nb_kernel131_ia32_sse2.intel_syntax.s

来自「最著名最快的分子模拟软件」· S 代码 · 共 2,209 行 · 第 1/4 页

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

	mov  esi, [ebp + nb131_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 + nb131_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 + nb131_c6]
	mulpd  xmm7, xmm4	 ;# fijD 
	mulpd  xmm5, xmm4	 ;# Vvdw6 

	;# put scalar force on stack Update Vvdwtot directly 
	addpd  xmm5, [esp + nb131_Vvdwtot]
	movapd xmm3, [esp + nb131_fstmp]
	mulpd  xmm7, [esp + nb131_tsc]
	subpd  xmm3, xmm7
	movapd [esp + nb131_fstmp], xmm3
	movapd [esp + nb131_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 + nb131_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 + nb131_c12]
	mulpd  xmm7, xmm4 
	mulpd  xmm5, xmm4  
	
	addpd  xmm5, [esp + nb131_Vvdwtot]
	movapd xmm3, [esp + nb131_fstmp]
	mulpd  xmm7, [esp + nb131_tsc]
	subpd  xmm3, xmm7
	movapd [esp + nb131_Vvdwtot], xmm5

	mulpd  xmm3, xmm0
		
	movapd xmm0, [esp + nb131_dxO]
	movapd xmm1, [esp + nb131_dyO]
	movapd xmm2, [esp + nb131_dzO]

	movd eax, mm0	
	movd ebx, mm1

	mov    edi, [ebp + nb131_faction]
	mulpd  xmm0, xmm3
	mulpd  xmm1, xmm3
	mulpd  xmm2, xmm3

	;# update O forces 
	movapd xmm3, [esp + nb131_fixO]
	movapd xmm4, [esp + nb131_fiyO]
	movapd xmm7, [esp + nb131_fizO]
	addpd  xmm3, xmm0
	addpd  xmm4, xmm1
	addpd  xmm7, xmm2
	movapd [esp + nb131_fixO], xmm3
	movapd [esp + nb131_fiyO], xmm4
	movapd [esp + nb131_fizO], xmm7
	;# update j forces with water O 
	movapd [esp + nb131_fjx], xmm0
	movapd [esp + nb131_fjy], xmm1
	movapd [esp + nb131_fjz], xmm2

	;# H1 interactions 
	movapd  xmm6, [esp + nb131_rinvH1]	
	movapd  xmm4, xmm6
	mulpd   xmm6, [esp + nb131_qqH] ;# vcoul 
	mulpd   xmm4, xmm4 ;# rinvsq
	mulpd   xmm4, xmm6 ;# vcoul*rinvsq
	
	addpd   xmm6, [esp + nb131_vctot]
	movapd [esp + nb131_vctot], xmm6

	movapd xmm0, [esp + nb131_dxH1]
	movapd xmm1, [esp + nb131_dyH1]
	movapd xmm2, [esp + nb131_dzH1]

	mulpd  xmm0, xmm4
	mulpd  xmm1, xmm4
	mulpd  xmm2, xmm4

	;# update H1 forces 
	movapd xmm3, [esp + nb131_fixH1]
	movapd xmm4, [esp + nb131_fiyH1]
	movapd xmm7, [esp + nb131_fizH1]
	addpd  xmm3, xmm0
	addpd  xmm4, xmm1
	addpd  xmm7, xmm2
	movapd [esp + nb131_fixH1], xmm3
	movapd [esp + nb131_fiyH1], xmm4
	movapd [esp + nb131_fizH1], xmm7
	;# update j forces with water H1 
	addpd  xmm0, [esp + nb131_fjx]
	addpd  xmm1, [esp + nb131_fjy]
	addpd  xmm2, [esp + nb131_fjz]
	movapd [esp + nb131_fjx], xmm0
	movapd [esp + nb131_fjy], xmm1
	movapd [esp + nb131_fjz], xmm2

