By: blaine (myname.delete@this.acm.org), December 20, 2020 12:40 pm
Room: Moderated Discussions
Konrad Schwarz (no.spam.delete@this.no.spam) on December 20, 2020 8:47 am wrote:
> anon (an.delete@this.n.com) on December 18, 2020 8:29 am wrote:
> > I read that C2X (the next C language revision) will add support for decimal floating point.
> > I know that POWER and SPARC CPUs have hardware support for decimal FP,
> > but are there any plans to include it on future x86 or ARM CPUs?
>
> The x87 supports a decimal floating point format, but I don't know if it is compatible to IEEE decimal floating
> point; the x87 format is a sign bit followed by 18 BCD digits and no exponent, so I expect not.
>
> As an aside, I recently read "The 8087 Primer" by John F. Palmer & Stephen P. Morse from
> 1983 and it is really a good introduction to floating point and the x87 -- I believe Palmer
> was the architect of the 8087. E.g., this clearly states that the round to infinity and
> round to minus infinity rounding modes were designed for interval arithmetic.
Palmer was the architect of the Intel floating point unit. There was a meeting where new guys were asked if they had taken numerical analysis, that how I ended up writing the first floating point test suite for Intel (me and a Tech Kurt?). Never raise your hand. Palmer was one of the nicest people that I ever worked with and I learned so much on the project. Kahan consulted. The first hardware we had to test was the sbc-310. Palmer also wrote a floating point simulator in PLM.
Morse was the architect of the 8086. He is brilliant. The x86 instruction architecture gets a lot of sand kicked on it, but given the constraints for compatibility and the technology of the day, he did a great job.
Indeed round to positive and negative infinities were there to support interval arithmetic. Non biased rounding was introduced also. There was talk of imprecise compares for floating point (that got squashed when I pointed out that it could break the Transitive property for equality.)
> anon (an.delete@this.n.com) on December 18, 2020 8:29 am wrote:
> > I read that C2X (the next C language revision) will add support for decimal floating point.
> > I know that POWER and SPARC CPUs have hardware support for decimal FP,
> > but are there any plans to include it on future x86 or ARM CPUs?
>
> The x87 supports a decimal floating point format, but I don't know if it is compatible to IEEE decimal floating
> point; the x87 format is a sign bit followed by 18 BCD digits and no exponent, so I expect not.
>
> As an aside, I recently read "The 8087 Primer" by John F. Palmer & Stephen P. Morse from
> 1983 and it is really a good introduction to floating point and the x87 -- I believe Palmer
> was the architect of the 8087. E.g., this clearly states that the round to infinity and
> round to minus infinity rounding modes were designed for interval arithmetic.
Palmer was the architect of the Intel floating point unit. There was a meeting where new guys were asked if they had taken numerical analysis, that how I ended up writing the first floating point test suite for Intel (me and a Tech Kurt?). Never raise your hand. Palmer was one of the nicest people that I ever worked with and I learned so much on the project. Kahan consulted. The first hardware we had to test was the sbc-310. Palmer also wrote a floating point simulator in PLM.
Morse was the architect of the 8086. He is brilliant. The x86 instruction architecture gets a lot of sand kicked on it, but given the constraints for compatibility and the technology of the day, he did a great job.
Indeed round to positive and negative infinities were there to support interval arithmetic. Non biased rounding was introduced also. There was talk of imprecise compares for floating point (that got squashed when I pointed out that it could break the Transitive property for equality.)
