Examining the CPU Voxel Sampling — Handmade Hero — Episode Guide

0:00Recap and set the stage for the day

🗩

0:00Recap and set the stage for the day

🗩

0:00Recap and set the stage for the day

🗩

0:31Determine to blur the lighting samples across voxels and reducing the flicker, after speeding up the performance of the grid ray tracer

🏃

0:31Determine to blur the lighting samples across voxels and reducing the flicker, after speeding up the performance of the grid ray tracer

🏃

0:31Determine to blur the lighting samples across voxels and reducing the flicker, after speeding up the performance of the grid ray tracer

🏃

3:02Prepare to enable hhlightprof to capture a grid-based ray cast run

6:56Update InternalLightingCore() to dump out the new source_lightboxes

9:47Compile in hhlightprof and update it to work with our grid ray caster

16:27Break in to InternalLightingCore()

🏃

16:27Break in to InternalLightingCore()

🏃

16:27Break in to InternalLightingCore()

🏃

18:24Reload, to see the walk table break

🏃

18:24Reload, to see the walk table break

🏃

18:24Reload, to see the walk table break

🏃

18:48Investigate the walk table breakage on hot reloading

📖

18:48Investigate the walk table breakage on hot reloading

📖

18:48Investigate the walk table breakage on hot reloading

📖

20:52Run in -Od, hot-reload and see the walk table break

🏃

20:52Run in -Od, hot-reload and see the walk table break

🏃

20:52Run in -Od, hot-reload and see the walk table break

🏃

21:32Refamiliarise ourselves with the walk table structure and code

📖

21:32Refamiliarise ourselves with the walk table structure and code

📖

21:32Refamiliarise ourselves with the walk table structure and code

📖

22:42Break in to InternalLightingCore() and inspect the LightSamplingWalkTable

🏃

22:42Break in to InternalLightingCore() and inspect the LightSamplingWalkTable

🏃

22:42Break in to InternalLightingCore() and inspect the LightSamplingWalkTable

🏃

25:28Break in to GridRayCast() and inspect the WalkTable usage

🏃

25:28Break in to GridRayCast() and inspect the WalkTable usage

🏃

25:28Break in to GridRayCast() and inspect the WalkTable usage

🏃

27:02Fix our walk table breakage by making ComputeWalkTable() block copy the SampleDirections, and set a fresh RayD and WalkTableOffset

33:32Run in -Od, hot-reload and see the walk table remain intact

🏃

33:32Run in -Od, hot-reload and see the walk table remain intact

🏃

33:32Run in -Od, hot-reload and see the walk table remain intact

🏃

34:08Run in -O2, hot-reload and see the walk table remain intact

🏃

34:08Run in -O2, hot-reload and see the walk table remain intact

🏃

34:08Run in -O2, hot-reload and see the walk table remain intact

🏃

34:42Enable then disable the LightBoxDumpTrigger, to dump the lighting

🏃

34:42Enable then disable the LightBoxDumpTrigger, to dump the lighting

🏃

34:42Enable then disable the LightBoxDumpTrigger, to dump the lighting

🏃

35:17Check out our lighting dump files, noting the large size of the source_lighting.dump

🗹

35:17Check out our lighting dump files, noting the large size of the source_lighting.dump

🗹

35:17Check out our lighting dump files, noting the large size of the source_lighting.dump

🗹

36:44Hit a read access violation in GetAlignmentOffset() from hhlightprof

🏃

36:44Hit a read access violation in GetAlignmentOffset() from hhlightprof

🏃

36:44Hit a read access violation in GetAlignmentOffset() from hhlightprof

🏃

37:14Make ProfileRun() push the SampleDirectionTable onto the TempArena

37:34Hit a write access violation in PushDebugLine()

🏃

37:34Hit a write access violation in PushDebugLine()

🏃

37:34Hit a write access violation in PushDebugLine()

🏃

37:44Make ProfileRun() disable UpdateDebugLines

38:03Run hhlightprof successfully

🏃

38:03Run hhlightprof successfully

🏃

38:03Run hhlightprof successfully

🏃

38:35Run an -O2 build of hhlightprof

🏃

38:35Run an -O2 build of hhlightprof

🏃

38:35Run an -O2 build of hhlightprof

🏃

38:59hhlightprof total seconds elapsed: 7.646482

🏃

38:59hhlightprof total seconds elapsed: 7.646482

🏃

38:59hhlightprof total seconds elapsed: 7.646482

🏃

40:03Disable LIGHTING_USE_GRID

40:20hhlightprof total seconds elapsed: 7.287836

🏃

40:20hhlightprof total seconds elapsed: 7.287836

🏃

40:20hhlightprof total seconds elapsed: 7.287836

🏃

40:42Save off our timings, enable LIGHTING_USE_GRID and make GridRayCast() return early if not Debugging

