nb_kernel333_x86_64_sse.intel_syntax.s

来自「最著名最快的分子模拟软件」· S 代码 · 共 2,385 行 · 第 1/5 页

	
	movd  eax, mm0		
	movd  ebx, mm1
	movd  ecx, mm2
	movd  edx, mm3

	movaps [rsp + nb333nf_c6], xmm4
	movaps [rsp + nb333nf_c12], xmm6

	mov rsi, [rbp + nb333nf_pos]   	;# base of pos[] 

	lea   rax, [rax + rax*2] 	;# replace jnr with j3 
	lea   rbx, [rbx + rbx*2]	
	lea   rcx, [rcx + rcx*2] 	;# replace jnr with j3 
	lea   rdx, [rdx + rdx*2]	

	;# move four coordinates to xmm0-xmm2 	
	movlps xmm4, [rsi + rax*4]
	movlps xmm5, [rsi + rcx*4]
	movss xmm2, [rsi + rax*4 + 8]
	movss xmm6, [rsi + rcx*4 + 8]

	movhps xmm4, [rsi + rbx*4]
	movhps xmm5, [rsi + rdx*4]

	movss xmm0, [rsi + rbx*4 + 8]
	movss xmm1, [rsi + rdx*4 + 8]

	shufps xmm2, xmm0, 0
	shufps xmm6, xmm1, 0
	
	movaps xmm0, xmm4
	movaps xmm1, xmm4

	shufps xmm2, xmm6, 136  ;# 10001000
	
	shufps xmm0, xmm5, 136  ;# 10001000
	shufps xmm1, xmm5, 221  ;# 11011101		

	;# move ixO-izO to xmm4-xmm6 
	movaps xmm4, [rsp + nb333nf_ixO]
	movaps xmm5, [rsp + nb333nf_iyO]
	movaps xmm6, [rsp + nb333nf_izO]

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

	;# square it 
	mulps xmm4,xmm4
	mulps xmm5,xmm5
	mulps xmm6,xmm6
	addps xmm4, xmm5
	addps xmm4, xmm6
	movaps xmm7, xmm4
	;# rsqO in xmm7

	;# move ixH1-izH1 to xmm4-xmm6 
	movaps xmm4, [rsp + nb333nf_ixH1]
	movaps xmm5, [rsp + nb333nf_iyH1]
	movaps xmm6, [rsp + nb333nf_izH1]

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

	;# square it 
	mulps xmm4,xmm4
	mulps xmm5,xmm5
	mulps xmm6,xmm6
	addps xmm6, xmm5
	addps xmm6, xmm4
	;# rsqH1 in xmm6 

	;# move ixH2-izH2 to xmm3-xmm5  
	movaps xmm3, [rsp + nb333nf_ixH2]
	movaps xmm4, [rsp + nb333nf_iyH2]
	movaps xmm5, [rsp + nb333nf_izH2]

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

	;# square it 
	mulps xmm3,xmm3
	mulps xmm4,xmm4
	mulps xmm5,xmm5
	addps xmm5, xmm4
	addps xmm5, xmm3
	
	;# move ixM-izM to xmm2-xmm4  
	movaps xmm3, [rsp + nb333nf_iyM]
	movaps xmm4, [rsp + nb333nf_izM]
	subps  xmm3, xmm1
	subps  xmm4, xmm2
	movaps xmm2, [rsp + nb333nf_ixM]
	subps  xmm2, xmm0	

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

	;# rsqH1 - seed in xmm2 
	rsqrtps xmm2, xmm6
	movaps  xmm3, xmm2
	mulps   xmm2, xmm2
	movaps  xmm0, [rsp + nb333nf_three]
	mulps   xmm2, xmm6	;# rsq*lu*lu 
	subps   xmm0, xmm2	;# 30-rsq*lu*lu 
	mulps   xmm0, xmm3	;# lu*(3-rsq*lu*lu) 
	mulps   xmm0, [rsp + nb333nf_half]
	movaps  [rsp + nb333nf_rinvH1], xmm0	;# rinvH1  
	mulps   xmm6, xmm0
	movaps  [rsp + nb333nf_rH1], xmm6

