nb_kernel312_ia32_sse2.intel_syntax.s

来自「最著名最快的分子模拟软件」· S 代码 · 共 2,210 行 · 第 1/5 页

S
2,210
字号
	movapd 	[esp + nb312_jyH2], xmm3	movapd 	[esp + nb312_jzH2], xmm4		movapd xmm0, [esp + nb312_ixO]	movapd xmm1, [esp + nb312_iyO]	movapd xmm2, [esp + nb312_izO]	movapd xmm3, [esp + nb312_ixO]	movapd xmm4, [esp + nb312_iyO]	movapd xmm5, [esp + nb312_izO]	subpd  xmm0, [esp + nb312_jxO]	subpd  xmm1, [esp + nb312_jyO]	subpd  xmm2, [esp + nb312_jzO]	subpd  xmm3, [esp + nb312_jxH1]	subpd  xmm4, [esp + nb312_jyH1]	subpd  xmm5, [esp + nb312_jzH1]	movapd [esp + nb312_dxOO], xmm0	movapd [esp + nb312_dyOO], xmm1	movapd [esp + nb312_dzOO], xmm2	mulpd  xmm0, xmm0	mulpd  xmm1, xmm1	mulpd  xmm2, xmm2	movapd [esp + nb312_dxOH1], xmm3	movapd [esp + nb312_dyOH1], xmm4	movapd [esp + nb312_dzOH1], xmm5	mulpd  xmm3, xmm3	mulpd  xmm4, xmm4	mulpd  xmm5, xmm5	addpd  xmm0, xmm1	addpd  xmm0, xmm2	addpd  xmm3, xmm4	addpd  xmm3, xmm5	movapd [esp + nb312_rsqOO], xmm0	movapd [esp + nb312_rsqOH1], xmm3	movapd xmm0, [esp + nb312_ixO]	movapd xmm1, [esp + nb312_iyO]	movapd xmm2, [esp + nb312_izO]	movapd xmm3, [esp + nb312_ixH1]	movapd xmm4, [esp + nb312_iyH1]	movapd xmm5, [esp + nb312_izH1]	subpd  xmm0, [esp + nb312_jxH2]	subpd  xmm1, [esp + nb312_jyH2]	subpd  xmm2, [esp + nb312_jzH2]	subpd  xmm3, [esp + nb312_jxO]	subpd  xmm4, [esp + nb312_jyO]	subpd  xmm5, [esp + nb312_jzO]	movapd [esp + nb312_dxOH2], xmm0	movapd [esp + nb312_dyOH2], xmm1	movapd [esp + nb312_dzOH2], xmm2	mulpd  xmm0, xmm0	mulpd  xmm1, xmm1	mulpd  xmm2, xmm2	movapd [esp + nb312_dxH1O], xmm3	movapd [esp + nb312_dyH1O], xmm4	movapd [esp + nb312_dzH1O], xmm5	mulpd  xmm3, xmm3	mulpd  xmm4, xmm4	mulpd  xmm5, xmm5	addpd  xmm0, xmm1	addpd  xmm0, xmm2	addpd  xmm3, xmm4	addpd  xmm3, xmm5	movapd [esp + nb312_rsqOH2], xmm0	movapd [esp + nb312_rsqH1O], xmm3	movapd xmm0, [esp + nb312_ixH1]	movapd xmm1, [esp + nb312_iyH1]	movapd xmm2, [esp + nb312_izH1]	movapd xmm3, [esp + nb312_ixH1]	movapd xmm4, [esp + nb312_iyH1]	movapd xmm5, [esp + nb312_izH1]	subpd  xmm0, [esp + nb312_jxH1]	subpd  xmm1, [esp + nb312_jyH1]	subpd  xmm2, [esp + nb312_jzH1]	subpd  xmm3, [esp + nb312_jxH2]	subpd  xmm4, [esp + nb312_jyH2]	subpd  xmm5, [esp + nb312_jzH2]	movapd [esp + nb312_dxH1H1], xmm0	movapd [esp + nb312_dyH1H1], xmm1	movapd [esp + nb312_dzH1H1], xmm2	mulpd  xmm0, xmm0	mulpd  xmm1, xmm1	mulpd  xmm2, xmm2	movapd [esp + nb312_dxH1H2], xmm3	movapd [esp + nb312_dyH1H2], xmm4	movapd [esp + nb312_dzH1H2], xmm5	mulpd  xmm3, xmm3	mulpd  xmm4, xmm4	mulpd  xmm5, xmm5	addpd  xmm0, xmm1	