More ATI Driver News from Derek Smart

[OpenGL guy]

]Because its running in software emu? Dunno. Ask OpenGL_guy, he might have an answer on what exactly was broken in TnL and subsequently fixed. I'm at a loss, as I do not know all the specifics.

Derek,

Don't make up crap you know nothing about. The driver bug that was causing the hangs in certain games (The Thing, BF 1942 and Mafia(?)) is fixed. Software emu? Give me a break.

I sent you a PM yesterday when you asked about the bug. Did you read it? If so, then why make up this garbage?

If you are trying to hurt ATi with slanderous comments, then you need to step back and take a break.

It was a simple error in a code path that didn't get hit by our testing. End of story.

P.S. Now that I am looking at this "multitexturing" bug, I can report all sort of "interesting" things that Battlecruiser Millenium is doing. Do you want me to post a bunch of stuff about your programming skills, Derek? Or your lack of knowledge of D3D specs?

 

[Joe DeFuria]

if TnL works in software (which defeats the purpose right there) and not in hardware (where its needed the most), its broken.

First of all, TnL (like apparently most of the issues being talked about here) doesn't appear to be a global issue. In other words, the drivers simply don't "revert" to software T&L, therefore the 9700 works. Hardware T&L seems to work just fine for most games.

There is apparently some aspect of T&L that is broken, so that it doesn't work in a game or two. And forcing software T&L is a temporary work around for those cases until the issue is resolved. (Which it apparently has been already).

We all understand your point. (You response was even predicted by Bigus Dickus). The request about shying away from blanket statements was made not because we believe you are technically wrong to assert "your definition" of broken. The request was made in the interests of maintaining a civil discussion. And if you are indeed interested in civil discussion (and quality debates), then it would be helpful if you'd throw everyone a bone and not make blanket statements unless of course they are true in the "blanket" case. Because they tend to incite the fanbois.

Yes, I'm aware that the "problem" can be attacked from the other side...that people not respond to "blanket" statements with emotional rhetoric. However, it's simply more efficient to nip it at the source.

(Edit - spelling / grammar).

 

[Derek Smart [3000AD]]

Dunno what to tell you Dave. If you are running that 09/10 DLL fix from ATI, then you've already got the fix for this.

Hopefully OpenGL_guy will shed some light on this when he shows up later today. I'm tired of being beat up by you bastards and speculation isn't going to cut it.

 

[Joe DeFuria]

P.S. Now that I am looking at this "multitexturing" bug, I can report all sort of "interesting" things that Battlecruiser Millenium is doing.

I would be curious to know, technically, why the issue cropped up. (Although in the interests of sanity, maybe we can try to leave direct commentary on the opinions of Derek's "skillz" out of it. )

 

[Derek Smart [3000AD]]

ok, lets not start THIS again. OK? Lets take this ONE step at a time. ZERO fuzzification, OK?

Don't make up crap you know nothing about. The driver bug that was causing the hangs in certain games (The Thing, BF 1942 and Mafia(?)) is fixed. Software emu? Give me a break.

uhm, NOW I'm confused.

As I understood it, that 09/10 ATI patch fix was for issues with those games *and* something to do with TnL.

I already KNEW about the TnL disabling issue, thats why I asked about it.

Yes, I still have your PM saved and with your permission, would post the contents as you orginally wrote them. You will clearly see that what you confirmed is what I went with.

So, I do NOT know what else you are talking about.

I sent you a PM yesterday when you asked about the bug. Did you read it? If so, then why make up this garbage?

See above

If you are trying to hurt ATi with slanderous comments, then you need to step back and take a break.

Don't be silly. I'm shocked that you would even make such statements!!! Quite frankly, you should be ashamed of yourself for going there. Thats just OUT OF LINE.

It was a simple error in a code path that didn't get hit by our testing. End of story.

I said ZERO FUZZIFICAITON, right? NOW, with that in mind, PLEASE explain EXACTLY what that simple code path was and WHAT it broke. My guess is, it has to do with TnL.

Which invariably means, I'm right and you're wrong and pissed.

P.S. Now that I am looking at this "multitexturing" bug, I can report all sort of "interesting" things that Battlecruiser Millenium is doing. Do you want me to post a bunch of stuff about your programming skills, Derek? Or your lack of knowledge of D3D specs?

Yes please. Be my guest. I like to watch myself laugh. I have an ENTIRE log of things you folks in driver dev are doing and which I have sent to Mike over several months and which ended up with fixes that solved problems with drivers.

So, post yours and I'll post mine. Even trade I'd say.

ALL I care about is what happened to MT on these 9xxx boards. If I'm doing something WRONG in D3D, it shouldn't work on ANY boards, let alone ATIs. And if its something specific to the 9xxx boards, then I need to know how to fix it. Thats your job. Not mine. I just want it to work.

