Hello,Here is my experience with the DK2 so far. I work with SDL2, OpenGL 3.3 in default profile (compatibility ?) and Windows 7 64 bits for now. I was surprised to get the Direct HMD mode working flawlessly (with this initialisation order : ovr_Initialize, ovrHmd_Create, window and GL context creation, ovrHmd_AttachToWindow, ovrHmd_ConfigureRendering).The only problem that disturbs me is that I can't get sRGB correction when outputing to the HMD in direct-HMD mode (colors are okay in extended desktop mode). I usually set my color textures to GL_SRGB8 format to get free conversion to linear space when reading textures, and when I output to color framebuffers (or FBO with GL_SRGB8 attached textures) I usually use the GL_FRAMEBUFFER_SRGB option.With the rift, I use two SRGB8 textures as framebuffers and use the SDK deformation renderer. The ovrDistortionCap_SRGB flag is set in the ovrHmd_ConfigureRendering call, and I get a correct image on my desktop screen, but not in the HMD (too dark, like if I was outputing the linear colors directly). If I omit this flag, the images on both my screen and the HMD are wrong (again, uncorrected linear colors). And this is in Direct-HMD mode only, I have troubles using extended desktop mode (not related to the SDK) but I can see that colors are correct. If I change the texture formats of my framebuffers from SRGB8 to RGB8 I just lose precision in dark colors (banding).I don't know if the situation is the same with DirectX, but apparently this is a bug in the SDK. Is there some other experience that contradicts mine, or some devs at Oculus who know what's happening in that part of the SDK ? I want to know if this is something that would get corrected in future versions or if I have to do the old trick of sqrt'ing color writes...Thanks !

It might be failing for the default framebuffer if the context initialization is calling SetPixelFormat twice on the same window because this isn't allowed ( see remarks at http://msdn.microsoft.com/en-us/library/windows/desktop/dd369049(v=vs.85).aspx ).The code at \oculussdk\samples\commonsrc\render\render_gl_win32_device.cpp in the Oculus 0.4.2 SDK calls SetPixelFormat twice on the same window/dc, so the second call where it tries to set WGL_FRAMEBUFFER_SRGB_CAPABLE_ARB is probably ignored.

Thanks for the hint. I dug a bit more, and correct me if I'm wrong. If we use the direct-HMD mode, what is displayed on the rift is not the surface where we created a DX/GL context (which is displayed on a window). I have no clear idea how the rift get those frames, I'm not even sure if it's part of the SDK.By the way, the sample demos from Oculus don't seem to use an sRGB framebuffer, so they're fine whatever mode is used.

"ockiller" wrote:I was surprised to get the Direct HMD mode working flawlessly (with this initialisation order : ovr_Initialize, ovrHmd_Create, window and GL context creation, ovrHmd_AttachToWindow, ovrHmd_ConfigureRendering).Thanks a lot for that piece of information! Following that exact order (and eliminating all other calls that might interfere), I finally got direct mode working!However, it vsyncs at 60Hz, which leads to stuttering. :( Did you encounter this problem? (SDK 0.4.3)"ockiller" wrote:The only problem that disturbs me is that I can't get sRGB correction when outputing to the HMD in direct-HMD mode (colors are okay in extended desktop mode). I usually set my color textures to GL_SRGB8 format to get free conversion to linear space when reading textures, and when I output to color framebuffers (or FBO with GL_SRGB8 attached textures) I usually use the GL_FRAMEBUFFER_SRGB option.I don't get correct colors in either mode. Normal (non-Rift) mode works fine, though.Are you using SDL_GL_SetAttribute(SDL_GL_FRAMEBUFFER_SRGB_CAPABLE, 1); ?Same here.

I didn't tried the SDK 0.4.3 yet, I will try it tomorrow. On 0.4.2 it vsyncs at 75 Hz, so when framerate is high enough I don't have stuttering. I overclocked my main monitor to 75 Hz though, it might explain why it works, I didn't tested with it at 60 Hz.In extended mode are you able to set the DK2 to 75 Hz ? Also, try to play with VSync option in nVidia Control Panel. I also put a SDL_GL_SetSwapInterval(0) to try to disable VSync. I think it can interferes badly with SDK VSync.

