nb_kernel232_ia32_sse.intel_syntax.s

来自「最著名最快的分子模拟软件」· S 代码 · 共 2,146 行 · 第 1/5 页

	mulps xmm1, [esp + nb232_dyOO]
	mulps xmm2, [esp + nb232_dzOO]
	subps xmm3, xmm0
	subps xmm4, xmm1
	subps xmm5, xmm2
	addps xmm0, [esp + nb232_fixO]
	addps xmm1, [esp + nb232_fiyO]
	addps xmm2, [esp + nb232_fizO]
	movaps [esp + nb232_fjxO], xmm3
	movaps [esp + nb232_fjyO], xmm4
	movaps [esp + nb232_fjzO], xmm5
	movaps [esp + nb232_fixO], xmm0
	movaps [esp + nb232_fiyO], xmm1
	movaps [esp + nb232_fizO], xmm2

	;# O-H1 interaction 
	movaps xmm0, [esp + nb232_rinvOH1]
	movaps xmm7, xmm0	;# xmm7=rinv 
	movaps xmm5, [esp + nb232_krf]
	movaps xmm1, xmm0
	mulps  xmm5, [esp + nb232_rsqOH1] ;# xmm5=krsq 
	movaps xmm4, xmm5
	addps  xmm4, xmm7	;# xmm4=rinv+ krsq 
	mulps  xmm0, xmm0
	subps  xmm4, [esp + nb232_crf]
	mulps  xmm4, [esp + nb232_qqOH] ;# xmm4=voul=qq*(rinv+ krsq) 
	mulps  xmm5, [esp + nb232_two]
	subps  xmm7, xmm5	;# xmm7=rinv-2*krsq 
	mulps  xmm7, [esp + nb232_qqOH] ;# xmm7 = coul part of fscal 
	addps  xmm6, xmm4	;# add to local vctot 
	mulps xmm0, xmm7	;# fsOH1  
	movaps xmm1, xmm0
	movaps xmm2, xmm0
	
	xorps xmm3, xmm3
	movaps xmm4, xmm3
	movaps xmm5, xmm3
	mulps xmm0, [esp + nb232_dxOH1]
	mulps xmm1, [esp + nb232_dyOH1]
	mulps xmm2, [esp + nb232_dzOH1]
	subps xmm3, xmm0
	subps xmm4, xmm1
	subps xmm5, xmm2
	addps xmm0, [esp + nb232_fixO]
	addps xmm1, [esp + nb232_fiyO]
	addps xmm2, [esp + nb232_fizO]
	movaps [esp + nb232_fjxH1], xmm3
	movaps [esp + nb232_fjyH1], xmm4
	movaps [esp + nb232_fjzH1], xmm5
	movaps [esp + nb232_fixO], xmm0
	movaps [esp + nb232_fiyO], xmm1
	movaps [esp + nb232_fizO], xmm2

	;# O-H2 interaction  
	movaps xmm0, [esp + nb232_rinvOH2]
	movaps xmm7, xmm0	;# xmm7=rinv 
	movaps xmm5, [esp + nb232_krf]	
	movaps xmm1, xmm0
	mulps  xmm5, [esp + nb232_rsqOH2] ;# xmm5=krsq 
	movaps xmm4, xmm5
	addps  xmm4, xmm7	;# xmm4=r inv+ krsq 
	mulps xmm0, xmm0
	subps  xmm4, [esp + nb232_crf]
	mulps  xmm4, [esp + nb232_qqOH] ;# xmm4=voul=qq*(rinv+ krsq) 
	mulps  xmm5, [esp + nb232_two]
	subps  xmm7, xmm5	;# xmm7=rinv-2*krsq 
	mulps  xmm7, [esp + nb232_qqOH] ;# xmm7 = coul part of fscal 
	addps  xmm6, xmm4	;# add to local vctot 
	mulps xmm0, xmm7	;# fsOH2 
	movaps xmm1, xmm0
	movaps xmm2, xmm0

