nb_kernel304_ia32_sse2.intel_syntax.s
来自「最著名最快的分子模拟软件」· S 代码 · 共 2,226 行 · 第 1/5 页
S
2,226 行
movapd [esp + nb304_rsqMM], xmm0 movapd [esp + nb304_rsqMH1], xmm3 movapd xmm0, [esp + nb304_ixM] movapd xmm1, [esp + nb304_iyM] movapd xmm2, [esp + nb304_izM] movapd xmm3, [esp + nb304_ixH1] movapd xmm4, [esp + nb304_iyH1] movapd xmm5, [esp + nb304_izH1] subpd xmm0, [esp + nb304_jxH2] subpd xmm1, [esp + nb304_jyH2] subpd xmm2, [esp + nb304_jzH2] subpd xmm3, [esp + nb304_jxM] subpd xmm4, [esp + nb304_jyM] subpd xmm5, [esp + nb304_jzM] movapd [esp + nb304_dxMH2], xmm0 movapd [esp + nb304_dyMH2], xmm1 movapd [esp + nb304_dzMH2], xmm2 mulpd xmm0, xmm0 mulpd xmm1, xmm1 mulpd xmm2, xmm2 movapd [esp + nb304_dxH1M], xmm3 movapd [esp + nb304_dyH1M], xmm4 movapd [esp + nb304_dzH1M], xmm5 mulpd xmm3, xmm3 mulpd xmm4, xmm4 mulpd xmm5, xmm5 addpd xmm0, xmm1 addpd xmm0, xmm2 addpd xmm3, xmm4 addpd xmm3, xmm5 movapd [esp + nb304_rsqMH2], xmm0 movapd [esp + nb304_rsqH1M], xmm3 movapd xmm0, [esp + nb304_ixH1] movapd xmm1, [esp + nb304_iyH1] movapd xmm2, [esp + nb304_izH1] movapd xmm3, [esp + nb304_ixH1] movapd xmm4, [esp + nb304_iyH1] movapd xmm5, [esp + nb304_izH1] subpd xmm0, [esp + nb304_jxH1] subpd xmm1, [esp + nb304_jyH1] subpd xmm2, [esp + nb304_jzH1] subpd xmm3, [esp + nb304_jxH2] subpd xmm4, [esp + nb304_jyH2] subpd xmm5, [esp + nb304_jzH2] movapd [esp + nb304_dxH1H1], xmm0 movapd [esp + nb304_dyH1H1], xmm1 movapd [esp + nb304_dzH1H1], xmm2 mulpd xmm0, xmm0 mulpd xmm1, xmm1 mulpd xmm2, xmm2 movapd [esp + nb304_dxH1H2], xmm3 movapd [esp + nb304_dyH1H2], xmm4 movapd [esp + nb304_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 + nb304_rsqH1H1], xmm0 movapd [esp + nb304_rsqH1H2], xmm3 movapd xmm0, [esp + nb304_ixH2] movapd xmm1, [esp + nb304_iyH2] movapd xmm2, [esp + nb304_izH2] movapd xmm3, [esp + nb304_ixH2] movapd xmm4, [esp + nb304_iyH2] movapd xmm5, [esp + nb304_izH2] subpd xmm0, [esp + nb304_jxM] subpd xmm1, [esp + nb304_jyM] subpd xmm2, [esp + nb304_jzM] subpd xmm3, [esp + nb304_jxH1] subpd xmm4, [esp + nb304_jyH1] subpd xmm5, [esp + nb304_jzH1] movapd [esp + nb304_dxH2M], xmm0 movapd [esp + nb304_dyH2M], xmm1 movapd [esp + nb304_dzH2M], xmm2 mulpd xmm0, xmm0 mulpd xmm1, xmm1 mulpd xmm2, xmm2 movapd [esp + nb304_dxH2H1], xmm3 movapd [esp + nb304_dyH2H1], xmm4 movapd [esp + nb304_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 + nb304_rsqH2M], xmm0 movapd [esp + nb304_rsqH2H1], xmm4 movapd xmm0, [esp + nb304_ixH2] movapd xmm1, [esp + nb304_iyH2] movapd xmm2, [esp + nb304_izH2] subpd xmm0, [esp + nb304_jxH2] subpd xmm1, [esp + nb304_jyH2] subpd xmm2, [esp + nb304_jzH2] movapd [esp + nb304_dxH2H2], xmm0 movapd [esp + nb304_dyH2H2], xmm1 movapd [esp + nb304_dzH2H2], xmm2 mulpd xmm0, xmm0 mulpd xmm1, xmm1 mulpd xmm2, xmm2 addpd xmm0, xmm1 addpd xmm0, xmm2 movapd [esp + nb304_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 + nb304_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 + nb304_half] ;# iter1 mulpd xmm7, [esp + nb304_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 + nb304_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 + nb304_half] ;# rinv mulpd xmm5, [esp + nb304_half] ;# rinv movapd [esp + nb304_rinvH2H2], xmm1 movapd [esp + nb304_rinvH2H1], xmm5 movapd xmm0, [esp + nb304_rsqMM] movapd xmm4, [esp + nb304_rsqMH1] 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 + nb304_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 + nb304_half] ;# iter1 of mulpd xmm7, [esp + nb304_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 + nb304_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 + nb304_half] ;# rinv mulpd xmm5, [esp + nb304_half] ;# rinv movapd [esp + nb304_rinvMM], xmm1 movapd [esp + nb304_rinvMH1], xmm5 movapd xmm0, [esp + nb304_rsqMH2] movapd xmm4, [esp + nb304_rsqH1M] 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 + nb304_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 + nb304_half] ;# iter1 mulpd xmm7, [esp + nb304_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 + nb304_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 + nb304_half] ;# rinv mulpd xmm5, [esp + nb304_half] ;# rinv movapd [esp + nb304_rinvMH2], xmm1 movapd [esp + nb304_rinvH1M], xmm5 movapd xmm0, [esp + nb304_rsqH1H1] movapd xmm4, [esp + nb304_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 + nb304_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 + nb304_half] ;# iter1a mulpd xmm7, [esp + nb304_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 + nb304_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 + nb304_half] ;# rinv mulpd xmm5, [esp + nb304_half] ;# rinv movapd [esp + nb304_rinvH1H1], xmm1 movapd [esp + nb304_rinvH1H2], xmm5 movapd xmm0, [esp + nb304_rsqH2M] cvtpd2ps xmm1, xmm0 rsqrtps xmm1, xmm1 cvtps2pd xmm1, xmm1 movapd xmm2, xmm1 ;# copy of luA mulpd xmm1, xmm1 ;# luA*luA movapd xmm3, [esp + nb304_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 + nb304_half] ;# iter1 movapd xmm2, xmm3 ;# copy of luA mulpd xmm3, xmm3 ;# luA*luA movapd xmm1, [esp + nb304_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 + nb304_half] ;# rinv movapd [esp + nb304_rinvH2M], xmm1 ;# start with MM interaction movapd xmm0, [esp + nb304_rinvMM] movapd xmm1, xmm0 mulpd xmm1, [esp + nb304_rsqMM] ;# xmm1=r mulpd xmm1, [esp + nb304_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 + nb304_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 + nb304_two] ;# two*Heps2 movapd xmm3, [esp + nb304_qqMM] 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 + nb304_vctot] xorpd xmm2, xmm2 movapd [esp + nb304_vctot], xmm5 mulpd xmm3, [esp + nb304_tsc] subpd xmm2, xmm3 mulpd xmm0, xmm2 ;# mult by rinv movapd xmm1, xmm0 movapd xmm2, xmm0 xorpd xmm3, xmm3 movapd xmm4, xmm3 movapd xmm5, xmm3 mulpd xmm0, [esp + nb304_dxMM] mulpd xmm1, [esp + nb304_dyMM] mulpd xmm2, [esp + nb304_dzMM] subpd xmm3, xmm0 subpd xmm4, xmm1 subpd xmm5, xmm2 addpd xmm0, [esp + nb304_fixM] addpd xmm1, [esp + nb304_fiyM] addpd xmm2, [esp + nb304_fizM] movapd [esp + nb304_fjxM], xmm3 movapd [esp + nb304_fjyM], xmm4 movapd [esp + nb304_fjzM], xmm5 movapd [esp + nb304_fixM], xmm0 movapd [esp + nb304_fiyM], xmm1 movapd [esp + nb304_fizM], xmm2 ;# M-H1 interaction movapd xmm0, [esp + nb304_rinvMH1] movapd xmm1, xmm0 mulpd xmm1, [esp + nb304_rsqMH1] ;# xmm1=r mulpd xmm1, [esp + nb304_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 mov esi, [ebp + nb304_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 + nb304_two] ;# two*Heps2 movapd xmm3, [esp + nb304_qqMH] 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 addpd xmm5, [esp + nb304_vctot] movapd [esp + nb304_vctot], xmm5 xorpd xmm1, xmm1 mulpd xmm3, [esp + nb304_tsc] mulpd xmm3, xmm0 subpd xmm1, xmm3 movapd xmm0, xmm1 movapd xmm2, xmm1 xorpd xmm3, xmm3 movapd xmm4, xmm3 movapd xmm5, xmm3 mulpd xmm0, [esp + nb304_dxMH1] mulpd xmm1, [esp + nb304_dyMH1] mulpd xmm2, [esp + nb304_dzMH1]⌨️ 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?