Vincent Diepeveen (diep@xs4all.nl) on 4/20/11 wrote:
---------------------------
>Moritz (Better@not.tell) on 4/15/11 wrote:
>---------------------------
>Where AMD and Nvidia shut up to much about is the actual amount of resources they have to execute executions X.
>For example a conditional move is an instruction you can see from AMD's instruction
>set they have in abundance, but how many of the 4 units can execute this?

My understanding is that non of them can, conditionals or branching is not done by the execution-units.

>There is in the 6900 manual not even a diagram showing the 4 execution units, let
>alone what sort of instructions which unit can execute.

That is because they are all the same now.

>This is something that's very crucial to know for programmers. If you look at agner
>fog, he has measured some latencies on cpu's; there is nothing like that about gpu's.

because latency doesn't matter, the idea is that there is always something else to do.

>Yet they leak so fragmentaric, that it's not even remotely enough to write really
>fast code for a gpu right from scratch.
>Only trial and error brings that.
>Very simple question is for example how fast is a shift-right?
>Then there is an opencl compiler in between that can mess things up, in case of
>AMD; yet i pray it'll get slowly better.

I think they are at it. intel is at an advantage because they have the compiler know-how in-house.
I guess you are not supposed to think that much about the actual code. They tell you how big the memorys are and how wide the different levels of units are, the rest is up to the compiler. Therefore you should write your code on a high level of abstraction and not use tricks making assumptions. Just write "a /= 2;" instead of shifting and hope the compiler is that smart.

>What i really wonder about is the small amount of contract jobs i see where companies/organisations
>ask for capable gpu programmers.

maybe there are just not that many programs that would benefit. Also not every customer has a graphicscard that is supported and faster than a multicore CPU.