	;# rsqH2 - seed to xmm2 
	rsqrtps xmm2, xmm5
	movaps  xmm3, xmm2
	mulps   xmm2, xmm2
	movaps  xmm0, [rsp + nb333nf_three]
	mulps   xmm2, xmm5	;# rsq*lu*lu 
	subps   xmm0, xmm2	;# 30-rsq*lu*lu 
	mulps   xmm0, xmm3	;# lu*(3-rsq*lu*lu) 
	mulps   xmm0, [rsp + nb333nf_half]
	movaps  [rsp + nb333nf_rinvH2], xmm0	;# rinvH2 
	mulps   xmm5, xmm0
	movaps  [rsp + nb333nf_rH2], xmm5

	;# rsqM - seed to xmm2 
	rsqrtps xmm2, xmm4
	movaps  xmm3, xmm2
	mulps   xmm2, xmm2
	movaps  xmm0, [rsp + nb333nf_three]
	mulps   xmm2, xmm4	;# rsq*lu*lu 
	subps   xmm0, xmm2	;# 30-rsq*lu*lu 
	mulps   xmm0, xmm3	;# lu*(3-rsq*lu*lu) 
	mulps   xmm0, [rsp + nb333nf_half]
	movaps  [rsp + nb333nf_rinvM], xmm0	;# rinvM 
	mulps   xmm4, xmm0
	movaps  [rsp + nb333nf_rM], xmm4	
	
	;# Do the O LJ table interaction directly.
	rsqrtps xmm2, xmm7
	movaps  xmm3, xmm2
	mulps   xmm2, xmm2
	movaps  xmm0, [rsp + nb333nf_three]
	mulps   xmm2, xmm7	;# rsq*lu*lu 
	subps   xmm0, xmm2	;# 30-rsq*lu*lu 
	mulps   xmm0, xmm3	;# lu*(3-rsq*lu*lu) 
	mulps   xmm0, [rsp + nb333nf_half] ;# rinv
	
	movaps xmm1, xmm0
	mulps  xmm1, xmm7	;# xmm1=r
	mulps  xmm1, [rsp + nb333nf_tsc] ;# r*tabscale
	
	movhlps xmm2, xmm1
	cvttps2pi mm6, xmm1
	cvttps2pi mm7, xmm2 	;# mm6/mm7 contain lu indices 
	cvtpi2ps xmm3, mm6
	cvtpi2ps xmm2, mm7
	movlhps  xmm3, xmm2
	subps    xmm1, xmm3	;# xmm1=eps 
	movaps xmm2, xmm1
	mulps  xmm2, xmm2   	;# xmm2=eps2 
	pslld   mm6, 2
	pslld   mm7, 2
	
	movd mm0, eax
	movd mm1, ebx
	movd mm2, ecx
	movd mm3, edx

	mov  rsi, [rbp + nb333nf_VFtab]
	movd eax, mm6
	psrlq mm6, 32
	movd ecx, mm7
	psrlq mm7, 32
	movd ebx, mm6
	movd edx, mm7

	lea   rax, [rax + rax*2]
	lea   rbx, [rbx + rbx*2]
	lea   rcx, [rcx + rcx*2]
	lea   rdx, [rdx + rdx*2]
	
	;# load dispersion table data into xmm4-xmm7
	movlps xmm5, [rsi + rax*4 + 16]
	movlps xmm7, [rsi + rcx*4 + 16]
	movhps xmm5, [rsi + rbx*4 + 16]
	movhps xmm7, [rsi + rdx*4 + 16] ;# got half coulomb table 

	movaps xmm4, xmm5
	shufps xmm4, xmm7, 136  ;# 10001000
	shufps xmm5, xmm7, 221  ;# 11011101

	movlps xmm7, [rsi + rax*4 + 24]
	movlps xmm3, [rsi + rcx*4 + 24]
	movhps xmm7, [rsi + rbx*4 + 24]
	movhps xmm3, [rsi + rdx*4 + 24] ;# other half of coulomb table  
	movaps xmm6, xmm7
	shufps xmm6, xmm3, 136  ;# 10001000
	shufps xmm7, xmm3, 221  ;# 11011101
	;# dispersion table YFGH ready in xmm4-xmm7
	mulps  xmm6, xmm1   	;# xmm6=Geps 
	mulps  xmm7, xmm2   	;# xmm7=Heps2 
	addps  xmm5, xmm6
	addps  xmm5, xmm7   	;# xmm5=Fp 
	mulps  xmm5, xmm1 ;# xmm5=eps*Fp 
	addps  xmm5, xmm4 ;# xmm5=VV 

