Something to try: Geometry Based D3D FSAA Viewer
- Thread starter Colourless
- Start date
Passed by the young guns. 
I guess that's the punishment for being an lazy ass.
Well done Colourless. Now just add the true edges, the polygon intersection tests, and the texture mode, so I can get DaveBaumann off my back.
One more addition you could do that I haven't done in my prog. Add modes where you just write to every second pixel horizontaly and vertically, and a way to select odd or even pixels in both directions. Sample positions should still be the same as before. That way it's easier to see the "X.5"-modes that mix subpixels from different pixels. (I.e. R8500 performance modes.)
And if you do it once for each odd/even position with a different primary color for each, and then blend the images. Then it should be easy to see it in one image. (Well, except that you'd need to invent a fourth prmary color.) This last trick would of course fail on cards where framegrabs don't work well.
I guess that's the punishment for being an lazy ass. Well done Colourless. Now just add the true edges, the polygon intersection tests, and the texture mode, so I can get DaveBaumann off my back.
One more addition you could do that I haven't done in my prog. Add modes where you just write to every second pixel horizontaly and vertically, and a way to select odd or even pixels in both directions. Sample positions should still be the same as before. That way it's easier to see the "X.5"-modes that mix subpixels from different pixels. (I.e. R8500 performance modes.)
And if you do it once for each odd/even position with a different primary color for each, and then blend the images. Then it should be easy to see it in one image. (Well, except that you'd need to invent a fourth prmary color.
) This last trick would of course fail on cards where framegrabs don't work well.pocketmoon_
Newcomer
Colourless said:
GF3, Det 42.86, 6Xs 32x32 1/8...
All new and improved....
New features:
- Texture Mode
- Ability to shift the grid
- Horizontal+Vertical Edges for both Geometry and Textures
- And a special mode to compare Geometry and Texture AA
Here's a shot of REFRAST 9x from the new version
Interesting thing that is visible is the refrast actually does Multisampling, not Supersampling.
Things to note:
The colourled words "Geometry Both Texture" indicate what colours represent the samples on screen. Of course, the 'Both' colour may not always match what is rendered but you should be able to work it out.
The top and right lines test Texture Antialiasing (note the coloured line on the very edge) and the bottom and left lines test Geometry Antialiasing (again note the line)
The Geometry Sampling Grid can be shifted with the Cursor keys
The Texture Sampling Grid can be shifted with the Cursor keys while holding Shift.
Use the 'M' key, or the menu to change which mode the program is running in (Geometry, Texture, Both, Three Square Comparison).
Download here: http://www.users.on.net/triforce/d3d_fsaaviewer/d3d_fsaaviewer-3.zip
Those 8500 shots are still puzzling me. Since we know the card is supersampling, the new Texture mode should hopefully show what it's doing.
-Colourless
New features:
- Texture Mode
- Ability to shift the grid
- Horizontal+Vertical Edges for both Geometry and Textures
- And a special mode to compare Geometry and Texture AA
Here's a shot of REFRAST 9x from the new version
Interesting thing that is visible is the refrast actually does Multisampling, not Supersampling.
Things to note:
The colourled words "Geometry Both Texture" indicate what colours represent the samples on screen. Of course, the 'Both' colour may not always match what is rendered but you should be able to work it out.
The top and right lines test Texture Antialiasing (note the coloured line on the very edge) and the bottom and left lines test Geometry Antialiasing (again note the line)
The Geometry Sampling Grid can be shifted with the Cursor keys
The Texture Sampling Grid can be shifted with the Cursor keys while holding Shift.
Use the 'M' key, or the menu to change which mode the program is running in (Geometry, Texture, Both, Three Square Comparison).
Download here: http://www.users.on.net/triforce/d3d_fsaaviewer/d3d_fsaaviewer-3.zip
Those 8500 shots are still puzzling me. Since we know the card is supersampling, the new Texture mode should hopefully show what it's doing.
-Colourless
Joe DeFuria
Legend
Very slick, colourless!
