nb_kernel204_x86_64_sse.intel_syntax.s

来自「最著名最快的分子模拟软件」· S 代码 · 共 2,224 行 · 第 1/5 页

    shufps xmm2, xmm3,  136  ;# 10001000 => jzMa jzMb jzMc jzMd

    ;# xmm0 = Mx
    ;# xmm1 = My
    ;# xmm2 = Mz
        
    movaps xmm3, xmm0
    movaps xmm4, xmm1
    movaps xmm5, xmm2
    movaps xmm6, xmm0
    movaps xmm7, xmm1
    movaps xmm8, xmm2
    
    subps xmm0, [rsp + nb204_ixH1]
    subps xmm1, [rsp + nb204_iyH1]
    subps xmm2, [rsp + nb204_izH1]
    subps xmm3, [rsp + nb204_ixH2]
    subps xmm4, [rsp + nb204_iyH2]
    subps xmm5, [rsp + nb204_izH2]
    subps xmm6, [rsp + nb204_ixM]
    subps xmm7, [rsp + nb204_iyM]
    subps xmm8, [rsp + nb204_izM]
    
	movaps [rsp + nb204_dxH1M], xmm0
	movaps [rsp + nb204_dyH1M], xmm1
	movaps [rsp + nb204_dzH1M], xmm2
	mulps  xmm0, xmm0
	mulps  xmm1, xmm1
	mulps  xmm2, xmm2
	movaps [rsp + nb204_dxH2M], xmm3
	movaps [rsp + nb204_dyH2M], xmm4
	movaps [rsp + nb204_dzH2M], xmm5
	mulps  xmm3, xmm3
	mulps  xmm4, xmm4
	mulps  xmm5, xmm5
	movaps [rsp + nb204_dxMM], xmm6
	movaps [rsp + nb204_dyMM], xmm7
	movaps [rsp + nb204_dzMM], xmm8
	mulps  xmm6, xmm6
	mulps  xmm7, xmm7
	mulps  xmm8, xmm8
	addps  xmm0, xmm1
	addps  xmm0, xmm2
	addps  xmm3, xmm4
	addps  xmm3, xmm5
    addps  xmm6, xmm7
    addps  xmm6, xmm8

	;# start doing invsqrt for jM atoms
	rsqrtps xmm1, xmm0
	rsqrtps xmm4, xmm3
    rsqrtps xmm7, xmm6
	
	movaps  xmm2, xmm1
	movaps  xmm5, xmm4
    movaps  xmm8, xmm7
    
	mulps   xmm1, xmm1 ;# lu*lu
	mulps   xmm4, xmm4 ;# lu*lu
    mulps   xmm7, xmm7 ;# lu*lu
		
	movaps  xmm9, [rsp + nb204_three]
	movaps  xmm10, xmm9
    movaps  xmm11, xmm9

	mulps   xmm1, xmm0 ;# rsq*lu*lu
	mulps   xmm4, xmm3 ;# rsq*lu*lu 
    mulps   xmm7, xmm6 ;# rsq*lu*lu
	
	subps   xmm9, xmm1
	subps   xmm10, xmm4
    subps   xmm11, xmm7 ;# 3-rsq*lu*lu

	mulps   xmm9, xmm2
	mulps   xmm10, xmm5
    mulps   xmm11, xmm8 ;# lu*(3-rsq*lu*lu)

	movaps  xmm4, [rsp + nb204_half]
	mulps   xmm9, xmm4  ;# rinvH1M
	mulps   xmm10, xmm4 ;# rinvH2M
    mulps   xmm11, xmm4 ;# rinvMM
	
