nb_kernel133_x86_64_sse2.intel_syntax.s

来自「最著名最快的分子模拟软件」· S 代码 · 共 2,375 行 · 第 1/5 页

	movapd  xmm3, xmm2
	mulsd   xmm2, xmm2
	movapd  xmm1, [rsp + nb133_three]
	mulsd   xmm2, xmm5	;# rsq*lu*lu 
	subsd   xmm1, xmm2	;# 30-rsq*lu*lu 
	mulsd   xmm1, xmm3	;# lu*(3-rsq*lu*lu) 
	mulsd   xmm1, [rsp + nb133_half] ;# iter1 ( new lu) 

	movapd xmm3, xmm1
	mulsd xmm1, xmm1	;# lu*lu 
	mulsd xmm5, xmm1	;# rsq*lu*lu 
	movapd xmm1, [rsp + nb133_three]
	subsd xmm1, xmm5	;# 3-rsq*lu*lu 
	mulsd xmm1, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulsd xmm1, [rsp + nb133_half] ;# rinv 
	movapd [rsp + nb133_rinvH2], xmm1
	
	;# rsqM - seed in xmm2 
	cvtsd2ss xmm2, xmm4
	rsqrtss xmm2, xmm2
	cvtss2sd xmm2, xmm2

	movapd  xmm3, xmm2
	mulsd   xmm2, xmm2
	movapd  xmm1, [rsp + nb133_three]
	mulsd   xmm2, xmm4	;# rsq*lu*lu 
	subsd   xmm1, xmm2	;# 30-rsq*lu*lu 
	mulsd   xmm1, xmm3	;# lu*(3-rsq*lu*lu) 
	mulsd   xmm1, [rsp + nb133_half] ;# iter1 ( new lu) 

	movapd xmm3, xmm1
	mulsd xmm1, xmm1	;# lu*lu 
	mulsd xmm4, xmm1	;# rsq*lu*lu 
	movapd xmm1, [rsp + nb133_three]
	subsd xmm1, xmm4	;# 3-rsq*lu*lu 
	mulsd xmm1, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulsd xmm1, [rsp + nb133_half] ;# rinv 
	movapd [rsp + nb133_rinvM], xmm1

	;# rsqO - put seed in xmm2 
	cvtsd2ss xmm2, xmm7	
	rsqrtss xmm2, xmm2
	cvtss2sd xmm2, xmm2

	movsd  xmm3, xmm2
	mulsd   xmm2, xmm2
	movsd  xmm4, [rsp + nb133_three]
	mulsd   xmm2, xmm7	;# rsq*lu*lu 
	subsd   xmm4, xmm2	;# 30-rsq*lu*lu 
	mulsd   xmm4, xmm3	;# lu*(3-rsq*lu*lu) 
	mulsd   xmm4, [rsp + nb133_half] ;# iter1 ( new lu) 

	movsd xmm3, xmm4
	mulsd xmm4, xmm4	;# lu*lu 
	mulsd xmm7, xmm4	;# rsq*lu*lu 
	movsd xmm4, [rsp + nb133_three]
	subsd xmm4, xmm7	;# 3-rsq*lu*lu 
	mulsd xmm4, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulsd xmm4, [rsp + nb133_half] ;# rinv 
	movsd  xmm7, xmm4	;# rinvO in xmm7 
	
	movsd xmm4, [rsp + nb133_rsqO]
	movapd xmm0, xmm7
	;# LJ table interaction.
	mulsd xmm4, xmm7	;# xmm4=r 
	mulsd xmm4, [rsp + nb133_tsc]
	
	cvttsd2si ebx, xmm4	;# mm6 = lu idx 
	cvtsi2sd xmm5, ebx
	subpd xmm4, xmm5
	movapd xmm1, xmm4	;# xmm1=eps 
	movapd xmm2, xmm1	
	mulpd  xmm2, xmm2	;# xmm2=eps2 

	shl ebx, 3

	mov  rsi, [rbp + nb133_VFtab]

