Torque3D OpenGL Status

#101 @Timmy:
This may very well be an extreme corner-case, but on large DXT(1,3 or 5 at 512X8192 resolution) texture sheet, the

else if((format == GFXFormatDXT1 || format == GFXFormatDXT3 || format == GFXFormatDXT5) && numMipLevels >0)

chunk caused the objects sharing it to turn black.
attempted the resolution listed: http://www.opengl.org/discussion_boards/showthread.php/166930-DDS-mipmapping-doesn-t-work

via altering that chunk to:

retTex->mMipLevels = numMipLevels;
glTexParameteri(binding, GL_GENERATE_MIPMAP_SGIS, GL_FALSE);
glTexParameteri(binding, GL_TEXTURE_MAX_LEVEL, numMipLevels);

showed no difference, but I'm not entirely certain that the present workaround of tossing in:

glTexParameteri(binding, GL_TEXTURE_MAX_LEVEL, 0);

doesn't simply short-circuit the entire thing.

#102 @Azaezel:

Yep i believe glTexParameteri(binding, GL_TEXTURE_MAX_LEVEL, 0) would completely bypass the mip maps. Technically we should be settings the max level as it can cause problems on some drivers.

Are you on amd or nvidia?

I'll try and reproduce the problem, just so i understand to reproduce the problem i need a 512X8192 DXTn compressed texture with mip maps and have a few static mesh sharing the same material?

Just one last thing, could you change the part of the code

// Complete the texture
glTexParameteri(binding, GL_TEXTURE_MIN_FILTER, GL_LINEAR);

to

// Complete the texture
if(numMipLevels > 0)
  glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_LINEAR_MIPMAP_NEAREST);
else
  glTexParameteri(binding, GL_TEXTURE_MIN_FILTER, GL_LINEAR);

Just see if that fixes the black texture, i'm doubting it will though but it's quick and easy to try. Going to be a lot easier if i can reproduce the problem here.

Honestly that whole function innerCreateTexture function could do with a complete overhaul.

#103 Nvidia, and the code snippet you provided does appear to have corrected that, with glTexParameteri(binding, GL_TEXTURE_MAX_LEVEL, numMipLevels); making no difference either way that I can tell offhand.

Since it's just some programmer art at present if you wish to crosscheck it: github.com/Azaezel/Torque3D/tree/dev_linux_opengl_Acid_Test

A proper test would of course entail manually editing the MIPs to different colors for verification, but I've completely forgotten how to do so with gimp (not an application I typically use much.)

Much obliged.

#104 Thanks for the feedback Azaezel, i will fix it up. I have closed the original pull request i sent to Luis and i will work on fixing up that whole innerCreateTexture function.

One scenario that setting the GL_TEXTURE_MAX_LEVEL can help with is if a texture has an incomplete set of mip maps and is missing say the last 3 levels, this can cause troubles with some drivers if you don't set that max level. It's always good practice to set the GL_TEXTURE_MAX_LEVEL to the number of mip maps.

#105 New update on Github: github.com/LuisAntonRebollo/Torque3D/commits/dev_linux_opengl

• OpenGL Fix: Render upside down when PostFX are disabled.
• OpenGL Fix: GuiColorPickerCtrl not select color correctly
• OpenGL Fix: PrimBuild render incorrect when no texture are asigned.

#106 @Azaezel:

The code i posted above and previously for the amd fix is totally wrong, i have wrongly assumed ( yes we all know the saying about this lol ) about what the fields forceMips and numMipLevels actually mean. I will correct this and send a pull request to Luis once i can confirm it A)Actually corrects the AMD problem B)Doesn't cause any side affects to NVidia etc.

Once i get this sorted out i'm going to look into why fullscreen mode doesn't work under windows, i'm not sure the status of this on the linux/osx side?

#107 I have now sent a much revised pull request to luis for this amd loading problem. I have changed a bit of loading code in the texture manager so even if you don't have a amd gpu it would be much appreciated if you could test it out. Pull request is here github.com/LuisAntonRebollo/Torque3D/pull/37

#108 Seems to work, judging by the results of the revised texture sheets I just slapped up for verification. (Thanks Andrew.)

#109 No problem. (I had nothing to do with it haha)

#110 heh. No, the artist.

#111 New update on Github:github.com/LuisAntonRebollo/Torque3D/tree/dev_linux_opengl

• OpenGL Fix: Precipitation render as solid color quads. Thx to Tim Newell.
• OpenGL: Add support for setupGenericShaders GSTexture (FFP emulation). Thx to Tim Newell.
• OpenGL Fix: Typo in common/gl/basicCloudsV.glsl. gvTexCoord0 -> vTexCoord0. Thx to Tim Newell.
• OpenGL Windows Fix: Window color buffer are only 16bit.
• OpenGL: Generalize code for GFXGLWindowTarget on all platforms.
• Added ARB_debug_output and AMD_debug_output support. Thx to Timmy

#112 I tried the Conifer Forest Pack Demo with the OpenGL Build.

