nb_kernel102_x86_64_sse.intel_syntax.s

来自「最著名最快的分子模拟软件」· S 代码 · 共 2,233 行 · 第 1/5 页

	movss  xmm3, [rsi + rcx*4 + 8]
	movss  xmm4, [rsi + rcx*4 + 20]
	movss  xmm5, [rsi + rcx*4 + 32]

	movhps xmm0, [rsi + rbx*4 + 4]
	movhps xmm1, [rsi + rbx*4 + 16]
	movhps xmm2, [rsi + rbx*4 + 28]
	
	movhps xmm3, [rsi + rdx*4 + 4]
	movhps xmm4, [rsi + rdx*4 + 16]
	movhps xmm5, [rsi + rdx*4 + 28]
	
	shufps xmm0, xmm3, 204  ;# 11001100
	shufps xmm1, xmm4, 204  ;# 11001100
	shufps xmm2, xmm5, 204  ;# 11001100
	movaps [rsp + nb102nf_jzO],  xmm0
	movaps [rsp + nb102nf_jzH1],  xmm1
	movaps [rsp + nb102nf_jzH2],  xmm2

	movaps xmm0, [rsp + nb102nf_ixO]
	movaps xmm1, [rsp + nb102nf_iyO]
	movaps xmm2, [rsp + nb102nf_izO]
	movaps xmm3, [rsp + nb102nf_ixO]
	movaps xmm4, [rsp + nb102nf_iyO]
	movaps xmm5, [rsp + nb102nf_izO]
	subps  xmm0, [rsp + nb102nf_jxO]
	subps  xmm1, [rsp + nb102nf_jyO]
	subps  xmm2, [rsp + nb102nf_jzO]
	subps  xmm3, [rsp + nb102nf_jxH1]
	subps  xmm4, [rsp + nb102nf_jyH1]
	subps  xmm5, [rsp + nb102nf_jzH1]
	
	mulps  xmm0, xmm0
	mulps  xmm1, xmm1
	mulps  xmm2, xmm2

	mulps  xmm3, xmm3
	mulps  xmm4, xmm4
	mulps  xmm5, xmm5
	addps  xmm0, xmm1
	addps  xmm0, xmm2
	addps  xmm3, xmm4
	addps  xmm3, xmm5
	movaps [rsp + nb102nf_rsqOO], xmm0
	movaps [rsp + nb102nf_rsqOH1], xmm3

	movaps xmm0, [rsp + nb102nf_ixO]
	movaps xmm1, [rsp + nb102nf_iyO]
	movaps xmm2, [rsp + nb102nf_izO]
	movaps xmm3, [rsp + nb102nf_ixH1]
	movaps xmm4, [rsp + nb102nf_iyH1]
	movaps xmm5, [rsp + nb102nf_izH1]
	subps  xmm0, [rsp + nb102nf_jxH2]
	subps  xmm1, [rsp + nb102nf_jyH2]
	subps  xmm2, [rsp + nb102nf_jzH2]
	subps  xmm3, [rsp + nb102nf_jxO]
	subps  xmm4, [rsp + nb102nf_jyO]
	subps  xmm5, [rsp + nb102nf_jzO]

	mulps  xmm0, xmm0
	mulps  xmm1, xmm1
	mulps  xmm2, xmm2

	mulps  xmm3, xmm3
	mulps  xmm4, xmm4
	mulps  xmm5, xmm5
	addps  xmm0, xmm1
	addps  xmm0, xmm2
	addps  xmm3, xmm4
	addps  xmm3, xmm5
	movaps [rsp + nb102nf_rsqOH2], xmm0
	movaps [rsp + nb102nf_rsqH1O], xmm3

	movaps xmm0, [rsp + nb102nf_ixH1]
	movaps xmm1, [rsp + nb102nf_iyH1]
	movaps xmm2, [rsp + nb102nf_izH1]
	movaps xmm3, [rsp + nb102nf_ixH1]
	movaps xmm4, [rsp + nb102nf_iyH1]
	movaps xmm5, [rsp + nb102nf_izH1]
	subps  xmm0, [rsp + nb102nf_jxH1]
	subps  xmm1, [rsp + nb102nf_jyH1]
	subps  xmm2, [rsp + nb102nf_jzH1]
	subps  xmm3, [rsp + nb102nf_jxH2]
	subps  xmm4, [rsp + nb102nf_jyH2]
	subps  xmm5, [rsp + nb102nf_jzH2]

	mulps  xmm0, xmm0
	mulps  xmm1, xmm1
	mulps  xmm2, xmm2

	mulps  xmm3, xmm3
	mulps  xmm4, xmm4
	mulps  xmm5, xmm5
	addps  xmm0, xmm1
	addps  xmm0, xmm2
	addps  xmm3, xmm4
	addps  xmm3, xmm5
	movaps [rsp + nb102nf_rsqH1H1], xmm0
	movaps [rsp + nb102nf_rsqH1H2], xmm3

