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 xmm1, [rsp + nb133nf_three]
	subpd xmm1, xmm4	;# 3-rsq*lu*lu 
	mulpd xmm1, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulpd xmm1, [rsp + nb133nf_half] ;# rinv 
	movapd  [rsp + nb133nf_rinvM], xmm1	

		
	;# rsqO - put seed in xmm2 
	cvtpd2ps xmm2, xmm7	
	rsqrtps xmm2, xmm2
	cvtps2pd xmm2, xmm2

	movapd  xmm3, xmm2
	mulpd   xmm2, xmm2
	movapd  xmm4, [rsp + nb133nf_three]
	mulpd   xmm2, xmm7	;# rsq*lu*lu 
	subpd   xmm4, xmm2	;# 30-rsq*lu*lu 
	mulpd   xmm4, xmm3	;# lu*(3-rsq*lu*lu) 
	mulpd   xmm4, [rsp + nb133nf_half] ;# iter1 ( new lu) 

	movapd xmm3, xmm4
	mulpd xmm4, xmm4	;# lu*lu 
	mulpd xmm7, xmm4	;# rsq*lu*lu 
	movapd xmm4, [rsp + nb133nf_three]
	subpd xmm4, xmm7	;# 3-rsq*lu*lu 
	mulpd xmm4, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulpd xmm4, [rsp + nb133nf_half] ;# rinv 
	movapd  xmm7, xmm4	;# rinvO in xmm7 
	
	
	
	movapd xmm4, [rsp + nb133nf_rsqO]
	movapd xmm0, xmm7
	;# LJ table interaction.
	mulpd xmm4, xmm7	;# xmm4=r 
	mulpd xmm4, [rsp + nb133nf_tsc]
	
	cvttpd2pi mm6, xmm4	;# mm6 = lu idx 
	cvtpi2pd xmm5, mm6
	subpd xmm4, xmm5
	movapd xmm1, xmm4	;# xmm1=eps 
	movapd xmm2, xmm1	
	mulpd  xmm2, xmm2	;# xmm2=eps2 

	pslld mm6, 3		;# idx *= 8 
	
	mov  rsi, [rbp + nb133nf_VFtab]
	movd eax, mm6
	psrlq mm6, 32
	movd ebx, mm6

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

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

	movapd xmm4, [rsp + nb133nf_c6]
	mulpd  xmm5, xmm4	 ;# Vvdw6 

	;# Update Vvdwtot directly 
	addpd  xmm5, [rsp + nb133nf_Vvdwtot]
	movapd [rsp + nb133nf_Vvdwtot], xmm5

	;# repulsion 
	movlpd xmm4, [rsi + rax*8 + 32]	;# Y1 	
	movlpd xmm3, [rsi + rbx*8 + 32]	;# Y2 
	movhpd xmm4, [rsi + rax*8 + 40]	;# Y1 F1 	
	movhpd xmm3, [rsi + rbx*8 + 40]	;# Y2 F2 

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

	movlpd xmm6, [rsi + rax*8 + 48]	;# G1
	movlpd xmm3, [rsi + rbx*8 + 48]	;# G2
	movhpd xmm6, [rsi + rax*8 + 56]	;# G1 H1 	
	movhpd xmm3, [rsi + rbx*8 + 56]	;# G2 H2 

	movapd xmm7, xmm6
	unpcklpd xmm6, xmm3	;# G1 G2 
	unpckhpd xmm7, xmm3	;# H1 H2 
	
	;# table ready, in xmm4-xmm7 	
	mulpd  xmm6, xmm1	;# xmm6=Geps 
	mulpd  xmm7, xmm2	;# xmm7=Heps2 
	addpd  xmm5, xmm6
	addpd  xmm5, xmm7	;# xmm5=Fp 	
	mulpd  xmm5, xmm1 ;# xmm5=eps*Fp 
	addpd  xmm5, xmm4 ;# xmm5=VV 
	
	movapd xmm4, [rsp + nb133nf_c12]
	mulpd  xmm5, xmm4  
	
	addpd  xmm5, [rsp + nb133nf_Vvdwtot]
	movapd [rsp + nb133nf_Vvdwtot], xmm5

	;# H1/H2/M interactions 
	movapd  xmm6, [rsp + nb133nf_rinvH1] 
	addpd   xmm6, [rsp + nb133nf_rinvH2] 
	movapd  xmm7, [rsp + nb133nf_rinvM] 
	mulpd   xmm6, [rsp + nb133nf_qqH]
	mulpd   xmm7, [rsp + nb133nf_qqM]
	addpd   xmm6, xmm7
	addpd   xmm6, [rsp + nb133nf_vctot]
	movapd  [rsp + nb133nf_vctot], xmm6
	
