nb_kernel231_ia32_sse2.intel_syntax.s

来自「最著名最快的分子模拟软件」· S 代码 · 共 2,366 行 · 第 1/5 页

	movapd xmm0, xmm5
	movapd xmm1, xmm6
	movapd xmm2, xmm7

	mulpd  xmm0, [esp + nb231_krf]	
	mulpd  xmm1, [esp + nb231_krf]	
	mulpd  xmm2, [esp + nb231_krf]	

	movapd [esp + nb231_krsqH2], xmm0
	movapd [esp + nb231_krsqH1], xmm1
	movapd [esp + nb231_krsqO], xmm2
		
	;# start with rsqO - put seed in xmm2 
	cvtpd2ps xmm2, xmm7	
	rsqrtps xmm2, xmm2
	cvtps2pd xmm2, xmm2

	movapd  xmm3, xmm2
	mulpd   xmm2, xmm2
	movapd  xmm4, [esp + nb231_three]
	mulpd   xmm2, xmm7	;# rsq*lu*lu 
	subpd   xmm4, xmm2	;# 30-rsq*lu*lu 
	mulpd   xmm4, xmm3	;# lu*(3-rsq*lu*lu) 
	mulpd   xmm4, [esp + nb231_half] ;# iter1 ( new lu) 

	movapd xmm3, xmm4
	mulpd xmm4, xmm4	;# lu*lu 
	mulpd xmm7, xmm4	;# rsq*lu*lu 
	movapd xmm4, [esp + nb231_three]
	subpd xmm4, xmm7	;# 3-rsq*lu*lu 
	mulpd xmm4, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulpd xmm4, [esp + nb231_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 + nb231_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 + nb231_half] ;# iter1 ( new lu) 

	movapd xmm3, xmm4
	mulpd xmm4, xmm4	;# lu*lu 
	mulpd xmm6, xmm4	;# rsq*lu*lu 
	movapd xmm4, [esp + nb231_three]
	subpd xmm4, xmm6	;# 3-rsq*lu*lu 
	mulpd xmm4, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulpd xmm4, [esp + nb231_half] ;# rinv 
	movapd  xmm6, xmm4	;# rinvH1 in xmm6 
	movapd  [esp + nb231_rinvH1], xmm6
	
	;# rsqH2 - seed in xmm2 
	cvtpd2ps xmm2, xmm5	
	rsqrtps xmm2, xmm2
	cvtps2pd xmm2, xmm2

	movapd  xmm3, xmm2
	mulpd   xmm2, xmm2
	movapd  xmm4, [esp + nb231_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 + nb231_half] ;# iter1 ( new lu) 

	movapd xmm3, xmm4
	mulpd xmm4, xmm4	;# lu*lu 
	mulpd xmm5, xmm4	;# rsq*lu*lu 
	movapd xmm4, [esp + nb231_three]
	subpd xmm4, xmm5	;# 3-rsq*lu*lu 
	mulpd xmm4, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulpd xmm4, [esp + nb231_half] ;# rinv 
	movapd  xmm5, xmm4	;# rinvH2 in xmm5 
	movapd  [esp + nb231_rinvH2], xmm5

	;# do O interactions 
	movapd xmm0, xmm7
	movapd xmm6, xmm7
	movapd xmm1, [esp + nb231_krsqO]
	addpd  xmm0, xmm1
	mulpd  xmm1, [esp + nb231_two]
	subpd  xmm0, [esp + nb231_crf] ;# xmm0=rinv+ krsq-crf 
	subpd  xmm6, xmm1
	mulpd  xmm0, [esp + nb231_qqO]
	mulpd  xmm6, [esp + nb231_qqO]

	mulpd  xmm6, xmm7
	movapd [esp + nb231_fstmp], xmm6 ;# save to temp. storage

	addpd  xmm0, [esp + nb231_vctot]
	movapd [esp + nb231_vctot], xmm0

	movapd xmm0, xmm7
	movapd xmm4, [esp + nb231_rsqO]
	;# LJ table interaction. xmm0=rinv, xmm4=rsq
	
	mulpd xmm4, xmm0	;# xmm4=r 
	mulpd xmm4, [esp + nb231_tsc]
	
