nb_kernel331_x86_64_sse2.intel_syntax.s

来自「最著名最快的分子模拟软件」· S 代码 · 共 2,353 行 · 第 1/5 页

	
	;# Repulsion 
	movapd xmm4, [rsi + rax*8 + 64]	;# Y1 F1 	
	movapd xmm3, [rsi + rbx*8 + 64]	;# Y2 F2 
	movapd xmm5, xmm4
	unpcklpd xmm4, xmm3	;# Y1 Y2 
	unpckhpd xmm5, xmm3	;# F1 F2 

	movapd xmm6, [rsi + rax*8 + 80]	;# G1 H1 	
	movapd xmm3, [rsi + rbx*8 + 80]	;# 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  xmm5, xmm1 ;# xmm5=eps*Fp 
	addpd  xmm5, xmm4 ;# xmm5=VV 

	mulpd  xmm5, [rsp + nb331nf_c12] ;# Vvdw12 
	addpd  xmm5, [rsp + nb331nf_Vvdwtot]
	movapd [rsp + nb331nf_Vvdwtot], xmm5

	;# Done with O interactions - now H1! 
	movapd xmm7, [rsp + nb331nf_rH1]
	mulpd xmm7, [rsp + nb331nf_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  rsi, [rbp + nb331nf_VFtab]
	movd eax, mm6
	psrlq mm6, 32
	movd ebx, mm6		;# indices in eax/ebx 
	lea   rax, [rax + rax*2] ;# idx *= 3 (total *=12 now) 	
	lea   rbx, [rbx + rbx*2]
	
	movapd xmm4, [rsi + rax*8]	;# Y1 F1 	
	movapd xmm3, [rsi + rbx*8]	;# Y2 F2 
	movapd xmm5, xmm4
	unpcklpd xmm4, xmm3	;# Y1 Y2 
	unpckhpd xmm5, xmm3	;# F1 F2 

	movapd xmm6, [rsi + rax*8 + 16]	;# G1 H1 	
	movapd xmm3, [rsi + rbx*8 + 16]	;# 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 	
	movapd xmm3, [rsp + nb331nf_qqH]
	mulpd  xmm5, xmm1 ;# xmm5=eps*Fp 
	addpd  xmm5, xmm4 ;# xmm5=VV 
	mulpd  xmm5, xmm3 ;# vcoul=qq*VV  
    ;# at this point mm5 contains vcoul
    ;# increment vcoul 
    addpd  xmm5, [rsp + nb331nf_vctot]
    movapd [rsp + nb331nf_vctot], xmm5 

	;# Done with H1, finally we do H2 interactions 
	movapd xmm7, [rsp + nb331nf_rH2]
	mulpd   xmm7, [rsp + nb331nf_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  rsi, [rbp + nb331nf_VFtab]
	movd eax, mm6
	psrlq mm6, 32
	movd ebx, mm6		;# indices in eax/ebx 
	lea   rax, [rax + rax*2] ;# idx *= 3 (total *=12 now)
	lea   rbx, [rbx + rbx*2]

	movapd xmm4, [rsi + rax*8]	;# Y1 F1 	
	movapd xmm3, [rsi + rbx*8]	;# Y2 F2 
	movapd xmm5, xmm4
	unpcklpd xmm4, xmm3	;# Y1 Y2 
	unpckhpd xmm5, xmm3	;# F1 F2 

	movapd xmm6, [rsi + rax*8 + 16]	;# G1 H1 	
	movapd xmm3, [rsi + rbx*8 + 16]	;# 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 	
	movapd xmm3, [rsp + nb331nf_qqH]
	mulpd  xmm5, xmm1 ;# xmm5=eps*Fp 
	addpd  xmm5, xmm4 ;# xmm5=VV 
	mulpd  xmm5, xmm3 ;# vcoul=qq*VV  
    ;# at this point mm5 contains vcoul 
    ;# increment vcoul 
    addpd  xmm5, [rsp + nb331nf_vctot]
    movapd [rsp + nb331nf_vctot], xmm5 
	