42:34A few words on replacing language / compiler "Errors" and "Warnings" with "I could not compile this" and "Things I noticed about the code"

💢

🗩

42:34A few words on replacing language / compiler "Errors" and "Warnings" with "I could not compile this" and "Things I noticed about the code"

💢

🗩

42:34A few words on replacing language / compiler "Errors" and "Warnings" with "I could not compile this" and "Things I noticed about the code"

💢

🗩

45:29hhlightprof total seconds elapsed (without ray casting): 1.246065

🏃

45:29hhlightprof total seconds elapsed (without ray casting): 1.246065

🏃

45:29hhlightprof total seconds elapsed (without ray casting): 1.246065

🏃

46:44Let GridRayCast() do its work

46:54hhlightprof total seconds elapsed: 7.706672

🏃

46:54hhlightprof total seconds elapsed: 7.706672

🏃

46:54hhlightprof total seconds elapsed: 7.706672

🏃

47:10Reorganise GridRayCast() to decrement the CostMetric after the loops, and comment out debugging code

50:18hhlightprof total seconds elapsed: 7.479611

🏃

50:18hhlightprof total seconds elapsed: 7.479611

🏃

50:18hhlightprof total seconds elapsed: 7.479611

🏃

50:46Note why the AABB testing loop in GridRayCast() does not tend to use all four SIMD lanes, and the simplicity of ComputeWalkTable()

📖

50:46Note why the AABB testing loop in GridRayCast() does not tend to use all four SIMD lanes, and the simplicity of ComputeWalkTable()

📖

50:46Note why the AABB testing loop in GridRayCast() does not tend to use all four SIMD lanes, and the simplicity of ComputeWalkTable()

📖

52:53Determine to decouple the spatial and lighting voxel grids

📖

52:53Determine to decouple the spatial and lighting voxel grids

📖

52:53Determine to decouple the spatial and lighting voxel grids

📖

53:55Try decreasing the CostMetric from 16 to 4 in GridRayCast()

54:28hhlightprof total seconds elapsed: 6.321697

🏃

54:28hhlightprof total seconds elapsed: 6.321697

🏃

54:28hhlightprof total seconds elapsed: 6.321697

🏃

55:00Try decreasing the CostMetric from 4 to 0 in GridRayCast()

55:13hhlightprof total seconds elapsed: 4.679288

🏃

55:13hhlightprof total seconds elapsed: 4.679288

🏃

55:13hhlightprof total seconds elapsed: 4.679288

🏃

55:34Interpret our 4.679288 seconds performance when casting no rays

📖

55:34Interpret our 4.679288 seconds performance when casting no rays

📖

55:34Interpret our 4.679288 seconds performance when casting no rays

📖

58:16Alignment of Atlas Cells

🖌

58:16Alignment of Atlas Cells

🖌

58:16Alignment of Atlas Cells

🖌

1:06:48Determine to remove the mutex from our atlas traversal code

📖

1:06:48Determine to remove the mutex from our atlas traversal code

📖

1:06:48Determine to remove the mutex from our atlas traversal code

📖

1:10:18Change ComputeLightPropagationWork() to distribute the lighting computation along the Y axis

1:12:00hhlightprof total seconds elapsed: 4.422329

🏃

1:12:00hhlightprof total seconds elapsed: 4.422329

🏃

1:12:00hhlightprof total seconds elapsed: 4.422329

🏃

1:12:41Consider compacting the lighting atlases

📖

1:12:41Consider compacting the lighting atlases

📖

1:12:41Consider compacting the lighting atlases

📖

1:14:44Consider the performance of ComputeVoxelIrradianceAt()

📖

1:14:44Consider the performance of ComputeVoxelIrradianceAt()

📖

1:14:44Consider the performance of ComputeVoxelIrradianceAt()

📖

1:15:41Try greatly simplifying ComputeVoxelIrradianceAt()

1:16:46hhlightprof total seconds elapsed: 3.160531

🏃

1:16:46hhlightprof total seconds elapsed: 3.160531

🏃

1:16:46hhlightprof total seconds elapsed: 3.160531

🏃

1:16:59Try further simplifying ComputeVoxelIrradianceAt()

1:17:33hhlightprof total seconds elapsed: 1.815950

🏃

1:17:33hhlightprof total seconds elapsed: 1.815950

🏃

1:17:33hhlightprof total seconds elapsed: 1.815950

🏃

1:17:50Determine to speed up ComputeVoxelIrradianceAt()

📖

1:17:50Determine to speed up ComputeVoxelIrradianceAt()

📖

1:17:50Determine to speed up ComputeVoxelIrradianceAt()

📖

1:18:22Instrument ComputeVoxelIrradianceAt() to more specifically gauge its performance

