Clint (and Jafar),
Thanks. The arrays sound very interesting. Reading between the lines I
suspect that they are uniform type but more general than just reals.
I'm afraid that helping with a project that size is beyond the time I have to
offer. I really don't know my way around the code that well and I was a far
better assembler programmer than C. Just lack of exposure. Heck I haven't
had
time to try and recompile Unicon for Windows 64 let alone tinker at that level.
There are other projects I'd like to see/try. Like adding an SSL/TLS layer to
messaging.
As for the blocks, I'd noticed something odd and was seeing what could be
learned from &allocated. The answer is quite a lot if your careful. There
were some other minor surprises like the fact that calculating 2^31-1 creates
2
large integer blocks (one because 2^31 won't fit andanother because 1 is
coerced). It makes sense when you think about it. Or that 16r7fffffff
requires
a large integer - which isn't as clear.
David
________________________________
From: Clinton Jeffery <[email protected]>
To: David Gamey <[email protected]>
Cc: [email protected]; [email protected]
Sent: Wed, October 13, 2010 5:34:56 PM
Subject: Re: [Unicon-group] very curious result with &allocated and real
expressions generating block region allocation
Hmm. For many years we had architectures where sizeof(double) !=
wordsize. And I guess from your data you are running a 32-bit Unicon
for which this is still true. But David, for 64-bit compilers, we
should be able to place the reals in descriptors. It will save more
than the memory savings of 12-bytes-per-real in the heap, it will also
speed up programs by reducing memory accesses. But it will be a lot
of #ifdef's and detail-work, so it will take awhile. Does anyone want
to help?
As an aside, we are currently finishing up arrays as a special case of
the list type in Unicon, with more compact C-compatible
representation. It reduces the memory cost of big lists-of-reals by
even more than sticking reals into descriptors, and is complementary
to this idea.
Clint
On Wed, Oct 13, 2010 at 8:29 AM, David Gamey <[email protected]> wrote:
> I was probably having a flashback to SPITBOL/370 :)
>
> ... where reals like integers could be kept inside the variable descriptor.
> Of course that meant really weird 7 byte reals. Sometimes the underlying
> semantics are so similar they get blurred.
------------------------------------------------------------------------------
Beautiful is writing same markup. Internet Explorer 9 supports
standards for HTML5, CSS3, SVG 1.1, ECMAScript5, and DOM L2 & L3.
Spend less time writing and rewriting code and more time creating great
experiences on the web. Be a part of the beta today.
http://p.sf.net/sfu/beautyoftheweb
_______________________________________________
Unicon-group mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/unicon-group