	;# dispersion 
	movlpd xmm4, [rsi + rbx*8]	;# Y1 	
	movhpd xmm4, [rsi + rbx*8 + 8]	;# Y1 F1 	
	movapd xmm5, xmm4
	unpcklpd xmm4, xmm3	;# Y1 Y2 
	unpckhpd xmm5, xmm3	;# F1 F2 

	movlpd xmm6, [rsi + rbx*8 + 16]	;# G1
	movhpd xmm6, [rsi + rbx*8 + 24]	;# G1 H1 	
	movapd xmm7, xmm6
	unpcklpd xmm6, xmm3	;# G1 G2 
	unpckhpd xmm7, xmm3	;# H1 H2 
	;# dispersion table ready, in xmm4-xmm7 	
	mulsd  xmm6, xmm1	;# xmm6=Geps 
	mulsd  xmm7, xmm2	;# xmm7=Heps2 
	addsd  xmm5, xmm6
	addsd  xmm5, xmm7	;# xmm5=Fp 	
	mulsd  xmm7, [rsp + nb133_two]	;# two*Heps2 
	addsd  xmm7, xmm6
	addsd  xmm7, xmm5 ;# xmm7=FF 
	mulsd  xmm5, xmm1 ;# xmm5=eps*Fp 
	addsd  xmm5, xmm4 ;# xmm5=VV 

	movsd xmm4, [rsp + nb133_c6]
	mulsd  xmm7, xmm4	 ;# fijD 
	mulsd  xmm5, xmm4	 ;# Vvdw6 

	;# put scalar force on stack Update Vvdwtot directly 
	addsd  xmm5, [rsp + nb133_Vvdwtot]
	xorpd  xmm3, xmm3
	mulsd  xmm7, [rsp + nb133_tsc]
	subsd  xmm3, xmm7
	movsd [rsp + nb133_fstmp], xmm3
	movsd [rsp + nb133_Vvdwtot], xmm5

	;# repulsion 
	movlpd xmm4, [rsi + rbx*8 + 32]	;# Y1 	
	movhpd xmm4, [rsi + rbx*8 + 40]	;# Y1 F1 	

	movapd xmm5, xmm4
	unpcklpd xmm4, xmm3	;# Y1 Y2 
	unpckhpd xmm5, xmm3	;# F1 F2 

	movlpd xmm6, [rsi + rbx*8 + 48]	;# G1
	movhpd xmm6, [rsi + rbx*8 + 56]	;# G1 H1 	

	movapd xmm7, xmm6
	unpcklpd xmm6, xmm3	;# G1 G2 
	unpckhpd xmm7, xmm3	;# H1 H2 
	
	;# table ready, in xmm4-xmm7 	
	mulsd  xmm6, xmm1	;# xmm6=Geps 
	mulsd  xmm7, xmm2	;# xmm7=Heps2 
	addsd  xmm5, xmm6
	addsd  xmm5, xmm7	;# xmm5=Fp 	
	mulsd  xmm7, [rsp + nb133_two]	;# two*Heps2 
	addsd  xmm7, xmm6
	addsd  xmm7, xmm5 ;# xmm7=FF 
	mulsd  xmm5, xmm1 ;# xmm5=eps*Fp 
	addsd  xmm5, xmm4 ;# xmm5=VV 
	
	movsd xmm4, [rsp + nb133_c12]
	mulsd  xmm7, xmm4 
	mulsd  xmm5, xmm4  
	
	addsd  xmm5, [rsp + nb133_Vvdwtot]
	movsd xmm3, [rsp + nb133_fstmp]
	mulsd  xmm7, [rsp + nb133_tsc]
	subsd  xmm3, xmm7
	movsd [rsp + nb133_Vvdwtot], xmm5

	mulsd  xmm3, xmm0
		
		
	movsd xmm0, [rsp + nb133_dxO]
	movsd xmm1, [rsp + nb133_dyO]
	movsd xmm2, [rsp + nb133_dzO]

	mov    rdi, [rbp + nb133_faction]
	mulsd  xmm0, xmm3
	mulsd  xmm1, xmm3
	mulsd  xmm2, xmm3

	;# update O forces 
	movapd xmm3, [rsp + nb133_fixO]
	movapd xmm4, [rsp + nb133_fiyO]
	movapd xmm7, [rsp + nb133_fizO]
	addsd  xmm3, xmm0
	addsd  xmm4, xmm1
	addsd  xmm7, xmm2
	movsd [rsp + nb133_fixO], xmm3
	movsd [rsp + nb133_fiyO], xmm4
	movsd [rsp + nb133_fizO], xmm7
	;# update j forces with water O 
	movsd [rsp + nb133_fjx], xmm0
	movsd [rsp + nb133_fjy], xmm1
	movsd [rsp + nb133_fjz], xmm2