	xorps xmm3, xmm3
	movaps xmm4, xmm3
	movaps xmm5, xmm3
	mulps xmm0, [esp + nb232_dxOH2]
	mulps xmm1, [esp + nb232_dyOH2]
	mulps xmm2, [esp + nb232_dzOH2]
	subps xmm3, xmm0
	subps xmm4, xmm1
	subps xmm5, xmm2
	addps xmm0, [esp + nb232_fixO]
	addps xmm1, [esp + nb232_fiyO]
	addps xmm2, [esp + nb232_fizO]
	movaps [esp + nb232_fjxH2], xmm3
	movaps [esp + nb232_fjyH2], xmm4
	movaps [esp + nb232_fjzH2], xmm5
	movaps [esp + nb232_fixO], xmm0
	movaps [esp + nb232_fiyO], xmm1
	movaps [esp + nb232_fizO], xmm2

	;# H1-O interaction 
	movaps xmm0, [esp + nb232_rinvH1O]
	movaps xmm7, xmm0	;# xmm7=rinv 
	movaps xmm5, [esp + nb232_krf]	
	movaps xmm1, xmm0
	mulps  xmm5, [esp + nb232_rsqH1O] ;# xmm5=krsq 
	movaps xmm4, xmm5
	addps  xmm4, xmm7	;# xmm4=rinv+ krsq 
	mulps xmm0, xmm0
	subps  xmm4, [esp + nb232_crf]
	mulps  xmm4, [esp + nb232_qqOH] ;# xmm4=voul=qq*(rinv+ krsq) 
	mulps  xmm5, [esp + nb232_two]
	subps  xmm7, xmm5	;# xmm7=rinv-2*krsq 
	mulps  xmm7, [esp + nb232_qqOH] ;# xmm7 = coul part of fscal 
	addps  xmm6, xmm4	;# add to local vctot 
	mulps xmm0, xmm7	;# fsOH2 
	movaps xmm1, xmm0
	movaps xmm2, xmm0

	movaps xmm3, [esp + nb232_fjxO]
	movaps xmm4, [esp + nb232_fjyO]
	movaps xmm5, [esp + nb232_fjzO]
	mulps xmm0, [esp + nb232_dxH1O]
	mulps xmm1, [esp + nb232_dyH1O]
	mulps xmm2, [esp + nb232_dzH1O]
	subps xmm3, xmm0
	subps xmm4, xmm1
	subps xmm5, xmm2
	addps xmm0, [esp + nb232_fixH1]
	addps xmm1, [esp + nb232_fiyH1]
	addps xmm2, [esp + nb232_fizH1]
	movaps [esp + nb232_fjxO], xmm3
	movaps [esp + nb232_fjyO], xmm4
	movaps [esp + nb232_fjzO], xmm5
	movaps [esp + nb232_fixH1], xmm0
	movaps [esp + nb232_fiyH1], xmm1
	movaps [esp + nb232_fizH1], xmm2

	;# H1-H1 interaction 
	movaps xmm0, [esp + nb232_rinvH1H1]
	movaps xmm7, xmm0	;# xmm7=rinv 
	movaps xmm5, [esp + nb232_krf]	
	movaps xmm1, xmm0
	mulps  xmm5, [esp + nb232_rsqH1H1] ;# xmm5=krsq 
	movaps xmm4, xmm5
	addps  xmm4, xmm7	;# xmm4=r inv+ krsq 
	subps  xmm4, [esp + nb232_crf]
	mulps xmm0, xmm0
	mulps  xmm4, [esp + nb232_qqHH] ;# xmm4=voul=qq*(rinv+ krsq) 
	mulps  xmm5, [esp + nb232_two]
	subps  xmm7, xmm5	;# xmm7=rinv-2*krsq 
	mulps  xmm7, [esp + nb232_qqHH] ;# xmm7 = coul part of fscal 
	addps  xmm6, xmm4	;# add to local vctot 
	mulps xmm0, xmm7	;# fsOH2 
	movaps xmm1, xmm0
	movaps xmm2, xmm0

