Looking for GPU Performance Issues | Handmade Hero Episode Guide

0:02Welcome to the stream

🗩

0:02Welcome to the stream

🗩

0:02Welcome to the stream

🗩

0:14A few words on RenderDoc's crash message yesterday, with praise for their tech support, and plans to enable the game to fail gracefully when launched with incorrect parameters

🗩

0:14A few words on RenderDoc's crash message yesterday, with praise for their tech support, and plans to enable the game to fail gracefully when launched with incorrect parameters

🗩

0:14A few words on RenderDoc's crash message yesterday, with praise for their tech support, and plans to enable the game to fail gracefully when launched with incorrect parameters

🗩

3:48Launch Handmade Hero in RenderDoc

🏃

3:48Launch Handmade Hero in RenderDoc

🏃

3:48Launch Handmade Hero in RenderDoc

🏃

4:13Set the wrong Working Directory in RenderDoc

🗹

4:13Set the wrong Working Directory in RenderDoc

🗹

4:13Set the wrong Working Directory in RenderDoc

🗹

4:34Crash RenderDoc upon launching Handmade Hero

🏃

4:34Crash RenderDoc upon launching Handmade Hero

🏃

4:34Crash RenderDoc upon launching Handmade Hero

🏃

6:12Launch Handmade Hero in RemedyBG with the Working Directory set wrong

🏃

6:12Launch Handmade Hero in RemedyBG with the Working Directory set wrong

🏃

6:12Launch Handmade Hero in RemedyBG with the Working Directory set wrong

🏃

7:15Hit our assertion in GetFontInfo()

🏃

7:15Hit our assertion in GetFontInfo()

🏃

7:15Hit our assertion in GetFontInfo()

🏃

7:39Make GetFontInfo() additionally assert that the Asset's Type is HHAAsset_None

8:34Hit our TextureIndex assertion in PushQuad()

🏃

8:34Hit our TextureIndex assertion in PushQuad()

🏃

8:34Hit our TextureIndex assertion in PushQuad()

🏃

9:55Enable all our asset Get*() functions to handle the absence of assets

16:43Run successfully with our incorrect Working Directory

🏃

16:43Run successfully with our incorrect Working Directory

🏃

16:43Run successfully with our incorrect Working Directory

🏃

17:13Enable AllocateGameAssets() to issue a warning notification when no assets were available

23:17Crash RemedyBG apparently on a jump-to-zero

🏃

23:17Crash RemedyBG apparently on a jump-to-zero

🏃

23:17Crash RemedyBG apparently on a jump-to-zero

🏃

23:58Create jump_to_zero_crash.cpp

25:07Run jump_to_zero to find that RemedyBG is fine with it

🏃

25:07Run jump_to_zero to find that RemedyBG is fine with it

🏃

25:07Run jump_to_zero to find that RemedyBG is fine with it

🏃

25:48Run Handmade Hero successfully with our new warning code

🏃

25:48Run Handmade Hero successfully with our new warning code

🏃

25:48Run Handmade Hero successfully with our new warning code

🏃

26:38Introduce Win32ErrorMessage()

28:31Run and close cleanly

🏃

28:31Run and close cleanly

🏃

28:31Run and close cleanly

🏃

28:44Make Win32ErrorMessage() print out an error message¹

34:16Run and close with our warning box

🏃

34:16Run and close with our warning box

🏃

34:16Run and close with our warning box

🏃

34:39Fix the MBoxType in Win32ErrorMessage()²

35:46Run and close with our warning box

🏃

35:46Run and close with our warning box

🏃

35:46Run and close with our warning box

🏃

35:51Temporarily try making AllocateGameAssets() produce a Fatal error

36:17Run and close with our error box

🏃

36:17Run and close with our error box

🏃

36:17Run and close with our error box

🏃

36:29Make WinMainCRTStartup() emit errors using Win32ErrorMessage()

38:58Run and close with our warning box

🏃

38:58Run and close with our warning box

🏃

38:58Run and close with our warning box

🏃

39:06Run unsuccessfully with the correct Working Directory

🏃

39:06Run unsuccessfully with the correct Working Directory

🏃

39:06Run unsuccessfully with the correct Working Directory

🏃

41:16We've got a saucy bean

🗩

41:16We've got a saucy bean

🗩

41:16We've got a saucy bean

🗩

42:12Fix GetBitmap() to correctly set the TextureHandle

43:45Run successfully

🏃

43:45Run successfully

🏃

43:45Run successfully

🏃

44:02Run in RenderDoc with the wrong Working Directory

🏃

44:02Run in RenderDoc with the wrong Working Directory

🏃

44:02Run in RenderDoc with the wrong Working Directory

🏃

44:25Crash RenderDoc post-exit

🏃

44:25Crash RenderDoc post-exit

🏃

44:25Crash RenderDoc post-exit

🏃

45:40Run in RenderDoc with the correct Working Directory, noting that we sometimes miss 60 FPS

