nb_kernel314_x86_64_sse.intel_syntax.s

来自「最著名最快的分子模拟软件」· S 代码 · 共 2,319 行 · 第 1/5 页

	addss   xmm1, [rsp + nb314_fiyO]
	addss   xmm2, [rsp + nb314_fizO]
	movss  [rsp + nb314_fixO], xmm0
	movss  [rsp + nb314_fiyO], xmm1
	movss  [rsp + nb314_fizO], xmm2

	;# do  H1 coulomb interaction
	movaps xmm0, [rsp + nb314_rinvH1H1] ;# rinv 
	movaps xmm1, xmm0
	mulps  xmm1, [rsp + nb314_rsqH1H1] 	;# r
	mulps xmm1, [rsp + nb314_tsc]
	
	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
	
	psrlq mm6, 32
	movd ebx, mm6
	movd ecx, mm7
	psrlq mm7, 32
	movd edx, mm7		;# table indices in ebx,ecx,edx 

	mov rsi, [rbp + nb314_VFtab]

	movlps xmm4, [rsi + rbx*4]
	movlps xmm3, [rsi + rcx*4]
	movlps xmm7, [rsi + rdx*4]
	movhps xmm4, [rsi + rbx*4 + 8]
	movhps xmm3, [rsi + rcx*4 + 8]
	movhps xmm7, [rsi + rdx*4 + 8]
	movaps xmm6, xmm3
	unpcklps xmm6, xmm7
	unpckhps xmm3, xmm7
	movaps xmm5, xmm4
	movaps xmm7, xmm4
	shufps xmm4, xmm6, 0x40
	shufps xmm5, xmm6, 0xE4
	movaps xmm6, xmm7
	shufps xmm6, xmm3, 0x48
	shufps xmm7, xmm3, 0xEC
	;# coulomb table ready, in xmm4-xmm7
	
	mulps  xmm6, xmm1   	;# xmm6=Geps 
	mulps  xmm7, xmm2   	;# xmm7=Heps2 
	addps  xmm5, xmm6
	addps  xmm5, xmm7   	;# xmm5=Fp 
	addps  xmm7, xmm7  	;# two*Heps2 

	;# fetch charges to xmm3 (temporary) 
	xorps  xmm3, xmm3
	movss   xmm3, [rsp + nb314_qqHH]
	movhps  xmm3, [rsp + nb314_qqMH]
	shufps  xmm3, xmm3, 193 ;# 11000001 

	addps  xmm7, xmm6
	addps  xmm7, xmm5 ;# xmm7=FF 
	mulps  xmm5, xmm1 ;# xmm5=eps*Fp 
	addps  xmm5, xmm4 ;# xmm5=VV 
	mulps  xmm5, xmm3 ;# vcoul=qq*VV  
	mulps  xmm3, xmm7 ;# fijC=FF*qq 
	;# at this point xmm5 contains vcoul and xmm3 fijC 
	
	addps  xmm5, [rsp + nb314_vctot]
	movaps [rsp + nb314_vctot], xmm5

	mulps  xmm3, [rsp + nb314_tsc]
	xorps  xmm2, xmm2
	subps  xmm2, xmm3
	mulps  xmm0, xmm2
	movaps xmm1, xmm0
	movaps xmm2, xmm0			

	mulps   xmm0, [rsp + nb314_dxH1H1]
	mulps   xmm1, [rsp + nb314_dyH1H1]
	mulps   xmm2, [rsp + nb314_dzH1H1]
	;# update forces H1 - j water 
	movaps  xmm3, [rsp + nb314_fjxO]
	movaps  xmm4, [rsp + nb314_fjyO]
	movaps  xmm5, [rsp + nb314_fjzO]
	addps   xmm3, xmm0
	addps   xmm4, xmm1
	addps   xmm5, xmm2
	movaps  [rsp + nb314_fjxO], xmm3
	movaps  [rsp + nb314_fjyO], xmm4
	movaps  [rsp + nb314_fjzO], xmm5
	addps   xmm0, [rsp + nb314_fixH1]
	addps   xmm1, [rsp + nb314_fiyH1]
	addps   xmm2, [rsp + nb314_fizH1]
	movaps  [rsp + nb314_fixH1], xmm0
	movaps  [rsp + nb314_fiyH1], xmm1
	movaps  [rsp + nb314_fizH1], xmm2	
	