addpd  xmm0, xmm2	addpd  xmm3, xmm4	addpd  xmm3, xmm5	movapd [esp + nb312_rsqH1H1], xmm0	movapd [esp + nb312_rsqH1H2], xmm3	movapd xmm0, [esp + nb312_ixH2]	movapd xmm1, [esp + nb312_iyH2]	movapd xmm2, [esp + nb312_izH2]	movapd xmm3, [esp + nb312_ixH2]	movapd xmm4, [esp + nb312_iyH2]	movapd xmm5, [esp + nb312_izH2]	subpd  xmm0, [esp + nb312_jxO]	subpd  xmm1, [esp + nb312_jyO]	subpd  xmm2, [esp + nb312_jzO]	subpd  xmm3, [esp + nb312_jxH1]	subpd  xmm4, [esp + nb312_jyH1]	subpd  xmm5, [esp + nb312_jzH1]	movapd [esp + nb312_dxH2O], xmm0	movapd [esp + nb312_dyH2O], xmm1	movapd [esp + nb312_dzH2O], xmm2	mulpd  xmm0, xmm0	mulpd  xmm1, xmm1	mulpd  xmm2, xmm2	movapd [esp + nb312_dxH2H1], xmm3	movapd [esp + nb312_dyH2H1], xmm4	movapd [esp + nb312_dzH2H1], xmm5	mulpd  xmm3, xmm3	mulpd  xmm4, xmm4	mulpd  xmm5, xmm5	addpd  xmm0, xmm1	addpd  xmm0, xmm2	addpd  xmm4, xmm3	addpd  xmm4, xmm5	movapd [esp + nb312_rsqH2O], xmm0	movapd [esp + nb312_rsqH2H1], xmm4	movapd xmm0, [esp + nb312_ixH2]	movapd xmm1, [esp + nb312_iyH2]	movapd xmm2, [esp + nb312_izH2]	subpd  xmm0, [esp + nb312_jxH2]	subpd  xmm1, [esp + nb312_jyH2]	subpd  xmm2, [esp + nb312_jzH2]	movapd [esp + nb312_dxH2H2], xmm0	movapd [esp + nb312_dyH2H2], xmm1	movapd [esp + nb312_dzH2H2], xmm2	mulpd xmm0, xmm0	mulpd xmm1, xmm1	mulpd xmm2, xmm2	addpd xmm0, xmm1	addpd xmm0, xmm2	movapd [esp + nb312_rsqH2H2], xmm0			;# start doing invsqrt use rsq values in xmm0, xmm4 	cvtpd2ps xmm1, xmm0		cvtpd2ps xmm5, xmm4		rsqrtps xmm1, xmm1	rsqrtps xmm5, xmm5	cvtps2pd xmm1, xmm1	cvtps2pd xmm5, xmm5		movapd  xmm2, xmm1	;# copy of luA 	movapd  xmm6, xmm5	;# copy of luB 	mulpd   xmm1, xmm1	;# luA*luA 	mulpd   xmm5, xmm5	;# luB*luB 	movapd  xmm3, [esp + nb312_three]	mulpd   xmm1, xmm0	;# rsqA*luA*luA 	mulpd   xmm5, xmm4	;# rsqB*luB*luB 		movapd  xmm7, xmm3	subpd   xmm3, xmm1	;# 3-rsqA*luA*luA 	subpd   xmm7, xmm5	;# 3-rsqB*luB*luB 	mulpd   xmm3, xmm2	;# luA*(3-rsqA*luA*luA) 	mulpd   xmm7, xmm6	;# luB*(3-rsqB*luB*luB) 	mulpd   xmm3, [esp + nb312_half] ;# iter1 	mulpd   xmm7, [esp + nb312_half] ;# iter1 	movapd  xmm2, xmm3	;# copy of luA 	movapd  xmm6, xmm7	;# copy of luB 	mulpd   xmm3, xmm3	;# luA*luA 	mulpd   xmm7, xmm7	;# luB*luB 	movapd  xmm1, [esp + nb312_three]	mulpd   xmm3, xmm0	;# rsqA*luA*luA 	mulpd   xmm7, xmm4	;# rsqB*luB*luB 		movapd  xmm5, xmm1	subpd   xmm1, xmm3	;# 3-rsqA*luA*luA 	subpd   xmm5, xmm7	;# 3-rsqB*luB*luB 	mulpd   xmm1, xmm2	;# luA*(3-rsqA*luA*luA) 	mulpd   xmm5, xmm6	;# luB*(3-rsqB*luB*luB) 	mulpd   xmm1, [esp + nb312_half] ;# rinv 	