	;# M interactions 
    ;# rsq in xmm0,xmm3,xmm6  
    ;# rinv in xmm9, xmm10, xmm11

    movaps xmm1, xmm9 ;# copy of rinv
    movaps xmm4, xmm10
    movaps xmm7, xmm11
    movaps xmm2, [rsp + nb204_krf]    
    mulps  xmm9, xmm9   ;# rinvsq
    mulps  xmm10, xmm10
    mulps  xmm11, xmm11
    mulps  xmm0, xmm2  ;# k*rsq
    mulps  xmm3, xmm2
    mulps  xmm6, xmm2
    movaps xmm2, xmm0 ;# copy of k*rsq
    movaps xmm5, xmm3
    movaps xmm8, xmm6
    addps  xmm2, xmm1  ;# rinv+krsq
    addps  xmm5, xmm4
    addps  xmm8, xmm7
    movaps xmm14, [rsp + nb204_crf]
    subps  xmm2, xmm14   ;# rinv+krsq-crf
    subps  xmm5, xmm14
    subps  xmm8, xmm14
    movaps xmm12, [rsp + nb204_qqMH]
    movaps xmm13, [rsp + nb204_qqMM]    
    mulps  xmm2, xmm12 ;# xmm6=voul=qq*(rinv+ krsq-crf)
    mulps  xmm5, xmm12 ;# xmm6=voul=qq*(rinv+ krsq-crf)
    mulps  xmm8, xmm13 ;# xmm6=voul=qq*(rinv+ krsq-crf)
    addps  xmm0, xmm0 ;# 2*krsq
    addps  xmm3, xmm3 
    addps  xmm6, xmm6 
    subps  xmm1, xmm0 ;# rinv-2*krsq
    subps  xmm4, xmm3
    subps  xmm7, xmm6
    mulps  xmm1, xmm12   ;# (rinv-2*krsq)*qq
    mulps  xmm4, xmm12
    mulps  xmm7, xmm13
    addps  xmm5, xmm8
    addps  xmm2, xmm15
    addps  xmm2, xmm5
    movaps [rsp + nb204_vctot], xmm2
    
    mulps  xmm1, xmm9   ;# fscal
    mulps  xmm4, xmm10
    mulps  xmm7, xmm11

	;# move j M forces to local temp variables 
    movlps xmm9, [rdi + rax*4 + 36] ;# jxMa jyMa  -   -
    movlps xmm10, [rdi + rcx*4 + 36] ;# jxMc jyMc  -   -
    movhps xmm9, [rdi + rbx*4 + 36] ;# jxMa jyMa jxMb jyMb 
    movhps xmm10, [rdi + rdx*4 + 36] ;# jxMc jyMc jxMd jyMd 

    movss  xmm11, [rdi + rax*4 + 44] ;# jzMa  -  -  -
    movss  xmm12, [rdi + rcx*4 + 44] ;# jzMc  -  -  -
    movss  xmm2,  [rdi + rbx*4 + 44] ;# jzMb  -  -  -
    movss  xmm3,  [rdi + rdx*4 + 44] ;# jzMd  -  -  -
    movlhps xmm11, xmm2 ;# jzMa  -  jzMb  -
    movlhps xmm12, xmm3 ;# jzMc  -  jzMd -
    
    shufps xmm11, xmm12,  136  ;# 10001000 => jzMa jzMb jzMc jzMd

    ;# xmm9: jxMa jyMa jxMb jyMb 
    ;# xmm10: jxMc jyMc jxMd jyMd
    ;# xmm11: jzMa jzMb jzMc jzMd

    movaps xmm0, xmm1
    movaps xmm2, xmm1
    movaps xmm3, xmm4
    movaps xmm5, xmm4
    movaps xmm6, xmm7
    movaps xmm8, xmm7

	mulps xmm0, [rsp + nb204_dxH1M]
	mulps xmm1, [rsp + nb204_dyH1M]
	mulps xmm2, [rsp + nb204_dzH1M]
	mulps xmm3, [rsp + nb204_dxH2M]
	mulps xmm4, [rsp + nb204_dyH2M]
	mulps xmm5, [rsp + nb204_dzH2M]
	mulps xmm6, [rsp + nb204_dxMM]
	mulps xmm7, [rsp + nb204_dyMM]
	mulps xmm8, [rsp + nb204_dzMM]

    movaps xmm13, xmm0
    movaps xmm14, xmm1
    addps xmm11, xmm2
    addps xmm0, [rsp + nb204_fixH1]
    addps xmm1, [rsp + nb204_fiyH1]
    addps xmm2, [rsp + nb204_fizH1]