	;# i H2 & M simultaneously first get i particle coords: 
    movaps  xmm0, [rsp + nb314_jxO]
    movaps  xmm1, [rsp + nb314_jyO]
    movaps  xmm2, [rsp + nb314_jzO]
    movaps  xmm3, xmm0
    movaps  xmm4, xmm1
    movaps  xmm5, xmm2
	subps   xmm0, [rsp + nb314_ixH2]
	subps   xmm1, [rsp + nb314_iyH2]
	subps   xmm2, [rsp + nb314_izH2]
	subps   xmm3, [rsp + nb314_ixM] 
	subps   xmm4, [rsp + nb314_iyM] 
	subps   xmm5, [rsp + nb314_izM] 
	movaps [rsp + nb314_dxH2H2], xmm0
	movaps [rsp + nb314_dyH2H2], xmm1
	movaps [rsp + nb314_dzH2H2], xmm2
	movaps [rsp + nb314_dxMM], xmm3
	movaps [rsp + nb314_dyMM], xmm4
	movaps [rsp + nb314_dzMM], xmm5
	mulps xmm0, xmm0
	mulps xmm1, xmm1
	mulps xmm2, xmm2
	mulps xmm3, xmm3
	mulps xmm4, xmm4
	mulps xmm5, xmm5
	addps xmm0, xmm1
	addps xmm4, xmm3
	addps xmm0, xmm2	;# have rsqH2 in xmm0 
	addps xmm4, xmm5	;# have rsqM in xmm4 

	;# start with H2, save data 
	movaps [rsp + nb314_rsqH2H2], xmm0
	movaps [rsp + nb314_rsqMM], xmm4	
	;# do invsqrt 
	rsqrtps xmm1, xmm0
	rsqrtps xmm5, xmm4
	movaps  xmm2, xmm1
	movaps  xmm6, xmm5
	mulps   xmm1, xmm1
	mulps   xmm5, xmm5
	movaps  xmm3, [rsp + nb314_three]
	movaps  xmm7, xmm3
	mulps   xmm1, xmm0
	mulps   xmm5, xmm4
	subps   xmm3, xmm1
	subps   xmm7, xmm5
	mulps   xmm3, xmm2
	mulps   xmm7, xmm6
	mulps   xmm3, [rsp + nb314_half] ;# rinv H2 - j water 
	mulps   xmm7, [rsp + nb314_half] ;# rinv M - j water  

	movaps [rsp + nb314_rinvH2H2], xmm3
	movaps [rsp + nb314_rinvMM], xmm7

	movaps xmm1, xmm3
	mulps  xmm1, [rsp + nb314_rsqH2H2]	;# xmm1=r 
	movaps xmm0, xmm3	;# xmm0=rinv 
	mulps  xmm1, [rsp + nb314_tsc]
	
	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

	psrlq mm6, 32
	movd ebx, mm6
	movd ecx, mm7
	psrlq mm7, 32
	movd edx, mm7		;# table indices in ebx,ecx,edx 

	movlps xmm4, [rsi + rbx*4]
	movlps xmm3, [rsi + rcx*4]
	movlps xmm7, [rsi + rdx*4]
	movhps xmm4, [rsi + rbx*4 + 8]
	movhps xmm3, [rsi + rcx*4 + 8]
	movhps xmm7, [rsi + rdx*4 + 8]
	movaps xmm6, xmm3
	unpcklps xmm6, xmm7
	unpckhps xmm3, xmm7
	movaps xmm5, xmm4
	movaps xmm7, xmm4
	shufps xmm4, xmm6, 0x40
	shufps xmm5, xmm6, 0xE4
	movaps xmm6, xmm7
	shufps xmm6, xmm3, 0x48
	shufps xmm7, xmm3, 0xEC
	;# coulomb table ready, in xmm4-xmm7
	mulps  xmm6, xmm1   	;# xmm6=Geps 
	mulps  xmm7, xmm2   	;# xmm7=Heps2 
	addps  xmm5, xmm6
	addps  xmm5, xmm7   	;# xmm5=Fp 
	mulps  xmm7, [rsp + nb314_two]   	;# two*Heps2 

