nb_kernel104_x86_64_sse2.intel_syntax.s

来自「最著名最快的分子模拟软件」· S 代码 · 共 2,312 行 · 第 1/5 页

	
	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 + nb104_half]
	mulpd   xmm9, xmm15  ;# first iteration for rinvH1H1 
	mulpd   xmm10, xmm15 ;# first iteration for rinvH2H1
    mulpd   xmm11, xmm15 ;# first iteration for rinvMH1	

    ;# 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 + nb104_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 + nb104_half]
	mulpd   xmm9, xmm15  ;#  rinvH1H1 
	mulpd   xmm10, xmm15 ;#   rinvH2H1
    mulpd   xmm11, xmm15 ;#   rinvMH1
	
	;# H1 interactions 
    movapd xmm0, xmm9
    movapd xmm1, xmm10
    movapd xmm2, xmm11
    mulpd  xmm9, xmm9
    mulpd  xmm10, xmm10
    mulpd  xmm11, xmm11
    mulpd  xmm0, [rsp + nb104_qqHH] 
    mulpd  xmm1, [rsp + nb104_qqHH] 
    mulpd  xmm2, [rsp + nb104_qqMH] 
    mulpd  xmm9, xmm0
    mulpd  xmm10, xmm1
    mulpd  xmm11, xmm2
    
    addpd xmm0, [rsp + nb104_vctot] 
    addpd xmm1, xmm2
    addpd xmm0, xmm1
    movapd [rsp + nb104_vctot], xmm0
    
    ;# move j H1 forces to xmm0-xmm2
	mov   rdi, [rbp + nb104_faction]	
	movlpd xmm0, [rdi + rax*8 + 24]
	movlpd xmm1, [rdi + rax*8 + 32]
	movlpd xmm2, [rdi + rax*8 + 40]
	movhpd xmm0, [rdi + rbx*8 + 24]
	movhpd xmm1, [rdi + rbx*8 + 32]
	movhpd xmm2, [rdi + rbx*8 + 40]

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

	mulpd xmm7, [rsp + nb104_dxH1H1]
	mulpd xmm8, [rsp + nb104_dyH1H1]
	mulpd xmm9, [rsp + nb104_dzH1H1]
	mulpd xmm10, [rsp + nb104_dxH2H1]
	mulpd xmm11, [rsp + nb104_dyH2H1]
	mulpd xmm12, [rsp + nb104_dzH2H1]
	mulpd xmm13, [rsp + nb104_dxMH1]
	mulpd xmm14, [rsp + nb104_dyMH1]
	mulpd xmm15, [rsp + nb104_dzMH1]

    addpd xmm0, xmm7
    addpd xmm1, xmm8
    addpd xmm2, xmm9
    addpd xmm7, [rsp + nb104_fixH1]
    addpd xmm8, [rsp + nb104_fiyH1]
    addpd xmm9, [rsp + nb104_fizH1]

    addpd xmm0, xmm10
    addpd xmm1, xmm11
    addpd xmm2, xmm12
    addpd xmm10, [rsp + nb104_fixH2]
    addpd xmm11, [rsp + nb104_fiyH2]
    addpd xmm12, [rsp + nb104_fizH2]

    addpd xmm0, xmm13
    addpd xmm1, xmm14
    addpd xmm2, xmm15
    addpd xmm13, [rsp + nb104_fixM]
    addpd xmm14, [rsp + nb104_fiyM]
    addpd xmm15, [rsp + nb104_fizM]

    movapd [rsp + nb104_fixH1], xmm7
    movapd [rsp + nb104_fiyH1], xmm8
    movapd [rsp + nb104_fizH1], xmm9
    movapd [rsp + nb104_fixH2], xmm10
    movapd [rsp + nb104_fiyH2], xmm11
    movapd [rsp + nb104_fizH2], xmm12
    movapd [rsp + nb104_fixM], xmm13
    movapd [rsp + nb104_fiyM], xmm14
    movapd [rsp + nb104_fizM], xmm15
   
