nb_kernel430_ia32_sse2.intel_syntax.s

来自「最著名最快的分子模拟软件」· S 代码 · 共 1,716 行 · 第 1/4 页

	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 + nb430_two]	;# two*Heps2 
	movapd xmm3, [esp + nb430_qq]
	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 
	;# get jnr from regs
	movd ecx, mm2
	movd edx, mm3
	mov esi, [ebp + nb430_dvda]
	
	;# Calculate dVda
	xorpd xmm7, xmm7
	mulpd xmm3, [esp + nb430_gbscale]
	movapd xmm6, xmm3
	mulpd  xmm6, [esp + nb430_r]
	addpd  xmm6, xmm5
	addpd  xmm5, [esp + nb430_vctot]
	movapd [esp + nb430_vctot], xmm5 

	;# xmm6=(vcoul+fijC*r)
	subpd  xmm7, xmm6
	movapd xmm6, xmm7
	
	;# update dvdasum
	addpd  xmm7, [esp + nb430_dvdasum]
	movapd [esp + nb430_dvdasum], xmm7 

	;# update j atoms dvdaj
	movhlps xmm7, xmm6
	addsd  xmm6, [esi + ecx*8]
	addsd  xmm7, [esi + edx*8]
	movsd  [esi + ecx*8], xmm6
	movsd  [esi + edx*8], xmm7
	
	;# put scalar force on stack temporarily 
	movapd [esp + nb430_fscal], xmm3

	movapd xmm4, [esp + nb430_r]
	mulpd  xmm4, [esp + nb430_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

	mov  esi, [ebp + nb430_VFtab]

	movd ecx, mm6
	psrlq mm6, 32
	movd edx, mm6		;# indices in eax/ebx 

	;# Dispersion 
	movapd xmm4, [esi + ecx*8]	;# Y1 F1 	
	movapd xmm3, [esi + edx*8]	;# Y2 F2 
	movapd xmm5, xmm4
	unpcklpd xmm4, xmm3	;# Y1 Y2 
	unpckhpd xmm5, xmm3	;# F1 F2 

	movapd xmm6, [esi + ecx*8 + 16]	;# G1 H1 	
	movapd xmm3, [esi + edx*8 + 16]	;# 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 + nb430_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 + nb430_c6]
	mulpd  xmm7, xmm4	 ;# fijD 
	mulpd  xmm5, xmm4	 ;# Vvdw6
	mulpd  xmm7, [esp + nb430_tsc]
	addpd  xmm7, [esp + nb430_fscal] ;# add to fscal 

	;# put scalar force back on stack Update Vvdwtot directly 
	addpd  xmm5, [esp + nb430_Vvdwtot]
	movapd [esp + nb430_fscal], xmm7
	movapd [esp + nb430_Vvdwtot], xmm5

	;# Repulsion 
	movapd xmm4, [esi + ecx*8 + 32]	;# Y1 F1 	
	movapd xmm3, [esi + edx*8 + 32]	;# Y2 F2 
	movapd xmm5, xmm4
	unpcklpd xmm4, xmm3	;# Y1 Y2 
	unpckhpd xmm5, xmm3	;# F1 F2 

	movapd xmm6, [esi + ecx*8 + 48]	;# G1 H1 	
	movapd xmm3, [esi + edx*8 + 48]	;# 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 + nb430_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 + nb430_c12]
	mulpd  xmm7, xmm4 ;# fijR 
	mulpd  xmm5, xmm4 ;# Vvdw12 
	mulpd  xmm7, [esp + nb430_tsc]
	addpd  xmm7, [esp + nb430_fscal] 
	
	addpd  xmm5, [esp + nb430_Vvdwtot]
	movapd [esp + nb430_Vvdwtot], xmm5
	xorpd  xmm4, xmm4

	mulpd xmm7, xmm0
	subpd xmm4, xmm7

	movapd xmm0, [esp + nb430_dx]
	movapd xmm1, [esp + nb430_dy]
	movapd xmm2, [esp + nb430_dz]

	mov    edi, [ebp + nb430_faction]
	mulpd  xmm0, xmm4
	mulpd  xmm1, xmm4
	mulpd  xmm2, xmm4
	;# xmm0-xmm2 contains tx-tz (partial force) 
	;# now update f_i 
	movapd xmm3, [esp + nb430_fix]
	movapd xmm4, [esp + nb430_fiy]
	movapd xmm5, [esp + nb430_fiz]
	addpd  xmm3, xmm0
	addpd  xmm4, xmm1
	addpd  xmm5, xmm2
	movapd [esp + nb430_fix], xmm3
	movapd [esp + nb430_fiy], xmm4
	movapd [esp + nb430_fiz], xmm5
	;# 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 + nb430_innerk],  2
	jl    .nb430_checksingle
	jmp   .nb430_unroll_loop
.nb430_checksingle:
	mov   edx, [esp + nb430_innerk]
	and   edx, 1
	jnz    .nb430_dosingle
	jmp    .nb430_updateouterdata
.nb430_dosingle:
	mov esi, [ebp + nb430_charge]
	mov edx, [ebp + nb430_invsqrta]
	mov edi, [ebp + nb430_pos]
	mov   ecx, [esp + nb430_innerjjnr]
	mov   eax, [ecx]	

