nb_kernel313_x86_64_sse2.intel_syntax.s

来自「最著名最快的分子模拟软件」· S 代码 · 共 2,374 行 · 第 1/5 页

	subsd xmm4, [rsp + nb313_ixM]
	subsd xmm3, [rsp + nb313_iyM]
	subsd xmm2, [rsp + nb313_izM]

	;# store dr 
	movapd [rsp + nb313_dxM], xmm4
	movapd [rsp + nb313_dyM], xmm3
	movapd [rsp + nb313_dzM], xmm2

	;# square it 
	mulpd xmm2,xmm2
	mulpd xmm3,xmm3
	mulpd xmm4,xmm4
	addpd xmm4, xmm3
	addpd xmm4, xmm2	
	;# rsqM in xmm4, rsqH2 in xmm5, rsqH1 in xmm6, rsqO in xmm7 

	;# 1/x for O - rsqO is in xmm7
	cvtsd2ss xmm2, xmm7
	movsd   xmm3, xmm7
	rcpps    xmm2, xmm2
	cvtss2sd xmm2, xmm2
	movsd   xmm1, [rsp + nb313_two]
	movsd   xmm0, xmm1
	mulsd   xmm7, xmm2
	subsd   xmm1, xmm7
	mulsd   xmm2, xmm1 ;# iter1 
	mulsd   xmm3, xmm2
	subsd   xmm0, xmm3
	mulsd   xmm0, xmm2 ;# xmm0=rinvsq
	movsd  [rsp + nb313_rinvsqO], xmm0

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

	movapd  xmm3, xmm2
	mulsd   xmm2, xmm2
	movapd  xmm0, [rsp + nb313_three]
	mulsd   xmm2, xmm6	;# rsq*lu*lu 
	subsd   xmm0, xmm2	;# 30-rsq*lu*lu 
	mulsd   xmm0, xmm3	;# lu*(3-rsq*lu*lu) 
	mulsd   xmm0, [rsp + nb313_half] ;# iter1 ( new lu) 

	movapd xmm2, xmm6
	movapd xmm3, xmm0
	mulsd xmm0, xmm0	;# lu*lu 
	mulsd xmm2, xmm0	;# rsq*lu*lu 
	movapd xmm0, [rsp + nb313_three]
	subsd xmm0, xmm2	;# 3-rsq*lu*lu 
	mulsd xmm0, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulsd xmm0, [rsp + nb313_half] ;# rinv 
	movapd [rsp + nb313_rinvH1], xmm0	;# rinvH1 
	mulsd  xmm6, xmm0
	movapd [rsp + nb313_rH1], xmm6	;# rH1 
	
	;# rsqH2 - seed in xmm2 
	cvtsd2ss xmm2, xmm5	
	rsqrtss xmm2, xmm2
	cvtss2sd xmm2, xmm2

	movapd  xmm3, xmm2
	mulsd   xmm2, xmm2
	movapd  xmm0, [rsp + nb313_three]
	mulsd   xmm2, xmm5	;# rsq*lu*lu 
	subsd   xmm0, xmm2	;# 30-rsq*lu*lu 
	mulsd   xmm0, xmm3	;# lu*(3-rsq*lu*lu) 
	mulsd   xmm0, [rsp + nb313_half] ;# iter1 ( new lu) 

	movapd xmm2, xmm5
	movapd xmm3, xmm0
	mulsd xmm0, xmm0	;# lu*lu 
	mulsd xmm2, xmm0	;# rsq*lu*lu 
	movapd xmm0, [rsp + nb313_three]
	subsd xmm0, xmm2	;# 3-rsq*lu*lu 
	mulsd xmm0, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulsd xmm0, [rsp + nb313_half] ;# rinv 
	movapd [rsp + nb313_rinvH2], xmm0 ;# rinv 
	mulsd xmm5, xmm0
	movapd [rsp + nb313_rH2], xmm5 ;# r 