	movaps xmm0, [rsp + nb102nf_ixH2]
	movaps xmm1, [rsp + nb102nf_iyH2]
	movaps xmm2, [rsp + nb102nf_izH2]
	movaps xmm3, [rsp + nb102nf_ixH2]
	movaps xmm4, [rsp + nb102nf_iyH2]
	movaps xmm5, [rsp + nb102nf_izH2]
	subps  xmm0, [rsp + nb102nf_jxO]
	subps  xmm1, [rsp + nb102nf_jyO]
	subps  xmm2, [rsp + nb102nf_jzO]
	subps  xmm3, [rsp + nb102nf_jxH1]
	subps  xmm4, [rsp + nb102nf_jyH1]
	subps  xmm5, [rsp + nb102nf_jzH1]

	mulps  xmm0, xmm0
	mulps  xmm1, xmm1
	mulps  xmm2, xmm2

	mulps  xmm3, xmm3
	mulps  xmm4, xmm4
	mulps  xmm5, xmm5
	addps  xmm0, xmm1
	addps  xmm0, xmm2
	addps  xmm4, xmm3
	addps  xmm4, xmm5
	movaps [rsp + nb102nf_rsqH2O], xmm0
	movaps [rsp + nb102nf_rsqH2H1], xmm4

	movaps xmm0, [rsp + nb102nf_ixH2]
	movaps xmm1, [rsp + nb102nf_iyH2]
	movaps xmm2, [rsp + nb102nf_izH2]
	subps  xmm0, [rsp + nb102nf_jxH2]
	subps  xmm1, [rsp + nb102nf_jyH2]
	subps  xmm2, [rsp + nb102nf_jzH2]

	mulps xmm0, xmm0
	mulps xmm1, xmm1
	mulps xmm2, xmm2
	addps xmm0, xmm1
	addps xmm0, xmm2
	movaps [rsp + nb102nf_rsqH2H2], xmm0
	
	;# start doing invsqrt use rsq values in xmm0, xmm4 
	rsqrtps xmm1, xmm0
	rsqrtps xmm5, xmm4
	movaps  xmm2, xmm1
	movaps  xmm6, xmm5
	mulps   xmm1, xmm1
	mulps   xmm5, xmm5
	movaps  xmm3, [rsp + nb102nf_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 + nb102nf_half] ;# rinvH2H2 
	mulps   xmm7, [rsp + nb102nf_half] ;# rinvH2H1 
	movaps  [rsp + nb102nf_rinvH2H2], xmm3
	movaps  [rsp + nb102nf_rinvH2H1], xmm7
	
	rsqrtps xmm1, [rsp + nb102nf_rsqOO]
	rsqrtps xmm5, [rsp + nb102nf_rsqOH1]
	movaps  xmm2, xmm1
	movaps  xmm6, xmm5
	mulps   xmm1, xmm1
	mulps   xmm5, xmm5
	movaps  xmm3, [rsp + nb102nf_three]
	movaps  xmm7, xmm3
	mulps   xmm1, [rsp + nb102nf_rsqOO]
	mulps   xmm5, [rsp + nb102nf_rsqOH1]
	subps   xmm3, xmm1
	subps   xmm7, xmm5
	mulps   xmm3, xmm2
	mulps   xmm7, xmm6
	mulps   xmm3, [rsp + nb102nf_half] 
	mulps   xmm7, [rsp + nb102nf_half]
	movaps  [rsp + nb102nf_rinvOO], xmm3
	movaps  [rsp + nb102nf_rinvOH1], xmm7
	
	rsqrtps xmm1, [rsp + nb102nf_rsqOH2]
	rsqrtps xmm5, [rsp + nb102nf_rsqH1O]
	movaps  xmm2, xmm1
	movaps  xmm6, xmm5
	mulps   xmm1, xmm1
	mulps   xmm5, xmm5
	movaps  xmm3, [rsp + nb102nf_three]
	movaps  xmm7, xmm3
	mulps   xmm1, [rsp + nb102nf_rsqOH2]
	mulps   xmm5, [rsp + nb102nf_rsqH1O]
	subps   xmm3, xmm1
	subps   xmm7, xmm5
	mulps   xmm3, xmm2
	mulps   xmm7, xmm6
	mulps   xmm3, [rsp + nb102nf_half] 
	mulps   xmm7, [rsp + nb102nf_half]
	movaps  [rsp + nb102nf_rinvOH2], xmm3
	movaps  [rsp + nb102nf_rinvH1O], xmm7
	