    addps xmm13, xmm3
    addps xmm14, xmm4
    addps xmm11, xmm5
    addps xmm3, [rsp + nb204_fixH2]
    addps xmm4, [rsp + nb204_fiyH2]
    addps xmm5, [rsp + nb204_fizH2]

    addps xmm13, xmm6
    addps xmm14, xmm7
    addps xmm11, xmm8
    addps xmm6, [rsp + nb204_fixM]
    addps xmm7, [rsp + nb204_fiyM]
    addps xmm8, [rsp + nb204_fizM]

    movaps [rsp + nb204_fixH1], xmm0
    movaps [rsp + nb204_fiyH1], xmm1
    movaps [rsp + nb204_fizH1], xmm2
    movaps [rsp + nb204_fixH2], xmm3
    movaps [rsp + nb204_fiyH2], xmm4
    movaps [rsp + nb204_fizH2], xmm5
    movaps [rsp + nb204_fixM], xmm6
    movaps [rsp + nb204_fiyM], xmm7
    movaps [rsp + nb204_fizM], xmm8
    
    ;# xmm0 = fMx
    ;# xmm1 = fMy
    ;# xmm2 = fMz
    movaps xmm15, xmm13
    unpcklps xmm13, xmm14
    unpckhps xmm15, xmm14
    
    addps xmm9, xmm13
    addps xmm10, xmm15

    movhlps  xmm12, xmm11 ;# fMzc fMzd
    
    movlps [rdi + rax*4 + 36], xmm9
    movhps [rdi + rbx*4 + 36], xmm9
    movlps [rdi + rcx*4 + 36], xmm10
    movhps [rdi + rdx*4 + 36], xmm10
    movss  [rdi + rax*4 + 44], xmm11
    movss  [rdi + rcx*4 + 44], xmm12
    shufps xmm11, xmm11, 1
    shufps xmm12, xmm12, 1
    movss  [rdi + rbx*4 + 44], xmm11
    movss  [rdi + rdx*4 + 44], xmm12
	
	;# should we do one more iteration? 
	sub dword ptr [rsp + nb204_innerk],  4
	jl    .nb204_single_check
	jmp   .nb204_unroll_loop
.nb204_single_check:
	add dword ptr [rsp + nb204_innerk],  4
	jnz   .nb204_single_loop
	jmp   .nb204_updateouterdata
.nb204_single_loop:
	mov   rdx, [rsp + nb204_innerjjnr] 	;# pointer to jjnr[k] 
	mov   eax, [rdx]	
	add qword ptr [rsp + nb204_innerjjnr],  4	

	mov rsi, [rbp + nb204_pos]
	lea   rax, [rax + rax*2]  

	;# fetch j coordinates 
	xorps xmm0, xmm0
	xorps xmm1, xmm1
	xorps xmm2, xmm2
	movss xmm0, [rsi + rax*4 + 36]		;# jxM  -  -  -
	movss xmm1, [rsi + rax*4 + 40]		;# jyM  -  -  -
	movss xmm2, [rsi + rax*4 + 44]		;# jzM  -  -  -  

	movlps xmm6, [rsi + rax*4 + 12]		;# xmm6 = jxH1 jyH1   -    -
	movss  xmm7, [rsi + rax*4 + 20]		;# xmm7 = jzH1   -    -    - 
	movhps xmm6, [rsi + rax*4 + 24]		;# xmm6 = jxH1 jyH1 jxH2 jyH2
	movss  xmm5, [rsi + rax*4 + 32]		;# xmm5 = jzH2   -    -    -
	
	;# have all coords, time for some shuffling.

	shufps xmm6, xmm6, 216 ;# 11011000	;# xmm6 = jxH1 jxH2 jyH1 jyH2 
	unpcklps xmm7, xmm5			;# xmm7 = jzH1 jzH2   -    -

	movlhps xmm0, xmm6			;# xmm0 = jxM   0   jxH1 jxH2 
	shufps  xmm1, xmm6, 228 ;# 11100100	;# xmm1 = jyM   0   jyH1 jyH2 
	shufps  xmm2, xmm7, 68  ;# 01000100	;# xmm2 = jzM   0   jzH1 jzH2

