Fixing Depth Peel Artifacts — Handmade Hero — Episode Guide

0:22Recap and set the stage for the day, improving the depth peeling

2:14Run the game, note the rendering artifacts while depth peeling, and consider what may be causing them

7:23handmade_opengl.cpp: Consider how CompilePeelComposite() is computing the ResultColor

10:16handmade_opengl.cpp: Make CompilePeelComposite() clamp the ResultColor to 0 to 1¹

10:57Run the game to see that that did not improve our situation, but did change our background colour

11:43handmade_opengl.cpp: Investigate how CompilePeelComposite() is drawing the peels

15:58Run the game to determine that the alpha inverse computation is wrong

16:44handmade_opengl.cpp: Prevent CompilePeelComposite() from putting alpha in Peel3

17:48handmade_opengl.cpp: Make CompilePeelComposite() clamp the colours again, and run the game to determine that this part of the composite is not the problem

19:16handmade_opengl.cpp: Make CompilePeelComposite() colour the peels

21:10Run the game to see the coloured peels

22:43handmade_opengl.cpp: Make CompilePeelComposite() multiply the peel colour by its alpha, and run the game to see that the error occurs where many peels overlap

25:17handmade_opengl.cpp: Enable CompileZBiasProgram() to fog pixels out before their alpha is 0

26:19Run the game and erroneously see colour on pixels that should have been fogged completely out

28:07handmade_opengl.cpp: Make CompileZBiasProgram() correctly clip out transparent pixels from the depth peel

29:41Run the game to see that we're a lot better

31:47handmade_opengl.cpp: Try making CompileZBiasProgram() draw Peel3 as bright pink and run the game to see purple where the artifact occurs

33:50Consider how best to mix the peels together

36:26Blackboard: Depth Peel Sandwich

40:26Run the game and think about what's actually happening with the artifact

43:24handmade_opengl.cpp: Enable CompileZBiasProgram() to draw something in the last peel above a given alpha threshold

48:18Run the game to see that that did help

50:51handmade_opengl.cpp: Make OpenGLRenderCommands() clear the alpha to 1, and the shader to undo the premultiplied alpha

52:20Run the game to see that it looks worse

52:53handmade_opengl.cpp: Note that our art asset packer does not handle sRGB

55:54handmade_opengl.cpp: Consider that clamped values cannot be used at the vertex level

57:12Blackboard: Incorrect clamping

58:10handmade_opengl.cpp: Make the FragmentCode itself in CompileZBiasProgram() clamp and compute the fog

58:54Blackboard: Vertex and Fragment Shader Output

1:00:23Run the game, crash in shader compilation and reorganise the code

1:01:39Run the game to see that the effect is very different, and the artifact is gone

1:04:38Note that we probably would still want to use multisampling on fast GPUs

1:06:46handmade_opengl.cpp: Begin to enable OpenGLRenderCommands() to conditionally perform multisampling

1:13:08Consider changing the renderer to draw at the size at which the game is running

1:17:07handmade_opengl.cpp: Change OpenGLRenderCommands() to draw at the correct aspect ratio

1:21:06Run the game to see the correct letterboxing

1:21:41handmade_opengl.cpp: Enable OpenGLBindFramebuffer() to blit to the correct size

1:25:11Run the game to see that it's nice and clean

1:25:38win32_handmade.cpp: Draw to a lower resolution, stretched up

1:26:02Run the game and see some weird artifacts and investigate those artifacts

1:28:33handmade_opengl.cpp: Try to force real sRGB off

1:30:07Run the game and consider that, because the alpha is not computed with sRGB, it does not line up right when stretched

1:31:18handmade_opengl.cpp: Consider that the bilinear filter on the depth peel composite may not be correct

1:33:47handmade_opengl.cpp: Make OpenGLRenderCommands() use a raw texel sample in the depth peel

1:34:38Run the game to see that the fringing while stretching has gone

1:35:02Blackboard: Correctly sampling from texels for stretching

1:39:25handmade_opengl.cpp: Determine to composite to an intermediate framebuffer, and then stretch that to the final image

1:42:57handmade_opengl.cpp: Introduce CompileFinalStretch()

1:48:34handmade_opengl.cpp: Introduce CreateFramebuffer()

1:57:03handmade_opengl.cpp: Enable CreateFramebuffer() to conditionally apply filtering²

1:58:37handmade_opengl.cpp: Make OpenGLRenderCommands() create the final GlobalResolveFramebuffer

2:02:34Run the game to admire the correct composite and then stretching

2:03:35win32_handmade.cpp: Draw to decreasingly smaller resolutions

2:05:58Run the game in pixel art mode^α

2:07:04Q&A

🗩

2:07:04Q&A

🗩

2:07:04Q&A

🗩

2:07:36naysayer88 OpenGL sucks

🗪

2:07:36naysayer88 OpenGL sucks

🗪

2:07:36naysayer88 OpenGL sucks

🗪

2:10:56naysayer88 The problem is there is no way for me to know what the "good API" is any more...

