cinera_handmade.network/cmuratori/hero/code/code649.hmml

[video output=day649 member=cmuratori stream_platform=twitch stream_username=handmade_hero  project=code title="Removing Lighting Walk Tables" vod_platform=youtube id=kcmku4TMTek annotator=Miblo]
[0:02][Recap and set the stage for the day debugging :lighting][:speech]
[1:03][Demo our light bleed bug][:lighting :run]
[1:45][Demo our light copying bug][:lighting :run]
[6:00][Describe BlockCopyAtlas()][:lighting :research]
[7:09][Assert in BlockCopyAtlas() that the Z loop doesn't happen][:lighting]
[7:50][Hit our Z loop assertion in BlockCopyAtlas()][:lighting :run]
[8:49][Move our Z loop assertion in BlockCopyAtlas() down to the Z clearing loop][:lighting]
[9:11][Hit our Z clearing loop assertion in BlockCopyAtlas()][:lighting :run]
[9:45][Inspect the BlockCopyAtlas() values in an -Od build][:lighting :run]
[11:26][Comment out the Z clearing loop in BlockCopyAtlas()][:lighting]
[11:51][Try to eyeball our light copying bug][:lighting :run]
[12:14][Toggle off DebugDrawSpatialGrid() in UpdateLighting()][:"debug visualisation" :lighting]
[12:44][Our light copying / flashing bug seems to still occur][:lighting :run]
[14:47][Uncomment the Z clearing loop in BlockCopyAtlas()][:lighting]
[15:14][Scour BlockCopyAtlas() for bugs][:lighting :research]
[18:07][Remove the minus 1 from the StopZ computation in BlockCopyAtlas()][:lighting]
[18:35][We still see a little bit of flashing][:lighting :run]
[18:54][Enable LIGHT_ATLAS_ASSERT][:lighting]
[19:25][Immediately trap on VALIDATE_TEXEL_ATLAS() of the SpecAtlas][:lighting :run]
[19:35][Toggle off the VALIDATE_TEXEL_ATLAS() calls in UpdateLighting()][:lighting]
[19:47][Trap on VALIDATE_TEXEL_ATLAS() in BlockCopyAtlas()][:lighting :run]
[19:54][Toggle off the VALIDATE_TEXEL_ATLAS() calls using #if TEMPORARY][:lighting]
[20:19][Trap on VALIDATE_TEXEL_ATLAS() at the end of BlockCopyAtlas()][:lighting :run]
[20:29][Toggle off the final VALIDATE_TEXEL_ATLAS() call using #if TEMPORARY][:lighting]
[20:43][Successfully :run without fetching out of bounds][:lighting :run]
[21:02][Remove the minus 1 from the StopX and StopY computations in BlockCopyAtlas()][:lighting]
[21:22][We have :lighting movement across tiles][:run]
[23:25][Decrease tUpdateBlend from 1 to 15/60 in UpdateLighting()][:lighting]
[23:49][Still see the :lighting shift][:run]
[24:13][Decrease tUpdateBlend from 15/60 to 1/60 in UpdateLighting()][:lighting]
[24:29][The shift is due to the light being recomputed differently][:lighting :run]
[25:22][Toggle off the -1 to 1 origin cube in UpdateAndRenderWorld()][:"debug visualisation" :lighting]
[26:59][The wall tiles darken after shifting][:lighting :run]
[28:07][Toggle off PushLight() on the DebugLightP in UpdateAndRenderWorld()][:lighting]
[28:26][The room still seems to brighten when we enter][:lighting :run]
[30:18][The light bleed takes longer with our slower tUpdateBlend][:lighting :run]
[31:01][Increase tUpdateBlend from 1/60 to 1 in UpdateLighting()][:lighting]
[31:50][The copy is now fine, but the ray casting differs when shifting the lighting voxel][:lighting :run]
[34:12][Prevent UpdateLighting() from offsetting VoxCameraOffset.Y by 2][:lighting]
[35:03][The lighting region is now centred][:lighting :run]
[36:38][Scour GridRayCast() for a reason for differences when shifting the lighting voxel][:lighting :research]
[41:16][Scour ComputeWalkTable() for bugs][:lighting :research]
[42:40][Make WalkTableOffset be an array of 4 in sample_direction, for ComputeWalkTable() to use][:"data structure" :lighting]
[43:12][The light doesn't seem to have changed][:lighting :run]
[43:58][Double-check ComputeWalkTable() for correctness][:lighting :research]
[45:16][Consider removing lighting walk tables][:lighting :research]
[46:56][The ray casting still differs when shifting the lighting voxel][:lighting :run]
[47:27][Consider making UpdateLighting() force dVoxel to be a multiple of 2 so the quadrant doesn't change][:lighting :research]
[48:32][@gir33][The problem changed, though, there's still a dark spot but it doesn't move now][:lighting]
[48:51][Demo the moving dark spot][:lighting :run]
[49:46][@gir33][I mean it seems like it only moved when leaving the room but not while moving within the room as before, no?][:lighting]
[50:07][Demo the changing dark spot when the lighting voxel alignment changes within a room][:lighting :run]
[50:59][Consider removing lighting walk tables][:lighting :research]
[55:15][@mallesbixie][FruFru is a great variable name, though]
[55:20][Consider removing lighting walk tables][:lighting :research]
