Difference between revisions of "3D Fundamentals Tutorial 11"
From Chilipedia
(→Downloads) |
(→Video) |
||
| Line 4: | Line 4: | ||
The tutorial video is on YouTube [https://youtu.be/HyVc0X9JKpg here]. | The tutorial video is on YouTube [https://youtu.be/HyVc0X9JKpg here]. | ||
| + | |||
| + | <div class="mw-collapsible mw-collapsed"><br /> | ||
| + | * Occlusion for multiple objects [https://youtu.be/HyVc0X9JKpg?t=0m24s 0:24] | ||
| + | <div class="mw-collapsible-content"> | ||
| + | :* We used culling of triangles when drawing convex shapes | ||
| + | :* But this doen't solve our problem with occlusion (e.g. multiple meshes in the same scene) | ||
| + | :* One could considers the painter's algorithm, but that's inefficient and of limited use | ||
| + | </div> | ||
| + | * Solution: the z-buffer: simple, elegant & performant [https://youtu.be/HyVc0X9JKpg?t=2m27s 2:27] | ||
| + | <div class="mw-collapsible-content"> | ||
| + | :* Essentially sorts triangles based on their depth at the pixel level | ||
| + | :* Uses two 2D-arrays (for x,y coordinates): a back-buffer (of colors, in <code>gfx</code> in our case) as a render target, and a z-buffer (of floats) to store depth | ||
| + | :* We initialize the z-buffer with infinity | ||
| + | :* Whenever we want to write to the back-buffer, we check to see if z-value of the pixel we are drawing is smaller (and set the z-buffer in that case) | ||
| + | :* This is a special sorting case [https://youtu.be/HyVc0X9JKpg?t=6m26s 6:26] | ||
| + | </div> | ||
| + | * Implementation in code [https://youtu.be/HyVc0X9JKpg?t=7m06s 7:06] | ||
| + | <div class="mw-collapsible-content"> | ||
| + | :* The <code>ZBuffer</code> class | ||
| + | :* Changes to the <code>Pipeline</code> class, <code>if (zb.TestAndSet(x,y,z))</code> before the <code>PutPixel</code> call | ||
| + | </div> | ||
| + | * If done well, Z-testing can reduce rendering time [https://youtu.be/HyVc0X9JKpg?t=9m54s 9:54] | ||
| + | <div class="mw-collapsible-content"> | ||
| + | :* Works best if geometries are sorted and drawn from front to back (more Z-test rejections) | ||
| + | :* This is contrary to intuition and old/other techniques (e.g. in 2D: drawing sprites and tilemaps) | ||
| + | </div> | ||
| + | * Scene definitions and demo [https://youtu.be/HyVc0X9JKpg?t=10m58s 10:58] | ||
| + | * Differences with Hardware APIs [https://youtu.be/HyVc0X9JKpg?t=11m50s 11:50] | ||
| + | <div class="mw-collapsible-content"> | ||
| + | :* These provide more options for configuring the z-buffer, such as depth comparison operators, choice of formats/types | ||
| + | :* These typically store 1/z in the buffer, requiring corrections for depth precision (high in the near plane, low in the far plane) | ||
| + | </div> | ||
| + | * Dealing with translucent pixels and depth [https://youtu.be/HyVc0X9JKpg?t=13m58s 13:58] | ||
| + | <div class="mw-collapsible-content"> | ||
| + | :* You will need to store the order of all the translucent pixels up until the first opaque one | ||
| + | </div> | ||
| + | * Stencil Buffers [https://youtu.be/HyVc0X9JKpg?t=14m28s 14:28] | ||
| + | <div class="mw-collapsible-content"> | ||
| + | :* Provide more flexibility to reject pixels & have more control (general purpose bufffers) | ||
| + | :* E.g. for UIs, HUDs, mirrors, shadow effects etc. | ||
| + | </div> | ||
| + | * Next up: vertex shaders and working towards dynamic lighting effects [https://youtu.be/HyVc0X9JKpg?t=15m00s 15:00] | ||
| + | </div> | ||
== Downloads == | == Downloads == | ||
Revision as of 03:27, 15 June 2020
Things that are behind get covered up by things that are in front.
Video
The tutorial video is on YouTube here.
- Occlusion for multiple objects 0:24
- We used culling of triangles when drawing convex shapes
- But this doen't solve our problem with occlusion (e.g. multiple meshes in the same scene)
- One could considers the painter's algorithm, but that's inefficient and of limited use
- Solution: the z-buffer: simple, elegant & performant 2:27
- Essentially sorts triangles based on their depth at the pixel level
- Uses two 2D-arrays (for x,y coordinates): a back-buffer (of colors, in
gfxin our case) as a render target, and a z-buffer (of floats) to store depth - We initialize the z-buffer with infinity
- Whenever we want to write to the back-buffer, we check to see if z-value of the pixel we are drawing is smaller (and set the z-buffer in that case)
- This is a special sorting case 6:26
- Implementation in code 7:06
- The
ZBufferclass - Changes to the
Pipelineclass,if (zb.TestAndSet(x,y,z))before thePutPixelcall
- The
- If done well, Z-testing can reduce rendering time 9:54
- Works best if geometries are sorted and drawn from front to back (more Z-test rejections)
- This is contrary to intuition and old/other techniques (e.g. in 2D: drawing sprites and tilemaps)
- These provide more options for configuring the z-buffer, such as depth comparison operators, choice of formats/types
- These typically store 1/z in the buffer, requiring corrections for depth precision (high in the near plane, low in the far plane)
- Dealing with translucent pixels and depth 13:58
- You will need to store the order of all the translucent pixels up until the first opaque one
- Stencil Buffers 14:28
- Provide more flexibility to reject pixels & have more control (general purpose bufffers)
- E.g. for UIs, HUDs, mirrors, shadow effects etc.
- Next up: vertex shaders and working towards dynamic lighting effects 15:00
Downloads
The GitHub repository for the tutorial code is here.
Paper on depth precision