🏃

45:40Run in RenderDoc with the correct Working Directory, noting that we sometimes miss 60 FPS

🏃

45:40Run in RenderDoc with the correct Working Directory, noting that we sometimes miss 60 FPS

🏃

46:54Capture a frame in RenderDoc and see that it took 35590 µs

🏃

46:54Capture a frame in RenderDoc and see that it took 35590 µs

🏃

46:54Capture a frame in RenderDoc and see that it took 35590 µs

🏃

48:48Look at our four Colour Passes, and plan to submit vertices economically

🏃

48:48Look at our four Colour Passes, and plan to submit vertices economically

🏃

48:48Look at our four Colour Passes, and plan to submit vertices economically

🏃

51:37Make OpenGLInit() disable RequestVSync

52:40Find that our frame time hovers around 14 ms per frame

🏃

52:40Find that our frame time hovers around 14 ms per frame

🏃

52:40Find that our frame time hovers around 14 ms per frame

🏃

53:27Reacquaint ourselves with our render dispatch in OpenGLEndFrame()

📖

53:27Reacquaint ourselves with our render dispatch in OpenGLEndFrame()

📖

53:27Reacquaint ourselves with our render dispatch in OpenGLEndFrame()

📖

55:31Understanding glMapBuffer()³

📖

55:31Understanding glMapBuffer()³

📖

55:31Understanding glMapBuffer()³

📖

57:57Nsight rendering time: 14ms to 18ms / frame

🏃

57:57Nsight rendering time: 14ms to 18ms / frame

🏃

57:57Nsight rendering time: 14ms to 18ms / frame

🏃

59:37Scrub through Events to see that glDrawArrays takes 1.27ms

🏃

59:37Scrub through Events to see that glDrawArrays takes 1.27ms

🏃

59:37Scrub through Events to see that glDrawArrays takes 1.27ms

🏃

1:00:35Scrutinise ResolveMultisample() with a view to speeding it up

📖

1:00:35Scrutinise ResolveMultisample() with a view to speeding it up

📖

1:00:35Scrutinise ResolveMultisample() with a view to speeding it up

📖

1:05:00Make CompileResolveMultisample() bake the SampleCount in to the shader, to hopefully permit the loop to be unrolled

1:08:03Find that it actually works

🏃

1:08:03Find that it actually works

🏃

1:08:03Find that it actually works

🏃

1:08:19Make CompileResolveMultisample() bake InvSampleCount in to the shader, and slightly reorganise it

1:12:35Find that that made no difference to the frame time, and that UpdateAndRenderEntities() takes a while

🏃

1:12:35Find that that made no difference to the frame time, and that UpdateAndRenderEntities() takes a while

🏃

1:12:35Find that that made no difference to the frame time, and that UpdateAndRenderEntities() takes a while

🏃

1:14:04Temporarily Disable DrawGroundCover()

1:14:37Find that Game Update takes less of the total time, but we are not hitting 60 FPS

🏃

1:14:37Find that Game Update takes less of the total time, but we are not hitting 60 FPS

🏃

1:14:37Find that Game Update takes less of the total time, but we are not hitting 60 FPS

🏃

1:17:11Disable HANDMADE_SLOW

1:18:23Find that Debug Collation takes a lot of the total time

🏃

1:18:23Find that Debug Collation takes a lot of the total time

🏃

1:18:23Find that Debug Collation takes a lot of the total time

🏃

1:18:46Compile out some of the debug system if HANDMADE_SLOW is off

1:20:43See the debug UI

🏃

1:20:43See the debug UI

🏃

1:20:43See the debug UI

🏃

1:21:07Instead compile out that part of the debug system if HANDMADE_SLOW is on, and rearrange the code to fix compile errors

1:23:33Find that lots of the debug system is absent

🏃

1:23:33Find that lots of the debug system is absent

🏃

1:23:33Find that lots of the debug system is absent

🏃

1:24:06Compile out timing if HANDMADE_SLOW is off

1:25:05Still see jerkiness with debug collation off

🏃

1:25:05Still see jerkiness with debug collation off

🏃

1:25:05Still see jerkiness with debug collation off

🏃

1:26:09Compile in our frame marker in all situations

1:26:45See that our frame time is well below 16ms, but we are not actually seeing 60 FPS

