nb_kernel201_ia32_sse2.intel_syntax.s

来自「最著名最快的分子模拟软件」· S 代码 · 共 1,790 行 · 第 1/4 页

	mulpd   xmm2, xmm7	;# rsq*lu*lu 
	subpd   xmm4, xmm2	;# 30-rsq*lu*lu 
	mulpd   xmm4, xmm3	;# lu*(3-rsq*lu*lu) 
	mulpd   xmm4, [esp + nb201_half] ;# iter1 ( new lu) 

	movapd xmm3, xmm4
	mulpd xmm4, xmm4	;# lu*lu 
	mulpd xmm7, xmm4	;# rsq*lu*lu 
	movapd xmm4, [esp + nb201_three]
	subpd xmm4, xmm7	;# 3-rsq*lu*lu 
	mulpd xmm4, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulpd xmm4, [esp + nb201_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 + nb201_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 + nb201_half] ;# iter1 ( new lu) 

	movapd xmm3, xmm4
	mulpd xmm4, xmm4	;# lu*lu 
	mulpd xmm6, xmm4	;# rsq*lu*lu 
	movapd xmm4, [esp + nb201_three]
	subpd xmm4, xmm6	;# 3-rsq*lu*lu 
	mulpd xmm4, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulpd xmm4, [esp + nb201_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 + nb201_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 + nb201_half] ;# iter1 ( new lu) 

	movapd xmm3, xmm4
	mulpd xmm4, xmm4	;# lu*lu 
	mulpd xmm5, xmm4	;# rsq*lu*lu 
	movapd xmm4, [esp + nb201_three]
	subpd xmm4, xmm5	;# 3-rsq*lu*lu 
	mulpd xmm4, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulpd xmm4, [esp + nb201_half] ;# rinv 
	movapd  xmm5, xmm4	;# rinvH2 in xmm5 

	;# do O interactions 
	movapd  xmm4, xmm7	
	mulpd   xmm4, xmm4	;# xmm6=rinv, xmm4=rinvsq 
	movapd  xmm3, xmm7
	movapd  xmm0, [esp + nb201_krsqO]
	addpd   xmm7, xmm0	;# xmm6=rinv+ krsq 
	mulpd   xmm0, [esp + nb201_two]
	subpd   xmm7, [esp + nb201_crf]
	subpd   xmm3, xmm0	;# xmm7=rinv-2*krsq 
	mulpd   xmm7, [esp + nb201_qqO] ;# vcoul 
	mulpd   xmm3, [esp + nb201_qqO]
	mulpd  xmm4, xmm3	;# total fsH1 in xmm4 
	
	addpd  xmm7, [esp + nb201_vctot]

	movapd xmm0, [esp + nb201_dxO]
	movapd xmm1, [esp + nb201_dyO]
	movapd xmm2, [esp + nb201_dzO]
	movapd [esp + nb201_vctot], xmm7
	mulpd  xmm0, xmm4
	mulpd  xmm1, xmm4
	mulpd  xmm2, xmm4
	
	;# update O forces 
	movapd xmm3, [esp + nb201_fixO]
	movapd xmm4, [esp + nb201_fiyO]
	movapd xmm7, [esp + nb201_fizO]
	addpd  xmm3, xmm0
	addpd  xmm4, xmm1
	addpd  xmm7, xmm2
	movapd [esp + nb201_fixO], xmm3
	movapd [esp + nb201_fiyO], xmm4
	movapd [esp + nb201_fizO], xmm7
	;# update j forces with water O 
	movapd [esp + nb201_fjx], xmm0
	movapd [esp + nb201_fjy], xmm1
	movapd [esp + nb201_fjz], xmm2

	;# H1 interactions 
	movapd  xmm4, xmm6	
	mulpd   xmm4, xmm4	;# xmm6=rinv, xmm4=rinvsq 
	movapd  xmm7, xmm6
	movapd  xmm0, [esp + nb201_krsqH1]
	addpd   xmm6, xmm0	;# xmm6=rinv+ krsq 
	mulpd   xmm0, [esp + nb201_two]
	subpd   xmm6, [esp + nb201_crf]
	subpd   xmm7, xmm0	;# xmm7=rinv-2*krsq 
	mulpd   xmm6, [esp + nb201_qqH] ;# vcoul 
	mulpd   xmm7, [esp + nb201_qqH]
	mulpd  xmm4, xmm7		;# total fsH1 in xmm4 
	
	addpd  xmm6, [esp + nb201_vctot]

	movapd xmm0, [esp + nb201_dxH1]
	movapd xmm1, [esp + nb201_dyH1]
	movapd xmm2, [esp + nb201_dzH1]
	movapd [esp + nb201_vctot], xmm6
	mulpd  xmm0, xmm4
	mulpd  xmm1, xmm4
	mulpd  xmm2, xmm4