	rsqrtps xmm1, [rsp + nb102nf_rsqH1H1]
	rsqrtps xmm5, [rsp + nb102nf_rsqH1H2]
	movaps  xmm2, xmm1
	movaps  xmm6, xmm5
	mulps   xmm1, xmm1
	mulps   xmm5, xmm5
	movaps  xmm3, [rsp + nb102nf_three]
	movaps  xmm7, xmm3
	mulps   xmm1, [rsp + nb102nf_rsqH1H1]
	mulps   xmm5, [rsp + nb102nf_rsqH1H2]
	subps   xmm3, xmm1
	subps   xmm7, xmm5
	mulps   xmm3, xmm2
	mulps   xmm7, xmm6
	mulps   xmm3, [rsp + nb102nf_half] 
	mulps   xmm7, [rsp + nb102nf_half]
	movaps  [rsp + nb102nf_rinvH1H1], xmm3
	movaps  [rsp + nb102nf_rinvH1H2], xmm7
	
	rsqrtps xmm1, [rsp + nb102nf_rsqH2O]
	movaps  xmm2, xmm1
	mulps   xmm1, xmm1
	movaps  xmm3, [rsp + nb102nf_three]
	mulps   xmm1, [rsp + nb102nf_rsqH2O]
	subps   xmm3, xmm1
	mulps   xmm3, xmm2
	mulps   xmm3, [rsp + nb102nf_half] 
	movaps  [rsp + nb102nf_rinvH2O], xmm3

	;# sum OO pot in xmm0, OH in xmm1 HH in xmm2 
	movaps xmm0, [rsp + nb102nf_rinvOO]
	movaps xmm1, [rsp + nb102nf_rinvOH1]
	movaps xmm2, [rsp + nb102nf_rinvH1H1]
	addps  xmm1, [rsp + nb102nf_rinvOH2]
	addps  xmm2, [rsp + nb102nf_rinvH1H2]
	addps  xmm1, [rsp + nb102nf_rinvH1O]
	addps  xmm2, [rsp + nb102nf_rinvH2H1]
	addps  xmm1, [rsp + nb102nf_rinvH2O]
	addps  xmm2, [rsp + nb102nf_rinvH2H2]
	
	mulps  xmm0, [rsp + nb102nf_qqOO]
	mulps  xmm1, [rsp + nb102nf_qqOH]
	mulps  xmm2, [rsp + nb102nf_qqHH]
	addps  xmm0, [rsp + nb102nf_vctot]
	addps  xmm1, xmm2
	addps  xmm0, xmm1
	movaps 	[rsp + nb102nf_vctot], xmm0

	;# should we do one more iteration? 
	sub dword ptr [rsp + nb102nf_innerk],  4
	jl    .nb102nf_single_check
	jmp   .nb102nf_unroll_loop
.nb102nf_single_check:
	add dword ptr [rsp + nb102nf_innerk],  4
	jnz   .nb102nf_single_loop
	jmp   .nb102nf_updateouterdata
.nb102nf_single_loop:
	mov   rdx, [rsp + nb102nf_innerjjnr]     ;# pointer to jjnr[k] 
	mov   eax, [rdx]	
	add qword ptr [rsp + nb102nf_innerjjnr],  4	

	mov rsi, [rbp + nb102nf_pos]
	lea   rax, [rax + rax*2]  

	;# fetch j coordinates 
	xorps xmm3, xmm3
	xorps xmm4, xmm4
	xorps xmm5, xmm5
	
	movss xmm3, [rsi + rax*4]		;# jxO  -  -  -
	movss xmm4, [rsi + rax*4 + 4]		;# jyO  -  -  -
	movss xmm5, [rsi + rax*4 + 8]		;# jzO  -  -  -  

	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  xmm2, [rsi + rax*4 + 32]		;# xmm2 = jzH2   -    -    -
	
	;# have all coords, time for some shuffling.

	shufps xmm6, xmm6, 216 ;# 11011000	;# xmm6 = jxH1 jxH2 jyH1 jyH2 
	unpcklps xmm7, xmm2			;# xmm7 = jzH1 jzH2   -    -
	movaps  xmm0, [rsp + nb102nf_ixO]     
	movaps  xmm1, [rsp + nb102nf_iyO]
	movaps  xmm2, [rsp + nb102nf_izO]	
	movlhps xmm3, xmm6			;# xmm3 = jxO   0   jxH1 jxH2 
	shufps  xmm4, xmm6, 228 ;# 11100100	;# xmm4 = jyO   0   jyH1 jyH2 
	shufps  xmm5, xmm7, 68  ;# 01000100	;# xmm5 = jzO   0   jzH1 jzH2

