nb_kernel111_ia32_sse2.intel_syntax.s

来自「最著名最快的分子模拟软件」· S 代码 · 共 1,864 行 · 第 1/4 页

	mulpd   xmm4, [esp + nb111_half] ;# iter1 ( new lu) 

	movapd xmm3, xmm4
	mulpd xmm4, xmm4	;# lu*lu 
	mulpd xmm7, xmm4	;# rsq*lu*lu 
	movapd xmm4, [esp + nb111_three]
	subpd xmm4, xmm7	;# 3-rsq*lu*lu 
	mulpd xmm4, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulpd xmm4, [esp + nb111_half] ;# rinv 
	movapd  xmm7, xmm4	;# rinvO in xmm7 
	
	;# rsqH1 - seed in xmm2 
	cvtpd2ps xmm2, xmm6	
	rsqrtps xmm2, xmm2
	cvtps2pd xmm2, xmm2

	movapd  xmm3, xmm2
	mulpd   xmm2, xmm2
	movapd  xmm4, [esp + nb111_three]
	mulpd   xmm2, xmm6	;# rsq*lu*lu 
	subpd   xmm4, xmm2	;# 30-rsq*lu*lu 
	mulpd   xmm4, xmm3	;# lu*(3-rsq*lu*lu) 
	mulpd   xmm4, [esp + nb111_half] ;# iter1 ( new lu) 

	movapd xmm3, xmm4
	mulpd xmm4, xmm4	;# lu*lu 
	mulpd xmm6, xmm4	;# rsq*lu*lu 
	movapd xmm4, [esp + nb111_three]
	subpd xmm4, xmm6	;# 3-rsq*lu*lu 
	mulpd xmm4, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulpd xmm4, [esp + nb111_half] ;# rinv 
	movapd  xmm6, xmm4	;# rinvH1 in xmm6 
	
	;# rsqH2 - seed in xmm2 
	cvtpd2ps xmm2, xmm5	
	rsqrtps xmm2, xmm2
	cvtps2pd xmm2, xmm2

	movapd  xmm3, xmm2
	mulpd   xmm2, xmm2
	movapd  xmm4, [esp + nb111_three]
	mulpd   xmm2, xmm5	;# rsq*lu*lu 
	subpd   xmm4, xmm2	;# 30-rsq*lu*lu 
	mulpd   xmm4, xmm3	;# lu*(3-rsq*lu*lu) 
	mulpd   xmm4, [esp + nb111_half] ;# iter1 ( new lu) 

	movapd xmm3, xmm4
	mulpd xmm4, xmm4	;# lu*lu 
	mulpd xmm5, xmm4	;# rsq*lu*lu 
	movapd xmm4, [esp + nb111_three]
	subpd xmm4, xmm5	;# 3-rsq*lu*lu 
	mulpd xmm4, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulpd xmm4, [esp + nb111_half] ;# rinv 
	movapd  xmm5, xmm4	;# rinvH2 in xmm5 

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

	addpd  xmm7, [esp + nb111_vctot]
	
	movapd [esp + nb111_Vvdwtot], xmm3
	movapd [esp + nb111_vctot], xmm7

	movapd xmm0, [esp + nb111_dxO]
	movapd xmm1, [esp + nb111_dyO]
	movapd xmm2, [esp + nb111_dzO]
	mulpd  xmm0, xmm4
	mulpd  xmm1, xmm4
	mulpd  xmm2, xmm4

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

	;# H1 interactions 
	movapd  xmm4, xmm6	
	mulpd   xmm4, xmm4	;# xmm6=rinv, xmm4=rinvsq 
	mulpd  xmm6, [esp + nb111_qqH]	;# xmm6=vcoul 
	mulpd  xmm4, xmm6		;# total fsH1 in xmm4 
	
	addpd  xmm6, [esp + nb111_vctot]

	movapd xmm0, [esp + nb111_dxH1]
	movapd xmm1, [esp + nb111_dyH1]
	movapd xmm2, [esp + nb111_dzH1]
	movapd [esp + nb111_vctot], xmm6
	mulpd  xmm0, xmm4
	mulpd  xmm1, xmm4
	mulpd  xmm2, xmm4

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

	;# H2 interactions 
	movapd  xmm4, xmm5	
	mulpd   xmm4, xmm4	;# xmm5=rinv, xmm4=rinvsq 
	mulpd  xmm5, [esp + nb111_qqH]	;# xmm5=vcoul 
	mulpd  xmm4, xmm5		;# total fsH1 in xmm4 
	
	addpd  xmm5, [esp + nb111_vctot]

	movapd xmm0, [esp + nb111_dxH2]
	movapd xmm1, [esp + nb111_dyH2]
	movapd xmm2, [esp + nb111_dzH2]
	movapd [esp + nb111_vctot], xmm5
	mulpd  xmm0, xmm4
	mulpd  xmm1, xmm4
	mulpd  xmm2, xmm4

