Uniformity analysis. (#3704)

* Uniformity analysis.

* Add [NonUniformReturn] decorations to some hlsl intrinsic functions.

20 files changed, 673 insertions, 9 deletions

diff --git a/build/visual-studio/slang/slang.vcxproj b/build/visual-studio/slang/slang.vcxproj
index 6314e660c..6ca03e63a 100644
--- a/build/visual-studio/slang/slang.vcxproj
+++ b/build/visual-studio/slang/slang.vcxproj
@@ -470,6 +470,7 @@ IF EXIST ..\..\..\external\slang-glslang\bin\windows-aarch64\release\slang-glsla
     <ClInclude Include="..\..\..\source\slang\slang-ir-strip.h" />

     <ClInclude Include="..\..\..\source\slang\slang-ir-synthesize-active-mask.h" />

     <ClInclude Include="..\..\..\source\slang\slang-ir-translate-glsl-global-var.h" />

+    <ClInclude Include="..\..\..\source\slang\slang-ir-uniformity.h" />

     <ClInclude Include="..\..\..\source\slang\slang-ir-use-uninitialized-out-param.h" />

     <ClInclude Include="..\..\..\source\slang\slang-ir-util.h" />

     <ClInclude Include="..\..\..\source\slang\slang-ir-validate.h" />

@@ -690,6 +691,7 @@ IF EXIST ..\..\..\external\slang-glslang\bin\windows-aarch64\release\slang-glsla
     <ClCompile Include="..\..\..\source\slang\slang-ir-strip.cpp" />

     <ClCompile Include="..\..\..\source\slang\slang-ir-synthesize-active-mask.cpp" />

     <ClCompile Include="..\..\..\source\slang\slang-ir-translate-glsl-global-var.cpp" />

+    <ClCompile Include="..\..\..\source\slang\slang-ir-uniformity.cpp" />

     <ClCompile Include="..\..\..\source\slang\slang-ir-use-uninitialized-out-param.cpp" />

     <ClCompile Include="..\..\..\source\slang\slang-ir-util.cpp" />

     <ClCompile Include="..\..\..\source\slang\slang-ir-validate.cpp" />

diff --git a/build/visual-studio/slang/slang.vcxproj.filters b/build/visual-studio/slang/slang.vcxproj.filters
index 9a029dcb7..eceb57b9f 100644
--- a/build/visual-studio/slang/slang.vcxproj.filters
+++ b/build/visual-studio/slang/slang.vcxproj.filters
@@ -498,6 +498,9 @@
     <ClInclude Include="..\..\..\source\slang\slang-ir-translate-glsl-global-var.h">

       <Filter>Header Files</Filter>

     </ClInclude>

+    <ClInclude Include="..\..\..\source\slang\slang-ir-uniformity.h">

+      <Filter>Header Files</Filter>

+    </ClInclude>

     <ClInclude Include="..\..\..\source\slang\slang-ir-use-uninitialized-out-param.h">

       <Filter>Header Files</Filter>

     </ClInclude>

@@ -1154,6 +1157,9 @@
     <ClCompile Include="..\..\..\source\slang\slang-ir-translate-glsl-global-var.cpp">

       <Filter>Source Files</Filter>

     </ClCompile>

+    <ClCompile Include="..\..\..\source\slang\slang-ir-uniformity.cpp">

+      <Filter>Source Files</Filter>

+    </ClCompile>

     <ClCompile Include="..\..\..\source\slang\slang-ir-use-uninitialized-out-param.cpp">

       <Filter>Source Files</Filter>

     </ClCompile>

diff --git a/slang.h b/slang.h
index b826707aa..1547ca552 100644
--- a/slang.h
+++ b/slang.h
@@ -924,6 +924,7 @@ extern "C"
             FileSystem,
             Heterogeneous,
             NoMangle,
+            ValidateUniformity,
             AllowGLSL,
 
             // Internal
diff --git a/source/core/slang-uint-set.cpp b/source/core/slang-uint-set.cpp
index 02e142706..e973cbc3a 100644
--- a/source/core/slang-uint-set.cpp
+++ b/source/core/slang-uint-set.cpp
@@ -27,7 +27,7 @@ UIntSet& UIntSet::operator=(const UIntSet& other)
     return *this;
 }
 
-HashCode UIntSet::getHashCode()
+HashCode UIntSet::getHashCode() const
 {
     int rs = 0;
     for (auto val : m_buffer)
diff --git a/source/core/slang-uint-set.h b/source/core/slang-uint-set.h
index 4912ae504..0f2165bab 100644
--- a/source/core/slang-uint-set.h
+++ b/source/core/slang-uint-set.h
@@ -26,7 +26,7 @@ public:
     UIntSet& operator=(UIntSet&& other);
     UIntSet& operator=(const UIntSet& other);
 
-    HashCode getHashCode();
+    HashCode getHashCode() const;
 
         /// Return the count of all bits directly represented
     Int getCount() const { return Int(m_buffer.getCount()) * kElementSize; }
diff --git a/source/slang/core.meta.slang b/source/slang/core.meta.slang
index 3131224b1..54a82c4f0 100644
--- a/source/slang/core.meta.slang
+++ b/source/slang/core.meta.slang
@@ -2308,7 +2308,8 @@ int __naturalStrideOf()
     return __naturalStrideOf_impl(__declVal<T>());
 }
 
-
+__intrinsic_op($(kIROp_TreatAsDynamicUniform))
+T asDynamicUniform<T>(T v);
 
 // Binding Attributes
 
@@ -2555,3 +2556,6 @@ attribute_syntax [PreferCheckpoint] : PreferCheckpointAttribute;
 
