Raytracing pipeline demo #173

kevyuu · 2025-01-22T17:07:23Z

Implement a demo to cover various ray tracing pipeline use cases.

multiple closest hit and any hit shaders
multiple miss shaders
multiple callable shaders

devshgraphicsprogramming · 2025-01-22T20:17:54Z

common/include/CCamera.hpp

-	void mouseProcess(const nbl::ui::IMouseEventChannel::range_t& events)
+	// return whether camera is moved by mouse
+	bool mouseProcess(const nbl::ui::IMouseEventChannel::range_t& events)
 	{
+		bool cameraMoved = false;


are these your changes?

coordinate with @AnastaZIuk

It is my changes.

this bool is not a reliable way to find out if the camera moved during a frame (nor is the keyboard one)

I'd perform any such checks in endInputProcessing by comparing matrices (or pos and target) before and after setting position and target

Signed-off-by: kevyuu <[email protected]>

- Multiple HitGroup. Each with closesthit and anythit shader - Multiple Miss Shader Group. Signed-off-by: kevyuu <[email protected]>

Signed-off-by: kevyuu <[email protected]>

Signed-off-by: Ali Cheraghi <[email protected]> Signed-off-by: kevyuu <[email protected]>

Signed-off-by: kevyuu <[email protected]>

devshgraphicsprogramming

main.cpp I'll review later

devshgraphicsprogramming · 2025-01-24T20:53:07Z

71_RayTracingPipeline/app_resources/common.hlsl

+enum E_LIGHT_TYPE : uint16_t
+{
+    ELT_DIRECTIONAL,
+    ELT_POINT,
+    ELT_SPOT,
+    ELT_COUNT
+};
+
+struct Light
+{
+    float32_t3 direction;
+    float32_t3 position;
+    float32_t outerCutoff;
+    uint16_t type;
+
+
+#ifndef __HLSL_VERSION
+    bool operator==(const Light&) const = default;
+#endif
+
+};


keep example simple, do just directional light

devshgraphicsprogramming · 2025-01-24T20:54:06Z

71_RayTracingPipeline/app_resources/random.hlsl

+// Generate a random unsigned int from two unsigned int values, using 16 pairs
+// of rounds of the Tiny Encryption Algorithm. See Zafar, Olano, and Curtis,
+// "GPU Random Numbers via the Tiny Encryption Algorithm"
+uint32_t tea(uint32_t val0, uint32_t val1)
+{
+  uint32_t v0 = val0;
+  uint32_t v1 = val1;
+  uint32_t s0 = 0;
+
+  for(uint32_t n = 0; n < 16; n++)
+  {
+    s0 += 0x9e3779b9;
+    v0 += ((v1 << 4) + 0xa341316c) ^ (v1 + s0) ^ ((v1 >> 5) + 0xc8013ea4);
+    v1 += ((v0 << 4) + 0xad90777d) ^ (v0 + s0) ^ ((v0 >> 5) + 0x7e95761e);
+  }
+
+  return v0;
+}
+
+// Generate a random unsigned int in [0, 2^24) given the previous RNG state
+// using the Numerical Recipes linear congruential generator
+uint32_t lcg(inout uint32_t prev)
+{
+  uint32_t LCG_A = 1664525u;
+  uint32_t LCG_C = 1013904223u;
+  prev       = (LCG_A * prev + LCG_C);
+  return prev & 0x00FFFFFF;
+}
+
+// Generate a random float32_t in [0, 1) given the previous RNG state
+float32_t rnd(inout uint32_t prev)
+{
+  return (float32_t(lcg(prev)) / float32_t(0x01000000));
+}


you may want to rework them into something resembling xoroshiro.hlsl in the Nabla repo and pcg.hlsl from the HLSL BxDF PR https://github.com/Devsh-Graphics-Programming/Nabla/pull/811/files#diff-717f65cfd315d91ced5e7da9f817f786fb1f8c1b1c06a570df4f61be3660a643

devshgraphicsprogramming · 2025-01-24T21:03:25Z

71_RayTracingPipeline/app_resources/raytrace.rahit.hlsl

+    uint32_t seed = p.seed;
+    if (geom.material.dissolve == 0.0)
+        IgnoreHit();
+    else if (rnd(seed) > geom.material.dissolve)
+        IgnoreHit();


~~ok so you stochastically do alpha, btw for this to add up, you'd need to "boost" the color (throughput in the color payload) by the probability of the path taken, so:~~

~~if you ignore the hit, you mul by 1.f/(1.f-dissolve)~~

~~if you don't ignore the hit, you mul by 1.f/dissolve~~

~~basically same as importance sampling.~~ Nvm they cancel out and you never multiplied them into the throughput.