	;# store all j coordinates in jM 
	movaps [rsp + nb204_jxM], xmm0
	movaps [rsp + nb204_jyM], xmm1
	movaps [rsp + nb204_jzM], xmm2
	subps  xmm0, [rsp + nb204_ixM]
	subps  xmm1, [rsp + nb204_iyM]
	subps  xmm2, [rsp + nb204_izM]
	movaps [rsp + nb204_dxMM], xmm0
	movaps [rsp + nb204_dyMM], xmm1
	movaps [rsp + nb204_dzMM], xmm2

	mulps xmm0, xmm0
	mulps xmm1, xmm1
	mulps xmm2, xmm2
	addps xmm0, xmm1
	addps xmm0, xmm2	;# have rsq in xmm0 

	movaps xmm6, xmm0
	
	;# do invsqrt 
	rsqrtps xmm1, xmm0
	mulps   xmm6, [rsp + nb204_krf] ;# xmm6=krsq 
	movaps  xmm2, xmm1
	movaps  xmm7, xmm6 	;# xmm7=krsq 
	mulps   xmm1, xmm1
	movaps  xmm3, [rsp + nb204_three]
	mulps   xmm1, xmm0
	subps   xmm3, xmm1
	mulps   xmm3, xmm2
	mulps   xmm3, [rsp + nb204_half] ;# rinv iO - j water 
	
	addps   xmm6, xmm3	;# xmm6=rinv+ krsq 
	mulps   xmm7, [rsp + nb204_two]
	subps   xmm6, [rsp + nb204_crf] ;# xmm6=rinv+ krsq-crf 
	
	xorps   xmm1, xmm1
	movaps  xmm0, xmm3
	subps   xmm3, xmm7	;# xmm3=rinv-2*krsq 
	xorps   xmm4, xmm4
	mulps   xmm0, xmm0	;# xmm0=rinvsq 
	;# fetch charges to xmm4 (temporary) 
	movss   xmm4, [rsp + nb204_qqMM]
	movhps  xmm4, [rsp + nb204_qqMH]

	mulps xmm6, xmm4	;# vcoul  
	mulps xmm3, xmm4	;# coul part of fs  


	addps   xmm6, [rsp + nb204_vctot]
	mulps   xmm0, xmm3	;# total fscal 
	movaps  [rsp + nb204_vctot], xmm6	

	movaps  xmm1, xmm0
	movaps  xmm2, xmm0
	mulps   xmm0, [rsp + nb204_dxMM]
	mulps   xmm1, [rsp + nb204_dyMM]
	mulps   xmm2, [rsp + nb204_dzMM]
	
	;# initial update for j forces 
	xorps   xmm3, xmm3
	xorps   xmm4, xmm4
	xorps   xmm5, xmm5
	addps   xmm3, xmm0
	addps   xmm4, xmm1
	addps   xmm5, xmm2
	movaps  [rsp + nb204_fjxM], xmm3
	movaps  [rsp + nb204_fjyM], xmm4
	movaps  [rsp + nb204_fjzM], xmm5
	addps   xmm0, [rsp + nb204_fixM]
	addps   xmm1, [rsp + nb204_fiyM]
	addps   xmm2, [rsp + nb204_fizM]
	movaps  [rsp + nb204_fixM], xmm0
	movaps  [rsp + nb204_fiyM], xmm1
	movaps  [rsp + nb204_fizM], xmm2

	
	;# done with i M Now do i H1 & H2 simultaneously first get i particle coords: 
    movaps  xmm0, [rsp + nb204_jxM]
    movaps  xmm1, [rsp + nb204_jyM]
    movaps  xmm2, [rsp + nb204_jzM]
    movaps  xmm3, xmm0
    movaps  xmm4, xmm1
    movaps  xmm5, xmm2
 
	subps   xmm0, [rsp + nb204_ixH1]
	subps   xmm1, [rsp + nb204_iyH1]
	subps   xmm2, [rsp + nb204_izH1]
	subps   xmm3, [rsp + nb204_ixH2] 
	subps   xmm4, [rsp + nb204_iyH2]
	subps   xmm5, [rsp + nb204_izH2] 

	movaps [rsp + nb204_dxH1M], xmm0
	movaps [rsp + nb204_dyH1M], xmm1
	movaps [rsp + nb204_dzH1M], xmm2
	movaps [rsp + nb204_dxH2M], xmm3
	movaps [rsp + nb204_dyH2M], xmm4
	movaps [rsp + nb204_dzH2M], 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 rsqH1 in xmm0 
	addps xmm4, xmm5	;# have rsqH2 in xmm4 
	