1:21:48hhlightprof total seconds elapsed: 2.806293

🏃

1:21:48hhlightprof total seconds elapsed: 2.806293

🏃

1:21:48hhlightprof total seconds elapsed: 2.806293

🏃

1:22:00Break in to ComputeVoxelIrradianceAt() and inspect the assembly

🏃

1:22:00Break in to ComputeVoxelIrradianceAt() and inspect the assembly

🏃

1:22:00Break in to ComputeVoxelIrradianceAt() and inspect the assembly

🏃

1:23:31Determine to optimise out some of the math in ComputeVoxelIrradianceAt()

📖

1:23:31Determine to optimise out some of the math in ComputeVoxelIrradianceAt()

📖

1:23:31Determine to optimise out some of the math in ComputeVoxelIrradianceAt()

📖

1:26:46Make ComputeVoxelIrradianceAt() operate wide¹^,2

1:49:07Dependents, and Cycle Ordering

1:52:32Continue to make ComputeVoxelIrradianceAt() operate wide³^,4

2:07:23Let GridRayCast() use the original CostMetric

2:07:48We are back to normal

🏃

2:07:48We are back to normal

🏃

2:07:48We are back to normal

🏃

2:07:53Q&A

🗩

2:07:53Q&A

🗩

2:07:53Q&A

🗩

2:08:15sagian2005 Q: Would things go better if you started with U, V and W each 4 wide?

🗪

2:08:15sagian2005 Q: Would things go better if you started with U, V and W each 4 wide?

🗪

2:08:15sagian2005 Q: Would things go better if you started with U, V and W each 4 wide?

🗪

2:09:36cirdanvalen Q: Could you not pad by 8 bytes to fix the overlap?

🗪

2:09:36cirdanvalen Q: Could you not pad by 8 bytes to fix the overlap?

🗪

2:09:36cirdanvalen Q: Could you not pad by 8 bytes to fix the overlap?

🗪

2:10:03mindmark42 Q: The L0 determines how many cache lines the CPU can hold, right?

🗪

2:10:03mindmark42 Q: The L0 determines how many cache lines the CPU can hold, right?

🗪

2:10:03mindmark42 Q: The L0 determines how many cache lines the CPU can hold, right?

🗪

2:10:43rooctag Q: Would you ever consider just going over explaining how an operations takes x amount of CPU ops? Or will that be in the Intro to C?

🗪

2:10:43rooctag Q: Would you ever consider just going over explaining how an operations takes x amount of CPU ops? Or will that be in the Intro to C?

🗪

2:10:43rooctag Q: Would you ever consider just going over explaining how an operations takes x amount of CPU ops? Or will that be in the Intro to C?

🗪

2:10:59ali4410 Q: Hi, do you recommend learning vi keybindings, emacs keybindings, or neither?

🗪

2:10:59ali4410 Q: Hi, do you recommend learning vi keybindings, emacs keybindings, or neither?

🗪

2:10:59ali4410 Q: Hi, do you recommend learning vi keybindings, emacs keybindings, or neither?

🗪

2:12:00mindmark42 Q: The memory caches on the CPU

🗪

2:12:00mindmark42 Q: The memory caches on the CPU

🗪

2:12:00mindmark42 Q: The memory caches on the CPU

🗪

2:13:17Hardware Caches⁵^,6

📖

2:13:17Hardware Caches⁵^,6

📖

2:13:17Hardware Caches⁵^,6

📖

2:22:50mindmark42 Q: Yes, that answers my question. I just was just off by one index

🗪

2:22:50mindmark42 Q: Yes, that answers my question. I just was just off by one index

🗪

2:22:50mindmark42 Q: Yes, that answers my question. I just was just off by one index

🗪

2:23:15billdstrong Q: How much time are you expecting to shave off from making this routine wide? About 2 / 3 or so? What are your expectations for your grid walk optimization? Are you trying to get under 4 seconds or lower?

🗪

2:23:15billdstrong Q: How much time are you expecting to shave off from making this routine wide? About 2 / 3 or so? What are your expectations for your grid walk optimization? Are you trying to get under 4 seconds or lower?

🗪

2:23:15billdstrong Q: How much time are you expecting to shave off from making this routine wide? About 2 / 3 or so? What are your expectations for your grid walk optimization? Are you trying to get under 4 seconds or lower?

🗪

2:23:51mindmark42 Q: Yeah, I mentioned those L caches. I was wondering how to determine how many cache lines a core holds

🗪

2:23:51mindmark42 Q: Yeah, I mentioned those L caches. I was wondering how to determine how many cache lines a core holds

🗪

2:23:51mindmark42 Q: Yeah, I mentioned those L caches. I was wondering how to determine how many cache lines a core holds

🗪

2:24:238-Way Caches⁷^,8

📖

2:24:238-Way Caches⁷^,8

📖

2:24:238-Way Caches⁷^,8

📖

2:29:01Thanks, everyone

🗩

2:29:01Thanks, everyone

🗩

2:29:01Thanks, everyone

🗩

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