For the shadow ray there is a more performant way actually, instead of importance sampling the mix

this*alpha+next*(1-alpha)

you can recognise that the shadow ray doesn't have to be a boolean and you can multiply the dissolve into it same you would an alpha value (you do need to create the BLAS with the no-duplicate hit flag though).

Also at this point you can make two separate anyhit HLSL shader sources.

devshgraphicsprogramming · 2025-01-24T21:05:21Z

71_RayTracingPipeline/app_resources/raytrace.rahit.hlsl

+    if (geom.material.illum != 4)
+        return;


you can build your TLAS with instance flags that prevent ANY_HIT being called on them (or what it only BLAS?).

This is useful because you don't end up executing the vk::RawBufferLoad for the things that are not transparent to check if they are transparent (saving Bandwidth in the meantime).

devshgraphicsprogramming · 2025-01-24T21:06:44Z

71_RayTracingPipeline/app_resources/common.hlsl

+
+struct [raypayload] ShadowPayload
+{
+	bool isShadowed : read(caller) : write(caller,miss);


as I've mentioned in the comment in your anyhit shader, its better to have a float32_t opacity and have the anyhit multiply into it, which allows for transparent shadows, you have to use the 4 bytes in your payload anyway.

devshgraphicsprogramming · 2025-01-24T22:07:38Z

71_RayTracingPipeline/app_resources/raytrace.rint.hlsl

+struct Attrib
+{
+    float3 HitAttribute;
+};


you never fill this out?

devshgraphicsprogramming · 2025-01-24T22:20:19Z

71_RayTracingPipeline/app_resources/raytrace_procedural.rchit.hlsl

+[[vk::binding(0, 0)]] RaytracingAccelerationStructure topLevelAS;
+
+[shader("closesthit")]
+void main(inout ColorPayload p, in BuiltInTriangleIntersectionAttributes attribs)


BuiltInTriangleIntersectionAttributes? but its not a triangle?

devshgraphicsprogramming · 2025-01-24T22:21:28Z

71_RayTracingPipeline/app_resources/raytrace_procedural.rchit.hlsl

+    SProceduralGeomInfo sphere = vk::RawBufferLoad < SProceduralGeomInfo > (pc.proceduralGeomInfoBuffer + primID * sizeof(SProceduralGeomInfo));
+
+    // Computing the normal at hit position
+    float32_t3 worldNormal = normalize(worldPosition - sphere.center);


you already had to load and compute some of this in the intersection shader, is it not possible to pass it in the payload/custom attribs to the closest hit shader?

devshgraphicsprogramming · 2025-01-24T22:25:50Z

71_RayTracingPipeline/app_resources/raytrace_procedural.rchit.hlsl

+        uint flags =
+            RAY_FLAG_ACCEPT_FIRST_HIT_AND_END_SEARCH | RAY_FLAG_FORCE_OPAQUE |
+            RAY_FLAG_SKIP_CLOSEST_HIT_SHADER;
+
+        ShadowPayload shadowPayload;
+        shadowPayload.isShadowed = true;
+        shadowPayload.seed = p.seed;
+        TraceRay(topLevelAS, flags, 0xFF, ERT_OCCLUSION, 0, EMT_OCCLUSION, rayDesc, shadowPayload);


btw if you follow my advice on transparent shadows, the flags will have to change

P.S. do IgnoreHit() and modifying the payload mix? (Idea is that accepted hit will not cause the invocation of a miss shader or any more anyhits).

devshgraphicsprogramming · 2025-01-24T22:39:38Z

71_RayTracingPipeline/app_resources/raytrace.rchit.hlsl

+        rayDesc.TMin = 0.01;
+        rayDesc.TMax = cLight.outLightDistance;
+
+        uint flags = RAY_FLAG_SKIP_CLOSEST_HIT_SHADER;


what's the reason for not having RAY_FLAG_ACCEPT_FIRST_HIT_AND_END_SEARCH here?