mulpd   xmm5, [esp + nb312_half] ;# rinv 	movapd [esp + nb312_rinvH2H2], xmm1	movapd [esp + nb312_rinvH2H1], xmm5	movapd xmm0, [esp + nb312_rsqOO]	movapd xmm4, [esp + nb312_rsqOH1]		cvtpd2ps xmm1, xmm0		cvtpd2ps xmm5, xmm4		rsqrtps xmm1, xmm1	rsqrtps xmm5, xmm5	cvtps2pd xmm1, xmm1	cvtps2pd xmm5, xmm5		movapd  xmm2, xmm1	;# copy of luA 	movapd  xmm6, xmm5	;# copy of luB 	mulpd   xmm1, xmm1	;# luA*luA 	mulpd   xmm5, xmm5	;# luB*luB 	movapd  xmm3, [esp + nb312_three]	mulpd   xmm1, xmm0	;# rsqA*luA*luA 	mulpd   xmm5, xmm4	;# rsqB*luB*luB 		movapd  xmm7, xmm3	subpd   xmm3, xmm1	;# 3-rsqA*luA*luA 	subpd   xmm7, xmm5	;# 3-rsqB*luB*luB 	mulpd   xmm3, xmm2	;# luA*(3-rsqA*luA*luA) 	mulpd   xmm7, xmm6	;# luB*(3-rsqB*luB*luB) 	mulpd   xmm3, [esp + nb312_half] ;# iter1 of  	mulpd   xmm7, [esp + nb312_half] ;# iter1 of  	movapd  xmm2, xmm3	;# copy of luA 	movapd  xmm6, xmm7	;# copy of luB 	mulpd   xmm3, xmm3	;# luA*luA 	mulpd   xmm7, xmm7	;# luB*luB 	movapd  xmm1, [esp + nb312_three]	mulpd   xmm3, xmm0	;# rsqA*luA*luA 	mulpd   xmm7, xmm4	;# rsqB*luB*luB 		movapd  xmm5, xmm1	subpd   xmm1, xmm3	;# 3-rsqA*luA*luA 	subpd   xmm5, xmm7	;# 3-rsqB*luB*luB 	mulpd   xmm1, xmm2	;# luA*(3-rsqA*luA*luA) 	mulpd   xmm5, xmm6	;# luB*(3-rsqB*luB*luB) 	mulpd   xmm1, [esp + nb312_half] ;# rinv 	mulpd   xmm5, [esp + nb312_half] ;# rinv	movapd [esp + nb312_rinvOO], xmm1	movapd [esp + nb312_rinvOH1], xmm5	movapd xmm0, [esp + nb312_rsqOH2]	movapd xmm4, [esp + nb312_rsqH1O]		cvtpd2ps xmm1, xmm0		cvtpd2ps xmm5, xmm4		rsqrtps xmm1, xmm1	rsqrtps xmm5, xmm5	cvtps2pd xmm1, xmm1	cvtps2pd xmm5, xmm5		movapd  xmm2, xmm1	;# copy of luA 	movapd  xmm6, xmm5	;# copy of luB 	mulpd   xmm1, xmm1	;# luA*luA 	mulpd   xmm5, xmm5	;# luB*luB 	movapd  xmm3, [esp + nb312_three]	mulpd   xmm1, xmm0	;# rsqA*luA*luA 	mulpd   xmm5, xmm4	;# rsqB*luB*luB 		movapd  xmm7, xmm3	subpd   xmm3, xmm1	;# 3-rsqA*luA*luA 	subpd   xmm7, xmm5	;# 3-rsqB*luB*luB 	mulpd   xmm3, xmm2	;# luA*(3-rsqA*luA*luA) 	mulpd   xmm7, xmm6	;# luB*(3-rsqB*luB*luB) 	mulpd   xmm3, [esp + nb312_half] ;# iter1 	mulpd   xmm7, [esp + nb312_half] ;# iter1 	movapd  xmm2, xmm3	;# copy of luA 	movapd  xmm6, xmm7	;# copy of luB 	mulpd   xmm3, xmm3	;# luA*luA 	mulpd   xmm7, xmm7	;# luB*luB 	movapd  xmm1, [esp + nb312_three]	mulpd   xmm3, xmm0	;# rsqA*luA*luA 	mulpd   xmm7, xmm4	;# rsqB*luB*luB 		movapd  xmm5, xmm1	subpd   xmm1, xmm3	;# 3-rsqA*luA*luA 	subpd   xmm5, xmm7	;# 3-rsqB*luB*luB 	mulpd   xmm1, xmm2	;# luA*(3-rsqA*luA*luA) 	mulpd   xmm5, xmm6	;# luB*(3-rsqB*luB*luB) 	