SIMD Raycasting — Handmade Hero — Episode Guide

0:02Recap and set the stage for the day with a few words on the pace of the project

🗩

0:02Recap and set the stage for the day with a few words on the pace of the project

🗩

0:02Recap and set the stage for the day with a few words on the pace of the project

🗩

4:09Run the game to see our current software-rendered lighting, with the determination to see how much performance we can get out of the CPU

🏃

4:09Run the game to see our current software-rendered lighting, with the determination to see how much performance we can get out of the CPU

🏃

4:09Run the game to see our current software-rendered lighting, with the determination to see how much performance we can get out of the CPU

🏃

6:47Prevent RayCast() from summing the TotalCastsInitiated

7:03Run the game to see that that does not appreciably affect our performance

🏃

7:03Run the game to see that that does not appreciably affect our performance

🏃

7:03Run the game to see that that does not appreciably affect our performance

🏃

8:05Augment lighting_work with some of the lighting_solution, to avoid bouncing cache lines around

16:08Run the game and inspect the profiler

🏃

16:08Run the game and inspect the profiler

🏃

16:08Run the game and inspect the profiler

🏃

16:52Make lighting_work and lighting_solution cache-aligned

19:53Cache-alignment and false sharing considerations when threading

🖌

19:53Cache-alignment and false sharing considerations when threading

🖌

19:53Cache-alignment and false sharing considerations when threading

🖌

24:15Introduce InitLighting() to align our lighting data

31:21Run the game and determine to double check that everything is aligned and figure out why the tests are slow

🏃

31:21Run the game and determine to double check that everything is aligned and figure out why the tests are slow

🏃

31:21Run the game and determine to double check that everything is aligned and figure out why the tests are slow

🏃

32:22Assert in ComputeLightPropagation() that the lighting_work is aligned

33:17Run the game and hit that assertion

🏃

33:17Run the game and hit that assertion

🏃

33:17Run the game and hit that assertion

🏃

34:01Pad the lighting_work to 64 bytes

34:40Run the game to see that we are running at full speed again

🏃

34:40Run the game to see that we are running at full speed again

🏃

34:40Run the game to see that we are running at full speed again

🏃

35:53Profile PartitionsPerCast and LeavesPerCast in ComputeLightPropagation()

37:54Run the game and consult the profiler to see that our PartitionsPerCast is high, and consider how to reduce them

🏃

37:54Run the game and consult the profiler to see that our PartitionsPerCast is high, and consider how to reduce them

🏃

37:54Run the game and consult the profiler to see that our PartitionsPerCast is high, and consider how to reduce them

🏃

41:50Add a TIMED_FUNCTION() in RayCast()

42:28Run the game and consult the profiler to see that our RayCast() is not too bad

🏃

42:28Run the game and consult the profiler to see that our RayCast() is not too bad

🏃

42:28Run the game and consult the profiler to see that our RayCast() is not too bad

🏃

45:24Remove the recursion in RayCastRecurse()

49:25Run the game to see that that greatly improved our performance

🏃

49:25Run the game to see that that greatly improved our performance

🏃

49:25Run the game to see that that greatly improved our performance

🏃

50:35Just make RayCast() perform our new code from RayCastRecurse()

52:06Run the game to see that that doesn't change our runtime

🏃

52:06Run the game to see that that doesn't change our runtime

🏃

52:06Run the game to see that that doesn't change our runtime

🏃

52:13Prevent RayCast() from taking Solution and add a TIMED_FUNCTION() in it

53:09Add a TIMED_FUNCTION() in RayCast()

53:25Run the game and see the performance of RayCast(), disable HANDMADE_SLOW and consider what to tackle next

🏃

53:25Run the game and see the performance of RayCast(), disable HANDMADE_SLOW and consider what to tackle next

🏃

53:25Run the game and see the performance of RayCast(), disable HANDMADE_SLOW and consider what to tackle next

🏃

55:45Consider performing RayCast() in SIMD

📖

55:45Consider performing RayCast() in SIMD

📖

55:45Consider performing RayCast() in SIMD

📖

1:00:21Reduce RayCount from 64 to 16 in ComputeLightPropagation()

1:00:36Run the game to see what kind of a speedup SIMD could provide

🏃

1:00:36Run the game to see what kind of a speedup SIMD could provide

🏃

1:00:36Run the game to see what kind of a speedup SIMD could provide

🏃

1:01:42Enable ComputeLightPropagation() to call SampleHemisphere() in SIMD, introducing V3_4x()

1:10:19Update RayCast() to work with our SIMD data, introducing GetComponent()

1:22:52Run the game, crash in RayCast() and investigate why

🏃

1:22:52Run the game, crash in RayCast() and investigate why

🏃

1:22:52Run the game, crash in RayCast() and investigate why

🏃

1:24:30Temporarily disable threading in ComputeLightPropagation()

