diff options
30 files changed, 2140 insertions, 180 deletions
diff --git a/source/slang/core.meta.slang b/source/slang/core.meta.slang index e1eb9c776..0a6278ec1 100644 --- a/source/slang/core.meta.slang +++ b/source/slang/core.meta.slang @@ -2797,6 +2797,9 @@ __attributeTarget(VarDeclBase) attribute_syntax [__vulkanCallablePayload(location : int = -1)] : VulkanCallablePayloadAttribute; __attributeTarget(VarDeclBase) +attribute_syntax [__vulkanHitObjectAttributes(location : int = -1)] : VulkanHitObjectAttributesAttribute; + +__attributeTarget(VarDeclBase) attribute_syntax [__vulkanHitAttributes] : VulkanHitAttributesAttribute; __attributeTarget(FunctionDeclBase) diff --git a/source/slang/hlsl.meta.slang b/source/slang/hlsl.meta.slang index 2a9a9f9d3..81df2b5e4 100644 --- a/source/slang/hlsl.meta.slang +++ b/source/slang/hlsl.meta.slang @@ -5557,12 +5557,44 @@ struct VkSubpassInputMS<T> /// /// https://developer.nvidia.com/rtx/path-tracing/nvapi/get-started /// +/// For VK the specification is currently in this PR +/// +/// https://github.com/KhronosGroup/GLSL/pull/196/files + +/// Internal helper functions + +// This is a bit of a hack for GLSL HitObjectAttributes +// It relies on [ForceInline] removing the surrounding function and just inserting the *contained* `t` as a global +// The __ref should indicate the desire for the returned value to not be a copy of t, but *t*. +// In practive __ref doesn't have this effect in practice. +// +// We need this to be able access the payload outside of a function (which is all that TraceRay for example needs) +// We access the HitObjectAttributes via this function for the desired type, and it acts *as if* it's just an access +// to the global t. +[ForceInline] +__ref T __hitObjectAttributes<T>() +{ + [__vulkanHitObjectAttributes] + static T t; + return t; +} + +// Next is the custom intrinsic that will compute the hitObjectAttributes location +// for GLSL-based targets. +// +__generic<Payload> +__target_intrinsic(__glslRayTracing, "$XH") +[__readNone] +int __hitObjectAttributesLocation(Payload payload); /// Immutable data type representing a ray hit or a miss. Can be used to invoke hit or miss shading, /// or as a key in ReorderThread. Created by one of several methods described below. HitObject /// and its related functions are available in raytracing shader types only. -__target_intrinsic(hlsl, NvHitObject) [__requiresNVAPI] +__target_intrinsic(hlsl, NvHitObject) +__glsl_extension(GL_NV_shader_invocation_reorder) +__glsl_extension(GL_EXT_ray_tracing) +__target_intrinsic(glsl, hitObjectNV) struct HitObject { /// Executes ray traversal (including anyhit and intersection shaders) like TraceRay, but returns the @@ -5577,20 +5609,104 @@ struct HitObject uint MissShaderIndex, RayDesc Ray, inout payload_t Payload) - { - HitObject hitObj; - __traceRay( - AccelerationStructure, - RayFlags, - InstanceInclusionMask, - RayContributionToHitGroupIndex, - MultiplierForGeometryContributionToHitGroupIndex, - MissShaderIndex, - Ray, - Payload, - hitObj); - return hitObj; - } + { + HitObject hitObj; + __hlslTraceRay( + AccelerationStructure, + RayFlags, + InstanceInclusionMask, + RayContributionToHitGroupIndex, + MultiplierForGeometryContributionToHitGroupIndex, + MissShaderIndex, + Ray, + Payload, + hitObj); + return hitObj; + } + + [ForceInline] + __specialized_for_target(glsl) + static HitObject TraceRay<payload_t>( + RaytracingAccelerationStructure AccelerationStructure, + uint RayFlags, + uint InstanceInclusionMask, + uint RayContributionToHitGroupIndex, + uint MultiplierForGeometryContributionToHitGroupIndex, + uint MissShaderIndex, + RayDesc Ray, + inout payload_t Payload) + { + HitObject hitObj; + + [__vulkanRayPayload] + static payload_t p; + + // Save the payload + p = Payload; + + __glslTraceRay( + hitObj, + AccelerationStructure, + RayFlags, // Assumes D3D/VK have some RayFlags values + InstanceInclusionMask, // cullMask + RayContributionToHitGroupIndex, // sbtRecordOffset + MultiplierForGeometryContributionToHitGroupIndex, // sbtRecordStride + MissShaderIndex, + Ray.Origin, + Ray.TMin, + Ray.Direction, + Ray.TMax, + __rayPayloadLocation(p)); + + // Write the payload out + Payload = p; + + return hitObj; + } + + /// Executes motion ray traversal (including anyhit and intersection shaders) like TraceRay, but returns the + /// resulting hit information as a HitObject and does not trigger closesthit or miss shaders. + [ForceInline] + __specialized_for_target(glsl) + static HitObject TraceMotionRay<payload_t>( + RaytracingAccelerationStructure AccelerationStructure, + uint RayFlags, + uint InstanceInclusionMask, + uint RayContributionToHitGroupIndex, + uint MultiplierForGeometryContributionToHitGroupIndex, + uint MissShaderIndex, + RayDesc Ray, + float CurrentTime, + inout payload_t Payload) + { + HitObject hitObj; + + [__vulkanRayPayload] + static payload_t p; + + // Save the payload + p = Payload; + + __glslTraceMotionRay( + hitObj, + AccelerationStructure, + RayFlags, // Assumes D3D/VK have some RayFlags values + InstanceInclusionMask, // cullMask + RayContributionToHitGroupIndex, // sbtRecordOffset + MultiplierForGeometryContributionToHitGroupIndex, // sbtRecordStride + MissShaderIndex, + Ray.Origin, + Ray.TMin, + Ray.Direction, + Ray.TMax, + CurrentTime, + __rayPayloadLocation(p)); + + // Write the payload out + Payload = p; + + return hitObj; + } /// Creates a HitObject representing a hit based on values explicitly passed as arguments, without /// tracing a ray. The primitive specified by AccelerationStructure, InstanceIndex, GeometryIndex, @@ -5611,8 +5727,8 @@ struct HitObject attr_t attributes) { HitObject hitObj; - __makeHit( - AccelerationStructure, + __hlslMakeHit( + AccelerationStructure, InstanceIndex, GeometryIndex, PrimitiveIndex, @@ -5625,6 +5741,84 @@ struct HitObject return hitObj; } + [ForceInline] + __specialized_for_target(glsl) + static HitObject MakeHit<attr_t>( + RaytracingAccelerationStructure AccelerationStructure, + uint InstanceIndex, + uint GeometryIndex, + uint PrimitiveIndex, + uint HitKind, + uint RayContributionToHitGroupIndex, + uint MultiplierForGeometryContributionToHitGroupIndex, + RayDesc Ray, + attr_t attributes) + { + HitObject hitObj; + + // Save the attributes + __ref attr_t attr = __hitObjectAttributes<attr_t>(); + + attr = attributes; + + __glslMakeHit(hitObj, + AccelerationStructure, + InstanceIndex, + PrimitiveIndex, + GeometryIndex, + HitKind, + RayContributionToHitGroupIndex, /// sbtRecordOffset? + MultiplierForGeometryContributionToHitGroupIndex, /// sbtRecordStride? + Ray.Origin, + Ray.TMin, + Ray.Direction, + Ray.TMax, + __hitObjectAttributesLocation(__hitObjectAttributes<attr_t>())); + + return hitObj; + } + + /// See MakeHit but handles Motion + /// Currently only supported on VK + [ForceInline] + __specialized_for_target(glsl) + static HitObject MakeMotionHit<attr_t>( + RaytracingAccelerationStructure AccelerationStructure, + uint InstanceIndex, + uint GeometryIndex, + uint PrimitiveIndex, + uint HitKind, + uint RayContributionToHitGroupIndex, + uint MultiplierForGeometryContributionToHitGroupIndex, + RayDesc Ray, + float CurrentTime, + attr_t attributes) + { + HitObject hitObj; + + // Save the attributes + __ref attr_t attr = __hitObjectAttributes<attr_t>(); + + attr = attributes; + + __glslMakeMotionHit(hitObj, + AccelerationStructure, + InstanceIndex, + PrimitiveIndex, + GeometryIndex, + HitKind, + RayContributionToHitGroupIndex, /// sbtRecordOffset? + MultiplierForGeometryContributionToHitGroupIndex, /// sbtRecordStride? + Ray.Origin, + Ray.TMin, + Ray.Direction, + Ray.TMax, + CurrentTime, + __hitObjectAttributesLocation(__hitObjectAttributes<attr_t>())); + + return hitObj; + } + /// Creates a HitObject representing a hit based on values explicitly passed as arguments, without /// tracing a ray. The primitive specified by AccelerationStructure, InstanceIndex, GeometryIndex, /// and PrimitiveIndex must exist. The shader table index is explicitly provided as an argument @@ -5644,9 +5838,9 @@ struct HitObject attr_t attributes) { HitObject hitObj; - __makeHitWithRecordIndex( - HitGroupRecordIndex, - AccelerationStructure, + __hlslMakeHitWithRecordIndex( + HitGroupRecordIndex, + AccelerationStructure, InstanceIndex, GeometryIndex, PrimitiveIndex, @@ -5657,81 +5851,221 @@ struct HitObject return hitObj; } + [ForceInline] + __specialized_for_target(glsl) + static HitObject MakeHit<attr_t>( + uint HitGroupRecordIndex, + RaytracingAccelerationStructure AccelerationStructure, + uint InstanceIndex, + uint GeometryIndex, + uint PrimitiveIndex, + uint HitKind, + RayDesc Ray, + attr_t attributes) + { + HitObject hitObj; + + // Save the attributes + __ref attr_t attr = __hitObjectAttributes<attr_t>(); + attr = attributes; + + __glslMakeHitWithIndex(hitObj, + AccelerationStructure, + InstanceIndex, ///? Same as instanceid ? + GeometryIndex, + PrimitiveIndex, + HitKind, /// Assuming HitKinds are compatible + HitGroupRecordIndex, /// sbtRecordIndex + Ray.Origin, + Ray.TMin, + Ray.Direction, + Ray.TMax, + __hitObjectAttributesLocation(__hitObjectAttributes<attr_t>())); + + return hitObj; + } + + /// See MakeHit but handles Motion + /// Currently only supported on VK + [ForceInline] + __specialized_for_target(glsl) + static HitObject MakeMotionHit<attr_t>( + uint HitGroupRecordIndex, + RaytracingAccelerationStructure AccelerationStructure, + uint InstanceIndex, + uint GeometryIndex, + uint PrimitiveIndex, + uint HitKind, + RayDesc Ray, + float CurrentTime, + attr_t attributes) + { + HitObject hitObj; + + // Save the attributes + __ref attr_t attr = __hitObjectAttributes<attr_t>(); + attr = attributes; + + __glslMakeMotionHitWithIndex(hitObj, + AccelerationStructure, + InstanceIndex, ///? Same as instanceid ? + GeometryIndex, + PrimitiveIndex, + HitKind, /// Assuming HitKinds are compatible + HitGroupRecordIndex, /// sbtRecordIndex + Ray.Origin, + Ray.TMin, + Ray.Direction, + Ray.TMax, + CurrentTime, + __hitObjectAttributesLocation(__hitObjectAttributes<attr_t>())); + + return hitObj; + } + /// Creates a HitObject representing a miss based on values explicitly passed as arguments, without /// tracing a ray. The provided shader table index must reference a valid miss record in the shader /// table. - __target_intrinsic(hlsl, "NvMakeMiss") [__requiresNVAPI] - static HitObject MakeMiss( - uint MissShaderIndex, + __target_intrinsic(hlsl, "NvMakeMiss") + static HitObject MakeMiss( + uint MissShaderIndex, RayDesc Ray); + [ForceInline] + __specialized_for_target(glsl) + static HitObject MakeMiss( + uint MissShaderIndex, + RayDesc Ray) + { + HitObject hitObj; + __glslMakeMiss(hitObj, MissShaderIndex, Ray.Origin, Ray.TMin, Ray.Direction, Ray.TMax); + return hitObj; + } + + /// See MakeMiss but handles Motion + /// Currently only supported on VK + [ForceInline] + __specialized_for_target(glsl) + static HitObject MakeMotionMiss( + uint MissShaderIndex, + RayDesc Ray, + float CurrentTime) + { + HitObject hitObj; + __glslMakeMotionMiss(hitObj, MissShaderIndex, Ray.Origin, Ray.TMin, Ray.Direction, Ray.TMax, CurrentTime); + return hitObj; + } + /// Creates a HitObject representing “NOP” (no operation) which is neither a hit nor a miss. Invoking a /// NOP hit object using HitObject::Invoke has no effect. Reordering by hit objects using /// ReorderThread will group NOP hit objects together. This can be useful in some reordering /// scenarios where future control flow for some threads is known to process neither a hit nor a /// miss. - __target_intrinsic(hlsl, "NvMakeNop") [__requiresNVAPI] + __target_intrinsic(hlsl, "NvMakeNop") static HitObject MakeNop(); + [ForceInline] + __specialized_for_target(glsl) + static HitObject MakeNop() + { + HitObject hitObj; + __glslMakeNop(hitObj); + return hitObj; + } + /// Invokes closesthit or miss shading for the specified hit object. In case of a NOP HitObject, no /// shader is invoked. - __target_intrinsic(hlsl, "NvInvokeHitObject") [__requiresNVAPI] + __target_intrinsic(hlsl, "NvInvokeHitObject") static