	;# rsqM - seed in xmm2 
	cvtsd2ss xmm2, xmm4	
	rsqrtss xmm2, xmm2
	cvtss2sd xmm2, xmm2

	movapd  xmm3, xmm2
	mulsd   xmm2, xmm2
	movapd  xmm0, [rsp + nb313_three]
	mulsd   xmm2, xmm4	;# rsq*lu*lu 
	subsd   xmm0, xmm2	;# 30-rsq*lu*lu 
	mulsd   xmm0, xmm3	;# lu*(3-rsq*lu*lu) 
	mulsd   xmm0, [rsp + nb313_half] ;# iter1 ( new lu) 

	movapd xmm2, xmm4
	movapd xmm3, xmm0
	mulsd xmm0, xmm0	;# lu*lu 
	mulsd xmm2, xmm0	;# rsq*lu*lu 
	movapd xmm0, [rsp + nb313_three]
	subsd xmm0, xmm2	;# 3-rsq*lu*lu 
	mulsd xmm0, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulsd xmm0, [rsp + nb313_half] ;# rinv 
	movapd [rsp + nb313_rinvM], xmm0 ;# rinv 
	mulsd xmm4, xmm0
	movapd [rsp + nb313_rM], xmm4 ;# r 

	;# do O interactions 
	movapd  xmm0, [rsp + nb313_rinvsqO]
	movapd  xmm1, xmm0	
	mulsd   xmm1, xmm1 ;# rinv4
	mulsd   xmm1, xmm0 ;#rinvsix
	movapd  xmm2, xmm1
	mulsd	xmm2, xmm2 ;# rinvtwelve
	mulsd  xmm1, [rsp + nb313_c6]
	mulsd  xmm2, [rsp + nb313_c12]
	movapd xmm3, xmm2
	subsd  xmm3, xmm1	;# Vvdw=Vvdw12-Vvdw6 		
	addsd  xmm3, [rsp + nb313_Vvdwtot]
	mulsd  xmm1, [rsp + nb313_six]
	mulsd  xmm2, [rsp + nb313_twelve]
	subsd  xmm2, xmm1
	mulsd  xmm2, xmm0
	movapd xmm4, xmm2 ;# total fsO 
	movsd [rsp + nb313_Vvdwtot], xmm3

	movapd xmm0, [rsp + nb313_dxO]
	movapd xmm1, [rsp + nb313_dyO]
	movapd xmm2, [rsp + nb313_dzO]
	mulsd  xmm0, xmm4
	mulsd  xmm1, xmm4
	mulsd  xmm2, xmm4
	;# tx in xmm0-xmm2 

	;# update O forces 
	movapd xmm3, [rsp + nb313_fixO]
	movapd xmm4, [rsp + nb313_fiyO]
	movapd xmm7, [rsp + nb313_fizO]
	addsd  xmm3, xmm0
	addsd  xmm4, xmm1
	addsd  xmm7, xmm2
	movlpd [rsp + nb313_fixO], xmm3
	movlpd [rsp + nb313_fiyO], xmm4
	movlpd [rsp + nb313_fizO], xmm7
	;# update j forces with water O 
	movlpd [rsp + nb313_fjx], xmm0
	movlpd [rsp + nb313_fjy], xmm1
	movlpd [rsp + nb313_fjz], xmm2

	;# Done with O interactions - now H1! 
	movapd xmm7, [rsp + nb313_rH1]
	mulpd xmm7, [rsp + nb313_tsc]
	cvttsd2si r8d, xmm7	;# mm6 = lu idx 
	cvtsi2sd xmm6, r8d
	subpd xmm7, xmm6
	movapd xmm1, xmm7	;# xmm1=eps 
	movapd xmm2, xmm1	
	mulpd  xmm2, xmm2	;# xmm2=eps2 
	
	shl r8d, 2		;# idx *= 4 
	mov  rsi, [rbp + nb313_VFtab]

	movapd xmm4, [rsi + r8*8]	;# Y1 F1 	
	xorpd xmm3, xmm3
	movapd xmm5, xmm4
	unpcklpd xmm4, xmm3	;# Y1 
	unpckhpd xmm5, xmm3	;# F1 