	;# H1 interactions
	movsd  xmm6, [rsp + nb133_rinvH1] 
	movsd  xmm4, xmm6
	mulsd   xmm4, xmm4	;# xmm6=rinv, xmm4=rinvsq 
	mulsd   xmm6, [rsp + nb133_qqH] ;# vcoul 
	mulsd   xmm4, xmm6    ;# fscal
	addsd  xmm6, [rsp + nb133_vctot]
	movsd [rsp + nb133_vctot], xmm6

	movapd xmm0, [rsp + nb133_dxH1]
	movapd xmm1, [rsp + nb133_dyH1]
	movapd xmm2, [rsp + nb133_dzH1]
	mulsd  xmm0, xmm4
	mulsd  xmm1, xmm4
	mulsd  xmm2, xmm4

	;# update H1 forces 
	movapd xmm3, [rsp + nb133_fixH1]
	movapd xmm4, [rsp + nb133_fiyH1]
	movapd xmm7, [rsp + nb133_fizH1]
	addsd  xmm3, xmm0
	addsd  xmm4, xmm1
	addsd  xmm7, xmm2
	movsd [rsp + nb133_fixH1], xmm3
	movsd [rsp + nb133_fiyH1], xmm4
	movsd [rsp + nb133_fizH1], xmm7
	;# update j forces with water H1 
	addsd  xmm0, [rsp + nb133_fjx]
	addsd  xmm1, [rsp + nb133_fjy]
	addsd  xmm2, [rsp + nb133_fjz]
	movsd [rsp + nb133_fjx], xmm0
	movsd [rsp + nb133_fjy], xmm1
	movsd [rsp + nb133_fjz], xmm2

	;# H2 interactions 
	movsd  xmm6, [rsp + nb133_rinvH2] 
	movsd  xmm4, xmm6
	mulsd   xmm4, xmm4	;# xmm6=rinv, xmm4=rinvsq 
	mulsd   xmm6, [rsp + nb133_qqH] ;# vcoul 
	mulsd   xmm4, xmm6    ;# fscal
	addsd  xmm6, [rsp + nb133_vctot]
	movsd [rsp + nb133_vctot], xmm6

	movapd xmm0, [rsp + nb133_dxH2]
	movapd xmm1, [rsp + nb133_dyH2]
	movapd xmm2, [rsp + nb133_dzH2]
	mulsd  xmm0, xmm4
	mulsd  xmm1, xmm4
	mulsd  xmm2, xmm4

	;# update H2 forces 
	movapd xmm3, [rsp + nb133_fixH2]
	movapd xmm4, [rsp + nb133_fiyH2]
	movapd xmm7, [rsp + nb133_fizH2]
	addsd  xmm3, xmm0
	addsd  xmm4, xmm1
	addsd  xmm7, xmm2
	movsd [rsp + nb133_fixH2], xmm3
	movsd [rsp + nb133_fiyH2], xmm4
	movsd [rsp + nb133_fizH2], xmm7
	;# update j forces with water H2 
	addsd  xmm0, [rsp + nb133_fjx]
	addsd  xmm1, [rsp + nb133_fjy]
	addsd  xmm2, [rsp + nb133_fjz]
	movsd [rsp + nb133_fjx], xmm0
	movsd [rsp + nb133_fjy], xmm1
	movsd [rsp + nb133_fjz], xmm2

	;# M interactions 
	movsd  xmm6, [rsp + nb133_rinvM] 
	movsd  xmm4, xmm6
	mulsd   xmm4, xmm4	;# xmm6=rinv, xmm4=rinvsq 
	mulsd   xmm6, [rsp + nb133_qqM] ;# vcoul 
	mulsd   xmm4, xmm6    ;# fscal
	addsd  xmm6, [rsp + nb133_vctot]
	movsd [rsp + nb133_vctot], xmm6

	movapd xmm0, [rsp + nb133_dxM]
	movapd xmm1, [rsp + nb133_dyM]
	movapd xmm2, [rsp + nb133_dzM]
	mulsd  xmm0, xmm4
	mulsd  xmm1, xmm4
	mulsd  xmm2, xmm4