	;# should we do one more iteration? 
	sub dword ptr [rsp + nb133nf_innerk],  2
	jl   .nb133nf_checksingle
	jmp  .nb133nf_unroll_loop
.nb133nf_checksingle:	
	mov   edx, [rsp + nb133nf_innerk]
	and   edx, 1
	jnz  .nb133nf_dosingle
	jmp  .nb133nf_updateouterdata
.nb133nf_dosingle:
	mov   rdx, [rsp + nb133nf_innerjjnr]     ;# pointer to jjnr[k] 
	mov   eax, [rdx]	
	add qword ptr [rsp + nb133nf_innerjjnr],  4

	mov rsi, [rbp + nb133nf_charge]    ;# base of charge[] 

	xorpd xmm3, xmm3
	movlpd xmm3, [rsi + rax*8]
	movapd xmm4, xmm3
	mulsd  xmm3, [rsp + nb133nf_iqM]
	mulsd  xmm4, [rsp + nb133nf_iqH]

	movd  mm0, eax		;# use mmx registers as temp storage 

	movapd  [rsp + nb133nf_qqM], xmm3
	movapd  [rsp + nb133nf_qqH], xmm4
	
	mov rsi, [rbp + nb133nf_type]
	mov eax, [rsi + rax*4]
	mov rsi, [rbp + nb133nf_vdwparam]
	shl eax, 1	
	mov edi, [rsp + nb133nf_ntia]
	add eax, edi

	movlpd xmm6, [rsi + rax*8]	;# c6a
	movhpd xmm6, [rsi + rax*8 + 8]	;# c6a c12a 

	xorpd xmm7, xmm7
	movapd xmm4, xmm6
	unpcklpd xmm4, xmm7
	unpckhpd xmm6, xmm7
	
	movd  eax, mm0
	movd  ebx, mm1
	movapd [rsp + nb133nf_c6], xmm4
	movapd [rsp + nb133nf_c12], xmm6
	
	mov rsi, [rbp + nb133nf_pos]       ;# base of pos[] 

	lea   rax, [rax + rax*2]     ;# replace jnr with j3 

	;# move coordinates to xmm0-xmm2 
	movlpd xmm0, [rsi + rax*8]
	movlpd xmm1, [rsi + rax*8 + 8]
	movlpd xmm2, [rsi + rax*8 + 16]

	;# move ixO-izO to xmm4-xmm6 
	movapd xmm4, [rsp + nb133nf_ixO]
	movapd xmm5, [rsp + nb133nf_iyO]
	movapd xmm6, [rsp + nb133nf_izO]

	;# calc dr 
	subsd xmm4, xmm0
	subsd xmm5, xmm1
	subsd xmm6, xmm2

	;# square it 
	mulsd xmm4,xmm4
	mulsd xmm5,xmm5
	mulsd xmm6,xmm6
	addsd xmm4, xmm5
	addsd xmm4, xmm6
	movapd xmm7, xmm4
	;# rsqO in xmm7 
	movapd [rsp + nb133nf_rsqO], xmm7
	
	;# move ixH1-izH1 to xmm4-xmm6 
	movapd xmm4, [rsp + nb133nf_ixH1]
	movapd xmm5, [rsp + nb133nf_iyH1]
	movapd xmm6, [rsp + nb133nf_izH1]

	;# calc dr 
	subsd xmm4, xmm0
	subsd xmm5, xmm1
	subsd xmm6, xmm2

	;# square it 
	mulsd xmm4,xmm4
	mulsd xmm5,xmm5
	mulsd xmm6,xmm6
	addsd xmm6, xmm5
	addsd xmm6, xmm4
	;# rsqH1 in xmm6 