	;# H2 interactions 
	movapd  xmm6, [esp + nb131_rinvH2]	
	movapd  xmm4, xmm6
	mulpd   xmm6, [esp + nb131_qqH] ;# vcoul 
	mulpd   xmm4, xmm4 ;# rinvsq
	mulpd   xmm4, xmm6 ;# vcoul*rinvsq

	addpd  xmm6, [esp + nb131_vctot]

	movapd xmm0, [esp + nb131_dxH2]
	movapd xmm1, [esp + nb131_dyH2]
	movapd xmm2, [esp + nb131_dzH2]
	movapd [esp + nb131_vctot], xmm6
	mulpd  xmm0, xmm4
	mulpd  xmm1, xmm4
	mulpd  xmm2, xmm4

	;# update H2 forces 
	movapd xmm3, [esp + nb131_fixH2]
	movapd xmm4, [esp + nb131_fiyH2]
	movapd xmm7, [esp + nb131_fizH2]
	addpd  xmm3, xmm0
	addpd  xmm4, xmm1
	addpd  xmm7, xmm2
	movapd [esp + nb131_fixH2], xmm3
	movapd [esp + nb131_fiyH2], xmm4
	movapd [esp + nb131_fizH2], xmm7

	mov edi, [ebp + nb131_faction]
	;# update j forces 
	addpd  xmm0, [esp + nb131_fjx]
	addpd  xmm1, [esp + nb131_fjy]
	addpd  xmm2, [esp + nb131_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 + nb131_innerk],  2
	jl    .nb131_checksingle
	jmp   .nb131_unroll_loop
.nb131_checksingle:	
	mov   edx, [esp + nb131_innerk]
	and   edx, 1
	jnz   .nb131_dosingle
	jmp   .nb131_updateouterdata
.nb131_dosingle:
	mov   edx, [esp + nb131_innerjjnr]     ;# pointer to jjnr[k] 
	mov   eax, [edx]	
	add dword ptr [esp + nb131_innerjjnr],  4	

	mov esi, [ebp + nb131_charge]    ;# base of charge[] 
	xorpd xmm3, xmm3
	movlpd xmm3, [esi + eax*8]
	movapd xmm4, xmm3
	mulpd  xmm3, [esp + nb131_iqO]
	mulpd  xmm4, [esp + nb131_iqH]

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

	movapd  [esp + nb131_qqO], xmm3
	movapd  [esp + nb131_qqH], xmm4
	
	mov esi, [ebp + nb131_type]
	mov eax, [esi + eax*4]
	mov esi, [ebp + nb131_vdwparam]
	shl eax, 1	
	mov edi, [esp + nb131_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 + nb131_c6], xmm4
	movapd [esp + nb131_c12], xmm6
	
	mov esi, [ebp + nb131_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 + nb131_ixO]
	movapd xmm5, [esp + nb131_iyO]
	movapd xmm6, [esp + nb131_izO]

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

	;# store dr 
	movapd [esp + nb131_dxO], xmm4
	movapd [esp + nb131_dyO], xmm5
	movapd [esp + nb131_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 + nb131_rsqO], xmm7
	
	;# move ixH1-izH1 to xmm4-xmm6 
	movapd xmm4, [esp + nb131_ixH1]
	movapd xmm5, [esp + nb131_iyH1]
	movapd xmm6, [esp + nb131_izH1]

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

	;# store dr 
	movapd [esp + nb131_dxH1], xmm4
	movapd [esp + nb131_dyH1], xmm5
	movapd [esp + nb131_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 + nb131_ixH2]
	movapd xmm4, [esp + nb131_iyH2]
	movapd xmm5, [esp + nb131_izH2]