alichraghi · 2025-01-25T00:24:08Z

71_RayTracingPipeline/app_resources/common.hlsl

+    float32_t3 direction;
+    float32_t3 position;
+    float32_t outerCutoff;
+    uint16_t type;


Don't use 16-bit types in PushConstants. they require a feature to be enabled and no AMD GPU supports them afaik

Signed-off-by: kevyuu <[email protected]>

kevyuu force-pushed the raytracing_pipeline_demo branch from 9e2eb0f to bcbd729 Compare January 22, 2025 17:07

devshgraphicsprogramming reviewed Jan 22, 2025

View reviewed changes

kevyuu force-pushed the raytracing_pipeline_demo branch from 8f16aad to 34a3fa9 Compare January 24, 2025 16:44

Fletterio and others added 14 commits January 24, 2025 23:44

Update Bloom example, removed memory barriers on FFT

881e40b

Signed-off-by: kevyuu <[email protected]>

Implement Ray Tracing Demo

444c917

- Multiple HitGroup. Each with closesthit and anythit shader - Multiple Miss Shader Group. Signed-off-by: kevyuu <[email protected]>

Add ImGui Overlay

c991e20

Signed-off-by: kevyuu <[email protected]>

Implement multiple light type

6ac8f88

Signed-off-by: kevyuu <[email protected]>

Implement callable shader

9091da1

Signed-off-by: kevyuu <[email protected]>

Implement procedural geometry intersection shader demo

d303356

Signed-off-by: kevyuu <[email protected]>

Add Readme

f261f7c

Signed-off-by: kevyuu <[email protected]>

Fix alignment issue of scratch buffer

7b5059c

Signed-off-by: kevyuu <[email protected]>

Fix alignment issue of scratch buffer

8cbb0dc

Signed-off-by: kevyuu <[email protected]>

Fix alignment issue of scratch buffer

a9d5f8b

Signed-off-by: kevyuu <[email protected]>

Reduce Push Constant Size and skip compacting procedural geometries

feedf65

Signed-off-by: kevyuu <[email protected]>

Fix occlusion of procedural geometries

e369368

Signed-off-by: kevyuu <[email protected]>

71: cache shaders

c2c82d4

Signed-off-by: Ali Cheraghi <[email protected]> Signed-off-by: kevyuu <[email protected]>

Remove unnecesary log

34a3fa9

Signed-off-by: kevyuu <[email protected]>

devshgraphicsprogramming reviewed Jan 24, 2025

View reviewed changes

alichraghi reviewed Jan 25, 2025

View reviewed changes

kevyuu added 4 commits January 25, 2025 16:22

Change recursion depth to 1 so the demo can be run on more devices.

ec88268

Signed-off-by: kevyuu <[email protected]>

Temporarily remove procedural geometries to debug crash on amd cards.

d21c22c

Signed-off-by: kevyuu <[email protected]>

Use Nabla AABB type instead creating another aabb type

bc093c7

Signed-off-by: kevyuu <[email protected]>

Fix query compact blas size issue

1dc682a

Signed-off-by: kevyuu <[email protected]>

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Raytracing pipeline demo #173

Raytracing pipeline demo #173

kevyuu commented Jan 22, 2025

devshgraphicsprogramming Jan 22, 2025

kevyuu Jan 23, 2025

devshgraphicsprogramming Jan 24, 2025

devshgraphicsprogramming left a comment

devshgraphicsprogramming Jan 24, 2025

devshgraphicsprogramming Jan 24, 2025

devshgraphicsprogramming Jan 24, 2025

devshgraphicsprogramming Jan 24, 2025

devshgraphicsprogramming Jan 24, 2025

devshgraphicsprogramming Jan 24, 2025

devshgraphicsprogramming Jan 24, 2025

devshgraphicsprogramming Jan 24, 2025

devshgraphicsprogramming Jan 24, 2025

devshgraphicsprogramming Jan 24, 2025

alichraghi Jan 25, 2025

Raytracing pipeline demo #173

Are you sure you want to change the base?

Raytracing pipeline demo #173

Conversation

kevyuu commented Jan 22, 2025

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

devshgraphicsprogramming left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment