nb_kernel314_ia32_sse2.intel_syntax.s

来自「最著名最快的分子模拟软件」· S 代码 · 共 2,193 行 · 第 1/5 页

	
	pslld mm6, 2		;# idx *= 4 
	mov  esi, [ebp + nb314_VFtab]
	movd eax, mm6
	psrlq mm6, 32
	movd ebx, mm6		;# indices in eax/ebx 

	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 + nb314_two]	;# two*Heps2 
	movapd xmm3, [esp + nb314_qqMH]
	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 

    	addpd  xmm5, [esp + nb314_vctot]
    	movapd [esp + nb314_vctot], xmm5
	xorpd  xmm1, xmm1
	mulpd  xmm3,  [esp + nb314_tsc]
	mulpd  xmm3, xmm0
	subpd  xmm1, xmm3

	movapd xmm0, xmm1
	movapd xmm2, xmm1
	
	movapd xmm3, [esp + nb314_fjxM]
	movapd xmm4, [esp + nb314_fjyM]
	movapd xmm5, [esp + nb314_fjzM]
	mulpd xmm0, [esp + nb314_dxH2M]
	mulpd xmm1, [esp + nb314_dyH2M]
	mulpd xmm2, [esp + nb314_dzH2M]
	subpd xmm3, xmm0
	subpd xmm4, xmm1
	subpd xmm5, xmm2
	addpd xmm0, [esp + nb314_fixH2]
	addpd xmm1, [esp + nb314_fiyH2]
	addpd xmm2, [esp + nb314_fizH2]
	movapd [esp + nb314_fjxM], xmm3
	movapd [esp + nb314_fjyM], xmm4
	movapd [esp + nb314_fjzM], xmm5
	movapd [esp + nb314_fixH2], xmm0
	movapd [esp + nb314_fiyH2], xmm1
	movapd [esp + nb314_fizH2], xmm2

	;# M-H1 interaction 
	movapd xmm0, [esp + nb314_rinvMH1]
	movapd xmm1, xmm0
	mulpd  xmm1, [esp + nb314_rsqMH1] ;# xmm1=r 
	mulpd  xmm1, [esp + nb314_tsc]	
	cvttpd2pi mm6, xmm1	;# mm6 = lu idx 
	cvtpi2pd xmm6, mm6
	subpd xmm1, xmm6	;# xmm1=eps 
	movapd xmm2, xmm1	
	mulpd  xmm2, xmm2	;# xmm2=eps2 
	
	pslld mm6, 2		;# idx *= 4 
	mov  esi, [ebp + nb314_VFtab]
	movd eax, mm6
	psrlq mm6, 32
	movd ebx, mm6		;# indices in eax/ebx 

	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 + nb314_two]	;# two*Heps2 
	movapd xmm3, [esp + nb314_qqMH]
	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 

    	addpd  xmm5, [esp + nb314_vctot]
    	movapd [esp + nb314_vctot], xmm5
	xorpd  xmm1, xmm1
	mulpd  xmm3,  [esp + nb314_tsc]
	mulpd  xmm3, xmm0
	subpd  xmm1, xmm3

	movapd xmm0, xmm1
	movapd xmm2, xmm1

	movapd xmm3, [esp + nb314_fjxH1]
	movapd xmm4, [esp + nb314_fjyH1]
	movapd xmm5, [esp + nb314_fjzH1]
	mulpd xmm0, [esp + nb314_dxMH1]
	mulpd xmm1, [esp + nb314_dyMH1]
	mulpd xmm2, [esp + nb314_dzMH1]
	subpd xmm3, xmm0
	subpd xmm4, xmm1
	subpd xmm5, xmm2
	addpd xmm0, [esp + nb314_fixM]
	addpd xmm1, [esp + nb314_fiyM]
	addpd xmm2, [esp + nb314_fizM]
	movapd [esp + nb314_fjxH1], xmm3
	movapd [esp + nb314_fjyH1], xmm4
	movapd [esp + nb314_fjzH1], xmm5
	movapd [esp + nb314_fixM], xmm0
	movapd [esp + nb314_fiyM], xmm1
	movapd [esp + nb314_fizM], xmm2

	;# M-H2 interaction 
	movapd xmm0, [esp + nb314_rinvMH2]
	movapd xmm1, xmm0
	mulpd  xmm1, [esp + nb314_rsqMH2] ;# xmm1=r 
	mulpd  xmm1, [esp + nb314_tsc]
	cvttpd2pi mm6, xmm1	;# mm6 = lu idx 
	cvtpi2pd xmm6, mm6
	subpd xmm1, xmm6	;# xmm1=eps 
	movapd xmm2, xmm1	
	mulpd  xmm2, xmm2	;# xmm2=eps2 
	
	pslld mm6, 2		;# idx *= 4 
	mov  esi, [ebp + nb314_VFtab]
	movd eax, mm6
	psrlq mm6, 32
	movd ebx, mm6		;# indices in eax/ebx 

	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 + nb314_two]	;# two*Heps2 
	movapd xmm3, [esp + nb314_qqMH]
	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 

    	addpd  xmm5, [esp + nb314_vctot]
    	movapd [esp + nb314_vctot], xmm5
	xorpd  xmm1, xmm1
	mulpd  xmm3,  [esp + nb314_tsc]
	mulpd  xmm3, xmm0
	subpd  xmm1, xmm3

	movapd xmm0, xmm1
	movapd xmm2, xmm1
	
	movapd xmm3, [esp + nb314_fjxH2]
	movapd xmm4, [esp + nb314_fjyH2]
	movapd xmm5, [esp + nb314_fjzH2]
	mulpd xmm0, [esp + nb314_dxMH2]
	mulpd xmm1, [esp + nb314_dyMH2]
	mulpd xmm2, [esp + nb314_dzMH2]
	subpd xmm3, xmm0
	subpd xmm4, xmm1
	subpd xmm5, xmm2
	addpd xmm0, [esp + nb314_fixM]
	addpd xmm1, [esp + nb314_fiyM]
	addpd xmm2, [esp + nb314_fizM]
	movapd [esp + nb314_fjxH2], xmm3
	movapd [esp + nb314_fjyH2], xmm4
	movapd [esp + nb314_fjzH2], xmm5
	movapd [esp + nb314_fixM], xmm0
	movapd [esp + nb314_fiyM], xmm1
	movapd [esp + nb314_fizM], xmm2

	;# M-M interaction 
	movapd xmm0, [esp + nb314_rinvMM]
	movapd xmm1, xmm0
	mulpd  xmm1, [esp + nb314_rsqMM] ;# xmm1=r 
	mulpd  xmm1, [esp + nb314_tsc]	
	cvttpd2pi mm6, xmm1	;# mm6 = lu idx 
	cvtpi2pd xmm6, mm6
	subpd xmm1, xmm6	;# xmm1=eps 
	movapd xmm2, xmm1	
	mulpd  xmm2, xmm2	;# xmm2=eps2 
	
	pslld mm6, 2		;# idx *= 4 
	mov  esi, [ebp + nb314_VFtab]
	movd eax, mm6
	psrlq mm6, 32
	movd ebx, mm6		;# indices in eax/ebx 

	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 + nb314_two]	;# two*Heps2 
	movapd xmm3, [esp + nb314_qqMM]
	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 
	
    	addpd  xmm5, [esp + nb314_vctot]
    	movapd [esp + nb314_vctot], xmm5
	xorpd  xmm1, xmm1
	mulpd  xmm3,  [esp + nb314_tsc]
	mulpd  xmm3, xmm0
	subpd  xmm1, xmm3