It would be most interesting to see how the 8500 works...it has been suspected for a while now that the sample patterns for textures might be different than that from the geometry. I'm at work now, but I'll give it a go if no one else does it by then...
It would be most interesting to see how the 8500 works...it has been suspected for a while now that the sample patterns for textures might be different than that from the geometry. I'm at work now, but I'll give it a go if no one else does it by then...
Got to wonder then if it's a precision problem with the R8500. I 'might' add in some really low precions modes to see if the pattern can be cleared up.
Then again, it might be using a different pattern for every other pixel or something strange.
Then again, it might be using a different pattern for every other pixel or something strange.
OK, first pass thinking this through. Could you correct me where I'm going wrong? I couldn't find the thread with the prior explanation I mentioned earlier, so I guess I'll have to duplicate things here:
Each square repeats sub pixel position changes a few times horizontally and vertically...hmm...I'm not sure how many times...3? Also, this same grouping is echoed in the pixel spacing of the sub pixel triangles throughout the square. In my understanding, this represents the "theoretical pixels" range of the square.
Geometry mode is based on sub pixel sized triangles. Their sub pixel "hit position" within the pixel corresponds to their pixel position within one of the theoretical pixels in the "geometry" and "both" squares, correct?
Texture mode is based on what? Pixel sized squares textured with a dot in the texture corresponding to the sub pixel "hit position" (again, this hit position corresponding to the overall position within the "texture" and "both" squares)?
First, it would be really nice if someone had time to correct my mistakes in the above.
Second, if this is the case what would "lighter" and "darker" squares mean? Based on my attempt at understanding above, I'd think it would represent less use of a sub pixel position from that area, either through mathematical weighting or less uses based on the screen position of the sample.
Bleh. I guess I'll search for that explanation again...I'm probably missing a really simple search term to find it.
Each square repeats sub pixel position changes a few times horizontally and vertically...hmm...I'm not sure how many times...3? Also, this same grouping is echoed in the pixel spacing of the sub pixel triangles throughout the square. In my understanding, this represents the "theoretical pixels" range of the square.
Geometry mode is based on sub pixel sized triangles. Their sub pixel "hit position" within the pixel corresponds to their pixel position within one of the theoretical pixels in the "geometry" and "both" squares, correct?
Texture mode is based on what? Pixel sized squares textured with a dot in the texture corresponding to the sub pixel "hit position" (again, this hit position corresponding to the overall position within the "texture" and "both" squares)?
First, it would be really nice if someone had time to correct my mistakes in the above.
Second, if this is the case what would "lighter" and "darker" squares mean? Based on my attempt at understanding above, I'd think it would represent less use of a sub pixel position from that area, either through mathematical weighting or less uses based on the screen position of the sample.
Bleh. I guess I'll search for that explanation again...I'm probably missing a really simple search term to find it.
Tried the updated program on an 8500, every combo of quality vs performance at 2x through 6x, including the odd modes 3x and 5x showed something different. No idea what it means though. 8500 had the multisampling selector disabled, so had to use r3d to change modes. The 9700 basically has a purple dot in the third box that never changes and you can change the values for multisampling in the program.
8500 had the multisampling selector disabled, so had to use r3d to change modes. The 9700 basically has a purple dot in the third box that never changes and you can change the values for multisampling in the program.
Bambers said:For the non ordered quality modes on the 8500 the texture positions don't match the geometry ones nor does the antialiasing effect on the edges match.
This normal
Basic has been saying this for some time about R200 FSAA, this just looks to be the confirmation of this.
BTW, thanks again to Colorless for the new version and of course to Basic for the initial versions!
If colourless put in an option to put the relevant fog on and off then with fog on the geometry positions would drop back to the texture ones in all drivers except 7206 and its XP equivalent.
Similarly in basics opengl app in all drivers except ^ the geometry positions are the same as the textures, a simple ordered grid.
Similarly in basics opengl app in all drivers except ^ the geometry positions are the same as the textures, a simple ordered grid.