	;# move ixH2-izH2 to xmm3-xmm5  
	movapd xmm3, [rsp + nb133nf_ixH2]
	movapd xmm4, [rsp + nb133nf_iyH2]
	movapd xmm5, [rsp + nb133nf_izH2]

	;# calc dr 
	subsd xmm3, xmm0
	subsd xmm4, xmm1
	subsd xmm5, xmm2

	;# square it 
	mulsd xmm3,xmm3
	mulsd xmm4,xmm4
	mulsd xmm5,xmm5
	addsd xmm5, xmm4
	addsd xmm5, xmm3
	;# move ixM-izM to xmm2-xmm4  
	movapd xmm3, [rsp + nb133nf_iyM]
	movapd xmm4, [rsp + nb133nf_izM]
	subpd  xmm3, xmm1
	subpd  xmm4, xmm2
	movapd xmm2, [rsp + nb133nf_ixM]
	subpd  xmm2, xmm0	

	;# square it 
	mulpd xmm2,xmm2
	mulpd xmm3,xmm3
	mulpd xmm4,xmm4
	addpd xmm4, xmm3
	addpd xmm4, xmm2	
	;# rsqM in xmm4, rsqH2 in xmm5, rsqH1 in xmm6, rsqO in xmm7 

	;# start with rsqH1 - put seed in xmm2 
	cvtsd2ss xmm2, xmm6	
	rsqrtss xmm2, xmm2
	cvtss2sd xmm2, xmm2

	movapd  xmm3, xmm2
	mulsd   xmm2, xmm2
	movapd  xmm1, [rsp + nb133nf_three]
	mulsd   xmm2, xmm6	;# rsq*lu*lu 
	subsd   xmm1, xmm2	;# 30-rsq*lu*lu 
	mulsd   xmm1, xmm3	;# lu*(3-rsq*lu*lu) 
	mulsd   xmm1, [rsp + nb133nf_half] ;# iter1 ( new lu) 

	movapd xmm3, xmm1
	mulsd xmm1, xmm1	;# lu*lu 
	mulsd xmm6, xmm1	;# rsq*lu*lu 
	movapd xmm1, [rsp + nb133nf_three]
	subsd xmm1, xmm6	;# 3-rsq*lu*lu 
	mulsd xmm1, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulsd xmm1, [rsp + nb133nf_half] ;# rinv 
	movapd [rsp + nb133nf_rinvH1], xmm1
	
	;# rsqH2 - seed in xmm2 
	cvtsd2ss xmm2, xmm5	
	rsqrtss xmm2, xmm2
	cvtss2sd xmm2, xmm2

	movapd  xmm3, xmm2
	mulsd   xmm2, xmm2
	movapd  xmm1, [rsp + nb133nf_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 + nb133nf_half] ;# iter1 ( new lu) 

	movapd xmm3, xmm1
	mulsd xmm1, xmm1	;# lu*lu 
	mulsd xmm5, xmm1	;# rsq*lu*lu 
	movapd xmm1, [rsp + nb133nf_three]
	subsd xmm1, xmm5	;# 3-rsq*lu*lu 
	mulsd xmm1, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulsd xmm1, [rsp + nb133nf_half] ;# rinv 
	movapd [rsp + nb133nf_rinvH2], xmm1
	
	;# rsqM - seed in xmm2 
	cvtsd2ss xmm2, xmm4
	rsqrtss xmm2, xmm2
	cvtss2sd xmm2, xmm2

	movapd  xmm3, xmm2
	mulsd   xmm2, xmm2
	movapd  xmm1, [rsp + nb133nf_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 + nb133nf_half] ;# iter1 ( new lu) 

	movapd xmm3, xmm1
	mulsd xmm1, xmm1	;# lu*lu 
	mulsd xmm4, xmm1	;# rsq*lu*lu 
	movapd xmm1, [rsp + nb133nf_three]
	subsd xmm1, xmm4	;# 3-rsq*lu*lu 
	mulsd xmm1, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulsd xmm1, [rsp + nb133nf_half] ;# rinv 
	movapd [rsp + nb133nf_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 + nb133nf_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 + nb133nf_half] ;# iter1 ( new lu) 

