diff --git a/test/Feature/RT/callable-shader.test b/test/Feature/RT/callable-shader.test
new file mode 100644
index 000000000..91b289576
--- /dev/null
+++ b/test/Feature/RT/callable-shader.test
@@ -0,0 +1,87 @@
+#--- source.hlsl
+
+// Callable shaders receive an inout user-defined parameter — Vulkan / DXR
+// don't have a Built-in convention beyond "size matches what the caller
+// declared". One uint of payload is enough to confirm routing.
+struct CallableParams {
+  uint Value;
+};
+
+[[vk::binding(0, 0)]] RWStructuredBuffer<uint> Output : register(u0);
+
+[shader("raygeneration")]
+void RayGen() {
+  // Each lane calls a different callable shader index. CallShader picks
+  // the callable record at the given index in the SBT's Callable region
+  // — inline RT has no equivalent of this stage.
+  const uint Idx = DispatchRaysIndex().x;
+  CallableParams P;
+  P.Value = 0;
+  CallShader(Idx, P);
+  Output[Idx] = P.Value;
+}
+
+[shader("callable")]
+void CallableA(inout CallableParams P) {
+  P.Value = 0xAAAA;
+}
+
+[shader("callable")]
+void CallableB(inout CallableParams P) {
+  P.Value = 0xBBBB;
+}
+//--- pipeline.yaml
+---
+Shaders:
+  - Stage: RayGeneration
+    Entry: RayGen
+  - Stage: Callable
+    Entry: CallableA
+  - Stage: Callable
+    Entry: CallableB
+Buffers:
+  - Name: Output
+    Format: UInt32
+    Stride: 4
+    FillSize: 8
+  - Name: Expected
+    Format: UInt32
+    Stride: 4
+    Data: [ 0xAAAA, 0xBBBB ]
+RayTracingPipelineConfig:
+  # No TraceRay calls — depth 1 is sufficient (CallShader doesn't count
+  # against ray recursion).
+  MaxTraceRecursionDepth: 1
+  MaxPayloadSizeInBytes: 4
+ShaderBindingTable:
+  RayGen:
+    ShaderName: RayGen
+  Callable:
+    - ShaderName: CallableA
+    - ShaderName: CallableB
+DescriptorSets:
+  - Resources:
+    - Name: Output
+      Kind: RWStructuredBuffer
+      DirectXBinding:
+        Register: 0
+        Space: 0
+      VulkanBinding:
+        Binding: 0
+DispatchParameters:
+  DispatchGroupCount: [ 2, 1, 1 ]
+Results:
+  - Result: CallableRouting
+    Rule: BufferExact
+    Actual: Output
+    Expected: Expected
+...
+#--- end
+
+# REQUIRES: raytracing-pipeline
+# Unimplemented https://github.com/llvm/offload-test-suite/issues/1268
+# XFAIL: Clang
+
+# RUN: split-file %s %t
+# RUN: %dxc_target_lib -T lib_6_5 -Fo %t.o %t/source.hlsl
+# RUN: %offloader %t/pipeline.yaml %t.o
diff --git a/test/Feature/RT/ray-flag-skip-closest-hit.test b/test/Feature/RT/ray-flag-skip-closest-hit.test
new file mode 100644
index 000000000..db7ec4683
--- /dev/null
+++ b/test/Feature/RT/ray-flag-skip-closest-hit.test
@@ -0,0 +1,127 @@
+#--- source.hlsl
+
+struct Payload {
+  uint Value;
+};
+
+[[vk::binding(0, 0)]] RaytracingAccelerationStructure Scene : register(t0);
+[[vk::binding(1, 0)]] RWStructuredBuffer<uint> Output : register(u0);
+
+[shader("raygeneration")]
+void RayGen() {
+  // Both lanes fire identical rays that hit the same triangle. Lane 0
+  // uses RAY_FLAG_NONE so the closest-hit shader runs and writes
+  // 0xBEEF into the payload. Lane 1 sets
+  // RAY_FLAG_SKIP_CLOSEST_HIT_SHADER (a PSO-only flag — inline RT can't
+  // express this) so the traversal records a hit but the closest-hit
+  // shader is never invoked. With initial payload = 0xAAAA the output
+  // is the sentinel for lane 1 and 0xBEEF for lane 0.