	movaps xmm4, [rsp + nb333nf_c6]
	mulps  xmm5, xmm4	;# Vvdw6 

	;# Update Vvdwtot directly	
	addps  xmm5, [rsp + nb333nf_Vvdwtot]
	movaps [rsp + nb333nf_Vvdwtot], xmm5

	;# load repulsion table data into xmm4-xmm7
	movlps xmm5, [rsi + rax*4 + 32]
	movlps xmm7, [rsi + rcx*4 + 32]
	movhps xmm5, [rsi + rbx*4 + 32]
	movhps xmm7, [rsi + rdx*4 + 32] ;# got half coulomb table 

	movaps xmm4, xmm5
	shufps xmm4, xmm7, 136  ;# 10001000
	shufps xmm5, xmm7, 221  ;# 11011101

	movlps xmm7, [rsi + rax*4 + 40]
	movlps xmm3, [rsi + rcx*4 + 40]
	movhps xmm7, [rsi + rbx*4 + 40]
	movhps xmm3, [rsi + rdx*4 + 40] ;# other half of coulomb table  
	movaps xmm6, xmm7
	shufps xmm6, xmm3, 136  ;# 10001000
	shufps xmm7, xmm3, 221  ;# 11011101
	;# repulsion table YFGH ready in xmm4-xmm7
	
	mulps  xmm6, xmm1   	;# xmm6=Geps 
	mulps  xmm7, xmm2   	;# xmm7=Heps2 
	addps  xmm5, xmm6
	addps  xmm5, xmm7   	;# xmm5=Fp 
	mulps  xmm5, xmm1 ;# xmm5=eps*Fp 
	addps  xmm5, xmm4 ;# xmm5=VV 
 
	movaps xmm4, [rsp + nb333nf_c12]
	mulps  xmm5, xmm4 ;# Vvdw12 
	addps  xmm5, [rsp + nb333nf_Vvdwtot]
	movaps [rsp + nb333nf_Vvdwtot], xmm5

	;# Do H1 interaction
	mov  rsi, [rbp + nb333nf_VFtab]
	
	movaps xmm7, [rsp + nb333nf_rH1]
	mulps   xmm7, [rsp + nb333nf_tsc]
	movhlps xmm4, xmm7
	cvttps2pi mm6, xmm7
	cvttps2pi mm7, xmm4	;# mm6/mm7 contain lu indices 
	
	cvtpi2ps xmm3, mm6
	cvtpi2ps xmm4, mm7
	movlhps xmm3, xmm4
	
	subps xmm7, xmm3
	movaps xmm1, xmm7	;# xmm1=eps 
	movaps xmm2, xmm1
	mulps  xmm2, xmm2	;# xmm2=eps2 
	pslld mm6, 2
	pslld mm7, 2
	
	movd eax, mm6
	psrlq mm6, 32
	movd ecx, mm7
	psrlq mm7, 32
	movd ebx, mm6
	movd edx, mm7

	lea   rax, [rax + rax*2]
	lea   rbx, [rbx + rbx*2]
	lea   rcx, [rcx + rcx*2]
	lea   rdx, [rdx + rdx*2]

	movlps xmm5, [rsi + rax*4]
	movlps xmm7, [rsi + rcx*4]
	movhps xmm5, [rsi + rbx*4]
	movhps xmm7, [rsi + rdx*4] ;# got half coulomb table 

	movaps xmm4, xmm5
	shufps xmm4, xmm7, 136  ;# 10001000
	shufps xmm5, xmm7, 221  ;# 11011101

