nb_kernel111_x86_64_sse2.intel_syntax.s

来自「最著名最快的分子模拟软件」· S 代码 · 共 1,816 行 · 第 1/4 页

	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 + nb111_half]
	mulpd   xmm9, xmm15  ;# first iteration for rinvO
	mulpd   xmm10, xmm15 ;# first iteration for rinvH1
    mulpd   xmm11, xmm15 ;# first iteration for rinvH2

    ;# 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 + nb111_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 + nb111_half]
	mulpd   xmm9, xmm15  ;#  rinvO 
	mulpd   xmm10, xmm15 ;#   rinvH1
    mulpd   xmm11, xmm15 ;#   rinvH2
	
	;# interactions 
    movapd xmm0, xmm9
    movapd xmm1, xmm10
    movapd xmm2, xmm11
    mulpd  xmm9, xmm9    ;# rinvsq
    mulpd  xmm10, xmm10
    mulpd  xmm11, xmm11
    movapd xmm12, xmm9
    mulpd  xmm12, xmm12 ;# rinv4
    mulpd  xmm12, xmm9  ;# rinv6
    mulpd  xmm0, [rsp + nb111_qqO] 
    mulpd  xmm1, [rsp + nb111_qqH] 
    mulpd  xmm2, [rsp + nb111_qqH] 
    movapd xmm13, xmm12 ;# rinv6
    mulpd  xmm12, xmm12 ;# rinv12
	mulpd  xmm13, [rsp + nb111_c6]
	mulpd  xmm12, [rsp + nb111_c12]
    movapd xmm14, xmm12
    subpd  xmm14, xmm13
    
	addpd  xmm14, [rsp + nb111_Vvdwtot]
	mulpd  xmm13, [rsp + nb111_six]
	mulpd  xmm12, [rsp + nb111_twelve]
	movapd [rsp + nb111_Vvdwtot], xmm14
    subpd  xmm12, xmm13 ;# LJ fscal        

    addpd  xmm12, xmm0
    
    mulpd  xmm9, xmm12
    mulpd  xmm10, xmm1
    mulpd  xmm11, xmm2
    
    addpd xmm0, [rsp + nb111_vctot] 
    addpd xmm1, xmm2
    addpd xmm0, xmm1
    movapd [rsp + nb111_vctot], xmm0
    
    ;# move j forces to xmm0-xmm2
    mov rdi, [rbp + nb111_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 xmm7, xmm9
    movapd xmm8, xmm9
    movapd xmm13, xmm11
    movapd xmm14, xmm11
    movapd xmm15, xmm11
    movapd xmm11, xmm10
    movapd xmm12, xmm10

	mulpd xmm7, [rsp + nb111_dxO]
	mulpd xmm8, [rsp + nb111_dyO]
	mulpd xmm9, [rsp + nb111_dzO]
	mulpd xmm10, [rsp + nb111_dxH1]
	mulpd xmm11, [rsp + nb111_dyH1]
	mulpd xmm12, [rsp + nb111_dzH1]
	mulpd xmm13, [rsp + nb111_dxH2]
	mulpd xmm14, [rsp + nb111_dyH2]
	mulpd xmm15, [rsp + nb111_dzH2]

    addpd xmm0, xmm7
    addpd xmm1, xmm8
    addpd xmm2, xmm9
    addpd xmm7, [rsp + nb111_fixO]
    addpd xmm8, [rsp + nb111_fiyO]
    addpd xmm9, [rsp + nb111_fizO]

    addpd xmm0, xmm10
    addpd xmm1, xmm11
    addpd xmm2, xmm12
    addpd xmm10, [rsp + nb111_fixH1]
    addpd xmm11, [rsp + nb111_fiyH1]
    addpd xmm12, [rsp + nb111_fizH1]

    addpd xmm0, xmm13
    addpd xmm1, xmm14
    addpd xmm2, xmm15
    addpd xmm13, [rsp + nb111_fixH2]
    addpd xmm14, [rsp + nb111_fiyH2]
    addpd xmm15, [rsp + nb111_fizH2]

    movapd [rsp + nb111_fixO], xmm7
    movapd [rsp + nb111_fiyO], xmm8
    movapd [rsp + nb111_fizO], xmm9
    movapd [rsp + nb111_fixH1], xmm10
    movapd [rsp + nb111_fiyH1], xmm11
    movapd [rsp + nb111_fizH1], xmm12
    movapd [rsp + nb111_fixH2], xmm13
    movapd [rsp + nb111_fiyH2], xmm14
    movapd [rsp + nb111_fizH2], xmm15
   
