nb_kernel331_ia32_sse2.intel_syntax.s

来自「最著名最快的分子模拟软件」· S 代码 · 共 2,337 行 · 第 1/5 页

	mulpd   xmm2, xmm2
	movapd  xmm4, [esp + nb331_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 + nb331_half] ;# iter1 ( new lu) 

	movapd xmm2, xmm7
	movapd xmm3, xmm4
	mulpd xmm4, xmm4	;# lu*lu 
	mulpd xmm2, xmm4	;# rsq*lu*lu 
	movapd xmm4, [esp + nb331_three]
	subpd xmm4, xmm2	;# 3-rsq*lu*lu 
	mulpd xmm4, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulpd xmm4, [esp + nb331_half] ;# rinv 
	movapd  [esp + nb331_rinvO], xmm4	;# rinvO in xmm4 
	mulpd   xmm7, xmm4
	movapd  [esp + nb331_rO], xmm7	;# r in xmm7 
	
	;# rsqH1 - seed in xmm2 
	cvtpd2ps xmm2, xmm6	
	rsqrtps xmm2, xmm2
	cvtps2pd xmm2, xmm2

	movapd  xmm3, xmm2
	mulpd   xmm2, xmm2
	movapd  xmm4, [esp + nb331_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 + nb331_half] ;# iter1 ( new lu) 

	movapd xmm2, xmm6
	movapd xmm3, xmm4
	mulpd xmm4, xmm4	;# lu*lu 
	mulpd xmm2, xmm4	;# rsq*lu*lu 
	movapd xmm4, [esp + nb331_three]
	subpd xmm4, xmm2	;# 3-rsq*lu*lu 
	mulpd xmm4, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulpd xmm4, [esp + nb331_half] ;# rinv 
	movapd [esp + nb331_rinvH1], xmm4	;# rinvH1 
	mulpd  xmm6, xmm4
	movapd [esp + nb331_rH1], xmm6	;# rH1 
	
	;# rsqH2 - seed in xmm2 
	cvtpd2ps xmm2, xmm5	
	rsqrtps xmm2, xmm2
	cvtps2pd xmm2, xmm2

	movapd  xmm3, xmm2
	mulpd   xmm2, xmm2
	movapd  xmm4, [esp + nb331_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 + nb331_half] ;# iter1 ( new lu) 

	movapd xmm2, xmm5
	movapd xmm3, xmm4
	mulpd xmm4, xmm4	;# lu*lu 
	mulpd xmm2, xmm4	;# rsq*lu*lu 
	movapd xmm4, [esp + nb331_three]
	subpd xmm4, xmm2	;# 3-rsq*lu*lu 
	mulpd xmm4, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulpd xmm4, [esp + nb331_half] ;# rinv 
	movapd [esp + nb331_rinvH2], xmm4 ;# rinv 
	mulpd xmm5, xmm4
	movapd [esp + nb331_rH2], xmm5 ;# r 

	;# do O interactions 
	;# rO is still in xmm7 
	movapd xmm0, [esp + nb331_rinvO]
	mulpd   xmm7, [esp + nb331_tsc]
	cvttpd2pi mm6, xmm7	;# mm6 = lu idx 
	cvtpi2pd xmm6, mm6
	subpd xmm7, xmm6
	movapd xmm1, xmm7	;# xmm1=eps 
	movapd xmm2, xmm1	
	mulpd  xmm2, xmm2	;# xmm2=eps2 
	
	pslld mm6, 2		;# idx *= 4 
	movd mm0, eax	
	movd mm1, ebx
	mov  esi, [ebp + nb331_VFtab]
	movd eax, mm6
	psrlq mm6, 32
	movd ebx, mm6		;# indices in eax/ebx 
	lea   eax, [eax + eax*2] ;# idx *= 3 (total *=12 now) 
	lea   ebx, [ebx + ebx*2]
	
	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 
	;# coulomb 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 + nb331_two]       ;# two*Heps2 
    movapd xmm0, [esp + nb331_qqO]
    addpd  xmm7, xmm6
    addpd  xmm7, xmm5 ;# xmm7=FF 
    mulpd  xmm5, xmm1 ;# xmm5=eps*Fp 
    addpd  xmm5, xmm4 ;# xmm5=VV 
    mulpd  xmm5, xmm0 ;# vcoul=qq*VV  
    mulpd  xmm0, xmm7 ;# fijC=FF*qq 
	;# at this point xmm5 contains vcoul and xmm0 fijC 
	;# increment vcoul - then we can get rid of mm5 
	;# update vctot 
	addpd  xmm5, [esp + nb331_vctot]
	movapd [esp + nb331_vctot], xmm5 
	