🗪

2:10:56naysayer88 The problem is there is no way for me to know what the "good API" is any more...

🗪

2:10:56naysayer88 The problem is there is no way for me to know what the "good API" is any more...

🗪

2:14:30jezzi23 In Win32InitOpenGL(), why do you use two different forms for assigning modern OpenGL function pointers, i.e. using Win32GetOpenGLFunction(Name) macro for some but assigning others directly?

🗪

2:14:30jezzi23 In Win32InitOpenGL(), why do you use two different forms for assigning modern OpenGL function pointers, i.e. using Win32GetOpenGLFunction(Name) macro for some but assigning others directly?

🗪

2:14:30jezzi23 In Win32InitOpenGL(), why do you use two different forms for assigning modern OpenGL function pointers, i.e. using Win32GetOpenGLFunction(Name) macro for some but assigning others directly?

🗪

2:15:01serialqwiller Just so you know, your stream no longer shows up in either the "Game Development" or "Programming" categories on twitch. It's almost a hidden stream now just for those who already followed

🗪

2:15:01serialqwiller Just so you know, your stream no longer shows up in either the "Game Development" or "Programming" categories on twitch. It's almost a hidden stream now just for those who already followed

🗪

2:15:01serialqwiller Just so you know, your stream no longer shows up in either the "Game Development" or "Programming" categories on twitch. It's almost a hidden stream now just for those who already followed

🗪

2:16:10nyeecola I tried your anonymous union and structs trick and the compiler (g++) screams at me saying I need some label in the struct name to conform to the standard. So I looked it up on google and it appears to be undefined behavior to use this trick. Should I be afraid of that?

🗪

2:16:10nyeecola I tried your anonymous union and structs trick and the compiler (g++) screams at me saying I need some label in the struct name to conform to the standard. So I looked it up on google and it appears to be undefined behavior to use this trick. Should I be afraid of that?

🗪

2:16:10nyeecola I tried your anonymous union and structs trick and the compiler (g++) screams at me saying I need some label in the struct name to conform to the standard. So I looked it up on google and it appears to be undefined behavior to use this trick. Should I be afraid of that?

🗪

2:19:56soysaucethekid The cursor always seems to be on the loading(?) icon. Is there a way to fix that? (I'm running into this issue and am not sure if there is some windows message I need to process)

🗪

2:19:56soysaucethekid The cursor always seems to be on the loading(?) icon. Is there a way to fix that? (I'm running into this issue and am not sure if there is some windows message I need to process)

🗪

2:19:56soysaucethekid The cursor always seems to be on the loading(?) icon. Is there a way to fix that? (I'm running into this issue and am not sure if there is some windows message I need to process)

🗪

2:20:18win32_handmade.cpp: Try to stop the wait cursor displaying once we have finished loading

2:28:38fr0styninja You can set the cursor on the WNDCLASS struct .hCursor = LoadCursor(0, IDC_ARROW);

🗪

2:28:38fr0styninja You can set the cursor on the WNDCLASS struct .hCursor = LoadCursor(0, IDC_ARROW);

🗪

2:28:38fr0styninja You can set the cursor on the WNDCLASS struct .hCursor = LoadCursor(0, IDC_ARROW);

🗪

2:30:06nyeecola You said on one stream that you don't need header guards because you have only one translation unit. I guess you meant that you're sure you only include each header once? Because even if you only have one translation unit you still can have problems if you include the same header twice

🗪

2:30:06nyeecola You said on one stream that you don't need header guards because you have only one translation unit. I guess you meant that you're sure you only include each header once? Because even if you only have one translation unit you still can have problems if you include the same header twice

🗪

2:30:06nyeecola You said on one stream that you don't need header guards because you have only one translation unit. I guess you meant that you're sure you only include each header once? Because even if you only have one translation unit you still can have problems if you include the same header twice

🗪

2:31:37nyeecola Multiple anonymous unions inside one struct, is that allowed?

🗪

2:31:37nyeecola Multiple anonymous unions inside one struct, is that allowed?

🗪

2:31:37nyeecola Multiple anonymous unions inside one struct, is that allowed?

🗪

2:32:37Wrap it up with a plug of Lysa³

📖

2:32:37Wrap it up with a plug of Lysa³

📖

2:32:37Wrap it up with a plug of Lysa³

📖

2:35:19Promote CaptainKraft's Patreon⁴

📖

2:35:19Promote CaptainKraft's Patreon⁴

📖

2:35:19Promote CaptainKraft's Patreon⁴

📖

2:37:01That's about it for today, with a glimpse into the future

🗩

2:37:01That's about it for today, with a glimpse into the future

🗩

2:37:01That's about it for today, with a glimpse into the future

🗩

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