    ;# store back j H1 forces from xmm0-xmm2
	movlpd [rdi + rax*8 + 24], xmm0
	movlpd [rdi + rax*8 + 32], xmm1
	movlpd [rdi + rax*8 + 40], xmm2
	movhpd [rdi + rbx*8 + 24], xmm0
	movhpd [rdi + rbx*8 + 32], xmm1
	movhpd [rdi + rbx*8 + 40], xmm2

	;# move j H2 coordinates to local temp variables 
	mov   rsi, [rbp + nb104_pos]	
    movlpd xmm0, [rsi + rax*8 + 48] 
    movlpd xmm1, [rsi + rax*8 + 56] 
    movlpd xmm2, [rsi + rax*8 + 64] 
    movhpd xmm0, [rsi + rbx*8 + 48] 
    movhpd xmm1, [rsi + rbx*8 + 56] 
    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 + nb104_ixH1]
    subpd xmm1, [rsp + nb104_iyH1]
    subpd xmm2, [rsp + nb104_izH1]
    subpd xmm3, [rsp + nb104_ixH2]
    subpd xmm4, [rsp + nb104_iyH2]
    subpd xmm5, [rsp + nb104_izH2]
    subpd xmm6, [rsp + nb104_ixM]
    subpd xmm7, [rsp + nb104_iyM]
    subpd xmm8, [rsp + nb104_izM]
    
	movapd [rsp + nb104_dxH1H2], xmm0
	movapd [rsp + nb104_dyH1H2], xmm1
	movapd [rsp + nb104_dzH1H2], xmm2
	mulpd  xmm0, xmm0
	mulpd  xmm1, xmm1
	mulpd  xmm2, xmm2
	movapd [rsp + nb104_dxH2H2], xmm3
	movapd [rsp + nb104_dyH2H2], xmm4
	movapd [rsp + nb104_dzH2H2], xmm5
	mulpd  xmm3, xmm3
	mulpd  xmm4, xmm4
	mulpd  xmm5, xmm5
	movapd [rsp + nb104_dxMH2], xmm6
	movapd [rsp + nb104_dyMH2], xmm7
	movapd [rsp + nb104_dzMH2], 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 + nb104_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 + nb104_half]
	mulpd   xmm9, xmm15  ;# first iteration for rinvH1H2 
	mulpd   xmm10, xmm15 ;# first iteration for rinvH2H2
    mulpd   xmm11, xmm15 ;# first iteration for rinvMH1H2

    ;# 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 + nb104_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 + nb104_half]
	mulpd   xmm9, xmm15  ;#  rinvH1H2
	mulpd   xmm10, xmm15 ;#   rinvH2H2
    mulpd   xmm11, xmm15 ;#   rinvMH2
	
	;# H2 interactions 
    movapd xmm0, xmm9
    movapd xmm1, xmm10
    movapd xmm2, xmm11
    mulpd  xmm9, xmm9
    mulpd  xmm10, xmm10
    mulpd  xmm11, xmm11
    mulpd  xmm0, [rsp + nb104_qqHH] 
    mulpd  xmm1, [rsp + nb104_qqHH] 
    mulpd  xmm2, [rsp + nb104_qqMH] 
    mulpd  xmm9, xmm0
    mulpd  xmm10, xmm1
    mulpd  xmm11, xmm2
    
    addpd xmm0, [rsp + nb104_vctot] 
    addpd xmm1, xmm2
    addpd xmm0, xmm1
    movapd [rsp + nb104_vctot], xmm0
    
    ;# move j H2 forces to xmm0-xmm2
	mov   rdi, [rbp + nb104_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 + nb104_dxH1H2]
	mulpd xmm8, [rsp + nb104_dyH1H2]
	mulpd xmm9, [rsp + nb104_dzH1H2]
	mulpd xmm10, [rsp + nb104_dxH2H2]
	mulpd xmm11, [rsp + nb104_dyH2H2]
	mulpd xmm12, [rsp + nb104_dzH2H2]
	mulpd xmm13, [rsp + nb104_dxMH2]
	mulpd xmm14, [rsp + nb104_dyMH2]
	mulpd xmm15, [rsp + nb104_dzMH2]