mulpd   xmm1, [esp + nb312_half] ;# rinv 	mulpd   xmm5, [esp + nb312_half] ;# rinv 	movapd [esp + nb312_rinvOH2], xmm1	movapd [esp + nb312_rinvH1O], xmm5	movapd xmm0, [esp + nb312_rsqH1H1]	movapd xmm4, [esp + nb312_rsqH1H2]		cvtpd2ps xmm1, xmm0		cvtpd2ps xmm5, xmm4		rsqrtps xmm1, xmm1	rsqrtps xmm5, xmm5	cvtps2pd xmm1, xmm1	cvtps2pd xmm5, xmm5		movapd  xmm2, xmm1	;# copy of luA 	movapd  xmm6, xmm5	;# copy of luB 	mulpd   xmm1, xmm1	;# luA*luA 	mulpd   xmm5, xmm5	;# luB*luB 	movapd  xmm3, [esp + nb312_three]	mulpd   xmm1, xmm0	;# rsqA*luA*luA 	mulpd   xmm5, xmm4	;# rsqB*luB*luB 		movapd  xmm7, xmm3	subpd   xmm3, xmm1	;# 3-rsqA*luA*luA 	subpd   xmm7, xmm5	;# 3-rsqB*luB*luB 	mulpd   xmm3, xmm2	;# luA*(3-rsqA*luA*luA) 	mulpd   xmm7, xmm6	;# luB*(3-rsqB*luB*luB) 	mulpd   xmm3, [esp + nb312_half] ;# iter1a 	mulpd   xmm7, [esp + nb312_half] ;# iter1b 	movapd  xmm2, xmm3	;# copy of luA 	movapd  xmm6, xmm7	;# copy of luB 	mulpd   xmm3, xmm3	;# luA*luA 	mulpd   xmm7, xmm7	;# luB*luB 	movapd  xmm1, [esp + nb312_three]	mulpd   xmm3, xmm0	;# rsqA*luA*luA 	mulpd   xmm7, xmm4	;# rsqB*luB*luB 		movapd  xmm5, xmm1	subpd   xmm1, xmm3	;# 3-rsqA*luA*luA 	subpd   xmm5, xmm7	;# 3-rsqB*luB*luB 	mulpd   xmm1, xmm2	;# luA*(3-rsqA*luA*luA) 	mulpd   xmm5, xmm6	;# luB*(3-rsqB*luB*luB) 	mulpd   xmm1, [esp + nb312_half] ;# rinv 	mulpd   xmm5, [esp + nb312_half] ;# rinv 	movapd [esp + nb312_rinvH1H1], xmm1	movapd [esp + nb312_rinvH1H2], xmm5	movapd xmm0, [esp + nb312_rsqH2O]	cvtpd2ps xmm1, xmm0		rsqrtps xmm1, xmm1	cvtps2pd xmm1, xmm1		movapd  xmm2, xmm1	;# copy of luA 	mulpd   xmm1, xmm1	;# luA*luA 	movapd  xmm3, [esp + nb312_three]	mulpd   xmm1, xmm0	;# rsqA*luA*luA 	subpd   xmm3, xmm1	;# 3-rsqA*luA*luA 	mulpd   xmm3, xmm2	;# luA*(3-rsqA*luA*luA) 	mulpd   xmm3, [esp + nb312_half] ;# iter1 	movapd  xmm2, xmm3	;# copy of luA 	mulpd   xmm3, xmm3	;# luA*luA 	movapd  xmm1, [esp + nb312_three]	mulpd   xmm3, xmm0	;# rsqA*luA*luA 	subpd   xmm1, xmm3	;# 3-rsqA*luA*luA 	mulpd   xmm1, xmm2	;# luA*(3-rsqA*luA*luA) 	mulpd   xmm1, [esp + nb312_half] ;# rinv 	movapd [esp + nb312_rinvH2O], xmm1		;# start with OO interaction 	movapd xmm0, [esp + nb312_rinvOO]	movapd xmm1, xmm0	mulpd  xmm1, [esp + nb312_rsqOO] ;# xmm1=r 	mulpd  xmm1, [esp + nb312_tsc]	cvttpd2pi mm6, xmm1	;# mm6 = lu idx 	cvtpi2pd xmm6, mm6	subpd xmm1, xmm6	;# xmm1=eps 	movapd xmm2, xmm1		mulpd  xmm2, xmm2	;# xmm2=eps2 		pslld mm6, 2		;# idx *= 4 	movd mm0, eax		movd mm1, ebx	mov  esi, [ebp + nb312_VFtab]	movd eax, mm6	psrlq mm6, 32	movd ebx, mm6		