Framebuffer bloom test

[dmn_clown]

I compiled test binaries to see how it plays on other hardware.

experimental/fbo-test/ in the svn.

[Ylle]

Using the precompiled windows binary:
Nothing on my integrated Intel Express. Not that suprising, but eh, going to try it on a more sensible card in a few moments.

Edit: C:\Documents and Settings\%USERNAME%\Application Data\OpenArena\baseoa\ files untouched.

[fromhell]

Also for some reason my q3config in personaloafolder/gamedir disappeared

[Ylle]

I tried
/r_bloom 1
/r_framebuffer_bloom 1

Edit: nothing on a ati radeon 9250 either.

[randombloke]

Well that was interesting.  Using:

ioquake3.x86.exe

seta r_framebuffer 1
seta r_framebuffer_bloom 1

This generates a red screen with an inverse open arena logo on it.  Printed on top is the english alphabet, as it would appear in a mirror on a horizontal line.

Using:

ioquake3.x86.exe

seta r_framebuffer 1
seta r_framebuffer_bloom 1
seta r_framebuffer_rotoscope 1

Very similar, only a black background this time.  An extended character set can be observed as well, again, it too is a mirror reflection.

As this has no effect on my .exe.

//Correction: I don't use SDL as its not required when using windows.  I just added it to allow the experimental .exe to execute.  So, I assume I don't have the hooks for these values in my .exe.

[kit89]

Worked fine for me. Tho it crippled the framerate.

r_bloom 0
r_framebuffer 1
r_framebuffer_bloom 1



and

r_bloom 0
r_framebuffer 1
r_framebuffer_bloom 1
r_framebuffer_rotoscope 1



That was using a Nvidia Geforce 6 6600 256MB

[fromhell]

oh btw the rotoscoping works, and i get high framerates with that on. The bloom never works

in rotoscope the model shading appears flat shaded than gourad very faceted.

[Crash!!!]

ok the bloom when turned on is not visible and it cripples framerate

[dmn_clown]

Framerates would drop, its all done on the GPU instead of the CPU.

Have you tried playing around with the defaults?  The effect is subtle, like bloom lighting should be.

 

Code:

    r_framebuffer_bloom_brightness = "0.85"
    r_framebuffer_bloom_sharpness = "0.75"
    r_framebuffer_blur_ammount = "7"
    r_framebuffer_blur_size = "128"

[iLeft.bye]

ati x1600 linux 8.37 -> nothing works
windows coming soon!

[next_ghost]

Gentoo Linux, ATi Mobility Radeon x1600, proprietary drivers v8.41, framebuffer+bloom works, framebuffer withou bloom did work, then stopped working, rotoscope looks like fromhell's screenshots.

[iLeft.bye]

ati x1600 linux 8.37 -> nothing works
windows coming soon!

edit:
my bad fbo bloom and rotoscope works (you can combine two rotobloom bloomscope) on linux even with standard ubuntu gutsy restricted ATi driver (version 8.37)

windows fails badly I couldnt even enable fbo -> some error message with an emo face ->

further edit:
I applied the patch myself and built one on linux
but it wont work... plus my binary size is bigger ... how did you build your linux x86 binary?

[iLeft.bye]

   //create the framebuffer texture space
   if (glConfig.stencilBits > 0) {
      R_FrameBufferCreate_ZS(&screenFbuffer, glConfig.vidWidth, glConfig.vidHeight);
   } else {
      R_FrameBufferCreate_Z(&screenFbuffer, glConfig.vidWidth, glConfig.vidHeight);
   }

R_FrameBufferCreate_ZS is the problem here
so if you set r_stencilbits to 0 you have more chance to run fbo

edit:
after setting r_stencilbits to 0 it my build worked too

[dmn_clown]

further edit:
I applied the patch myself and built one on linux
but it wont work... plus my binary size is bigger ... how did you build your linux x86 binary?

Code:

  OPTIMIZE = -O2 -march=i586 -fomit-frame-pointer \
      -falign-loops=2 -falign-jumps=2 \
      -falign-functions=2 -fstrength-reduce
    HAVE_VM_COMPILED=true

no ffast-math no funroll-loops (great website, bad compile flag).

btw - setting r_framebuffer_blur_size to 10 makes the effect far less subtle for the people that aren't sure if something is happening.

