FP blending in R420?

[Dave Baumann]

Well, let's see how Mint reacts. As for wary, I am anonymous here and don't represent any company. I'll be the first to admit I am a real bastard sometimes. But I'm a real bastard whose behavior doesn't reflect on my company.

And you are being a real bastard that may prevent useful information coming on this forum. The course of action you are taking at the moment, I can guarantee, is not one that will help promote a response. Regardless of whether you like OpenGL guys posting style or not he is a route to direct information and that is more useful to more people that frequent this forum than your personal feelings.

 

[dksuiko]

You call this

OpenGL guy, I was hoping you would be able to answer this. Could you try to find out?

Rude and disrepectful?!?!

Of course it's rude. How dare Mint ask OpenGL Guy a question without first basking him with compliments on how cool it must be to work at ATI and to know what he knows. </sarcasm> You're not the only one annoyed with OpenGL's attitude. Yeah, somtimes makes he informative posts - but many times it seems like he's only posting to throw jabs and let you know he works at ATI. A stark contrast to sireric's posts.

 

[martrox]

DC, while I won't comment on wether you are a bastard or not, I can say for certain you are single minded and and bullheaded....... and for the most part, pretty thick skinned - thank the powers that be! However, there are times when it is best to just go against your demons and just stay out of the storm.... and this is probably one of those times..... If only for B3d's sake.....

 

[Evildeus]

This is going of topic, but i think plenty of people would need to be cut-off then

 

[DemoCoder]

It's not direct route if one won't answer people's polite questions because at some time in the past, they said something one perceived as negative towards ATI HW.

Anyway, I'll take your advice martox.

Maybe we can get this thread back to the earlier discussion, which is the performance of FP blending vs having to do blends in the pixel shader when you lack it.

 

[OpenGL guy]

You call this

OpenGL guy, I was hoping you would be able to answer this. Could you try to find out?

Rude and disrepectful?!?!

OpenGL Guy is the one who has become rude and disrepectful. Many of his posts lately have become snide sniping remarks at people and I have received PMs to that effect that from several forum members who were frankly surprised that an ATI employee has been acting like that in public forums.

Go ahead and take my comments out of context. Go ahead and misrepresent what I was referring to.

Mintmaster, send me a PM if you have a problem with what I said.

 

[martrox]

Thanks, DC......

 

[Mintmaster]

Holy shit! A lot can happen here in a few hours! I was about to answer soon after OpenGL guy's post with a pseudo apology, but look what happened now!

DC, Dave is kinda right. I appreciate the support (though I don't know what to think when you say I'm the anti-Chalnoth - I like to think my arguments make a lot of sense and have no bias, not that I'm an ATI version of Chalnoth), but you don't want to make things worse than they are. That's the basis for some of my political views as well, as you may have noticed.

Well, let's see how this goes now...

So it looks like most people are pretty sure there's no FP blending, but it's not confirmed yet. OpenGL guy, I was hoping you would be able to answer this. Could you try to find out?

With comments like:

This sucks. R300->R420 is almost as bad as GF3->GF4. You get a massive performance leap, but you only get a couple of new features (longer instruction length, 3Dc) that in my book aren't very important.

Why should I feel compelled to answer anything?

I know the answer, there is no "finding out".

Sorry, didn't mean to disrespect your work or ATI's. I've just been thinking about HDR for a while now and then, trying to get a good general solution without FP blending. I think my earlier suggetion of doing alpha blending on the final 8-bit per channel buffer is good enough for now.

I had always hoped that there was I16 blending, but when some ATI reps told me face to face it wasn't possible, is sort of felt like a dagger to my hopes of HDR support in the near future. Guess I was just venting my frustration.

But remember, I never saw a big need for long instruction support way back when R300 was out and later when NV3x came out, so it would be sort of hypocritical for me to support it now, right? And 3Dc is great for a nice incremental IQ/perf boost, but it doesn't let me do anything new.

I was exaggerating in my GF3/GF4 comparison, as the performance boost is much, MUCH greater, and software developers are even further behind the hardware now than they were then. ATI has every right to skip adding new features this time around given the amount they got shafted by developers in the past wrt the features of the Radeon and Radeon 8500.

In light of my support for ATI in the past (and the fact that I worked there 2 summers), can you please just clear some things up regarding blending? Is I16 blending supported in R3xx? R420? If so, why isn't it supported in DX, or is that an old issue that's now resolved?

 

[DemoCoder]

Mint, I meant anti in the sense of little to no favoritism. Just clearing that up.

 

[Hyp-X]

I don't like the "SM" approach at all. We are talking two different things:

PS2.0 vs PS3.0
Don't expect any visual difference here. PS3.0 is more flexible and it will be easier to develop on PS3.0 capable hardware - but at the end of the day (read at the optimization stage) it might turn out that most of the shaders run faster with the ps_2_a profile.
So it won't affect consumers that much - if at all.

First of all, consider that PS 3.0, with current hardware, effectively also means support for FP blending/filtering (ATI will apparently also include FP blend support on their SM 3.0 parts).

No it does not mean that either effectively or not.
It's two different things.
The effects that need FP blending/filtering are not neccessarily the same that need PS3.0.

