nb_kernel104_x86_64_sse.intel_syntax.s

来自「最著名最快的分子模拟软件」· S 代码 · 共 2,223 行 · 第 1/5 页

    subps   xmm11, xmm7 ;# 3-rsq*lu*lu

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

	movaps  xmm0, [rsp + nb104_half]
	mulps   xmm9, xmm0  ;# rinvH1M 
	mulps   xmm10, xmm0 ;# rinvH2M
    mulps   xmm11, xmm0 ;# rinvMM
	
	;# M interactions 
    movaps xmm0, xmm9
    movaps xmm1, xmm10
    movaps xmm2, xmm11
    mulps  xmm9, xmm9
    mulps  xmm10, xmm10
    mulps  xmm11, xmm11
    mulps  xmm0, [rsp + nb104_qqMH] 
    mulps  xmm1, [rsp + nb104_qqMH] 
    mulps  xmm2, [rsp + nb104_qqMM] 
    mulps  xmm9, xmm0
    mulps  xmm10, xmm1
    mulps  xmm11, xmm2
    
    addps xmm0, [rsp + nb104_vctot] 
    addps xmm1, xmm2
    addps xmm0, xmm1
    movaps [rsp + nb104_vctot], xmm0
    
	;# move j M forces to local temp variables 
    movlps xmm0, [rdi + rax*4 + 36] ;# jxMa jyMa  -   -
    movlps xmm1, [rdi + rcx*4 + 36] ;# jxMc jyMc  -   -
    movhps xmm0, [rdi + rbx*4 + 36] ;# jxMa jyMa jxMb jyMb 
    movhps xmm1, [rdi + rdx*4 + 36] ;# jxMc jyMc jxMd jyMd 

    movss  xmm2, [rdi + rax*4 + 44] ;# jzMa  -  -  -
    movss  xmm3, [rdi + rcx*4 + 44] ;# jzMc  -  -  -
    movss  xmm7, [rdi + rbx*4 + 44] ;# jzMb  -  -  -
    movss  xmm8, [rdi + rdx*4 + 44] ;# jzMd  -  -  -
    movlhps xmm2, xmm7 ;# jzMa  -  jzMb  -
    movlhps xmm3, xmm8 ;# jzMc  -  jzMd -
    
    shufps xmm2, xmm3,  136  ;# 10001000 => jzMa jzMb jzMc jzMd

    ;# xmm0: jxMa jyMa jxMb jyMb 
    ;# xmm1: jxMc jyMc jxMd jyMd
    ;# xmm2: jzMa jzMb jzMc jzMd

    movaps xmm7, xmm9
    movaps xmm8, xmm9
    movaps xmm13, xmm11
    movaps xmm14, xmm11
    movaps xmm15, xmm11
    movaps xmm11, xmm10
    movaps xmm12, xmm10

	mulps xmm7, [rsp + nb104_dxH1M]
	mulps xmm8, [rsp + nb104_dyH1M]
	mulps xmm9, [rsp + nb104_dzH1M]
	mulps xmm10, [rsp + nb104_dxH2M]
	mulps xmm11, [rsp + nb104_dyH2M]
	mulps xmm12, [rsp + nb104_dzH2M]
	mulps xmm13, [rsp + nb104_dxMM]
	mulps xmm14, [rsp + nb104_dyMM]
	mulps xmm15, [rsp + nb104_dzMM]

    movaps xmm3, xmm7
    movaps xmm4, xmm8
    addps xmm2, xmm9
    addps xmm7, [rsp + nb104_fixH1]
    addps xmm8, [rsp + nb104_fiyH1]
    addps xmm9, [rsp + nb104_fizH1]

    addps xmm3, xmm10
    addps xmm4, xmm11
    addps xmm2, xmm12
    addps xmm10, [rsp + nb104_fixH2]
    addps xmm11, [rsp + nb104_fiyH2]
    addps xmm12, [rsp + nb104_fizH2]