	;# update M forces 
	movapd xmm3, [rsp + nb133_fixM]
	movapd xmm4, [rsp + nb133_fiyM]
	movapd xmm7, [rsp + nb133_fizM]
	addsd  xmm3, xmm0
	addsd  xmm4, xmm1
	addsd  xmm7, xmm2
	movsd [rsp + nb133_fixM], xmm3
	movsd [rsp + nb133_fiyM], xmm4
	movsd [rsp + nb133_fizM], xmm7

	mov rdi, [rbp + nb133_faction]
	;# update j forces 
	addsd  xmm0, [rsp + nb133_fjx]
	addsd  xmm1, [rsp + nb133_fjy]
	addsd  xmm2, [rsp + nb133_fjz]
	movlpd xmm3, [rdi + rax*8]
	movlpd xmm4, [rdi + rax*8 + 8]
	movlpd xmm5, [rdi + rax*8 + 16]
	addsd xmm3, xmm0
	addsd xmm4, xmm1
	addsd xmm5, xmm2
	movlpd [rdi + rax*8], xmm3
	movlpd [rdi + rax*8 + 8], xmm4
	movlpd [rdi + rax*8 + 16], xmm5

.nb133_updateouterdata:
	mov   ecx, [rsp + nb133_ii3]
	mov   rdi, [rbp + nb133_faction]
	mov   rsi, [rbp + nb133_fshift]
	mov   edx, [rsp + nb133_is3]

	;# accumulate  Oi forces in xmm0, xmm1, xmm2 
	movapd xmm0, [rsp + nb133_fixO]
	movapd xmm1, [rsp + nb133_fiyO]
	movapd xmm2, [rsp + nb133_fizO]

	movhlps xmm3, xmm0
	movhlps xmm4, xmm1
	movhlps xmm5, xmm2
	addsd  xmm0, xmm3
	addsd  xmm1, xmm4
	addsd  xmm2, xmm5 ;# sum is in low xmm0-xmm2 

	;# increment i force 
	movsd  xmm3, [rdi + rcx*8]
	movsd  xmm4, [rdi + rcx*8 + 8]
	movsd  xmm5, [rdi + rcx*8 + 16]
	subsd  xmm3, xmm0
	subsd  xmm4, xmm1
	subsd  xmm5, xmm2
	movsd  [rdi + rcx*8],     xmm3
	movsd  [rdi + rcx*8 + 8], xmm4
	movsd  [rdi + rcx*8 + 16], xmm5

	;# accumulate force in xmm6/xmm7 for fshift 
	movapd xmm6, xmm0
	movsd xmm7, xmm2
	unpcklpd xmm6,xmm1 

	;# accumulate H1i forces in xmm0, xmm1, xmm2 
	movapd xmm0, [rsp + nb133_fixH1]
	movapd xmm1, [rsp + nb133_fiyH1]
	movapd xmm2, [rsp + nb133_fizH1]

	movhlps xmm3, xmm0
	movhlps xmm4, xmm1
	movhlps xmm5, xmm2
	addsd  xmm0, xmm3
	addsd  xmm1, xmm4
	addsd  xmm2, xmm5 ;# sum is in low xmm0-xmm2 

	;# increment i force 
	movsd  xmm3, [rdi + rcx*8 + 24]
	movsd  xmm4, [rdi + rcx*8 + 32]
	movsd  xmm5, [rdi + rcx*8 + 40]
	subsd  xmm3, xmm0
	subsd  xmm4, xmm1
	subsd  xmm5, xmm2
	movsd  [rdi + rcx*8 + 24], xmm3
	movsd  [rdi + rcx*8 + 32], xmm4
	movsd  [rdi + rcx*8 + 40], xmm5

	;# accumulate force in xmm6/xmm7 for fshift 
	addsd xmm7, xmm2
	unpcklpd xmm0, xmm1
	addpd xmm6, xmm0

	;# accumulate H2i forces in xmm0, xmm1, xmm2 
	movapd xmm0, [rsp + nb133_fixH2]
	movapd xmm1, [rsp + nb133_fiyH2]
	movapd xmm2, [rsp + nb133_fizH2]

	movhlps xmm3, xmm0
	movhlps xmm4, xmm1
	movhlps xmm5, xmm2
	addsd  xmm0, xmm3
	addsd  xmm1, xmm4
	addsd  xmm2, xmm5 ;# sum is in low xmm0-xmm2 