    addpd xmm0, xmm7
    addpd xmm1, xmm8
    addpd xmm2, xmm9
    addpd xmm7, [rsp + nb104_fixH1]
    addpd xmm8, [rsp + nb104_fiyH1]
    addpd xmm9, [rsp + nb104_fizH1]

    addpd xmm0, xmm10
    addpd xmm1, xmm11
    addpd xmm2, xmm12
    addpd xmm10, [rsp + nb104_fixH2]
    addpd xmm11, [rsp + nb104_fiyH2]
    addpd xmm12, [rsp + nb104_fizH2]

    addpd xmm0, xmm13
    addpd xmm1, xmm14
    addpd xmm2, xmm15
    addpd xmm13, [rsp + nb104_fixM]
    addpd xmm14, [rsp + nb104_fiyM]
    addpd xmm15, [rsp + nb104_fizM]

    movapd [rsp + nb104_fixH1], xmm7
    movapd [rsp + nb104_fiyH1], xmm8
    movapd [rsp + nb104_fizH1], xmm9
    movapd [rsp + nb104_fixH2], xmm10
    movapd [rsp + nb104_fiyH2], xmm11
    movapd [rsp + nb104_fizH2], xmm12
    movapd [rsp + nb104_fixM], xmm13
    movapd [rsp + nb104_fiyM], xmm14
    movapd [rsp + nb104_fizM], 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
       
	;# move j M coordinates to local temp variables 
	mov   rsi, [rbp + nb104_pos]	
    movlpd xmm0, [rsi + rax*8 + 72] 
    movlpd xmm1, [rsi + rax*8 + 80] 
    movlpd xmm2, [rsi + rax*8 + 88] 
    movhpd xmm0, [rsi + rbx*8 + 72] 
    movhpd xmm1, [rsi + rbx*8 + 80] 
    movhpd xmm2, [rsi + rbx*8 + 88] 

    ;# xmm0 = Mx
    ;# xmm1 = My
    ;# xmm2 = Mz
        
    movapd xmm3, xmm0
    movapd xmm4, xmm1
    movapd xmm5, xmm2
    movapd xmm6, xmm0
    movapd xmm7, xmm1
    movapd xmm8, xmm2
    
    subpd xmm0, [rsp + nb104_ixH1]
    subpd xmm1, [rsp + nb104_iyH1]
    subpd xmm2, [rsp + nb104_izH1]
    subpd xmm3, [rsp + nb104_ixH2]
    subpd xmm4, [rsp + nb104_iyH2]
    subpd xmm5, [rsp + nb104_izH2]
    subpd xmm6, [rsp + nb104_ixM]
    subpd xmm7, [rsp + nb104_iyM]
    subpd xmm8, [rsp + nb104_izM]
    
	movapd [rsp + nb104_dxH1M], xmm0
	movapd [rsp + nb104_dyH1M], xmm1
	movapd [rsp + nb104_dzH1M], xmm2
	mulpd  xmm0, xmm0
	mulpd  xmm1, xmm1
	mulpd  xmm2, xmm2
	movapd [rsp + nb104_dxH2M], xmm3
	movapd [rsp + nb104_dyH2M], xmm4
	movapd [rsp + nb104_dzH2M], xmm5
	mulpd  xmm3, xmm3
	mulpd  xmm4, xmm4
	mulpd  xmm5, xmm5
	movapd [rsp + nb104_dxMM], xmm6
	movapd [rsp + nb104_dyMM], xmm7
	movapd [rsp + nb104_dzMM], 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 jM 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 + nb104_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 + nb104_half]
	mulpd   xmm9, xmm15  ;# first iteration for rinvH1M 
	mulpd   xmm10, xmm15 ;# first iteration for rinvH2M
    mulpd   xmm11, xmm15 ;# first iteration for rinvMM

    ;# 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 + nb104_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 + nb104_half]
	mulpd   xmm9, xmm15  ;#  rinvH1M
	mulpd   xmm10, xmm15 ;#   rinvH2M
    mulpd   xmm11, xmm15 ;#   rinvMM