	movapd xmm6, [rsi + r8*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, [rsp + nb313_two]	;# two*Heps2 
	movapd xmm3, [rsp + nb313_qqH]
	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 
    ;# at this point mm5 contains vcoul and xmm3 fijC 
    ;# increment vcoul 
	xorpd  xmm4, xmm4
    addsd  xmm5, [rsp + nb313_vctot]
	mulsd  xmm3, [rsp + nb313_rinvH1]
    movlpd [rsp + nb313_vctot], xmm5 
	mulsd  xmm3, [rsp + nb313_tsc]
	subsd xmm4, xmm3

	movapd xmm0, [rsp + nb313_dxH1]
	movapd xmm1, [rsp + nb313_dyH1]
	movapd xmm2, [rsp + nb313_dzH1]
	mulsd  xmm0, xmm4
	mulsd  xmm1, xmm4
	mulsd  xmm2, xmm4

	;# update H1 forces 
	movapd xmm3, [rsp + nb313_fixH1]
	movapd xmm4, [rsp + nb313_fiyH1]
	movapd xmm7, [rsp + nb313_fizH1]
	addsd  xmm3, xmm0
	addsd  xmm4, xmm1
	addsd  xmm7, xmm2
	movlpd [rsp + nb313_fixH1], xmm3
	movlpd [rsp + nb313_fiyH1], xmm4
	movlpd [rsp + nb313_fizH1], xmm7
	;# update j forces with water H1 
	addsd  xmm0, [rsp + nb313_fjx]
	addsd  xmm1, [rsp + nb313_fjy]
	addsd  xmm2, [rsp + nb313_fjz]
	movlpd [rsp + nb313_fjx], xmm0
	movlpd [rsp + nb313_fjy], xmm1
	movlpd [rsp + nb313_fjz], xmm2

	;#  H2 interactions 
	movapd xmm7, [rsp + nb313_rH2]
	mulsd   xmm7, [rsp + nb313_tsc]
	cvttsd2si r8d, xmm7	;# mm6 = lu idx 
	cvtsi2sd xmm6, r8d
	subsd xmm7, xmm6
	movapd xmm1, xmm7	;# xmm1=eps 
	movapd xmm2, xmm1	
	mulsd  xmm2, xmm2	;# xmm2=eps2 
	
	shl r8d, 2		;# idx *= 4 
	mov  rsi, [rbp + nb313_VFtab]

	movapd xmm4, [rsi + r8*8]	;# Y1 F1 	
	xorpd xmm3, xmm3
	movapd xmm5, xmm4
	unpcklpd xmm4, xmm3	;# Y1 
	unpckhpd xmm5, xmm3	;# F1 

	movapd xmm6, [rsi + r8*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, [rsp + nb313_two]	;# two*Heps2 
	movapd xmm3, [rsp + nb313_qqH]
	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 
    	;# at this point mm5 contains vcoul and xmm3 fijC 
    	;# increment vcoul 
	xorpd  xmm4, xmm4
    	addsd  xmm5, [rsp + nb313_vctot]
	mulsd  xmm3, [rsp + nb313_rinvH2]
    	movlpd [rsp + nb313_vctot], xmm5 
	mulsd  xmm3, [rsp + nb313_tsc]
	subsd  xmm4, xmm3

	movapd xmm0, [rsp + nb313_dxH2]
	movapd xmm1, [rsp + nb313_dyH2]
	movapd xmm2, [rsp + nb313_dzH2]
	mulsd  xmm0, xmm4
	mulsd  xmm1, xmm4
	mulsd  xmm2, xmm4
	