	movlps xmm7, [rsi + rax*4 + 8]
	movlps xmm3, [rsi + rcx*4 + 8]
	movhps xmm7, [rsi + rbx*4 + 8]
	movhps xmm3, [rsi + rdx*4 + 8] ;# other half of coulomb table  
	movaps xmm6, xmm7
	shufps xmm6, xmm3, 136  ;# 10001000
	shufps xmm7, xmm3, 221  ;# 11011101
	;# coulomb table ready, in xmm4-xmm7      
        
	mulps  xmm6, xmm1   	;# xmm6=Geps 
	mulps  xmm7, xmm2   	;# xmm7=Heps2 
	addps  xmm5, xmm6
	addps  xmm5, xmm7   	;# xmm5=Fp        
	movaps xmm0, [rsp + nb333nf_qqH]
	mulps  xmm5, xmm1 ;# xmm5=eps*Fp 
	addps  xmm5, xmm4 ;# xmm5=VV 
	mulps  xmm5, xmm0 ;# vcoul=qq*VV  
	;# at this point mm5 contains vcoul 
	;# increment vcoul 
	addps  xmm5, [rsp + nb333nf_vctot]
	movaps [rsp + nb333nf_vctot], xmm5 

	;# Done with H1, do H2 interactions 
	movaps xmm7, [rsp + nb333nf_rH2]
	mulps   xmm7, [rsp + nb333nf_tsc]
	movhlps xmm4, xmm7
	cvttps2pi mm6, xmm7
	cvttps2pi mm7, xmm4	;# mm6/mm7 contain lu indices 
	
	cvtpi2ps xmm3, mm6
	cvtpi2ps xmm4, mm7
	movlhps xmm3, xmm4
	
	subps xmm7, xmm3
	movaps xmm1, xmm7	;# xmm1=eps 
	movaps xmm2, xmm1
	mulps  xmm2, xmm2	;# xmm2=eps2 
	pslld mm6, 2
	pslld mm7, 2
	
	movd eax, mm6
	psrlq mm6, 32
	movd ecx, mm7
	psrlq mm7, 32
	movd ebx, mm6
	movd edx, mm7

	lea   rax, [rax + rax*2]
	lea   rbx, [rbx + rbx*2]
	lea   rcx, [rcx + rcx*2]
	lea   rdx, [rdx + rdx*2]

	movlps xmm5, [rsi + rax*4]
	movlps xmm7, [rsi + rcx*4]
	movhps xmm5, [rsi + rbx*4]
	movhps xmm7, [rsi + rdx*4] ;# got half coulomb table 

	movaps xmm4, xmm5
	shufps xmm4, xmm7, 136  ;# 10001000
	shufps xmm5, xmm7, 221  ;# 11011101

	movlps xmm7, [rsi + rax*4 + 8]
	movlps xmm3, [rsi + rcx*4 + 8]
	movhps xmm7, [rsi + rbx*4 + 8]
	movhps xmm3, [rsi + rdx*4 + 8] ;# other half of coulomb table  
	movaps xmm6, xmm7
	shufps xmm6, xmm3, 136  ;# 10001000
	shufps xmm7, xmm3, 221  ;# 11011101
	;# coulomb table ready, in xmm4-xmm7      
        
	mulps  xmm6, xmm1   	;# xmm6=Geps 
	mulps  xmm7, xmm2   	;# xmm7=Heps2 
	addps  xmm5, xmm6
	addps  xmm5, xmm7   	;# xmm5=Fp        
	movaps xmm0, [rsp + nb333nf_qqH]
	mulps  xmm5, xmm1 ;# xmm5=eps*Fp 
	addps  xmm5, xmm4 ;# xmm5=VV 
	mulps  xmm5, xmm0 ;# vcoul=qq*VV  
	;# at this point mm5 contains vcoul 
	;# increment vcoul 
	addps  xmm5, [rsp + nb333nf_vctot]
	movaps [rsp + nb333nf_vctot], xmm5 