So, I would tread very, very carefully if you decide to go along this path. I'm good at what I do. DAMN good in fact. As such, the end result might NOT be what you are expecting.

Bring it on

EDIT

Just wanted to say, good points Joe.

 

[Dave Baumann]

Dunno what to tell you Dave. If you are running that 09/10 DLL fix from ATI, then you've already got the fix for this.

No, that was unpatched release drivers (7.75).

 

[Joe DeFuria]

So, I do NOT know what else you are talking about.

Let me clear this one up.

Derek, you previous posts in this thread have you speculating that hardware T&L is simply disabeld in Radeon 9700 drivers, and software emulation at least sometimes used, which is why "it works" in some cases.

Obviously Open_GL said something in his PMs to you that he believes should be enough for you to know that at no time is hardware T&L disabled in the drivers an software emu activated. That there was a bug which caused some cases of hardware T&L to fail.

OK? End of confusion. Simple misunderstanding....move on.

 

[ben6]

hey Derek/OpenGL , while I or others might find this slightly amusing , I think we can do without the "here's a list of things wrong with drivers/here's a list of things wrong with BCM" in this thread. Just my opinion after reading 15 pages of posts

 

[Derek Smart [3000AD]]

So, I do NOT know what else you are talking about.

Let me clear this one up.

Derek, you previous posts in this thread have you speculating that hardware T&L is simply disabeld in Radeon 9700 drivers, and software emulation at least sometimes used, which is why "it works" in some cases.

Obviously Open_GL said something in his PMs to you that he believes should be enough for you to know that at no time is hardware T&L disabled in the drivers an software emu activated. That there was a bug which caused some cases of hardware T&L to fail.

OK? End of confusion. Simple misunderstanding....move on.

Yes, you are right about the above.

HOWEVER

1. An unsupported code path can lead to HW TnL not working. It doesn't mean it [HW TnL] is broken*

2. While I cannot disclose the contents of his PM, he basically asserted that HW TnL was never broken.

Now, lets get back to the discussion of what the definition of broken is. To me

doesn't work because some code path is wrong == its broken

And considering that, whatever in that code path that caused it to fail, to the extent of requiring a fix, means thats, well, it was broken.

Just because something would cause an anomaly in some cases and not another, doesn't mean that its not broken. Fer crying out loud, HOW did we get to this?

YMMV

 

[Joe DeFuria]

I think we can do without the "here's a list of things wrong with drivers/here's a list of things wrong with BCM" in this thread.

Actually, I would like to see it, and not for pure entertainment value.

If it can be done constructively, it may very well be a perfect illustration of just how "difficult" it is to pin down these types of issues. How when one party swears it's "the other one's" fault, it actually might be:

1) His own
2) The other party
3) The "fault" of a thrid party
4) All of the above.

For all the finger pointing that goes on whenever problems erupt, seeing exactly why people point the in certain directions, and then working out the issue and identifying where the figers should actually be pointed, could be an insightful exercise.

 

[ExitiosuS]

You know what Ben6, I'd have to say that every last word of these fifteen pages is good reading.
I don't think hearing more information can hurt anyone, now can it? Hehe.
Unless they're sensitive or whatever, in which case, close your eyes and move on to a different site.

 

[ben6]

Let me be clear, constructive discussion about problems, I have absolutely 0 problem with. Civilized discussion about the cause of a problem I have no problem with and in fact welcome.

A he said/she said argument , I have a problem with .

 

[Joe DeFuria]

And considering that, whatever in that code path that caused it to fail, to the extent of requiring a fix, means thats, well, it was broken.

Hey. I agree with you.

THAT BEING SAID, again, the purpose of the request to not make blanket statements (and thus being more precise) is merely one of trying to maintain civil and focused discussion.

Case in point: this thread. Note that at your mentioning of "T&L being broken", multiple posts followed to the extent of basically trying to clarify "in what way? It works here, etc." Note the confusion on all sides about "what was said", inflated emotions,etc.

I couldv'e predicted it. (In fact, I think I did...)

So again, if you are interested in quality discussion then simply try and be more precise with your points. You appear to loath "fuzzification". A statement like "T&L is broken" is about as fuzzy as one can get. Help yourself out and simply be more precise.

 

[ExitiosuS]

I was being facieses, ben6. 8)

 

[Rodéric]

I think I agree with Mr Smart when he says that something that doesn't always work is broken.

At least that's what my boss and clients always thought...

I also agree with Joe Defuria.

 

[ExitiosuS]

I was always told to think of it like a chain... If one link is busted, than it breaks the chain.
Therefore, you have a "broken" chain... or two chains? Hehe.
I hope I don't have to clarify that, because that's just what I've been told by teachers (gah!) about what makes something broken.
Oh well.

 

[Derek Smart [3000AD]]

In case there was any doubt, according to Dictionary.com, here is the definition of broken.