"ockiller" wrote:In extended mode are you able to set the DK2 to 75 Hz ?I used to get 75 Hz in extended mode - without judder, but with a visible tear line in the right eye."ockiller" wrote:Also, try to play with VSync option in nVidia Control Panel. I also put a SDL_GL_SetSwapInterval(0) to try to disable VSync. I think it can interferes badly with SDK VSync.Thanks! Setting VSync in the nVidia driver to application controlled did the trick (I had it set to adaptive VSync before; never had an application affected by it in Direct Mode).Direct Mode is now perfectly smooth and running at 75 Hz.However, I now get terrible judder in extended mode (even though it's running at 75 Hz)... *sigh*Changing VSync settings via SDL_GL_SetSwapInterval() doesn't help and worst of all, I can't even get back to the more-or-less working state I had before. I'm at this point again, where I get the feeling that either Windows, the Oculus SDK or the nVidia driver is trolling me. :? However, since Direct Mode is more important going forward, I'm going to concentrate on that for the time being. Next step: Get sRGB working: viewtopic.php?f=34&t=16186&p=211893#p211909 [edit]: WORKS!

No sRGB inside HMD (OpenGL, SDK 0.4.3) | Meta Community Forums

20 Replies

sth
Honored Guest
11 years ago
I get the <1fps problem as well after a short while. :(
HartLabs
Honored Guest
11 years ago
Could someone who got OpenGL Direct Mode working provide an example of their initialization code (specifically from ovr_Initialize to ovrHmd_ConfigureRendering)? I have this order but still not functional. I get the same "OVRcreateDXGIFactory2 result 0x0" error popping up in my console that I did in prior SDK versions. An example would help alot!
skynet
Honored Guest
11 years ago
BTW: Is TimeWarp working for you?

Latency testing might be broken with sRGB. In the 0.4.2 code I saw that in the 'latency pixel' render code is trying to 'counter' sRGB encoding in the shader instead of just disabling GL_FRAMEBUFFER_SRGB when rendering that pixel. In conjunction with the other anormalies with sRGB, we could assume that it all doesn't work well.

Anonymous

11 years ago

"HartLabs" wrote:
Could someone who got OpenGL Direct Mode working provide an example of their initialization code (specifically from ovr_Initialize to ovrHmd_ConfigureRendering)? I have this order but still not functional. I get the same "OVRcreateDXGIFactory2 result 0x0" error popping up in my console that I did in prior SDK versions. An example would help alot!

It's not straightforward to give my init code, as it's not in only one place, but it's almost that :

ovr_Initialize();
hmd = ovrHmd_Create(0);
// stripped fallback if hmd == NULL
ovrHmd_SetEnabledCaps(hmd, ovrHmdCap_LowPersistence | ovrHmdCap_DynamicPrediction);
ovrHmd_ConfigureTracking(hmd, ovrTrackingCap_Orientation | ovrTrackingCap_MagYawCorrection | ovrTrackingCap_Position, 0);

Somewhere else ('x', 'y', 'width' and 'height' are set to hmd->WindowPos and hmd->Resolution values) :

SDL_Init(SDL_INIT_VIDEO);
window = SDL_CreateWindow(title, x, y, width, height, SDL_WINDOW_OPENGL | flags);
context = SDL_GL_CreateContext(window);
glewInit();

And then :

    SDL_SysWMinfo info;
    SDL_VERSION(&info.version);
    SDL_GetWindowWMInfo(getWindowState()->window, &info);

    union ovrGLConfig glcfg;
    memset(&glcfg, 0, sizeof(glcfg));
    glcfg.OGL.Header.API = ovrRenderAPI_OpenGL;
    glcfg.OGL.Header.RTSize = hmd->Resolution;
    glcfg.OGL.Header.Multisample = 0;
#if defined(OVR_OS_WIN32)
    glcfg.OGL.Window = info.info.win.window;
    glcfg.OGL.DC = wglGetCurrentDC();
    if (!(hmd->HmdCaps & ovrHmdCap_ExtendDesktop))
        ovrHmd_AttachToWindow(hmd, info.info.win.window, NULL, NULL);
#elif defined(OVR_OS_LINUX)
    glcfg.OGL.Disp = info.info.x11.display;
    glcfg.OGL.Win = info.info.x11.window;
#endif

    for (int i = 0; i < 2; i++) {
        ovrSizei size = ovrHmd_GetFovTextureSize(hmd, ovrEye_Left + i, hmd->DefaultEyeFov[i], pixelDensity);

        frameFBO[i] = createFBO(size.w, size.h, blahblahblah...);

        frameDesc[i].OGL.Header.API = ovrRenderAPI_OpenGL;
        frameDesc[i].OGL.Header.TextureSize.w = size.w;
        frameDesc[i].OGL.Header.TextureSize.h = size.h;
        frameDesc[i].OGL.Header.RenderViewport.Pos.x = 0;
        frameDesc[i].OGL.Header.RenderViewport.Pos.y = 0;
        frameDesc[i].OGL.Header.RenderViewport.Size.w = size.w;
        frameDesc[i].OGL.Header.RenderViewport.Size.h = size.h;
        frameDesc[i].OGL.TexId = frameFBO[i]->color[0]->id;
    }

