[SER] Basic execution tests #7379

simoll · 2025-04-25T16:03:27Z

All trivial scalar/vector/matrix getters
HitObject::FromRayQuery with procedural hit
HitObject::GetAttributes with custom attributes and procedural hit

SER implementation tracker: #7214

- All trivial scalar/vector/matrix getters - HitObject::FromRayQuery with procedural hit - HitObject::GetAttributes<T> with custom attributes and procedural hit SER implementation tracker: microsoft#7214

damyanp

There's a lot of code here to review and I've not had a chance to go through it all in detail.

However, I'll note that before this can go into main we'll need to get it updated to follow the LLVM coding standards.

For the short term maybe this should go into the staging branch?

tools/clang/unittests/HLSLExec/DXRUtil.h

tools/clang/unittests/HLSLExec/ExecutionTest.cpp

damyanp · 2025-04-25T19:33:30Z

tools/clang/unittests/HLSLExec/ExecutionTest.cpp

@@ -2078,6 +2095,707 @@ void ExecutionTest::RunRWByteBufferComputeTest(ID3D12Device *pDevice,
  WaitForSignal(pCommandQueue, FO);
 }

+CComPtr<ID3D12Resource> ExecutionTest::RunDXRTest(


Any reason this all has to be in this file?

clandrew · 2025-04-25T20:34:52Z

Thanks! Taking a look at this now. It might be that we can simply ingest the test code based on this staging branch and it does not necessarily have to be merged.

simoll · 2025-04-29T11:37:54Z

There's a lot of code here to review and I've not had a chance to go through it all in detail.

However, I'll note that before this can go into main we'll need to get it updated to follow the LLVM coding standards.

For the short term maybe this should go into the staging branch?

Should close this PR and re-open against staging-sm6.9?

…ocal constants / simplifications

simoll · 2025-04-30T05:11:34Z

There are merge conflicts with the sm69 staging branch just because this was originally based on main. Will reopen as a new pR.

… simplified tests/ added SERInvokeNoSBTTest

simoll · 2025-04-30T10:59:07Z

Decided to reopen and updated to make sure we can keep reviewing this while main gets merged into staging.

Reworked variable names to follow LLVM Coding Standards
Removed some redundant code with a new CreateDXRDevice function
Added another test (Invoke called on a HitObject-from-RQ that has no SBT set)

…atch

- Array of 8 HitObjects, each with different ray flags - Samples at 'random' positions to block SROA from breaking down the array

(procedural) - Support for procedural geometry and triangles at the same time - fix: make IS non-optional - Made payload/attributeCount in RunDXRTest non-defaulting - Use a circle for procedural geoemtry and make sure it fits into the AABB

jenatali · 2025-05-07T23:38:24Z

tools/clang/unittests/HLSLExec/ExecutionTest_SER.h

+void anyhit(inout PerRayData payload, in Attrs attrs)
+{
+    payload.visited |= 2U;
+    AcceptHitAndEndSearch();


There is a test bug here:

The test attempts to very that threads fall into 3 buckets of work:

Threads that hit nothing (payload ends up as 0)

Threads that hit the AABB, which doesn't commit the hit, and that's all (payload ends up as 8)

Threads that hit the AABB and the triangle (payload ends up as 2 | 8 | 16 = 26)

For that last bucket, there's no requirement on the ordering between the hits of the AABB and the triangle. If the triangle is hit first, then this function call causes the AABB to not be hit, and therefore the 8 is missing from the final payload (value 18). WARP hits this case 100% of the time.

However, even without this, if the triangle is hit first and is in front of the procedural primitive (which it apparently is, half of the time, according to what I'm seeing with WARP), then accepting the triangle will cause the procedural primitive or its any-hit to be skipped.

I changed the test to no longer rely on anyhit/intersections. Instead, we hit or miss the triangle (aabbs ignored by ray flag), then mutate to an aabb hit. Only testing for correct invocation of the miss/closesthit/chAABB shaders

tools/clang/unittests/HLSLExec/DXRUtil.h

jenatali · 2025-05-08T03:25:41Z

tools/clang/unittests/HLSLExec/ExecutionTest_SER.h

+    if (relHitPos.y >= 0.0f)
+        hitKind = 1;
+    hitKind *= 2;
+    if (relHitPos.x >= 0.0f)
+        hitKind += 1;
+    hitKind *= 2;
+    if (relHitPos.z >= 0.0f)
+        hitKind += 1;
+
+    CustomAttrs attrs;
+    attrs.dist = length(relHitPos);


It seems something in here is too dependent on floating point precision. If I allow FMA fusing in WARP (which literally just was hooked up a few weeks ago) then the test passes, but if I disable it then I get a bucketing which is off-by-one from the test expectations.

Made the hitkind more robust by replacing the fp comparison and with launchIndex-based comparisons to generate a meaningful hitKind. Please test on your end, there may still be fp precision issues with the aabb intersection logic itself.

jenatali · 2025-05-08T03:48:57Z

tools/clang/unittests/HLSLExec/ExecutionTest_SER.h

+[shader("anyhit")]
+void anyhit(inout PerRayData payload, in Attrs attrs)
+{
+    AcceptHitAndEndSearch();


This test appears to suffer from the same issue I described for SERShaderTableIndexTest, where the order of triangle vs procedural hits in the acceleration structure determines the overall bucketing.

Hit order issue should be resolved. Keeping this open until confirmed.

… arith for hitKind computation

…/tri hit order

github-actions · 2025-05-08T08:52:38Z

✅ With the latest revision this PR passed the C/C++ code formatter.