	;# update H2 forces 
	movapd xmm3, [rsp + nb313_fixH2]
	movapd xmm4, [rsp + nb313_fiyH2]
	movapd xmm7, [rsp + nb313_fizH2]
	addsd  xmm3, xmm0
	addsd  xmm4, xmm1
	addsd  xmm7, xmm2
	movlpd [rsp + nb313_fixH2], xmm3
	movlpd [rsp + nb313_fiyH2], xmm4
	movlpd [rsp + nb313_fizH2], xmm7
	;# update j forces with water H1 
	addsd  xmm0, [rsp + nb313_fjx]
	addsd  xmm1, [rsp + nb313_fjy]
	addsd  xmm2, [rsp + nb313_fjz]
	movlpd [rsp + nb313_fjx], xmm0
	movlpd [rsp + nb313_fjy], xmm1
	movlpd [rsp + nb313_fjz], xmm2

	;# M interactions 
	movapd xmm7, [rsp + nb313_rM]
	mulsd   xmm7, [rsp + nb313_tsc]
	cvttsd2si r8d, xmm7	;# mm6 = lu idx 
	cvtsi2sd xmm6, r8d
	subsd xmm7, xmm6
	movapd xmm1, xmm7	;# xmm1=eps 
	movapd xmm2, xmm1	
	mulsd  xmm2, xmm2	;# xmm2=eps2 
	
	shl r8d, 2		;# idx *= 4 
	mov  rsi, [rbp + nb313_VFtab]

	movapd xmm4, [rsi + r8*8]	;# Y1 F1 	
	xorpd xmm3, xmm3
	movapd xmm5, xmm4
	unpcklpd xmm4, xmm3	;# Y1 
	unpckhpd xmm5, xmm3	;# F1 

	movapd xmm6, [rsi + r8*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, [rsp + nb313_two]	;# two*Heps2 
	movapd xmm3, [rsp + nb313_qqM]
	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 
    	;# at this point mm5 contains vcoul and xmm3 fijC 
    	;# increment vcoul 
	xorpd  xmm4, xmm4
    	addsd  xmm5, [rsp + nb313_vctot]
	mulsd  xmm3, [rsp + nb313_rinvM]
    	movlpd [rsp + nb313_vctot], xmm5 
	mulsd  xmm3, [rsp + nb313_tsc]
	subsd  xmm4, xmm3

	movapd xmm0, [rsp + nb313_dxM]
	movapd xmm1, [rsp + nb313_dyM]
	movapd xmm2, [rsp + nb313_dzM]
	mulsd  xmm0, xmm4
	mulsd  xmm1, xmm4
	mulsd  xmm2, xmm4
	
	;# update M forces 
	movapd xmm3, [rsp + nb313_fixM]
	movapd xmm4, [rsp + nb313_fiyM]
	movapd xmm7, [rsp + nb313_fizM]
	addsd  xmm3, xmm0
	addsd  xmm4, xmm1
	addsd  xmm7, xmm2
	movlpd [rsp + nb313_fixM], xmm3
	movlpd [rsp + nb313_fiyM], xmm4
	movlpd [rsp + nb313_fizM], xmm7

	mov rdi, [rbp + nb313_faction]
	;# update j forces 
	addsd  xmm0, [rsp + nb313_fjx]
	addsd  xmm1, [rsp + nb313_fjy]
	addsd  xmm2, [rsp + nb313_fjz]

	;# the fj's - start by accumulating forces from memory 
	movlpd xmm3, [rdi + rax*8]
	movlpd xmm4, [rdi + rax*8 + 8]
	movlpd xmm5, [rdi + rax*8 + 16]
	addsd xmm3, xmm0
	addsd xmm4, xmm1
	addsd xmm5, xmm2
	movlpd [rdi + rax*8], xmm3
	movlpd [rdi + rax*8 + 8], xmm4
	movlpd [rdi + rax*8 + 16], xmm5

.nb313_updateouterdata:
	mov   ecx, [rsp + nb313_ii3]
	mov   rdi, [rbp + nb313_faction]
	mov   rsi, [rbp + nb313_fshift]
	mov   edx, [rsp + nb313_is3]

