Emulator

I updated the emulator and had the same problem.
He begins to appear exactly from r5025. Before it was just uncheck the "Enable CPU -> EFB access and everything was normal.

Core2Duo E5400
4GB RAM
video onboard Intel(R)G33/31 Express Chipset Family

Sorry, is r6025.

More details:

.--------------------------------------------------------------------------.

Issue 3274 has been merged into this issue.

This is not a dupe issue, skid. I do not have an on-board GPU.

Both this issue and issue 3274 are reporting that r6025 causes graphical issues. The reporting stating that on-board GPU's are affected is extra information.

Maybe we can narrow down what my GPU has that some on-board cards have (or doesn't have)that is causing this. I can provide more info if needed.

I'm using the VisionTek Radeon HD3650, 1GB DDR2 with the Catalyst 10.9 AGP Hotfix.

fwiw, what jhonn.copperfield said above kind of doesn't make sense... r6025 only enabled a code path which was disabled before if EFB access is disabled as well. So nvidia onboard gpus should either work or at least fail with EFB access enabled before r6025.

However, if that is not the case, this issue makes even less sense to me...

Sorry, NeoBrainX, the table submited was wrong, correct is:
More details:

.-------------------------------------------------------------------------------.

Yeah, that certainly makes a lot of more sense...
Still, I don't get what's so bad about creating an additional texture.
I'll create a test program for you to check why the driver fails to create the textures... Might take a few days though.

Ok, I will be waiting for your test program.

I'd like to test this as well seeing how my video card that is not on-board and is capable of good performance) suffers from this problem also.

Recently I've made some tests and discovered more details:

1. This issue occurs with other on-board GPUs manufacturers too, not only Intel GMA and nVidia.

2. As Tommy confirmed, some off-board GPUs have this issue too (generally the low-end GPUs, but the same GPUs has a higher performance than the on-board GPUs if you test them in r4598 or older)

3. I've tested Zelda Twilight Princess on Dolphin r6482 with a Intel GMA and new errors appear, not just the Framebuffer Manage error (see attachments)

Fwiw, could you guys with the problem run dxdiag and tell me what D3D version your GPU supports?
run dxdiag, select the "display" or "graphics" tab and tell me the DDI-version.

By Wikipedia informations, the Intel GMA 945G (the GPU that i've used to run the tests of comment 22), support Pixel Shader 2.0 (3.0 by software), OpenGL 1.4 and DirectX 9.0c.

The DxDiag say that the video card DDI version is 9Ex and the driver is WHQL certified. I've attached the log generated by DxDiag, if you want to obtain more information.

10.1

One thing I noticed is in the screenshots of the warning that have been posted so far, either the width or the height is an odd number. Maybe simply rounding the framebuffer dimensions up or down to an even number would fix it? I think they could test this by either running in fullscreen or changing the Window Size in the Display options to have even numbers in both fields.

d'oh... it's actually pretty obvious.
If the hardware doesn't support NPOT2 textures it's clear why this fails...

I always thought this would be an issue in dx11 as well (where all GPUs need to support NPOT2 textures), but apparently everyone here was using dx9-only...

Should have a fix soon.

A fix for it will cause regression in DX9 (some slowdown or break compatibility with some game)?

Oh yeah, lack of NPOT2 support sounds like it could be the reason. I don't know why I thought certain cards had "NMOT2" issues instead.

Actually, another reason I can think of is, maybe those cards don't support any of the 5 depth formats that are currently listed in the FramebufferManager constructor? I noticed that D3DFMT_D16 is not among them, and I think certain cards only support that one. That would make some sense too because it would only affect nVidia cards since ATI cards would usually support FOURCC_DF16 instead, which is already listed.

We could implement a fallback system (only drop NPOT2 textures support if errors are found, otherwise use the normal NPOT2 textures as is now)

... or just check the D3D9 device caps ...

idk, there's not much we can do if the hw doesn't support the depth formats we need.. we could just disable efb access altogether maybe

Checking the device caps sounds kind of unreliable to me, actually, since it's not clear whether some cards that don't normally support NPOT2 textures make a special exception for textures that are associated with the frame buffer, and there's also this "conditional" NPOT2 support cards can say they have, which maybe we can assume means it's supported for all cases we care about but I'm not completely sure on that. (Mine says POW2=yes, NONPOW2CONDITIONAL=yes, and I haven't run into this sort of failure so far.)

The current depth format of last resort is D3DFMT_D24X8, but it looks like if that format gets used, EFB access acts (partially?) disabled. I'm not sure what the point of the D24X8 format is if we can't use it to get depth information like the other formats, but if we can actually use it then I would think we could use D16 too. (My card doesn't support either of those so I can't check how they behave at the moment.)

Recheck with r6725 plskthxbye.

I've tested many GameCube games and Super Mario Galaxy on Dolphin r6725 with a Intel GMA and the same errors occur, but Framebuffer Manage error now have a double size (see attachments). I've noticed too that many textures gets rendered green (in all games, but in ZTP and SMG they are more visible).

Little correction, I've tested on Dolphin r6728, but the same issues occur on r6725.

Working on a proper fix, give me a few minutes...

I forgot something, the green textures can be fixed by disabling "Emulate format changes".

Alright, try again with r6730 :)

Ok, just waiting Mamario build to test it :D

Same as in r6728 or r6725, the only difference is the FrameBuffer Manager error that now is on line 96 instead of line 80.

r6731?

Same issues, the only thing is that the Frame Buffer Manager error was replaced by the alert "No supported depth format found, disabling depth buffers. Expect glitches."

Erm, what happens if you use dx11 anyway?

uh, someone else posted this here before and for whatever reason removed it again:
"According to intel gma9xx specs: In Direct3D, depth textures are not supported. "

No way to fix this bug then, sorry. Still wondering how DX11 behaves, or does the gma9xx not even support DX 10.0?

With DX11 two errors occur (see attachments), then Dolphin crash ("Dolphin.exe was stopped working").

Is it any different if you set the fullscreen resolution to your desktop resolution AND enable "start renderer in fullscreen"?

No, occur the same as in comment 35 and comment 47.

But Enable CPU -> EFB access changes of r6025 that caused this issue, before r6025 Dolphin works without these glitches if you disable Enable CPU -> EFB access. I think that r6025 would be reverted or a patch (or hack) would be implemented.

done, r6737.

I think that is fixed, sometimes a Pixel Shader error appear but apparently no glitches are caused (see attachments). Some games also have slowdowns with Dolphin saying that they are running at 100%, but I think that this is another issue not related to this. Thanks NeoBrain.

uh, someone else posted this here before and for whatever reason removed it again:
"According to intel gma9xx specs: In Direct3D, depth textures are not supported. "

Yes, it was me. Removed the entry because the spec was actually for intel gma945 and not for the 965.
But as it seems... intel still hasn't changed this :( (Well the g45 should support depth textures now)

Only for notice, r6731 fixed this issue on nVidia on-board cards.