 __attributeTarget(DeclBase)
 attribute_syntax [KnownBuiltin(name : String)] : KnownBuiltinAttribute;
+
+__attributeTarget(FunctionDeclBase)
+attribute_syntax [NonUniformReturn] : NonDynamicUniformAttribute;
diff --git a/source/slang/hlsl.meta.slang b/source/slang/hlsl.meta.slang
index 5c97de525..f220a68a5 100644
--- a/source/slang/hlsl.meta.slang
+++ b/source/slang/hlsl.meta.slang
@@ -6423,6 +6423,7 @@ uint NonUniformResourceIndex(uint index)
 __glsl_extension(GL_EXT_nonuniform_qualifier)
 [__readNone]
 [ForceInline]
+[NonUniformReturn]
 int NonUniformResourceIndex(int index)
 {
     __target_switch
@@ -8730,6 +8731,7 @@ bool WaveActiveAnyTrue(bool condition)
 
 __glsl_extension(GL_KHR_shader_subgroup_ballot)
 __spirv_version(1.3)
+[NonUniformReturn]
 uint4 WaveActiveBallot(bool condition)
 {
     __target_switch
@@ -8757,6 +8759,7 @@ uint WaveActiveCountBits(bool value)
 
 __glsl_extension(GL_KHR_shader_subgroup_basic)
 __spirv_version(1.3)
+[NonUniformReturn]
 uint WaveGetLaneCount()
 {
     __target_switch
@@ -8775,6 +8778,7 @@ uint WaveGetLaneCount()
 
 __glsl_extension(GL_KHR_shader_subgroup_basic)
 __spirv_version(1.3)
+[NonUniformReturn]
 uint WaveGetLaneIndex()
 {
     __target_switch
@@ -8793,6 +8797,7 @@ uint WaveGetLaneIndex()
 
 __glsl_extension(GL_KHR_shader_subgroup_basic)
 __spirv_version(1.3)
+[NonUniformReturn]
 bool WaveIsFirstLane()
 {
     __target_switch
@@ -9802,6 +9807,7 @@ void AcceptHitAndEndSearch()
 
 // 10.4.1 - Ray Dispatch System Values
 
+[NonUniformReturn]
 uint3 DispatchRaysIndex()
 {
     __target_switch
@@ -9832,6 +9838,7 @@ uint3 DispatchRaysDimensions()
 
 // 10.4.2 - Ray System Values
 
+[NonUniformReturn]
 float3 WorldRayOrigin()
 {
     __target_switch
@@ -9846,6 +9853,7 @@ float3 WorldRayOrigin()
     }
 }
 
+[NonUniformReturn]
 float3 WorldRayDirection()
 {
     __target_switch
@@ -9860,6 +9868,7 @@ float3 WorldRayDirection()
     }
 }
 
+[NonUniformReturn]
 float RayTMin()
 {
     __target_switch
@@ -9884,6 +9893,7 @@ float RayTMin()
 // we should simply provide two overloads here, specialized
 // to the appropriate Vulkan stages.
 //
+[NonUniformReturn]
 float RayTCurrent()
 {
     __target_switch
@@ -9914,6 +9924,7 @@ uint RayFlags()
 
 // 10.4.3 - Primitive/Object Space System Values
 
+[NonUniformReturn]
 uint InstanceIndex()
 {
     __target_switch
@@ -9928,6 +9939,7 @@ uint InstanceIndex()
     }
 }
 
+[NonUniformReturn]
 uint InstanceID()
 {
     __target_switch
@@ -9942,6 +9954,7 @@ uint InstanceID()
     }
 }
 
+[NonUniformReturn]
 uint PrimitiveIndex()
 {
     __target_switch
@@ -9956,6 +9969,7 @@ uint PrimitiveIndex()
     }
 }
 
+[NonUniformReturn]
 float3 ObjectRayOrigin()
 {
     __target_switch
@@ -9970,6 +9984,7 @@ float3 ObjectRayOrigin()
     }
 }
 
+[NonUniformReturn]
 float3 ObjectRayDirection()
 {
     __target_switch
@@ -9986,6 +10001,7 @@ float3 ObjectRayDirection()
 
 // TODO: optix has an optixGetObjectToWorldTransformMatrix function that returns 12
 // floats by reference.
+[NonUniformReturn]
 float3x4 ObjectToWorld3x4()
 {
     __target_switch
@@ -10000,6 +10016,7 @@ float3x4 ObjectToWorld3x4()
     }
 }
 
+[NonUniformReturn]
 float3x4 WorldToObject3x4()
 {
     __target_switch
@@ -10014,6 +10031,7 @@ float3x4 WorldToObject3x4()
     }
 }
 
+[NonUniformReturn]
 float4x3 ObjectToWorld4x3()
 {
     __target_switch
@@ -10027,6 +10045,7 @@ float4x3 ObjectToWorld4x3()
     }
 }
 
+[NonUniformReturn]
 float4x3 WorldToObject4x3()
 {
     __target_switch
@@ -10047,6 +10066,7 @@ float4x3 WorldToObject4x3()
 __glsl_version(460)
 __glsl_extension(GL_NV_ray_tracing_motion_blur)
 __glsl_extension(GL_EXT_ray_tracing)
+[NonUniformReturn]
 float RayCurrentTime()
 {
     __target_switch
@@ -10069,10 +10089,11 @@ float RayCurrentTime()
 // declarations, so that users can know they aren't coding
 // against the final spec?
 //
-float3x4 ObjectToWorld() { return ObjectToWorld3x4(); }
-float3x4 WorldToObject() { return WorldToObject3x4(); }
+[NonUniformReturn] float3x4 ObjectToWorld() { return ObjectToWorld3x4(); }
+[NonUniformReturn] float3x4 WorldToObject() { return WorldToObject3x4(); }
 
 // 10.4.4 - Hit Specific System values
+[NonUniformReturn]
 uint HitKind()
 {
     __target_switch
@@ -10270,6 +10291,7 @@ extension __TextureImpl<T,__Shape2D, 1, 0, 0, $(kStdlibResourceAccessFeedback),
 