	movapd xmm0, xmm1
	movapd xmm2, xmm1
	
	movapd xmm3, [esp + nb314_fjxM]
	movapd xmm4, [esp + nb314_fjyM] 
	movapd xmm5, [esp + nb314_fjzM]
	mulpd xmm0, [esp + nb314_dxMM]
	mulpd xmm1, [esp + nb314_dyMM]
	mulpd xmm2, [esp + nb314_dzMM]
	subpd xmm3, xmm0
	subpd xmm4, xmm1
	subpd xmm5, xmm2
	addpd xmm0, [esp + nb314_fixM]
	addpd xmm1, [esp + nb314_fiyM]
	addpd xmm2, [esp + nb314_fizM]
	movapd [esp + nb314_fjxM], xmm3
	movapd [esp + nb314_fjyM], xmm4
	movapd [esp + nb314_fjzM], xmm5
	movapd [esp + nb314_fixM], xmm0
	movapd [esp + nb314_fiyM], xmm1
	movapd [esp + nb314_fizM], xmm2

	mov edi, [ebp + nb314_faction]

	movd eax, mm0
	movd ebx, mm1
	
	;# Did all interactions - now update j forces 
	;# Step1 - transpose fjxO, fjyO and fjzO, fjxH1
	movapd xmm0, [esp + nb314_fjxO]
	movapd xmm1, [esp + nb314_fjyO]
	movapd xmm2, [esp + nb314_fjzO]
	movapd xmm3, [esp + nb314_fjxH1]
	movapd xmm4, xmm0
	movapd xmm5, xmm2
	unpcklpd xmm0, xmm1   ;# fjOxA fjOyA
	unpckhpd xmm4, xmm1   ;# fjOxB fjOyB
	unpcklpd xmm2, xmm3   ;# fjOzA fjH1xA
	unpckhpd xmm5, xmm3   ;# fjOzB fjH1xB
	movlpd xmm1, [edi + eax*8]
	movlpd xmm3, [edi + ebx*8]
	movhpd xmm1, [edi + eax*8 + 8]
	movhpd xmm3, [edi + ebx*8 + 8]
	movlpd xmm6, [edi + eax*8 + 16]
	movlpd xmm7, [edi + ebx*8 + 16]
	movhpd xmm6, [edi + eax*8 + 24]
	movhpd xmm7, [edi + ebx*8 + 24]
	addpd  xmm1, xmm0
	addpd  xmm3, xmm4
	addpd  xmm6, xmm2
	addpd  xmm7, xmm5
	movlpd [edi + eax*8],      xmm1
	movlpd [edi + ebx*8],      xmm3
	movhpd [edi + eax*8 + 8],  xmm1
	movhpd [edi + ebx*8 + 8],  xmm3
	movlpd [edi + eax*8 + 16], xmm6
	movlpd [edi + ebx*8 + 16], xmm7
	movhpd [edi + eax*8 + 24], xmm6
	movhpd [edi + ebx*8 + 24], xmm7

	;# Step2 - transpose fjyH1, fjzH1 and fjxH2, fjyH2
	movapd xmm0, [esp + nb314_fjyH1]
	movapd xmm1, [esp + nb314_fjzH1]
	movapd xmm2, [esp + nb314_fjxH2]
	movapd xmm3, [esp + nb314_fjyH2]
	movapd xmm4, xmm0
	movapd xmm5, xmm2
	unpcklpd xmm0, xmm1   ;# fjOxA fjOyA
	unpckhpd xmm4, xmm1   ;# fjOxB fjOyB
	unpcklpd xmm2, xmm3   ;# fjOzA fjH1xA
	unpckhpd xmm5, xmm3   ;# fjOzB fjH1xB
	movlpd xmm1, [edi + eax*8 + 32]
	movlpd xmm3, [edi + ebx*8 + 32]
	movhpd xmm1, [edi + eax*8 + 40]
	movhpd xmm3, [edi + ebx*8 + 40]
	movlpd xmm6, [edi + eax*8 + 48]
	movlpd xmm7, [edi + ebx*8 + 48]
	movhpd xmm6, [edi + eax*8 + 56]
	movhpd xmm7, [edi + ebx*8 + 56]
	addpd  xmm1, xmm0
	addpd  xmm3, xmm4
	addpd  xmm6, xmm2
	addpd  xmm7, xmm5
	movlpd [edi + eax*8 + 32], xmm1
	movlpd [edi + ebx*8 + 32], xmm3
	movhpd [edi + eax*8 + 40], xmm1
	movhpd [edi + ebx*8 + 40], xmm3
	movlpd [edi + eax*8 + 48], xmm6
	movlpd [edi + ebx*8 + 48], xmm7
	movhpd [edi + eax*8 + 56], xmm6
	movhpd [edi + ebx*8 + 56], xmm7