;# indices in eax/ebx 	movlpd xmm4, [esi + eax*8]	;# Y1		movlpd xmm3, [esi + ebx*8]	;# Y2	movhpd xmm4, [esi + eax*8 + 8]	;# Y1 F1 		movhpd xmm3, [esi + ebx*8 + 8]	;# Y2 F2 	movapd xmm5, xmm4	unpcklpd xmm4, xmm3	;# Y1 Y2 	unpckhpd xmm5, xmm3	;# F1 F2 	movlpd xmm6, [esi + eax*8 + 16]	;# G1	movlpd xmm3, [esi + ebx*8 + 16]	;# G2	movhpd xmm6, [esi + eax*8 + 24]	;# G1 H1 		movhpd xmm3, [esi + ebx*8 + 24]	;# G2 H2 	movapd xmm7, xmm6	unpcklpd xmm6, xmm3	;# G1 G2 	unpckhpd xmm7, xmm3	;# H1 H2 	;# coulomb table ready, in xmm4-xmm7  			mulpd  xmm6, xmm1	;# xmm6=Geps 	mulpd  xmm7, xmm2	;# xmm7=Heps2 	addpd  xmm5, xmm6	addpd  xmm5, xmm7	;# xmm5=Fp 		mulpd  xmm7, [esp + nb312_two]	;# two*Heps2 	movapd xmm3, [esp + nb312_qqOO]	addpd  xmm7, xmm6	addpd  xmm7, xmm5 ;# xmm7=FF 	mulpd  xmm5, xmm1 ;# xmm5=eps*Fp 	addpd  xmm5, xmm4 ;# xmm5=VV 	mulpd  xmm5, xmm3 ;# vcoul=qq*VV  	mulpd  xmm3, xmm7 ;# fijC=FF*qq     ;# at this point mm5 contains vcoul and xmm3 fijC     ;# increment vcoul - then we can get rid of mm5     ;# update vctot     addpd  xmm5, [esp + nb312_vctot]	xorpd  xmm2, xmm2    movapd [esp + nb312_vctot], xmm5	mulpd  xmm3, [esp + nb312_tsc]	;# start doing lj 	movapd xmm2, xmm0	mulpd  xmm2, xmm2	movapd xmm1, xmm2	mulpd  xmm1, xmm2	mulpd  xmm1, xmm2	;# xmm1=rinvsix 	movapd xmm2, xmm1	mulpd  xmm2, xmm2	;# xmm2=rinvtwelve 	mulpd  xmm1, [esp + nb312_c6]	mulpd  xmm2, [esp + nb312_c12]	movapd xmm4, xmm2	subpd  xmm4, xmm1	addpd  xmm4, [esp + nb312_Vvdwtot]	mulpd  xmm1, [esp + nb312_six]	mulpd  xmm2, [esp + nb312_twelve]	movapd [esp + nb312_Vvdwtot], xmm4	subpd  xmm2, xmm1	mulpd  xmm2, xmm0	subpd  xmm2, xmm3	mulpd  xmm0, xmm2		movapd xmm1, xmm0	movapd xmm2, xmm0			xorpd xmm3, xmm3	movapd xmm4, xmm3	movapd xmm5, xmm3	mulpd xmm0, [esp + nb312_dxOO]	mulpd xmm1, [esp + nb312_dyOO]	mulpd xmm2, [esp + nb312_dzOO]	subpd xmm3, xmm0	subpd xmm4, xmm1	subpd xmm5, xmm2	addpd xmm0, [esp + nb312_fixO]	addpd xmm1, [esp + nb312_fiyO]	addpd xmm2, [esp + nb312_fizO]	movapd [esp + nb312_fjxO], xmm3	movapd [esp + nb312_fjyO], xmm4	movapd [esp + nb312_fjzO], xmm5	movapd [esp + nb312_fixO], xmm0	movapd [esp + nb312_fiyO], xmm1	movapd [esp + nb312_fizO], xmm2	;# O-H1 interaction 	movapd xmm0, [esp + nb312_rinvOH1]	movapd xmm1, xmm0	mulpd  xmm1, [esp + nb312_rsqOH1] ;# xmm1=r 	mulpd  xmm1, [esp + nb312_tsc]	cvttpd2pi mm6, xmm1	;# mm6 = lu idx 	cvtpi2pd xmm6, mm6	subpd xmm1, xmm6	;# xmm1=eps 	movapd xmm2, xmm1		mulpd  xmm2, xmm2	;# xmm2=eps2 	

⌨️ 快捷键说明

复制代码Ctrl + C
搜索代码Ctrl + F
全屏模式F11
增大字号Ctrl + =
减小字号Ctrl + -
显示快捷键?