+  const uint Idx = DispatchRaysIndex().x;
+  const uint Flags =
+      Idx == 0 ? RAY_FLAG_NONE : RAY_FLAG_SKIP_CLOSEST_HIT_SHADER;
+
+  Payload P;
+  P.Value = 0xAAAA;
+  RayDesc Ray;
+  Ray.Origin = float3(0, 0, 1);
+  Ray.Direction = float3(0, 0, -1);
+  Ray.TMin = 0.0;
+  Ray.TMax = 100.0;
+  TraceRay(Scene, Flags, 0xFF, 0, 1, 0, Ray, P);
+  Output[Idx] = P.Value;
+}
+
+[shader("miss")]
+void MissMain(inout Payload P) {
+  // Sentinel — every ray hits the triangle, so this never runs.
+  P.Value = 0xDEAD;
+}
+
+[shader("closesthit")]
+void ClosestHitMain(inout Payload P,
+                   in BuiltInTriangleIntersectionAttributes Attr) {
+  P.Value = 0xBEEF;
+}
+//--- pipeline.yaml
+---
+Shaders:
+  - Stage: RayGeneration
+    Entry: RayGen
+  - Stage: Miss
+    Entry: MissMain
+  - Stage: ClosestHit
+    Entry: ClosestHitMain
+Buffers:
+  - Name: Vertices
+    Format: Float32
+    Stride: 12
+    Data: [ 0.0, 1.0, 0.0, -1.0, -1.0, 0.0, 1.0, -1.0, 0.0 ]
+  - Name: Output
+    Format: UInt32
+    Stride: 4
+    FillSize: 8
+  - Name: Expected
+    Format: UInt32
+    Stride: 4
+    # Lane 0: closest-hit runs, writes 0xBEEF.
+    # Lane 1: closest-hit skipped, payload stays at the initial 0xAAAA.
+    Data: [ 0xBEEF, 0xAAAA ]
+AccelerationStructures:
+  BLAS:
+    - Name: TriangleBLAS
+      Triangles:
+        - VertexBuffer: Vertices
+          VertexFormat: RGB32Float
+          VertexStride: 12
+          VertexCount: 3
+  TLAS:
+    - Name: Scene
+      Instances:
+        - BLAS: TriangleBLAS
+RayTracingPipelineConfig:
+  MaxTraceRecursionDepth: 1
+  MaxPayloadSizeInBytes: 4
+HitGroups:
+  - Name: TriangleHitGroup
+    Type: Triangles
+    ClosestHit: ClosestHitMain
+ShaderBindingTable:
+  RayGen:
+    ShaderName: RayGen
+  Miss:
+    - ShaderName: MissMain
+  HitGroup:
+    - ShaderName: TriangleHitGroup
+DescriptorSets:
+  - Resources:
+    - Name: Scene
+      Kind: AccelerationStructure
+      DirectXBinding:
+        Register: 0
+        Space: 0
+      VulkanBinding:
+        Binding: 0
+    - Name: Output
+      Kind: RWStructuredBuffer
+      DirectXBinding:
+        Register: 0
+        Space: 0
+      VulkanBinding:
+        Binding: 1
+DispatchParameters:
+  DispatchGroupCount: [ 2, 1, 1 ]
+Results:
+  - Result: RayFlagSkipClosestHit
+    Rule: BufferExact
+    Actual: Output
+    Expected: Expected
+...
+#--- end
+
+# REQUIRES: raytracing-pipeline
+# Unimplemented https://github.com/llvm/offload-test-suite/issues/1268
+# XFAIL: Clang
+
+# RUN: split-file %s %t
+# RUN: %dxc_target_lib -T lib_6_5 -Fo %t.o %t/source.hlsl
+# RUN: %offloader %t/pipeline.yaml %t.o
diff --git a/test/Feature/RT/trace-ray-recursion.test b/test/Feature/RT/trace-ray-recursion.test
new file mode 100644
index 000000000..98d509ba3
--- /dev/null
+++ b/test/Feature/RT/trace-ray-recursion.test
@@ -0,0 +1,134 @@
+#--- source.hlsl
+
+struct Payload {
+  uint Value;
+};
+
+[[vk::binding(0, 0)]] RaytracingAccelerationStructure Scene : register(t0);
+[[vk::binding(1, 0)]] RWStructuredBuffer<uint> Output : register(u0);
+
+[shader("raygeneration")]
+void RayGen() {
+  Payload P;
+  P.Value = 0;
+  RayDesc Ray;
+  Ray.Origin = float3(0, 0, 1);
+  Ray.Direction = float3(0, 0, -1);
+  Ray.TMin = 0.0;
+  Ray.TMax = 100.0;
+  TraceRay(Scene, RAY_FLAG_NONE, 0xFF, 0, 1, 0, Ray, P);
+  Output[0] = P.Value;
+}
+
+[shader("miss")]
+void MissMain(inout Payload P) {
+  // Sentinel — never observed (every ray hits the triangle).