	;# accumulate  Oi forces in xmm0, xmm1, xmm2 
	movapd xmm0, [rsp + nb313_fixO]
	movapd xmm1, [rsp + nb313_fiyO]
	movapd xmm2, [rsp + nb313_fizO]

	movhlps xmm3, xmm0
	movhlps xmm4, xmm1
	movhlps xmm5, xmm2
	addsd  xmm0, xmm3
	addsd  xmm1, xmm4
	addsd  xmm2, xmm5 ;# sum is in low xmm0-xmm2 

	movapd xmm3, xmm0	
	movapd xmm4, xmm1	
	movapd xmm5, xmm2	

	;# increment i force 
	movsd  xmm3, [rdi + rcx*8]
	movsd  xmm4, [rdi + rcx*8 + 8]
	movsd  xmm5, [rdi + rcx*8 + 16]
	subsd  xmm3, xmm0
	subsd  xmm4, xmm1
	subsd  xmm5, xmm2
	movsd  [rdi + rcx*8],     xmm3
	movsd  [rdi + rcx*8 + 8], xmm4
	movsd  [rdi + rcx*8 + 16], xmm5

	;# accumulate force in xmm6/xmm7 for fshift 
	movapd xmm6, xmm0
	movsd xmm7, xmm2
	unpcklpd xmm6, xmm1

	;# accumulate H1i forces in xmm0, xmm1, xmm2 
	movapd xmm0, [rsp + nb313_fixH1]
	movapd xmm1, [rsp + nb313_fiyH1]
	movapd xmm2, [rsp + nb313_fizH1]

	movhlps xmm3, xmm0
	movhlps xmm4, xmm1
	movhlps xmm5, xmm2
	addsd  xmm0, xmm3
	addsd  xmm1, xmm4
	addsd  xmm2, xmm5 ;# sum is in low xmm0-xmm2 

	;# increment i force 
	movsd  xmm3, [rdi + rcx*8 + 24]
	movsd  xmm4, [rdi + rcx*8 + 32]
	movsd  xmm5, [rdi + rcx*8 + 40]
	subsd  xmm3, xmm0
	subsd  xmm4, xmm1
	subsd  xmm5, xmm2
	movsd  [rdi + rcx*8 + 24], xmm3
	movsd  [rdi + rcx*8 + 32], xmm4
	movsd  [rdi + rcx*8 + 40], xmm5

	;# accumulate force in xmm6/xmm7 for fshift 
	addsd xmm7, xmm2
	unpcklpd xmm0, xmm1
	addpd xmm6, xmm0

	;# accumulate H2i forces in xmm0, xmm1, xmm2 
	movapd xmm0, [rsp + nb313_fixH2]
	movapd xmm1, [rsp + nb313_fiyH2]
	movapd xmm2, [rsp + nb313_fizH2]

	movhlps xmm3, xmm0
	movhlps xmm4, xmm1
	movhlps xmm5, xmm2
	addsd  xmm0, xmm3
	addsd  xmm1, xmm4
	addsd  xmm2, xmm5 ;# sum is in low xmm0-xmm2 

	movapd xmm3, xmm0	
	movapd xmm4, xmm1	
	movapd xmm5, xmm2	

	;# increment i force 
	movsd  xmm3, [rdi + rcx*8 + 48]
	movsd  xmm4, [rdi + rcx*8 + 56]
	movsd  xmm5, [rdi + rcx*8 + 64]
	subsd  xmm3, xmm0
	subsd  xmm4, xmm1
	subsd  xmm5, xmm2
	movsd  [rdi + rcx*8 + 48], xmm3
	movsd  [rdi + rcx*8 + 56], xmm4
	movsd  [rdi + rcx*8 + 64], xmm5

	;# accumulate force in xmm6/xmm7 for fshift 
	addsd xmm7, xmm2
	unpcklpd xmm0, xmm1
	addpd xmm6, xmm0

	;# accumulate Mi forces in xmm0, xmm1, xmm2 
	movapd xmm0, [rsp + nb313_fixM]
	movapd xmm1, [rsp + nb313_fiyM]
	movapd xmm2, [rsp + nb313_fizM]