 // Get the index of the geometry that was hit in an intersection, any-hit, or closest-hit shader
 __glsl_extension(GL_EXT_ray_tracing)
+[NonUniformReturn]
 uint GeometryIndex()
 {
     __target_switch
@@ -10532,6 +10554,7 @@ struct RayQuery <let rayFlagsGeneric : RAY_FLAG = RAY_FLAG_NONE>
     __glsl_extension(GL_EXT_ray_query)
     __glsl_version(460)
     [__NoSideEffect]
+    [NonUniformReturn]
     CANDIDATE_TYPE CandidateType()
     {
         __target_switch
@@ -10550,6 +10573,7 @@ struct RayQuery <let rayFlagsGeneric : RAY_FLAG = RAY_FLAG_NONE>
     __glsl_extension(GL_EXT_ray_query)
     __glsl_version(460)
     [__NoSideEffect]
+    [NonUniformReturn]
     COMMITTED_STATUS CommittedStatus()
     {
         __target_switch
@@ -10568,6 +10592,7 @@ struct RayQuery <let rayFlagsGeneric : RAY_FLAG = RAY_FLAG_NONE>
     __glsl_extension(GL_EXT_ray_query)
     __glsl_version(460)
     [__NoSideEffect]
+    [NonUniformReturn]
     bool CandidateProceduralPrimitiveNonOpaque()
     {
         __target_switch
@@ -10587,6 +10612,7 @@ struct RayQuery <let rayFlagsGeneric : RAY_FLAG = RAY_FLAG_NONE>
     __glsl_extension(GL_EXT_ray_query)
     __glsl_version(460)
     [__NoSideEffect]
+    [NonUniformReturn]
     float CandidateTriangleRayT()
     {
         __target_switch
@@ -10604,6 +10630,7 @@ struct RayQuery <let rayFlagsGeneric : RAY_FLAG = RAY_FLAG_NONE>
     __glsl_extension(GL_EXT_ray_query)
     __glsl_version(460)
     [__NoSideEffect]
+    [NonUniformReturn]
     float CommittedRayT()
     {
         __target_switch
@@ -10675,6 +10702,7 @@ ${{{{
     __glsl_extension(GL_EXT_ray_query)
     __glsl_version(460)
     [__NoSideEffect]
+    [NonUniformReturn]
     float3x4 $(ccName)$(matName)3x4()
     {
         __target_switch
@@ -10693,6 +10721,7 @@ ${{{{
     __glsl_extension(GL_EXT_ray_query)
     __glsl_version(460)
     [__readNone]
+    [NonUniformReturn]
     float4x3 $(ccName)$(matName)4x3()
     {
         __target_switch
@@ -10734,6 +10763,7 @@ ${{{{
     __glsl_extension(GL_EXT_ray_query)
     __glsl_version(460)
     [__NoSideEffect]
+    [NonUniformReturn]
     $(method.type) $(ccName)$(method.hlslName)()
     {
         __target_switch
@@ -10757,6 +10787,7 @@ ${{{{
     __glsl_extension(GL_EXT_ray_query)
     __glsl_version(460)
     [__NoSideEffect]
+    [NonUniformReturn]
     uint RayFlags()
     {
         __target_switch
@@ -10773,6 +10804,7 @@ ${{{{
     __glsl_extension(GL_EXT_ray_query)
     __glsl_version(460)
     [__NoSideEffect]
+    [NonUniformReturn]
     float3 WorldRayOrigin()
     {
         __target_switch
@@ -10789,6 +10821,7 @@ ${{{{
     __glsl_extension(GL_EXT_ray_query)
     __glsl_version(460)
     [__NoSideEffect]
+    [NonUniformReturn]
     float3 WorldRayDirection()
     {
         __target_switch
@@ -10805,6 +10838,7 @@ ${{{{
     __glsl_extension(GL_EXT_ray_query)
     __glsl_version(460)
     [__NoSideEffect]
+    [NonUniformReturn]
     float RayTMin()
     {
         __target_switch
@@ -12082,6 +12116,7 @@ void debugBreak();
 [__requiresNVAPI]
 __glsl_extension(GL_EXT_shader_realtime_clock)
 [require(shaderclock)]
+[NonUniformReturn]
 uint getRealtimeClockLow()
 {
     __target_switch
@@ -12100,12 +12135,14 @@ uint getRealtimeClockLow()
 }
 
 __target_intrinsic(cpp, "std::chrono::high_resolution_clock::now().time_since_epoch().count()")
-__target_intrinsic(cuda, "clock64")
+    __target_intrinsic(cuda, "clock64")
+[NonUniformReturn]
 int64_t __cudaCppGetRealtimeClock();
 
 [__requiresNVAPI]
 __glsl_extension(GL_EXT_shader_realtime_clock)
 [require(shaderclock)]
+[NonUniformReturn]
 uint2 getRealtimeClock()
 {
     __target_switch
@@ -12134,14 +12171,17 @@ uint2 getRealtimeClock()
 
 __target_intrinsic(cuda, "(threadIdx)")
 [__readNone]
+[NonUniformReturn]
 uint3 cudaThreadIdx();
 
 __target_intrinsic(cuda, "(blockIdx)")
 [__readNone]
+[NonUniformReturn]
 uint3 cudaBlockIdx();
 
 __target_intrinsic(cuda, "(blockDim)")
 [__readNone]
+[NonUniformReturn]
 uint3 cudaBlockDim();
 
 //
diff --git a/source/slang/slang-ast-modifier.h b/source/slang/slang-ast-modifier.h
index c5d18f4f8..8e5120fad 100644
--- a/source/slang/slang-ast-modifier.h
+++ b/source/slang/slang-ast-modifier.h
@@ -634,6 +634,10 @@ class AttributeUsageAttribute : public Attribute
     SyntaxClass<NodeBase> targetSyntaxClass;
 };
 
+class NonDynamicUniformAttribute : public Attribute
+{
+    SLANG_AST_CLASS(NonDynamicUniformAttribute)
+};
 
 class RequireCapabilityAttribute : public Attribute
 {
@@ -1482,7 +1486,7 @@ class GLSLPatchModifier : public SimpleModifier
 //
 class BitFieldModifier : public Modifier
 {
-    SLANG_ABSTRACT_AST_CLASS(BitFieldModifier)
+    SLANG_AST_CLASS(BitFieldModifier)
 
     IntegerLiteralValue width;
 
@@ -1491,4 +1495,9 @@ class BitFieldModifier : public Modifier
     DeclRef<VarDecl> backingDeclRef;
 };
 
