Exophase (exophase@gmail.com) on 4/27/12 wrote:
---------------------------
>David Kanter (dkanter@realworldtech.com) on 4/26/12 wrote:
>---------------------------
>>Maybe just do LD-->LD in the same pipe?
>
>Hm, okay. So the AGU could at least forward to itself (I guess that was implied?)
>Still, wouldn't being able to forward from AGU to AGU at least offer theoretically
>less execution resource pressure in some scenarios? Not saying they should have
>done it, was just wondering if it'd really be worth >nothing.

That's what I'm asking...what is the situation where you think it would help? And how common is it?

>>JF is a nice guy, but relying on him for architectural >>details is a bit silly.
>>
>
>I don't think it's silly to believe that a relatively high level and public employee
>at AMD would be passing on accurate technical information, >especially with no disclaimer
>up front in his posts that he isn't an engineer.

JF is highly visible, he's definitely in marketing. Those two facts alone indicate he's not likely to be a source of good information.

>His blog >posts even have a deliberate
>function of attempting to convey technical information. I don't think one should
>expect much in the way of low level details but it's reasonable to expect the things
>he says to not be blatantly false. AMD should have probably been more careful about
>what they let him post, and when that included >misinformation he should have been urged to correct it.

Even his informative posts aren't very technical and only talk about vague generalities. If you compare that to posts by Mark Buxton at Intel (http://software.intel.com/en-us/blogs/2009/01/29/recent-intelr-avx-architectural-changes/), there's quite a contrast.

In a large part, that's because Mark Buxton is an engineer, while John Fruehe is not. That doesn't make one evil or good, but it does mean that one is a much more reliable source of information concerning architecture.

>>Actually the AGU is quite useful for a design where you can't re-order.
>
>Yeah, just so you know where I'm coming from I thought this also because JF said
>the third AGU was useless >_> (yes, I do get what you said about listening to him
>;P) Which I took to mean that the two load/store pipelines were closely coupled
>to two AGUs and the third one couldn't actually submit >results to it.

>If all three AGUs can forward to two L/S paths then it's >useful even with memory
>reordering, as is the case in Sandy Bridge, right?

Absolutely. In fact, it's more useful in the context of the K8 than Sandy Bridge. The K8 cannot re-order so you REALLY want to calculate every address as soon as possible.

David