joel@gould9.UUCP (Joel West) (12/03/85)
This is probably a trivial question -- I don't currently have a Sun-3, but may evenutally port software to one. I understand two Sun-3 configurations are distinct in their approach to floating point arithmetic: * 68020 + 68881 * 68020 + Weitek (?) FP board The second is somewhat faster and is expected to be more popular, I'm told. Are executables for these two variants binary compatible, i.e., does the compiler have to know what machine it is targetting? If so, I assume this means the FP board is emulating the 68881 instruction set, or some such. The machine-language equivalent of the following VAX sequence for one or both configurations would be appreciated. /* A = B * C + D */ movd B,r0 muld2 C,r0 addd2 D,r0 movd r0,A -- Joel West (619) 457-9681 CACI, Inc. Federal, 3344 N. Torrey Pines Ct., La Jolla, CA 92037 {cbosgd,ihnp4,pyramid,sdcsvax,ucla-cs}!gould9!joel gould9!joel@nosc.ARPA
henry@utzoo.UUCP (Henry Spencer) (12/04/85)
> I understand two Sun-3 configurations are distinct in their > approach to floating point arithmetic: > * 68020 + 68881 > * 68020 + Weitek (?) FP board > ... > Are executables for these two variants binary compatible...? Almost certainly not, since the Weitek boards I know about pre-date the introduction, and probably the design, of the 68881. There has also been some recent commentary about what the Weitek designers should learn from the 68881, which again strongly suggests the two are not compatible. -- Henry Spencer @ U of Toronto Zoology {allegra,ihnp4,linus,decvax}!utzoo!henry
henry@utzoo.UUCP (Henry Spencer) (12/08/85)
> > Are executables for these two [Sun 3] variants binary compatible...? > > Almost certainly not, since the Weitek boards I know about pre-date the > introduction, and probably the design, of the 68881... Turns out I was right, but for the wrong reasons. The 68881 interface has been frozen for much longer than I thought, and the Sun people who designed the fast-floating-point board using the Weitek chips knew about it. But they opted for a more specialized interface for higher speed, so the Sun-3 Weitek floating-point option is *not* binary compatible with the 68881 floating point. That word comes from multiple sources within Sun, so I think it's definitive. -- Henry Spencer @ U of Toronto Zoology {allegra,ihnp4,linus,decvax}!utzoo!henry
andrew@aimmi.UUCP (Andrew Stewart) (12/10/85)
In article <196@gould9.UUCP> joel@gould9.UUCP (Joel West) writes: >I understand two Sun-3 configurations are distinct in their >approach to floating point arithmetic: > * 68020 + 68881 > * 68020 + Weitek (?) FP board >The second is somewhat faster and is expected to be more popular, I'm told. > >Are executables for these two variants binary compatible, ..... Yes - as I understand it, the SUN software checks for the FPA board at boot time. If it's there, it ships stuff over the fast local bus to the FPA instead of the 68881 chip. The clever bit is that the FPA doesn't need to emulate the 68881 - in certain out-of-range cases, the FPA board gives up in disgust, so the CPU passes the sum to the 68881 which gives the correct-according-to-IEEE answer. Since this happens in a very small % of cases, the speed loss is trivial. And the binaries are compatible. Neat. -- ------------------------------------------- Andrew Stewart UUCP: andrew@aimmi.UUCP ARPA: andrew@aimmi.ac.uk "Once the subject is clearly understood, the explanation will seem sensible."
kds@intelca.UUCP (Ken Shoemaker) (12/13/85)
> Yes - as I understand it, the SUN software checks for the FPA board at > boot time. If it's there, it ships stuff over the fast local bus to the > FPA instead of the 68881 chip. > The clever bit is that the FPA doesn't need to emulate the 68881 - in certain > out-of-range cases, the FPA board gives up in disgust, so the CPU passes the I would think that there are still some problems with this, with respect to round-off models, but oh well...Can anyone explain to me how this works? If the FPA board is there, does the 68020 trap on every floating point op and ship the stuff over by hand, or is there something in the coprocessor interface that the startup code sets to make the 020 use the FPA rather than the 881? As they say, pardon my ignorance... -- yes, some uncomplicated peoples still believe this myth... Ken Shoemaker, Santa Clara, Ca. {pur-ee,hplabs,amd,scgvaxd,dual,qantel}!intelca!kds ---the above views are personal.