And I do think that FP blending/filtering is much more useful feature of the NV40 than its PS3.0 support...

Regardless, some PS 3.0 shaders will just be unfeasible to run in PS 2.0. While it may be fundamentally possible to do the same math, it will be vastly too slow to do practically. These shaders won't appear in games for some time, though, but that still only makes your statement true in the near-term.

Near term: before SM4.0 cards will arrive right?

And don't forget that PS 3.0 shaders are easier to develop for, given the additional freedom available, and thus we may see game developers start developing PS 3.0 shaders with advanced effects and not bothering to write the PS 2.0 fallback (for example, if it exceeds one of a number of limits seen in PS 2.0) with equivalent quality.

Not all will write PS2.0 fallbacks only the ones planning to stay in this business

And you don't develop PS3.0 shaders or PS2.0 shader.
You write them in HLSL and compile.
Of course there are hitting limits but that have to be dealt with regardless of PS level (see R300 vs R420)

And I don't believe for an instant that any shaders will run faster with the PS_2_a profile than they would with a PS_3_0 profile.

Given the penalties of dynamic branching that shouldn't be a surprise...

 

[Hyp-X]

I had always hoped that there was I16 blending, but when some ATI reps told me face to face it wasn't possible, is sort of felt like a dagger to my hopes of HDR support in the near future.

Well, I still say there is "I16" blending in R300.
I know I used it - unless it was removed from newer drivers it's there.

The only reason the driver doesn't expose D3DUSAGE_QUERY_POSTPIXELSHADER_BLENDING for those formats is because there's no support for all of the stuff that required for that (eg. there's no alpha test support).

If you want information use the dx list or fire an e-mail to a devrel guy like rhuddy.
That's much more effective than trying to pump OpenGL Guy for info here.

 

[Mintmaster]

Thanks Hyp-X, I'll give it a shot. Seems funny that there's no alpha test support. That's a lot easier to do than blending, AFAIK.

As for devrel, well, like I said earlier, ATI reps at ATI Mojo Day Reloaded told me face to face (after some deliberation with each other, no less) that it's not supported, so I don't see how devrel would give me any more accurate of an answer. They likely have plenty of emails from more important people, especially at this time of year, and I doubt the real experts would answer me.

I16 will do just fine for me, especially if it's supported in R300.

 

[LeStoffer]

Edit: Nevermind - See post below...

 

[OpenGL guy]

Thanks Hyp-X, I'll give it a shot. Seems funny that there's no alpha test support. That's a lot easier to do than blending, AFAIK.

As for devrel, well, like I said earlier, ATI reps at ATI Mojo Day Reloaded told me face to face (after some deliberation with each other, no less) that it's not supported, so I don't see how devrel would give me any more accurate of an answer. They likely have plenty of emails from more important people, especially at this time of year, and I doubt the real experts would answer me.

I16 will do just fine for me, especially if it's supported in R300.

The CAPS bits we expose is what the HW supports. I16 and FP16 blending is not supported for R3x0 or R420.

Hyp-X, it's possible that the driver was not disabling blending for surfaces that don't support blending, but don't count on the blend being correct in these cases.

 

[Geo]

Btw, we should assume "compelled" was an unfortunate choice of words, rather than a modus operandi for your participation? I'd hate to be walking around with an image of you under the lash around here.

 

[Mintmaster]

Thanks OpenGL guy. I guess I shouldn't waste time trying it out.

 

[Hyp-X]

Hyp-X, it's possible that the driver was not disabling blending for surfaces that don't support blending, but don't count on the blend being correct in these cases.

So, it is/was an incorrect behaviour?
Kinda like the FP "pseudo" filtering?

It might have done bytewise alpha blending ... I have to think about it but it still might have limited usability.

Btw, I have only tried D3DFMT_G16R16, not sure about D3DFMT_A16B16G16R16 at all.

 

[ben6]

Thanks for clearing it up OpenGL Guy

 

[OpenGL guy]

Hyp-X, it's possible that the driver was not disabling blending for surfaces that don't support blending, but don't count on the blend being correct in these cases.

So, it is/was an incorrect behaviour?
Kinda like the FP "pseudo" filtering?

The driver was not disabling filtering on FP surfaces, because it assumed the application would do the right thing. This was corrected some time ago. I didn't check the driver's behavior with blending, but what I stated stands: No FP or I16 blending is supported on R3x0 or R420.

 

[pat777]

Does anyone know if R420 supports:
- FP16 blending?
- FP32 (well actually, FP24 internally) blending?
- Blending with other formats, like I16?
- FP filtering?

I'm sure the R420 only supports the pixel formats the R3xx supported.

And by the way, FP framebuffer blending is required for complete support of HDR rendering. If you want to blend while doing HDR rendering otherwise, you would need to write to a separate texture each time a draw call is made (and you'd have to guarantee that your draw call doesn't overlap itself), reading that texture into the next pass. The performance hit wouldn't be remotely bearable for such a situation.

The only way around this is to severely limit yourself as to what you can render when doing HDR rendering.

Also, without geometry instancing the R420 has no way of reducing draw calls. Then again, Half-Life 2 got HDR to work well on X800 XT.