	;# should we do one more iteration? 
	sub dword ptr [rsp + nb331nf_innerk],  2
	jl    .nb331nf_checksingle
	jmp   .nb331nf_unroll_loop
.nb331nf_checksingle:	
	mov   edx, [rsp + nb331nf_innerk]
	and   edx, 1
	jnz   .nb331nf_dosingle
	jmp   .nb331nf_updateouterdata
.nb331nf_dosingle:
	mov   rdx, [rsp + nb331nf_innerjjnr]     ;# pointer to jjnr[k] 
	mov   eax, [rdx]

	mov rsi, [rbp + nb331nf_charge]    ;# base of charge[] 
	xorpd xmm3, xmm3	
	movlpd xmm3, [rsi + rax*8]
	movapd xmm4, xmm3	     
	mulpd  xmm3, [rsp + nb331nf_iqO]
	mulpd  xmm4, [rsp + nb331nf_iqH]

	movd  mm0, eax		;# use mmx registers as temp storage 
	movapd  [rsp + nb331nf_qqO], xmm3
	movapd  [rsp + nb331nf_qqH], xmm4	
	
	mov rsi, [rbp + nb331nf_type]
	mov eax, [rsi + rax*4]
	mov rsi, [rbp + nb331nf_vdwparam]
	shl eax, 1	
	mov edi, [rsp + nb331nf_ntia]
	add eax, edi

	movlpd xmm6, [rsi + rax*8]	;# c6a
	movhpd xmm6, [rsi + rax*8 + 8]	;# c6a c12a 
	xorpd xmm7, xmm7
	movapd xmm4, xmm6
	unpcklpd xmm4, xmm7
	unpckhpd xmm6, xmm7
	
	movd  eax, mm0
	movapd [rsp + nb331nf_c6], xmm4
	movapd [rsp + nb331nf_c12], xmm6
	
	mov rsi, [rbp + nb331nf_pos]       ;# base of pos[] 
	lea   rax, [rax + rax*2]     ;# replace jnr with j3 
	
	;# move coords to xmm0-xmm2 
	movlpd xmm0, [rsi + rax*8]
	movlpd xmm1, [rsi + rax*8 + 8]
	movlpd xmm2, [rsi + rax*8 + 16]

	;# move ixO-izO to xmm4-xmm6 
	movapd xmm4, [rsp + nb331nf_ixO]
	movapd xmm5, [rsp + nb331nf_iyO]
	movapd xmm6, [rsp + nb331nf_izO]

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

	;# square it 
	mulsd xmm4,xmm4
	mulsd xmm5,xmm5
	mulsd xmm6,xmm6
	addsd xmm4, xmm5
	addsd xmm4, xmm6
	movapd xmm7, xmm4
	;# rsqO in xmm7 

	;# move ixH1-izH1 to xmm4-xmm6 
	movapd xmm4, [rsp + nb331nf_ixH1]
	movapd xmm5, [rsp + nb331nf_iyH1]
	movapd xmm6, [rsp + nb331nf_izH1]

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

	;# 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, [rsp + nb331nf_ixH2]
	movapd xmm4, [rsp + nb331nf_iyH2]
	movapd xmm5, [rsp + nb331nf_izH2]

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

	;# 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, [rsp + nb331nf_three]
	mulsd   xmm2, xmm7	;# rsq*lu*lu 
	subsd   xmm4, xmm2	;# 30-rsq*lu*lu 
	mulsd   xmm4, xmm3	;# lu*(3-rsq*lu*lu) 
	mulsd   xmm4, [rsp + nb331nf_half] ;# iter1 ( new lu) 

	movapd xmm2, xmm7
	movapd xmm3, xmm4
	mulsd xmm4, xmm4	;# lu*lu 
	mulsd xmm2, xmm4	;# rsq*lu*lu 
	movapd xmm4, [rsp + nb331nf_three]
	subsd xmm4, xmm2	;# 3-rsq*lu*lu 
	mulsd xmm4, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulsd xmm4, [rsp + nb331nf_half] ;# rinv 
	movapd  [rsp + nb331nf_rinvO], xmm4	;# rinvO in xmm4 
	mulsd   xmm7, xmm4
	movapd  [rsp + nb331nf_rO], xmm7	;# r in xmm7 