void Invoke<payload_t>( RaytracingAccelerationStructure AccelerationStructure, HitObject HitOrMiss, inout payload_t Payload); + __specialized_for_target(glsl) + static void Invoke<payload_t>( + RaytracingAccelerationStructure AccelerationStructure, + HitObject HitOrMiss, + inout payload_t Payload) + { + [__vulkanRayPayload] + static payload_t p; + + // Save the payload + p = Payload; + + __glslInvoke(HitOrMiss, __rayPayloadLocation(p)); + + // Write payload result + Payload = p; + } + /// Returns true if the HitObject encodes a miss, otherwise returns false. - __target_intrinsic(hlsl) [__requiresNVAPI] + __target_intrinsic(hlsl) + __target_intrinsic(glsl, "hitObjectIsMissNV($0)") bool IsMiss(); /// Returns true if the HitObject encodes a hit, otherwise returns false. - __target_intrinsic(hlsl) [__requiresNVAPI] + __target_intrinsic(hlsl) + __target_intrinsic(glsl, "hitObjectIsHitNV($0)") bool IsHit(); /// Returns true if the HitObject encodes a nop, otherwise returns false. - __target_intrinsic(hlsl) [__requiresNVAPI] + __target_intrinsic(hlsl) + __target_intrinsic(glsl, "hitObjectIsEmptyNV($0)") bool IsNop(); /// Queries ray properties from HitObject. Valid if the hit object represents a hit or a miss. - __target_intrinsic(hlsl) [__requiresNVAPI] + __target_intrinsic(hlsl) RayDesc GetRayDesc(); + __specialized_for_target(glsl) + RayDesc GetRayDesc() + { + RayDesc ray = { __glslGetRayWorldOrigin(), __glslGetTMin(), __glslGetRayDirection(), __glslGetTMax() }; + return ray; + } + /// Queries shader table index from HitObject. Valid if the hit object represents a hit or a miss. - __target_intrinsic(hlsl) [__requiresNVAPI] + __target_intrinsic(hlsl) + __target_intrinsic(glsl, "hitObjectGetShaderBindingTableRecordIndexNV($0)") uint GetShaderTableIndex(); /// Returns the instance index of a hit. Valid if the hit object represents a hit. - __target_intrinsic(hlsl) [__requiresNVAPI] + __target_intrinsic(hlsl) + __target_intrinsic(glsl, "hitObjectGetInstanceCustomIndexNV($0)") uint GetInstanceIndex(); /// Returns the instance ID of a hit. Valid if the hit object represents a hit. - __target_intrinsic(hlsl) [__requiresNVAPI] + __target_intrinsic(hlsl) + __target_intrinsic(glsl, "hitObjectGetInstanceIdNV($0)") uint GetInstanceID(); /// Returns the geometry index of a hit. Valid if the hit object represents a hit. - __target_intrinsic(hlsl) [__requiresNVAPI] + __target_intrinsic(hlsl) + __target_intrinsic(glsl, "hitObjectGetGeometryIndexNV($0)") uint GetGeometryIndex(); /// Returns the primitive index of a hit. Valid if the hit object represents a hit. - __target_intrinsic(hlsl) [__requiresNVAPI] + __target_intrinsic(hlsl) + __target_intrinsic(glsl, "hitObjectGetPrimitiveIndexNV($0)") uint GetPrimitiveIndex(); /// Returns the hit kind. Valid if the hit object represents a hit. - __target_intrinsic(hlsl) [__requiresNVAPI] + __target_intrinsic(hlsl) + __target_intrinsic(glsl, "hitObjectGetHitKindNV($0)") uint GetHitKind(); /// Returns the attributes of a hit. Valid if the hit object represents a hit or a miss. @@ -5739,61 +6073,245 @@ struct HitObject attr_t GetAttributes<attr_t>() { attr_t v; - __getAttributesFromHitObject(v); + __hlslGetAttributesFromHitObject(v); return v; } + __specialized_for_target(glsl) + attr_t GetAttributes<attr_t>() + { + // Work out the location + int attributeLocation = __hitObjectAttributesLocation(__hitObjectAttributes<attr_t>()); + + // Load the attributes from the location + __glslGetAttributes(attributeLocation); + + // Return the attributes + return __hitObjectAttributes<attr_t>(); + } /// Loads a root constant from the local root table referenced by the hit object. Valid if the hit object /// represents a hit or a miss. RootConstantOffsetInBytes must be a multiple of 4. __target_intrinsic(hlsl) [__requiresNVAPI] uint LoadLocalRootTableConstant(uint RootConstantOffsetInBytes); - /// - /// !!!! Internal impl. Do not use! - /// + /// + /// !!!! Internal NVAPI HLSL impl. Not part of interface! !!!!!!!!!!!! + /// __target_intrinsic(hlsl, "NvGetAttributesFromHitObject($0, $1)") [__requiresNVAPI] - void __getAttributesFromHitObject<T>(out T t); + void __hlslGetAttributesFromHitObject<T>(out T t); __target_intrinsic(hlsl, "NvMakeHitWithRecordIndex") [__requiresNVAPI] - static void __makeHitWithRecordIndex<attr_t>(uint HitGroupRecordIndex, - RaytracingAccelerationStructure AccelerationStructure, - uint InstanceIndex, - uint GeometryIndex, - uint PrimitiveIndex, - uint HitKind, - RayDesc Ray, - attr_t attributes, + static void __hlslMakeHitWithRecordIndex<attr_t>(uint HitGroupRecordIndex, + RaytracingAccelerationStructure AccelerationStructure, + uint InstanceIndex, + uint GeometryIndex, + uint PrimitiveIndex, + uint HitKind, + RayDesc Ray, + attr_t attributes, out HitObject hitObj); __target_intrinsic(hlsl, "NvMakeHit") [__requiresNVAPI] - static void __makeHit<attr_t>(RaytracingAccelerationStructure AccelerationStructure, - uint InstanceIndex, - uint GeometryIndex, - uint PrimitiveIndex, - uint HitKind, - uint RayContributionToHitGroupIndex, - uint MultiplierForGeometryContributionToHitGroupIndex, - RayDesc Ray, - attr_t attributes, + static void __hlslMakeHit<attr_t>(RaytracingAccelerationStructure AccelerationStructure, + uint InstanceIndex, + uint GeometryIndex, + uint PrimitiveIndex, + uint HitKind, + uint RayContributionToHitGroupIndex, + uint MultiplierForGeometryContributionToHitGroupIndex, + RayDesc Ray, + attr_t attributes, out HitObject hitObj); __target_intrinsic(hlsl, "NvTraceRayHitObject") [__requiresNVAPI] - static void __traceRay<payload_t>( - RaytracingAccelerationStructure AccelerationStructure, - uint RayFlags, - uint InstanceInclusionMask, - uint RayContributionToHitGroupIndex, - uint MultiplierForGeometryContributionToHitGroupIndex, - uint MissShaderIndex, - RayDesc Ray, + static void __hlslTraceRay<payload_t>( + RaytracingAccelerationStructure AccelerationStructure, + uint RayFlags, + uint InstanceInclusionMask, + uint RayContributionToHitGroupIndex, + uint MultiplierForGeometryContributionToHitGroupIndex, + uint MissShaderIndex, + RayDesc Ray, inout payload_t Payload, out HitObject hitObj); + + /// + /// !!!! Internal GLSL GL_NV_shader_invocation_reorder impl. Not part of interface! !!!!!!!!!!!! + /// + + __glsl_extension(GL_NV_shader_invocation_reorder) + __glsl_extension(GL_EXT_ray_tracing) + __target_intrinsic(glsl, "hitObjectRecordMissNV") + static void __glslMakeMiss( + HitObject hitObj, + uint MissShaderIndex, + float3 Origin, + float TMin, + float3 Direction, + float TMax); + + __glsl_extension(GL_NV_shader_invocation_reorder) + __glsl_extension(GL_EXT_ray_tracing) + __glsl_extension(GLSL_NV_ray_tracing_motion_blur) + __target_intrinsic(glsl, "hitObjectRecordMissNV") + static void __glslMakeMotionMiss( + HitObject hitObj, + uint MissShaderIndex, + float3 Origin, + float TMin, + float3 Direction, + float TMax, + float CurrentTime); + + __glsl_extension(GL_NV_shader_invocation_reorder) + __glsl_extension(GL_EXT_ray_tracing) + __target_intrinsic(glsl, "hitObjectRecordEmptyNV($0)") + static void __glslMakeNop(HitObject hitObj); + + __glsl_extension(GL_NV_shader_invocation_reorder) + __target_intrinsic(glsl, "hitObjectGetObjectRayDirectionNV($0)") + float3 __glslGetRayDirection(); + + __glsl_extension(GL_NV_shader_invocation_reorder) + __target_intrinsic(glsl, "hitObjectGetWorldRayOriginNV($0)") + float3 __glslGetRayWorldOrigin(); + + __glsl_extension(GL_NV_shader_invocation_reorder) + __target_intrinsic(glsl, "hitObjectGetRayTMaxNV($0)") + float __glslGetTMax(); + + __glsl_extension(GL_NV_shader_invocation_reorder) + __target_intrinsic(glsl, "hitObjectGetRayTMinNV($0)") + float __glslGetTMin(); + + __glsl_extension(GL_NV_shader_invocation_reorder) + __glsl_extension(GL_EXT_ray_tracing) + __target_intrinsic(glsl, "hitObjectRecordHitWithIndexNV") + static void __glslMakeHitWithIndex( + HitObject hitObj, + RaytracingAccelerationStructure accelerationStructure, + uint instanceid, + uint primitiveid, + uint geometryindex, + uint hitKind, + uint sbtRecordIndex, + float3 origin, + float Tmin, + float3 direction, + float Tmax, + int attributeLocation); + + __glsl_extension(GL_NV_shader_invocation_reorder) + __glsl_extension(GL_EXT_ray_tracing) + __glsl_extension(GLSL_NV_ray_tracing_motion_blur) + __target_intrinsic(glsl, "hitObjectRecordHitWithIndexMotionNV ") + static void __glslMakeMotionHitWithIndex( + HitObject hitObj, + RaytracingAccelerationStructure accelerationStructure, + uint instanceid, + uint primitiveid, + uint geometryindex, + uint hitKind, + uint sbtRecordIndex, + float3 origin, + float Tmin, + float3 direction, + float Tmax, + float CurrentTime, + int attributeLocation); + + __glsl_extension(GL_EXT_ray_tracing) + __glsl_extension(GL_NV_shader_invocation_reorder) + __target_intrinsic(glsl, "hitObjectRecordHitNV") + static void __glslMakeHit( + HitObject hitObj, + RaytracingAccelerationStructure accelerationStructure, + uint instanceid, + uint primitiveid, + uint geometryindex, + uint hitKind, + uint sbtRecordOffset, + uint sbtRecordStride, + float3 origin, + float Tmin, + float3 direction, + float Tmax, + int attributeLocation); + + __glsl_extension(GL_EXT_ray_tracing) + __glsl_extension(GL_NV_shader_invocation_reorder) + __glsl_extension(GLSL_NV_ray_tracing_motion_blur) + __target_intrinsic(glsl, "hitObjectRecordHitMotionNV") + static void __glslMakeMotionHit( + HitObject hitObj, + RaytracingAccelerationStructure accelerationStructure, + uint instanceid, + uint primitiveid, + uint geometryindex, + uint hitKind, + uint sbtRecordOffset, + uint sbtRecordStride, + float3 origin, + float Tmin, + float3 direction, + float Tmax, + float CurrentTime, + int attributeLocation); + + + __glsl_extension(GL_EXT_ray_tracing) + __glsl_extension(GL_NV_shader_invocation_reorder) + __target_intrinsic(glsl, "hitObjectGetAttributesNV($0, $1)") + void __glslGetAttributes(int attributeLocation); + + __glsl_extension(GL_EXT_ray_tracing) + __glsl_extension(GL_NV_shader_invocation_reorder) + __target_intrinsic(glsl, "hitObjectTraceRayNV") + static void __glslTraceRay( + HitObject hitObj, + RaytracingAccelerationStructure accelerationStructure, + uint rayFlags, + uint cullMask, + uint sbtRecordOffset, + uint sbtRecordStride, + uint missIndex, + float3 origin, + float Tmin, + float3 direction, + float Tmax, + int payload); + + __glsl_extension(GL_EXT_ray_tracing) + __glsl_extension(GL_NV_shader_invocation_reorder) + __glsl_extension(GLSL_NV_ray_tracing_motion_blur) + __target_intrinsic(glsl, "hitObjectTraceRayMotionNV") + static void __glslTraceMotionRay( + HitObject hitObj, + RaytracingAccelerationStructure accelerationStructure, + uint rayFlags, + uint cullMask, + uint sbtRecordOffset, + uint sbtRecordStride, + uint missIndex, + float3 origin, + float Tmin, + float3 direction, + float Tmax, + float currentTime, + int payload); + + __glsl_extension(GL_EXT_ray_tracing) + __glsl_extension(GL_NV_shader_invocation_reorder) + __target_intrinsic(glsl, "hitObjectExecuteShaderNV") + static void __glslInvoke( + HitObject hitObj, + int payload); }; /// Reorders threads based on a coherence hint value. NumCoherenceHintBits indicates how many of @@ -5803,8 +6321,11 @@ struct HitObject /// NumCoherenceHintBits. /// Where possible, reordering will also attempt to retain locality in the thread’s launch indices /// (DispatchRaysIndex in DXR). -__target_intrinsic(hlsl, "NvReorderThread") [__requiresNVAPI] +__target_intrinsic(hlsl, "NvReorderThread") +__glsl_extension(GL_NV_shader_invocation_reorder) +__glsl_extension(GL_EXT_ray_tracing) +__target_intrinsic(glsl, "reorderThreadNV") void ReorderThread( uint CoherenceHint, uint NumCoherenceHintBitsFromLSB ); /// Reorders threads based on a hit object, optionally extended by a coherence hint value. Coherence @@ -5822,8 +6343,11 @@ void ReorderThread( uint CoherenceHint, uint NumCoherenceHintBitsFromLSB ); /// same shader ID. (Miss shaders and NOP HitObjects are grouped separately). Within each of these /// groups, it will attempt to order threads