…atch

nadaOuf · 2025-05-06T19:43:36Z

tools/clang/unittests/HLSLExec/ExecutionTest.cpp

+  TEST_METHOD(SERInvokeNoSBTTest);
+  TEST_METHOD(SERMaybeReorderThreadTest)
+  TEST_METHOD(SERDynamicHitObjectArrayTest);
+  TEST_METHOD(SERWaveIncoherentHitTest);


I am reformatting these test to fit our d3d12 hlk test structures. @simoll since these tests don't need to run as part of dxilconf I am planning to reject this PR. Do you still need it open to push more changes? or these all the tests you are adding?

nadaOuf · 2025-05-07T18:25:45Z

tools/clang/unittests/HLSLExec/DXRUtil.h

+    D3D12_RESOURCE_DESC Desc = CD3DX12_RESOURCE_DESC::Buffer(
+        TotalSizeInBytes, D3D12_RESOURCE_FLAG_NONE,
+        std::max(D3D12_RAYTRACING_SHADER_RECORD_BYTE_ALIGNMENT,
+                 D3D12_DEFAULT_RESOURCE_PLACEMENT_ALIGNMENT));


why are we getting max here? those are known values and the default is always going to be bigger.

[SER] Basic execution tests

6cb6843

- All trivial scalar/vector/matrix getters - HitObject::FromRayQuery with procedural hit - HitObject::GetAttributes<T> with custom attributes and procedural hit SER implementation tracker: microsoft#7214

github-project-automation bot added this to HLSL Roadmap Apr 25, 2025

github-project-automation bot moved this to New in HLSL Roadmap Apr 25, 2025

simoll mentioned this pull request Apr 24, 2025

SER implementation tracker #7214

Open

41 tasks

tex3d added this to HLSL Support Apr 25, 2025

tex3d self-assigned this Apr 25, 2025

damyanp moved this to Active in HLSL Support Apr 25, 2025

damyanp requested changes Apr 25, 2025

View reviewed changes

github-project-automation bot moved this from New to In progress in HLSL Roadmap Apr 25, 2025

damyanp requested review from nadaOuf and clandrew April 25, 2025 20:10

Add SetShaderTableIndex+LoadLocalRootConstant tests / host code for l…

6bcb151

…ocal constants / simplifications

simoll changed the base branch from main to staging-sm6.9 April 30, 2025 05:10

simoll closed this Apr 30, 2025

github-project-automation bot moved this from Active to Closed in HLSL Support Apr 30, 2025

github-project-automation bot moved this from In progress to Done in HLSL Roadmap Apr 30, 2025

simoll reopened this Apr 30, 2025

simoll changed the base branch from staging-sm6.9 to main April 30, 2025 10:49

Uppercased vars for coding standards / added CreateDXRDevice helper /…

144a083

… simplified tests/ added SERInvokeNoSBTTest

simoll changed the base branch from main to staging-sm6.9 April 30, 2025 14:28

simoll added 6 commits April 30, 2025 17:56

Merge remote-tracking branch 'msft/staging-sm6.9' into ser_exectest_p…

5a31d33

…atch

Test all MaybeReorderThread variants (in wave-incoherent execution)

c2e2dee

DynamicHitObjectArrayTest

6ab79fe

- Array of 8 HitObjects, each with different ray flags - Samples at 'random' positions to block SROA from breaking down the array

WaveIncoherentHitTest

8f1c598

SERReorderCoherentTest

e1df92b

jenatali reviewed May 7, 2025

View reviewed changes

jenatali reviewed May 8, 2025

View reviewed changes

tools/clang/unittests/HLSLExec/DXRUtil.h Outdated Show resolved Hide resolved

jenatali reviewed May 8, 2025

View reviewed changes

simoll added 4 commits May 8, 2025 10:09

Fixed SERShaderTableIndexTest (not relying on is/ah any longer)

c6fce45

Turn ShaderTable::Init into a ctor with initializer list

c3c399f

Fix GetAttributesTest: procedural not contained in AABB / use integer…

15140b1

… arith for hitKind computation

Fix SERWaveIncoherentHitTest: Use ray_flags to be independent of aabb…

6e42757

…/tri hit order

simoll added 2 commits May 8, 2025 10:54

nfc: formatting

0aee7ff

Merge remote-tracking branch 'msft/staging-sm6.9' into ser_exectest_p…

03e013a

…atch

damyanp moved this from Closed to Active in HLSL Support May 9, 2025

simoll marked this pull request as draft May 9, 2025 17:00

damyanp unassigned tex3d May 9, 2025

nadaOuf reviewed May 9, 2025

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[SER] Basic execution tests #7379

[SER] Basic execution tests #7379

simoll commented Apr 25, 2025

damyanp left a comment

damyanp Apr 25, 2025

clandrew commented Apr 25, 2025

simoll commented Apr 29, 2025

simoll commented Apr 30, 2025

simoll commented Apr 30, 2025

jenatali May 7, 2025

simoll May 8, 2025

jenatali May 8, 2025

simoll May 8, 2025

jenatali May 8, 2025

simoll May 8, 2025

github-actions bot commented May 8, 2025 •

edited

Loading

nadaOuf May 6, 2025

nadaOuf May 7, 2025

[SER] Basic execution tests #7379

Are you sure you want to change the base?

[SER] Basic execution tests #7379

Conversation

simoll commented Apr 25, 2025

damyanp left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

clandrew commented Apr 25, 2025

simoll commented Apr 29, 2025

simoll commented Apr 30, 2025

simoll commented Apr 30, 2025

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

github-actions bot commented May 8, 2025 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

github-actions bot commented May 8, 2025 •

edited

Loading