Replacing rand() and Preparing for SIMD — Handmade Ray — Episode Guide

0:06Recap and set the stage for the day

🗩

0:06Recap and set the stage for the day

🗩

0:06Recap and set the stage for the day

🗩

1:38Note that we're building in optimised mode

🗩

1:38Note that we're building in optimised mode

🗩

1:38Note that we're building in optimised mode

🗩

2:15Run and see our output image

🏃

2:15Run and see our output image

🏃

2:15Run and see our output image

🏃

3:39ray.cpp: Walk through the code

🗩

3:39ray.cpp: Walk through the code

🗩

3:39ray.cpp: Walk through the code

🗩

5:23Consider two areas of optimisation: 1) Bounding Volume Hierarchy

🗩

5:23Consider two areas of optimisation: 1) Bounding Volume Hierarchy

🗩

5:23Consider two areas of optimisation: 1) Bounding Volume Hierarchy

🗩

6:572) Using better math operations

🗩

6:572) Using better math operations

🗩

6:572) Using better math operations

🗩

7:42Step into RenderTile() and inspect the asm, noting down routines to improve

15:51Check out PCG, A Family of Better Random Number Generators¹ with a recommendation to read the full paper²

📖

15:51Check out PCG, A Family of Better Random Number Generators¹ with a recommendation to read the full paper²

📖

15:51Check out PCG, A Family of Better Random Number Generators¹ with a recommendation to read the full paper²

📖

24:13Check out the x86 SSE2 shift-left instructions³

📖

24:13Check out the x86 SSE2 shift-left instructions³

📖

24:13Check out the x86 SSE2 shift-left instructions³

📖

27:59Read 6.3 - Specific Implementations⁴ and the Xorshift wiki article⁵

📖

27:59Read 6.3 - Specific Implementations⁴ and the Xorshift wiki article⁵

📖

27:59Read 6.3 - Specific Implementations⁴ and the Xorshift wiki article⁵

📖

31:15Introduce XOrShift32() from Wikipedia⁶ with a check into doing this in a 64-bit⁷

37:45Run our program to get a benchmark timing

🏃

37:45Run our program to get a benchmark timing

🏃

37:45Run our program to get a benchmark timing

🏃

38:59Replace rand() with our new XOrShift32(), packing Entropy in the work_order struct

46:39Run to see no obvious problems with our output, and note our dramatically improved performance

🏃

46:39Run to see no obvious problems with our output, and note our dramatically improved performance

🏃

46:39Run to see no obvious problems with our output, and note our dramatically improved performance

🏃

48:47Step into the code and inspect the asm to see a lot of mulss calls

51:30Introduce CastSampleRays() to do some of the work of RenderTile()

58:21Run to see that we lose some speed

🏃

58:21Run to see that we lose some speed

🏃

58:21Run to see that we lose some speed

🏃

59:12Make RenderTile() only use a random_series in its inner loop

1:00:07Run to see that that's a little bit better

🏃

1:00:07Run to see that that's a little bit better

🏃

1:00:07Run to see that that's a little bit better

🏃

1:01:07Rename cast_result to cast_state which contains both the input and output data

1:08:12Run to see some busted imagery

🏃

1:08:12Run to see some busted imagery

🏃

1:08:12Run to see some busted imagery

🏃

1:08:58Fix RenderTile() to correctly fill out the cast_state State

1:12:20Run to see that that helps

🏃

1:12:20Run to see that that helps

🏃

1:12:20Run to see that that helps

🏃

1:13:05Consider how to perform this ray casting wide

🗩

1:13:05Consider how to perform this ray casting wide

🗩

1:13:05Consider how to perform this ray casting wide

🗩

1:18:04Transform CastSampleRays() to handle the notion of operating wide

🗩

1:18:04Transform CastSampleRays() to handle the notion of operating wide

🗩

1:18:04Transform CastSampleRays() to handle the notion of operating wide

🗩

1:19:06Run to see that it runs roughly four times faster, and that the image now contains tile-boundary artifacts

🏃

1:19:06Run to see that it runs roughly four times faster, and that the image now contains tile-boundary artifacts

🏃

1:19:06Run to see that it runs roughly four times faster, and that the image now contains tile-boundary artifacts

🏃

1:21:08Temporarily revert RandomUnlateral() to use rand()

1:21:38Run to see no artifacts, and note that the XOrShift32() needs improving

🏃

1:21:38Run to see no artifacts, and note that the XOrShift32() needs improving

🏃

1:21:38Run to see no artifacts, and note that the XOrShift32() needs improving

🏃

1:22:46Sketch in the code to enable CastSampleRays() to operate wide

1:33:17Describe our current situation

🗩

1:33:17Describe our current situation

🗩

1:33:17Describe our current situation

🗩

1:34:11Set up CastSampleRays() to let all rays in all lanes finish

1:38:21Consider how to track the materials wide

🗩

1:38:21Consider how to track the materials wide

🗩

1:38:21Consider how to track the materials wide

🗩

1:40:08Set up CastSampleRays() to track the materials wide and collate all the computations

1:52:29Create ray_lane.h to #define the lanes, and introduce RandomBilateralLane(), various permutations of ConditionalAssign(), a Max(), MaskIsZeroed() and versions of HorizontalAdd()

2:03:39Run and see totally busted imagery

🏃

2:03:39Run and see totally busted imagery

🏃

2:03:39Run and see totally busted imagery

🏃

2:04:23Build in debug mode and on one core

2:05:40Step in to CastSampleRays() and inspect its values

2:05:56Make CastSampleRays() set FilmX and FilmY to their centres

2:07:14Step in to CastSampleRays() and see that the State->Series and Order->Entropy are both 0

2:08:36Make CastSampleRays() offset the Entropy and use different random series per ray

2:09:27Step in to CastSampleRays() and note that the ConditionalAssign() is wrong

2:10:44Make ConditionalAssign() zero the Mask if there is nothing set in it

2:11:20Step in to ConditionalAssign() to see that that is better

2:11:41Run to see how the picture looks

🏃

2:11:41Run to see how the picture looks

🏃

2:11:41Run to see how the picture looks

🏃

2:13:24View the image

🏃

2:13:24View the image

🏃

2:13:24View the image

🏃

2:13:49Reduce the RayCount and increase the CoreCount

2:14:49Investigate the summation

2:17:53Make CastSampleRays() correctly set the LaneMask

2:18:35Run and see a more correct image

🏃

2:18:35Run and see a more correct image

🏃

2:18:35Run and see a more correct image

🏃

2:18:52Switch back to the optimised version, with more RaysPerPixel

2:19:09Run to see that we're darker

🏃

2:19:09Run to see that we're darker

🏃

2:19:09Run to see that we're darker

🏃

2:20:13Correctly set the LaneWidth

2:21:20Run and see that the images are basically indistinguishable

🏃

2:21:20Run and see that the images are basically indistinguishable

🏃

2:21:20Run and see that the images are basically indistinguishable

🏃

2:22:12Set up to support a constrained set of LANE_WIDTH values

2:30:05Run to see that XOrShift32() is actually fine

🏃

2:30:05Run to see that XOrShift32() is actually fine

🏃

2:30:05Run to see that XOrShift32() is actually fine

🏃

2:31:45Do LANE_WIDTH==8 too

2:32:43Q&A

🗩

2:32:43Q&A

🗩

2:32:43Q&A

🗩

2:33:46yurasniper Q: How would one implement something like bloom effect in a raytracer?

🗪

2:33:46yurasniper Q: How would one implement something like bloom effect in a raytracer?

🗪

2:33:46yurasniper Q: How would one implement something like bloom effect in a raytracer?

🗪

2:39:46Run our program to capture its performance statistics

🏃

2:39:46Run our program to capture its performance statistics

🏃

2:39:46Run our program to capture its performance statistics

🏃

2:42:07macielda Q: Is the Halton 2,3 sequence a good way to generate sample positions? I've heard about some people using it. It is a low discrepancy series

🗪

2:42:07macielda Q: Is the Halton 2,3 sequence a good way to generate sample positions? I've heard about some people using it. It is a low discrepancy series

🗪

2:42:07macielda Q: Is the Halton 2,3 sequence a good way to generate sample positions? I've heard about some people using it. It is a low discrepancy series

🗪

2:43:11Rename our image and stat files

🗹

2:43:11Rename our image and stat files

🗹

2:43:11Rename our image and stat files

🗹

2:44:30vaualbus Q: When you learn this way of doing SIMD? I remember in Handmade Hero when we had optimized the renderer we use __m128 every way

🗪

2:44:30vaualbus Q: When you learn this way of doing SIMD? I remember in Handmade Hero when we had optimized the renderer we use __m128 every way

🗪

2:44:30vaualbus Q: When you learn this way of doing SIMD? I remember in Handmade Hero when we had optimized the renderer we use __m128 every way

🗪

2:46:05macielda Q: What is your take on AA methods? I'm currently looking for one for my game. I see The Witness has MSAA option only (no FXAA, TXAA and friends)?

🗪

2:46:05macielda Q: What is your take on AA methods? I'm currently looking for one for my game. I see The Witness has MSAA option only (no FXAA, TXAA and friends)?

🗪

2:46:05macielda Q: What is your take on AA methods? I'm currently looking for one for my game. I see The Witness has MSAA option only (no FXAA, TXAA and friends)?

🗪

2:46:31longboolean Q: Are there any machines with hardware RNG that just puts random values into a register with one instruction?⁸

🗪

2:46:31longboolean Q: Are there any machines with hardware RNG that just puts random values into a register with one instruction?⁸

🗪

2:46:31longboolean Q: Are there any machines with hardware RNG that just puts random values into a register with one instruction?⁸

🗪

2:48:44pseudonym73 Q: G'day, long time no stream. Low-discrepancy sequences do exhibit blue noise behaviours if you do them right, but their main advantage is that you can access the quasi-random streams in an arbitrary order. Not really relevant yet. Also, you can do better than 2,3 Halton

🗪

2:48:44pseudonym73 Q: G'day, long time no stream. Low-discrepancy sequences do exhibit blue noise behaviours if you do them right, but their main advantage is that you can access the quasi-random streams in an arbitrary order. Not really relevant yet. Also, you can do better than 2,3 Halton

🗪

2:48:44pseudonym73 Q: G'day, long time no stream. Low-discrepancy sequences do exhibit blue noise behaviours if you do them right, but their main advantage is that you can access the quasi-random streams in an arbitrary order. Not really relevant yet. Also, you can do better than 2,3 Halton

🗪

2:49:37macielda Q: Do shader languages expose things like "Conditional Assign"?

🗪

2:49:37macielda Q: Do shader languages expose things like "Conditional Assign"?

🗪

2:49:37macielda Q: Do shader languages expose things like "Conditional Assign"?

🗪

2:51:14Ensure that everything is in good shape

🗹

2:51:14Ensure that everything is in good shape

🗹

2:51:14Ensure that everything is in good shape

🗹

2:52:14Shut down

🗩

2:52:14Shut down

🗩

2:52:14Shut down

🗩

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