Voxel-Indexed Lighting Samples — Handmade Hero — Episode Guide

0:08Recap and set the stage for the day

0:42Run the game and consider the problem of reconstructing per-pixel lighting from our coarse computations

8:35Blackboard: Sending down four interpolation points per quad, instead of doing a global gather

15:04Blackboard: Radial-normalised interpolation

17:31Run the game and consider whether or not to interpolate across primitives

20:41Blackboard: Lighting Lookups

24:45Blackboard: Storage format for a spherical query

25:34Blackboard: Quick spatial query

31:00Consider the problem of packaging up this data for parallel access by the GPU

39:18Consult the Texture Lookup Functions in the OpenGL Shading Language 1.50 Quick Reference Card for sampling from a 3D texture¹

41:15Blackboard: Artifacts across cell boundaries

44:28Determine to create a voxel-based lighting solution

45:14handmade_render_group.cpp: Introduce OutputLightingTextures()

46:25handmade_world_mode.h: Augment lighting_solution to contain voxel and storage information

52:05handmade_world_mode.h: Introduce lighting_texel linked list for fast access lookup

55:54handmade_world_mode.h: Introduce lighting_textures

56:46handmade_render_group.cpp: Begin to enable OutputLightingTextures() to pack our lighting data

59:34handmade_render_group.cpp: Determine to prevent OutputLightingQuads() from considering the reflector's own colour

1:00:12Run the game to see that it's running slower than expected

1:02:17Run the game and note that we want to stop the reflectors from considering their own colour during the lighting pass

1:02:51handmade_render_group.cpp: Change LightingTest() to multiply in the reflector's colour at the start, and ExtractReflectorsFromVerts() and ExtractReflectorsFromQuads() to not use the reflector's own emission colour

1:06:22Run the game to see that the red and blue reflectors do not show their own colour, but that these colours are also not bouncing off other reflectors, and investigate why

1:11:53handmade_render_group.cpp: Continue to enable OutputLightingTextures() to pack our lighting data into a linked list

1:16:14Blackboard: Mapping world space positions into a screen space lookup table

1:20:45handmade_render_group.cpp: Make OutputLightingTextures() index into voxels

1:25:23Blackboard: Picking a cell from a grid

1:27:00handmade_render_group.cpp: Make OutputLightingTextures() map the lighting into our voxel

1:32:39handmade_world_mode.cpp: Make UpdateAndRenderWorld() call OutputLightingTextures() on test textures

1:35:31handmade_render_group.cpp: Introduce OutputTextureDebug() for OutputLighting() to call

1:43:13Run the game, toggle to the lighting view and see nothing

1:43:56Step in to OutputLightingTextures() and...

1:45:40Wish for the ability to optimise just one routine while waiting for our lighting solution to be computed

1:46:37...inspect the values produced by OutputLightingTextures()

1:48:32handmade_render_group.cpp: Fix OutputTextureDebug() to index into the correct locations

1:49:10Run the game in release mode and see weirdly positioned light voxels

1:50:45handmade_render_group.cpp: Try to put #pragma optimise around OutputTextureDebug() to disable optimisations

1:51:10Step through OutputTextureDebug() to see that the colour is not a sane value

1:53:06handmade_render_group.cpp: Try to make OutputTextureDebug() set the colour to white, and run the game to see that we are storing lighting samples, but that only the colour value is incorrectly computed

1:54:12handmade_render_group.cpp: Fix OutputLightingTextures() to multiply the colour into the correct space

1:54:31Run the game and see our voxelised lighting samples

1:54:55Q&A

🗩

1:54:55Q&A

🗩

1:54:55Q&A

🗩

1:54:57handmade_render_group.cpp: Remove the #pragma optimize

1:55:23Run the game and express interest in seeing how this works

1:57:49Brian Greetings Casey! For after the stream, here is a collaborative work from us in #hero for a help guide to setting up WeeChat

🗪

1:57:49Brian Greetings Casey! For after the stream, here is a collaborative work from us in #hero for a help guide to setting up WeeChat

🗪

1:57:49Brian Greetings Casey! For after the stream, here is a collaborative work from us in #hero for a help guide to setting up WeeChat

🗪

2:00:07roam00010011 Maybe it's a non-issue, but because the entity is tilted, when it walks in front of a cube it is being cut. Is there an elegant solution for it?

🗪

2:00:07roam00010011 Maybe it's a non-issue, but because the entity is tilted, when it walks in front of a cube it is being cut. Is there an elegant solution for it?

🗪

2:00:07roam00010011 Maybe it's a non-issue, but because the entity is tilted, when it walks in front of a cube it is being cut. Is there an elegant solution for it?

🗪

2:01:34handmade_world_mode.cpp: Make AddPlayer() draw the hero bitmap and run the game

2:02:10handmade_render_group.cpp: Increase the ZBias in PushBitmap() and run the game to see that the hero gets clipped differently by the geometry

2:05:15alexkelbo When we made the high and low priority thread queues, how did we make the high queue actually have a higher priority? I can't find the difference between them in the implementation, if that makes sense

🗪

2:05:15alexkelbo When we made the high and low priority thread queues, how did we make the high queue actually have a higher priority? I can't find the difference between them in the implementation, if that makes sense

🗪

2:05:15alexkelbo When we made the high and low priority thread queues, how did we make the high queue actually have a higher priority? I can't find the difference between them in the implementation, if that makes sense

🗪

2:11:23roam00010011 If I was to guess, I would say the problem is not the 2D entities in 3D space, but rather 2D entities with 3D entities. If those cubes were sprites like the trees, it would work just fine

🗪

2:11:23roam00010011 If I was to guess, I would say the problem is not the 2D entities in 3D space, but rather 2D entities with 3D entities. If those cubes were sprites like the trees, it would work just fine

🗪

2:11:23roam00010011 If I was to guess, I would say the problem is not the 2D entities in 3D space, but rather 2D entities with 3D entities. If those cubes were sprites like the trees, it would work just fine

🗪

2:13:03alexkelbo In the software renderer we aligned the memory to 16 bytes for SIMD but then we used the unaligned load and store instructions. Why is that?

🗪

2:13:03alexkelbo In the software renderer we aligned the memory to 16 bytes for SIMD but then we used the unaligned load and store instructions. Why is that?

🗪

2:13:03alexkelbo In the software renderer we aligned the memory to 16 bytes for SIMD but then we used the unaligned load and store instructions. Why is that?

🗪

2:16:18Blackboard: False sharing of one cache line by two threads

2:18:12That's about it for today

🗩

2:18:12That's about it for today

🗩

2:18:12That's about it for today

🗩

Keyboard Navigation

Global Keys

Menu toggling

In-Menu and Index Controls

Quotes and References Menus and Index

Quotes, References and Credits Menus

Filter Menu

Filter and Link Menus

Credits Menu