	;# Step3 - transpose fjzH2, fjxM and fjyM, fjzM
	movapd xmm0, [esp + nb314_fjzH2]
	movapd xmm1, [esp + nb314_fjxM]
	movapd xmm2, [esp + nb314_fjyM]
	movapd xmm3, [esp + nb314_fjzM]
	movapd xmm4, xmm0
	movapd xmm5, xmm2
	unpcklpd xmm0, xmm1   ;# fjOxA fjOyA
	unpckhpd xmm4, xmm1   ;# fjOxB fjOyB
	unpcklpd xmm2, xmm3   ;# fjOzA fjH1xA
	unpckhpd xmm5, xmm3   ;# fjOzB fjH1xB
	movlpd xmm1, [edi + eax*8 + 64]
	movlpd xmm3, [edi + ebx*8 + 64]
	movhpd xmm1, [edi + eax*8 + 72]
	movhpd xmm3, [edi + ebx*8 + 72]
	movlpd xmm6, [edi + eax*8 + 80]
	movlpd xmm7, [edi + ebx*8 + 80]
	movhpd xmm6, [edi + eax*8 + 88]
	movhpd xmm7, [edi + ebx*8 + 88]
	addpd  xmm1, xmm0
	addpd  xmm3, xmm4
	addpd  xmm6, xmm2
	addpd  xmm7, xmm5
	movlpd [edi + eax*8 + 64], xmm1
	movlpd [edi + ebx*8 + 64], xmm3
	movhpd [edi + eax*8 + 72], xmm1
	movhpd [edi + ebx*8 + 72], xmm3
	movlpd [edi + eax*8 + 80], xmm6
	movlpd [edi + ebx*8 + 80], xmm7
	movhpd [edi + eax*8 + 88], xmm6
	movhpd [edi + ebx*8 + 88], xmm7

	;# should we do one more iteration? 
	sub dword ptr [esp + nb314_innerk],  2
	jl    .nb314_checksingle
	jmp   .nb314_unroll_loop
.nb314_checksingle:
	mov   edx, [esp + nb314_innerk]
	and   edx, 1
	jnz   .nb314_dosingle
	jmp   .nb314_updateouterdata
.nb314_dosingle:
	mov   edx, [esp + nb314_innerjjnr]     ;# pointer to jjnr[k] 
	mov   eax, [edx]

	mov esi, [ebp + nb314_pos]       ;# base of pos[] 

	lea   eax, [eax + eax*2]     ;# replace jnr with j3 
	
	;# move j coordinates to local temp variables 
	;# load ox, oy, oz, h1x
	movlpd xmm0, [esi + eax*8]
	movhpd xmm0, [esi + eax*8 + 8]
	movlpd xmm1, [esi + eax*8 + 16]
	movhpd xmm1, [esi + eax*8 + 24]
	movlpd xmm2, [esi + eax*8 + 32]
	movhpd xmm2, [esi + eax*8 + 40]
	movlpd xmm3, [esi + eax*8 + 48]
	movhpd xmm3, [esi + eax*8 + 56]
	movlpd xmm4, [esi + eax*8 + 64]
	movhpd xmm4, [esi + eax*8 + 72]
	movlpd xmm5, [esi + eax*8 + 80]
	movhpd xmm5, [esi + eax*8 + 88]
	movsd  [esp + nb314_jxO], xmm0
	movsd  [esp + nb314_jzO], xmm1
	movsd  [esp + nb314_jyH1], xmm2
	movsd  [esp + nb314_jxH2], xmm3