	cvttpd2pi mm6, xmm4	;# mm6 = lu idx 
	cvtpi2pd xmm5, mm6
	subpd xmm4, xmm5
	movapd xmm1, xmm4	;# xmm1=eps 
	movapd xmm2, xmm1	
	mulpd  xmm2, xmm2	;# xmm2=eps2 

	pslld mm6, 3		;# idx *= 8 
	
	movd mm0, eax	
	movd mm1, ebx

	mov  esi, [ebp + nb231_VFtab]
	movd eax, mm6
	psrlq mm6, 32
	movd ebx, mm6

	;# dispersion 
	movlpd xmm4, [esi + eax*8]	;# Y1 	
	movlpd xmm3, [esi + ebx*8]	;# Y2 
	movhpd xmm4, [esi + eax*8 + 8]	;# Y1 F1 	
	movhpd xmm3, [esi + ebx*8 + 8]	;# Y2 F2 
	movapd xmm5, xmm4
	unpcklpd xmm4, xmm3	;# Y1 Y2 
	unpckhpd xmm5, xmm3	;# F1 F2 

	movlpd xmm6, [esi + eax*8 + 16]	;# G1
	movlpd xmm3, [esi + ebx*8 + 16]	;# G2
	movhpd xmm6, [esi + eax*8 + 24]	;# G1 H1 	
	movhpd xmm3, [esi + ebx*8 + 24]	;# G2 H2 
	movapd xmm7, xmm6
	unpcklpd xmm6, xmm3	;# G1 G2 
	unpckhpd xmm7, xmm3	;# H1 H2 
	;# dispersion table ready, in xmm4-xmm7 	
	mulpd  xmm6, xmm1	;# xmm6=Geps 
	mulpd  xmm7, xmm2	;# xmm7=Heps2 
	addpd  xmm5, xmm6
	addpd  xmm5, xmm7	;# xmm5=Fp 	
	mulpd  xmm7, [esp + nb231_two]	;# two*Heps2 
	addpd  xmm7, xmm6
	addpd  xmm7, xmm5 ;# xmm7=FF 
	mulpd  xmm5, xmm1 ;# xmm5=eps*Fp 
	addpd  xmm5, xmm4 ;# xmm5=VV 

	movapd xmm4, [esp + nb231_c6]
	mulpd  xmm7, xmm4	 ;# fijD 
	mulpd  xmm5, xmm4	 ;# Vvdw6 

	;# put scalar force on stack Update Vvdwtot directly 
	addpd  xmm5, [esp + nb231_Vvdwtot]
	movapd xmm3, [esp + nb231_fstmp]
	mulpd  xmm7, [esp + nb231_tsc]
	subpd  xmm3, xmm7
	movapd [esp + nb231_fstmp], xmm3
	movapd [esp + nb231_Vvdwtot], xmm5

	;# repulsion 
	movlpd xmm4, [esi + eax*8 + 32]	;# Y1 	
	movlpd xmm3, [esi + ebx*8 + 32]	;# Y2 
	movhpd xmm4, [esi + eax*8 + 40]	;# Y1 F1 	
	movhpd xmm3, [esi + ebx*8 + 40]	;# Y2 F2 

	movapd xmm5, xmm4
	unpcklpd xmm4, xmm3	;# Y1 Y2 
	unpckhpd xmm5, xmm3	;# F1 F2 

	movlpd xmm6, [esi + eax*8 + 48]	;# G1
	movlpd xmm3, [esi + ebx*8 + 48]	;# G2
	movhpd xmm6, [esi + eax*8 + 56]	;# G1 H1 	
	movhpd xmm3, [esi + ebx*8 + 56]	;# G2 H2 

	movapd xmm7, xmm6
	unpcklpd xmm6, xmm3	;# G1 G2 
	unpckhpd xmm7, xmm3	;# H1 H2 
	
	;# table ready, in xmm4-xmm7 	
	mulpd  xmm6, xmm1	;# xmm6=Geps 
	mulpd  xmm7, xmm2	;# xmm7=Heps2 
	addpd  xmm5, xmm6
	addpd  xmm5, xmm7	;# xmm5=Fp 	
	mulpd  xmm7, [esp + nb231_two]	;# two*Heps2 
	addpd  xmm7, xmm6
	addpd  xmm7, xmm5 ;# xmm7=FF 
	mulpd  xmm5, xmm1 ;# xmm5=eps*Fp 
	addpd  xmm5, xmm4 ;# xmm5=VV 
	