+class DynamicUniformModifier : public Modifier
+{
+    SLANG_AST_CLASS(DynamicUniformModifier)
+};
+
 } // namespace Slang
diff --git a/source/slang/slang-check-modifier.cpp b/source/slang/slang-check-modifier.cpp
index 2d9107431..0f49891d0 100644
--- a/source/slang/slang-check-modifier.cpp
+++ b/source/slang/slang-check-modifier.cpp
@@ -1083,6 +1083,7 @@ namespace Slang
         case ASTNodeType::HLSLCentroidModifier:
         case ASTNodeType::PerVertexModifier:
         case ASTNodeType::HLSLUniformModifier:
+        case ASTNodeType::DynamicUniformModifier:
             return (as<VarDeclBase>(decl) && (isGlobalDecl(decl) || as<StructDecl>(getParentDecl(decl)))) || as<ParamDecl>(decl);
 
         case ASTNodeType::HLSLSemantic:
diff --git a/source/slang/slang-diagnostic-defs.h b/source/slang/slang-diagnostic-defs.h
index 2150be5ad..8264d256b 100644
--- a/source/slang/slang-diagnostic-defs.h
+++ b/source/slang/slang-diagnostic-defs.h
@@ -716,6 +716,10 @@ DIAGNOSTIC(41904, Error, unableToAlignOf, "alignof could not be performed for ty
 DIAGNOSTIC(42001, Error, invalidUseOfTorchTensorTypeInDeviceFunc, "invalid use of TorchTensor type in device/kernel functions. use `TensorView` instead.")
 
 DIAGNOSTIC(45001, Error, unresolvedSymbol, "unresolved external symbol '$0'.")
+
+DIAGNOSTIC(41201, Warning, expectDynamicUniformArgument, "argument for '$0' is not a dynamic uniform.")
+DIAGNOSTIC(41201, Warning, expectDynamicUniformValue, "value stored at this location must be dynamic uniform.")
+
 //
 // 5xxxx - Target code generation.
 //
diff --git a/source/slang/slang-emit.cpp b/source/slang/slang-emit.cpp
index 2217bc143..1d72fd233 100644
--- a/source/slang/slang-emit.cpp
+++ b/source/slang/slang-emit.cpp
@@ -71,6 +71,7 @@
 #include "slang-ir-string-hash.h"
 #include "slang-ir-simplify-for-emit.h"
 #include "slang-ir-pytorch-cpp-binding.h"
+#include "slang-ir-uniformity.h"
 #include "slang-ir-vk-invert-y.h"
 #include "slang-legalize-types.h"
 #include "slang-lower-to-ir.h"
@@ -371,6 +372,13 @@ Result linkAndOptimizeIR(
 
     simplifyIR(targetProgram, irModule, IRSimplificationOptions::getDefault(), sink);
 
+    if (targetProgram->getOptionSet().getBoolOption(CompilerOptionName::ValidateUniformity))
+    {
+        validateUniformity(irModule, sink);
+        if (sink->getErrorCount() != 0)
+            return SLANG_FAIL;
+    }
+
     // Fill in default matrix layout into matrix types that left layout unspecified.
     specializeMatrixLayout(codeGenContext->getTargetProgram(), irModule);
 
diff --git a/source/slang/slang-ir-entry-point-uniforms.cpp b/source/slang/slang-ir-entry-point-uniforms.cpp
index 518f6ae2c..a269ec171 100644
--- a/source/slang/slang-ir-entry-point-uniforms.cpp
+++ b/source/slang/slang-ir-entry-point-uniforms.cpp
@@ -161,6 +161,8 @@ bool isVaryingParameter(IRTypeLayout* typeLayout)
 
 bool isVaryingParameter(IRVarLayout* varLayout)
 {
+    if (!varLayout)
+        return false;
     return isVaryingParameter(varLayout->getTypeLayout());
 }
 
diff --git a/source/slang/slang-ir-inst-defs.h b/source/slang/slang-ir-inst-defs.h
index 656b3d320..4865aa0b5 100644
--- a/source/slang/slang-ir-inst-defs.h
+++ b/source/slang/slang-ir-inst-defs.h
@@ -838,6 +838,9 @@ INST(HighLevelDeclDecoration,               highLevelDecl,          1, 0)
         // Marks a type to be non copyable, causing SSA pass to skip turning variables of the the type into SSA values.
     INST(NonCopyableTypeDecoration, nonCopyable, 0, 0)
 
+        // Marks a value to be dynamically uniform.
+    INST(DynamicUniformDecoration, DynamicUniform, 0, 0)
+
         /// A call to the decorated function should always be folded into its use site.
     INST(AlwaysFoldIntoUseSiteDecoration, alwaysFold, 0, 0)
 
@@ -951,6 +954,9 @@ INST(HighLevelDeclDecoration,               highLevelDecl,          1, 0)
 
     INST_RANGE(CheckpointHintDecoration, PreferCheckpointDecoration, PreferRecomputeDecoration)
 
+        /// Marks a function whose return value is never dynamic uniform.
+    INST(NonDynamicUniformReturnDecoration, NonDynamicUniformReturnDecoration, 0, 0)
+
         /// Marks a class type as a COM interface implementation, which enables
         /// the witness table to be easily picked up by emit.
     INST(COMWitnessDecoration, COMWitnessDecoration, 1, 0)
@@ -1023,6 +1029,7 @@ INST(CastPtrToInt, CastPtrToInt, 1, 0)
 INST(CastIntToPtr, CastIntToPtr, 1, 0)
 INST(CastToVoid, castToVoid, 1, 0)
 INST(PtrCast, PtrCast, 1, 0)
+INST(TreatAsDynamicUniform, TreatAsDynamicUniform, 1, 0)
 
 INST(SizeOf,                            sizeOf,                     1, 0)
 INST(AlignOf,                           alignOf,                    1, 0)
diff --git a/source/slang/slang-ir-insts.h b/source/slang/slang-ir-insts.h
index a80f7fb29..92a9d473f 100644
--- a/source/slang/slang-ir-insts.h
+++ b/source/slang/slang-ir-insts.h
@@ -685,6 +685,8 @@ struct IROffsetDecoration : IRDecoration
     IRIntegerValue getOffset() { return getOffsetOperand()->getValue(); }
 };
 
