nb_kernel232_ia32_sse2.intel_syntax.s

来自「最著名最快的分子模拟软件」· S 代码 · 共 2,152 行 · 第 1/5 页

S
2,152
字号
	movhpd xmm7, [esi + ebx*8 + 40]	movapd 	[esp + nb232_jxO], xmm2	movapd 	[esp + nb232_jyO], xmm3	movapd 	[esp + nb232_jzO], xmm4	movapd 	[esp + nb232_jxH1], xmm5	movapd 	[esp + nb232_jyH1], xmm6	movapd 	[esp + nb232_jzH1], xmm7	movlpd xmm2, [esi + eax*8 + 48]	movlpd xmm3, [esi + eax*8 + 56]	movlpd xmm4, [esi + eax*8 + 64]	movhpd xmm2, [esi + ebx*8 + 48]	movhpd xmm3, [esi + ebx*8 + 56]	movhpd xmm4, [esi + ebx*8 + 64]	movapd 	[esp + nb232_jxH2], xmm2	movapd 	[esp + nb232_jyH2], xmm3	movapd 	[esp + nb232_jzH2], xmm4		movapd xmm0, [esp + nb232_ixO]	movapd xmm1, [esp + nb232_iyO]	movapd xmm2, [esp + nb232_izO]	movapd xmm3, [esp + nb232_ixO]	movapd xmm4, [esp + nb232_iyO]	movapd xmm5, [esp + nb232_izO]	subpd  xmm0, [esp + nb232_jxO]	subpd  xmm1, [esp + nb232_jyO]	subpd  xmm2, [esp + nb232_jzO]	subpd  xmm3, [esp + nb232_jxH1]	subpd  xmm4, [esp + nb232_jyH1]	subpd  xmm5, [esp + nb232_jzH1]	movapd [esp + nb232_dxOO], xmm0	movapd [esp + nb232_dyOO], xmm1	movapd [esp + nb232_dzOO], xmm2	mulpd  xmm0, xmm0	mulpd  xmm1, xmm1	mulpd  xmm2, xmm2	movapd [esp + nb232_dxOH1], xmm3	movapd [esp + nb232_dyOH1], xmm4	movapd [esp + nb232_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 + nb232_rsqOO], xmm0	movapd [esp + nb232_rsqOH1], xmm3	movapd xmm0, [esp + nb232_ixO]	movapd xmm1, [esp + nb232_iyO]	movapd xmm2, [esp + nb232_izO]	movapd xmm3, [esp + nb232_ixH1]	movapd xmm4, [esp + nb232_iyH1]	movapd xmm5, [esp + nb232_izH1]	subpd  xmm0, [esp + nb232_jxH2]	subpd  xmm1, [esp + nb232_jyH2]	subpd  xmm2, [esp + nb232_jzH2]	subpd  xmm3, [esp + nb232_jxO]	subpd  xmm4, [esp + nb232_jyO]	subpd  xmm5, [esp + nb232_jzO]	movapd [esp + nb232_dxOH2], xmm0	movapd [esp + nb232_dyOH2], xmm1	movapd [esp + nb232_dzOH2], xmm2	mulpd  xmm0, xmm0	mulpd  xmm1, xmm1	mulpd  xmm2, xmm2	movapd [esp + nb232_dxH1O], xmm3	movapd [esp + nb232_dyH1O], xmm4	movapd [esp + nb232_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 + nb232_rsqOH2], xmm0	movapd [esp + nb232_rsqH1O], xmm3	movapd xmm0, [esp + nb232_ixH1]	movapd xmm1, [esp + nb232_iyH1]	movapd xmm2, [esp + nb232_izH1]	movapd xmm3, [esp + nb232_ixH1]	movapd xmm4, [esp + nb232_iyH1]	movapd xmm5, [esp + nb232_izH1]	subpd  xmm0, [esp + nb232_jxH1]	subpd  xmm1, [esp + nb232_jyH1]	subpd  xmm2, [esp + nb232_jzH1]	subpd  xmm3, [esp + nb232_jxH2]	subpd  xmm4, [esp + nb232_jyH2]	subpd  xmm5, [esp + nb232_jzH2]	movapd [esp + nb232_dxH1H1], xmm0	movapd [esp + nb232_dyH1H1], xmm1	movapd [esp + nb232_dzH1H1], xmm2	mulpd  xmm0, xmm0	mulpd  xmm1, xmm1	mulpd  xmm2, xmm2	movapd [esp + nb232_dxH1H2], xmm3	movapd [esp + nb232_dyH1H2], xmm4	movapd [esp + nb232_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 + nb232_rsqH1H1], xmm0	movapd [esp + nb232_rsqH1H2], xmm3	movapd xmm0, [esp + nb232_ixH2]	movapd xmm1, [esp + nb232_iyH2]	movapd xmm2, [esp + nb232_izH2]	movapd xmm3, [esp + nb232_ixH2]	movapd xmm4, [esp + nb232_iyH2]	movapd xmm5, [esp + nb232_izH2]	subpd  xmm0, [esp + nb232_jxO]	subpd  xmm1, [esp + nb232_jyO]	subpd  xmm2, [esp + nb232_jzO]	subpd  xmm3, [esp + nb232_jxH1]	subpd  xmm4, [esp + nb232_jyH1]	subpd  xmm5, [esp + nb232_jzH1]	movapd [esp + nb232_dxH2O], xmm0	movapd [esp + nb232_dyH2O], xmm1	movapd [esp + nb232_dzH2O], xmm2	mulpd  xmm0, xmm0	mulpd  xmm1, xmm1	mulpd  xmm2, xmm2	movapd [esp + nb232_dxH2H1], xmm3	movapd [esp + nb232_dyH2H1], xmm4	movapd [esp + nb232_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 + nb232_rsqH2O], xmm0	movapd [esp + nb232_rsqH2H1], xmm4	movapd xmm0, [esp + nb232_ixH2]	movapd xmm1, [esp + nb232_iyH2]	movapd xmm2, [esp + nb232_izH2]	subpd  xmm0, [esp + nb232_jxH2]	subpd  xmm1, [esp + nb232_jyH2]	subpd  xmm2, [esp + nb232_jzH2]	movapd [esp + nb232_dxH2H2], xmm0	movapd [esp + nb232_dyH2H2], xmm1	movapd [esp + nb232_dzH2H2], xmm2	mulpd xmm0, xmm0	mulpd xmm1, xmm1	mulpd xmm2, xmm2	addpd xmm0, xmm1	addpd xmm0, xmm2	movapd [esp + nb232_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 + nb232_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 + nb232_half] ;# iter1 	mulpd   xmm7, [esp + nb232_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 + nb232_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 + nb232_half] ;# rinv 	mulpd   xmm5, [esp + nb232_half] ;# rinv 	movapd [esp + nb232_rinvH2H2], xmm1	movapd [esp + nb232_rinvH2H1], xmm5	movapd xmm0, [esp + nb232_rsqOO]	movapd xmm4, [esp + nb232_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 + nb232_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 + nb232_half] ;# iter1 of  	mulpd   xmm7, [esp + nb232_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 + nb232_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 + nb232_half] ;# rinv 	mulpd   xmm5, [esp + nb232_half] ;# rinv	movapd [esp + nb232_rinvOO], xmm1	movapd [esp + nb232_rinvOH1], xmm5	movapd xmm0, [esp + nb232_rsqOH2]	movapd xmm4, [esp + nb232_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 + nb232_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 + nb232_half] ;# iter1 	mulpd   xmm7, [esp + nb232_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 + nb232_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 + nb232_half] ;# rinv 	mulpd   xmm5, [esp + nb232_half] ;# rinv 	movapd [esp + nb232_rinvOH2], xmm1	movapd [esp + nb232_rinvH1O], xmm5	movapd xmm0, [esp + nb232_rsqH1H1]	movapd xmm4, [esp + nb232_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 + nb232_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 + nb232_half] ;# iter1a 	mulpd   xmm7, [esp + nb232_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 + nb232_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 + nb232_half] ;# rinv 	mulpd   xmm5, [esp + nb232_half] ;# rinv 	movapd [esp + nb232_rinvH1H1], xmm1	movapd [esp + nb232_rinvH1H2], xmm5	movapd xmm0, [esp + nb232_rsqH2O]	cvtpd2ps xmm1, xmm0		rsqrtps xmm1, xmm1	cvtps2pd xmm1, xmm1		movapd  xmm2, xmm1	;# copy of luA 	mulpd   xmm1, xmm1	;# luA*luA 	movapd  xmm3, [esp + nb232_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 + nb232_half] ;# iter1 	movapd  xmm2, xmm3	;# copy of luA 	mulpd   xmm3, xmm3	;# luA*luA 	movapd  xmm1, [esp + nb232_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 + nb232_half] ;# rinv 	movapd [esp + nb232_rinvH2O], xmm1		;# start with OO interaction 		movapd xmm0, [esp + nb232_rinvOO]	movapd xmm7, xmm0		;# xmm7=rinv 	movapd xmm5, [esp + nb232_krf]	mulpd  xmm5, [esp + nb232_rsqOO]	;# xmm5=krsq 	movapd xmm6, xmm5	addpd  xmm6, xmm7		;# xmm6=rinv+ krsq 	subpd  xmm6, [esp + nb232_crf]	;# rinv+krsq-crf 		mulpd  xmm6, [esp + nb232_qqOO]	;# xmm6=voul=qq*(rinv+ krsq-crf) 	mulpd  xmm5, [esp + nb232_two]	;# 2*krsq 	subpd  xmm7, xmm5		;# xmm7=rinv-2*krsq 	mulpd  xmm7, [esp + nb232_qqOO]	;# xmm7 = qq*(rinv-2*krsq) 	mulpd  xmm7, xmm0		addpd  xmm6, [esp + nb232_vctot]	movapd [esp + nb232_vctot], xmm6	movapd [esp + nb232_fstmp], xmm7	;# LJ table interaction	movapd xmm4, [esp + nb232_rsqOO]		mulpd xmm4, xmm0	;# xmm4=r 	mulpd xmm4, [esp + nb232_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 		movd mm0, eax		movd mm1, ebx	mov  esi, [ebp + nb232_VFtab]	movd eax, mm6	psrlq mm6, 32	movd ebx, mm6	;# dispersion 	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 	;# 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  xmm7, [esp + nb232_two]	;# two*Heps2 	addpd  xmm7, xmm6	addpd  xmm7, xmm5 ;# xmm7=FF 	mulpd  xmm5, xmm1 ;# xmm5=eps*Fp 	addpd  xmm5, xmm4 ;# xmm5=VV 	movapd xmm4, [esp + nb232_c6]	mulpd  xmm7, xmm4	 ;# fijD 	mulpd  xmm5, xmm4	 ;# Vvdw6 	;# put scalar force on stack Update Vvdwtot directly 	addpd  xmm5, [esp + nb232_Vvdwtot]	movapd xmm3, [esp + nb232_fstmp]	mulpd  xmm7, [esp + nb232_tsc]	subpd  xmm3, xmm7	movapd [esp + nb232_fstmp], xmm3	movapd [esp + nb232_Vvdwtot], xmm5	;# repulsion 	movlpd xmm4, [esi + eax*8 + 32]	;# Y1 		movlpd xmm3, [esi + ebx*8 + 32]	;# Y2 	movhpd xmm4, [esi + eax*8 + 40]	;# Y1 F1 		movhpd xmm3, [esi + ebx*8 + 40]	;# Y2 F2 	movapd xmm5, xmm4	unpcklpd xmm4, xmm3	;# Y1 Y2 	unpckhpd xmm5, xmm3	;# F1 F2 

⌨️ 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?