	movaps xmm3, [esp + nb232_fjxH1]
	movaps xmm4, [esp + nb232_fjyH1]
	movaps xmm5, [esp + nb232_fjzH1]
	mulps xmm0, [esp + nb232_dxH1H1]
	mulps xmm1, [esp + nb232_dyH1H1]
	mulps xmm2, [esp + nb232_dzH1H1]
	subps xmm3, xmm0
	subps xmm4, xmm1
	subps xmm5, xmm2
	addps xmm0, [esp + nb232_fixH1]
	addps xmm1, [esp + nb232_fiyH1]
	addps xmm2, [esp + nb232_fizH1]
	movaps [esp + nb232_fjxH1], xmm3
	movaps [esp + nb232_fjyH1], xmm4
	movaps [esp + nb232_fjzH1], xmm5
	movaps [esp + nb232_fixH1], xmm0
	movaps [esp + nb232_fiyH1], xmm1
	movaps [esp + nb232_fizH1], xmm2

	;# H1-H2 interaction 
	movaps xmm0, [esp + nb232_rinvH1H2]
	movaps xmm7, xmm0	;# xmm7=rinv 
	movaps xmm5, [esp + nb232_krf]	
	movaps xmm1, xmm0
	mulps  xmm5, [esp + nb232_rsqH1H2] ;# xmm5=krsq 
	movaps xmm4, xmm5
	addps  xmm4, xmm7	;# xmm4=r inv+ krsq 
	mulps xmm0, xmm0
	subps  xmm4, [esp + nb232_crf]
	mulps  xmm4, [esp + nb232_qqHH] ;# xmm4=voul=qq*(rinv+ krsq) 
	mulps  xmm5, [esp + nb232_two]
	subps  xmm7, xmm5	;# xmm7=rinv-2*krsq 
	mulps  xmm7, [esp + nb232_qqHH] ;# xmm7 = coul part of fscal 
	addps  xmm6, xmm4	;# add to local vctot 
	mulps xmm0, xmm7	;# fsOH2 
	movaps xmm1, xmm0
	movaps xmm2, xmm0
	
	movaps xmm3, [esp + nb232_fjxH2]
	movaps xmm4, [esp + nb232_fjyH2]
	movaps xmm5, [esp + nb232_fjzH2]
	mulps xmm0, [esp + nb232_dxH1H2]
	mulps xmm1, [esp + nb232_dyH1H2]
	mulps xmm2, [esp + nb232_dzH1H2]
	subps xmm3, xmm0
	subps xmm4, xmm1
	subps xmm5, xmm2
	addps xmm0, [esp + nb232_fixH1]
	addps xmm1, [esp + nb232_fiyH1]
	addps xmm2, [esp + nb232_fizH1]
	movaps [esp + nb232_fjxH2], xmm3
	movaps [esp + nb232_fjyH2], xmm4
	movaps [esp + nb232_fjzH2], xmm5
	movaps [esp + nb232_fixH1], xmm0
	movaps [esp + nb232_fiyH1], xmm1
	movaps [esp + nb232_fizH1], xmm2

	;# H2-O interaction 
	movaps xmm0, [esp + nb232_rinvH2O]
	movaps xmm7, xmm0	;# xmm7=rinv 
	movaps xmm5, [esp + nb232_krf]	
	movaps xmm1, xmm0
	mulps  xmm5, [esp + nb232_rsqH2O] ;# xmm5=krsq 
	movaps xmm4, xmm5
	addps  xmm4, xmm7	;# xmm4=r inv+ krsq 
	subps  xmm4, [esp + nb232_crf]
	mulps xmm0, xmm0
	mulps  xmm4, [esp + nb232_qqOH] ;# xmm4=voul=qq*(rinv+ krsq) 
	mulps  xmm5, [esp + nb232_two]
	subps  xmm7, xmm5	;# xmm7=rinv-2*krsq 
	mulps  xmm7, [esp + nb232_qqOH] ;# xmm7 = coul part of fscal 
	addps  xmm6, xmm4	;# add to local vctot 
	mulps xmm0, xmm7	;# fsOH2 
	movaps xmm1, xmm0
	movaps xmm2, xmm0