[56:18][Check GridRayCast() to inform our lighting walk table removal decision][:lighting :research]
[59:50][Toggle on WalkTable stepping in GridRayCast()][:lighting]
[1:00:06][Traverse the orphanage into a relatively stably lit room][:lighting :run]
[1:00:17][Change the CostMetric-based ray casting loop in GridRayCast() to iterate over a hard-bounded GridWalkIteration][:lighting]
[1:00:47][The lighting looks the same][:lighting :run]
[1:00:56][Remove CostMetric entirely from GridRayCast()][:lighting]
[1:01:23][The lighting remains fine][:lighting :run]
[1:01:46][Consider formulating an exit criterion in GridRayCast() upon leaving the spatial partition][:lighting :research]
[1:04:29][Determine to make GridRayCast() work without a walk table][:lighting :research]
[1:09:25][Sketch out walk table-eliminated grid stepping in GridRayCast()][:lighting]
[1:45:42][@spacenaming][DirCurP is 2*DirCurP+dX/Y at the moment][:lighting]
[1:46:06][@spacenaming][XStepV have it included already][:lighting]
[1:46:35][Determine to remove lighting walk tables][:lighting :research]
[1:47:31][Break][:admin]
[1:48:10][:afk]
[1:50:06][Return and centre the camera][:admin]
[1:52:25][Implement on-the-fly grid stepping in GridRayCast()][:lighting]
[2:13:50][Hit an overflow read access violation on SpatialGridLeaves in GridRayCast()][:lighting :run]
[2:14:03][Introduce a CheckGridIndex to separate out the two grid stepping routines in GridRayCast()][:lighting]
[2:14:46][The lighting behaves the same as before][:lighting :run]
[2:15:11][Break in to GridRayCast() and compare the values of the two grid stepping routines][:lighting :run]
[2:16:33][Fix the dGridIndexXBase and dGridIndexYBase computations in GridRayCast()][:lighting]
[2:16:51][Break in to GridRayCast() and compare the values of the two grid stepping routines][:lighting :run]
[2:17:51][Assert in GridRayCast() that GridIndex and CheckGridIndex match][:lighting]
[2:18:14][:Run the game][:lighting]
[2:18:24][Update ~remedybg from 0.3.5.0 to 0.3.6.4][:admin]
[2:20:21][Hit the GridIndex and CheckGridIndex matching assertion in GridRayCast()][:lighting :run]
[2:21:49][Swap in the new on-the-fly grid stepping in GridRayCast()][:lighting]
[2:22:49][Our ray casting correctly no longer differs when shifting the lighting voxel][:lighting :run]
[2:24:19][Consider how to update tTerminate][:lighting :research]
[2:29:46][Make GridRayCast() compute tTerminate on the fly][:lighting]
[2:38:43][Our lighting happily looks no different][:lighting :run]
[2:39:08][Remove the :lighting walk tables][:lighting]
[2:41:08][:Run successfully and briefly consider our remaining light bleed bug][:lighting]
[2:44:28][Reflect on today's work and consider further simplifications][:lighting :speech]
[2:45:35][\~82% frame time (\~400,000,000 cycles) on ComputeLightPropagationWork][:lighting :performance :run]
[2:46:43][Make GridRayCast() only cast one ray][:lighting]
[2:47:05][\~35% frame time (\~35,000,000 cycles) on ComputeLightPropagationWork][:lighting :performance :run]
[2:47:36][Make GridRayCast() cast the full LIGHTING_OCTAHEDRAL_RAYS_PER_TEXEL][:lighting]
[2:47:42][\~82% frame time (\~400,000,000 cycles) on ComputeLightPropagationWork][:lighting :performance :run]
[2:48:00][Make GridRayCast() only cast no rays][:lighting]
[2:48:08][\~11% frame time (\~8,800,000 cycles) on ComputeLightPropagationWork][:lighting :performance :run]
[2:48:24][Make GridRayCast() cast the full LIGHTING_OCTAHEDRAL_RAYS_PER_TEXEL][:lighting]
[2:48:27][\~82% frame time (\~400,000,000 cycles) on ComputeLightPropagationWork][:lighting :performance :run]
[2:48:58][Make GridRayCast() cast half of the LIGHTING_OCTAHEDRAL_RAYS_PER_TEXEL][:lighting]
[2:49:12][Admire the cool red light bleed][:lighting :run]
[2:49:49][Make GridRayCast() double the RayWeight to compensate for the halved number of rays][:lighting]
[2:50:09][Admire the apparently structured light bleed bug][:lighting :run]
[2:51:28][Let GridRayCast() cast all the rays, at their appropriate weight][:lighting]
[2:51:51][Disable LIGHT_ATLAS_ASSERT][:lighting]
[2:52:10][\~28ms frame time][:lighting :performance :run]
[2:52:36][Disable HANDMADE_SLOW]
[2:52:48][\~24ms frame time][:lighting :performance :run]
[2:53:33][That's it for today][:speech]
[2:54:12][@gir33][Are the GPU shaders just lerping from the results that the CPU is doing?][:lighting]
[2:57:07][@gir33][Is that like a spherical harmonics encoding?][:lighting]
[3:01:02][That's it for today, with a glimpse into the future and reflections on the engine][:speech]
[3:02:42][@vaualbus][Can Z be re-added in an easier way now that the system is easier?][:lighting]
[3:05:13][That's it][:speech]
[/video]