    addps xmm3, xmm13
    addps xmm4, xmm14
    addps xmm2, xmm15
    addps xmm13, [rsp + nb104_fixM]
    addps xmm14, [rsp + nb104_fiyM]
    addps xmm15, [rsp + nb104_fizM]
    
    movaps [rsp + nb104_fixH1], xmm7
    movaps [rsp + nb104_fiyH1], xmm8
    movaps [rsp + nb104_fizH1], xmm9
    movaps [rsp + nb104_fixH2], xmm10
    movaps [rsp + nb104_fiyH2], xmm11
    movaps [rsp + nb104_fizH2], xmm12
    movaps [rsp + nb104_fixM], xmm13
    movaps [rsp + nb104_fiyM], xmm14
    movaps [rsp + nb104_fizM], xmm15

    ;# xmm3 = fMx , xmm4 = fMy
    movaps xmm5, xmm3
    unpcklps xmm3, xmm4
    unpckhps xmm5, xmm4
    
    addps xmm0, xmm3
    addps xmm1, xmm5

    movhlps  xmm3, xmm2 ;# fMzc fMzd
    
    movlps [rdi + rax*4 + 36], xmm0
    movhps [rdi + rbx*4 + 36], xmm0
    movlps [rdi + rcx*4 + 36], xmm1
    movhps [rdi + rdx*4 + 36], xmm1
    movss  [rdi + rax*4 + 44], xmm2
    movss  [rdi + rcx*4 + 44], xmm3
    shufps xmm2, xmm2, 1
    shufps xmm3, xmm3, 1
    movss  [rdi + rbx*4 + 44], xmm2
    movss  [rdi + rdx*4 + 44], xmm3

	;# should we do one more iteration? 
	sub dword ptr [rsp + nb104_innerk],  4
	jl    .nb104_single_check
	jmp   .nb104_unroll_loop
.nb104_single_check:
	add dword ptr [rsp + nb104_innerk],  4
	jnz   .nb104_single_loop
	jmp   .nb104_updateouterdata
.nb104_single_loop:
	mov   rdx, [rsp + nb104_innerjjnr] 	;# pointer to jjnr[k] 
	mov   eax, [rdx]	
	add qword ptr [rsp + nb104_innerjjnr],  4	

	mov rsi, [rbp + nb104_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 + nb104_jxM], xmm0
	movaps [rsp + nb104_jyM], xmm1
	movaps [rsp + nb104_jzM], xmm2
	subps  xmm0, [rsp + nb104_ixM]
	subps  xmm1, [rsp + nb104_iyM]
	subps  xmm2, [rsp + nb104_izM]
	movaps [rsp + nb104_dxMM], xmm0
	movaps [rsp + nb104_dyMM], xmm1
	movaps [rsp + nb104_dzMM], xmm2
	mulps xmm0, xmm0
	mulps xmm1, xmm1
	mulps xmm2, xmm2
	addps xmm0, xmm1
	addps xmm0, xmm2	;# have rsq in xmm0 
	
	;# do invsqrt 
	rsqrtps xmm1, xmm0
	movaps  xmm2, xmm1	
	mulps   xmm1, xmm1
	movaps  xmm3, [rsp + nb104_three]
	mulps   xmm1, xmm0
	subps   xmm3, xmm1
	mulps   xmm3, xmm2
	mulps   xmm3, [rsp + nb104_half] ;# rinv iM- j water 

	xorps   xmm1, xmm1
	movaps  xmm0, xmm3
	xorps   xmm4, xmm4
	mulps   xmm0, xmm0	;# xmm0=rinvsq
	