1:26:14Step through RayCast() to try and see what's going wrong

🏃

1:26:14Step through RayCast() to try and see what's going wrong

🏃

1:26:14Step through RayCast() to try and see what's going wrong

🏃

1:32:12Assert in RayCast() that the Depth is < BoxStack

1:33:04Step back into RayCast(), and eventually hit that assertion

🏃

1:33:04Step back into RayCast(), and eventually hit that assertion

🏃

1:33:04Step back into RayCast(), and eventually hit that assertion

🏃

1:35:05Increase BoxStack from 32 to 64 in RayCast()

1:35:30Run the game and see nothing

🏃

1:35:30Run the game and see nothing

🏃

1:35:30Run the game and see nothing

🏃

1:35:56Prevent RayCast() from pushing on the boxes four times

1:37:42Run the game to see that our performance has decreased

🏃

1:37:42Run the game to see that our performance has decreased

🏃

1:37:42Run the game to see that our performance has decreased

🏃

1:38:18Enable RayCast() to call IsInRectangleCenterHalfDim() in SIMD

1:45:35Introduce Any3TrueInAtLeastOneLane(), All3TrueInAtLeastOneLane(), AbsoluteValue() and v3_4x versions of <=, −, | and &¹

1:59:56Step in to RayCast() and inspect our SIMD data

🏃

1:59:56Step in to RayCast() and inspect our SIMD data

🏃

1:59:56Step in to RayCast() and inspect our SIMD data

🏃

2:01:59Prevent AbsoluteValue() from converting the Mask to a float

2:03:15Step back in to AbsoluteValue() and through our other new SIMD functions in RayCast()

🏃

2:03:15Step back in to AbsoluteValue() and through our other new SIMD functions in RayCast()

🏃

2:03:15Step back in to AbsoluteValue() and through our other new SIMD functions in RayCast()

🏃

2:06:13Run the game to see that our lighting is a little bit messed up

🏃

2:06:13Run the game to see that our lighting is a little bit messed up

🏃

2:06:13Run the game to see that our lighting is a little bit messed up

🏃

2:07:19Make RayCast() perform the old scalar code after our SIMD code, to verify the SIMD

2:11:28Run the game and fail to hit our verification assertion

🏃

2:11:28Run the game and fail to hit our verification assertion

🏃

2:11:28Run the game and fail to hit our verification assertion

🏃

2:13:10Prevent RayCast() from erroneously breaking out of the child box loop

2:13:48Run the game to see that we are correct

🏃

2:13:48Run the game to see that we are correct

🏃

2:13:48Run the game to see that we are correct

🏃

2:14:07Streamline and switch RayCast() almost entirely over to SIMD, introducing ZeroV34x(), versions of V3_4x() that load four scalars, and one wide set, into lanes, v3_4x versions of − for full 4-wide negation, < and *, a f32_4x struct and versions of −, * and / that use this type²

3:19:18Step in to RayCast() to inspect its values

🏃

3:19:18Step in to RayCast() to inspect its values

🏃

3:19:18Step in to RayCast() to inspect its values

🏃

3:21:33Make RayCast() subtract only one element of the FaceRelOrigin, introducing a f32_4x version of -=

3:28:15Step in to RayCast() and inspect the FaceRelOrigin

🏃

3:28:15Step in to RayCast() and inspect the FaceRelOrigin

🏃

3:28:15Step in to RayCast() and inspect the FaceRelOrigin

🏃

3:29:19Run the game to see that we are producing the correct lighting a little faster

🏃

3:29:19Run the game to see that we are producing the correct lighting a little faster

🏃

3:29:19Run the game to see that we are producing the correct lighting a little faster

🏃

3:29:50Enable RayCast() to perform the bounds checking in SIMD

3:41:11Organise handmade_simd.h and introduce f32_4x versions of every operator and function we need

4:02:30Run the game to see that it's totally fine

🏃

4:02:30Run the game to see that it's totally fine

🏃

4:02:30Run the game to see that it's totally fine

🏃

4:03:01Read through RayCast() with the determination to finish doing it all in SIMD

📖

4:03:01Read through RayCast() with the determination to finish doing it all in SIMD

📖

4:03:01Read through RayCast() with the determination to finish doing it all in SIMD

📖

4:04:22Start to enable RayCast() to update and load out the boxes in SIMD, before saving it for tomorrow

4:11:01Q&A

🗩

4:11:01Q&A

🗩

4:11:01Q&A

🗩

4:12:01Clarify the video capture card situation mentioned in the pre-stream

🗩

4:12:01Clarify the video capture card situation mentioned in the pre-stream

🗩

4:12:01Clarify the video capture card situation mentioned in the pre-stream

🗩

4:12:58wired_life Q: I think you call AnyTrue() from within AllTrue(). Can you check that's intended?

🗪

4:12:58wired_life Q: I think you call AnyTrue() from within AllTrue(). Can you check that's intended?

🗪

4:12:58wired_life Q: I think you call AnyTrue() from within AllTrue(). Can you check that's intended?

🗪

4:14:29"And" and "All" in a matrix

🖌

4:14:29"And" and "All" in a matrix

🖌

4:14:29"And" and "All" in a matrix

🖌

4:15:59Call it there

🗩

4:15:59Call it there

🗩

4:15:59Call it there

🗩

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