by the value of their coherence hints. And within ranges /// of equal coherence hints, it will attempt to maximize locality in 3D space of the ray hit (if any). -__target_intrinsic(hlsl, "NvReorderThread") [__requiresNVAPI] +__target_intrinsic(hlsl, "NvReorderThread") +__glsl_extension(GL_NV_shader_invocation_reorder) +__glsl_extension(GL_EXT_ray_tracing) +__target_intrinsic(glsl, "reorderThreadNV") void ReorderThread( HitObject HitOrMiss, uint CoherenceHint, uint NumCoherenceHintBitsFromLSB ); /// Is equivalent to @@ -5833,4 +6357,6 @@ void ReorderThread( HitObject HitOrMiss, uint CoherenceHint, uint NumCoherenceHi /// With CoherenceHint and NumCoherenceHintBitsFromLSB as 0, meaning they are ignored. [__requiresNVAPI] __target_intrinsic(hlsl, "NvReorderThread") +__glsl_extension(GL_NV_shader_invocation_reorder) +__target_intrinsic(glsl, "reorderThreadNV") void ReorderThread( HitObject HitOrMiss ); diff --git a/source/slang/slang-ast-modifier.h b/source/slang/slang-ast-modifier.h index 8419facce..9146c21d9 100644 --- a/source/slang/slang-ast-modifier.h +++ b/source/slang/slang-ast-modifier.h @@ -829,6 +829,16 @@ class VulkanHitAttributesAttribute : public Attribute SLANG_AST_CLASS(VulkanHitAttributesAttribute) }; +// A `[__vulkanHitObjectAttributes(location)]` attribute, which is used in the +// standard library implementation to indicate that a variable +// actually represents the attributes on a HitObject as part of +// Shader ExecutionReordering +class VulkanHitObjectAttributesAttribute : public Attribute +{ + SLANG_AST_CLASS(VulkanHitObjectAttributesAttribute) + + int location; +}; // A `[mutating]` attribute, which indicates that a member // function is allowed to modify things through its `this` diff --git a/source/slang/slang-check-modifier.cpp b/source/slang/slang-check-modifier.cpp index b8ac21e2d..aa28571a7 100644 --- a/source/slang/slang-check-modifier.cpp +++ b/source/slang/slang-check-modifier.cpp @@ -604,6 +604,15 @@ namespace Slang callablePayloadAttr->location = (int32_t)val->value; } + else if (auto hitObjectAttributesAttr = as<VulkanHitObjectAttributesAttribute>(attr)) + { + SLANG_ASSERT(attr->args.getCount() == 1); + auto val = checkConstantIntVal(attr->args[0]); + + if (!val) return false; + + hitObjectAttributesAttr->location = (int32_t)val->value; + } else if (auto forwardDerivativeAttr = as<ForwardDerivativeAttribute>(attr)) { SLANG_ASSERT(attr->args.getCount() == 1); diff --git a/source/slang/slang-compiler.h b/source/slang/slang-compiler.h index eb6d6c1a5..14ac63531 100755 --- a/source/slang/slang-compiler.h +++ b/source/slang/slang-compiler.h @@ -2267,7 +2267,7 @@ namespace Slang /// ProgramLayout* getOrCreateLayout(DiagnosticSink* sink); - /// Get the layout for the program on the taarget. + /// Get the layout for the program on the target. /// /// This routine assumes that `getOrCreateLayout` /// has already been called previously. diff --git a/source/slang/slang-emit-c-like.cpp b/source/slang/slang-emit-c-like.cpp index 09a18a31c..e872bffb1 100644 --- a/source/slang/slang-emit-c-like.cpp +++ b/source/slang/slang-emit-c-like.cpp @@ -41,6 +41,56 @@ struct CLikeSourceEmitter::ComputeEmitActionsContext List<EmitAction>* actions; }; +/* !!!!!!!!!!!!!!!!!!!!!!!!!!!! LocationTracker !!!!!!!!!!!!!!!!!!!!!!!!!! */ + +/* static */LocationTracker::Kind LocationTracker::getKindFromDecoration(IRDecoration* decoration) +{ + switch (decoration->getOp()) + { + case kIROp_VulkanRayPayloadDecoration: return Kind::RayPayload; + case kIROp_VulkanCallablePayloadDecoration: return Kind::CallablePayload; + case kIROp_VulkanHitObjectAttributesDecoration: return Kind::HitObjectAttribute; + default: break; + } + return Kind::Invalid; +} + +Index LocationTracker::getValue(IRInst* inst, IRDecoration* decoration) +{ + const Kind kind = getKindFromDecoration(decoration); + SLANG_RELEASE_ASSERT(kind != Kind::Invalid); + if (kind == Kind::Invalid) + { + return -1; + } + + return getValue(kind, inst, decoration); +} + +Index LocationTracker::getValue(Kind kind, IRInst* inst, IRDecoration* decoration) +{ + if (decoration->getOperandCount() > 0) + { + // TODO(JS): + // There could be a clash with the auto generated location, and the user set value/ + // Perhaps the implication in practice is that either all are marked or none. + const int explicitLocation = int(getIntVal(decoration->getOperand(0))); + if (explicitLocation >= 0) + return UInt(explicitLocation); + } + + auto& nextValue = m_nextValueForKind[Index(kind)]; + + const Location defaultLocation{kind, nextValue}; + const Location foundLocation = m_mapIRToLocations.GetOrAddValue(inst, defaultLocation); + + // Increase if it was the default + nextValue += Index(defaultLocation == foundLocation); + + // Has to match the kind + return (foundLocation.kind == kind) ? foundLocation.value : -1; +} + /* !!!!!!!!!!!!!!!!!!!!!!!!!!!! CLikeSourceEmitter !!!!!!!!!!!!!!!!!!!!!!!!!! */ /* static */SourceLanguage CLikeSourceEmitter::getSourceLanguage(CodeGenTarget target) @@ -3009,46 +3059,6 @@ void CLikeSourceEmitter::emitInterpolationModifiers(IRInst* varInst, IRType* val emitInterpolationModifiersImpl(varInst, valueType, layout); } -UInt CLikeSourceEmitter::getRayPayloadLocation(IRInst* inst) -{ - if (auto rayPayloadDecoration = inst->findDecoration<IRVulkanRayPayloadDecoration>()) - { - int explicitLocation = int(getIntVal(rayPayloadDecoration->getOperand(0))); - - if (explicitLocation >= 0) - return UInt(explicitLocation); - } - - auto& map = m_mapIRValueToRayPayloadLocation; - UInt value = 0; - if(map.TryGetValue(inst, value)) - return value; - - value = map.Count(); - map.Add(inst, value); - return value; -} - -UInt CLikeSourceEmitter::getCallablePayloadLocation(IRInst* inst) -{ - if (auto callablePayloadDecoration = inst->findDecoration<IRVulkanCallablePayloadDecoration>()) - { - int explicitLocation = int(getIntVal(callablePayloadDecoration->getOperand(0))); - - if (explicitLocation >= 0) - return UInt(explicitLocation); - } - - auto& map = m_mapIRValueToCallablePayloadLocation; - UInt value = 0; - if(map.TryGetValue(inst, value)) - return value; - - value = map.Count(); - map.Add(inst, value); - return value; -} - /// Emit modifiers that should apply even for a declaration of an SSA temporary. void CLikeSourceEmitter::emitTempModifiers(IRInst* temp) { diff --git a/source/slang/slang-emit-c-like.h b/source/slang/slang-emit-c-like.h index d85e336e0..e702dbf01 100644 --- a/source/slang/slang-emit-c-like.h +++ b/source/slang/slang-emit-c-like.h @@ -17,6 +17,48 @@ namespace Slang { +class LocationTracker +{ +public: + enum class Kind + { + Invalid = -1, + RayPayload, ///< GLSL rayPayload + CallablePayload, ///< GLSL callableData + HitObjectAttribute, ///< GLSL hitObjectAttribute + CountOf, + }; + + /// Given a decoration returns the Kind, or Kind::Invalid if that is not appropriate + static Kind getKindFromDecoration(IRDecoration* decoration); + + /// Get the location value associated with inst (and decoration). + /// Will return -1, if no location is associated + Index getValue(IRInst* inst, IRDecoration* decoration); + + /// Get the location value associated with inst (and decoration). + /// The kind must match that for the decoration. + /// Will return -1, if no location is associated + Index getValue(Kind kind, IRInst* inst, IRDecoration* decoration); + +protected: + struct Location + { + typedef Location ThisType; + + bool operator==(const ThisType& rhs) const { return kind == rhs.kind && value == rhs.value; } + bool operator!=(const ThisType& rhs) const { return !(*this == rhs); } + + Kind kind; ///< The kind of location + Index value; ///< The value of the location. Must be >= 0 + }; + + Index m_nextValueForKind[Count(Kind::CountOf)] = { 0, }; + + Dictionary<IRInst*, Location> m_mapIRToLocations; +}; + + class CLikeSourceEmitter: public SourceEmitterBase { public: @@ -201,6 +243,8 @@ public: {} }; + + /// Must be called before used virtual SlangResult init(); @@ -233,6 +277,8 @@ public: ComponentType* getProgram() { return m_codeGenContext->getProgram(); } TargetProgram* getTargetProgram() { return m_codeGenContext->getTargetProgram(); } + LocationTracker& getLocationTracker() { return m_locationTracker; } + // // Types // @@ -357,16 +403,12 @@ public: void emitStruct(IRStructType* structType); void emitClass(IRClassType* structType); - - /// Emit type attributes that should appear after, e.g., a `struct` keyword void emitPostKeywordTypeAttributes(IRInst* inst) { emitPostKeywordTypeAttributesImpl(inst); } void emitInterpolationModifiers(IRInst* varInst, IRType* valueType, IRVarLayout* layout); - UInt getRayPayloadLocation(IRInst* inst); - - UInt getCallablePayloadLocation(IRInst* inst); + /// Emit modifiers that should apply even for a declaration of an SSA temporary. virtual void emitTempModifiers(IRInst* temp); @@ -540,8 +582,9 @@ public: // to use for it when emitting code. Dictionary<IRInst*, String> m_mapInstToName; - Dictionary<IRInst*, UInt> m_mapIRValueToRayPayloadLocation; - Dictionary<IRInst*, UInt> m_mapIRValueToCallablePayloadLocation; + // Maps instructions to locations. Used for GLSL output for locations, but could potentially + // be used for other kinds of location. + LocationTracker m_locationTracker; }; } diff --git a/source/slang/slang-emit-glsl.cpp b/source/slang/slang-emit-glsl.cpp index 48dcd21f7..e187f5e59 100644 --- a/source/slang/slang-emit-glsl.cpp +++ b/source/slang/slang-emit-glsl.cpp @@ -2240,45 +2240,62 @@ void GLSLSourceEmitter::emitVarDecorationsImpl(IRInst* varDecl) // the payload won't automatically get a layout applied // (it isn't part of the user-visible interface...) // - if (varDecl->findDecoration<IRVulkanRayPayloadDecoration>()) - { - m_writer->emit("layout(location = "); - m_writer->emit(getRayPayloadLocation(varDecl)); - m_writer->emit(")\n"); - if( getTargetCaps().implies(CapabilityAtom::GL_NV_ray_tracing) ) - { - m_writer->emit("rayPayloadNV\n"); - } - else - { - m_writer->emit("rayPayloadEXT\n"); - } - } - if (varDecl->findDecoration<IRVulkanCallablePayloadDecoration>()) + + for (auto decoration : varDecl->getDecorations()) { - m_writer->emit("layout(location = "); - m_writer->emit(getCallablePayloadLocation(varDecl)); - m_writer->emit(")\n"); - if( getTargetCaps().implies(CapabilityAtom::GL_NV_ray_tracing) ) + typedef LocationTracker::Kind LocationKind; + + LocationKind locationKind = LocationKind::Invalid; + UnownedStringSlice prefix; + if (as<IRVulkanHitAttributesDecoration>(decoration)) { - m_writer->emit("callableDataNV\n"); + prefix = toSlice("hitAttribute"); } else { - m_writer->emit("callableDataEXT\n"); + // Handle attributes that have location + const LocationKind decorationLocationKind = LocationTracker::getKindFromDecoration(decoration); + if (decorationLocationKind == LocationKind::Invalid) + { + // Next decoration + continue; + } + + locationKind = decorationLocationKind; + + // Get the location value + const auto locationValue = m_locationTracker.getValue(locationKind, varDecl, decoration); + + m_writer->emit(toSlice("layout(location = ")); + m_writer->emit(locationValue); + m_writer->emit(toSlice(")\n")); + + switch (locationKind) + { + case LocationKind::CallablePayload: prefix = toSlice("callableData"); break; + case LocationKind::HitObjectAttribute: prefix = toSlice("hitObjectAttribute"); break; + case LocationKind::RayPayload: prefix = toSlice("rayPayload"); break; + default: break; + } } - } - if (varDecl->findDecoration<IRVulkanHitAttributesDecoration>()) - { - if( getTargetCaps().implies(CapabilityAtom::GL_NV_ray_tracing) ) + SLANG_ASSERT(prefix.getLength()); + m_writer->emit(prefix); + + // Special case hitObjectAttribute as is only NV currently + if (locationKind == LocationKind::HitObjectAttribute || + getTargetCaps().implies(CapabilityAtom::GL_NV_ray_tracing)) { - m_writer->emit("hitAttributeNV\n"); + m_writer->emit(toSlice("NV")); } else { - m_writer->emit("hitAttributeEXT\n"); + m_writer->emit(toSlice("EXT")); } + m_writer->emit(toSlice("\n")); + + // If we emit a location we