	movaps xmm3, [esp + nb232_fjxO]
	movaps xmm4, [esp + nb232_fjyO]
	movaps xmm5, [esp + nb232_fjzO]
	mulps xmm0, [esp + nb232_dxH2O]
	mulps xmm1, [esp + nb232_dyH2O]
	mulps xmm2, [esp + nb232_dzH2O]
	subps xmm3, xmm0
	subps xmm4, xmm1
	subps xmm5, xmm2
	addps xmm0, [esp + nb232_fixH2]
	addps xmm1, [esp + nb232_fiyH2]
	addps xmm2, [esp + nb232_fizH2]
	movaps [esp + nb232_fjxO], xmm3
	movaps [esp + nb232_fjyO], xmm4
	movaps [esp + nb232_fjzO], xmm5
	movaps [esp + nb232_fixH2], xmm0
	movaps [esp + nb232_fiyH2], xmm1
	movaps [esp + nb232_fizH2], xmm2

	;# H2-H1 interaction 
	movaps xmm0, [esp + nb232_rinvH2H1]
	movaps xmm7, xmm0	;# xmm7=rinv 
	movaps xmm5, [esp + nb232_krf]	
	movaps xmm1, xmm0
	mulps  xmm5, [esp + nb232_rsqH2H1] ;# xmm5=krsq 
	movaps xmm4, xmm5
	addps  xmm4, xmm7	;# xmm4=r inv+ krsq 
	subps  xmm4, [esp + nb232_crf]
	mulps xmm0, xmm0
	mulps  xmm4, [esp + nb232_qqHH] ;# xmm4=voul=qq*(rinv+ krsq) 
	mulps  xmm5, [esp + nb232_two]
	subps  xmm7, xmm5	;# xmm7=rinv-2*krsq 
	mulps  xmm7, [esp + nb232_qqHH] ;# xmm7 = coul part of fscal 
	addps  xmm6, xmm4	;# add to local vctot 
	mulps xmm0, xmm7	;# fsOH2 
	movaps xmm1, xmm0
	movaps xmm2, xmm0

	movaps xmm3, [esp + nb232_fjxH1]
	movaps xmm4, [esp + nb232_fjyH1]
	movaps xmm5, [esp + nb232_fjzH1]
	mulps xmm0, [esp + nb232_dxH2H1]
	mulps xmm1, [esp + nb232_dyH2H1]
	mulps xmm2, [esp + nb232_dzH2H1]
	subps xmm3, xmm0
	subps xmm4, xmm1
	subps xmm5, xmm2
	addps xmm0, [esp + nb232_fixH2]
	addps xmm1, [esp + nb232_fiyH2]
	addps xmm2, [esp + nb232_fizH2]
	movaps [esp + nb232_fjxH1], xmm3
	movaps [esp + nb232_fjyH1], xmm4
	movaps [esp + nb232_fjzH1], xmm5
	movaps [esp + nb232_fixH2], xmm0
	movaps [esp + nb232_fiyH2], xmm1
	movaps [esp + nb232_fizH2], xmm2

	;# H2-H2 interaction 
	movaps xmm0, [esp + nb232_rinvH2H2]
	movaps xmm7, xmm0	;# xmm7=rinv 
	movaps xmm5, [esp + nb232_krf]	
	movaps xmm1, xmm0
	mulps  xmm5, [esp + nb232_rsqH2H2] ;# xmm5=krsq 
	movaps xmm4, xmm5
	addps  xmm4, xmm7	;# xmm4=r inv+ krsq 
	subps  xmm4, [esp + nb232_crf]
	mulps xmm0, xmm0
	mulps  xmm4, [esp + nb232_qqHH] ;# xmm4=voul=qq*(rinv+ krsq) 
	mulps  xmm5, [esp + nb232_two]
	subps  xmm7, xmm5	;# xmm7=rinv-2*krsq 
	mulps  xmm7, [esp + nb232_qqHH] ;# xmm7 = coul part of fscal 
	addps  xmm6, xmm4	;# add to local vctot 
	mulps xmm0, xmm7	;# fsOH2 
	movaps xmm1, xmm0
	movaps xmm2, xmm0