	xorps  xmm3, xmm3
	
	;# fetch charges to xmm3 (temporary) 
	movss   xmm3, [rsp + nb314_qqHH]
	movhps  xmm3, [rsp + nb314_qqMH]
	shufps  xmm3, xmm3, 193 ;# 11000001
	
	addps  xmm7, xmm6
	addps  xmm7, xmm5 ;# xmm7=FF 
	mulps  xmm5, xmm1 ;# xmm5=eps*Fp 
	addps  xmm5, xmm4 ;# xmm5=VV 
	mulps  xmm5, xmm3 ;# vcoul=qq*VV  
	mulps  xmm3, xmm7 ;# fijC=FF*qq 
	;# at this point xmm5 contains vcoul and xmm3 fijC 
	addps  xmm5, [rsp + nb314_vctot]
	movaps [rsp + nb314_vctot], xmm5	

	xorps  xmm1, xmm1

	mulps xmm3, [rsp + nb314_tsc]
	mulps xmm3, xmm0
	subps  xmm1, xmm3
	
	movaps  xmm0, xmm1
	movaps  xmm2, xmm1
	mulps   xmm0, [rsp + nb314_dxH2H2]
	mulps   xmm1, [rsp + nb314_dyH2H2]
	mulps   xmm2, [rsp + nb314_dzH2H2]
	;# update forces H1 - j water 
	movaps  xmm3, [rsp + nb314_fjxO]
	movaps  xmm4, [rsp + nb314_fjyO]
	movaps  xmm5, [rsp + nb314_fjzO]
	addps   xmm3, xmm0
	addps   xmm4, xmm1
	addps   xmm5, xmm2
	movaps  [rsp + nb314_fjxO], xmm3
	movaps  [rsp + nb314_fjyO], xmm4
	movaps  [rsp + nb314_fjzO], xmm5
	addps   xmm0, [rsp + nb314_fixH2]
	addps   xmm1, [rsp + nb314_fiyH2]
	addps   xmm2, [rsp + nb314_fizH2]
	movaps  [rsp + nb314_fixH2], xmm0
	movaps  [rsp + nb314_fiyH2], xmm1
	movaps  [rsp + nb314_fizH2], xmm2
	
	;# do table for i M - j water interaction 
	movaps xmm0, [rsp + nb314_rinvMM]
	movaps xmm1, xmm0
	mulps  xmm1, [rsp + nb314_rsqMM]	;# xmm0=rinv, xmm1=r 
	mulps  xmm1, [rsp + nb314_tsc]
	
	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
 
	psrlq mm6, 32
	movd ebx, mm6
	movd ecx, mm7
	psrlq mm7, 32
	movd edx, mm7		;# table indices in ebx,ecx,edx 

	movlps xmm4, [rsi + rbx*4]
	movlps xmm3, [rsi + rcx*4]
	movlps xmm7, [rsi + rdx*4]
	movhps xmm4, [rsi + rbx*4 + 8]
	movhps xmm3, [rsi + rcx*4 + 8]
	movhps xmm7, [rsi + rdx*4 + 8]
	movaps xmm6, xmm3
	unpcklps xmm6, xmm7
	unpckhps xmm3, xmm7
	movaps xmm5, xmm4
	movaps xmm7, xmm4
	shufps xmm4, xmm6, 0x40
	shufps xmm5, xmm6, 0xE4
	movaps xmm6, xmm7
	shufps xmm6, xmm3, 0x48
	shufps xmm7, xmm3, 0xEC