are done. + break; } if (varDecl->findDecoration<IRGloballyCoherentDecoration>()) diff --git a/source/slang/slang-intrinsic-expand.cpp b/source/slang/slang-intrinsic-expand.cpp index 48186b94a..64ef4e761 100644 --- a/source/slang/slang-intrinsic-expand.cpp +++ b/source/slang/slang-intrinsic-expand.cpp @@ -710,46 +710,53 @@ const char* IntrinsicExpandContext::_emitSpecial(const char* cursor) // shaders. case 'X': { + typedef LocationTracker::Kind LocationKind; + SLANG_RELEASE_ASSERT(*cursor); - switch (*cursor++) + const auto kindChar = *cursor++; + + LocationKind kind = LocationKind::Invalid; + + // The `$XP`/`$XC`/`$XH` case handles looking up + // the associated `location` for a variable + // used as the argument. + switch (kindChar) { - case 'P': - { - // The `$XP` case handles looking up - // the associated `location` for a variable - // used as the argument ray payload at a - // trace call site. - - Index argIndex = 0; - SLANG_RELEASE_ASSERT(m_argCount > argIndex); - auto arg = m_args[argIndex].get(); - auto argLoad = as<IRLoad>(arg); - SLANG_RELEASE_ASSERT(argLoad); - auto argVar = argLoad->getOperand(0); - m_writer->emit(m_emitter->getRayPayloadLocation(argVar)); - } - break; + case 'P': kind = LocationKind::RayPayload; break; + case 'C': kind = LocationKind::CallablePayload; break; + case 'H': kind = LocationKind::HitObjectAttribute; break; + default: break; + } - case 'C': + SLANG_ASSERT(kind != LocationKind::Invalid); + + if (kind != LocationKind::Invalid) + { + Index argIndex = 0; + SLANG_RELEASE_ASSERT(m_argCount > argIndex); + auto arg = m_args[argIndex].get(); + auto argLoad = as<IRLoad>(arg); + SLANG_RELEASE_ASSERT(argLoad); + + auto argVar = argLoad->getOperand(0); + + // Find the associated decoration + IRDecoration* foundDecoration = nullptr; + for (auto decoration : argVar->getDecorations()) { - // The `$XC` case handles looking up - // the associated `location` for a variable - // used as the argument callable payload at a - // call site. - - Index argIndex = 0; - SLANG_RELEASE_ASSERT(m_argCount > argIndex); - auto arg = m_args[argIndex].get(); - auto argLoad = as<IRLoad>(arg); - SLANG_RELEASE_ASSERT(argLoad); - auto argVar = argLoad->getOperand(0); - m_writer->emit(m_emitter->getCallablePayloadLocation(argVar)); + const auto curKind = LocationTracker::getKindFromDecoration(decoration); + if (curKind == kind) + { + foundDecoration = decoration; + break; + } } - break; - default: - SLANG_RELEASE_ASSERT(false); - break; + // Must have found the decoration + SLANG_ASSERT(foundDecoration); + + const auto location = m_emitter->getLocationTracker().getValue(kind, argVar, foundDecoration); + m_writer->emit(location); } } break; diff --git a/source/slang/slang-ir-inst-defs.h b/source/slang/slang-ir-inst-defs.h index 51811f59e..9f3111b87 100644 --- a/source/slang/slang-ir-inst-defs.h +++ b/source/slang/slang-ir-inst-defs.h @@ -586,6 +586,8 @@ INST(HighLevelDeclDecoration, highLevelDecl, 1, 0) INST(VulkanRayPayloadDecoration, vulkanRayPayload, 0, 0) INST(VulkanHitAttributesDecoration, vulkanHitAttributes, 0, 0) + INST(VulkanHitObjectAttributesDecoration, vulkanHitObjectAttributes, 0, 0) + INST(RequireSPIRVVersionDecoration, requireSPIRVVersion, 1, 0) INST(RequireGLSLVersionDecoration, requireGLSLVersion, 1, 0) INST(RequireGLSLExtensionDecoration, requireGLSLExtension, 1, 0) diff --git a/source/slang/slang-ir-insts.h b/source/slang/slang-ir-insts.h index 5eea12de8..f5c3d10ae 100644 --- a/source/slang/slang-ir-insts.h +++ b/source/slang/slang-ir-insts.h @@ -239,6 +239,11 @@ IR_SIMPLE_DECORATION(VulkanCallablePayloadDecoration) /// to it. IR_SIMPLE_DECORATION(VulkanHitAttributesDecoration) +/// A decoration that indicates that a variable represents +/// vulkan hit object attributes, and should have a location assigned +/// to it. +IR_SIMPLE_DECORATION(VulkanHitObjectAttributesDecoration) + struct IRRequireGLSLVersionDecoration : IRDecoration { enum { kOp = kIROp_RequireGLSLVersionDecoration }; @@ -3358,6 +3363,12 @@ public: { addDecoration(inst, kIROp_VulkanCallablePayloadDecoration, getIntValue(getIntType(), location)); } + + void addVulkanHitObjectAttributesDecoration(IRInst* inst, int location) + { + addDecoration(inst, kIROp_VulkanHitObjectAttributesDecoration, getIntValue(getIntType(), location)); + } + }; void addHoistableInst( diff --git a/source/slang/slang-ir.cpp b/source/slang/slang-ir.cpp index 6112beaf2..7aa51f31b 100644 --- a/source/slang/slang-ir.cpp +++ b/source/slang/slang-ir.cpp @@ -48,6 +48,7 @@ namespace Slang case kIROp_VulkanCallablePayloadDecoration: case kIROp_VulkanHitAttributesDecoration: case kIROp_VulkanRayPayloadDecoration: + case kIROp_VulkanHitObjectAttributesDecoration: { return true; } diff --git a/source/slang/slang-lower-to-ir.cpp b/source/slang/slang-lower-to-ir.cpp index a0158cf38..cf3781415 100644 --- a/source/slang/slang-lower-to-ir.cpp +++ b/source/slang/slang-lower-to-ir.cpp @@ -2115,7 +2115,11 @@ void addVarDecorations( { builder->addVulkanCallablePayloadDecoration(inst, callablePayloadAttr->location); } - else if(as<VulkanHitAttributesAttribute>(mod)) + else if (auto hitObjectAttr = as<VulkanHitObjectAttributesAttribute>(mod)) + { + builder->addVulkanHitObjectAttributesDecoration(inst, hitObjectAttr->location); + } + else if (as<VulkanHitAttributesAttribute>(mod)) { builder->addSimpleDecoration<IRVulkanHitAttributesDecoration>(inst); } diff --git a/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-array.slang b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-array.slang new file mode 100644 index 000000000..79b8411b0 --- /dev/null +++ b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-array.slang @@ -0,0 +1,100 @@ +// hit-object-array.slang + +//TEST:SIMPLE: -target dxil -entry rayGenerationMain -stage raygeneration -profile sm_6_5 -DNV_SHADER_EXTN_SLOT=u0 + +/* This doesn't work correctly currently because it produces assignments to the HitObject array in GLSL output. +But that is not valid for hitObjectNV int he extension */ + +//DISABLE_TEST:SIMPLE: -target glsl -entry rayGenerationMain -stage raygeneration -profile sm_6_5 -line-directive-mode none + +//DISABLE_TEST(compute):COMPARE_COMPUTE:-d3d12 -output-using-type -use-dxil -profile sm_6_5 -render-feature ray-query +//DISABLE_TEST(compute):COMPARE_COMPUTE:-vk -output-using-type -render-feature ray-query + +//TEST_INPUT: set scene = AccelerationStructure +uniform RaytracingAccelerationStructure scene; + +//TEST_INPUT:set outputBuffer = out ubuffer(data=[0, 0, 0, 0], stride=4) +RWStructuredBuffer<uint> outputBuffer; + +struct SomeValues +{ + int a; + float b; +}; + +uint calcValue(HitObject hit) +{ + uint r = 0; + + if (hit.IsHit()) + { + uint instanceIndex = hit.GetInstanceIndex(); + uint instanceID = hit.GetInstanceID(); + uint geometryIndex = hit.GetGeometryIndex(); + uint primitiveIndex = hit.GetPrimitiveIndex(); + uint hitKind = hit.GetHitKind(); + + r += hitKind; + r += instanceIndex; + r += instanceID; + r += geometryIndex; + r += primitiveIndex; + + RayDesc ray = hit.GetRayDesc(); + + r += uint(ray.TMin > 0); + r += uint(ray.TMax < ray.TMin); + + SomeValues objSomeValues = hit.GetAttributes<SomeValues>(); + + r += objSomeValues.a; + } + else if (hit.IsMiss()) + { + r += 1; + } + + return r; +} + +void rayGenerationMain() +{ + int2 launchID = int2(DispatchRaysIndex().xy); + int2 launchSize = int2(DispatchRaysDimensions().xy); + + int idx = launchID.x; + + SomeValues someValues = { idx, idx * 2.0f }; + + RayDesc ray; + ray.Origin = float3(idx, 0, 0); + ray.TMin = 0.01f; + ray.Direction = float3(0, 1, 0); + ray.TMax = 1e4f; + + uint hitKind = 0; + + HitObject hitObjs[2]; + + hitObjs[0] = HitObject::MakeHit(0, scene, idx, idx * 2, idx * 3, hitKind, ray, someValues); + + { + int rayContributionToHitGroupIndex = 0; + int multiplierForGeometryContributionToHitGroupIndex = 4; + + hitObjs[1] = HitObject::MakeHit(scene, + idx, + idx * 2, + idx * 3, + hitKind, + rayContributionToHitGroupIndex, + multiplierForGeometryContributionToHitGroupIndex, + ray, + someValues); + } + + + uint r = calcValue(hitObjs[0]) + calcValue(hitObjs[1]); + + outputBuffer[idx] = r; +} diff --git a/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-array.slang.expected b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-array.slang.expected new file mode 100644 index 000000000..fdc5cf0d7 --- /dev/null +++ b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-array.slang.expected @@ -0,0 +1,439 @@ +result code = 0 +standard error = { +} +standard output = { +; +; Note: shader requires additional functionality: +; UAVs at every shader stage +; +; shader hash: 1376fe501fef02e2587aacc2b18e252f +; +; Buffer Definitions: +; +; Resource bind info for g_NvidiaExt +; { +; +; struct struct.NvShaderExtnStruct +; { +; +; uint opcode; ; Offset: 0 +; uint rid; ; Offset: 4 +; uint sid; ; Offset: 8 +; uint4 dst1u; ; Offset: 12 +; uint4 src3u; ; Offset: 28 +; uint4 src4u; ; Offset: 44 +; uint4 src5u; ; Offset: 60 +; uint4 src0u; ; Offset: 76 +; uint4 src1u; ; Offset: 92 +; uint4 src2u; ; Offset: 108 +; uint4 dst0u; ; Offset: 124 +; uint markUavRef; ; Offset: 140 +; uint numOutputsForIncCounter; ; Offset: 144 +; float padding1[27]; ; Offset: 148 +; +; } $Element; ; Offset: 0 Size: 256 +; +; } +; +; Resource bind info for outputBuffer_0 +; { +; +; uint $Element; ; Offset: 0 Size: 4 +; +; } +; +; +; Resource Bindings: +; +; Name Type Format Dim ID HLSL Bind Count +; ------------------------------ ---------- ------- ----------- ------- -------------- ------ +; scene_0 texture i32 ras T0 t0 1 +; g_NvidiaExt UAV struct r/w+cnt U0 u0 1 +; outputBuffer_0 UAV struct r/w U1 u1 1 +; +target datalayout = "e-m:e-p:32:32-i1:32-i8:8-i16:16-i32:32-i64:64-f16:16-f32:32-f64:64-n8:16:32:64" +target triple = "dxil-ms-dx" + +%"class.RWStructuredBuffer<NvShaderExtnStruct>" = type { %struct.NvShaderExtnStruct } +%struct.NvShaderExtnStruct = type { i32, i32, i32, <4 x i32>, <4 x i32>, <4 x i32>, <4 x i32>, <4 x i32>, <4 x i32>, <4 x i32>, <4 x i32>, i32, i32, [27 x float] } +%struct.RaytracingAccelerationStructure = type { i32 } +%"class.RWStructuredBuffer<unsigned int>" = type { i32 } +%struct.SomeValues_0 = type { i32, float } +%struct.NvHitObjectMacroDummyPayloadType = type { i32 } +%dx.types.Handle = type { i8* } + +@"\01?g_NvidiaExt@@3V?$RWStructuredBuffer@UNvShaderExtnStruct@@@@A" = external constant %"class.RWStructuredBuffer<NvShaderExtnStruct>", align 4 +@"\01?scene_0@@3URaytracingAccelerationStructure@@A" = external constant %struct.RaytracingAccelerationStructure, align 4 +@"\01?outputBuffer_0@@3V?$RWStructuredBuffer@I@@A" = external constant %"class.RWStructuredBuffer<unsigned int>", align 4 + +; Function Attrs: nounwind +define void @"\01?rayGenerationMain@@YAXXZ"() #0 { + %1 = load %struct.RaytracingAccelerationStructure, %struct.RaytracingAccelerationStructure* @"\01?scene_0@@3URaytracingAccelerationStructure@@A", align 4, !noalias !18 + %2 = load %"class.RWStructuredBuffer<unsigned int>", %"class.RWStructuredBuffer<unsigned int>"* @"\01?outputBuffer_0@@3V?$RWStructuredBuffer@I@@A", align 4 + %3 = load %"class.RWStructuredBuffer<NvShaderExtnStruct>", %"class.RWStructuredBuffer<NvShaderExtnStruct>"* @"\01?g_NvidiaExt@@3V?