Finishing the Main SIMD Raycasting Loop | Handmade Hero Episode Guide

0:01Recap and set the stage for the day finishing SIMD optimising the lighting

🗩

0:01Recap and set the stage for the day finishing SIMD optimising the lighting

🗩

0:01Recap and set the stage for the day finishing SIMD optimising the lighting

🗩

0:47Toggle on the threading and add a b32 Hit to raycast_result for RayCast() to encode that a ray did not hit

5:57Run the game to find that we're running at 32ms

🏃

5:57Run the game to find that we're running at 32ms

🏃

5:57Run the game to find that we're running at 32ms

🏃

6:29Toggle off the threading

6:49Run the game to see that we're running at 128ms per frame

🏃

6:49Run the game to see that we're running at 128ms per frame

🏃

6:49Run the game to see that we're running at 128ms per frame

🏃

7:18Toggle on the threading

7:26Run the game and consider our 32ms per frame rate

🏃

7:26Run the game and consider our 32ms per frame rate

🏃

7:26Run the game and consider our 32ms per frame rate

🏃

9:43Excise from RayCast() the non-SIMD tRay code, and start to consider how to retire rays hits

11:42Preserving ray hits vs traversing the spatial hierarchy, when threading

🖌

11:42Preserving ray hits vs traversing the spatial hierarchy, when threading

🖌

11:42Preserving ray hits vs traversing the spatial hierarchy, when threading

🖌

15:57Enable RayCast() to record ray hits for each SIMD component before traversing the spatial hierarchy

20:02Run the game to see that we're running at the same 32ms

🏃

20:02Run the game to see that we're running at the same 32ms

🏃

20:02Run the game to see that we're running at the same 32ms

🏃

20:37Revert RayCast() to traverse the spatial hierarchy, applying the ray hit mask for each component, and streamline how this works, introducing f32_4x versions of &= and |=

26:58Run the game to see that that's fine

🏃

26:58Run the game to see that that's fine

🏃

26:58Run the game to see that that's fine

🏃

27:06Start to streamline the tRay setting code

28:56Fix the CloseEnough check in RayCast()

30:05Run the game to see not much difference

🏃

30:05Run the game to see not much difference

🏃

30:05Run the game to see not much difference

🏃

30:20Introduce Select() to streamline the tRay setting code¹

34:15Run the game to see that we are at ~28ms per frame

🏃

34:15Run the game to see that we are at ~28ms per frame

🏃

34:15Run the game to see that we are at ~28ms per frame

🏃

34:44Make RayCast() set the BoxIndex and BoxSurface in SIMD using Select()

42:30Run the game and crash in ComputeLightPropagation()

🏃

42:30Run the game and crash in ComputeLightPropagation()

🏃

42:30Run the game and crash in ComputeLightPropagation()

🏃

44:45Step in to GetBox() to see that our BoxIndex is busted

🏃

44:45Step in to GetBox() to see that our BoxIndex is busted

🏃

44:45Step in to GetBox() to see that our BoxIndex is busted

🏃

46:33Step through RayCast() to see what's happening

🏃

46:33Step through RayCast() to see what's happening

🏃

46:33Step through RayCast() to see what's happening

🏃

49:46Make RayCast() actually set the BoxIndex and BoxSurfaceIndex

🦉

🖮

49:46Make RayCast() actually set the BoxIndex and BoxSurfaceIndex

🦉

🖮

49:46Make RayCast() actually set the BoxIndex and BoxSurfaceIndex

🦉

🖮

50:55Run the game with the selection happening

🏃

🦉

50:55Run the game with the selection happening

🏃

🦉

50:55Run the game with the selection happening

🏃

🦉

51:09Make RayCast() set the RayP in SIMD using a v3_4x version of Select()

🦉

🖮

51:09Make RayCast() set the RayP in SIMD using a v3_4x version of Select()

🦉

🖮

51:09Make RayCast() set the RayP in SIMD using a v3_4x version of Select()

🦉

🖮

53:38Run the game to see that we are down to ~26ms

🏃

🦉

53:38Run the game to see that we are down to ~26ms

🏃

🦉

53:38Run the game to see that we are down to ~26ms

🏃

🦉

54:22Add a TIMED_FUNCTION() in RayCast()

🦉

🖮

54:22Add a TIMED_FUNCTION() in RayCast()

🦉

🖮

54:22Add a TIMED_FUNCTION() in RayCast()

🦉

🖮

54:41Run the game to consult the profiler

🏃

🦉

54:41Run the game to consult the profiler

🏃

🦉

54:41Run the game to consult the profiler

🏃

🦉

54:55Add a TIMED_BLOCK() around the startup code in RayCast()

🦉

🖮

54:55Add a TIMED_BLOCK() around the startup code in RayCast()

🦉

🖮

54:55Add a TIMED_BLOCK() around the startup code in RayCast()

🦉

🖮

55:35Run the game and consult the profiler to see that the startup cost is not high

