Next: Where is balance? -- Re: Academic priorities
From: jsavard on 2 Apr 2005 03:02
Tom Van Vleck wrote:

> (I was not a 650 user, but I did program and operate the
> 7070 in 1962 and 1963. It was my impression that the 650
> was not transistorized and did not have core, two
> advantages claimed for our machine, along with magnetic
> tape which Bemer says was the business requirement.)

Your impression is correct. The 650 used vacuum tubes, and had a
magnetic drum as its main memory.

I don't know if IBM did sneak in a transistorized 650-4, although I
tend to doubt it; in all cases I know of, IBM used a new model number
when they converted an architecture from vacuum tubes to transistors.

John Savard

From: Jeff Kenton on 3 Apr 2005 18:45
Terje Mathisen wrote:
> PDP is interesting. From reading Mike Cowlishaw's description of some of
> its key benefits I figured out how to do it, only it turned out that I
> had generated a nearly but not totally equivalent packing method.
>
> Mike's original version can be implemented with less fixed logic (and/or
> lookup tables), but for a pure sw implementation of decimal, it is
> possible that it would be faster to skip it and simply work in base-1000
> all the time, and then suffer some additional cost for unpacking/packing
> each time a decimal shift of one or two digits was needed.

Or use base-1,000,000,000 which fits nicely within 32-bit words for SW
implementation.
From: Terje Mathisen on 4 Apr 2005 06:23
Jeff Kenton wrote:

> Terje Mathisen wrote:
>
>> PDP is interesting. From reading Mike Cowlishaw's description of some
>> of its key benefits I figured out how to do it, only it turned out
>> that I had generated a nearly but not totally equivalent packing method.
>>
>> Mike's original version can be implemented with less fixed logic
>> (and/or lookup tables), but for a pure sw implementation of decimal,
>> it is possible that it would be faster to skip it and simply work in
>> base-1000 all the time, and then suffer some additional cost for
>> unpacking/packing each time a decimal shift of one or two digits was
>> needed.
>
>
> Or use base-1,000,000,000 which fits nicely within 32-bit words for SW
> implementation.

The is afaik the standard approach of the perl bignum libs.

The problem, as seen from the DPD world, is that it gives away two bits
in each 32-bit block, but this is only one additional decimal digit in a
64-bit word. (18 vs 19 digits)

So if you happen to have an application that uses a bunch of 19-wide
decimal numbers, you have three alternatives:

a) You can allocate 96 bits, allowing up to 27 digits.

b) Use DPD with the useful property that the last 4 bits encode one
additional digit.

c) Simply use base-1e19 which does fit inside a 64-bit word.
:-)

I.e. I'm firmly in the 'pick a suitably large decimal base and work with
standard binary numbers' camp. :-)

Terje

--
- <Terje.Mathisen(a)hda.hydro.com>
"almost all programming can be viewed as an exercise in caching"
From: jsavard on 4 Apr 2005 09:13
Terje Mathisen wrote:

> I.e. I'm firmly in the 'pick a suitably large decimal base and work
with
> standard binary numbers' camp. :-)

That certainly makes much more sense in software.

This kind of encoding is much faster in hardware, and it is intended
for computers that have actual decimal hardware, to allow them to
rapidly store data in a compact format.

John Savard

From: Terje Mathisen on 4 Apr 2005 12:38
jsavard(a)ecn.ab.ca wrote:

> Terje Mathisen wrote:
>
>>I.e. I'm firmly in the 'pick a suitably large decimal base and work
>> with standard binary numbers' camp. :-)
>
> That certainly makes much more sense in software.
>
> This kind of encoding is much faster in hardware, and it is intended
> for computers that have actual decimal hardware, to allow them to
> rapidly store data in a compact format.

Does it really save any time and/or hw resources?

If all your math is add/sub, then decimal is relatively efficient, but
if you have to do quite a bit of multiplication as well, i.e. rate
calculation with 5 decimals, then it would seem that you need a _lot_ of
decimal multipliers?

Terje

--
- <Terje.Mathisen(a)hda.hydro.com>
"almost all programming can be viewed as an exercise in caching"
First  |  Prev  |  Next  |  Last
Pages: 1 2 3 4 5 6 7
Next: Where is balance? -- Re: Academic priorities