	;# rsqH1 - seed in xmm2 
	cvtsd2ss xmm2, xmm6	
	rsqrtss xmm2, xmm2
	cvtss2sd xmm2, xmm2

	movapd  xmm3, xmm2
	mulsd   xmm2, xmm2
	movapd  xmm4, [rsp + nb331nf_three]
	mulsd   xmm2, xmm6	;# rsq*lu*lu 
	subsd   xmm4, xmm2	;# 30-rsq*lu*lu 
	mulsd   xmm4, xmm3	;# lu*(3-rsq*lu*lu) 
	mulsd   xmm4, [rsp + nb331nf_half] ;# iter1 ( new lu) 

	movapd xmm2, xmm6
	movapd xmm3, xmm4
	mulsd xmm4, xmm4	;# lu*lu 
	mulsd xmm2, xmm4	;# rsq*lu*lu 
	movapd xmm4, [rsp + nb331nf_three]
	subsd xmm4, xmm2	;# 3-rsq*lu*lu 
	mulsd xmm4, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulsd xmm4, [rsp + nb331nf_half] ;# rinv 
	movapd [rsp + nb331nf_rinvH1], xmm4	;# rinvH1 
	mulsd  xmm6, xmm4
	movapd [rsp + nb331nf_rH1], xmm6	;# rH1 
	
	;# rsqH2 - seed in xmm2 
	cvtsd2ss xmm2, xmm5	
	rsqrtss xmm2, xmm2
	cvtss2sd xmm2, xmm2

	movapd  xmm3, xmm2
	mulsd   xmm2, xmm2
	movapd  xmm4, [rsp + nb331nf_three]
	mulsd   xmm2, xmm5	;# rsq*lu*lu 
	subsd   xmm4, xmm2	;# 30-rsq*lu*lu 
	mulsd   xmm4, xmm3	;# lu*(3-rsq*lu*lu) 
	mulsd   xmm4, [rsp + nb331nf_half] ;# iter1 ( new lu) 

	movapd xmm2, xmm5
	movapd xmm3, xmm4
	mulsd xmm4, xmm4	;# lu*lu 
	mulsd xmm2, xmm4	;# rsq*lu*lu 
	movapd xmm4, [rsp + nb331nf_three]
	subsd xmm4, xmm2	;# 3-rsq*lu*lu 
	mulsd xmm4, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulsd xmm4, [rsp + nb331nf_half] ;# rinv 
	movapd [rsp + nb331nf_rinvH2], xmm4 ;# rinv 
	mulsd xmm5, xmm4
	movapd [rsp + nb331nf_rH2], xmm5 ;# r 

	;# do O interactions 
	movd mm0, eax	
	;# rO is still in xmm7 
	mulsd   xmm7, [rsp + nb331nf_tsc]
	cvttsd2si eax, xmm7	;# lu idx 
	cvtsi2sd xmm6, eax
	subsd xmm7, xmm6
	movapd xmm1, xmm7	;# xmm1=eps 
	movapd xmm2, xmm7
	mulsd  xmm2, xmm2	;# xmm2=eps2 
	 	
	shl eax, 2		
	mov  rsi, [rbp + nb331nf_VFtab]
	lea   rax, [rax + rax*2] ;# idx *= 3 (total *=12 now) 	
	
	movapd xmm4, [rsi + rax*8]	;# Y1 F1 
	xorpd xmm3, xmm3
	movapd xmm5, xmm4
	unpcklpd xmm4, xmm3	;# Y1 
	unpckhpd xmm5, xmm3	;# F1 

	movapd xmm6, [rsi + rax*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	;# F+Geps 
    addsd  xmm5, xmm7       ;# xmm5=Fp=F+Geps+Heps2        
    mulsd  xmm5, xmm1 ;# xmm5=eps*Fp 
    addsd  xmm5, xmm4 ;# xmm5=VV 
	
    movapd  xmm0, [rsp + nb331nf_qqO]
    mulsd  xmm5, xmm0 ;# vcoul=qq*VV  
	;# at this point xmm5 contains vcoul 
	;# increment vcoul - then we can get rid of mm5 
	;# update vctot 
	addsd  xmm5, [rsp + nb331nf_vctot]
	movlpd [rsp + nb331nf_vctot], xmm5 
	
