nb_kernel332_ia32_3dnow.intel_syntax.s

最著名最快的分子模拟软件

	;# coulomb table 
	mov edx, [ebp + nb332nf_VFtab]
	mov ecx, [esp + nb332nf_n1]
	lea ecx, [ecx + ecx*2]
	shl ecx, 2

	;# load all values we need 
	movd mm4, [edx + ecx*4]
	movd mm5, [edx + ecx*4 + 4]
	movd mm6, [edx + ecx*4 + 8]
	movd mm7, [edx + ecx*4 + 12]
	
	pfmul mm6, mm0  ;# mm6 = Geps 		
	pfmul mm7, mm2	;# mm7 = Heps2 
	
	pfadd mm5, mm6
	pfadd mm5, mm7	;# mm5 = Fp 

	pfmul mm5, mm0  ;# mm5=eps*Fp 
	pfadd mm5, mm4	;#  mm5= VV 

	pfmul mm5, [esp + nb332nf_qqOO]	;# vcoul=qq*VV 
	;# update vctot directly, use mm3 for fscal sum. 
	pfadd mm5, [esp + nb332nf_vctot]
	movq [esp + nb332nf_vctot], mm5
	
	;# dispersion table 
	;# load all the table values we need 
	movd mm4, [edx + ecx*4 + 16]
	movd mm5, [edx + ecx*4 + 20]
	movd mm6, [edx + ecx*4 + 24]
	movd mm7, [edx + ecx*4 + 28]
	pfmul mm6, mm0  ;# mm6 = Geps 		
	pfmul mm7, mm2	;# mm7 = Heps2 
	pfadd mm5, mm6
	pfadd mm5, mm7	;# mm5 = Fp 
	pfmul mm5, mm0  ;# mm5=eps*Fp 
	pfadd mm5, mm4	;#  mm5= VV 	

	movq mm4, [esp + nb332nf_c6]
	pfmul mm5, mm4	;# Vvdw6            
	;# update Vvdwtot to release mm5! 
	pfadd mm5, [esp + nb332nf_Vvdwtot]      ;# add the earlier value 
	movq [esp + nb332nf_Vvdwtot], mm5       ;# store the sum       

	;# repulsion table 
	;# load all the table values we need 
	movd mm4, [edx + ecx*4 + 32]
	movd mm5, [edx + ecx*4 + 36]
	movd mm6, [edx + ecx*4 + 40]
	movd mm7, [edx + ecx*4 + 44]

	pfmul mm6, mm0  ;# mm6 = Geps 		
	pfmul mm7, mm2	;# mm7 = Heps2 
	pfadd mm5, mm6
	pfadd mm5, mm7	;# mm5 = Fp 
	pfmul mm5, mm0  ;# mm5=eps*Fp 
	pfadd mm5, mm4	;#  mm5= VV 

	movq mm6, [esp + nb332nf_c12]
	pfmul mm5, mm6	;# Vvdw12 
	;# change sign of fscal and multiply with rinv  
 	;# update Vvdwtot  
	pfadd mm5, [esp + nb332nf_Vvdwtot]      ;# add the earlier value 
	movq [esp + nb332nf_Vvdwtot], mm5       ;# store the sum       
	
	;# Ready with the oxygen - time for hydrogens 
	
	movq mm0, [esp + nb332nf_tmprsqH]

	pfrsqrt mm1, mm0
	pswapd mm0,mm0
	pfrsqrt mm2, mm0
	pswapd mm0,mm0
	punpckldq mm1,mm2	;# seeds are in mm1 now, and rsq in mm0. 

	movq mm2, mm1
	pfmul mm1,mm1
    pfrsqit1 mm1,mm0				
    pfrcpit2 mm1,mm2	;# mm1=invsqrt 
	
	pfmul mm0,mm1		;# mm0=r 
	pfmul mm0, [esp + nb332nf_tsc]
	pf2iw mm4, mm0
	movq [esp + nb332nf_n1], mm4
	pi2fd mm4,mm4
	pfsub mm0, mm4               ;# now mm0 is eps and mm4 n0 
	movq  mm2, mm0
	pfmul mm2, mm2		;# mm0 is eps, mm2 eps2 
	