	;# Done with H2, do M interactions 
	movaps xmm7, [rsp + nb333nf_rM]
	mulps   xmm7, [rsp + nb333nf_tsc]
	movhlps xmm4, xmm7
	cvttps2pi mm6, xmm7
	cvttps2pi mm7, xmm4	;# mm6/mm7 contain lu indices 
	
	cvtpi2ps xmm3, mm6
	cvtpi2ps xmm4, mm7
	movlhps xmm3, xmm4
	
	subps xmm7, xmm3
	movaps xmm1, xmm7	;# xmm1=eps 
	movaps xmm2, xmm1
	mulps  xmm2, xmm2	;# xmm2=eps2 
	pslld mm6, 2
	pslld mm7, 2
	
	movd eax, mm6
	psrlq mm6, 32
	movd ecx, mm7
	psrlq mm7, 32
	movd ebx, mm6
	movd edx, mm7

	lea   rax, [rax + rax*2]
	lea   rbx, [rbx + rbx*2]
	lea   rcx, [rcx + rcx*2]
	lea   rdx, [rdx + rdx*2]

	movlps xmm5, [rsi + rax*4]
	movlps xmm7, [rsi + rcx*4]
	movhps xmm5, [rsi + rbx*4]
	movhps xmm7, [rsi + rdx*4] ;# got half coulomb table 

	movaps xmm4, xmm5
	shufps xmm4, xmm7, 136  ;# 10001000
	shufps xmm5, xmm7, 221  ;# 11011101

	movlps xmm7, [rsi + rax*4 + 8]
	movlps xmm3, [rsi + rcx*4 + 8]
	movhps xmm7, [rsi + rbx*4 + 8]
	movhps xmm3, [rsi + rdx*4 + 8] ;# other half of coulomb table  
	movaps xmm6, xmm7
	shufps xmm6, xmm3, 136  ;# 10001000
	shufps xmm7, xmm3, 221  ;# 11011101
	;# coulomb table ready, in xmm4-xmm7      
        
	mulps  xmm6, xmm1   	;# xmm6=Geps 
	mulps  xmm7, xmm2   	;# xmm7=Heps2 
	addps  xmm5, xmm6
	addps  xmm5, xmm7   	;# xmm5=Fp        
	movaps xmm0, [rsp + nb333nf_qqM]
	mulps  xmm5, xmm1 ;# xmm5=eps*Fp 
	addps  xmm5, xmm4 ;# xmm5=VV 
	mulps  xmm5, xmm0 ;# vcoul=qq*VV  
	;# at this point mm5 contains vcoul 
	;# increment vcoul 
	addps  xmm5, [rsp + nb333nf_vctot]
	movaps [rsp + nb333nf_vctot], xmm5 
	;# should we do one more iteration? 
	sub dword ptr [rsp + nb333nf_innerk],  4
	jl    .nb333nf_odd_inner
	jmp   .nb333nf_unroll_loop
.nb333nf_odd_inner:	
	add dword ptr [rsp + nb333nf_innerk],  4
	jnz   .nb333nf_odd_loop
	jmp   .nb333nf_updateouterdata
.nb333nf_odd_loop:
	mov   rdx, [rsp + nb333nf_innerjjnr] 	;# pointer to jjnr[k] 
	mov   eax, [rdx]	
	add qword ptr [rsp + nb333nf_innerjjnr],  4	

 	xorps xmm4, xmm4  	;# clear reg.
	movss xmm4, [rsp + nb333nf_iqM]
	mov rsi, [rbp + nb333nf_charge] 
	movhps xmm4, [rsp + nb333nf_iqH]  ;# [qM  0  qH  qH] 
	shufps xmm4, xmm4, 41	;# [0 qH qH qM]

	movss xmm3, [rsi + rax*4]	;# charge in xmm3 
	shufps xmm3, xmm3, 0
	mulps xmm3, xmm4
	movaps [rsp + nb333nf_qqM], xmm3	;# use dummy qq for storage 
	
	xorps xmm6, xmm6
	mov rsi, [rbp + nb333nf_type]
	mov ebx, [rsi + rax*4]
	mov rsi, [rbp + nb333nf_vdwparam]
	shl ebx, 1	
	add ebx, [rsp + nb333nf_ntia]
	movlps xmm6, [rsi + rbx*4]
	movaps xmm7, xmm6
	shufps xmm6, xmm6, 252  ;# 11111100
	shufps xmm7, xmm7, 253  ;# 11111101
	movaps [rsp + nb333nf_c6], xmm6
	movaps [rsp + nb333nf_c12], xmm7
	
	mov