[SM6.10] VEC9 TriangleObjectPositions / CHECK-pass validation tests #7831

simoll · 2025-10-22T21:30:09Z

DXIL validation tests showing correct uses pass

SM6.10 tracking bug: #7824

V-FEXrt · 2025-10-24T01:13:55Z

Hey Simon!

We've actually been talking about DXIL op stuff internally that this PR is going to fall under. Specially we want a better system to mark ops as experimental before they are upgraded to stable!

I have a proposal up (microsoft/hlsl-specs#698) but due to scheduling conflicts we aren't going to be able to make a final decision until ~Nov 4th.

Since the decision there has impacts on this PR we are holding off on the review for just a little bit. I wanted to give you a heads up so you weren't surprised by the silence from our side

V-FEXrt · 2025-11-12T22:31:45Z

microsoft/hlsl-specs#729

The proposal is up here, once the implementation lands the ops here will need to be renumbered. I suspect the implementation will land quite quickly

V-FEXrt · 2025-11-14T00:11:12Z

Spec has been merged! I'm on vacation the next few weeks but @tex3d should be uploading the infrastructure change soon. Then this can be rebased on that

simoll · 2025-12-01T15:16:49Z

Spec has been merged! I'm on vacation the next few weeks but @tex3d should be uploading the infrastructure change soon. Then this can be rebased on that

@tex3d What's the status of the experimental opcode infra change to unblock this PR?

tex3d · 2025-12-01T23:32:37Z

@tex3d What's the status of the experimental opcode infra change to unblock this PR?

I'm trying to get a second review on this: #7947. Once merged, you can update the branch and move the intrinsic definitions in hctdb.py into experimental and update any references to the opcode numbers that changed.

tex3d

Now that #7947 has merged, you'll need to rebase and move things to experimental ops according to the comments I made.

utils/hct/hctdb.py

tex3d · 2025-12-08T21:01:08Z

I believe we'll also want to reserve high-level intrinsic IDs for these ASAP. It might be good to do this in the same PR, but a separate PR for that would be ok too.

That means updates to gen_intrin_main.txt, generated files, as well as manually adding EmptyLower entries for gLowerTable in HLOperationLower.cpp for each new high-level intrinsic operation.

This could serve as a good example for how to reserve HLSL intrinsics and DXIL opcodes for a feature before it's implemented, using the new experimental DXIL op table.

tex3d · 2025-12-10T03:41:37Z

I've created a PR to reserve HL and DXIL ops for these operations here: #7995.

So, you should be able to re-apply just the changes necessary on top of main when that PR merges.

simoll · 2025-12-10T06:10:04Z

Thanks @tex3d. I'll rebase when #7995 lands

SM6.10 tracking bug: microsoft#7824

tex3d · 2025-12-18T19:01:56Z

tools/clang/test/DXILValidation/clusterid_passing.ll

+
+; Function Attrs: nounwind
+define void @"\01?test_closesthit@@YAXUPayload@@UBuiltInTriangleIntersectionAttributes@@@Z"(%struct.Payload* noalias nocapture %payload, %struct.BuiltInTriangleIntersectionAttributes* nocapture readnone %attr) #0 {
+  %1 = call i32 @dx.op.clusterID(i32 2147483651)  ; ClusterID()


When this is round-tripped through llvm, i32 2147483651 will become i32 -2147483645. In lowering tests, for example, CHECK lines will need to use i32 -2147483645 instead of the unsigned value.

Using the unsigned value here could create confusion and additional difficulty for updating tests when moving ops to final DXIL because one will need to use two search/replace expressions to update each opcode, in case the unsigned form is used somewhere.

For this reason, I recommend considering the signed i32 form the canonical one to use in all cases for consistency.

Suggested change

%1 = call i32 @dx.op.clusterID(i32 2147483651) ; ClusterID()

%1 = call i32 @dx.op.clusterID(i32 -2147483645) ; ClusterID()

tex3d

I think this looks good, with just one concern:

The signed form of the i32 opcode is the one llvm will print and the one that will be needed in CHECK lines of tests. I think it would be best to stick to the signed form for input IR as well, so that each opcode doesn't have two textural forms to search/replace when updating ops in tests for released versions.

tex3d

I added suggested change comments to make updating the opcodes to canonical form easier.

tex3d · 2025-12-18T19:28:59Z

tools/clang/test/DXILValidation/clusterid_passing.ll

+
+; Function Attrs: nounwind
+define void @"\01?test_closesthit@@YAXUPayload@@UBuiltInTriangleIntersectionAttributes@@@Z"(%struct.Payload* noalias nocapture %payload, %struct.BuiltInTriangleIntersectionAttributes* nocapture readnone %attr) #0 {
+  %1 = call i32 @dx.op.clusterID(i32 2147483651)  ; ClusterID()