	;# coulomb table 
	mov edx, [ebp + nb332nf_VFtab]
	mov ecx, [esp + nb332nf_n1]
	lea ecx, [ecx + ecx*2]
	shl ecx, 2
	;# load all values we need 
	movd mm4, [edx + ecx*4]
	movd mm5, [edx + ecx*4 + 4]
	movd mm6, [edx + ecx*4 + 8]
	movd mm7, [edx + ecx*4 + 12]
	mov ecx, [esp + nb332nf_n1+4]
	lea ecx, [ecx + ecx*2]
	shl ecx, 2
	punpckldq mm4, [edx + ecx*4]
	punpckldq mm5, [edx + ecx*4 + 4]
	punpckldq mm6, [edx + ecx*4 + 8]
	punpckldq mm7, [edx + ecx*4 + 12]
	
	pfmul mm6, mm0  ;# mm6 = Geps 		
	pfmul mm7, mm2	;# mm7 = Heps2 
	
	pfadd mm5, mm6
	pfadd mm5, mm7	;# mm5 = Fp 

	pfmul mm5, mm0  ;# mm5=eps*Fp 
	pfadd mm5, mm4	;#  mm5= VV 

	pfmul mm5, [esp + nb332nf_qqOH]	;# vcoul=qq*VV 
	;# update vctot 
	pfadd mm5, [esp + nb332nf_vctot]
	movq [esp + nb332nf_vctot], mm5

	;# interactions with j H1 
	movq  mm0, [esi + eax*4 + 12]
	movd  mm1, [esi + eax*4 + 20]
	;# copy & expand to mm2-mm4 for the H interactions 
	movq  mm2, mm0
	movq  mm3, mm0
	movq  mm4, mm1
	punpckldq mm2,mm2
	punpckhdq mm3,mm3
	punpckldq mm4,mm4
	
	pfsubr mm0, [esp + nb332nf_ixO]
	pfsubr mm1, [esp + nb332nf_izO]
		
	pfmul mm0,mm0
	pfmul mm1,mm1
	pfacc mm0, mm1
	pfadd mm0, mm1		;# mm0=rsqO 
	
	punpckldq mm2, mm2
	punpckldq mm3, mm3
	punpckldq mm4, mm4  ;# mm2-mm4 is jx-jz 
	pfsubr mm2, [esp + nb332nf_ixH]
	pfsubr mm3, [esp + nb332nf_iyH]
	pfsubr mm4, [esp + nb332nf_izH] ;# mm2-mm4 is dxH-dzH 
	
	pfmul mm2,mm2
	pfmul mm3,mm3
	pfmul mm4,mm4

	pfadd mm3,mm2
	pfadd mm3,mm4		;# mm3=rsqH 
	movq [esp + nb332nf_tmprsqH], mm3

    pfrsqrt mm1,mm0

    movq mm2,mm1
    pfmul mm1,mm1
    pfrsqit1 mm1,mm0				
    pfrcpit2 mm1,mm2	;# mm1=invsqrt 
	pfmul mm0, mm1		;# mm0=rsq  
	
	pfmul mm0, [esp + nb332nf_tsc]
	pf2iw mm4, mm0
	movd [esp + nb332nf_n1], mm4
	pi2fd mm4,mm4
	pfsub mm0, mm4               ;# now mm0 is eps and mm4 n0 
	movq  mm2, mm0
	pfmul mm2, mm2		;# mm0 is eps, mm2 eps2 

	;# coulomb table 
	mov edx, [ebp + nb332nf_VFtab]
	mov ecx, [esp + nb332nf_n1]
	lea ecx, [ecx + ecx*2]
	shl ecx, 2

	;# load all values we need 
	movd mm4, [edx + ecx*4]
	movd mm5, [edx + ecx*4 + 4]
	movd mm6, [edx + ecx*4 + 8]
	movd mm7, [edx + ecx*4 + 12]
	
	pfmul mm6, mm0  ;# mm6 = Geps 		
	pfmul mm7, mm2	;# mm7 = Heps2 
	
	pfadd mm5, mm6
	pfadd mm5, mm7	;# mm5 = Fp 

	pfmul mm5, mm0  ;# mm5=eps*Fp 
	pfadd mm5, mm4	;#  mm5= VV 

	pfmul mm5, [esp + nb332nf_qqOH]	;# vcoul=qq*VV 
	;# update vctot directly 
	pfadd mm5, [esp + nb332nf_vctot]
	movq [esp + nb332nf_vctot], mm5
	
	movq mm0, [esp + nb332nf_tmprsqH]
	pfrsqrt mm1, mm0
	pswapd mm0,mm0
	pfrsqrt mm2, mm0
	pswapd mm0,mm0
	punpckldq mm1,mm2	;# seeds are in mm1 now, and rsq in mm0. 

	movq mm2, mm1
	pfmul mm1,mm1
    pfrsqit1 mm1,mm0				
    pfrcpit2 mm1,mm2	;# mm1=invsqrt 
	
	pfmul mm0,mm1		;# mm0=r 
	pfmul mm0, [esp + nb332nf_tsc]
	pf2iw mm4, mm0
	movq [esp + nb332nf_n1], mm4
	pi2fd mm4,mm4
	pfsub mm0, mm4               ;# now mm0 is eps and mm4 n0 
	movq  mm2, mm0
	pfmul mm2, mm2		;# mm0 is eps, mm2 eps2 
	
	;# coulomb table 
	mov edx, [ebp + nb332nf_VFtab]
	mov ecx, [esp + nb332nf_n1]
	lea ecx, [ecx + ecx*2]
	shl ecx, 2
	;# load all values we need 
	movd mm4, [edx + ecx*4]
	movd mm5, [edx + ecx*4 + 4]
	movd mm6, [edx + ecx*4 + 8]
	movd mm7, [edx + ecx*4 + 12]
	mov ecx, [esp + nb332nf_n1+4]
	lea ecx, [ecx + ecx*2]
	shl ecx, 2
	punpckldq mm4, [edx + ecx*4]
	punpckldq mm5, [edx + ecx*4 + 4]
	punpckldq mm6, [edx + ecx*4 + 8]
	punpckldq mm7, [edx + ecx*4 + 12]

	
	pfmul mm6, mm0  ;# mm6 = Geps 		
	pfmul mm7, mm2	;# mm7 = Heps2 
	
	pfadd mm5, mm6
	pfadd mm5, mm7	;# mm5 = Fp 

	pfmul mm5, mm0  ;# mm5=eps*Fp 
	pfadd mm5, mm4	;#  mm5= VV 

	pfmul mm5, [esp + nb332nf_qqHH]	;# vcoul=qq*VV 
	;# update vctot 
	pfadd mm5, [esp + nb332nf_vctot]
	movq [esp + nb332nf_vctot], mm5
	
	;# interactions with j H2 
	movq  mm0, [esi + eax*4 + 24]
	movd  mm1, [esi + eax*4 + 32]
	;# copy & expand to mm2-mm4 for the H interactions 
	movq  mm2, mm0
	movq  mm3, mm0
	movq  mm4, mm1
	punpckldq mm2,mm2
	punpckhdq mm3,mm3
	punpckldq mm4,mm4

	pfsubr mm0, [esp + nb332nf_ixO]
	pfsubr mm1, [esp + nb332nf_izO]
		
	pfmul mm0,mm0
	pfmul mm1,mm1
	pfacc mm0, mm1
	pfadd mm0, mm1		;# mm0=rsqO 
	
	punpckldq mm2, mm2
	punpckldq mm3, mm3
	punpckldq mm4, mm4  ;# mm2-mm4 is jx-jz 
	pfsubr mm2, [esp + nb332nf_ixH]
	pfsubr mm3, [esp + nb332nf_iyH]
	pfsubr mm4, [esp + nb332nf_izH] ;# mm2-mm4 is dxH-dzH 
	
	pfmul mm2,mm2
	pfmul mm3,mm3
	pfmul mm4,mm4

	pfadd mm3,mm2
	pfadd mm3,mm4		;# mm3=rsqH 
	movq [esp + nb332nf_tmprsqH], mm3

    pfrsqrt mm1,mm0

    movq mm2,mm1
    pfmul mm1,mm1
    pfrsqit1 mm1,mm0				
    pfrcpit2 mm1,mm2	;# mm1=invsqrt 
	pfmul mm0, mm1

	pfmul mm0, [esp + nb332nf_tsc]
	pf2iw mm4, mm0
	movd [esp + nb332nf_n1], mm4
	pi2fd mm4,mm4
	pfsub mm0, mm4               ;# now mm0 is eps and mm4 n0 
	movq  mm2, mm0
	pfmul mm2, mm2		;# mm0 is eps, mm2 eps2 

	;# coulomb table 
	mov edx, [ebp + nb332nf_VFtab]
	mov ecx, [esp + nb332nf_n1]
	lea ecx, [ecx + ecx*2]
	shl ecx, 2

	;# load all values we need 
	movd mm4, [edx + ecx*4]
	movd mm5, [edx + ecx*4 + 4]
	movd mm6, [edx + ecx*4 + 8]
	movd mm7, [edx + ecx*4 + 12]
	
	pfmul mm6, mm0  ;# mm6 = Geps 		
	pfmul mm7, mm2	;# mm7 = Heps2 
	
	pfadd mm5, mm6
	pfadd mm5, mm7	;# mm5 = Fp 

	pfmul mm5, mm0  ;# mm5=eps*Fp 
	pfadd mm5, mm4	;#  mm5= VV 

	pfmul mm5, [esp + nb332nf_qqOH]	;# vcoul=qq*VV 
	;# update vctot directly 
	pfadd mm5, [esp + nb332nf_vctot]
	movq [esp + nb332nf_vctot], mm5
	
	movq mm0, [esp + nb332nf_tmprsqH]
	pfrsqrt mm1, mm0
	pswapd mm0,mm0
	pfrsqrt mm2, mm0
	pswapd mm0,mm0
	punpckldq mm1,mm2	;# seeds are in mm1 now, and rsq in mm0. 

	movq mm2, mm1
	pfmul mm1,mm1
    pfrsqit1 mm1,mm0				
    pfrcpit2 mm1,mm2	;# mm1=invsqrt 
	
	pfmul mm0,mm1		;# mm0=r 
	pfmul mm0, [esp + nb332nf_tsc]
	pf2iw mm4, mm0
	movq [esp + nb332nf_n1], mm4
	pi2fd mm4,mm4
	pfsub mm0, mm4               ;# now mm0 is eps and mm4 n0 
	movq  mm2, mm0
	pfmul mm2, mm2		;# mm0 is eps, mm2 eps2 
	
	;# coulomb table 
	mov edx, [ebp + nb332nf_VFtab]
	mov ecx, [esp + nb332nf_n1]
	lea ecx, [ecx + ecx*2]
	shl ecx, 2
	;# load all values we need 
	movd mm4, [edx + ecx*4]
	movd mm5, [edx + ecx*4 + 4]
	movd mm6, [edx + ecx*4 + 8]
	movd mm7, [edx + ecx*4 + 12]
	mov ecx, [esp + nb332nf_n1+4] ;# mm5 = Fp 
	lea ecx, [ecx + ecx*2]
	shl ecx, 2
	punpckldq mm4, [edx + ecx*4]
	punpckldq mm5, [edx + ecx*4 + 4]
	punpckldq mm6, [edx + ecx*4 + 8]
	punpckldq mm7, [edx + ecx*4 + 12]

	
	pfmul mm6, mm0  ;# mm6 = Geps 		
	pfmul mm7, mm2	;# mm7 = Heps2 
	
	pfadd mm5, mm6
	pfadd mm5, mm7	;# mm5 = Fp 

	pfmul mm5, mm0  ;# mm5=eps*Fp 
	pfadd mm5, mm4	;#  mm5= VV 

	pfmul mm5, [esp + nb332nf_qqHH]	;# vcoul=qq*VV 
	;# update vctot 
	pfadd mm5, [esp + nb332nf_vctot]
	movq [esp + nb332nf_vctot], mm5
		
	;#  done  - one more? 
	dec dword ptr [esp + nb332nf_innerk]
	jz  .nb332nf_updateouterdata
	jmp .nb332nf_inner_loop	
.nb332nf_updateouterdata:	
	;# get n from stack
	mov esi, [esp + nb332nf_n]
        ;# get group index for i particle 
        mov   edx, [ebp + nb332nf_gid]      	;# base of gid[]
        mov   edx, [edx + esi*4]		;# ggid=gid[n]

	movq  mm7, [esp + nb332nf_vctot]     
	pfacc mm7,mm7	          ;# get and sum the two parts of total potential 

	mov   eax, [ebp + nb332nf_Vc]
	movd  mm6, [eax + edx*4] 
	pfadd mm6, mm7
	movd  [eax + edx*4], mm6          ;# increment vc[gid] 

	movq  mm7, [esp + nb332nf_Vvdwtot]     
	pfacc mm7,mm7	          ;# get and sum the two parts of total potential 

	mov   eax, [ebp + nb332nf_Vvdw]
	movd  mm6, [eax + edx*4] 
	pfadd mm6, mm7
	movd  [eax + edx*4], mm6          ;# increment Vvdwtot[gid] 
       	;# finish if last 
        mov ecx, [esp + nb332nf_nn1]
	;# esi already loaded with n
	inc esi
        sub ecx, esi
        jecxz .nb332nf_outerend

        ;# not last, iterate outer loop once more!  
        mov [esp + nb332nf_n], esi
        jmp .nb332nf_outer
.nb332nf_outerend:
        ;# check if more outer neighborlists remain
        mov   ecx, [esp + nb332nf_nri]
	;# esi already loaded with n above
        sub   ecx, esi
        jecxz .nb332nf_end
        ;# non-zero, do one more workunit
        jmp   .nb332nf_threadloop
.nb332nf_end:
	femms
	mov eax, [esp + nb332nf_nouter] 	
	mov ebx, [esp + nb332nf_ninner]
	mov ecx, [ebp + nb332nf_outeriter]
	mov edx, [ebp + nb332nf_inneriter]
	mov [ecx], eax
	mov [edx], ebx

	add esp, 152
	pop edi
	pop esi
    	pop edx
    	pop ecx
    	pop ebx
    	pop eax
	leave
	ret