🏃

🦉

55:35Run the game and consult the profiler to see that the startup cost is not high

🏃

🦉

55:35Run the game and consult the profiler to see that the startup cost is not high

🏃

🦉

56:00Perform SampleHemisphere() in SIMD

🦉

🖮

56:00Perform SampleHemisphere() in SIMD

🦉

🖮

56:00Perform SampleHemisphere() in SIMD

🦉

🖮

1:01:22Run the game to see that we're down to 22ms per frame

🏃

🦉

1:01:22Run the game to see that we're down to 22ms per frame

🏃

🦉

1:01:22Run the game to see that we're down to 22ms per frame

🏃

🦉

1:02:04Temporarily make SampleHemisphere() use complete randomisation

🦉

🖮

1:02:04Temporarily make SampleHemisphere() use complete randomisation

🦉

🖮

1:02:04Temporarily make SampleHemisphere() use complete randomisation

🦉

🖮

1:02:20Run the game to see that this would put us back up to 30ms per frame, and note why

🏃

🦉

1:02:20Run the game to see that this would put us back up to 30ms per frame, and note why

🏃

🦉

1:02:20Run the game to see that this would put us back up to 30ms per frame, and note why

🏃

🦉

1:04:08Drop the RayCount down to 4 in ComputeLightPropagation()

🦉

🖮

1:04:08Drop the RayCount down to 4 in ComputeLightPropagation()

🦉

🖮

1:04:08Drop the RayCount down to 4 in ComputeLightPropagation()

🦉

🖮

1:04:25Run the game and unexpectedly see no speed improvement

🏃

🦉

1:04:25Run the game and unexpectedly see no speed improvement

🏃

🦉

1:04:25Run the game and unexpectedly see no speed improvement

🏃

🦉

1:05:45Remove variable suffixes in RayCast()

🦉

🖮

1:05:45Remove variable suffixes in RayCast()

🦉

🖮

1:05:45Remove variable suffixes in RayCast()

🦉

🖮

1:08:55Consider removing the Depth loop in RayCast() and reposition the AnyTrue(Mask) test

1:10:31Run the game and consider where to go from here

🏃

1:10:31Run the game and consider where to go from here

🏃

1:10:31Run the game and consider where to go from here

🏃

1:11:25Inspect the assembly of RayCast()

1:14:32Remove the Mask tests from RayCast() entirely

1:15:16Run the game to see no real difference

🏃

1:15:16Run the game to see no real difference

🏃

1:15:16Run the game to see no real difference

🏃

1:15:40Try removing the AnyTrue(tCheck)

1:15:55Run the game to see that that would put us up to ~25ms per frame

🏃

1:15:55Run the game to see that that would put us up to ~25ms per frame

🏃

1:15:55Run the game to see that that would put us up to ~25ms per frame

🏃

1:16:51Compute RayP at the very end of RayCast()

1:18:30Run the game to see no difference

🏃

1:18:30Run the game to see no difference

🏃

1:18:30Run the game to see no difference

🏃

1:18:41Replace RayP with tRay in RayCast()

1:20:40Run the game to see no difference

🏃

1:20:40Run the game to see no difference

🏃

1:20:40Run the game to see no difference

🏃

1:21:01Let RayCast() break if(AllTrue(Mask))

1:21:12Run the game to see no difference

🏃

1:21:12Run the game to see no difference

🏃

1:21:12Run the game to see no difference

🏃

1:21:37Toggle off the snake

1:21:47Run the game with our consistently lit scene

🏃

1:21:47Run the game with our consistently lit scene

🏃

1:21:47Run the game with our consistently lit scene

🏃

1:22:28Inline AccumulateSample() in ComputeLightPropagation()

1:25:37Run the game to see no difference, and consider further improvements

🏃

1:25:37Run the game to see no difference, and consider further improvements

🏃

1:25:37Run the game to see no difference, and consider further improvements

🏃

1:27:06Make SampleHemisphere() operate entirely SIMD, introducing RandomBilateral_4x() and versions of Inner(), LengthSq() and NOZ() that take v3_4x

1:41:35A few words on _mm_rsqrt_ps and _mm_sqrt_ps²

📖

1:41:35A few words on _mm_rsqrt_ps and _mm_sqrt_ps²

📖

1:41:35A few words on _mm_rsqrt_ps and _mm_sqrt_ps²

📖

1:44:20Rename our new NOZ() to ApproxNOZ() for SampleHemisphere() to call, and introduce ApproxInvSquareRoot() using _mm_rsqrt_ps³

1:49:24Run the game to see no difference

🏃

1:49:24Run the game to see no difference

🏃

1:49:24Run the game to see no difference

🏃

1:49:52Inline SampleHemisphere() in ComputeLightPropagation()

1:51:51Run the game at ~22ms per frame, and consider that this CPU rendered lighting is performant enough for us

🏃

1:51:51Run the game at ~22ms per frame, and consider that this CPU rendered lighting is performant enough for us

🏃

1:51:51Run the game at ~22ms per frame, and consider that this CPU rendered lighting is performant enough for us

🏃

1:52:52Q&A

🗩

1:52:52Q&A

🗩

1:52:52Q&A

🗩

1:53:22printf_armin How much % of the CPU does it drain?