+IR_SIMPLE_DECORATION(DynamicUniformDecoration)
+
 struct IRBuiltinDecoration : IRDecoration
 {
     enum
@@ -4619,6 +4621,11 @@ public:
         addDecoration(value, kIROp_NonCopyableTypeDecoration);
     }
 
+    void addDynamicUniformDecoration(IRInst* value)
+    {
+        addDecoration(value, kIROp_DynamicUniformDecoration);
+    }
+
         /// Add a decoration that indicates that the given `inst` depends on the given `dependency`.
         ///
         /// This decoration can be used to ensure that a value that an instruction
diff --git a/source/slang/slang-ir-uniformity.cpp b/source/slang/slang-ir-uniformity.cpp
new file mode 100644
index 000000000..9c463f530
--- /dev/null
+++ b/source/slang/slang-ir-uniformity.cpp
@@ -0,0 +1,474 @@
+#include "slang-ir-uniformity.h"
+
+#include "slang-ir.h"
+#include "slang-ir-insts.h"
+#include "slang-ir-util.h"
+#include "slang-ir-dominators.h"
+
+namespace Slang
+{
+    struct ValidateUniformityContext
+    {
+        IRModule* module;
+        DiagnosticSink* sink;
+
+        HashSet<IRInst*> nonUniformInsts;
+        ValidateUniformityContext* parentContext = nullptr;
+        IRCall* call = nullptr;
+        IRFunc* currentCallee = nullptr;
+
+        bool isInstNonUniform(IRInst* inst)
+        {
+            auto context = this;
+            while (context)
+            {
+                if (context->nonUniformInsts.contains(inst))
+                    return true;
+                context = context->parentContext;
+            }
+            return false;
+        }
+
+        struct FunctionNonUniformInfoKey
+        {
+            IRFunc* func;
+            UIntSet nonUniformParams;
+
+            bool operator==(const FunctionNonUniformInfoKey& other) const
+            {
+                return func == other.func && nonUniformParams == other.nonUniformParams;
+            }
+            HashCode getHashCode() const
+            {
+                return combineHash(Slang::getHashCode(func), nonUniformParams.getHashCode());
+            }
+        };
+
+        struct FunctionNonUniformInfo
+        {
+            UIntSet nonUniformParams;
+            bool isResultNonUniform = false;
+        };
+
+        Dictionary<FunctionNonUniformInfoKey, FunctionNonUniformInfo> functionNonUniformInfos;
+
+        template<typename F>
+        void traverseControlDependentBlocks(IRDominatorTree* dom, IRInst* inst, const F& f)
+        {
+            auto block = as<IRBlock>(inst->getParent());
+            if (!block)
+                return;
+            for (auto idom = dom->getImmediateDominator(block); idom; idom = dom->getImmediateDominator(idom))
+            {
+                if (as<IRUnconditionalBranch>(idom->getTerminator()))
+                    continue;
+                if (auto ifelse = as<IRIfElse>(idom->getTerminator()))
+                {
+                    if (dom->dominates(ifelse->getAfterBlock(), block))
+                        continue;
+                }
+                else if (auto switchInst = as<IRSwitch>(idom->getTerminator()))
+                {
+                    if (dom->dominates(switchInst->getBreakLabel(), block))
+                        continue;
+                }
+                else if (auto loopInst = as<IRLoop>(idom->getTerminator()))
+                {
+                    if (dom->dominates(loopInst->getBreakBlock(), block))
+                        continue;
+                }
+                f(idom);
+            }
+        }
+
+        FunctionNonUniformInfo* getFunctionNonUniformInfo(IRCall* callInst, const FunctionNonUniformInfoKey& key)
+        {
+            if (auto rs = functionNonUniformInfos.tryGetValue(key))
+                return rs;
+
+            // Is the function already being analyzed? If so exit early to avoid infinite recursion.
+            for (auto context = this; context; context = context->parentContext)
+            {
+                if (context->currentCallee == key.func)
+                    return nullptr;
+            }
+
+            // If the function body has target intrinsic, we can't analyze it, and we
+            // will use the fallback behavior (result is non-uniform if any of its arguments are non-uniform).
+            for (auto block : key.func->getBlocks())
+            {
+                if (auto genAsm = as<IRGenericAsm>(block->getTerminator()))
+                {
+                    return nullptr;
+                }
+            }
+
+            ValidateUniformityContext subContext;
+            subContext.module = module;
+            subContext.sink = sink;
+            subContext.parentContext = this;
+
+            List<IRInst*> workList;
+            Index paramIndex = 0;
+            for (auto param : key.func->getParams())
+            {
+                if (key.nonUniformParams.contains(UInt(paramIndex)))
+                {
+                    subContext.nonUniformInsts.add(param);
+                    workList.add(param);
+                }
+                paramIndex++;
+            }
+            subContext.call = callInst;
+            subContext.currentCallee = key.func;
+            subContext.propagateNonUniform(key.func, workList);
+
+            FunctionNonUniformInfo info;
+            info.nonUniformParams = key.nonUniformParams;
+            paramIndex = 0;
+            for (auto param : key.func->getParams())
+            {
+                if (subContext.nonUniformInsts.contains(param))
+                {
+                    info.nonUniformParams.add(paramIndex);
+                }
+                paramIndex++;
+            }
+
+            // If the function has [NonUniformReturn] attribute,
+            // treat its return value as non uniform.
+            if (key.func->findDecorationImpl(kIROp_NonDynamicUniformReturnDecoration))
+            {
+                info.isResultNonUniform = true;
+            }
+            else
+            {
+                // The return value is non-uniform if the any values used in IRReturn is
+                // non-uniform, or if the return insts are control-dependent on non-uniform
+                // values.
+                for (auto bb : key.func->getBlocks())
+                {
+                    if (auto ret = as<IRReturn>(bb->getTerminator()))
+                    {
+                        if (subContext.isInstNonUniform(ret->getVal()) || subContext.isInstNonUniform(ret))
+                        {
+                            info.isResultNonUniform = true;
+                            break;
+                        }
+                    }
+                }
+            }
+            functionNonUniformInfos[key] = info;
+            return functionNonUniformInfos.tryGetValue(key);
+        }
+
+        bool isDynamicUniformLocation(IRInst* addr)
+        {
+            while (addr)
+            {
+                switch (addr->getOp())
+                {
+                case kIROp_FieldAddress:
+                    if (as<IRFieldAddress>(addr)->getField()->findDecoration<IRDynamicUniformDecoration>())
+                        return true;
+                    addr = as<IRFieldAddress>(addr)->getBase();
+                    break;
+                case kIROp_GetElementPtr:
+                    addr = as<IRGetElementPtr>(addr)->getBase();
+                    break;
+                case kIROp_GetOffsetPtr:
+                    addr = addr->getOperand(0);