	xorpd  xmm6, xmm6
	movapd xmm7, xmm6
	movsd  xmm7, [edx + eax*8]
	movlpd xmm6, [esi + eax*8]	;# xmm6(0) has the charge
	mulsd  xmm7, [esp + nb430_isai]
	movapd [esp + nb430_isaprod], xmm7
	movapd xmm1, xmm7
	mulpd xmm1, [esp + nb430_gbtsc]
	movapd [esp + nb430_gbscale], xmm1
	
	mulsd  xmm7, [esp + nb430_iq]
	mulsd  xmm6, xmm7
	movapd [esp + nb430_qq], xmm6
	
	mov esi, [ebp + nb430_type]
	mov edx, [esi + eax*4]
	mov esi, [ebp + nb430_vdwparam]
	shl edx, 1
	mov edi, [esp + nb430_ntia]
	add edx, edi

	movlpd xmm6, [esi + edx*8]	;# c6a
	movhpd xmm6, [esi + edx*8 + 8]	;# c6a c12a 

	xorpd xmm7, xmm7
	movapd xmm4, xmm6
	unpcklpd xmm4, xmm7
	unpckhpd xmm6, xmm7
	
	movapd [esp + nb430_c6], xmm4
	movapd [esp + nb430_c12], xmm6
	
	mov esi, [ebp + nb430_pos]		;# base of pos[]
	
	movd  mm2, eax
	lea   eax, [eax + eax*2]     ;# replace jnr with j3 

	;# move two coordinates to xmm0-xmm2 
	movlpd xmm0, [esi + eax*8]
	movlpd xmm1, [esi + eax*8 + 8]
	movlpd xmm2, [esi + eax*8 + 16]

	mov    edi, [ebp + nb430_faction]

	;# move nb430_ix-iz to xmm4-xmm6 
	movapd xmm4, [esp + nb430_ix]
	movapd xmm5, [esp + nb430_iy]
	movapd xmm6, [esp + nb430_iz]

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

	;# store dr 
	movapd [esp + nb430_dx], xmm4
	movapd [esp + nb430_dy], xmm5
	movapd [esp + nb430_dz], xmm6
	;# square it 
	mulsd xmm4,xmm4
	mulsd xmm5,xmm5
	mulsd xmm6,xmm6
	addsd xmm4, xmm5
	addsd xmm4, xmm6
	;# rsq in xmm4 

	cvtsd2ss xmm5, xmm4	
	rsqrtss xmm5, xmm5
	cvtss2sd xmm2, xmm5	;# lu in low xmm2 

	;# lookup seed in xmm2 
	movapd xmm5, xmm2	;# copy of lu 
	mulsd xmm2, xmm2	;# lu*lu 
	movapd xmm1, [esp + nb430_three]
	mulsd xmm2, xmm4	;# rsq*lu*lu 			
	movapd xmm0, [esp + nb430_half]
	subsd xmm1, xmm2	;# 30-rsq*lu*lu 
	mulsd xmm1, xmm5	
	mulsd xmm1, xmm0	;# xmm0=iter1 of rinv (new lu) 

	movapd xmm5, xmm1	;# copy of lu 
	mulsd xmm1, xmm1	;# lu*lu 
	movapd xmm2, [esp + nb430_three]
	mulsd xmm1, xmm4	;# rsq*lu*lu 			
	movapd xmm0, [esp + nb430_half]
	subsd xmm2, xmm1	;# 30-rsq*lu*lu 
	mulsd xmm2, xmm5	
	mulsd xmm0, xmm2	;# xmm0=iter2 of rinv (new lu) 
	mulsd xmm4, xmm0	;# xmm4=r 
	movsd [esp + nb430_r], xmm4
	mulsd xmm4, [esp + nb430_gbscale]
	
	cvttsd2si edx, xmm4	;# mm6 = lu idx 
	cvtsi2sd xmm5, edx
	subsd xmm4, xmm5
	movapd xmm1, xmm4	;# xmm1=eps 
	movapd xmm2, xmm1	
	mulsd  xmm2, xmm2	;# xmm2=eps2 
	
	shl edx, 2		;# idx *= 4 
	mov  esi, [ebp + nb430_GBtab]

	;# Coulomb 
	movapd xmm4, [esi + edx*8]	;# Y1 F1 	
	xorpd xmm3, xmm3
	movapd xmm5, xmm4
	unpcklpd xmm4, xmm3	;# Y1 
	unpckhpd xmm5, xmm3	;# F1 

	movapd xmm6, [esi + edx*8 + 16]	;# G1 H1 	
	xorpd xmm3, xmm3
	movapd xmm7, xmm6
	unpcklpd xmm6, xmm3	;# G1 
	unpckhpd xmm7, xmm3	;# H1 
	;# coulomb 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 + nb430_two]	;# two*Heps2 
	movapd xmm3, [esp + nb430_qq]
	addsd  xmm7, xmm6
	addsd  xmm7, xmm5 ;# xmm7=FF 
	mulsd  xmm5, xmm1 ;# xmm5=eps*Fp 
	addsd  xmm5, xmm4 ;# xmm5=VV 
	mulsd  xmm5, xmm3 ;# vcoul=qq*VV  
	mulsd  xmm3, xmm7 ;# fijC=FF*qq 
	;# get jnr from regs
	movd ebx, mm2
	mov esi, [ebp + nb430_dvda]
	