🗪

1:53:22printf_armin How much % of the CPU does it drain?

🗪

1:53:22printf_armin How much % of the CPU does it drain?

🗪

1:56:07tbodt_ Q: What version control system do you use?

🗪

1:56:07tbodt_ Q: What version control system do you use?

🗪

1:56:07tbodt_ Q: What version control system do you use?

🗪

1:56:50nxsy Q: Can you look at the thread profile view when you lower the samples from 16 to 4 and don’t improve frame rate?

🗪

🏃

1:56:50nxsy Q: Can you look at the thread profile view when you lower the samples from 16 to 4 and don’t improve frame rate?

🗪

🏃

1:56:50nxsy Q: Can you look at the thread profile view when you lower the samples from 16 to 4 and don’t improve frame rate?

🗪

🏃

1:57:48Temporarily Toggle off VSync

1:58:46Run the game and consult the profiler to determine that the pixel shader is too slow

🏃

1:58:46Run the game and consult the profiler to determine that the pixel shader is too slow

🏃

1:58:46Run the game and consult the profiler to determine that the pixel shader is too slow

🏃

2:02:17mallesbixie Q: You cast the input value in the U32_4x loader to a float. Why is that?⁴

🗪

2:02:17mallesbixie Q: You cast the input value in the U32_4x loader to a float. Why is that?⁴

🗪

2:02:17mallesbixie Q: You cast the input value in the U32_4x loader to a float. Why is that?⁴

🗪

2:05:32jamoflaw Q: How does the mask replace an if in the SIMD instructions?

🗪

2:05:32jamoflaw Q: How does the mask replace an if in the SIMD instructions?

🗪

2:05:32jamoflaw Q: How does the mask replace an if in the SIMD instructions?

🗪

2:05:46Masking in SIMD⁵^,6

🖌

2:05:46Masking in SIMD⁵^,6

🖌

2:05:46Masking in SIMD⁵^,6

🖌

2:19:38Read about PBLENDVB - 'Variable Blend Packed Bytes' in the Intel 64 and IA-32 Architectures Software Developer Manuals⁷ and consult the Steam Hardware Survey⁸ for instruction set use

📖

2:19:38Read about PBLENDVB - 'Variable Blend Packed Bytes' in the Intel 64 and IA-32 Architectures Software Developer Manuals⁷ and consult the Steam Hardware Survey⁸ for instruction set use

📖

2:19:38Read about PBLENDVB - 'Variable Blend Packed Bytes' in the Intel 64 and IA-32 Architectures Software Developer Manuals⁷ and consult the Steam Hardware Survey⁸ for instruction set use

📖

2:25:23alexkelbo Q: Why does the light flicker even when the cube is not moving?

🗪

2:25:23alexkelbo Q: Why does the light flicker even when the cube is not moving?

🗪

2:25:23alexkelbo Q: Why does the light flicker even when the cube is not moving?

🗪

2:25:42vaualbus Q: Could we ship with SSE4, so we have _m256 and get more performance improvement?

🗪

2:25:42vaualbus Q: Could we ship with SSE4, so we have _m256 and get more performance improvement?

🗪

2:25:42vaualbus Q: Could we ship with SSE4, so we have _m256 and get more performance improvement?

🗪

2:25:52sgtrumbi Q: What can you see in TaskManager's GPU tab?

🗪

2:25:52sgtrumbi Q: What can you see in TaskManager's GPU tab?

🗪

2:25:52sgtrumbi Q: What can you see in TaskManager's GPU tab?

🗪

2:27:25longboolean Q: Do you have any tips on talking with non programmers about programming related concepts?

🗪

2:27:25longboolean Q: Do you have any tips on talking with non programmers about programming related concepts?

🗪

2:27:25longboolean Q: Do you have any tips on talking with non programmers about programming related concepts?

🗪

2:27:34tbodt_ Q: How does your profiler work? Does it hook into the compiler or something?

🗪

2:27:34tbodt_ Q: How does your profiler work? Does it hook into the compiler or something?

🗪

2:27:34tbodt_ Q: How does your profiler work? Does it hook into the compiler or something?

🗪

2:27:51alexkelbo Q: Could we compile one exe with SSE4 that falls back to SSE2 if it's not present on the CPU, or would we need to compile into several exe's?

🗪

2:27:51alexkelbo Q: Could we compile one exe with SSE4 that falls back to SSE2 if it's not present on the CPU, or would we need to compile into several exe's?

🗪

2:27:51alexkelbo Q: Could we compile one exe with SSE4 that falls back to SSE2 if it's not present on the CPU, or would we need to compile into several exe's?

🗪

2:28:50Close up shop

🗩

2:28:50Close up shop

🗩

2:28:50Close up shop

🗩

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