 

[JoshMST]

Hey Dave, I think that asking Derek about the T&L in 3D Mark is a bit of an odd thing. We should probably be asking Mad Onion how exactly it does the T&L. Here is what I am thinking.

Since 3D 2001 SE is a DX8.1 based benchmark, so it probably has 3 settings for T&L: Software, DX7 T&L, and DX8.1 T&L (aka Vertex Shader). Now, since 3D Mark 2001 is able to distinquish between the three types of T&L that graphics chips support, do you know if it automatically loads up the code to use for each type? Or is it using the DX7 T&L for the majority of the tests (minus the VS tests)? I would think that it uses the Vertex Shaders on graphics chips that support it. If that is so, then there is probably nothing wrong with the Vertex Shader on the 9700, but rather the DX7 T&L emulation that the 9700 Vertex Shader does could very well be the broken part. As you know, Vertex Shaders up til now have the hardcoded DX7 T&L running in emulation (which is why in DX7 benchmarks the GeForce 4 MX does better in T&L than the GeForce 4 Ti, the emulation is slightly slower than the real thing)

So, if I were to give an answer off the top of my head, I would say that when 3D 2001 runs on a DX8.1 part, it strictly uses Vertex Shading code, and not hardwired T&L code. One way to test this out would be to load up 3D Mark 2000 and try the software vs hardware T&L (as it is strictly using the old hardcoded T&L)

Please let me know if I am totally off base with this. Because from what I see of BCM, it does not have DX8.1 code, and therefore would not use the Vertex Shader but rather the hardware T&L emulation that the Vertex Shaders support.

OpenGL Guy could probably answer this one.

 

[demalion]

]In case there was any doubt, according to Dictionary.com, here is the definition of broken.

This is a bit facetious. You've commented on your English education, and I believe you, but your comment here ignores the very important nature of context in commnunication.
For example, literal applications of the definition you provide could justify calling every single piece of software "broken", if you were to decide to attempt to justify "Forcibly separated into two or more pieces", or make some other literally justifiable but meangingless application. To argue against such an application, I believe you'd have to agree that the context of the usage and how meaningful it is to communicate an idea is more important than supporting a literal application of a definition.

There are even more interesting definitions that could be used to justify all sorts of useless comments that ignore context and effective communication, and I hope you can recognize atleast that context and communication is important even when it is possible to logically justify a definition. Please recognize that I'm not advocating that such a meaninglessly literal application of the definition is valid, but simply making a point about context and effective communication.

Moving on to the particular definition I think you'd use: "Not functioning; out of order" Now, this definition does not specify a scope. By your application here, there are all sorts of things that can be said to be broken when a common agreement on scope and context is ignored. For example, if I feel free to place things in my own context arbitrarily, I could say BC3K is broken in using W buffers. I could even then proceed to back this up with the comment that DirectX doesn't seem to demand W buffer support, and argue that dependence on it and any abnormal behavior that occurs qualifies the game as broken. I hope you realize that I'm not actually advocating this stance, but illustrating that without agreeing on some common scope or context that someone could argue this.

Now, on to your statement: The problem here is that you have your own scope and context, and I'm not saying it is invalid from your viewpoint. But you're not talking to yourself, you're talking to others. From there it should be evident what others are trying to say, basically that their context is not yours. It just doesn't make sense, to them, and I hope you can see why, for a function that demonstrably works to be called "broken" in this case...after all, you could make the case that ANY driver with ANY bugs is "broken" if you don't apply a reasonable scope to your label.

Another problem: you post a link to this definition, but the scope is not specified there, so it doesn't help resolve this at all, only confuse it, as everyone else is discussing the scope of your statements, not the literal definition you are applying.

Finally, in case you are thinking I'm trying to insult you or flame, or whatever, I wish to make clear that 1) I see a gap in the discussion of this term, a breakdown in communication even 2) I'm posting my take on it in an attempt to clarify what I see as the problem 3) the detail in my posts isn't because I'm trying to treat you as "dumb" or some such, but because I'm trying to be as absolutely clear and specific as I can to get my point across, instead of being vague to start and specifying more detail over several replies.

From there, you can go wherever you want, but I hope it is someplace productive.

EDIT: added "go"

 

[Derek Smart [3000AD]]

Because from what I see of BCM, it does not have DX8.1 code, and therefore would not use the Vertex Shader but rather the hardware T&L emulation that the Vertex Shaders support.

Correct

From there, you can wherever you want, but I hope it is someplace productive.

I just read your post.

I SWEAR if you do that again, I'm gonna......oh wait, you don't live next door. Never mind

No seriously though, if I ever saw a really good fuzzification job, you just posted it. Good job. At the end of the day, I'm going to refer you, once again, to the definition of what BROKEN is, m'kay?