Suggested change

%1 = call i32 @dx.op.clusterID(i32 2147483651) ; ClusterID()

%1 = call i32 @dx.op.clusterID(i32 -2147483645) ; ClusterID()

tex3d · 2025-12-18T19:29:38Z

tools/clang/test/DXILValidation/clusterid_passing.ll

+  %1 = call %dx.types.HitObject @dx.op.hitObject_MakeNop(i32 266)  ; HitObject_MakeNop()
+  %2 = call i32 @dx.op.hitObject_StateScalar.i32(i32 2147483654, %dx.types.HitObject %1)  ; HitObject_ClusterID(hitObject)
+  %3 = call i32 @dx.op.allocateRayQuery(i32 178, i32 0)  ; AllocateRayQuery(constRayFlags)
+  %4 = call i32 @dx.op.rayQuery_StateScalar.i32(i32 2147483652, i32 %3)  ; RayQuery_CandidateClusterID(rayQueryHandle)
+  %5 = call i32 @dx.op.rayQuery_StateScalar.i32(i32 2147483653, i32 %3)  ; RayQuery_CommittedClusterID(rayQueryHandle)


Suggested change

%1 = call %dx.types.HitObject @dx.op.hitObject_MakeNop(i32 266) ; HitObject_MakeNop()

%2 = call i32 @dx.op.hitObject_StateScalar.i32(i32 2147483654, %dx.types.HitObject %1) ; HitObject_ClusterID(hitObject)

%3 = call i32 @dx.op.allocateRayQuery(i32 178, i32 0) ; AllocateRayQuery(constRayFlags)

%4 = call i32 @dx.op.rayQuery_StateScalar.i32(i32 2147483652, i32 %3) ; RayQuery_CandidateClusterID(rayQueryHandle)

%5 = call i32 @dx.op.rayQuery_StateScalar.i32(i32 2147483653, i32 %3) ; RayQuery_CommittedClusterID(rayQueryHandle)

%1 = call %dx.types.HitObject @dx.op.hitObject_MakeNop(i32 266) ; HitObject_MakeNop()

%2 = call i32 @dx.op.hitObject_StateScalar.i32(i32 -2147483642, %dx.types.HitObject %1) ; HitObject_ClusterID(hitObject)

%3 = call i32 @dx.op.allocateRayQuery(i32 178, i32 0) ; AllocateRayQuery(constRayFlags)

%4 = call i32 @dx.op.rayQuery_StateScalar.i32(i32 -2147483644, i32 %3) ; RayQuery_CandidateClusterID(rayQueryHandle)

%5 = call i32 @dx.op.rayQuery_StateScalar.i32(i32 -2147483643, i32 %3) ; RayQuery_CommittedClusterID(rayQueryHandle)

tex3d · 2025-12-18T19:30:01Z

tools/clang/test/DXILValidation/triangle_position_passing.ll

+
+; Function Attrs: nounwind
+define void @"\01?test_closesthit@@YAXUPayload@@UBuiltInTriangleIntersectionAttributes@@@Z"(%struct.Payload* noalias nocapture %payload, %struct.BuiltInTriangleIntersectionAttributes* nocapture readnone %attr) #0 {
+  %1 = call <9 x float> @dx.op.triangleObjectPosition.f32(i32 2147483655)  ; TriangleObjectPosition()


Suggested change

%1 = call <9 x float> @dx.op.triangleObjectPosition.f32(i32 2147483655) ; TriangleObjectPosition()

%1 = call <9 x float> @dx.op.triangleObjectPosition.f32(i32 -2147483641) ; TriangleObjectPosition()

tex3d · 2025-12-18T19:30:30Z

tools/clang/test/DXILValidation/triangle_position_passing.ll

+  %1 = call %dx.types.HitObject @dx.op.hitObject_MakeNop(i32 266)  ; HitObject_MakeNop()
+  %2 = call <9 x float> @dx.op.hitObject_TriangleObjectPosition.f32(i32 2147483658, %dx.types.HitObject %1)  ; HitObject_TriangleObjectPosition(hitObject)
+  %3 = call i32 @dx.op.allocateRayQuery(i32 178, i32 0)  ; AllocateRayQuery(constRayFlags)
+  %4 = call <9 x float> @dx.op.rayQuery_CandidateTriangleObjectPosition.f32(i32 2147483656, i32 %3)  ; RayQuery_CandidateTriangleObjectPosition(rayQueryHandle)
+  %5 = call <9 x float> @dx.op.rayQuery_CommittedTriangleObjectPosition.f32(i32 2147483657, i32 %3)  ; RayQuery_CommittedTriangleObjectPosition(rayQueryHandle)


Suggested change

%1 = call %dx.types.HitObject @dx.op.hitObject_MakeNop(i32 266) ; HitObject_MakeNop()