    ;# store back j 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

	;# should we do one more iteration? 
	sub dword ptr [rsp + nb111_innerk],  2
	jl   .nb111_checksingle
	jmp  .nb111_unroll_loop
.nb111_checksingle:	
	mov   edx, [rsp + nb111_innerk]
	and   edx, 1
	jnz  .nb111_dosingle
	jmp  .nb111_updateouterdata
.nb111_dosingle:
	mov   rdx, [rsp + nb111_innerjjnr]     ;# pointer to jjnr[k] 
	mov   eax, [rdx]	
	add qword ptr [rsp + nb111_innerjjnr],  4	

	mov rsi, [rbp + nb111_charge]    ;# base of charge[] 

	xorpd xmm3, xmm3
	movlpd xmm3, [rsi + rax*8]
	movapd xmm4, xmm3
	mulpd  xmm3, [rsp + nb111_iqO]
	mulpd  xmm4, [rsp + nb111_iqH]

	movapd  [rsp + nb111_qqO], xmm3
	movapd  [rsp + nb111_qqH], xmm4
	
	mov rsi, [rbp + nb111_type]
	mov r8d, [rsi + rax*4]
	mov rsi, [rbp + nb111_vdwparam]
	shl r8d, 1	
	mov edi, [rsp + nb111_ntia]
	add r8d, edi

	movlpd xmm6, [rsi + r8*8]	;# c6a
	movhpd xmm6, [rsi + r8*8 + 8]	;# c6a c12a 

	xorpd xmm7, xmm7
	movapd xmm4, xmm6
	unpcklpd xmm4, xmm7
	unpckhpd xmm6, xmm7
	
	movapd [rsp + nb111_c6], xmm4
	movapd [rsp + nb111_c12], xmm6
	
	mov rsi, [rbp + nb111_pos]       ;# base of pos[] 

	lea   rax, [rax + rax*2]     ;# replace jnr with j3 

	;# move coordinates to xmm0-xmm2 	
	movlpd xmm4, [rsi + rax*8]
	movlpd xmm5, [rsi + rax*8 + 8]
	movlpd xmm6, [rsi + rax*8 + 16]
    movapd xmm0, xmm4
    movapd xmm1, xmm5
    movapd xmm2, xmm6

	;# calc dr 
	subsd xmm4, [rsp + nb111_ixO]
	subsd xmm5, [rsp + nb111_iyO]
	subsd xmm6, [rsp + nb111_izO]

	;# store dr 
	movapd [rsp + nb111_dxO], xmm4
	movapd [rsp + nb111_dyO], xmm5
	movapd [rsp + nb111_dzO], xmm6
	;# square it 
	mulsd xmm4,xmm4
	mulsd xmm5,xmm5
	mulsd xmm6,xmm6
	addsd xmm4, xmm5
	addsd xmm4, xmm6
	movapd xmm7, xmm4
	;# rsqO in xmm7 

	;# move j coords to xmm4-xmm6 
	movapd xmm4, xmm0
	movapd xmm5, xmm1
	movapd xmm6, xmm2

	;# calc dr 
	subsd xmm4, [rsp + nb111_ixH1]
	subsd xmm5, [rsp + nb111_iyH1]
	subsd xmm6, [rsp + nb111_izH1]

	;# store dr 
	movapd [rsp + nb111_dxH1], xmm4
	movapd [rsp + nb111_dyH1], xmm5
	movapd [rsp + nb111_dzH1], xmm6
	;# square it 
	mulsd xmm4,xmm4
	mulsd xmm5,xmm5
	mulsd xmm6,xmm6
	addsd xmm6, xmm5
	addsd xmm6, xmm4
	;# rsqH1 in xmm6 

	;# move j coords to xmm3-xmm5
	movapd xmm3, xmm0
	movapd xmm4, xmm1
	movapd xmm5, xmm2

	;# calc dr 
	subsd xmm3, [rsp + nb111_ixH2]
	subsd xmm4, [rsp + nb111_iyH2]
	subsd xmm5, [rsp + nb111_izH2]

	;# store dr 
	movapd [rsp + nb111_dxH2], xmm3
	movapd [rsp + nb111_dyH2], xmm4
	movapd [rsp + nb111_dzH2], xmm5
	;# square it 
	mulsd xmm3,xmm3
	mulsd xmm4,xmm4
	mulsd xmm5,xmm5
	addsd xmm5, xmm4
	addsd xmm5, xmm3
	;# rsqH2 in xmm5, rsqH1 in xmm6, rsqO in xmm7 