+                    break;
+                case kIROp_Param:
+                case kIROp_Var:
+                    return addr->findDecoration<IRDynamicUniformDecoration>() != nullptr;
+                default:
+                    addr = nullptr;
+                }
+            }
+            return false;
+        }
+
+        void propagateNonUniform(IRFunc* root, List<IRInst*>& workList)
+        {
+            List<IRInst*>& nextWorkList = *module->getContainerPool().getList<IRInst>();
+            HashSet<IRInst*>& workListSet = *module->getContainerPool().getHashSet<IRInst>();
+            auto addToWorkList = [&](IRInst* inst)
+                {
+                    if (workListSet.add(inst))
+                    {
+                        nonUniformInsts.add(inst);
+                        nextWorkList.add(inst);
+                    }
+                };
+
+            // Go through the children first to identify initial non-uniform insts.
+            for (auto block : root->getBlocks())
+            {
+                for (auto inst = block->getFirstInst(); inst; inst = inst->getNextInst())
+                {
+                    switch (inst->getOp())
+                    {
+                    case kIROp_Call:
+                    {
+                        auto callInst = as<IRCall>(inst);
+                        auto callee = getResolvedInstForDecorations(callInst->getCallee());
+                        if (callee->findDecorationImpl(kIROp_NonDynamicUniformReturnDecoration))
+                        {
+                            addToWorkList(inst);
+                        }
+                        break;
+                    }
+                    }
+                }
+            }
+
+            auto dom = module->findOrCreateDominatorTree(root);
+
+            auto visitControlDependentBlock = [&](IRBlock* dependentBlock)
+                {
+                    if (!dependentBlock)
+                        return;
+                    for (auto block : dom->getProperlyDominatedBlocks(dependentBlock))
+                    {
+                        for (auto inst = block->getFirstInst(); inst; inst = inst->getNextInst())
+                        {
+                            switch (inst->getOp())
+                            {
+                            case kIROp_Store:
+                            case kIROp_SwizzledStore:
+                                addToWorkList(inst->getOperand(0));
+                                break;
+                            case kIROp_Return:
+                                addToWorkList(inst);
+                                break;
+                            case kIROp_Call:
+                                {
+                                    auto call = as<IRCall>(inst);
+                                    for (UInt i = 0; i < call->getArgCount(); i++)
+                                    {
+                                        if (as<IRPtrTypeBase>(call->getArg(i)))
+                                            addToWorkList(call->getArg(i));
+                                    }
+                                }
+                                break;
+                            }
+                        }
+                    }
+                };
+
+            while (workList.getCount())
+            {
+                for (Index i = 0; i < workList.getCount(); i++)
+                {
+                    auto inst = workList[i];
+                    for (auto use = inst->firstUse; use; use = use->nextUse)
+                    {
+                        auto user = use->getUser();
+                        if (as<IRAttr>(user))
+                            continue;
+                        if (as<IRDecoration>(user))
+                            continue;
+                        switch (user->getOp())
+                        {
+                        case kIROp_TreatAsDynamicUniform:
+                            continue;
+                        case kIROp_FieldAddress:
+                        {
+                            if (isDynamicUniformLocation(user))
+                                continue;
+                            break;
+                        }
+                        case kIROp_FieldExtract:
+                        {
+                            if (as<IRFieldExtract>(user)->findDecoration<IRDynamicUniformDecoration>())
+                                continue;
+                            break;
+                        }
+                        case kIROp_SwizzledStore:
+                        case kIROp_Store:
+                        {
+                            if (use == user->getOperands() + 1)
+                            {
+                                auto ptr = user->getOperand(0);
+                                addToWorkList(ptr);
+                                if (isDynamicUniformLocation(ptr))
+                                {
+                                    sink->diagnose(user->sourceLoc, Diagnostics::expectDynamicUniformValue, ptr);
+                                }
+                                else
+                                {
+                                    // Conservatively treat the entire composite at root addr as non-uniform.
+                                    auto addrRoot = getRootAddr(ptr);
+                                    addToWorkList(addrRoot);
+                                }
+                            }
+                            break;
+                        }
+                        case kIROp_ifElse:
+                        {
+                            auto ifElse = as<IRIfElse>(user);
+                            visitControlDependentBlock(ifElse->getTrueBlock());
+                            visitControlDependentBlock(ifElse->getFalseBlock());
+                            break;
+                        }
+                        case kIROp_Switch:
+                        {
+                            auto switchInst = as<IRSwitch>(user);
+                            for (UInt c = 0; c < switchInst->getCaseCount(); c++)
+                                visitControlDependentBlock(switchInst->getCaseLabel(c));
+                            visitControlDependentBlock(switchInst->getDefaultLabel());
+                            break;
+                        }
+                        case kIROp_Call:
+                        {
+                            auto callInst = as<IRCall>(user);
+                            auto callee = getResolvedInstForDecorations(callInst->getCallee());
+                            if (auto func = as<IRFunc>(callee))
+                            {
+                                if (func->getFirstBlock())
+                                {
+                                    FunctionNonUniformInfoKey key;
+                                    key.func = func;
+                                    for (UInt argi = 0; argi < callInst->getArgCount(); argi++)
+                                    {
+                                        if (nonUniformInsts.contains(callInst->getArg(argi)))
+                                        {
+                                            auto param = getParamAt(func->getFirstBlock(), argi);
+                                            if (param->findDecoration<IRDynamicUniformDecoration>())
+                                            {