+  P.Value = 0xDEAD;
+}
+
+[shader("closesthit")]
+void RecursiveCH(inout Payload P, in BuiltInTriangleIntersectionAttributes Attr) {
+  // First entry into the hit shader: payload still carries raygen's 0.
+  // Bump it to a marker, fire a secondary ray from a fresh origin above the
+  // triangle so we hit the same triangle again, then OR in the high marker
+  // when the secondary chain unwinds. The secondary invocation sees a
+  // non-zero payload and goes down the second-level branch.
+  if (P.Value == 0) {
+    P.Value = 0x1;
+    RayDesc Ray;
+    Ray.Origin = float3(0, 0, 2);
+    Ray.Direction = float3(0, 0, -1);
+    Ray.TMin = 0.0;
+    Ray.TMax = 100.0;
+    TraceRay(Scene, RAY_FLAG_NONE, 0xFF, 0, 1, 0, Ray, P);
+    P.Value |= 0x100;
+  } else {
+    // Second-level hit: write the low marker and don't recurse further.
+    P.Value = 0x10;
+  }
+}
+//--- pipeline.yaml
+---
+Shaders:
+  - Stage: RayGeneration
+    Entry: RayGen
+  - Stage: Miss
+    Entry: MissMain
+  - Stage: ClosestHit
+    Entry: RecursiveCH
+Buffers:
+  - Name: Vertices
+    Format: Float32
+    Stride: 12
+    Data: [ 0.0, 1.0, 0.0, -1.0, -1.0, 0.0, 1.0, -1.0, 0.0 ]
+  - Name: Output
+    Format: UInt32
+    Stride: 4
+    FillSize: 4
+  - Name: Expected
+    Format: UInt32
+    Stride: 4
+    # First-level CH sets Value=0x1 (overwritten by secondary CH to 0x10), then
+    # OR's in 0x100 after the secondary chain unwinds — final = 0x110.
+    Data: [ 0x110 ]
+AccelerationStructures:
+  BLAS:
+    - Name: TriangleBLAS
+      Triangles:
+        - VertexBuffer: Vertices
+          VertexFormat: RGB32Float
+          VertexStride: 12
+          VertexCount: 3
+  TLAS:
+    - Name: Scene
+      Instances:
+        - BLAS: TriangleBLAS
+RayTracingPipelineConfig:
+  # Raygen invokes TraceRay (depth 1) and the first-level CH invokes
+  # TraceRay again (depth 2). Both need to be allowed.
+  MaxTraceRecursionDepth: 2
+  MaxPayloadSizeInBytes: 4
+HitGroups:
+  - Name: TriangleHitGroup
+    Type: Triangles
+    ClosestHit: RecursiveCH
+ShaderBindingTable:
+  RayGen:
+    ShaderName: RayGen
+  Miss:
+    - ShaderName: MissMain
+  HitGroup:
+    - ShaderName: TriangleHitGroup
+DescriptorSets:
+  - Resources:
+    - Name: Scene
+      Kind: AccelerationStructure
+      DirectXBinding:
+        Register: 0
+        Space: 0
+      VulkanBinding:
+        Binding: 0
+    - Name: Output
+      Kind: RWStructuredBuffer
+      DirectXBinding:
+        Register: 0
+        Space: 0
+      VulkanBinding:
+        Binding: 1
+DispatchParameters:
+  DispatchGroupCount: [ 1, 1, 1 ]
+Results:
+  - Result: TraceRayRecursion
+    Rule: BufferExact
+    Actual: Output
+    Expected: Expected
+...
+#--- end
+
+# REQUIRES: raytracing-pipeline
+# Unimplemented https://github.com/llvm/offload-test-suite/issues/1268
+# XFAIL: Clang
+
+# RUN: split-file %s %t
+# RUN: %dxc_target_lib -T lib_6_5 -Fo %t.o %t/source.hlsl
+# RUN: %offloader %t/pipeline.yaml %t.o