🏃

1:26:45See that our frame time is well below 16ms, but we are not actually seeing 60 FPS

🏃

1:26:45See that our frame time is well below 16ms, but we are not actually seeing 60 FPS

🏃

1:27:22Nsight rendering time: 10ms to 14ms / frame

🏃

1:27:22Nsight rendering time: 10ms to 14ms / frame

🏃

1:27:22Nsight rendering time: 10ms to 14ms / frame

🏃

1:29:07Capture a frame in Nsight to see that glDrawArrays remains by far the most expensive call

🏃

1:29:07Capture a frame in Nsight to see that glDrawArrays remains by far the most expensive call

🏃

1:29:07Capture a frame in Nsight to see that glDrawArrays remains by far the most expensive call

🏃

1:33:47Hard set SampleCount to 1 in CompileResolveMultisample()

1:34:44Nsight rendering time: 4ms / frame

🏃

1:34:44Nsight rendering time: 4ms / frame

🏃

1:34:44Nsight rendering time: 4ms / frame

🏃

1:35:03Capture a frame in Nsight to find that the resolve and draw calls take similar times

🏃

1:35:03Capture a frame in Nsight to find that the resolve and draw calls take similar times

🏃

1:35:03Capture a frame in Nsight to find that the resolve and draw calls take similar times

🏃

1:36:09Understanding multisampling

🗩

1:36:09Understanding multisampling

🗩

1:36:09Understanding multisampling

🗩

1:37:47See our crispy lines, without multisampling

🏃

1:37:47See our crispy lines, without multisampling

🏃

1:37:47See our crispy lines, without multisampling

🏃

1:38:31Research multisampling in GLSL⁴^,5^,6^,7^,8

📖

1:38:31Research multisampling in GLSL⁴^,5^,6^,7^,8

📖

1:38:31Research multisampling in GLSL⁴^,5^,6^,7^,8

📖

1:45:42Spec out our desired fast path in CompileResolveMultisample(), based on there being only one sample in a multisample

1:48:00See nothing

🏃

1:48:00See nothing

🏃

1:48:00See nothing

🏃

1:48:04Revert CompileResolveMultisample()

1:50:08See everything

🏃

1:50:08See everything

🏃

1:50:08See everything

🏃

1:50:10Spec out our desired fast path in CompileResolveMultisample() again

1:51:40See nothing

🏃

1:51:40See nothing

🏃

1:51:40See nothing

🏃

1:51:52Cut out the else and the if(1) in CompileResolveMultisample()

1:52:41Find that the if(1) was the problem, somehow

🏃

1:52:41Find that the if(1) was the problem, somehow

🏃

1:52:41Find that the if(1) was the problem, somehow

🏃

1:53:08Reinsert the if(1) in CompileResolveMultisample()

1:53:30Capture a frame in Nsight, but still see no error message

🏃

1:53:30Capture a frame in Nsight, but still see no error message

🏃

1:53:30Capture a frame in Nsight, but still see no error message

🏃

1:55:24Enable HANDMADE_SLOW

1:55:46Trigger a fragment shader error: implicit cast from "int" to "bool"

🏃

1:55:46Trigger a fragment shader error: implicit cast from "int" to "bool"

🏃

1:55:46Trigger a fragment shader error: implicit cast from "int" to "bool"

🏃

1:56:21Change the if(1) to if(true) in CompileResolveMultisample()

1:57:20See our standard output

🏃

1:57:20See our standard output

🏃

1:57:20See our standard output

🏃

1:57:29Trigger our non-multisampled fast path in CompileResolveMultisample()

1:57:34See our crispy edges

🏃

1:57:34See our crispy edges

🏃

1:57:34See our crispy edges

🏃

1:57:47Make a note to ask GPU people about our fast path

1:58:28Q&A

🗩

1:58:28Q&A

🗩

1:58:28Q&A

🗩

1:59:13nxsy Q: textureSamples(sampler)?⁹

🗪

1:59:13nxsy Q: textureSamples(sampler)?⁹

🗪

1:59:13nxsy Q: textureSamples(sampler)?⁹

🗪

2:00:27Brian Q: You can have Windows automatically capture crash dumps for you. Check if this key exists: HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Windows\Windows Error Reporting\LocalDumps and, if so, it will automatically capture dumps to %LOCALAPPDATA%\CrashDumps. You can read more¹⁰

🗪

2:00:27Brian Q: You can have Windows automatically capture crash dumps for you. Check if this key exists: HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Windows\Windows Error Reporting\LocalDumps and, if so, it will automatically capture dumps to %LOCALAPPDATA%\CrashDumps. You can read more¹⁰

🗪

2:00:27Brian Q: You can have Windows automatically capture crash dumps for you. Check if this key exists: HKEY_LOCAL_MACHINE\SOFTWARE\Microsoft\Windows\Windows Error Reporting\LocalDumps and, if so, it will automatically capture dumps to %LOCALAPPDATA%\CrashDumps. You can read more¹⁰

🗪

2:03:47aaronnickovich Q: The Watch page¹¹ has no more scheduled appearances. When will you be streaming next?

🗪

2:03:47aaronnickovich Q: The Watch page¹¹ has no more scheduled appearances. When will you be streaming next?

🗪

2:03:47aaronnickovich Q: The Watch page¹¹ has no more scheduled appearances. When will you be streaming next?

🗪

2:04:14blazeitfury Q: You should switch to Linux

🗪

2:04:14blazeitfury Q: You should switch to Linux

🗪

2:04:14blazeitfury Q: You should switch to Linux

🗪

2:05:41rationalcoder Q: Someone had asked whether or not that would optimize anything since the GPU has to execute both branches

🗪

2:05:41rationalcoder Q: Someone had asked whether or not that would optimize anything since the GPU has to execute both branches

🗪

2:05:41rationalcoder Q: Someone had asked whether or not that would optimize anything since the GPU has to execute both branches

🗪

2:07:34bulmanator Q: I’m not sure if you were joking when you said to ask, but if you have the time and feel like it I would 100% be down to see lectures from you about the rest of the animation system like you did with skinning

🗪

2:07:34bulmanator Q: I’m not sure if you were joking when you said to ask, but if you have the time and feel like it I would 100% be down to see lectures from you about the rest of the animation system like you did with skinning

🗪

2:07:34bulmanator Q: I’m not sure if you were joking when you said to ask, but if you have the time and feel like it I would 100% be down to see lectures from you about the rest of the animation system like you did with skinning

🗪

2:09:26philliptrudeau Q: I think that putting it under "Chat" made it less clickable. Also, flashy thumbnails are really important

🗪

2:09:26philliptrudeau Q: I think that putting it under "Chat" made it less clickable. Also, flashy thumbnails are really important

🗪

2:09:26philliptrudeau Q: I think that putting it under "Chat" made it less clickable. Also, flashy thumbnails are really important

🗪

2:11:20blazeitfury Q: My girlfriend watches and tells you most of the things you said about Linux are old school issues

🗪

2:11:20blazeitfury Q: My girlfriend watches and tells you most of the things you said about Linux are old school issues

🗪

2:11:20blazeitfury Q: My girlfriend watches and tells you most of the things you said about Linux are old school issues

🗪

2:16:13aaronnickovich I have an Arch Linux machine. Its dependencies are now broken and will take weeks to get working again

🗪

2:16:13aaronnickovich I have an Arch Linux machine. Its dependencies are now broken and will take weeks to get working again

🗪

2:16:13aaronnickovich I have an Arch Linux machine. Its dependencies are now broken and will take weeks to get working again

🗪

2:18:37ivereadthesequel Q: Casey please dispel the myth that molly123 and I are the same person. rupan3 has a crazy conspiracy going on I can't shake

🗪

2:18:37ivereadthesequel Q: Casey please dispel the myth that molly123 and I are the same person. rupan3 has a crazy conspiracy going on I can't shake

🗪

2:18:37ivereadthesequel Q: Casey please dispel the myth that molly123 and I are the same person. rupan3 has a crazy conspiracy going on I can't shake

🗪

2:19:01pythno Q: For what reason do you like Linux if it breaks all the time?

🗪

2:19:01pythno Q: For what reason do you like Linux if it breaks all the time?

🗪

2:19:01pythno Q: For what reason do you like Linux if it breaks all the time?

🗪

2:23:12Wrap it up

🗩

2:23:12Wrap it up

🗩

2:23:12Wrap it up

🗩

Handmade Hero

Keyboard Navigation

Global Keys

Menu toggling

In-Menu Movement

Quotes and References Menus

Quotes, References and Credits Menus

Filter Menu

Filter and Link Menus

Credits Menu