	movapd xmm4, [esp + nb231_c12]
	mulpd  xmm7, xmm4 
	mulpd  xmm5, xmm4  
	
	addpd  xmm5, [esp + nb231_Vvdwtot]
	movapd xmm3, [esp + nb231_fstmp]
	mulpd  xmm7, [esp + nb231_tsc]
	subpd  xmm3, xmm7
	movapd [esp + nb231_Vvdwtot], xmm5

	mulpd  xmm3, xmm0
		
	movapd xmm0, [esp + nb231_dxO]
	movapd xmm1, [esp + nb231_dyO]
	movapd xmm2, [esp + nb231_dzO]

	movd eax, mm0	
	movd ebx, mm1

	mov    edi, [ebp + nb231_faction]
	mulpd  xmm0, xmm3
	mulpd  xmm1, xmm3
	mulpd  xmm2, xmm3

	;# update O forces 
	movapd xmm3, [esp + nb231_fixO]
	movapd xmm4, [esp + nb231_fiyO]
	movapd xmm7, [esp + nb231_fizO]
	addpd  xmm3, xmm0
	addpd  xmm4, xmm1
	addpd  xmm7, xmm2
	movapd [esp + nb231_fixO], xmm3
	movapd [esp + nb231_fiyO], xmm4
	movapd [esp + nb231_fizO], xmm7
	;# update j forces with water O 
	movapd [esp + nb231_fjx], xmm0
	movapd [esp + nb231_fjy], xmm1
	movapd [esp + nb231_fjz], xmm2

	;# H1 interactions 
	movapd  xmm6, [esp + nb231_rinvH1]	
	movapd  xmm4, xmm6
	mulpd   xmm4, xmm4	;# xmm6=rinv, xmm4=rinvsq 
	movapd  xmm7, xmm6
	movapd  xmm0, [esp + nb231_krsqH1]
	addpd   xmm6, xmm0	;# xmm6=rinv+ krsq 
	mulpd   xmm0, [esp + nb231_two]
	subpd   xmm6, [esp + nb231_crf]
	subpd   xmm7, xmm0	;# xmm7=rinv-2*krsq 
	mulpd   xmm6, [esp + nb231_qqH] ;# vcoul 
	mulpd   xmm7, [esp + nb231_qqH]
	mulpd  xmm4, xmm7		;# total fsH1 in xmm4 
	
	addpd  xmm6, [esp + nb231_vctot]

	movapd xmm0, [esp + nb231_dxH1]
	movapd xmm1, [esp + nb231_dyH1]
	movapd xmm2, [esp + nb231_dzH1]
	movapd [esp + nb231_vctot], xmm6
	mulpd  xmm0, xmm4
	mulpd  xmm1, xmm4
	mulpd  xmm2, xmm4

	;# update H1 forces 
	movapd xmm3, [esp + nb231_fixH1]
	movapd xmm4, [esp + nb231_fiyH1]
	movapd xmm7, [esp + nb231_fizH1]
	addpd  xmm3, xmm0
	addpd  xmm4, xmm1
	addpd  xmm7, xmm2
	movapd [esp + nb231_fixH1], xmm3
	movapd [esp + nb231_fiyH1], xmm4
	movapd [esp + nb231_fizH1], xmm7
	;# update j forces with water H1 
	addpd  xmm0, [esp + nb231_fjx]
	addpd  xmm1, [esp + nb231_fjy]
	addpd  xmm2, [esp + nb231_fjz]
	movapd [esp + nb231_fjx], xmm0
	movapd [esp + nb231_fjy], xmm1
	movapd [esp + nb231_fjz], xmm2