	;# increment i force 
	movsd  xmm3, [rdi + rcx*8 + 48]
	movsd  xmm4, [rdi + rcx*8 + 56]
	movsd  xmm5, [rdi + rcx*8 + 64]
	subsd  xmm3, xmm0
	subsd  xmm4, xmm1
	subsd  xmm5, xmm2
	movsd  [rdi + rcx*8 + 48], xmm3
	movsd  [rdi + rcx*8 + 56], xmm4
	movsd  [rdi + rcx*8 + 64], xmm5

	;# accumulate force in xmm6/xmm7 for fshift 
	addsd xmm7, xmm2
	unpcklpd xmm0, xmm1
	addpd xmm6, xmm0

	;# accumulate Mi forces in xmm0, xmm1, xmm2 
	movapd xmm0, [rsp + nb133_fixM]
	movapd xmm1, [rsp + nb133_fiyM]
	movapd xmm2, [rsp + nb133_fizM]

	movhlps xmm3, xmm0
	movhlps xmm4, xmm1
	movhlps xmm5, xmm2
	addsd  xmm0, xmm3
	addsd  xmm1, xmm4
	addsd  xmm2, xmm5 ;# sum is in low xmm0-xmm2 

	;# increment i force 
	movsd  xmm3, [rdi + rcx*8 + 72]
	movsd  xmm4, [rdi + rcx*8 + 80]
	movsd  xmm5, [rdi + rcx*8 + 88]
	subsd  xmm3, xmm0
	subsd  xmm4, xmm1
	subsd  xmm5, xmm2
	movsd  [rdi + rcx*8 + 72], xmm3
	movsd  [rdi + rcx*8 + 80], xmm4
	movsd  [rdi + rcx*8 + 88], xmm5

	;# accumulate force in xmm6/xmm7 for fshift 
	addsd xmm7, xmm2
	unpcklpd xmm0, xmm1
	addpd xmm6, xmm0

	;# increment fshift force 
	movlpd xmm3, [rsi + rdx*8]
	movhpd xmm3, [rsi + rdx*8 + 8]
	movsd  xmm4, [rsi + rdx*8 + 16]
	subpd  xmm3, xmm6
	subsd  xmm4, xmm7
	movlpd [rsi + rdx*8],      xmm3
	movhpd [rsi + rdx*8 + 8],  xmm3
	movsd  [rsi + rdx*8 + 16], xmm4

	;# get n from stack
	mov esi, [rsp + nb133_n]
        ;# get group index for i particle 
        mov   rdx, [rbp + nb133_gid]      	;# base of gid[]
        mov   edx, [rdx + rsi*4]		;# ggid=gid[n]

	;# accumulate total potential energy and update it 
	movapd xmm7, [rsp + nb133_vctot]
	;# accumulate 
	movhlps xmm6, xmm7
	addsd  xmm7, xmm6	;# low xmm7 has the sum now 
        
	;# add earlier value from mem 
	mov   rax, [rbp + nb133_Vc]
	addsd xmm7, [rax + rdx*8] 
	;# move back to mem 
	movsd [rax + rdx*8], xmm7 
	
	;# accumulate total lj energy and update it 
	movapd xmm7, [rsp + nb133_Vvdwtot]
	;# accumulate 
	movhlps xmm6, xmm7
	addsd  xmm7, xmm6	;# low xmm7 has the sum now 

	;# add earlier value from mem 
	mov   rax, [rbp + nb133_Vvdw]
	addsd xmm7, [rax + rdx*8] 
	;# move back to mem 
	movsd [rax + rdx*8], xmm7 
	
       ;# finish if last 
        mov ecx, [rsp + nb133_nn1]
	;# esi already loaded with n
	inc esi
        sub ecx, esi
        jz .nb133_outerend

        ;# not last, iterate outer loop once more!  
        mov [rsp + nb133_n], esi
        jmp .nb133_outer
.nb133_outerend:
        ;# check if more outer neighborlists remain
        mov   ecx, [rsp + nb133_nri]
	;# esi already loaded with n above
        sub   ecx, esi
        jz .nb133_end
        ;# non-zero, do one more workunit
        jmp   .nb133_threadloop
.nb133_end:
	mov eax, [rsp + nb133_nouter]
	mov ebx, [rsp + nb133_ninner]
	mov rcx, [rbp + nb133_outeriter]
	mov rdx, [rbp + nb133_inneriter]
	mov [rcx], eax
	mov [rdx], ebx

	add rsp, 1080
	emms


        pop r15