HDR_RGB10_MAX is disable in shaders/gl/torque.glsl at line 156. This cause a crash or a endless loading spamming the error message that HDR_RGB10_MAX is undefined. After uncomment the variable it load the map :)

So I guess it should be enabled.

#113 @Thomas, thx... added to my ToDo list.

I started working fulltime with BeamNG to make T3D multiplatform.

More info in blog: Torque3D is going multiplatform, thanks to BeamNG

Development moved to github.com/BeamNG/Torque3D/tree/dev_linux_opengl

#114 thank you very much for your efforts.

#115 just noticed someone else already reported this.

#116 quickie notes/queries: (Mar 04, 2014 baseline)

gfxD3D9PCDeviceProfiler.cpp - #include <d3d.h> deprecated for newer PC directx versions. Thinking bracket it in a compiler flag, or look for a directx version tag?

gfxGLDeviceProfiler.cpp - GFXDevice::GFXDeviceEventType::deInit 2008 didn't like that one. older compilers can get away with GFXDevice::deInit , or I could cook up and toss over one that encapsulates that enum in a subclass. Preference?

#117 If anyone is having a hard time looking through the OpenGL traces with apitrace, simply change the implementation of the following methods:

void GFXGLDevice::enterDebugEvent(ColorI color, const char *name)
{
	if (gglHasExtension(EXT_debug_marker))
	{
		glPushGroupMarkerEXT(dStrlen(name), name);
	}
}

void GFXGLDevice::leaveDebugEvent()
{
	if (gglHasExtension(EXT_debug_marker))
	{
		glPopGroupMarkerEXT();
	}
}

void GFXGLDevice::setDebugMarker(ColorI color, const char *name)
{
	if (gglHasExtension(EXT_debug_marker))
	{
		glInsertEventMarkerEXT(dStrlen(name), name);
	}
}

This will give you nice nested draw calls so you can see when each render bin is being rendered, as well as when each mesh is being rendered (assuming you are using antons latest code).

#118 New update on Github: github.com/BeamNG/Torque3D/tree/dev_linux_opengl

• Fixed imposter generator code.
• Remove GFXGLTextureTargetAUXBufferImpl
• Change GL_EXT_framebuffer_object to GL_ARB_framebuffer_object
• Clean GFXProfile.
• Support GL_ARB_gpu_shader5 on GLSL and fix fxaaP.glsl error:...
• Change context to 3.2 and remove more FFP
• Fix: HDR_RGB10_MAX is not declared on GLSL.
• Fix: Missions take 10 minutes to load on AMD cards. Thx to Timmy
• Add OpenGL cache for avoid glGet* calls.
• explicit asignment for auto &ssd = mDesc.samplers[i];
• Remove glMapBuffer, and use FrameAllocator for lock/unlock...
• Add HW instancing support.
• Add a simple gpu profiler.
• Separate sampler object from texture object.
• Fix: GFXDevice::disableShaders() set incorrect GFXDevice::setupGenericShaders.
• Add support for sampler1D
• Fix: Declare a glsl function with name "mul" cause problems with some drivers.
• Fix EOL in some files.

A good week of work in BeamNG :D


@Azaezel:
Thx for your review :D
About <d3d.h>... I only move existing code from Engine/source/gfx/D3D9/pc/gfxPCD3D9Device.cpp but yes it's deprecated.

About GFXDevice::GFXDeviceEventType::deInit, yes this will cause problems on some compilers. Added to my ToDo.

@James:
As always, another great contribution. Added to my ToDo.

#119 Nice work Luis, things are definitely moving along now :-)

#120 Is anyone here familiar with how the Advanced Lighting works? Our current OpenGL code is based on an earlier port with a few ideas from Anton's port (e.g. the vertex attributes in the shaders). I have the lighting pass as well as the spotlight shadows working, but the sunlight vector lighting is proving to be a challenge to fix. Essentially I am getting shadows but they are moving based on the camera position. In most cases they do not appear at all. Does anyone know of a good way to debug whats going on? I suspect its just a problem with the vector light shader as the spot lights still work, but I'm still at a loss as to which part of the shadowmap check is going wrong...

Unlike the port on github, we do not flip Y coordinates in the vertex shader, which I suspect is accounting for some of the oddness.

Also has anyone noticed any problems with the shaders rendering green / orange artefacts on OSX (typically when rendering the glow & lighting specular)?