	;# Calculate dVda
	xorpd xmm7, xmm7
	mulsd xmm3, [esp + nb430_gbscale]
	movsd xmm6, xmm3
	mulsd  xmm6, [esp + nb430_r]
	addsd  xmm6, xmm5
	addsd  xmm5, [esp + nb430_vctot]
	movsd [esp + nb430_vctot], xmm5 

	;# xmm6=(vcoul+fijC*r)
	subpd xmm7, xmm6
	movsd xmm6, xmm7
	
	;# update dvdasum
	addsd  xmm7, [esp + nb430_dvdasum]
	movsd [esp + nb430_dvdasum], xmm7 

	;# update j atoms dvdaj
	addsd  xmm6, [esi + ebx*8]
	movsd  [esi + ebx*8], xmm6
	
	;# put scalar force on stack temporarily 
	movsd [esp + nb430_fscal], xmm3

	movsd xmm4, [esp + nb430_r]
	mulsd  xmm4, [esp + nb430_tsc]
	cvttsd2si edx, xmm4	;# mm6 = lu idx 
	cvtsi2sd xmm5, edx
	subsd xmm4, xmm5
	movsd xmm1, xmm4	;# xmm1=eps 
	movsd xmm2, xmm1	
	mulsd  xmm2, xmm2	;# xmm2=eps2 

	shl edx, 3

	mov  esi, [ebp + nb430_VFtab]

	;# Dispersion 
	movapd xmm4, [esi + edx*8]	;# Y1 F1 	
	xorpd xmm3, xmm3
	movapd xmm5, xmm4
	unpcklpd xmm4, xmm3	;# Y1 
	unpckhpd xmm5, xmm3	;# F1 

	movapd xmm6, [esi + edx*8 + 16]	;# G1 H1 	
	xorpd xmm3, xmm3
	movapd xmm7, xmm6
	unpcklpd xmm6, xmm3	;# G1 
	unpckhpd xmm7, xmm3	;# H1 
	;# 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 + nb430_two]	;# two*Heps2 
	movapd xmm3, [esp + nb430_qq]
	addsd  xmm7, xmm6
	addsd  xmm7, xmm5 ;# xmm7=FF 
	mulsd  xmm5, xmm1 ;# xmm5=eps*Fp 
	addsd  xmm5, xmm4 ;# xmm5=VV 

	movapd xmm4, [esp + nb430_c6]
	mulsd  xmm7, xmm4	 ;# fijD 
	mulsd  xmm5, xmm4	 ;# Vvdw6
	mulpd  xmm7, [esp + nb430_tsc]
	addsd  xmm7, [esp + nb430_fscal] ;# add to fscal 

	;# put scalar force back on stack Update Vvdwtot directly 
	addsd  xmm5, [esp + nb430_Vvdwtot]
	movlpd [esp + nb430_fscal], xmm7
	movlpd [esp + nb430_Vvdwtot], xmm5

	;# Repulsion 
	movapd xmm4, [esi + edx*8 + 32]	;# Y1 F1 	
	xorpd xmm3, xmm3
	movapd xmm5, xmm4
	unpcklpd xmm4, xmm3	;# Y1 
	unpckhpd xmm5, xmm3	;# F1 

	movapd xmm6, [esi + edx*8 + 48]	;# G1 H1 	
	xorpd xmm3, xmm3
	movapd xmm7, xmm6
	unpcklpd xmm6, xmm3	;# G1 
	unpckhpd xmm7, xmm3	;# H1 
	;# 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 + nb430_two]	;# two*Heps2 
	movapd xmm3, [esp + nb430_qq]
	addsd  xmm7, xmm6
	addsd  xmm7, xmm5 ;# xmm7=FF 
	mulsd  xmm5, xmm1 ;# xmm5=eps*Fp 
	addsd  xmm5, xmm4 ;# xmm5=VV 

	movapd xmm4, [esp + nb430_c12]
	mulsd  xmm7, xmm4 ;# fijR 
	mulsd  xmm5, xmm4 ;# Vvdw12 
	mulpd  xmm7, [esp + nb430_tsc]
	addsd  xmm7, [esp + nb430_fscal] 
	
	addsd  xmm5, [esp + nb430_Vvdwtot]
	movlpd [esp + nb430_Vvdwtot], xmm5
	xorpd  xmm4, xmm4

	mulsd xmm7, xmm0
	subsd xmm4, xmm7

	movapd xmm0, [esp + nb430_dx]
	movapd xmm1, [esp + nb430_dy]
	movapd xmm2, [esp + nb430_dz]