	;# fetch charges to xmm4
	movss   xmm4, [rsp + nb104_qqMM] 
	movhps  xmm4, [rsp + nb104_qqMH]
	
	mulps   xmm3, xmm4	;# xmm3=vcoul 
	mulps   xmm0, xmm3	;# total fscal 
	addps   xmm3, [rsp + nb104_vctot]
	movaps  [rsp + nb104_vctot], xmm3	

	movaps  xmm1, xmm0
	movaps  xmm2, xmm0
	mulps   xmm0, [rsp + nb104_dxMM]
	mulps   xmm1, [rsp + nb104_dyMM]
	mulps   xmm2, [rsp + nb104_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 + nb104_fjxM], xmm3
	movaps  [rsp + nb104_fjyM], xmm4
	movaps  [rsp + nb104_fjzM], xmm5
	addps   xmm0, [rsp + nb104_fixM]
	addps   xmm1, [rsp + nb104_fiyM]
	addps   xmm2, [rsp + nb104_fizM]
	movaps  [rsp + nb104_fixM], xmm0
	movaps  [rsp + nb104_fiyM], xmm1
	movaps  [rsp + nb104_fizM], xmm2

	
	;# done with i M Now do i H1 & H2 simultaneously first get i particle coords: 
    movaps  xmm0, [rsp + nb104_jxM]
    movaps  xmm1, [rsp + nb104_jyM]
    movaps  xmm2, [rsp + nb104_jzM]
    movaps  xmm3, xmm0
    movaps  xmm4, xmm1
    movaps  xmm5, xmm2
 
	subps   xmm0, [rsp + nb104_ixH1]
	subps   xmm1, [rsp + nb104_iyH1]
	subps   xmm2, [rsp + nb104_izH1]
	subps   xmm3, [rsp + nb104_ixH2] 
	subps   xmm4, [rsp + nb104_iyH2]
	subps   xmm5, [rsp + nb104_izH2] 
    
	movaps [rsp + nb104_dxH1M], xmm0
	movaps [rsp + nb104_dyH1M], xmm1
	movaps [rsp + nb104_dzH1M], xmm2
	movaps [rsp + nb104_dxH2M], xmm3
	movaps [rsp + nb104_dyH2M], xmm4
	movaps [rsp + nb104_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 + nb104_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 + nb104_half] ;# rinv H1 - j water 
	mulps   xmm7, [rsp + nb104_half] ;# rinv H2 - j water  

	;# assemble charges in xmm6 
	xorps   xmm6, xmm6
	movss   xmm6, [rsp + nb104_qqMH]
	movhps  xmm6, [rsp + nb104_qqHH]
	
	;# do coulomb interaction 
	movaps  xmm0, xmm3
	movaps  xmm4, xmm7
	mulps   xmm0, xmm0	;# rinvsq 
	mulps   xmm4, xmm4	;# rinvsq 
	mulps   xmm3, xmm6	;# vcoul 
	mulps   xmm7, xmm6	;# vcoul 
	movaps  xmm2, xmm3
	addps   xmm2, xmm7	;# total vcoul 
	mulps   xmm0, xmm3	;# fscal 
	
	addps   xmm2, [rsp + nb104_vctot]
	mulps   xmm7, xmm4	;# fscal 
	movaps  [rsp + nb104_vctot], xmm2
	movaps  xmm1, xmm0
	movaps  xmm2, xmm0
	mulps   xmm0, [rsp + nb104_dxH1M]
	mulps   xmm1, [rsp + nb104_dyH1M]
	mulps   xmm2, [rsp + nb104_dzH1M]
	;# update forces H1 - j water 
	movaps  xmm3, [rsp + nb104_fjxM]
	movaps  xmm4, [rsp + nb104_fjyM]
	movaps  xmm5, [rsp + nb104_fjzM]
	addps   xmm3, xmm0
	addps   xmm4, xmm1
	addps   xmm5, xmm2
	movaps  [rsp + nb104_fjxM], xmm3
	movaps  [rsp + nb104_fjyM], xmm4
	movaps  [rsp + nb104_fjzM], xmm5
	addps   xmm0, [rsp + nb104_fixH1]
	addps   xmm1, [rsp + nb104_fiyH1]
	addps   xmm2, [rsp + nb104_fizH1]
	movaps  [rsp + nb104_fixH1], xmm0
	movaps  [rsp + nb104_fiyH1], xmm1
	movaps  [rsp + nb104_fizH1], xmm2
	;# do forces H2 - j water 
	movaps xmm0, xmm7
	movaps xmm1, xmm7
	movaps xmm2, xmm7
	mulps   xmm0, [rsp + nb104_dxH2M]
	mulps   xmm1, [rsp + nb104_dyH2M]
	mulps   xmm2, [rsp + nb104_dzH2M]
	movaps  xmm3, [rsp + nb104_fjxM]
	movaps  xmm4, [rsp + nb104_fjyM]
	movaps  xmm5, [rsp + nb104_fjzM]
	addps   xmm3, xmm0
	addps   xmm4, xmm1
	addps   xmm5, xmm2
	mov     rsi, [rbp + nb104_faction]
	movaps  [rsp + nb104_fjxM], xmm3
	movaps  [rsp + nb104_fjyM], xmm4
	movaps  [rsp + nb104_fjzM], xmm5
	addps   xmm0, [rsp + nb104_fixH2]
	addps   xmm1, [rsp + nb104_fiyH2]
	addps   xmm2, [rsp + nb104_fizH2]
	movaps  [rsp + nb104_fixH2], xmm0
	movaps  [rsp + nb104_fiyH2], xmm1
	movaps  [rsp + nb104_fizH2], xmm2

	;# update j water forces from local variables 
	movlps  xmm0, [rsi + rax*4 + 36]
	movlps  xmm1, [rsi + rax*4 + 12]
	movhps  xmm1, [rsi + rax*4 + 24]
	movaps  xmm3, [rsp + nb104_fjxM]
	movaps  xmm4, [rsp + nb104_fjyM]
	movaps  xmm5, [rsp + nb104_fjzM]
	movaps  xmm6, xmm5
	movaps  xmm7, xmm5
	shufps  xmm6, xmm6, 2 ;# 00000010
	shufps  xmm7, xmm7, 3 ;# 00000011
	addss   xmm5, [rsi + rax*4 + 44]
	addss   xmm6, [rsi + rax*4 + 20]
	addss   xmm7, [rsi + rax*4 + 32]
	movss   [rsi + rax*4 + 44], xmm5
	movss   [rsi + rax*4 + 20], xmm6
	movss   [rsi + rax*4 + 32], xmm7
	movaps   xmm5, xmm3
	unpcklps xmm3, xmm4
	unpckhps xmm5, xmm4
	addps    xmm0, xmm3
	addps    xmm1, xmm5
	movlps  [rsi + rax*4 + 36], xmm0 
	movlps  [rsi + rax*4 + 12], xmm1 
	movhps  [rsi + rax*4 + 24], xmm1 
	
	dec   dword ptr [rsp + nb104_innerk]
	jz    .nb104_updateouterdata
	jmp   .nb104_single_loop
.nb104_updateouterdata:
	mov   ecx, [rsp + nb104_ii3]
	mov   rdi, [rbp + nb104_faction]
	mov   rsi, [rbp + nb104_fshift]
	mov   edx, [rsp + nb104_is3]

	;# accumulate H1i forces in xmm0, xmm1, xmm2 
	movaps xmm0, [rsp + nb104_fixH1]
	movaps xmm1, [rsp + nb104_fiyH1] 
	movaps xmm2, [rsp + nb104_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 
	movaps xmm6, xmm0
	movss xmm7, xmm2
	movlhps xmm6, xmm1
	shufps  xmm6, xmm6, 8 ;# 00001000	

	;# accumulate H2i forces in xmm0, xmm1, xmm2 
	movaps xmm0, [rsp + nb104_fixH2]
	movaps xmm1, [rsp + nb104_fiyH2]
	movaps xmm2, [rsp + nb104_fizH2]

	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 + 24]
	movss  xmm4, [rdi + rcx*4 + 28]
	movss  xmm5, [rdi + rcx*4 + 32]
	subss  xmm3, xmm0
	subss  xmm4, xmm1
	subss  xmm5, xmm2
	movss  [rdi + rcx*4 + 24], xmm3
	movss  [rdi + rcx*4 + 28], xmm4
	movss  [rdi + rcx*4 + 32], xmm5

	;# accumulate force in xmm6/xmm7 for fshift 
	addss xmm7, xmm2