	;# do invsqrt 
	rsqrtps xmm1, xmm0
	rsqrtps xmm5, xmm4
	movaps  xmm2, xmm1
	movaps  xmm6, xmm5
	mulps   xmm1, xmm1
	mulps   xmm5, xmm5
	movaps  xmm3, [rsp + nb204_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 + nb204_half] ;# rinv H1 - j water 
	mulps   xmm7, [rsp + nb204_half] ;# rinv H2 - j water  

	mulps xmm0, [rsp + nb204_krf] ;# krsq 
	mulps xmm4, [rsp + nb204_krf] ;# krsq  

	;# assemble charges in xmm6 
	xorps   xmm6, xmm6
	movss   xmm6, [rsp + nb204_qqMH]
	movhps  xmm6, [rsp + nb204_qqHH]
	movaps  xmm1, xmm0
	movaps  xmm5, xmm4
	addps   xmm0, xmm3	;# krsq+ rinv 
	addps   xmm4, xmm7	;# krsq+ rinv 
	subps   xmm0, [rsp + nb204_crf]
	subps   xmm4, [rsp + nb204_crf]
	mulps   xmm1, [rsp + nb204_two]
	mulps   xmm5, [rsp + nb204_two]
	mulps   xmm0, xmm6	;# vcoul 
	mulps   xmm4, xmm6	;# vcoul 
	addps   xmm4, xmm0		
	addps   xmm4, [rsp + nb204_vctot]
	movaps  [rsp + nb204_vctot], xmm4
	movaps  xmm0, xmm3
	movaps  xmm4, xmm7
	mulps   xmm3, xmm3
	mulps   xmm7, xmm7
	subps   xmm0, xmm1
	subps   xmm4, xmm5
	mulps   xmm0, xmm6
	mulps   xmm4, xmm6
	mulps   xmm0, xmm3	;# fscal 
	mulps   xmm7, xmm4	;# fscal 
	
	movaps  xmm1, xmm0
	movaps  xmm2, xmm0
	mulps   xmm0, [rsp + nb204_dxH1M]
	mulps   xmm1, [rsp + nb204_dyH1M]
	mulps   xmm2, [rsp + nb204_dzH1M]
	;# update forces H1 - j water 
	movaps  xmm3, [rsp + nb204_fjxM]
	movaps  xmm4, [rsp + nb204_fjyM]
	movaps  xmm5, [rsp + nb204_fjzM]
	addps   xmm3, xmm0
	addps   xmm4, xmm1
	addps   xmm5, xmm2
	movaps  [rsp + nb204_fjxM], xmm3
	movaps  [rsp + nb204_fjyM], xmm4
	movaps  [rsp + nb204_fjzM], xmm5
	addps   xmm0, [rsp + nb204_fixH1]
	addps   xmm1, [rsp + nb204_fiyH1]
	addps   xmm2, [rsp + nb204_fizH1]
	movaps  [rsp + nb204_fixH1], xmm0
	movaps  [rsp + nb204_fiyH1], xmm1
	movaps  [rsp + nb204_fizH1], xmm2
	;# do forces H2 - j water 
	movaps xmm0, xmm7
	movaps xmm1, xmm7
	movaps xmm2, xmm7
	mulps   xmm0, [rsp + nb204_dxH2M]
	mulps   xmm1, [rsp + nb204_dyH2M]
	mulps   xmm2, [rsp + nb204_dzH2M]
	movaps  xmm3, [rsp + nb204_fjxM]
	movaps  xmm4, [rsp + nb204_fjyM]