	movaps xmm1, xmm0
	movaps [esp + nb232_vctot], xmm6
	movaps xmm2, xmm0
	
	movaps xmm3, [esp + nb232_fjxH2]
	movaps xmm4, [esp + nb232_fjyH2]
	movaps xmm5, [esp + nb232_fjzH2]
	mulps xmm0, [esp + nb232_dxH2H2]
	mulps xmm1, [esp + nb232_dyH2H2]
	mulps xmm2, [esp + nb232_dzH2H2]
	subps xmm3, xmm0
	subps xmm4, xmm1
	subps xmm5, xmm2
	addps xmm0, [esp + nb232_fixH2]
	addps xmm1, [esp + nb232_fiyH2]
	addps xmm2, [esp + nb232_fizH2]
	movaps [esp + nb232_fjxH2], xmm3
	movaps [esp + nb232_fjyH2], xmm4
	movaps [esp + nb232_fjzH2], xmm5
	movaps [esp + nb232_fixH2], xmm0
	movaps [esp + nb232_fiyH2], xmm1
	movaps [esp + nb232_fizH2], xmm2

	mov edi, [ebp + nb232_faction]
		
	;# Did all interactions - now update j forces 
	;# At this stage forces are still on the stack, in positions:
	;# fjxO, fjyO, fjzO, ... , fjzH2.
	;# Each position is a quadruplet of forces for the four 
	;# corresponding j waters, so we need to transpose them before
	;# adding to the memory positions.
	;# 
	;# This _used_ to be a simple transpose, but the resulting high number
	;# of unaligned 128-bit load/stores might trigger a possible hardware 
	;# bug on Athlon and Opteron chips, so I have worked around it
	;# to use 64-bit load/stores instead. The performance hit should be
	;# very modest, since the 128-bit unaligned memory instructions were
	;# slow anyway. 
	
		
	;# 4 j waters with three atoms each - first do Oxygen X & Y forces for 4 j particles 
	movaps xmm0, [esp + nb232_fjxO] ;# xmm0= fjxOa  fjxOb  fjxOc  fjxOd 
	movaps xmm2, [esp + nb232_fjyO] ;# xmm1= fjyOa  fjyOb  fjyOc  fjyOd
	movlps xmm3, [edi + eax*4]
	movlps xmm4, [edi + ecx*4]
	movaps xmm1, xmm0
	unpcklps xmm0, xmm2    	   ;# xmm0= fjxOa  fjyOa  fjxOb  fjyOb
	unpckhps xmm1, xmm2        ;# xmm1= fjxOc  fjyOc  fjxOd  fjyOd
	movhps xmm3, [edi + ebx*4]
	movhps xmm4, [edi + edx*4]
	addps  xmm3, xmm0
	addps  xmm4, xmm1
	movlps [edi + eax*4], xmm3
	movlps [edi + ecx*4], xmm4
	movhps [edi + ebx*4], xmm3
	movhps [edi + edx*4], xmm4
	