	;# update H1 forces 
	movapd xmm3, [esp + nb201_fixH1]
	movapd xmm4, [esp + nb201_fiyH1]
	movapd xmm7, [esp + nb201_fizH1]
	addpd  xmm3, xmm0
	addpd  xmm4, xmm1
	addpd  xmm7, xmm2
	movapd [esp + nb201_fixH1], xmm3
	movapd [esp + nb201_fiyH1], xmm4
	movapd [esp + nb201_fizH1], xmm7
	;# update j forces with water H1 
	addpd  xmm0, [esp + nb201_fjx]
	addpd  xmm1, [esp + nb201_fjy]
	addpd  xmm2, [esp + nb201_fjz]
	movapd [esp + nb201_fjx], xmm0
	movapd [esp + nb201_fjy], xmm1
	movapd [esp + nb201_fjz], xmm2

	;# H2 interactions 
	movapd  xmm4, xmm5	
	mulpd   xmm4, xmm4	;# xmm5=rinv, xmm4=rinvsq 
	movapd  xmm7, xmm5
	movapd  xmm0, [esp + nb201_krsqH2]
	addpd   xmm5, xmm0	;# xmm5=rinv+ krsq 
	mulpd   xmm0, [esp + nb201_two]
	subpd   xmm5, [esp + nb201_crf]
	subpd   xmm7, xmm0	;# xmm7=rinv-2*krsq 
	mulpd   xmm5, [esp + nb201_qqH] ;# vcoul 
	mulpd   xmm7, [esp + nb201_qqH]
	mulpd  xmm4, xmm7		;# total fsH2 in xmm4 
	
	addpd  xmm5, [esp + nb201_vctot]

	movapd xmm0, [esp + nb201_dxH2]
	movapd xmm1, [esp + nb201_dyH2]
	movapd xmm2, [esp + nb201_dzH2]
	movapd [esp + nb201_vctot], xmm5
	mulpd  xmm0, xmm4
	mulpd  xmm1, xmm4
	mulpd  xmm2, xmm4

	;# update H2 forces 
	movapd xmm3, [esp + nb201_fixH2]
	movapd xmm4, [esp + nb201_fiyH2]
	movapd xmm7, [esp + nb201_fizH2]
	addpd  xmm3, xmm0
	addpd  xmm4, xmm1
	addpd  xmm7, xmm2
	movapd [esp + nb201_fixH2], xmm3
	movapd [esp + nb201_fiyH2], xmm4
	movapd [esp + nb201_fizH2], xmm7

	mov edi, [ebp + nb201_faction]
	;# update j forces 
	addpd  xmm0, [esp + nb201_fjx]
	addpd  xmm1, [esp + nb201_fjy]
	addpd  xmm2, [esp + nb201_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 + nb201_innerk],  2
	jl    .nb201_checksingle
	jmp   .nb201_unroll_loop
.nb201_checksingle:	
	mov   edx, [esp + nb201_innerk]
	and   edx, 1
	jnz   .nb201_dosingle
	jmp   .nb201_updateouterdata
.nb201_dosingle:
	mov   edx, [esp + nb201_innerjjnr]     ;# pointer to jjnr[k] 
	mov   eax, [edx]	
	add dword ptr [esp + nb201_innerjjnr],  4	

	mov esi, [ebp + nb201_charge]    ;# base of charge[] 
	xorpd xmm3, xmm3
	movlpd xmm3, [esi + eax*8]
	movapd xmm4, xmm3
	mulpd  xmm3, [esp + nb201_iqO]
	mulpd  xmm4, [esp + nb201_iqH]
	movapd  [esp + nb201_qqO], xmm3
	movapd  [esp + nb201_qqH], xmm4
	
	mov esi, [ebp + nb201_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 + nb201_ixO]
	movapd xmm5, [esp + nb201_iyO]
	movapd xmm6, [esp + nb201_izO]

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

	;# store dr 
	movapd [esp + nb201_dxO], xmm4
	movapd [esp + nb201_dyO], xmm5
	movapd [esp + nb201_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 + nb201_ixH1]
	movapd xmm5, [esp + nb201_iyH1]
	movapd xmm6, [esp + nb201_izH1]

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

	;# store dr 
	movapd [esp + nb201_dxH1], xmm4
	movapd [esp + nb201_dyH1], xmm5
	movapd [esp + nb201_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 + nb201_ixH2]
	movapd xmm4, [esp + nb201_iyH2]
	movapd xmm5, [esp + nb201_izH2]

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

	;# store dr 
	movapd [esp + nb201_dxH2], xmm3
	movapd [esp + nb201_dyH2], xmm4
	movapd [esp + nb201_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 
	
	movapd xmm0, xmm5
	movapd xmm1, xmm6
	movapd xmm2, xmm7

	mulsd  xmm0, [esp + nb201_krf]	
	mulsd  xmm1, [esp + nb201_krf]	
	mulsd  xmm2, [esp + nb201_krf]	

	movapd [esp + nb201_krsqH2], xmm0
	movapd [esp + nb201_krsqH1], xmm1
	movapd [esp + nb201_krsqO], xmm2
	