%2 = call <9 x float> @dx.op.hitObject_TriangleObjectPosition.f32(i32 2147483658, %dx.types.HitObject %1) ; HitObject_TriangleObjectPosition(hitObject)

%3 = call i32 @dx.op.allocateRayQuery(i32 178, i32 0) ; AllocateRayQuery(constRayFlags)

%4 = call <9 x float> @dx.op.rayQuery_CandidateTriangleObjectPosition.f32(i32 2147483656, i32 %3) ; RayQuery_CandidateTriangleObjectPosition(rayQueryHandle)

%5 = call <9 x float> @dx.op.rayQuery_CommittedTriangleObjectPosition.f32(i32 2147483657, i32 %3) ; RayQuery_CommittedTriangleObjectPosition(rayQueryHandle)

%1 = call %dx.types.HitObject @dx.op.hitObject_MakeNop(i32 266) ; HitObject_MakeNop()

%2 = call <9 x float> @dx.op.hitObject_TriangleObjectPosition.f32(i32 -2147483638, %dx.types.HitObject %1) ; HitObject_TriangleObjectPosition(hitObject)

%3 = call i32 @dx.op.allocateRayQuery(i32 178, i32 0) ; AllocateRayQuery(constRayFlags)

%4 = call <9 x float> @dx.op.rayQuery_CandidateTriangleObjectPosition.f32(i32 -2147483640, i32 %3) ; RayQuery_CandidateTriangleObjectPosition(rayQueryHandle)

%5 = call <9 x float> @dx.op.rayQuery_CommittedTriangleObjectPosition.f32(i32 -2147483639, i32 %3) ; RayQuery_CommittedTriangleObjectPosition(rayQueryHandle)

github-project-automation bot added this to HLSL Roadmap Oct 22, 2025

github-project-automation bot moved this to New in HLSL Roadmap Oct 22, 2025

simoll mentioned this pull request Oct 22, 2025

SM6.10 implementation tracker: ClusterID and TriangleObjectPositions #7824

Open

5 tasks

tex3d reviewed Dec 8, 2025

View reviewed changes

utils/hct/hctdb.py Show resolved Hide resolved

utils/hct/hctdb.py Outdated Show resolved Hide resolved

[SM6.10] VEC9 TriangleObjectPositions / passing validation tests

63a5b8b

SM6.10 tracking bug: microsoft#7824

simoll changed the title ~~[SM6.10] DXIL opcodes for ClusterID and TriangleObjectPositions~~ [SM6.10] VEC9 TriangleObjectPositions / CHECK-pass validation tests Dec 15, 2025

simoll force-pushed the sm610_dxil_opcodes branch from b3a793f to 63a5b8b Compare December 15, 2025 05:59

simoll requested a review from tex3d December 15, 2025 06:01

tex3d reviewed Dec 18, 2025

View reviewed changes

	%1 = call i32 @dx.op.clusterID(i32 2147483651) ; ClusterID()
	%1 = call i32 @dx.op.clusterID(i32 -2147483645) ; ClusterID()

	%1 = call <9 x float> @dx.op.triangleObjectPosition.f32(i32 2147483655) ; TriangleObjectPosition()
	%1 = call <9 x float> @dx.op.triangleObjectPosition.f32(i32 -2147483641) ; TriangleObjectPosition()

[SM6.10] VEC9 TriangleObjectPositions / CHECK-pass validation tests #7831

Are you sure you want to change the base?

[SM6.10] VEC9 TriangleObjectPositions / CHECK-pass validation tests #7831

Conversation

simoll commented Oct 22, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

V-FEXrt commented Oct 24, 2025

Uh oh!

V-FEXrt commented Nov 12, 2025

Uh oh!

V-FEXrt commented Nov 14, 2025

Uh oh!

simoll commented Dec 1, 2025

Uh oh!

tex3d commented Dec 1, 2025

Uh oh!

tex3d left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

tex3d commented Dec 8, 2025

Uh oh!

tex3d commented Dec 10, 2025

Uh oh!

simoll commented Dec 10, 2025

Uh oh!

tex3d Dec 18, 2025

Choose a reason for hiding this comment

Uh oh!

tex3d Dec 18, 2025

Choose a reason for hiding this comment

Uh oh!

tex3d left a comment

Choose a reason for hiding this comment

Uh oh!

tex3d left a comment

Choose a reason for hiding this comment

Uh oh!

tex3d Dec 18, 2025

Choose a reason for hiding this comment

Uh oh!

tex3d Dec 18, 2025

Choose a reason for hiding this comment

Uh oh!

tex3d Dec 18, 2025

Choose a reason for hiding this comment

Uh oh!

tex3d Dec 18, 2025

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

simoll commented Oct 22, 2025 •

edited

Loading