	;# coulomb table ready, in xmm4-xmm7
	mulps  xmm6, xmm1   	;# xmm6=Geps 
	mulps  xmm7, xmm2   	;# xmm7=Heps2 
	addps  xmm5, xmm6
	addps  xmm5, xmm7   	;# xmm5=Fp 
	mulps  xmm7, [rsp + nb314_two]   	;# two*Heps2 

	xorps  xmm3, xmm3
	;# fetch charges to xmm3 (temporary) 
	movss   xmm3, [rsp + nb314_qqMH]
	movhps  xmm3, [rsp + nb314_qqMM]
	shufps  xmm3, xmm3, 193 ;# 11000001
	
	addps  xmm7, xmm6
	addps  xmm7, xmm5 ;# xmm7=FF 
	mulps  xmm5, xmm1 ;# xmm5=eps*Fp 
	addps  xmm5, xmm4 ;# xmm5=VV 
	mulps  xmm5, xmm3 ;# vcoul=qq*VV  
	mulps  xmm3, xmm7 ;# fijC=FF*qq 
	;# at this point xmm5 contains vcoul and xmm3 fijC 
	addps  xmm5, [rsp + nb314_vctot]
	movaps [rsp + nb314_vctot], xmm5	

	xorps  xmm1, xmm1

	mulps xmm3, [rsp + nb314_tsc]
	mulps xmm3, xmm0
	subps  xmm1, xmm3
	
	movaps  xmm0, xmm1
	movaps  xmm2, xmm1
	
	mulps   xmm0, [rsp + nb314_dxMM]
	mulps   xmm1, [rsp + nb314_dyMM]
	mulps   xmm2, [rsp + nb314_dzMM]
	movaps  xmm3, [rsp + nb314_fjxO]
	movaps  xmm4, [rsp + nb314_fjyO]
	movaps  xmm5, [rsp + nb314_fjzO]
	addps   xmm3, xmm0
	addps   xmm4, xmm1
	addps   xmm5, xmm2
	mov     rsi, [rbp + nb314_faction]
	movaps  [rsp + nb314_fjxO], xmm3
	movaps  [rsp + nb314_fjyO], xmm4
	movaps  [rsp + nb314_fjzO], xmm5
	addps   xmm0, [rsp + nb314_fixM]
	addps   xmm1, [rsp + nb314_fiyM]
	addps   xmm2, [rsp + nb314_fizM]
	movaps  [rsp + nb314_fixM], xmm0
	movaps  [rsp + nb314_fiyM], xmm1
	movaps  [rsp + nb314_fizM], xmm2

	;# update j water forces from local variables.
	;# transpose back first
	movaps  xmm0, [rsp + nb314_fjxO] ;# Ox H1x H2x Mx 
	movaps  xmm1, [rsp + nb314_fjyO] ;# Oy H1y H2y My
	movaps  xmm2, [rsp + nb314_fjzO] ;# Oz H1z H2z Mz
	 
	movaps  xmm3, xmm0
	movaps  xmm4, xmm0
	unpcklps xmm3, xmm1       	;# Ox Oy - -
	shufps  xmm4, xmm2, 0x1	  	;# h1x - Oz -
	movaps  xmm5, xmm1
	movaps  xmm6, xmm0
	unpcklps xmm5, xmm2 	  	;# - - H1y H1z
	unpckhps xmm6, xmm1	  	;# h2x h2y - - 
	unpckhps xmm1, xmm2	  	;# - - My Mz
	
	shufps   xmm2, xmm0, 0x32  ;# (00110010) h2z - Mx -
	shufps   xmm3, xmm4, 36  ;# 00100100 ;# Ox Oy Oz H1x 
	shufps   xmm5, xmm6, 78  ;# 01001110 ;# h1y h1z h2x h2y
	shufps   xmm2, xmm1, 232  ;# 11101000 ;# h2z mx my mz

	movlps  xmm0, [rsi + rax*4]
	movlps  xmm1, [rsi + rax*4 + 16]
	movlps  xmm4, [rsi + rax*4 + 32]
	movhps  xmm0, [rsi + rax*4 + 8]
	movhps  xmm1, [rsi + rax*4 + 24]
	movhps  xmm4, [rsi + rax*4 + 40]
	addps   xmm0, xmm3
	addps   xmm1, xmm5
	addps   xmm4, xmm2
	movlps   [rsi + rax*4], xmm0
	movlps   [rsi + rax*4 + 16], xmm1
	movlps   [rsi + rax*4 + 32], xmm4
	movhps   [rsi + rax*4 + 8], xmm0
	movhps   [rsi + rax*4 + 24], xmm1
	movhps   [rsi + rax*4 + 40], xmm4
	