$RWStructuredBuffer@UNvShaderExtnStruct@@@@A", align 4, !noalias !21 + %4 = alloca %struct.SomeValues_0, align 8 + %5 = alloca %struct.SomeValues_0, align 8 + %6 = alloca %struct.SomeValues_0, align 8 + %7 = alloca %struct.NvHitObjectMacroDummyPayloadType, align 4 + %8 = alloca %struct.SomeValues_0, align 8 + %9 = alloca %struct.NvHitObjectMacroDummyPayloadType, align 4 + %10 = call i32 @dx.op.dispatchRaysIndex.i32(i32 145, i8 0) ; DispatchRaysIndex(col) + %11 = sitofp i32 %10 to float + %12 = fmul fast float %11, 2.000000e+00 + %13 = sitofp i32 %10 to float + %14 = mul nsw i32 %10, 3 + %15 = shl nsw i32 %10, 1 + %16 = getelementptr inbounds %struct.SomeValues_0, %struct.SomeValues_0* %8, i32 0, i32 0 + store i32 %10, i32* %16, align 8 + %17 = getelementptr inbounds %struct.SomeValues_0, %struct.SomeValues_0* %8, i32 0, i32 1 + store float %12, float* %17, align 4 + %18 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %19 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %18, i8 1) ; BufferUpdateCounter(uav,inc) + %20 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %20, i32 %19, i32 0, i32 69, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %21 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %21, i32 %19, i32 144, i32 2, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %22 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %22, i32 %19, i32 76, i32 %10, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %23 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %23, i32 %19, i32 80, i32 %15, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %24 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %24, i32 %19, i32 84, i32 %14, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %25 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %25, i32 %19, i32 88, i32 0, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %26 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %26, i32 %19, i32 92, i32 0, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %27 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %28 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %27, i8 1) ; BufferUpdateCounter(uav,inc) + %29 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %30 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %29, i8 1) ; BufferUpdateCounter(uav,inc) + call void @dx.op.callShader.struct.SomeValues_0(i32 159, i32 %30, %struct.SomeValues_0* nonnull %8) ; CallShader(ShaderIndex,Parameter) + %31 = call %dx.types.Handle @dx.op.createHandleForLib.struct.RaytracingAccelerationStructure(i32 160, %struct.RaytracingAccelerationStructure %1) ; CreateHandleForLib(Resource) + call void @dx.op.traceRay.struct.NvHitObjectMacroDummyPayloadType(i32 157, %dx.types.Handle %31, i32 0, i32 0, i32 0, i32 0, i32 %30, float %13, float 0.000000e+00, float 0.000000e+00, float 0x3F847AE140000000, float 0.000000e+00, float 1.000000e+00, float 0.000000e+00, float 1.000000e+04, %struct.NvHitObjectMacroDummyPayloadType* nonnull %9) ; TraceRay(AccelerationStructure,RayFlags,InstanceInclusionMask,RayContributionToHitGroupIndex,MultiplierForGeometryContributionToShaderIndex,MissShaderIndex,Origin_X,Origin_Y,Origin_Z,TMin,Direction_X,Direction_Y,Direction_Z,TMax,payload) + %32 = mul nsw i32 %10, 3 + %33 = shl nsw i32 %10, 1 + %34 = getelementptr inbounds %struct.SomeValues_0, %struct.SomeValues_0* %6, i32 0, i32 0 + store i32 %10, i32* %34, align 8 + %35 = getelementptr inbounds %struct.SomeValues_0, %struct.SomeValues_0* %6, i32 0, i32 1 + store float %12, float* %35, align 4 + %36 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %37 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %36, i8 1) ; BufferUpdateCounter(uav,inc) + %38 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %38, i32 %37, i32 0, i32 68, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %39 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %39, i32 %37, i32 144, i32 2, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %40 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %40, i32 %37, i32 76, i32 %10, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %41 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %41, i32 %37, i32 80, i32 %33, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %42 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %42, i32 %37, i32 84, i32 %32, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %43 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %43, i32 %37, i32 88, i32 0, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %44 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %44, i32 %37, i32 92, i32 0, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %45 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %45, i32 %37, i32 96, i32 4, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %46 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %47 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %46, i8 1) ; BufferUpdateCounter(uav,inc) + %48 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %49 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %48, i8 1) ; BufferUpdateCounter(uav,inc) + call void @dx.op.callShader.struct.SomeValues_0(i32 159, i32 %49, %struct.SomeValues_0* nonnull %6) ; CallShader(ShaderIndex,Parameter) + %50 = call %dx.types.Handle @dx.op.createHandleForLib.struct.RaytracingAccelerationStructure(i32 160, %struct.RaytracingAccelerationStructure %1) ; CreateHandleForLib(Resource) + call void @dx.op.traceRay.struct.NvHitObjectMacroDummyPayloadType(i32 157, %dx.types.Handle %50, i32 0, i32 0, i32 0, i32 0, i32 %49, float %13, float 0.000000e+00, float 0.000000e+00, float 0x3F847AE140000000, float 0.000000e+00, float 1.000000e+00, float 0.000000e+00, float 1.000000e+04, %struct.NvHitObjectMacroDummyPayloadType* nonnull %7) ; TraceRay(AccelerationStructure,RayFlags,InstanceInclusionMask,RayContributionToHitGroupIndex,MultiplierForGeometryContributionToShaderIndex,MissShaderIndex,Origin_X,Origin_Y,Origin_Z,TMin,Direction_X,Direction_Y,Direction_Z,TMax,payload) + %51 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %52 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %51, i8 1) ; BufferUpdateCounter(uav,inc) + %53 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %53, i32 %52, i32 0, i32 83, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %54 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %54, i32 %52, i32 76, i32 %28, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %55 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %56 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %55, i8 1) ; BufferUpdateCounter(uav,inc) + %57 = icmp eq i32 %56, 0 + br i1 %57, label %130, label %58 + +; <label>:58 ; preds = %0 + %59 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %60 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %59, i8 1) ; BufferUpdateCounter(uav,inc) + %61 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %61, i32 %60, i32 0, i32 75, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %62 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %62, i32 %60, i32 76, i32 %28, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %63 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %64 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %63, i8 1) ; BufferUpdateCounter(uav,inc) + %65 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %66 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %65, i8 1) ; BufferUpdateCounter(uav,inc) + %67 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %67, i32 %66, i32 0, i32 74, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %68 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %68, i32 %66, i32 76, i32 %28, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %69 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %70 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %69, i8 1) ; BufferUpdateCounter(uav,inc) + %71 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %72 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %71, i8 1) ; BufferUpdateCounter(uav,inc) + %73 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %73, i32 %72, i32 0, i32 77, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %74 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %74, i32 %72, i32 76, i32 %28, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %75 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %76 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %75, i8 1) ; BufferUpdateCounter(uav,inc) + %77 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %78 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %77, i8 1) ; BufferUpdateCounter(uav,inc) + %79 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %79, i32 %78, i32 0, i32 76, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %80 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %80, i32 %78, i32 76, i32 %28, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %81 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %82 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %81, i8 1) ; BufferUpdateCounter(uav,inc) + %83 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %84 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %83, i8 1) ; BufferUpdateCounter(uav,inc) + %85 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %85, i32 %84, i32 0, i32 78, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %86 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %86, i32 %84, i32 76, i32 %28, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %87 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %88 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %87, i8 1) ; BufferUpdateCounter(uav,inc) + %89 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %90 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %89, i8 1) ; BufferUpdateCounter(uav,inc) + %91 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %91, i32 %90, i32 0, i32 79, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %92 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %92, i32 %90, i32 76, i32 %28, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %93 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %94 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %93, i8 1) ; BufferUpdateCounter(uav,inc) + %95 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %96 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %95, i8 1) ; BufferUpdateCounter(uav,inc) + %97 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %98 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %97, i8 1) ; BufferUpdateCounter(uav,inc) + %99 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %100 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %99, i8 1) ; BufferUpdateCounter(uav,inc) + %101 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %102 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %101, i8 1) ; BufferUpdateCounter(uav,inc) + %103 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %104 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %103, i8 1) ; BufferUpdateCounter(uav,inc) + %105 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %106 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %105, i8 1) ; BufferUpdateCounter(uav,inc) + %107 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %108 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %107, i8 1) ; BufferUpdateCounter(uav,inc) + %109 = bitcast i32 %94 to float + %110 = bitcast i32 %96 to float + %111 = fcmp fast ogt float %109, 0.000000e+00 + %112 = zext i1 %111 to i32 + %113 = fcmp fast olt float %110, %109 + %114 = zext i1 %113 to i32 + %115 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %116 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %115, i8 1) ; BufferUpdateCounter(uav,inc) + %117 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %117, i32 %116, i32 0, i32 80, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %118 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %118, i32 %116, i32 76, i32 %28, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %119 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %120 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %119, i8 1) ; BufferUpdateCounter(uav,inc) + call void @dx.op.callShader.struct.SomeValues_0(i32 159, i32 %120, %struct.SomeValues_0* nonnull %5) ; CallShader(ShaderIndex,Parameter) + %121 = getelementptr inbounds %struct.SomeValues_0, %struct.SomeValues_0* %5, i32 0, i32 0 + %122 = load i32, i32* %121, align 8 + %123 = add i32 %70, %64 + %124 = add i32 %123, %76 + %125 = add i32 %124, %82 + %126 = add i32 %125, %88 + %127 = add i32 %126, %112 + %128 = add i32 %127, %114 + %129 = add i32 %128, %122 + br label %139 + +; <label>:130 ; preds = %0 + %131 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %132 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %131, i8 