	;# Dispersion 
	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 
	;# 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 + nb331_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 + nb331_c6]
	mulpd  xmm7, xmm4	 ;# fijD 
	mulpd  xmm5, xmm4	 ;# Vvdw6 
	addpd  xmm0, xmm7	;# fscal summation 
		
	addpd  xmm5, [esp + nb331_Vvdwtot]
	movapd [esp + nb331_Vvdwtot], xmm5
	
	;# Repulsion 
	movlpd xmm4, [esi + eax*8 + 64]	;# Y1
	movlpd xmm3, [esi + ebx*8 + 64]	;# Y2
	movhpd xmm4, [esi + eax*8 + 72]	;# Y1 F1 	
	movhpd xmm3, [esi + ebx*8 + 72]	;# Y2 F2 

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

	movlpd xmm6, [esi + eax*8 + 80]	;# G1
	movlpd xmm3, [esi + ebx*8 + 80]	;# G2
	movhpd xmm6, [esi + eax*8 + 88]	;# G1 H1 	
	movhpd xmm3, [esi + ebx*8 + 88]	;# 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 + nb331_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 + nb331_c12]
	mulpd  xmm7, xmm4 ;# fijR 
	mulpd  xmm5, xmm4 ;# Vvdw12 
	addpd  xmm7, xmm0	;# fscal summation 
	mulpd xmm7, [esp + nb331_tsc]
	mulpd xmm7, [esp + nb331_rinvO]

	addpd  xmm5, [esp + nb331_Vvdwtot]
	movapd [esp + nb331_Vvdwtot], xmm5

	xorpd xmm4, xmm4
	subpd xmm4, xmm7
	movapd xmm0, [esp + nb331_dxO]
	movapd xmm1, [esp + nb331_dyO]
	movapd xmm2, [esp + nb331_dzO]

	movd eax, mm0	
	movd ebx, mm1
	
	movapd xmm0, [esp + nb331_dxO]
	movapd xmm1, [esp + nb331_dyO]
	movapd xmm2, [esp + nb331_dzO]
	mulpd  xmm0, xmm4
	mulpd  xmm1, xmm4
	mulpd  xmm2, xmm4	;# tx in xmm0-xmm2 

	;# update O forces 
	movapd xmm3, [esp + nb331_fixO]
	movapd xmm4, [esp + nb331_fiyO]
	movapd xmm7, [esp + nb331_fizO]
	addpd  xmm3, xmm0
	addpd  xmm4, xmm1
	addpd  xmm7, xmm2
	movapd [esp + nb331_fixO], xmm3
	movapd [esp + nb331_fiyO], xmm4
	movapd [esp + nb331_fizO], xmm7
	;# update j forces with water O 
	movapd [esp + nb331_fjx], xmm0
	movapd [esp + nb331_fjy], xmm1
	movapd [esp + nb331_fjz], xmm2

	;# Done with O interactions - now H1! 
	movapd xmm7, [esp + nb331_rH1]
	mulpd xmm7, [esp + nb331_tsc]
	cvttpd2pi mm6, xmm7	;# mm6 = lu idx 
	cvtpi2pd xmm6, mm6
	subpd xmm7, xmm6
	movapd xmm1, xmm7	;# xmm1=eps 
	movapd xmm2, xmm1	
	mulpd  xmm2, xmm2	;# xmm2=eps2 
	
	pslld mm6, 2		;# idx *= 4 
	mov  esi, [ebp + nb331_VFtab]
	movd eax, mm6
	psrlq mm6, 32
	movd ebx, mm6		;# indices in eax/ebx 
	lea   eax, [eax + eax*2] ;# idx *= 3 (total *=12 now) 	
	lea   ebx, [ebx + ebx*2]
	
	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 
	;# coulomb 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 + nb331_two]	;# two*Heps2 
	movapd xmm3, [esp + nb331_qqH]
	addpd  xmm7, xmm6
	addpd  xmm7, xmm5 ;# xmm7=FF 
	mulpd  xmm5, xmm1 ;# xmm5=eps*Fp 
	addpd  xmm5, xmm4 ;# xmm5=VV 
	mulpd  xmm5, xmm3 ;# vcoul=qq*VV  
	mulpd  xmm3, xmm7 ;# fijC=FF*qq 
    ;# at this point mm5 contains vcoul and xmm3 fijC 
    ;# increment vcoul 
	xorpd  xmm4, xmm4
    addpd  xmm5, [esp + nb331_vctot]
	mulpd  xmm3, [esp + nb331_rinvH1]
    movapd [esp + nb331_vctot], xmm5 
	mulpd  xmm3, [esp + nb331_tsc]
	subpd xmm4, xmm3

	movapd xmm0, [esp + nb331_dxH1]
	movapd xmm1, [esp + nb331_dyH1]
	movapd xmm2, [esp + nb331_dzH1]
	mulpd  xmm0, xmm4
	mulpd  xmm1, xmm4
	mulpd  xmm2, xmm4