	;# Dispersion 
	movapd xmm4, [rsi + rax*8 + 32]	;# Y1 F1 	
	xorpd xmm3, xmm3
	movapd xmm5, xmm4
	unpcklpd xmm4, xmm3	;# Y1 
	unpckhpd xmm5, xmm3	;# F1 

	movapd xmm6, [rsi + rax*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  xmm5, xmm1 ;# xmm5=eps*Fp 
	addsd  xmm5, xmm4 ;# xmm5=VV 

	mulsd  xmm5, [rsp + nb331nf_c6]	 ;# Vvdw6 
	
	addsd  xmm5, [rsp + nb331nf_Vvdwtot]
	movsd [rsp + nb331nf_Vvdwtot], xmm5
	
	;# Repulsion 
	movapd xmm4, [rsi + rax*8 + 64]	;# Y1 F1 	
	xorpd xmm3, xmm3
	movapd xmm5, xmm4
	unpcklpd xmm4, xmm3	;# Y1 
	unpckhpd xmm5, xmm3	;# F1 

	movapd xmm6, [rsi + rax*8 + 80]	;# 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  xmm5, xmm1 ;# xmm5=eps*Fp 
	addsd  xmm5, xmm4 ;# xmm5=VV 

	mulsd  xmm5, [rsp + nb331nf_c12] ;# Vvdw12 

	addsd  xmm5, [rsp + nb331nf_Vvdwtot]
	movsd [rsp + nb331nf_Vvdwtot], xmm5

	;# Done with O interactions - now H1! 
	movapd xmm7, [rsp + nb331nf_rH1]
	mulpd xmm7, [rsp + nb331nf_tsc]
	cvttsd2si eax, xmm7	;# mm6 = lu idx 
	cvtsi2sd xmm6, eax
	subpd xmm7, xmm6
	movapd xmm1, xmm7	;# xmm1=eps 
	movapd xmm2, xmm1	
	mulpd  xmm2, xmm2	;# xmm2=eps2 
	
	shl eax, 2		;# idx *= 4 
	mov  rsi, [rbp + nb331nf_VFtab]
	lea   rax, [rax + rax*2] ;# idx *= 3 (total *=12 now)	

	movapd xmm4, [rsi + rax*8]	;# Y1 F1 	
	xorpd xmm3, xmm3
	movapd xmm5, xmm4
	unpcklpd xmm4, xmm3	;# Y1 
	unpckhpd xmm5, xmm3	;# F1 

	movapd xmm6, [rsi + rax*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 	
	movapd xmm3, [rsp + nb331nf_qqH]
	mulsd  xmm5, xmm1 ;# xmm5=eps*Fp 
	addsd  xmm5, xmm4 ;# xmm5=VV 
	mulsd  xmm5, xmm3 ;# vcoul=qq*VV  
    ;# at this point mm5 contains vcoul 
    ;# increment vcoul 
    addsd  xmm5, [rsp + nb331nf_vctot]
    movlpd [rsp + nb331nf_vctot], xmm5 

	;# Done with H1, finally we do H2 interactions 
	movapd xmm7, [rsp + nb331nf_rH2]
	mulsd   xmm7, [rsp + nb331nf_tsc]
	cvttsd2si eax, xmm7	;# mm6 = lu idx 
	cvtsi2sd xmm6, eax
	subsd xmm7, xmm6
	movapd xmm1, xmm7	;# xmm1=eps 
	movapd xmm2, xmm1	
	mulsd  xmm2, xmm2	;# xmm2=eps2 
	
	shl eax, 2		;# idx *= 4 
	mov  rsi, [rbp + nb331nf_VFtab]
	lea   rax, [rax + rax*2] ;# idx *= 3 (total *=12 now)	

	movapd xmm4, [rsi + rax*8]	;# Y1 F1 	
	xorpd xmm3, xmm3
	movapd xmm5, xmm4
	unpcklpd xmm4, xmm3	;# Y1 
	unpckhpd xmm5, xmm3	;# F1 

	movapd xmm6, [rsi + rax*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 	
	movapd xmm3, [rsp + nb331nf_qqH]
	mulsd  xmm5, xmm1 ;# xmm5=eps*Fp 
	addsd  xmm5, xmm4 ;# xmm5=VV 
	mulsd  xmm5, xmm3 ;# vcoul=qq*VV  
    	;# at this point mm5 contains vcoul 
    	;# increment vcoul 
    	addsd  xmm5, [rsp + nb331nf_vctot]
    	movlpd [rsp + nb331nf_vctot], xmm5 

.nb331nf_updateouterdata:
	;# get n from stack
	mov esi, [rsp + nb331nf_n]
        ;# get group index for i particle 
        mov   rdx, [rbp + nb331nf_gid]      	;# base of gid[]
        mov   edx, [rdx + rsi*4]		;# ggid=gid[n]

	;