	dec dword ptr [rsp + nb314_innerk]
	jz    .nb314_updateouterdata
	jmp   .nb314_single_loop
.nb314_updateouterdata:
	mov   ecx, [rsp + nb314_ii3]
	mov   rdi, [rbp + nb314_faction]
	mov   rsi, [rbp + nb314_fshift]
	mov   edx, [rsp + nb314_is3]

	;# accumulate  Oi forces in xmm0, xmm1, xmm2 
	movaps xmm0, [rsp + nb314_fixO]
	movaps xmm1, [rsp + nb314_fiyO] 
	movaps xmm2, [rsp + nb314_fizO]

	movhlps xmm3, xmm0
	movhlps xmm4, xmm1
	movhlps xmm5, xmm2
	addps  xmm0, xmm3
	addps  xmm1, xmm4
	addps  xmm2, xmm5 ;# sum is in 1/2 in xmm0-xmm2 

	movaps xmm3, xmm0	
	movaps xmm4, xmm1	
	movaps xmm5, xmm2	

	shufps xmm3, xmm3, 1
	shufps xmm4, xmm4, 1
	shufps xmm5, xmm5, 1
	addss  xmm0, xmm3
	addss  xmm1, xmm4
	addss  xmm2, xmm5	;# xmm0-xmm2 has single force in pos0 

	;# increment i force 
	movss  xmm3, [rdi + rcx*4]
	movss  xmm4, [rdi + rcx*4 + 4]
	movss  xmm5, [rdi + rcx*4 + 8]
	subss  xmm3, xmm0
	subss  xmm4, xmm1
	subss  xmm5, xmm2
	movss  [rdi + rcx*4],     xmm3
	movss  [rdi + rcx*4 + 4], xmm4
	movss  [rdi + rcx*4 + 8], xmm5

	;# accumulate force in xmm6/xmm7 for fshift 
	movaps xmm6, xmm0
	movss xmm7, xmm2
	movlhps xmm6, xmm1
	shufps  xmm6, xmm6, 8 ;# 00001000	

	;# accumulate H1i forces in xmm0, xmm1, xmm2 
	movaps xmm0, [rsp + nb314_fixH1]
	movaps xmm1, [rsp + nb314_fiyH1]
	movaps xmm2, [rsp + nb314_fizH1]

	movhlps xmm3, xmm0
	movhlps xmm4, xmm1
	movhlps xmm5, xmm2
	addps  xmm0, xmm3
	addps  xmm1, xmm4
	addps  xmm2, xmm5 ;# sum is in 1/2 in xmm0-xmm2 

	movaps xmm3, xmm0	
	movaps xmm4, xmm1	
	movaps xmm5, xmm2	

	shufps xmm3, xmm3, 1
	shufps xmm4, xmm4, 1
	shufps xmm5, xmm5, 1
	addss  xmm0, xmm3
	addss  xmm1, xmm4
	addss  xmm2, xmm5	;# xmm0-xmm2 has single force in pos0 

	;# increment i force 
	movss  xmm3, [rdi + rcx*4 + 12]
	movss  xmm4, [rdi + rcx*4 + 16]
	movss  xmm5, [rdi + rcx*4 + 20]
	subss  xmm3, xmm0
	subss  xmm4, xmm1
	subss  xmm5, xmm2
	movss  [rdi + rcx*4 + 12], xmm3
	movss  [rdi + rcx*4 + 16], xmm4
	movss  [rdi + rcx*4 + 20], xmm5

	;# accumulate force in xmm6/xmm7 for fshift 
	addss xmm7, xmm2
	movlhps xmm0, xmm1
	shufps  xmm0, xmm0, 8 ;# 00001000	
	addps   xmm6, xmm0