Re: cpu type idea


"Wolfgang Kern" <nowhere@xxxxxxxx> wrote in message
news:g5v39m$5jq$1@xxxxxxxxxxxxxxxxxxxxxxxx

"cr88192" wrote:
...
in part, I was referencing some typical sci-fi cliches in movies and
games, > where very often some piece of machinery exists doing some
fantastical task,
which are usually presented in a certain way:
large spining objects, such as rotors, or those multiple-interconnected
spinning rings (yes, I have seen them do everything from interstellar
transportation to digitizing someone and sending them onto the internet
with
these);
some kind of loud noise, with metalic screeching being most common, maybe
followed by low-pitched rumbling;
usually there is some kind of bright lightsource in the middle (most
often
with a blue or green tint);

When I was about ten, I played with this huge light-bulb styled diodes,
which produced an astonishing blue light cloud if they were overpowered.
And I often find exactly this effect in early and latest sci-fi movies.


maybe that is it...

I had thought it would have been some sort of flourescent, but maybe this
could be it as well...


in general, these have replaced the lightning, jacob's ladders, and
vacuum
tubes that were more common in older movies. the more modern replacement
(typically for far less powerful side devices) being tesla coils, those
'lightning in a piece of plastic' things, and tubes filled with glowing
liquid.

in a lot of newer stuff, what would seem to be flourescent bulbs in some
kind of fancy housing, are also gaining in popularity (an example of this
would be the smaller 'arc-reactor' in the iron-man movie...).
albeit I have seen this approach also used a lot in star-trek and
friends...

then I ended up making a few minor new-age and web-culture references.

for example:

http://www.youtube.com/results?search_query=shoop+da+whoop&search_type=&aq=f

Oh, they actually made a cult out of it...


that and many other things, desu...

http://www.youtube.com/results?search_query=4chan&search_type=
http://www.youtube.com/results?search_query=desu&search_type=&aq=f

....



but, yes, the unrelenting nature and incoherent nature of the posts, and
their occurance on a number of different newsgroups (related to seemingly
arbitrary topics), make me uncertain as to whether the OP is an actual
person, or just some kind of chatbot (actually, I have seen a lot more
coherent and human-like responses made by chatbots in the past, making me
wonder if this would even be a very good one...).

Who knows, not too long ago I saw a suggestion to use A.L.A as a
test-group for social behaviour. Perhaps they now test our response to
'Artificial Nonsense' like 'Flying Bucket' posts and similar.


yes, maybe...



sadly, I don't have much on-topic to say, not having done a whole lot ASM or
compiler related recently.
of well, I have recently noted that both AMD and Intel have their own ideas
for how to extend C and C++ to handle large array-like vectors.

I am left with another idea:
how about we make a "conservative" set of extensions for this kind of thing:

_Vector float a[100], b[100], c[100];

c=a*a+b;

now, how to pull this off is left to the compiler...

this is beyond my personal extensions, which just consist of a few elements
from GLSL onto C's syntax.

it being my idea that we can add constructs that people already use, such as
vector geometry, quaternions, and matrices (of which I have done largely
absent adding any novel syntax).


vector-arrays could go into this set, but this is just how I imagine them:
as vector arrays.

as I see it, we could just change the semantics so that they are
pass-by-value, and add in code such that operations can be parallelized
(leaving many of the optimization details still in the hands of the
compiler).


(or, alternatively, people could just extend GCC's pre-existing albeit
horrid vector syntax to handle much larger vectors, and add more
bulk-parallelization support to GCC...).


now, here is what AMD proposes:

kernel void sum(float a<>, float b<>, out float c<>)
{
c = a + b;
}

int main(int argc, char** argv)
{
int i, j;
float a<10, 10>;
float b<10, 10>;
float c<10, 10>;
float input_a[10][10];
float input_b[10][10];
float input_c[10][10];
for(i=0; i<10; i++) {
for(j=0; j<10; j++) {
input_a[i][j] = (float) i;
input_b[i][j] = (float) j;
}
}
streamRead(a, input_a);
streamRead(b, input_b);
sum(a, b, c);
streamWrite(c, input_c);
...
}

and, personally, I don't see why...


(Intel's option is similar, but is C++ based).


so, as I could imagine the above:
_Vector float[] sum(_Vector float a[], _Vector float b[])
{
c = a + b;
}

int main(int argc, char** argv)
{
int i, j;
_Vector float a[10][10];
_Vector float b[10][10];
_Vector float c[10][10];

for(i=0; i<10; i++) {
for(j=0; j<10; j++) {
a[i][j] = (float) i;
b[i][j] = (float) j;
}
}
c=sum(a, b);
...
}

now, _Vector is a little ugly (but could be largely macro'ed away).

now, it would be expected than when directly manipulating these vectors,
they would be "serialized", and serious limitations would be put on the use
of pointers (probably creating something semantically equivalent, but
processed linearly).

to make it really spiffy, it could support certain kinds of tensor
operations:

for(i=0; i<10; i++)
for(j=0; j<10; j++)
c[i][j]=a[i][]^b[][j];

where A^B is a dot-product (me finding that A*B being a pairwise multiply is
more useful than it would seem, and so A^B is probably better for
dot-product, leaving A*B for the pairwise-multiply operation).

note that it will be expected that the compiler will notice that it can
parallelize these for's as well, but this is just assuming it is not enough
just to parallelize the vectror and tensor operations.


granted that, in any case, vectorization and parallelization, or for that
matter offloading things to the GPU, is likely to involve more than a little
voodoo magic in the compiler.

note that, in the above, the 'for' statements are likely to be
pattern-matched and regarded simply as a statement of the range of values
within the expression, rather than them actually being interpreted as for
statements (basically, it is conceptually similar to loop unrolling and
similar).


or such...


__
wolfgang


.

Relevant Pages

  • Re: getting avg of long and std::vector::size_type
    ... the compiler starts by evaluating a/b. ... int and b is an int, so it performs integer division on a and b. ... Now the compiler starts with the cast. ... first step is to convert a to a float. ...
    (comp.lang.cpp)
  • Re: matrix stuff (solving b = A*x) --> using numerical recipes
    ... Numerical recipes (can be seen on ... Ok, I added it now (actually I also had it at some earlier moment, but then I removed it since I couldn't see any compiler warnings/errors without stdio.h). ... void banmul(float **a, unsigned long n, int left, int right, float x, float b); ...
    (comp.lang.c)
  • Re: problem with cast and unions
    ... Think of it as a compiler for Small programs. ... to a Small "cell" (actually an int). ... can anyone please tell me how is the structure assignment implemented ...
    (comp.lang.c)
  • Re: OT: C++ overloading operators
    ... float *a,*b; ... void bar1(float * restrict a, const float * restrict b, float c, int N) ... If you're using the intel compiler, ...
    (comp.dsp)
  • Re: It Pays to Enrich Your C Skills
    ... Check if you can score a perfect 10 (without using a compiler). ... int main{ ... struct bitfield { ... out if it is a negative integer constant or a constant expression ...
    (comp.lang.c.moderated)