	;# update H1 forces 
	movapd xmm3, [esp + nb331_fixH1]
	movapd xmm4, [esp + nb331_fiyH1]
	movapd xmm7, [esp + nb331_fizH1]
	addpd  xmm3, xmm0
	addpd  xmm4, xmm1
	addpd  xmm7, xmm2
	movapd [esp + nb331_fixH1], xmm3
	movapd [esp + nb331_fiyH1], xmm4
	movapd [esp + nb331_fizH1], xmm7
	;# update j forces with water H1 
	addpd  xmm0, [esp + nb331_fjx]
	addpd  xmm1, [esp + nb331_fjy]
	addpd  xmm2, [esp + nb331_fjz]
	movapd [esp + nb331_fjx], xmm0
	movapd [esp + nb331_fjy], xmm1
	movapd [esp + nb331_fjz], xmm2

	;# Done with H1, finally we do H2 interactions 
	movapd xmm7, [esp + nb331_rH2]
	mulpd   xmm7, [esp + nb331_tsc]
	cvttpd2pi mm6, xmm7	;# mm6 = lu idx 
	cvtpi2pd xmm6, mm6
	subpd xmm7, xmm6
	movapd xmm1, xmm7	;# xmm1=eps 
	movapd xmm2, xmm1	
	mulpd  xmm2, xmm2	;# xmm2=eps2 
	
	pslld mm6, 2		;# idx *= 4 
	mov  esi, [ebp + nb331_VFtab]
	movd eax, mm6
	psrlq mm6, 32
	movd ebx, mm6		;# indices in eax/ebx 
	lea   eax, [eax + eax*2] ;# idx *= 3 (total *=12 now)
	lea   ebx, [ebx + ebx*2]

	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 
	;# coulomb 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 + nb331_two]	;# two*Heps2 
	movapd xmm3, [esp + nb331_qqH]
	addpd  xmm7, xmm6
	addpd  xmm7, xmm5 ;# xmm7=FF 
	mulpd  xmm5, xmm1 ;# xmm5=eps*Fp 
	addpd  xmm5, xmm4 ;# xmm5=VV 
	mulpd  xmm5, xmm3 ;# vcoul=qq*VV  
	mulpd  xmm3, xmm7 ;# fijC=FF*qq 
    ;# at this point mm5 contains vcoul and xmm3 fijC 
    ;# increment vcoul 
	xorpd  xmm4, xmm4
    addpd  xmm5, [esp + nb331_vctot]
	mulpd  xmm3, [esp + nb331_rinvH2]
    movapd [esp + nb331_vctot], xmm5 
	mulpd  xmm3, [esp + nb331_tsc]
	subpd  xmm4, xmm3

	movapd xmm0, [esp + nb331_dxH2]
	movapd xmm1, [esp + nb331_dyH2]
	movapd xmm2, [esp + nb331_dzH2]
	mulpd  xmm0, xmm4
	mulpd  xmm1, xmm4
	mulpd  xmm2, xmm4

    movd eax, mm0   
    movd ebx, mm1
	
	;# update H2 forces 
	movapd xmm3, [esp + nb331_fixH2]
	movapd xmm4, [esp + nb331_fiyH2]
	movapd xmm7, [esp + nb331_fizH2]
	addpd  xmm3, xmm0
	addpd  xmm4, xmm1
	addpd  xmm7, xmm2
	movapd [esp + nb331_fixH2], xmm3
	movapd [esp + nb331_fiyH2], xmm4
	movapd [esp + nb331_fizH2], xmm7

	mov edi, [ebp + nb331_faction]
	;# update j forces 
	addpd  xmm0, [esp + nb331_fjx]
	addpd  xmm1, [esp + nb331_fjy]
	addpd  xmm2, [esp + nb331_fjz]

	;# the fj's - start by accumulating forces from memory 
	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 + nb331_innerk],  2
	jl    .nb331_checksingle
	jmp   .nb331_unroll_loop
.nb331_checksingle:	
	mov   edx, [esp + nb331_innerk]
	and   edx, 1
	jnz   .nb331_dosingle
	jmp   .nb331_updateouterdata
.nb331_dosingle:
	mov   edx, [esp + nb331_innerjjnr]     ;# pointer to jjnr[k] 
	mov   eax, [edx]

	mov esi, [ebp + nb331_charge]    ;# base of charge[] 
	xorpd xmm3, xmm3	
	movlpd xmm3, [esi + eax*8]
	movapd xmm4, xmm3	     
	mulpd  xmm3, [esp + nb331_iqO]
	mulpd  xmm4, [esp + nb331_iqH]

	movd  mm0, eax		;# use mmx registers as temp storage 
	movapd  [esp + nb331_qqO], xmm3
	movapd  [esp + nb331_qqH], xmm4	
	
	mov esi, [ebp + nb331_type]
	mov eax, [esi + eax*4]
	mov esi, [ebp + nb331_vdwparam]
	shl eax, 1	
	mov edi, [esp + nb331_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 + nb331_c6], xmm4
	movapd [esp + nb331_c12], xmm6
	
	mov esi, [ebp + nb331_pos]       ;# base of pos[] 
	lea   eax, [eax + eax*2]     ;# replace jnr with j3 
	
	;# move coords 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 + nb331_ixO]
	movapd xmm5, [esp + nb331_iyO]