1) ; BufferUpdateCounter(uav,inc) + %133 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %133, i32 %132, i32 0, i32 73, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %134 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %134, i32 %132, i32 76, i32 %28, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %135 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %136 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %135, i8 1) ; BufferUpdateCounter(uav,inc) + %137 = icmp ne i32 %136, 0 + %138 = zext i1 %137 to i32 + br label %139 + +; <label>:139 ; preds = %130, %58 + %140 = phi i32 [ %129, %58 ], [ %138, %130 ] + %141 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %142 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %141, i8 1) ; BufferUpdateCounter(uav,inc) + %143 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %143, i32 %142, i32 0, i32 83, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %144 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %144, i32 %142, i32 76, i32 %47, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %145 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %146 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %145, i8 1) ; BufferUpdateCounter(uav,inc) + %147 = icmp eq i32 %146, 0 + br i1 %147, label %220, label %148 + +; <label>:148 ; preds = %139 + %149 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %150 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %149, i8 1) ; BufferUpdateCounter(uav,inc) + %151 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %151, i32 %150, i32 0, i32 75, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %152 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %152, i32 %150, i32 76, i32 %47, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %153 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %154 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %153, i8 1) ; BufferUpdateCounter(uav,inc) + %155 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %156 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %155, i8 1) ; BufferUpdateCounter(uav,inc) + %157 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %157, i32 %156, i32 0, i32 74, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %158 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %158, i32 %156, i32 76, i32 %47, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %159 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %160 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %159, i8 1) ; BufferUpdateCounter(uav,inc) + %161 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %162 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %161, i8 1) ; BufferUpdateCounter(uav,inc) + %163 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %163, i32 %162, i32 0, i32 77, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %164 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %164, i32 %162, i32 76, i32 %47, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %165 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %166 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %165, i8 1) ; BufferUpdateCounter(uav,inc) + %167 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %168 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %167, i8 1) ; BufferUpdateCounter(uav,inc) + %169 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %169, i32 %168, i32 0, i32 76, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %170 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %170, i32 %168, i32 76, i32 %47, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %171 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %172 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %171, i8 1) ; BufferUpdateCounter(uav,inc) + %173 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %174 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %173, i8 1) ; BufferUpdateCounter(uav,inc) + %175 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %175, i32 %174, i32 0, i32 78, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %176 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %176, i32 %174, i32 76, i32 %47, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %177 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %178 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %177, i8 1) ; BufferUpdateCounter(uav,inc) + %179 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %180 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %179, i8 1) ; BufferUpdateCounter(uav,inc) + %181 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %181, i32 %180, i32 0, i32 79, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %182 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %182, i32 %180, i32 76, i32 %47, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %183 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %184 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %183, i8 1) ; BufferUpdateCounter(uav,inc) + %185 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %186 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %185, i8 1) ; BufferUpdateCounter(uav,inc) + %187 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %188 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %187, i8 1) ; BufferUpdateCounter(uav,inc) + %189 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %190 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %189, i8 1) ; BufferUpdateCounter(uav,inc) + %191 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %192 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %191, i8 1) ; BufferUpdateCounter(uav,inc) + %193 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %194 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %193, i8 1) ; BufferUpdateCounter(uav,inc) + %195 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %196 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %195, i8 1) ; BufferUpdateCounter(uav,inc) + %197 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %198 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %197, i8 1) ; BufferUpdateCounter(uav,inc) + %199 = bitcast i32 %184 to float + %200 = bitcast i32 %186 to float + %201 = fcmp fast ogt float %199, 0.000000e+00 + %202 = zext i1 %201 to i32 + %203 = fcmp fast olt float %200, %199 + %204 = zext i1 %203 to i32 + %205 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %206 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %205, i8 1) ; BufferUpdateCounter(uav,inc) + %207 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %207, i32 %206, i32 0, i32 80, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %208 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %208, i32 %206, i32 76, i32 %47, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %209 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %210 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %209, i8 1) ; BufferUpdateCounter(uav,inc) + call void @dx.op.callShader.struct.SomeValues_0(i32 159, i32 %210, %struct.SomeValues_0* nonnull %4) ; CallShader(ShaderIndex,Parameter) + %211 = getelementptr inbounds %struct.SomeValues_0, %struct.SomeValues_0* %4, i32 0, i32 0 + %212 = load i32, i32* %211, align 8 + %213 = add i32 %160, %154 + %214 = add i32 %213, %166 + %215 = add i32 %214, %172 + %216 = add i32 %215, %178 + %217 = add i32 %216, %202 + %218 = add i32 %217, %204 + %219 = add i32 %218, %212 + br label %229 + +; <label>:220 ; preds = %139 + %221 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %222 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %221, i8 1) ; BufferUpdateCounter(uav,inc) + %223 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %223, i32 %222, i32 0, i32 73, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %224 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %224, i32 %222, i32 76, i32 %47, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %225 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32 160, %"class.RWStructuredBuffer<NvShaderExtnStruct>" %3) ; CreateHandleForLib(Resource) + %226 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %225, i8 1) ; BufferUpdateCounter(uav,inc) + %227 = icmp ne i32 %226, 0 + %228 = zext i1 %227 to i32 + br label %229 + +; <label>:229 ; preds = %220, %148 + %230 = phi i32 [ %219, %148 ], [ %228, %220 ] + %231 = add i32 %230, %140 + %232 = call %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<unsigned int>"(i32 160, %"class.RWStructuredBuffer<unsigned int>" %2) ; CreateHandleForLib(Resource) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %232, i32 %10, i32 0, i32 %231, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + ret void +} + +; Function Attrs: nounwind +declare i32 @dx.op.bufferUpdateCounter(i32, %dx.types.Handle, i8) #0 + +; Function Attrs: nounwind +declare void @dx.op.rawBufferStore.i32(i32, %dx.types.Handle, i32, i32, i32, i32, i32, i32, i8, i32) #0 + +; Function Attrs: nounwind +declare void @dx.op.callShader.struct.SomeValues_0(i32, i32, %struct.SomeValues_0*) #0 + +; Function Attrs: nounwind +declare void @dx.op.traceRay.struct.NvHitObjectMacroDummyPayloadType(i32, %dx.types.Handle, i32, i32, i32, i32, i32, float, float, float, float, float, float, float, float, %struct.NvHitObjectMacroDummyPayloadType*) #0 + +; Function Attrs: nounwind readnone +declare i32 @dx.op.dispatchRaysIndex.i32(i32, i8) #1 + +; Function Attrs: nounwind readonly +declare %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<NvShaderExtnStruct>"(i32, %"class.RWStructuredBuffer<NvShaderExtnStruct>") #2 + +; Function Attrs: nounwind readonly +declare %dx.types.Handle @dx.op.createHandleForLib.struct.RaytracingAccelerationStructure(i32, %struct.RaytracingAccelerationStructure) #2 + +; Function Attrs: nounwind readonly +declare %dx.types.Handle @"dx.op.createHandleForLib.class.RWStructuredBuffer<unsigned int>"(i32, %"class.RWStructuredBuffer<unsigned int>") #2 + +attributes #0 = { nounwind } +attributes #1 = { nounwind readnone } +attributes #2 = { nounwind readonly } + +!llvm.ident = !{!0} +!dx.version = !{!1} +!dx.valver = !{!2} +!dx.shaderModel = !{!3} +!dx.resources = !{!4} +!dx.entryPoints = !{!13, !15} + +!0 = !{!"clang version 3.7 (tags/RELEASE_370/final)"} +!1 = !{i32 1, i32 5} +!2 = !{i32 1, i32 7} +!3 = !{!"lib", i32 6, i32 5} +!4 = !{!5, !8, null, null} +!5 = !{!6} +!6 = !{i32 0, %struct.RaytracingAccelerationStructure* @"\01?scene_0@@3URaytracingAccelerationStructure@@A", !"scene_0", i32 0, i32 0, i32 1, i32 16, i32 0, !7} +!7 = !{i32 0, i32 4} +!8 = !{!9, !11} +!9 = !{i32 0, %"class.RWStructuredBuffer<NvShaderExtnStruct>"* @"\01?g_NvidiaExt@@3V?$RWStructuredBuffer@UNvShaderExtnStruct@@@@A", !"g_NvidiaExt", i32 0, i32 0, i32 1, i32 12, i1 false, i1 true, i1 false, !10} +!10 = !{i32 1, i32 256} +!11 = !{i32 1, %"class.RWStructuredBuffer<unsigned int>"* @"\01?outputBuffer_0@@3V?$RWStructuredBuffer@I@@A", !"outputBuffer_0", i32 0, i32 1, i32 1, i32 12, i1 false, i1 false, i1 false, !12} +!12 = !{i32 1, i32 4} +!13 = !{null, !"", null, !4, !14} +!14 = !{i32 0, i64 8454160} +!15 = !{void ()* @"\01?rayGenerationMain@@YAXXZ", !"\01?rayGenerationMain@@YAXXZ", null, null, !16} +!16 = !{i32 8, i32 7, i32 5, !17} +!17 = !{i32 0} +!18 = !{!19} +!19 = distinct !{!19, !20, !"\01?HitObject_MakeHit_1@@YA?AUNvHitObject@@URaytracingAccelerationStructure@@IIIIIIURayDesc@@USomeValues_0@@@Z: %agg.result"} +!20 = distinct !{!20, !"\01?HitObject_MakeHit_1@@YA?AUNvHitObject@@URaytracingAccelerationStructure@@IIIIIIURayDesc@@USomeValues_0@@@Z"} +!21 = !{!22} +!22 = distinct !{!22, !23, !"\01?HitObject_GetAttributes_0@@YA?AUSomeValues_0@@UNvHitObject@@@Z: %agg.result"} +!23 = distinct !{!23, !"