	;# H2 interactions 
	movapd  xmm5, [esp + nb231_rinvH2]	
	movapd  xmm4, xmm5
	mulpd   xmm4, xmm4	;# xmm5=rinv, xmm4=rinvsq 
	movapd  xmm7, xmm5
	movapd  xmm0, [esp + nb231_krsqH2]
	addpd   xmm5, xmm0	;# xmm5=rinv+ krsq 
	mulpd   xmm0, [esp + nb231_two]
	subpd   xmm5, [esp + nb231_crf]
	subpd   xmm7, xmm0	;# xmm7=rinv-2*krsq 
	mulpd   xmm5, [esp + nb231_qqH] ;# vcoul 
	mulpd   xmm7, [esp + nb231_qqH]
	mulpd  xmm4, xmm7		;# total fsH2 in xmm4 
	
	addpd  xmm5, [esp + nb231_vctot]

	movapd xmm0, [esp + nb231_dxH2]
	movapd xmm1, [esp + nb231_dyH2]
	movapd xmm2, [esp + nb231_dzH2]
	movapd [esp + nb231_vctot], xmm5
	mulpd  xmm0, xmm4
	mulpd  xmm1, xmm4
	mulpd  xmm2, xmm4

	;# update H2 forces 
	movapd xmm3, [esp + nb231_fixH2]
	movapd xmm4, [esp + nb231_fiyH2]
	movapd xmm7, [esp + nb231_fizH2]
	addpd  xmm3, xmm0
	addpd  xmm4, xmm1
	addpd  xmm7, xmm2
	movapd [esp + nb231_fixH2], xmm3
	movapd [esp + nb231_fiyH2], xmm4
	movapd [esp + nb231_fizH2], xmm7

	mov edi, [ebp + nb231_faction]
	;# update j forces 
	addpd  xmm0, [esp + nb231_fjx]
	addpd  xmm1, [esp + nb231_fjy]
	addpd  xmm2, [esp + nb231_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 + nb231_innerk],  2
	jl    .nb231_checksingle
	jmp   .nb231_unroll_loop
.nb231_checksingle:	
	mov   edx, [esp + nb231_innerk]
	and   edx, 1
	jnz   .nb231_dosingle
	jmp   .nb231_updateouterdata
.nb231_dosingle:
	mov   edx, [esp + nb231_innerjjnr]     ;# pointer to jjnr[k] 
	mov   eax, [edx]	
	add dword ptr [esp + nb231_innerjjnr],  4	

	mov esi, [ebp + nb231_charge]    ;# base of charge[] 
	xorpd xmm3, xmm3
	movlpd xmm3, [esi + eax*8]
	movapd xmm4, xmm3
	mulpd  xmm3, [esp + nb231_iqO]
	mulpd  xmm4, [esp + nb231_iqH]

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

	movapd  [esp + nb231_qqO], xmm3
	movapd  [esp + nb231_qqH], xmm4
	
	mov esi, [ebp + nb231_type]
	mov eax, [esi + eax*4]
	mov esi, [ebp + nb231_vdwparam]
	shl eax, 1	
	mov edi, [esp + nb231_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
	movapd [esp + nb231_c6], xmm4
	movapd [esp + nb231_c12], xmm6
	
	mov esi, [ebp + nb231_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 + nb231_ixO]
	movapd xmm5, [esp + nb231_iyO]
	movapd xmm6, [esp + nb231_izO]

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

	;# store dr 
	movapd [esp + nb231_dxO], xmm4
	movapd [esp + nb231_dyO], xmm5
	movapd [esp + nb231_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 
	movapd [esp + nb231_rsqO], xmm7
	
	;# move ixH1-izH1 to xmm4-xmm6 
	movapd xmm4, [esp + nb231_ixH1]
	movapd xmm5, [esp + nb231_iyH1]
	movapd xmm6, [esp + nb231_izH1]

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

	;# store dr 
	movapd [esp + nb231_dxH1], xmm4
	movapd [esp + nb231_dyH1], xmm5
	movapd [esp + nb231_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 + nb231_ixH2]
	movapd xmm4, [esp + nb231_iyH2]
	movapd xmm5, [esp + nb231_izH2]

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

	;# store dr 
	movapd [esp + nb231_dxH2], xmm3
	movapd [esp + nb231_dyH2], xmm4
	movapd [esp + nb231_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 + nb231_krf]	
	mulsd  xmm1, [esp + nb231_krf]	
	mulsd  xmm2, [esp + nb231_krf]	

	movapd [esp + nb231_krsqH2], xmm0
	movapd [esp + nb231_krsqH1], xmm1
	movapd [esp + nb231_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 + nb231_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 + nb231_half] ;# iter1 ( new lu) 

	movapd xmm3, xmm4