[randombloke]

So, why did I get the mirror image test card thingy?

//I know...my technical descriptions are very complex...

[dmn_clown]

I don't use SDL as its not required when using windows.

...waits for the obvious to sink in...

[gord]

Hello, I am the developer of this patch, there are a few things to note:
this patch requires a few extensions. i didn't apply more rigorous extension checking before but i'll add that tonight, for now make sure you defiantly have the following (windows users can use a program called glcaps or something like that, i forget. Linux has glxinfo of course).

GL_EXT_packed_depth_stencil (or GL_NV_packed_depth_stencil)
GL_ARB_depth_texture
GL_EXT_framebuffer_object

GL_ARB_vertex_program
GL_ARB_vertex_shader
GL_ARB_fragment_program
GL_ARB_fragment_shader
GL_ARB_shading_language_100
GL_ARB_texture_non_power_of_two

people having problems when stencil buffers are enabled i would guess do not have packed depth stencil support, this chart suggests that some ati cards for whatever reason do not have this at all: http://www.labri.fr/perso/granier/OpenGL/ext/videocards.html if someone with an ati card could clarify this for me that would be great, if it turns out that ATI really don't have any form of depth-stencil framebuffer support i will just auto-disable stencil so that it at least works.

for people who seem to be having problems with the lighting looking flat shaded, i would first make sure that its not a problem that already exists that is just 'highlighted' by the colour modification the rotoscope shader does, just turn off the bloom/rotoscope effects and take a very close look at the model in question (also if you could post a screenshot with that setup that would be great, so i can take a close look for myself)

for the person apparently having problems with sdl make sure that you are using the latest svn version, with that version there is only sdl support iirc.

i would also like to clarify some of the cvar names (i really need to make these more obvious)
"r_framebuffer" - toggles the framebuffer rendering path
"r_framebuffer_bloom" - toggles the bloom code
"r_framebuffer_blur_size" - the size (in pixels) of the blurred image, higher values will be a higher quality with less popping but obviously, slower (you may need to increase the blur amount too)
"r_framebuffer_blur_ammount" - the size in percentage of the blur, the 'best' value for this is 100/blur_samples.
"r_framebuffer_blur_samples" - only three values will work for this 5, 7 and 9. higher values = higher quality.

"r_framebuffer_bloom_sharpness" - the sharpness between the bright colours and the dark ones when blooming
"r_framebuffer_bloom_brightness" - pretty obvious, 0.85 is a sensible value. for testing i use 10.0 (to make it really show)

"r_framebuffer_rotoscope" - enables rotoscoping
"r_framebuffer_rotoscope_zedge" - uses a different method for edge detection which requires the zbuffer be bound to a texture.

finally how the effect 'should' look, this is my box with regular quake 3 demo four. http://s239.photobucket.com/albums/ff277/gordallott/

[gord]

latest patch should fix problems for some ati users
http://bugzilla.icculus.org/show_bug.cgi?id=3422

[dmn_clown]

updated the bins

[fromhell]

Still not seeing bloom, and still not keeping cfg.

[dmn_clown]

does /writeconfig work?

[gord]

the patch should not have anything todo with writing config files, a few cvars are added but not in any way that touches that code. i would suspect problems elsewhere. regarding it not working with that ati card could you please post some sort of information, ie: make sure its run in a terminal of some sort and see if there are any warning messages or anything (especially regarding shader info logs, it could be that your chip does not support the exp() glsl function). also make sure you really turn the bloom brightness up to make sure you arn't getting it.

[dmn_clown]

It doesn't behave nicely when the gl extensions aren't supported (radeon mobility r300e with xorg's ati driver) they are ignored when set before entering a map, but in game changes followed by vid_restarts crash the game.

[gord]

I should have a patch that might fix that up tommorow night (hopefully). i was getting a similar experiance but one that managed to crash xorg entirely.

[iLeft.bye]

I should have a patch that might fix that up tommorow night (hopefully). i was getting a similar experiance but one that managed to crash xorg entirely.

I love you
will you add vbo?