	movsd xmm3, xmm4
	mulsd xmm4, xmm4	;# lu*lu 
	mulsd xmm7, xmm4	;# rsq*lu*lu 
	movsd xmm4, [rsp + nb133nf_three]
	subsd xmm4, xmm7	;# 3-rsq*lu*lu 
	mulsd xmm4, xmm3	;# lu*(	3-rsq*lu*lu) 
	mulsd xmm4, [rsp + nb133nf_half] ;# rinv 
	movsd  xmm7, xmm4	;# rinvO in xmm7 
	
	movsd xmm4, [rsp + nb133nf_rsqO]
	movapd xmm0, xmm7
	;# LJ table interaction.
	mulsd xmm4, xmm7	;# xmm4=r 
	mulsd xmm4, [rsp + nb133nf_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 + nb133nf_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  xmm5, xmm1 ;# xmm5=eps*Fp 
	addsd  xmm5, xmm4 ;# xmm5=VV 

	movsd xmm4, [rsp + nb133nf_c6]
	mulsd  xmm5, xmm4	 ;# Vvdw6 

	;# put scalar force on stack Update Vvdwtot directly 
	addsd  xmm5, [rsp + nb133nf_Vvdwtot]
	movsd [rsp + nb133nf_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  xmm5, xmm1 ;# xmm5=eps*Fp 
	addsd  xmm5, xmm4 ;# xmm5=VV 
	
	movsd xmm4, [rsp + nb133nf_c12]
	mulsd  xmm5, xmm4  
	
	addsd  xmm5, [rsp + nb133nf_Vvdwtot]
	movsd [rsp + nb133nf_Vvdwtot], xmm5

	;# H1/H2/M interactions 
	movsd  xmm6, [rsp + nb133nf_rinvH1] 
	addsd  xmm6, [rsp + nb133nf_rinvH2] 
	movsd  xmm7, [rsp + nb133nf_rinvM] 
	mulsd  xmm6, [rsp + nb133nf_qqH]
	mulsd  xmm7, [rsp + nb133nf_qqM]
	addsd  xmm6, xmm7
	addsd  xmm6, [rsp + nb133nf_vctot]
	movsd  [rsp + nb133nf_vctot], xmm6
	
.nb133nf_updateouterdata:
	;# get n from stack
	mov esi, [rsp + nb133nf_n]
        ;# get group index for i particle 
        mov   rdx, [rbp + nb133nf_gid]      	;# base of gid[]
        mov   edx, [rdx + rsi*4]		;# ggid=gid[n]

	;# accumulate total potential energy and update it 
	movapd xmm7, [rsp + nb133nf_vctot]
	;# accumulate 
	movhlps xmm6, xmm7
	addsd  xmm7, xmm6	;# low xmm7 has the sum now 
        
	;# add earlier value from mem 
	mov   rax, [rbp + nb133nf_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 + nb133nf_Vvdwtot]
	;# accumulate 
	movhlps xmm6, xmm7
	addsd  xmm7, xmm6	;# low xmm7 has the sum now 

	;# add earlier value from mem 
	mov   rax, [rbp + nb133nf_Vvdw]
	addsd xmm7, [rax + rdx*8] 
	;# move back to mem 
	movsd [rax + rdx*8], xmm7 
	
       ;# finish if last 
        mov ecx, [rsp + nb133nf_nn1]
	;# esi already loaded with n
	inc esi
        sub ecx, esi
        jz .nb133nf_outerend

        ;# not last, iterate outer loop once more!  
        mov [rsp + nb133nf_n], esi
        jmp .nb133nf_outer
.nb133nf_outerend:
        ;# check if more outer neighborlists remain
        mov   ecx, [rsp + nb133nf_nri]
	;# esi already loaded with n above
        sub   ecx, esi
        jz .nb133nf_end
        ;# non-zero, do one more workunit
        jmp   .nb133nf_threadloop
.nb133nf_end:
	mov eax, [rsp + nb133nf_nouter]
	mov ebx, [rsp + nb133nf_ninner]
	mov rcx, [rbp + nb133nf_