	;# start with rsqO - put seed in xmm2 
	cvtsd2ss xmm2, xmm7	
	rsqrtss xmm2, xmm2
	cvtss2sd xmm2, xmm2

	movapd  xmm3, xmm2
	mulsd   xmm2, xmm2
	movapd  xmm4, [rsp + nb111_three]
	mulsd   xmm2, xmm7	;# rsq*lu*lu 
	subsd   xmm4, xmm2	;# 30-rsq*lu*lu 
	mulsd   xmm4, xmm3	;# lu*(3-rsq*lu*lu) 
	mulsd   xmm4, [rsp + nb111_half] ;# iter1 ( new lu) 

	movapd xmm3, xmm4
	mulsd xmm4, xmm4	;# lu*lu 
	mulsd xmm7, xmm4	;# rsq*lu*lu 
	movapd xmm4, [rsp + nb111_three]
	subsd xmm4, xmm7	;# 3-rsq*lu*lu 
	mulsd xmm4, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulsd xmm4, [rsp + nb111_half] ;# rinv 
	movapd  xmm7, xmm4	;# rinvO in xmm7 
	
	;# rsqH1 - seed in xmm2 
	cvtsd2ss xmm2, xmm6	
	rsqrtss xmm2, xmm2
	cvtss2sd xmm2, xmm2

	movapd  xmm3, xmm2
	mulsd   xmm2, xmm2
	movapd  xmm4, [rsp + nb111_three]
	mulsd   xmm2, xmm6	;# rsq*lu*lu 
	subsd   xmm4, xmm2	;# 30-rsq*lu*lu 
	mulsd   xmm4, xmm3	;# lu*(3-rsq*lu*lu) 
	mulsd   xmm4, [rsp + nb111_half] ;# iter1 ( new lu) 

	movapd xmm3, xmm4
	mulsd xmm4, xmm4	;# lu*lu 
	mulsd xmm6, xmm4	;# rsq*lu*lu 
	movapd xmm4, [rsp + nb111_three]
	subsd xmm4, xmm6	;# 3-rsq*lu*lu 
	mulsd xmm4, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulsd xmm4, [rsp + nb111_half] ;# rinv 
	movapd  xmm6, xmm4	;# rinvH1 in xmm6 
	
	;# rsqH2 - seed in xmm2 
	cvtsd2ss xmm2, xmm5	
	rsqrtss xmm2, xmm2
	cvtss2sd xmm2, xmm2

	movapd  xmm3, xmm2
	mulsd   xmm2, xmm2
	movapd  xmm4, [rsp + nb111_three]
	mulsd   xmm2, xmm5	;# rsq*lu*lu 
	subsd   xmm4, xmm2	;# 30-rsq*lu*lu 
	mulsd   xmm4, xmm3	;# lu*(3-rsq*lu*lu) 
	mulsd   xmm4, [rsp + nb111_half] ;# iter1 ( new lu) 

	movapd xmm3, xmm4
	mulsd xmm4, xmm4	;# lu*lu 
	mulsd xmm5, xmm4	;# rsq*lu*lu 
	movapd xmm4, [rsp + nb111_three]
	subsd xmm4, xmm5	;# 3-rsq*lu*lu 
	mulsd xmm4, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulsd xmm4, [rsp + nb111_half] ;# rinv 
	movapd  xmm5, xmm4	;# rinvH2 in xmm5 

	;# do O interactions 
	movapd  xmm4, xmm7	
	mulsd   xmm4, xmm4	;# xmm7=rinv, xmm4=rinvsq 
	movapd xmm1, xmm4
	mulsd  xmm1, xmm4
	mulsd  xmm1, xmm4	;# xmm1=rinvsix 
	movapd xmm2, xmm1
	mulsd  xmm2, xmm2	;# xmm2=rinvtwelve 
	mulsd  xmm7, [rsp + nb111_qqO]	;# xmm7=vcoul 
	
	mulsd  xmm1, [rsp + nb111_c6]
	mulsd  xmm2, [rsp + nb111_c12]
	movapd xmm3, xmm2
	subsd  xmm3, xmm1	;# Vvdw=Vvdw12-Vvdw6 		
	addsd  xmm3, [rsp + nb111_Vvdwtot]
	mulsd  xmm1, [rsp + nb111_six]
	mulsd  xmm2, [rsp + nb111_twelve]
	subsd  xmm2, xmm1
	addsd  xmm2, xmm7	
	mulsd  xmm4, xmm2	;# total fsO in xmm4 

	addsd  xmm7, [rsp + nb111_vctot]
	
	movsd [rsp + nb111_Vvdwtot], xmm3
	movsd [rsp + nb111_vctot], xmm7

	movapd xmm0, [rsp + nb111_dxO]
	movapd xmm1, [rsp + nb111_dyO]
	movapd xmm2, [rsp + nb111_dzO]
	mulsd  xmm0, xmm4
	mulsd  xmm1, xmm4
	mulsd  xmm2, xmm4