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

	;# store dr 
	movapd [esp + nb131_dxH2], xmm3
	movapd [esp + nb131_dyH2], xmm4
	movapd [esp + nb131_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 + nb131_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 + nb131_half] ;# iter1 ( new lu) 

	movapd xmm3, xmm4
	mulsd xmm4, xmm4	;# lu*lu 
	mulsd xmm7, xmm4	;# rsq*lu*lu 
	movapd xmm4, [esp + nb131_three]
	subsd xmm4, xmm7	;# 3-rsq*lu*lu 
	mulsd xmm4, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulsd xmm4, [esp + nb131_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 + nb131_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 + nb131_half] ;# iter1 ( new lu) 

	movapd xmm3, xmm4
	mulsd xmm4, xmm4	;# lu*lu 
	mulsd xmm6, xmm4	;# rsq*lu*lu 
	movapd xmm4, [esp + nb131_three]
	subsd xmm4, xmm6	;# 3-rsq*lu*lu 
	mulsd xmm4, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulsd xmm4, [esp + nb131_half] ;# rinv 
	movapd  xmm6, xmm4	;# rinvH1 in xmm6 
	movsd  [esp + nb131_rinvH1], xmm6
	
	;# rsqH2 - seed in xmm2 
	cvtsd2ss xmm2, xmm5	
	rsqrtss xmm2, xmm2
	cvtss2sd xmm2, xmm2

	movapd  xmm3, xmm2
	mulsd   xmm2, xmm2
	movapd  xmm4, [esp + nb131_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 + nb131_half] ;# iter1 ( new lu) 

	movapd xmm3, xmm4
	mulsd xmm4, xmm4	;# lu*lu 
	mulsd xmm5, xmm4	;# rsq*lu*lu 
	movapd xmm4, [esp + nb131_three]
	subsd xmm4, xmm5	;# 3-rsq*lu*lu 
	mulsd xmm4, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulsd xmm4, [esp + nb131_half] ;# rinv 
	movapd  xmm5, xmm4	;# rinvH2 in xmm5 
	movsd [esp + nb131_rinvH2], xmm5
	
	;# do O interactions 
	movsd xmm0, xmm7
	mulsd  xmm7, [esp + nb131_qqO] ;# vcoul
	movsd xmm6, xmm0
	mulsd  xmm6, xmm7 ;# vcoul*rinv

	movsd [esp + nb131_fstmp], xmm6 ;# save to temp. storage

	addsd  xmm7, [esp + nb131_vctot]
	movsd [esp + nb131_vctot], xmm7

	movsd xmm4, [esp + nb131_rsqO]

	;# LJ table interaction. xmm0=rinv, xmm4=rsq	
	mulsd xmm4, xmm0	;# xmm4=r 
	mulsd xmm4, [esp + nb131_tsc]
	
	cvttsd2si ebx, xmm4	;# mm6 = lu idx 
	cvtsi2sd xmm5, ebx
	subpd xmm4, xmm5
	movapd xmm1, xmm4	;# xmm1=eps 
	movapd xmm2, xmm1	
	mulpd  xmm2, xmm2	;# xmm2=eps2 

	shl ebx, 3
	
	mov  esi, [ebp + nb131_VFtab]

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

	movlpd xmm6, [esi + ebx*8 + 16]	;# G1
	movhpd xmm6, [esi + ebx*8 + 24]	;# G1 H1 	
	movapd xmm7, xmm6
	unpcklpd xmm6, xmm3	;# G1 G2 
	unpckhpd xmm7, xmm3	;# H1 H2 
	;# dispersion table ready, in xmm4-xmm7 	
	mulsd  xmm6, xmm1	;# xmm6=Geps 
	mulsd  xmm7, xmm2	;# xmm7=Heps2 
	addsd  xmm5, xmm6
	addsd  xmm5, xmm7	;# xmm5=Fp 	
	mulsd  xmm7, [esp + nb131_two]	;# two*Heps2 
	addsd  xmm7, xmm6
	addsd  xmm7, xmm5 ;# xmm7=FF 
	mulsd  xmm5, xmm1 ;# xmm5=eps*Fp 
	addsd  xmm5, xmm4 ;# xmm5=VV 

	movsd xmm4, [esp + nb131_c6]
	mulsd  xmm7, xmm4	 ;# fijD 
	mulsd  xmm5, xmm4	 ;# Vvdw6 

	;# put scalar force on stack Update Vvdwtot directly 
	addsd  xmm5, [esp + nb131_Vvdwtot]
	movsd xmm3, [esp + nb131_fstmp]
	mulsd  xmm7, [esp + nb131_tsc]
	subsd  xmm3, xmm7
	movsd [esp + nb131_fstmp], xmm3
	movsd [esp + nb131_Vvdwtot], xmm5