+                                                sink->diagnose(callInst->sourceLoc, Diagnostics::expectDynamicUniformArgument, param);
+                                            }
+                                            else
+                                            {
+                                                key.nonUniformParams.add(i);
+                                            }
+                                        }
+                                    }
+                                    if (auto funcInfo = getFunctionNonUniformInfo(callInst, key))
+                                    {
+                                        for (UInt argi = 0; argi < callInst->getArgCount(); argi++)
+                                        {
+                                            if (funcInfo->nonUniformParams.contains(argi))
+                                            {
+                                                addToWorkList(callInst->getArg(argi));
+                                            }
+                                            if (funcInfo->isResultNonUniform)
+                                            {
+                                                addToWorkList(callInst);
+                                            }
+                                        }
+                                        break;
+                                    }
+                                }
+                            }
+                            // The default behavior for calls is that the result is non-uniform if
+                            // any of its arguments are non-uniform.
+                            bool isNonUniformCall = callee->findDecorationImpl(kIROp_NonDynamicUniformReturnDecoration) != nullptr;
+                            if (!isNonUniformCall)
+                            {
+                                for (UInt argi = 0; argi < callInst->getArgCount(); argi++)
+                                {
+                                    if (nonUniformInsts.contains(callInst->getArg(argi)))
+                                    {
+                                        isNonUniformCall = true;
+                                        break;
+                                    }
+                                }
+                            }
+                            if (isNonUniformCall)
+                            {
+                                addToWorkList(callInst);
+                                for (UInt argi = 0; argi < callInst->getArgCount(); argi++)
+                                {
+                                    if (auto ptrType = as<IRPtrTypeBase>(callInst->getArg(argi)->getDataType()))
+                                    {
+                                        addToWorkList(callInst->getArg(argi));
+                                        // Conservatively treat the entire composite at root addr as non-uniform.
+                                        auto addrRoot = getRootAddr(callInst->getArg(argi));
+                                        addToWorkList(addrRoot);
+                                    }
+                                }
+                            }
+                            break;
+                        }
+                        default:
+                            break;
+                        }
+                        addToWorkList(user);
+                    }
+                }
+                workList.swapWith(nextWorkList);
+                nextWorkList.clear();
+            }
+        }
+
+        void analyzeModule()
+        {
+            List<IRInst*>& workList = *module->getContainerPool().getList<IRInst>();
+            for (auto globalInst : module->getGlobalInsts())
+            {
+                if (auto code = as<IRGlobalValueWithCode>(globalInst))
+                {
+                    auto func = getResolvedInstForDecorations(code);
+                    if (func->findDecorationImpl(kIROp_NonDynamicUniformReturnDecoration))
+                    {
+                        nonUniformInsts.add(code);
+                    }
+                }
+                if (auto entryPointDecor = globalInst->findDecoration<IREntryPointDecoration>())
+                {
+                    auto func = as<IRFunc>(globalInst);
+                    if (!func)
+                        continue;
+                    for (auto param : func->getParams())
+                    {
+                        auto varLayout = findVarLayout(param);
+                        if (isVaryingParameter(varLayout) || varLayout->findAttr<IRSystemValueSemanticAttr>())
+                        {
+                            nonUniformInsts.add(param);
+                            workList.add(param);
+                        }
+                    }
+                    currentCallee = func;
+                    call = nullptr;
+                    propagateNonUniform(func, workList);
+                }
+            }
+            workList.clear();
+
+            eliminateAsDynamicUniformInst();
+        }
+
+        void eliminateAsDynamicUniformInst()
+        {
+            List<IRInst*>& workList = *module->getContainerPool().getList<IRInst>();
+            workList.add(module->getModuleInst());
+            for (Index i = 0; i < workList.getCount(); i++)
+            {
+                auto inst = workList[i];
+                if (inst->getOp() == kIROp_TreatAsDynamicUniform)
+                {
+                    auto val = inst->getOperand(0);
+                    inst->replaceUsesWith(val);
+                    inst->removeAndDeallocate();
+                }
+                else
+                {
+                    for (auto child = inst->getFirstChild(); child; child = child->getNextInst())
+                    {
+                        workList.add(child);
+                    }
+                }
+            }
+        }
+    };
+
+    void validateUniformity(IRModule* module, DiagnosticSink* sink)
+    {
+        ValidateUniformityContext context;
+        context.module = module;
+        context.sink = sink;
+        context.analyzeModule();
+    }
+}
diff --git a/source/slang/slang-ir-uniformity.h b/source/slang/slang-ir-uniformity.h
new file mode 100644
index 000000000..db3c89bd6
--- /dev/null
+++ b/source/slang/slang-ir-uniformity.h
@@ -0,0 +1,10 @@
+// slang-ir-uniformity.h
+#pragma once
+
+namespace Slang
+{
+    struct IRModule;
+    class DiagnosticSink;
+
+    void validateUniformity(IRModule* module, DiagnosticSink* sink);
+}
diff --git a/source/slang/slang-lower-to-ir.cpp b/source/slang/slang-lower-to-ir.cpp
index 753b21589..146af452f 100644
--- a/source/slang/slang-lower-to-ir.cpp
+++ b/source/slang/slang-lower-to-ir.cpp
@@ -2187,6 +2187,10 @@ void addVarDecorations(
         {
             builder->addSemanticDecoration(inst, hlslSemantic->name.getContent());
         }
+        else if (auto dynamicUniform = as<DynamicUniformModifier>(mod))
+        {
+            builder->addDynamicUniformDecoration(inst);
+        }
         // TODO: what are other modifiers we need to propagate through?
     }
     if(auto t = composeGetters<IRMeshOutputType>(inst->getFullType(), &IROutTypeBase::getValueType))
@@ -8376,9 +8380,16 @@ struct DeclLoweringVisitor : DeclVisitor<DeclLoweringVisitor, LoweredValInfo>
                 fieldKey,
                 fieldType);
 