	;# store all j coordinates in jO  
	movaps [rsp + nb102nf_jxO], xmm3
	movaps [rsp + nb102nf_jyO], xmm4
	movaps [rsp + nb102nf_jzO], xmm5
	subps  xmm0, xmm3
	subps  xmm1, xmm4
	subps  xmm2, xmm5
	
	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 + nb102nf_three]
	mulps   xmm1, xmm0
	subps   xmm3, xmm1
	mulps   xmm3, xmm2							
	mulps   xmm3, [rsp + nb102nf_half] ;# rinv iO - j water 

	xorps   xmm1, xmm1

	xorps   xmm4, xmm4

	;# fetch charges to xmm4 (temporary) 
	movss   xmm4, [rsp + nb102nf_qqOO]

	movhps  xmm4, [rsp + nb102nf_qqOH]

	mulps   xmm3, xmm4	;# xmm3=vcoul 

	addps   xmm3, [rsp + nb102nf_vctot]
	movaps  [rsp + nb102nf_vctot], xmm3	

	;# done with i O Now do i H1 & H2 simultaneously: 
	movaps  xmm0, [rsp + nb102nf_ixH1]
	movaps  xmm1, [rsp + nb102nf_iyH1]
	movaps  xmm2, [rsp + nb102nf_izH1]	
	movaps  xmm3, [rsp + nb102nf_ixH2] 
	movaps  xmm4, [rsp + nb102nf_iyH2] 
	movaps  xmm5, [rsp + nb102nf_izH2] 
	subps   xmm0, [rsp + nb102nf_jxO]
	subps   xmm1, [rsp + nb102nf_jyO]
	subps   xmm2, [rsp + nb102nf_jzO]
	subps   xmm3, [rsp + nb102nf_jxO]
	subps   xmm4, [rsp + nb102nf_jyO]
	subps   xmm5, [rsp + nb102nf_jzO]
	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 + nb102nf_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 + nb102nf_half] ;# rinv H1 - j water 
	mulps   xmm7, [rsp + nb102nf_half] ;# rinv H2 - j water  

	;# assemble charges in xmm6 
	xorps   xmm6, xmm6
	;# do coulomb interaction 
	movaps  xmm0, xmm3
	movss   xmm6, [rsp + nb102nf_qqOH]
	movaps  xmm4, xmm7
	movhps  xmm6, [rsp + nb102nf_qqHH]
	mulps   xmm3, xmm6	;# vcoul 
	mulps   xmm7, xmm6	;# vcoul 
	addps   xmm3, xmm7	;# total vcoul 
	addps   xmm3, [rsp + nb102nf_vctot]
	movaps  [rsp + nb102nf_vctot], xmm3
	
	dec   dword ptr [rsp + nb102nf_innerk]
	jz    .nb102nf_updateouterdata
	jmp   .nb102nf_single_loop
.nb102nf_updateouterdata:
	;# get n from stack
	mov esi, [rsp + nb102nf_n]
        ;# get group index for i particle 
        mov   rdx, [rbp + nb102nf_gid]      	;# base of gid[]
        mov   edx, [rdx + rsi*4]		;# ggid=gid[n]

	;# accumulate total potential energy and update it 
	movaps xmm7, [rsp + nb102nf_vctot]
	;# accumulate 
	movhlps xmm6, xmm7
	addps  xmm7, xmm6	;# pos 0-1 in xmm7 have the sum now 
	movaps xmm6, xmm7
	shufps xmm6, xmm6, 1
	addss  xmm7, xmm6		

	;# add earlier value from mem 
	mov   rax, [rbp + nb102nf_Vc]
	addss xmm7, [rax + rdx*4] 
	;# move back to mem 
	movss [rax + rdx*4], xmm7 	

        ;# finish if last 
        mov ecx, [rsp + nb102nf_nn1]
	;# esi already loaded with n
	inc esi
        sub ecx, esi
        jz .nb102nf_outerend

        ;# not last, iterate outer loop once more!  
        mov [rsp + nb102nf_n], esi
        jmp .nb102nf_outer
.nb102nf_outerend:
        ;# check if more outer neighborlists remain
        mov   ecx, [rsp + nb102nf_nri]
	;# esi already loaded with n above
        sub   ecx, esi
        jz .nb102nf_end
        ;# non-zero, do one more workunit
        jmp   .nb102nf_threadloop
.nb102nf_end:
	

	mov eax, [rsp + nb102nf_nouter]
	mov ebx, [rsp + nb102nf_ninner]
	mov rcx, [rbp + nb102nf_outeriter]
	mov rdx, [rbp + nb102nf_inneriter]
	mov [rcx], eax
	mov [rdx], ebx

	add rsp, 776
	emms
	
	pop rbx
	pop	rbp
	ret