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

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

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

	movapd xmm7, xmm6
	unpcklpd xmm6, xmm3	;# G1 G2 
	unpckhpd xmm7, xmm3	;# H1 H2 
	
	;# table ready, in xmm4-xmm7 	
	mulsd  xmm6, xmm1	;# xmm6=Geps 
	mulsd  xmm7, xmm2	;# xmm7=Heps2 
	addsd  xmm5, xmm6
	addsd  xmm5, xmm7	;# xmm5=Fp 	
	mulsd  xmm7, [esp + nb131_two]	;# two*Heps2 
	addsd  xmm7, xmm6
	addsd  xmm7, xmm5 ;# xmm7=FF 
	mulsd  xmm5, xmm1 ;# xmm5=eps*Fp 
	addsd  xmm5, xmm4 ;# xmm5=VV 
	
	movsd xmm4, [esp + nb131_c12]
	mulsd  xmm7, xmm4 
	mulsd  xmm5, xmm4  
	
	addsd  xmm5, [esp + nb131_Vvdwtot]
	movsd xmm3, [esp + nb131_fstmp]
	mulsd  xmm7, [esp + nb131_tsc]
	subsd  xmm3, xmm7
	movsd [esp + nb131_Vvdwtot], xmm5

	mulsd  xmm3, xmm0
		
	movsd xmm0, [esp + nb131_dxO]
	movsd xmm1, [esp + nb131_dyO]
	movsd xmm2, [esp + nb131_dzO]

	mov    edi, [ebp + nb131_faction]
	mulsd  xmm0, xmm3
	mulsd  xmm1, xmm3
	mulsd  xmm2, xmm3

	;# update O forces 
	movapd xmm3, [esp + nb131_fixO]
	movapd xmm4, [esp + nb131_fiyO]
	movapd xmm7, [esp + nb131_fizO]
	addsd  xmm3, xmm0
	addsd  xmm4, xmm1
	addsd  xmm7, xmm2
	movlpd [esp + nb131_fixO], xmm3
	movlpd [esp + nb131_fiyO], xmm4
	movlpd [esp + nb131_fizO], xmm7
	;# update j forces with water O 
	movlpd [esp + nb131_fjx], xmm0
	movlpd [esp + nb131_fjy], xmm1
	movlpd [esp + nb131_fjz], xmm2

	;# H1 interactions 
	movsd  xmm6, [esp + nb131_rinvH1]	
	movsd  xmm4, xmm6
	mulsd   xmm6, [esp + nb131_qqH] ;# vcoul 
	mulsd   xmm4, xmm4 ;# rinvsq
	mulsd   xmm4, xmm6 ;# vcoul*rinvsq

	addsd  xmm6, [esp + nb131_vctot]

	movapd xmm0, [esp + nb131_dxH1]
	movapd xmm1, [esp + nb131_dyH1]
	movapd xmm2, [esp + nb131_dzH1]
	movlpd [esp + nb131_vctot], xmm6
	mulsd  xmm0, xmm4
	mulsd  xmm1, xmm4
	mulsd  xmm2, xmm4

	;# update H1 forces 
	movapd xmm3, [esp + nb131_fixH1]
	movapd xmm4, [esp + nb131_fiyH1]
	movapd xmm7, [esp + nb131_fizH1]
	addsd  xmm3, xmm0
	addsd  xmm4, xmm1
	addsd  xmm7, xmm2
	movlpd [esp + nb131_fixH1], xmm3
	movlpd [esp + nb131_fiyH1], xmm4
	movlpd [esp + nb131_fizH1], xmm7
	;# update j forces with water H1 
	addsd  xmm0, [esp + nb131_fjx]
	addsd  xmm1, [esp + nb131_fjy]
	addsd  xmm2, [esp + nb131_fjz]
	movlpd [esp + nb131_fjx], xmm0
	movlpd [esp + nb131_fjy], xmm1
	movlpd [esp + nb131_fjz], xmm2