    if (!ovrHmd_ConfigureRendering(hmd, &glcfg.Config, ovrDistortionCap_Chromatic | ovrDistortionCap_Vignette | ovrDistortionCap_TimeWarp | ovrDistortionCap_HqDistortion | 0*ovrDistortionCap_SRGB | ovrDistortionCap_Overdrive, hmd->DefaultEyeFov, eyeDesc))
        argh();

Only tested on Win 7 x64.

HartLabs
Honored Guest
11 years ago
Thanks! I got it working, kind of. I was calling ovr_Initialize() prior to creating the window but after allocating the window variable. Moving both to after ovr_Initialize() fixed it right up.

It is still not usable though. I am getting 37.5fps with minimal GPU load. Tried all versions of VSync settings in the NVidia control panel. Additionally on the rift (not the mirror) I get double images. I tried to disable mirroring to troubleshoot more but if in initialization I call:

ovrHmd_SetEnabledCaps(program.oculus.hmd, ovrHmdCap_LowPersistence | ovrHmdCap_DynamicPrediction | ovrHmdCap_NoMirrorToWindow);

I crash with the message "Interfering software is preventing your rift from activating...". Any ideas?
sth
Honored Guest
11 years ago
"HartLabs" wrote:
It is still not usable though. I am getting 37.5fps with minimal GPU load.

Sounds like you're calling SwapBuffers manually. This is already done by ovrHmd_EndFrame().
Anonymous
11 years ago
There are some (different ?) performance problems here and there for now. Some have sometime abyssal performance (way below 1 fps) like I did with 0.4.3. I now only create a window (not fullscreen) for direct and extended mode and I don't have this problem anymore. It's also the only way to make things work fine on computers with lower resolution than 1080p.

jspenguin (thanks a lot for the SDK DLLs !) also reported a fairly high cost of the function ovrHmd_EndFrame() especially since 0.4.3. An Oculus dev explained that they actually use 2 sync points to reduce latency, and that can hinder performance somewhat. Better solutions should come in the future. But with V-Sync off, I can't have more than ~160 fps with a spinning cube :( (okay, Core 2 Duo, 9800 GTX..., but still, not much room for nice graphics).
HartLabs
Honored Guest
11 years ago
"sth" wrote:
Sounds like you're calling SwapBuffers manually. This is already done by ovrHmd_EndFrame().

I had this problem when I first implemented the SDK back in 0.4, and the symptoms are identical but there are no extra SwapBuffers in my code. My present code works fine in extended which I would imagine would not be the case if I had an extra swap buffer somewhere.

Any idea how an extra Swap call might be made while in Direct? It seems to me it might have to do with the mirror somehow, but like I said I crash if I attempt to turn it off. I am rendering to a singe shared texture, I could see that somehow causing it as well maybe?

I tried turning off timewarp, dynamic prediction, only rendering one eye, turning off vsync everywhere possible, no change. Seems kind of strange it is not affected if I turn off Vsync.

Hmmm just noticed another weird symptom. If I unplug my second monitor my program hangs. Plugging it back in and it runs. But now plugged back in it has no output even though the system recognizes it.

I am creating a borderless fullscreen window, tried full windowed, fullscreen, no change.
sth
Honored Guest
11 years ago
Strange... Your setup sounds very similar to mine, but I "only" get the "less than 1fps after a while" bug. :?
HartLabs
Honored Guest
11 years ago
"sth" wrote:
I "only" get the "less than 1fps after a while" bug.

I actually just realized I get this bug as well. What I though was the program hang after unplugging my second monitor is actually the <1 FPS bug. I found it is cleared by alt tabbing out of my program. FPS instantly goes right back to 37.5 once that is done. I get this problem even if I boot without the second monitor at all. Strange.

Forum Discussion

🚨 This forum is archived and read-only. To submit a forum post, please visit our new Developer Forum. 🚨

No sRGB inside HMD (OpenGL, SDK 0.4.3)

20 Replies

Quick Links

Other Meta Support

Related Content

Photo Gallery Inside the World

Link just doesnt load inside the headset

Meta XR Simulator does not support OpenGL, while SDK does

[Unity] Can't use controller to select button inside UI scroll

Simple, one page, OpenGL example (0.4.3 beta)