	;# Oxygen Z & first hydrogen X forces for 4 j particles 
	movaps xmm0, [esp + nb232_fjzO]  ;# xmm0= fjzOa   fjzOb   fjzOc   fjzOd 
	movaps xmm2, [esp + nb232_fjxH1] ;# xmm1= fjxH1a  fjxH1b  fjxH1c  fjxH1d
	movlps xmm3, [edi + eax*4 + 8]
	movlps xmm4, [edi + ecx*4 + 8]
	movaps xmm1, xmm0
	unpcklps xmm0, xmm2    	   ;# xmm0= fjzOa  fjxH1a  fjzOb  fjxH1b
	unpckhps xmm1, xmm2        ;# xmm1= fjzOc  fjxH1c  fjzOd  fjxH1d
	movhps xmm3, [edi + ebx*4 + 8]
	movhps xmm4, [edi + edx*4 + 8]
	addps  xmm3, xmm0
	addps  xmm4, xmm1
	movlps [edi + eax*4 + 8], xmm3
	movlps [edi + ecx*4 + 8], xmm4
	movhps [edi + ebx*4 + 8], xmm3
	movhps [edi + edx*4 + 8], xmm4

	
	;# First hydrogen Y & Z forces for 4 j particles 
	movaps xmm0, [esp + nb232_fjyH1]  ;# xmm0= fjyH1a  fjyH1b  fjyH1c  fjyH1d 
	movaps xmm2, [esp + nb232_fjzH1] ;# xmm1= fjzH1a  fjzH1b  fjzH1c  fjzH1d
	movlps xmm3, [edi + eax*4 + 16]
	movlps xmm4, [edi + ecx*4 + 16]
	movaps xmm1, xmm0
	unpcklps xmm0, xmm2		;# xmm0= fjyH1a  fjzH1a  fjyH1b  fjzH1b
	unpckhps xmm1, xmm2		;# xmm1= fjyH1c  fjzH1c  fjyH1d  fjzH1d
	movhps xmm3, [edi + ebx*4 + 16]
	movhps xmm4, [edi + edx*4 + 16]
	addps  xmm3, xmm0
	addps  xmm4, xmm1
	movlps [edi + eax*4 + 16], xmm3
	movlps [edi + ecx*4 + 16], xmm4
	movhps [edi + ebx*4 + 16], xmm3
	movhps [edi + edx*4 + 16], xmm4

	
	;# Second hydrogen X & Y forces for 4 j particles 
	movaps xmm0, [esp + nb232_fjxH2]  ;# xmm0= fjxH2a  fjxH2b  fjxH2c  fjxH2d 
	movaps xmm2, [esp + nb232_fjyH2] ;# xmm1= fjyH2a  fjyH2b  fjyH2c  fjyH2d
	movlps xmm3, [edi + eax*4 + 24]
	movlps xmm4, [edi + ecx*4 + 24]
	movaps xmm1, xmm0
	unpcklps xmm0, xmm2		;# xmm0= fjxH2a  fjyH2a  fjxH2b  fjyH2b
	unpckhps xmm1, xmm2		;# xmm1= fjxH2c  fjyH2c  fjxH2d  fjyH2d
	movhps xmm3, [edi + ebx*4 + 24]
	movhps xmm4, [edi + edx*4 + 24]
	addps  xmm3, xmm0
	addps  xmm4, xmm1
	movlps [edi + eax*4 + 24], xmm3
	movlps [edi + ecx*4 + 24], xmm4
	movhps [edi + ebx*4 + 24], xmm3
	movhps [edi + edx*4 + 24], xmm4

	
	;# Second hydrogen Z forces for 4 j particles 
	;# Just load the four Z coords into one reg. each
	movss xmm4, [edi + eax*4 + 32]
	movss xmm5, [edi + ebx*4 + 32]
	movss xmm6, [edi + ecx*4 + 32]
	movss xmm7, [edi + edx*4 + 32]
	;# add what we have on the stack
	addss xmm4, [esp + nb232_fjzH2] 
	addss xmm5, [esp + nb232_fjzH2b] 
	addss xmm6, [esp + nb232_fjzH2c] 
	addss xmm7, [esp + nb232_fjzH2d]
	;# store back
	movss [edi + eax*4 + 32], xmm4
	movss [edi + ebx*4 + 32], xmm5
	movss [edi + ecx*4 + 32], xmm6
	movss [edi + edx*4 + 32], xmm7
	
	;# should we do one more iteration? 
	sub dword ptr [esp + nb232_innerk],  4