16]

    movapd xmm3, xmm4
    movapd xmm5, xmm4
    movapd xmm7, xmm8
    movapd xmm9, xmm8
    movapd xmm10, xmm12
    movapd xmm11, xmm12

	mulpd xmm3, [rsp + nb312_dxOO]
	mulpd xmm4, [rsp + nb312_dyOO]
	mulpd xmm5, [rsp + nb312_dzOO]
	mulpd xmm7, [rsp + nb312_dxH1O]
	mulpd xmm8, [rsp + nb312_dyH1O]
	mulpd xmm9, [rsp + nb312_dzH1O]
	mulpd xmm10, [rsp + nb312_dxH2O]
	mulpd xmm11, [rsp + nb312_dyH2O]
	mulpd xmm12, [rsp + nb312_dzH2O]

    addpd xmm0, xmm3
    addpd xmm1, xmm4
    addpd xmm2, xmm5
    addpd xmm3, [rsp + nb312_fixO]
    addpd xmm4, [rsp + nb312_fiyO]
    addpd xmm5, [rsp + nb312_fizO]

    addpd xmm0, xmm7
    addpd xmm1, xmm8
    addpd xmm2, xmm9
    addpd xmm7, [rsp + nb312_fixH1]
    addpd xmm8, [rsp + nb312_fiyH1]
    addpd xmm9, [rsp + nb312_fizH1]

    addpd xmm0, xmm10
    addpd xmm1, xmm11
    addpd xmm2, xmm12
    addpd xmm10, [rsp + nb312_fixH2]
    addpd xmm11, [rsp + nb312_fiyH2]
    addpd xmm12, [rsp + nb312_fizH2]

    movapd [rsp + nb312_fixO], xmm3
    movapd [rsp + nb312_fiyO], xmm4
    movapd [rsp + nb312_fizO], xmm5
    movapd [rsp + nb312_fixH1], xmm7
    movapd [rsp + nb312_fiyH1], xmm8
    movapd [rsp + nb312_fizH1], xmm9
    movapd [rsp + nb312_fixH2], xmm10
    movapd [rsp + nb312_fiyH2], xmm11
    movapd [rsp + nb312_fizH2], xmm12
   
    ;# store back j O forces from xmm0-xmm2
	movlpd [rdi + rax*8],      xmm0
	movlpd [rdi + rax*8 + 8],  xmm1
	movlpd [rdi + rax*8 + 16], xmm2
	movhpd [rdi + rbx*8],      xmm0
	movhpd [rdi + rbx*8 + 8],  xmm1
	movhpd [rdi + rbx*8 + 16], xmm2

	;# move j H1 coordinates to local temp variables 
    mov rsi, [rbp + nb312_pos]
    movlpd xmm0, [rsi + rax*8 + 24] 
    movlpd xmm1, [rsi + rax*8 + 32] 
    movlpd xmm2, [rsi + rax*8 + 40] 
    movhpd xmm0, [rsi + rbx*8 + 24] 
    movhpd xmm1, [rsi + rbx*8 + 32] 
    movhpd xmm2, [rsi + rbx*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
    
    subpd xmm0, [rsp + nb312_ixO]
    subpd xmm1, [rsp + nb312_iyO]
    subpd xmm2, [rsp + nb312_izO]
    subpd xmm3, [rsp + nb312_ixH1]
    subpd xmm4, [rsp + nb312_iyH1]
    subpd xmm5, [rsp + nb312_izH1]
    subpd xmm6, [rsp + nb312_ixH2]
    subpd xmm7, [rsp + nb312_iyH2]
    subpd xmm8, [rsp + nb312_izH2]
    
	movapd [rsp + nb312_dxOH1], xmm0
	movapd [rsp + nb312_dyOH1], xmm1
	movapd [rsp + nb312_dzOH1], xmm2
	mulpd  xmm0, xmm0
	mulpd  xmm1, xmm1
	mulpd  xmm2, xmm2
	movapd [rsp + nb312_dxH1H1], xmm3
	movapd [rsp + nb312_dyH1H1], xmm4
	movapd [rsp + nb312_dzH1H1], xmm5
	mulpd  xmm3, xmm3
	mulpd  xmm4, xmm4
	mulpd  xmm5, xmm5
	movapd [rsp + nb312_dxH2H1], xmm6
	movapd [rsp + nb312_dyH2H1], xmm7
	movapd [rsp + nb312_dzH2H1], 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 jH1 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 + nb312_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 + nb312_half]
	mulpd   xmm9, xmm15  ;# first iteration for rinvOH1 
	mulpd   xmm10, xmm15 ;# first iteration for rinvH1H1
    mulpd   xmm11, xmm15 ;# first iteration for rinvH2OH1

    ;# 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 + nb312_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 + nb312_half]
	mulpd   xmm9, xmm15  ;#  rinvOH1
	mulpd   xmm10, xmm15 ;#   rinvH1H1
    mulpd   xmm11, xmm15 ;#   rinvH2H1
	
	movapd  [rsp + nb312_rinvOH1], xmm9
	movapd  [rsp + nb312_rinvH1H1], xmm10
	movapd  [rsp + nb312_rinvH2H1], xmm11
	
	;# H1 interactions 
    ;# rsq in xmm0,xmm3,xmm6  
    ;# rinv in xmm9, xmm10, xmm11

    movapd xmm1, [rsp + nb312_tsc]
    mulpd  xmm0, xmm9  ;# r
    mulpd  xmm3, xmm10
    mulpd  xmm6, xmm11
    mulpd  xmm0, xmm1 ;# rtab
    mulpd  xmm3, xmm1
    mulpd  xmm6, xmm1
    
    ;# truncate and convert to integers
    cvttpd2dq xmm1, xmm0
    cvttpd2dq xmm4, xmm3
    cvttpd2dq xmm7, xmm6        

    ;# convert back to float
    cvtdq2pd  xmm2, xmm1
    cvtdq2pd  xmm5, xmm4
    cvtdq2pd  xmm8, xmm7
    
    ;# multiply by 4
    pslld   xmm1, 2
    pslld   xmm4, 2
    pslld   xmm7, 2
    
    ;# move to integer registers
    pshufd xmm13, xmm1, 1
    pshufd xmm14, xmm4, 1
    pshufd xmm15, xmm7, 1
    movd    r8d, xmm1
    movd    r10d, xmm4
    movd    r12d, xmm7
    movd    r9d, xmm13
    movd    r11d, xmm14
    movd    r13d, xmm15
        
    mov  rsi, [rbp + nb312_VFtab]

    ;# calculate eps
    subpd     xmm0, xmm2
    subpd     xmm3, xmm5
    subpd     xmm6, xmm8

    movapd    [rsp + nb312_epsO], xmm0
    movapd    [rsp + nb312_epsH1], xmm3
    movapd    [rsp + nb312_epsH2], xmm6

    ;# Load LOTS of table data
    movlpd xmm0,  [rsi + r8*8]
    movlpd xmm1,  [rsi + r8*8 + 8]
    movlpd xmm2,  [rsi + r8*8 + 16]
    movlpd xmm3,  [rsi + r8*8 + 24]
    movlpd xmm4,  [rsi + r10*8]
    movlpd xmm5,  [rsi + r10*8 + 8]
    movlpd xmm6,  [rsi + r10*8 + 16]