-            if (auto packOffsetModifier = fieldDecl->findModifier<HLSLPackOffsetSemantic>())
+            for (auto mod : fieldDecl->modifiers)
             {
-                lowerPackOffsetModifier(fieldKey, packOffsetModifier);
+                if (auto packOffsetModifier = as<HLSLPackOffsetSemantic>(mod))
+                {
+                    lowerPackOffsetModifier(fieldKey, packOffsetModifier);
+                }
+                else if (auto dynamicUniformModifer = as<DynamicUniformModifier>(mod))
+                {
+                    subBuilder->addDynamicUniformDecoration(fieldKey);
+                }
             }
         }
 
@@ -9669,6 +9680,8 @@ struct DeclLoweringVisitor : DeclVisitor<DeclLoweringVisitor, LoweredValInfo>
                 getBuilder()->addRequireSPIRVVersionDecoration(irFunc, spvVersion->version);
             else if (auto cudasmVersion = as<RequiredCUDASMVersionModifier>(modifier))
                 getBuilder()->addRequireCUDASMVersionDecoration(irFunc, cudasmVersion->version);
+            else if (auto nonUniform= as<NonDynamicUniformAttribute>(modifier))
+                getBuilder()->addDecoration(irFunc, kIROp_NonDynamicUniformReturnDecoration);
         }
 
         if (!isInline)
diff --git a/source/slang/slang-options.cpp b/source/slang/slang-options.cpp
index 6b56d5600..13984e530 100644
--- a/source/slang/slang-options.cpp
+++ b/source/slang/slang-options.cpp
@@ -516,6 +516,7 @@ void initCommandOptions(CommandOptions& options)
         "Set the filesystem hook to use for a compile request."},
         { OptionKind::Heterogeneous, "-heterogeneous", nullptr, "Output heterogeneity-related code." },
         { OptionKind::NoMangle, "-no-mangle", nullptr, "Do as little mangling of names as possible." },
+        { OptionKind::ValidateUniformity, "-validate-uniformity", nullptr, "Perform uniformity validation analysis." },
         { OptionKind::AllowGLSL, "-allow-glsl", nullptr, "Enable GLSL as an input language." },
     };
     _addOptions(makeConstArrayView(experimentalOpts), options);
@@ -1664,6 +1665,7 @@ SlangResult OptionsParser::_parse(
         switch (optionKind)
         {
             case OptionKind::NoMangle:
+            case OptionKind::ValidateUniformity:
             case OptionKind::AllowGLSL:
             case OptionKind::EmitIr:
             case OptionKind::DumpIntermediates:
diff --git a/source/slang/slang-parser.cpp b/source/slang/slang-parser.cpp
index 3f1ed3f76..ae40234bc 100644
--- a/source/slang/slang-parser.cpp
+++ b/source/slang/slang-parser.cpp
@@ -8025,6 +8025,7 @@ namespace Slang
         _makeParseModifier("uniform",       HLSLUniformModifier::kReflectClassInfo),
         _makeParseModifier("volatile",      HLSLVolatileModifier::kReflectClassInfo),
         _makeParseModifier("export",        HLSLExportModifier::kReflectClassInfo),
+        _makeParseModifier("dynamic_uniform", DynamicUniformModifier::kReflectClassInfo),
 
         // Modifiers for geometry shader input
         _makeParseModifier("point",         HLSLPointModifier::kReflectClassInfo),
diff --git a/tests/language-feature/dynamic-uniform/dynamic-uniform-1.slang b/tests/language-feature/dynamic-uniform/dynamic-uniform-1.slang
new file mode 100644
index 000000000..61583a84d
--- /dev/null
+++ b/tests/language-feature/dynamic-uniform/dynamic-uniform-1.slang
@@ -0,0 +1,73 @@
+
+
+//DIAGNOSTIC_TEST:SIMPLE(filecheck=CHECK): -target hlsl -entry main -profile cs_6_0 -validate-uniformity
+
+RWStructuredBuffer<int> output;
+void expectUniform(dynamic_uniform int v)
+{
+    output[0] = v;
+}
+
+[NonUniformReturn]
+int produceNonUniform() { return 0; }
+
+struct B
+{
+    int a;
+    int c;
+    dynamic_uniform int d;
+}
+
+int getBa(B obj)
+{
+    return obj.a;
+}
+
+int getBaFromInout(inout B obj)
+{
+    return obj.a;
+}
+
+int getBdFromInout(inout B obj)
+{
+    return obj.d;
+}
+
+ConstantBuffer<B> ub;
+void main(int tid: SV_DispatchThreadID)
+{
+    int d = ub.a + ub.c;
+    int f = ub.a + tid;
+    int ff = ub.a + produceNonUniform();
+    
+    // CHECK-DAG: ([[# @LINE+1]]): warning 41201
+    expectUniform(f);
+    // CHECK-DAG: ([[# @LINE+1]]): warning 41201
+    expectUniform(ff);
+
+    B obj = ub;
+    // CHECK-DAG: ([[# @LINE+1]]): warning 41201
+    expectUniform(getBaFromInout(obj));
+    obj.a = ff;
+    // CHECK-DAG: ([[# @LINE+1]]): warning 41201
+    expectUniform(getBa(obj));
+
+    // CHECK-DAG: ([[# @LINE+1]]): warning 41201
+    obj.d = ff;
+
+    // CHECK-NOT: ([[# @LINE+1]]): warning 41201
+    expectUniform(getBdFromInout(obj));
+    
+    // CHECK-NOT: ([[# @LINE+1]]): warning 41201
+    obj.d = d; // OK
+
+    // CHECK-NOT: ([[# @LINE+1]]): warning 41201
+    expectUniform(d);
+
+    // CHECK-NOT: ([[# @LINE+1]]): warning 41201
+    expectUniform(getBa(ub));
+
+    // CHECK-NOT: ([[# @LINE+1]]): warning 41201
+    expectUniform(asDynamicUniform(ff));
+
+}