	;# update O forces 
	movapd xmm3, [rsp + nb111_fixO]
	movapd xmm4, [rsp + nb111_fiyO]
	movapd xmm7, [rsp + nb111_fizO]
	addsd  xmm3, xmm0
	addsd  xmm4, xmm1
	addsd  xmm7, xmm2
	movsd [rsp + nb111_fixO], xmm3
	movsd [rsp + nb111_fiyO], xmm4
	movsd [rsp + nb111_fizO], xmm7
	;# update j forces with water O 
	movsd [rsp + nb111_fjx], xmm0
	movsd [rsp + nb111_fjy], xmm1
	movsd [rsp + nb111_fjz], xmm2

	;# H1 interactions 
	movapd  xmm4, xmm6	
	mulsd   xmm4, xmm4	;# xmm6=rinv, xmm4=rinvsq 
	mulsd  xmm6, [rsp + nb111_qqH]	;# xmm6=vcoul 
	mulsd  xmm4, xmm6		;# total fsH1 in xmm4 
	
	addsd  xmm6, [rsp + nb111_vctot]

	movapd xmm0, [rsp + nb111_dxH1]
	movapd xmm1, [rsp + nb111_dyH1]
	movapd xmm2, [rsp + nb111_dzH1]
	movsd [rsp + nb111_vctot], xmm6
	mulsd  xmm0, xmm4
	mulsd  xmm1, xmm4
	mulsd  xmm2, xmm4

	;# update H1 forces 
	movapd xmm3, [rsp + nb111_fixH1]
	movapd xmm4, [rsp + nb111_fiyH1]
	movapd xmm7, [rsp + nb111_fizH1]
	addsd  xmm3, xmm0
	addsd  xmm4, xmm1
	addsd  xmm7, xmm2
	movsd [rsp + nb111_fixH1], xmm3
	movsd [rsp + nb111_fiyH1], xmm4
	movsd [rsp + nb111_fizH1], xmm7
	;# update j forces with water H1 
	addsd  xmm0, [rsp + nb111_fjx]
	addsd  xmm1, [rsp + nb111_fjy]
	addsd  xmm2, [rsp + nb111_fjz]
	movsd [rsp + nb111_fjx], xmm0
	movsd [rsp + nb111_fjy], xmm1
	movsd [rsp + nb111_fjz], xmm2

	;# H2 interactions 
	movapd  xmm4, xmm5	
	mulsd   xmm4, xmm4	;# xmm5=rinv, xmm4=rinvsq 
	mulsd  xmm5, [rsp + nb111_qqH]	;# xmm5=vcoul 
	mulsd  xmm4, xmm5		;# total fsH1 in xmm4 
	
	addsd  xmm5, [rsp + nb111_vctot]

	movapd xmm0, [rsp + nb111_dxH2]
	movapd xmm1, [rsp + nb111_dyH2]
	movapd xmm2, [rsp + nb111_dzH2]
	movsd [rsp + nb111_vctot], xmm5
	mulsd  xmm0, xmm4
	mulsd  xmm1, xmm4
	mulsd  xmm2, xmm4

	;# update H2 forces 
	movapd xmm3, [rsp + nb111_fixH2]
	movapd xmm4, [rsp + nb111_fiyH2]
	movapd xmm7, [rsp + nb111_fizH2]
	addsd  xmm3, xmm0
	addsd  xmm4, xmm1
	addsd  xmm7, xmm2
	movsd [rsp + nb111_fixH2], xmm3
	movsd [rsp + nb111_fiyH2], xmm4
	movsd [rsp + nb111_fizH2], xmm7

	mov rdi, [rbp + nb111_faction]
	;# update j forces 
	addsd  xmm0, [rsp + nb111_fjx]
	addsd  xmm1, [rsp + nb111_fjy]
	addsd  xmm2, [rsp + nb111_fjz]
	movlpd xmm3, [rdi + rax*8]
	movlpd xmm4, [rdi + rax*8 + 8]
	movlpd xmm5, [rdi + rax*8 + 16]
	addsd xmm3, xmm0
	addsd xmm4, xmm1
	addsd xmm5, xmm2