	;# update H2 forces 
	movapd xmm3, [esp + nb111_fixH2]
	movapd xmm4, [esp + nb111_fiyH2]
	movapd xmm7, [esp + nb111_fizH2]
	addpd  xmm3, xmm0
	addpd  xmm4, xmm1
	addpd  xmm7, xmm2
	movapd [esp + nb111_fixH2], xmm3
	movapd [esp + nb111_fiyH2], xmm4
	movapd [esp + nb111_fizH2], xmm7

	mov edi, [ebp + nb111_faction]
	;# update j forces 
	addpd  xmm0, [esp + nb111_fjx]
	addpd  xmm1, [esp + nb111_fjy]
	addpd  xmm2, [esp + nb111_fjz]
	movlpd xmm3, [edi + eax*8]
	movlpd xmm4, [edi + eax*8 + 8]
	movlpd xmm5, [edi + eax*8 + 16]
	movhpd xmm3, [edi + ebx*8]
	movhpd xmm4, [edi + ebx*8 + 8]
	movhpd xmm5, [edi + ebx*8 + 16]
	subpd xmm3, xmm0
	subpd xmm4, xmm1
	subpd xmm5, xmm2
	movlpd [edi + eax*8], xmm3
	movlpd [edi + eax*8 + 8], xmm4
	movlpd [edi + eax*8 + 16], xmm5
	movhpd [edi + ebx*8], xmm3
	movhpd [edi + ebx*8 + 8], xmm4
	movhpd [edi + ebx*8 + 16], xmm5
	
	;# should we do one more iteration? 
	sub dword ptr [esp + nb111_innerk],  2
	jl   .nb111_checksingle
	jmp  .nb111_unroll_loop
.nb111_checksingle:	
	mov   edx, [esp + nb111_innerk]
	and   edx, 1
	jnz  .nb111_dosingle
	jmp  .nb111_updateouterdata
.nb111_dosingle:
	mov   edx, [esp + nb111_innerjjnr]     ;# pointer to jjnr[k] 
	mov   eax, [edx]	
	add dword ptr [esp + nb111_innerjjnr],  4	

	mov esi, [ebp + nb111_charge]    ;# base of charge[] 

	xorpd xmm3, xmm3
	movlpd xmm3, [esi + eax*8]
	movapd xmm4, xmm3
	mulpd  xmm3, [esp + nb111_iqO]
	mulpd  xmm4, [esp + nb111_iqH]

	movd  mm0, eax		;# use mmx registers as temp storage 

	movapd  [esp + nb111_qqO], xmm3
	movapd  [esp + nb111_qqH], xmm4
	
	mov esi, [ebp + nb111_type]
	mov eax, [esi + eax*4]
	mov esi, [ebp + nb111_vdwparam]
	shl eax, 1	
	mov edi, [esp + nb111_ntia]
	add eax, edi

	movlpd xmm6, [esi + eax*8]	;# c6a
	movhpd xmm6, [esi + eax*8 + 8]	;# c6a c12a 

	xorpd xmm7, xmm7
	movapd xmm4, xmm6
	unpcklpd xmm4, xmm7
	unpckhpd xmm6, xmm7
	
	movd  eax, mm0
	movd  ebx, mm1
	movapd [esp + nb111_c6], xmm4
	movapd [esp + nb111_c12], xmm6
	
	mov esi, [ebp + nb111_pos]       ;# base of pos[] 

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

	;# move coordinates to xmm0-xmm2 
	movlpd xmm0, [esi + eax*8]
	movlpd xmm1, [esi + eax*8 + 8]
	movlpd xmm2, [esi + eax*8 + 16]

	;# move ixO-izO to xmm4-xmm6 
	movapd xmm4, [esp + nb111_ixO]
	movapd xmm5, [esp + nb111_iyO]
	movapd xmm6, [esp + nb111_izO]

	;# calc dr 
	subsd xmm4, xmm0
	subsd xmm5, xmm1
	subsd xmm6, xmm2

	;# store dr 
	movapd [esp + nb111_dxO], xmm4
	movapd [esp + nb111_dyO], xmm5
	movapd [esp + 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 ixH1-izH1 to xmm4-xmm6 
	movapd xmm4, [esp + nb111_ixH1]
	movapd xmm5, [esp + nb111_iyH1]
	movapd xmm6, [esp + nb111_izH1]

	;# calc dr 
	subsd xmm4, xmm0
	subsd xmm5, xmm1
	subsd xmm6, xmm2

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

	;# move ixH2-izH2 to xmm3-xmm5  
	movapd xmm3, [esp + nb111_ixH2]
	movapd xmm4, [esp + nb111_iyH2]
	movapd xmm5, [esp + nb111_izH2]

	;# calc dr 
	subsd xmm3, xmm0
	subsd xmm4, xmm1
	subsd xmm5, xmm2

	;# store dr 
	movapd [esp + nb111_dxH2], xmm3
	movapd [esp + nb111_dyH2], xmm4
	movapd [esp + 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, [esp + 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, [esp + nb111_half] ;# iter1 ( new lu) 

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

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

	movapd xmm3, xmm4
	mulsd xmm4, xmm4	;# lu*lu 
	mulsd xmm5, xmm4	;# rsq*lu*lu 
	movapd xmm4, [esp + nb111_three]
	subsd xmm4, xmm5	;# 3-rsq*lu*lu 
	mulsd xmm4, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulsd xmm4, [esp + 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, [esp + nb111_qqO]	;# xmm7=vcoul 
	
	mulsd  xmm1, [esp + nb111_c6]
	mulsd  xmm2, [esp + nb111_c12]
	movapd xmm3, xmm2
	subsd  xmm3, xmm1	;# Vvdw=Vvdw12-Vvdw6 		
	addsd  xmm3, [esp + nb111_Vvdwtot]
	mulsd  xmm1, [esp + nb111_six]
	mulsd  xmm2, [esp + nb111_twelve]
	subsd  xmm2, xmm1
	addsd  xmm2, xmm7	
	mulsd  xmm4, xmm2	;# total fsO in xmm4 

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

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

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

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

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

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

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

	movapd xmm0, [esp + nb111_dxH2]
	movapd xmm1, [esp + nb111_dyH2]