	;# 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 + nb201_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 + nb201_half] ;# iter1 ( new lu) 

	movapd xmm3, xmm4
	mulsd xmm4, xmm4	;# lu*lu 
	mulsd xmm7, xmm4	;# rsq*lu*lu 
	movapd xmm4, [esp + nb201_three]
	subsd xmm4, xmm7	;# 3-rsq*lu*lu 
	mulsd xmm4, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulsd xmm4, [esp + nb201_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 + nb201_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 + nb201_half] ;# iter1 ( new lu) 

	movapd xmm3, xmm4
	mulsd xmm4, xmm4	;# lu*lu 
	mulsd xmm6, xmm4	;# rsq*lu*lu 
	movapd xmm4, [esp + nb201_three]
	subsd xmm4, xmm6	;# 3-rsq*lu*lu 
	mulsd xmm4, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulsd xmm4, [esp + nb201_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 + nb201_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 + nb201_half] ;# iter1 ( new lu) 

	movapd xmm3, xmm4
	mulsd xmm4, xmm4	;# lu*lu 
	mulsd xmm5, xmm4	;# rsq*lu*lu 
	movapd xmm4, [esp + nb201_three]
	subsd xmm4, xmm5	;# 3-rsq*lu*lu 
	mulsd xmm4, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulsd xmm4, [esp + nb201_half] ;# rinv 
	movapd  xmm5, xmm4	;# rinvH2 in xmm5 

	;# do O interactions 
	movapd  xmm4, xmm7	
	mulsd   xmm4, xmm4	;# xmm6=rinv, xmm4=rinvsq 
	movapd  xmm3, xmm7
	movapd  xmm0, [esp + nb201_krsqO]
	addsd   xmm7, xmm0	;# xmm6=rinv+ krsq 
	mulsd   xmm0, [esp + nb201_two]
	subsd   xmm7, [esp + nb201_crf]
	subsd   xmm3, xmm0	;# xmm7=rinv-2*krsq 
	mulsd   xmm7, [esp + nb201_qqO] ;# vcoul 
	mulsd   xmm3, [esp + nb201_qqO]
	mulsd  xmm4, xmm3	;# total fsH1 in xmm4 
	
	addsd  xmm7, [esp + nb201_vctot]

	movapd xmm0, [esp + nb201_dxO]
	movapd xmm1, [esp + nb201_dyO]
	movapd xmm2, [esp + nb201_dzO]
	movlpd [esp + nb201_vctot], xmm7
	mulsd  xmm0, xmm4
	mulsd  xmm1, xmm4
	mulsd  xmm2, xmm4

	;# update O forces 
	movapd xmm3, [esp + nb201_fixO]
	movapd xmm4, [esp + nb201_fiyO]
	movapd xmm7, [esp + nb201_fizO]
	addsd  xmm3, xmm0
	addsd  xmm4, xmm1
	addsd  xmm7, xmm2
	movlpd [esp + nb201_fixO], xmm3
	movlpd [esp + nb201_fiyO], xmm4
	movlpd [esp + nb201_fizO], xmm7
	;# update j forces with water O 
	movlpd [esp + nb201_fjx], xmm0
	movlpd [esp + nb201_fjy], xmm1
	movlpd [esp + nb201_fjz], xmm2

	;# H1 interactions 
	movapd  xmm4, xmm6	
	mulsd   xmm4, xmm4	;# xmm6=rinv, xmm4=rinvsq 
	movapd  xmm7, xmm6
	movapd  xmm0, [esp + nb201_krsqH1]
	addsd   xmm6, xmm0	;# xmm6=rinv+ krsq 
	mulsd   xmm0, [esp + nb201_two]
	subsd   xmm6, [esp + nb201_crf]
	subsd   xmm7, xmm0	;# xmm7=rinv-2*krsq 
	mulsd   xmm6, [esp + nb201_qqH] ;# vcoul 
	mulsd   xmm7, [esp + nb201_qqH]
	mulsd  xmm4, xmm7		;# total fsH1 in xmm4 
	
	addsd  xmm6, [esp + nb201_vctot]

	movapd xmm0, [esp + nb201_dxH1]
	movapd xmm1, [esp + nb201_dyH1]
	movapd xmm2, [esp + nb201_dzH1]
	movlpd [esp + nb201_vctot], xmm6
	mulsd  xmm0, xmm4
	mulsd  xmm1, xmm4
	mulsd  xmm2, xmm4

	;# update H1 forces 
	movapd xmm3, [esp + nb201_fixH1]
	movapd xmm4, [esp + nb201_fiyH1]
	movapd xmm7, [esp + nb201_fizH1]
	addsd  xmm3, xmm0
	addsd  xmm4, xmm1
	addsd  xmm7, xmm2
	movlpd [esp + nb201_fixH1], xmm3
	movlpd [esp + nb201_fiyH1], xmm4
	movlpd [esp + nb201_fizH1], xmm7
	;# update j forces with water H1 
	addsd  xmm0, [esp + nb201_fjx]
	addsd  xmm1, [esp + nb201_fjy]
	addsd  xmm2, [esp + nb201_fjz]
	movlpd [esp + nb201_fjx], xmm0
	movlpd [esp + nb201_fjy], xmm1
	movlpd [esp + nb201_fjz], xmm2

	;# H2 interactions