\01?HitObject_GetAttributes_0@@YA?AUSomeValues_0@@UNvHitObject@@@Z"} +} diff --git a/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-assign.slang b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-assign.slang new file mode 100644 index 000000000..c4b40c2ac --- /dev/null +++ b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-assign.slang @@ -0,0 +1,33 @@ +// hit-object-assign.slang + +//TEST:SIMPLE: -target dxil -entry computeMain -stage compute -profile sm_6_5 -DNV_SHADER_EXTN_SLOT=u0 +//TEST:SIMPLE: -target glsl -entry computeMain -stage compute -profile sm_6_5 -line-directive-mode none + +//DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -output-using-type -profile sm_6_5 -nvapi-slot u0 +//DISABLE_TEST(compute):COMPARE_COMPUTE:-d3d12 -output-using-type -use-dxil -profile sm_6_5 -render-feature ray-query +//DISABLE_TEST(compute):COMPARE_COMPUTE:-vk -output-using-type -render-feature ray-query + +//TEST_INPUT:ubuffer(data=[0 0 0 0 0], stride=4):out,name=outputBuffer +RWStructuredBuffer<uint> outputBuffer; + +[numthreads(4, 1, 1)] +void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID) +{ + int idx = int(dispatchThreadID.x); + + RayDesc ray; + ray.Origin = float3(idx, 0, 0); + ray.TMin = 0.01f; + ray.Direction = float3(0, 1, 0); + ray.TMax = 1e4f; + + HitObject hit = HitObject::MakeMiss(idx, ray); + + // Let's try assigning. This should work on VK + // because it will SSA out such there isn't an assignment in GLSL output. + hit = HitObject::MakeMiss(idx + 1, ray); + + int r = int(hit.IsMiss()); + + outputBuffer[idx] = r; +} diff --git a/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-assign.slang.1.expected b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-assign.slang.1.expected new file mode 100644 index 000000000..e7b370664 --- /dev/null +++ b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-assign.slang.1.expected @@ -0,0 +1,42 @@ +result code = 0 +standard error = { +} +standard output = { +#version 450 +#extension GL_EXT_ray_tracing : require +#extension GL_NV_shader_invocation_reorder : require +layout(row_major) uniform; +layout(row_major) buffer; +layout(std430, binding = 0) buffer _S1 { + uint _data[]; +} outputBuffer_0; +struct RayDesc_0 +{ + vec3 Origin_0; + float TMin_0; + vec3 Direction_0; + float TMax_0; +}; + +layout(local_size_x = 4, local_size_y = 1, local_size_z = 1) in; +void main() +{ + int idx_0 = int(gl_GlobalInvocationID.x); + RayDesc_0 ray_0; + ray_0.Origin_0 = vec3(float(idx_0), 0.00000000000000000000, 0.00000000000000000000); + ray_0.TMin_0 = 0.00999999977648258209; + ray_0.Direction_0 = vec3(0.00000000000000000000, 1.00000000000000000000, 0.00000000000000000000); + ray_0.TMax_0 = 10000.00000000000000000000; + RayDesc_0 _S2 = ray_0; + hitObjectNV hitObj_0; + hitObjectRecordMissNV(hitObj_0, uint(idx_0), _S2.Origin_0, _S2.TMin_0, _S2.Direction_0, _S2.TMax_0); + RayDesc_0 _S3 = ray_0; + hitObjectNV hitObj_1; + hitObjectRecordMissNV(hitObj_1, uint(idx_0 + 1), _S3.Origin_0, _S3.TMin_0, _S3.Direction_0, _S3.TMax_0); + bool _S4 = (hitObjectIsMissNV((hitObj_1))); + uint _S5 = uint(int(_S4)); + ((outputBuffer_0)._data[(uint(idx_0))]) = _S5; + return; +} + +} diff --git a/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-assign.slang.expected b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-assign.slang.expected new file mode 100644 index 000000000..db8b93407 --- /dev/null +++ b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-assign.slang.expected @@ -0,0 +1,153 @@ +result code = 0 +standard error = { +} +standard output = { +; +; Input signature: +; +; Name Index Mask Register SysValue Format Used +; -------------------- ----- ------ -------- -------- ------- ------ +; no parameters +; +; Output signature: +; +; Name Index Mask Register SysValue Format Used +; -------------------- ----- ------ -------- -------- ------- ------ +; no parameters +; shader hash: 04bfde91b0cbfbe401b763d621f07e2a +; +; Pipeline Runtime Information: +; +; Compute Shader +; NumThreads=(4,1,1) +; +; +; Buffer Definitions: +; +; Resource bind info for g_NvidiaExt +; { +; +; struct struct.NvShaderExtnStruct +; { +; +; uint opcode; ; Offset: 0 +; uint rid; ; Offset: 4 +; uint sid; ; Offset: 8 +; uint4 dst1u; ; Offset: 12 +; uint4 src3u; ; Offset: 28 +; uint4 src4u; ; Offset: 44 +; uint4 src5u; ; Offset: 60 +; uint4 src0u; ; Offset: 76 +; uint4 src1u; ; Offset: 92 +; uint4 src2u; ; Offset: 108 +; uint4 dst0u; ; Offset: 124 +; uint markUavRef; ; Offset: 140 +; uint numOutputsForIncCounter; ; Offset: 144 +; float padding1[27]; ; Offset: 148 +; +; } $Element; ; Offset: 0 Size: 256 +; +; } +; +; Resource bind info for outputBuffer_0 +; { +; +; uint $Element; ; Offset: 0 Size: 4 +; +; } +; +; +; Resource Bindings: +; +; Name Type Format Dim ID HLSL Bind Count +; ------------------------------ ---------- ------- ----------- ------- -------------- ------ +; g_NvidiaExt UAV struct r/w+cnt U0 u0 1 +; outputBuffer_0 UAV struct r/w U1 u1 1 +; +target datalayout = "e-m:e-p:32:32-i1:32-i8:8-i16:16-i32:32-i64:64-f16:16-f32:32-f64:64-n8:16:32:64" +target triple = "dxil-ms-dx" + +%dx.types.Handle = type { i8* } +%"class.RWStructuredBuffer<NvShaderExtnStruct>" = type { %struct.NvShaderExtnStruct } +%struct.NvShaderExtnStruct = type { i32, i32, i32, <4 x i32>, <4 x i32>, <4 x i32>, <4 x i32>, <4 x i32>, <4 x i32>, <4 x i32>, <4 x i32>, i32, i32, [27 x float] } +%"class.RWStructuredBuffer<unsigned int>" = type { i32 } + +define void @computeMain() { + %1 = call %dx.types.Handle @dx.op.createHandle(i32 57, i8 1, i32 1, i32 1, i1 false) ; CreateHandle(resourceClass,rangeId,index,nonUniformIndex) + %2 = call %dx.types.Handle @dx.op.createHandle(i32 57, i8 1, i32 0, i32 0, i1 false) ; CreateHandle(resourceClass,rangeId,index,nonUniformIndex) + %3 = call i32 @dx.op.threadId.i32(i32 93, i32 0) ; ThreadId(component) + %4 = sitofp i32 %3 to float + %5 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %2, i8 1) ; BufferUpdateCounter(uav,inc) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %2, i32 %5, i32 0, i32 70, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %2, i32 %5, i32 76, i32 %3, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %2, i32 %5, i32 80, i32 1008981770, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %2, i32 %5, i32 84, i32 1176256512, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %6 = bitcast float %4 to i32 + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %2, i32 %5, i32 92, i32 %6, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %2, i32 %5, i32 96, i32 0, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %2, i32 %5, i32 100, i32 0, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %2, i32 %5, i32 108, i32 0, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %2, i32 %5, i32 112, i32 1065353216, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %2, i32 %5, i32 116, i32 0, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %7 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %2, i8 1) ; BufferUpdateCounter(uav,inc) + %8 = add nsw i32 %3, 1 + %9 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %2, i8 1) ; BufferUpdateCounter(uav,inc) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %2, i32 %9, i32 0, i32 70, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %2, i32 %9, i32 76, i32 %8, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %2, i32 %9, i32 80, i32 1008981770, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %2, i32 %9, i32 84, i32 1176256512, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %10 = bitcast float %4 to i32 + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %2, i32 %9, i32 92, i32 %10, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %2, i32 %9, i32 96, i32 0, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %2, i32 %9, i32 100, i32 0, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %2, i32 %9, i32 108, i32 0, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %2, i32 %9, i32 112, i32 1065353216, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %2, i32 %9, i32 116, i32 0, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %11 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %2, i8 1) ; BufferUpdateCounter(uav,inc) + %12 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %2, i8 1) ; BufferUpdateCounter(uav,inc) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %2, i32 %12, i32 0, i32 73, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %2, i32 %12, i32 76, i32 %11, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + %13 = call i32 @dx.op.bufferUpdateCounter(i32 70, %dx.types.Handle %2, i8 1) ; BufferUpdateCounter(uav,inc) + %14 = icmp ne i32 %13, 0 + %15 = zext i1 %14 to i32 + call void @dx.op.rawBufferStore.i32(i32 140, %dx.types.Handle %1, i32 %3, i32 0, i32 %15, i32 undef, i32 undef, i32 undef, i8 1, i32 4) ; RawBufferStore(uav,index,elementOffset,value0,value1,value2,value3,mask,alignment) + ret void +} + +; Function Attrs: nounwind readnone +declare i32 @dx.op.threadId.i32(i32, i32) #0 + +; Function Attrs: nounwind +declare void @dx.op.rawBufferStore.i32(i32, %dx.types.Handle, i32, i32, i32, i32, i32, i32, i8, i32) #1 + +; Function Attrs: nounwind +declare i32 @dx.op.bufferUpdateCounter(i32, %dx.types.Handle, i8) #1 + +; Function Attrs: nounwind readonly +declare %dx.types.Handle @dx.op.createHandle(i32, i8, i32, i32, i1) #2 + +attributes #0 = { nounwind readnone } +attributes #1 = { nounwind } +attributes #2 = { nounwind readonly } + +!llvm.ident = !{!0} +!dx.version = !{!1} +!dx.valver = !{!2} +!dx.shaderModel = !{!3} +!dx.resources = !{!4} +!dx.entryPoints = !{!10} + +!0 = !{!"clang version 3.7 (tags/RELEASE_370/final)"} +!1 = !{i32 1, i32 5} +!2 = !{i32 1, i32 7} +!3 = !{!"cs", i32 6, i32 5} +!4 = !{null, !5, null, null} +!5 = !{!6, !8} +!6 = !{i32 0, %"class.RWStructuredBuffer<NvShaderExtnStruct>"* undef, !"", i32 0, i32 0, i32 1, i32 12, i1 false, i1 true, i1 false, !7} +!7 = !{i32 1, i32 256} +!8 = !{i32 1, %"class.RWStructuredBuffer<unsigned int>"* undef, !"", i32 0, i32 1, i32 1, i32 12, i1 false, i1 false, i1 false, !9} +!9 = !{i32 1, i32 4} +!10 = !{void ()* @computeMain, !"computeMain", null, !4, !11} +!11 = !{i32 0, i64 8388624, i32 4, !12} +!12 = !{i32 4, i32 1, i32 1} +} diff --git a/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-make-hit.slang b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-make-hit.slang index 9ba631c9f..e38b29446 100644 --- a/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-make-hit.slang +++ b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-make-hit.slang @@ -1,6 +1,7 @@ // hit-object-make-hit.slang -//DISABLE_TEST:SIMPLE: -target dxil -entry rayGenerationMain -stage raygeneration -profile sm_6_5 -DNV_SHADER_EXTN_SLOT=u0 +//TEST:SIMPLE: -target dxil -entry rayGenerationMain -stage raygeneration -profile sm_6_5 -DNV_SHADER_EXTN_SLOT=u0 +//TEST:SIMPLE: -target glsl -entry rayGenerationMain -stage raygeneration -profile sm_6_5 -line-directive-mode none //DISABLE_TEST(compute):COMPARE_COMPUTE:-d3d12 -output-using-type -use-dxil -profile sm_6_5 -render-feature ray-query //DISABLE_TEST(compute):COMPARE_COMPUTE:-vk -output-using-type -render-feature ray-query diff --git a/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-make-hit.slang.1.expected b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-make-hit.slang.1.expected new file mode 100644 index 000000000..167c93aff --- /dev/null +++ b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-make-hit.slang.1.expected @@ -0,0 +1,108 @@ +result code = 0 +standard error = { +} +standard output = { +#version 460 +#extension GL_EXT_ray_tracing : require +#extension GL_NV_shader_invocation_reorder : require +layout(row_major) uniform; +layout(row_major) buffer; +struct SomeValues_0 +{ + int a_0; + float b_0; +}; + +layout(location = 0) +hitObjectAttributeNV +SomeValues_0 t_0; + +layout(binding = 0) +uniform accelerationStructureEXT scene_0; + +struct RayDesc_0 +{ + vec3 Origin_0; + float TMin_0; + vec3 Direction_0; + float TMax_0; +}; + +RayDesc_0 HitObject_GetRayDesc_0(hitObjectNV this_0) +{ + vec3 _S1 = (hitObjectGetWorldRayOriginNV((this_0))); + float _S2 = (hitObjectGetRayTMinNV((this_0))); + vec3 _S3 = (hitObjectGetObjectRayDirectionNV((this_0))); + float _S4 = (hitObjectGetRayTMaxNV((this_0))); + RayDesc_0 ray_0 = { _S1, _S2, _S3, _S4 }; + return ray_0; +} + +SomeValues_0 HitObject_GetAttributes_0(hitObjectNV this_1) +{ + hitObjectGetAttributesNV((this_1), ((0))); + return t_0; +} + +uint calcValue_0(hitObjectNV hit_0) +{ + bool _S5 = (hitObjectIsHitNV((hit_0))); + uint r_0; + if(_S5) + { + uint instanceIndex_0 = (hitObjectGetInstanceCustomIndexNV((hit_0))); + uint instanceID_0 = (hitObjectGetInstanceIdNV((hit_0))); + uint geometryIndex_0 = (hitObjectGetGeometryIndexNV((hit_0))); + uint primitiveIndex_0 = (hitObjectGetPrimitiveIndexNV((hit_0))); + uint hitKind_0 = (hitObjectGetHitKindNV((hit_0))); + uint r_1 = 0U + hitKind_0 + instanceIndex_0 + instanceID_0 + geometryIndex_0 + primitiveIndex_0; + RayDesc_0 ray_1 = HitObject_GetRayDesc_0(hit_0); + uint r_2 = r_1 + uint(ray_1.TMin_0 > 0.00000000000000000000) + uint(ray_1.TMax_0 < ray_1.TMin_0); + SomeValues_0 objSomeValues_0 = HitObject_GetAttributes_0(hit_0); + r_0 = r_2 + uint(objSomeValues_0.a_0); + } + else + { + bool _S6 = (hitObjectIsMissNV((hit_0))); + uint r_3; + if(_S6) + { + r_3 = 1U; + } + else + { + r_3 = 0U; + } + r_0 = r_3; + } + return r_0; +} + +layout(std430, binding = 1) buffer _S7 { + uint _data[]; +} outputBuffer_0; +void main() +{ + uvec3 _S8 = ((gl_LaunchIDEXT)); + ivec2 launchID_0 = ivec2(_S8.xy); + uvec3 _S9 = ((gl_LaunchSizeEXT)); + int idx_0 = launchID_0.x; + RayDesc_0 ray_2; + ray_2.Origin_0 = vec3(float(idx_0), 0.00000000000000000000, 0.00000000000000000000); + ray_2.TMin_0 = 0.00999999977648258209; + ray_2.Direction_0 = vec3(0.00000000000000000000, 1.00000000000000000000, 0.00000000000000000000); + ray_2.TMax_0 = 10000.00000000000000000000; + RayDesc_0 _S10 = ray_2; + hitObjectNV hitObj_0; + hitObjectRecordHitWithIndexNV(hitObj_0, scene_0, uint(idx_0), uint(idx_0 * 2), uint(idx_0 * 3), 0U, 0U, _S10.Origin_0, _S10.TMin_0, _S10.Direction_0, _S10.TMax_0, (0)); + uint r_4 = calcValue_0(hitObj_0); + RayDesc_0 _S11 = ray_2; + hitObjectNV hitObj_1; + hitObjectRecordHitNV(hitObj_1, scene_0, uint(idx_0), uint(idx_0 * 3), uint(idx_0 * 2), 0U, 0U, 4U, _S11.Origin_0, _S11.TMin_0, _S11.Direction_0, _S11.TMax_0, (0)); + uint _S12 = calcValue_0(hitObj_1); + uint r_5 = r_4 + _S12; + ((outputBuffer_0)._data[(uint(idx_0))]) = r_5; + return; +} + +} diff --git a/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-make-miss.slang b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-make-miss.slang index cf4262c53..421063987 100644 --- a/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-make-miss.slang +++ b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-make-miss.slang @@ -1,6 +1,7 @@ // hit-object-make-miss.slang //TEST:SIMPLE: -target dxil -entry computeMain -stage compute -profile sm_6_5 -DNV_SHADER_EXTN_SLOT=u0 +//TEST:SIMPLE: -target glsl -entry computeMain -stage compute -profile sm_6_5 -line-directive-mode none //DISABLE_TEST:COMPARE_COMPUTE_EX:-slang -compute -dx12 -output-using-type -profile sm_6_5 -nvapi-slot u0 //DISABLE_TEST(compute):COMPARE_COMPUTE:-d3d12 -output-using-type -use-dxil -profile sm_6_5 -render-feature ray-query diff --git a/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-make-miss.slang.1.expected b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-make-miss.slang.1.expected new file mode 100644 index 000000000..33eb46da3 --- /dev/null +++ b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-make-miss.slang.1.expected @@ -0,0 +1,39 @@ +result code = 0 +standard error = { +} +standard output = { +#version 450 +#extension GL_EXT_ray_tracing : require +#extension GL_NV_shader_invocation_reorder : require +layout(row_major) uniform; +layout(row_major) buffer; +layout(std430, binding = 0) buffer _S1 { + uint _data[]; +} outputBuffer_0; +struct RayDesc_0 +{ + vec3 Origin_0; + float TMin_0; + vec3 Direction_0; + float TMax_0; +}; + +layout(local_size_x = 4, local_size_y = 1, local_size_z = 1) in; +void main() +{ + int idx_0 = int(gl_GlobalInvocationID.x); + RayDesc_0 ray_0; + ray_0.Origin_0 = vec3(float(idx_0), 0.00000000000000000000, 0.00000000000000000000); + ray_0.TMin_0 = 0.00999999977648258209; + ray_0.Direction_0 = vec3(0.00000000000000000000, 1.00000000000000000000, 0.00000000000000000000); + ray_0.TMax_0 = 10000.00000000000000000000; + RayDesc_0 _S2 = ray_0; + hitObjectNV hitObj_0; + hitObjectRecordMissNV(hitObj_0, uint(idx_0), _S2.Origin_0, _S2.TMin_0, _S2.Direction_0, _S2.TMax_0); + bool _S3 = (hitObjectIsMissNV((hitObj_0))); + uint _S4 = uint(int(_S3)); + ((outputBuffer_0)._data[(uint(idx_0))]) = _S4; + return; +} + +} diff --git a/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-make-nop.slang b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-make-nop.slang index c5f379898..b1d72c47e 100644 --- a/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-make-nop.slang +++ b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-make-nop.slang @@ -1,6 +1,7 @@ // hit-object-make-nop.slang //TEST:SIMPLE: -target dxil -entry rayGenerationMain -stage raygeneration -profile sm_6_5 -DNV_SHADER_EXTN_SLOT=u0 +//TEST:SIMPLE: -target glsl -entry rayGenerationMain -stage raygeneration -profile sm_6_5 -line-directive-mode none //DISABLE_TEST(compute):COMPARE_COMPUTE:-d3d12 -output-using-type -use-dxil -profile sm_6_5 -render-feature ray-query //DISABLE_TEST(compute):COMPARE_COMPUTE:-vk -output-using-type -render-feature ray-query diff --git a/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-make-nop.slang.1.expected b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-make-nop.slang.1.expected new file mode 100644 index 000000000..956dddad3 --- /dev/null +++ b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-make-nop.slang.1.expected @@ -0,0 +1,28 @@ +result code = 0 +standard error = { +} +standard output = { +#version 460 +#extension GL_EXT_ray_tracing : require +#extension GL_NV_shader_invocation_reorder : require +layout(row_major) uniform; +layout(row_major) buffer; +layout(std430, binding = 0) buffer _S1 { + uint _data[]; +} outputBuffer_0; +void main() +{ + uvec3 _S2 = ((gl_LaunchIDEXT)); + ivec2 launchID_0 = ivec2(_S2.xy); + uvec3 _S3 = ((gl_LaunchSizeEXT)); + int idx_0 = launchID_0.x; + hitObjectNV hitObj_0; + hitObjectRecordEmptyNV((hitObj_0)); + uint _S4 = uint(idx_0); + bool _S5 = (hitObjectIsEmptyNV((hitObj_0))); + uint _S6 = uint(_S5); + ((outputBuffer_0)._data[(_S4)]) = _S6; + return; +} + +} diff --git a/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-reorder-thread.slang b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-reorder-thread.slang index 89b01b29a..ed83b8d47 100644 --- a/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-reorder-thread.slang +++ b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-reorder-thread.slang @@ -1,6 +1,7 @@ // hit-object-reorder-thread.slang -//DISABLE_TEST:SIMPLE: -target dxil -entry rayGenerationMain -stage raygeneration -profile sm_6_5 -DNV_SHADER_EXTN_SLOT=u0 +//TEST:SIMPLE: -target dxil -entry rayGenerationMain -stage raygeneration -profile sm_6_5 -DNV_SHADER_EXTN_SLOT=u0 +//TEST:SIMPLE: -target glsl -entry rayGenerationMain -stage raygeneration -profile sm_6_5 -line-directive-mode none //DISABLE_TEST(compute):COMPARE_COMPUTE:-d3d12 -output-using-type -use-dxil -profile sm_6_5 -render-feature ray-query //DISABLE_TEST(compute):COMPARE_COMPUTE:-vk -output-using-type -render-feature ray-query diff --git a/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-reorder-thread.slang.1.expected b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-reorder-thread.slang.1.expected new file mode 100644 index 000000000..90223115b --- /dev/null +++ b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-reorder-thread.slang.1.expected @@ -0,0 +1,116 @@ +result code = 0 +standard error = { +} +standard output = { +#version 460 +#extension GL_EXT_ray_tracing : require +#extension GL_NV_shader_invocation_reorder : require +layout(row_major) uniform; +layout(row_major) buffer; +struct SomeValues_0 +{ + int a_0; + float b_0; +}; + +layout(location = 0) +rayPayloadEXT +SomeValues_0 p_0; + +layout(binding = 0) +uniform accelerationStructureEXT scene_0; + +layout(location = 0) +hitObjectAttributeNV +SomeValues_0 t_0; + +SomeValues_0 HitObject_GetAttributes_0(hitObjectNV this_0) +{ + hitObjectGetAttributesNV((this_0), ((0))); + return t_0; +} + +uint calcValue_0(hitObjectNV hit_0) +{ + bool _S1 = (hitObjectIsHitNV((hit_0))); + uint r_0; + if(_S1) + { + uint instanceIndex_0 = (hitObjectGetInstanceCustomIndexNV((hit_0))); + uint instanceID_0 = (hitObjectGetInstanceIdNV((hit_0))); + uint geometryIndex_0 = (hitObjectGetGeometryIndexNV((hit_0))); + uint primitiveIndex_0 = (hitObjectGetPrimitiveIndexNV((hit_0))); + SomeValues_0 objSomeValues_0 = HitObject_GetAttributes_0(hit_0); + r_0 = 0U + instanceIndex_0 + instanceID_0 + geometryIndex_0 + primitiveIndex_0 + uint(objSomeValues_0.a_0); + } + else + { + r_0 = 0U; + } + return r_0; +} + +layout(location = 1) +rayPayloadEXT +SomeValues_0 p_1; + +void HitObject_Invoke_0(accelerationStructureEXT AccelerationStructure_0, hitObjectNV HitOrMiss_0, inout SomeValues_0 Payload_0) +{ + p_1 = Payload_0; + hitObjectExecuteShaderNV(HitOrMiss_0, (1)); + Payload_0 = p_1; + return; +} + +layout(std430, binding = 1) buffer _S2 { + uint _data[]; +} outputBuffer_0; +struct RayDesc_0 +{ + vec3 Origin_0; + float TMin_0; + vec3 Direction_0; + float TMax_0; +}; + +void main() +{ + uvec3 _S3 = ((gl_LaunchIDEXT)); + ivec2 launchID_0 = ivec2(_S3.xy); + uvec3 _S4 = ((gl_LaunchSizeEXT)); + int idx_0 = launchID_0.x; + SomeValues_0 someValues_0 = { idx_0, float(idx_0) * 2.00000000000000000000 }; + RayDesc_0 ray_0; + ray_0.Origin_0 = vec3(float(idx_0), 0.00000000000000000000, 0.00000000000000000000); + ray_0.TMin_0 = 0.00999999977648258209; + ray_0.Direction_0 = vec3(0.00000000000000000000, 1.00000000000000000000, 0.00000000000000000000); + ray_0.TMax_0 = 10000.00000000000000000000; + RayDesc_0 _S5 = ray_0; + p_0 = someValues_0; + hitObjectNV hitObj_0; + hitObjectTraceRayNV(hitObj_0, scene_0, 20U, 255U, 0U, 4U, 0U, _S5.Origin_0, _S5.TMin_0, _S5.Direction_0, _S5.TMax_0, (0)); + uint r_1 = calcValue_0(hitObj_0); + reorderThreadNV(hitObj_0); + SomeValues_0 otherValues_0; + SomeValues_0 _S6 = { idx_0 * -1, float(idx_0) * 4.00000000000000000000 }; + otherValues_0 = _S6; + HitObject_Invoke_0(scene_0, hitObj_0, otherValues_0); + uint _S7 = calcValue_0(hitObj_0); + uint r_2 = r_1 + _S7; + reorderThreadNV(hitObj_0, uint(idx_0 & 3), 2U); + SomeValues_0 _S8 = { idx_0 * -2, float(idx_0) * 8.00000000000000000000 }; + otherValues_0 = _S8; + HitObject_Invoke_0(scene_0, hitObj_0, otherValues_0); + uint _S9 = calcValue_0(hitObj_0); + uint r_3 = r_2 + _S9; + reorderThreadNV(uint(idx_0 & 1), 1U); + SomeValues_0 _S10 = { idx_0 * -4, float(idx_0) * 16.00000000000000000000 }; + otherValues_0 = _S10; + HitObject_Invoke_0(scene_0, hitObj_0, otherValues_0); + uint _S11 = calcValue_0(hitObj_0); + uint r_4 = r_3 + _S11; + ((outputBuffer_0)._data[(uint(idx_0))]) = r_4; + return; +} + +} diff --git a/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-trace-motion-ray.slang b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-trace-motion-ray.slang new file mode 100644 index 000000000..f57ecf02a --- /dev/null +++ b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-trace-motion-ray.slang @@ -0,0 +1,79 @@ +// hit-object-trace-motion-ray.slang + +// Motion rays not supported on HLSL impl currently +//DISABLE_TEST:SIMPLE: -target dxil -entry rayGenerationMain -stage raygeneration -profile sm_6_5 -DNV_SHADER_EXTN_SLOT=u0 +//TEST:SIMPLE: -target glsl -entry rayGenerationMain -stage raygeneration -profile sm_6_5 -line-directive-mode none + +//DISABLE_TEST(compute):COMPARE_COMPUTE:-d3d12 -output-using-type -use-dxil -profile sm_6_6 -render-feature ray-query +//DISABLE_TEST(compute):COMPARE_COMPUTE:-vk -output-using-type -render-feature ray-query + +//TEST_INPUT: set scene = AccelerationStructure +uniform RaytracingAccelerationStructure scene; + +//TEST_INPUT:set outputBuffer = out ubuffer(data=[0, 0, 0, 0], stride=4) +RWStructuredBuffer<uint> outputBuffer; + +struct SomeValues +{ + int a; + float b; +}; + +uint calcValue(HitObject hit) +{ + uint r = 0; + + if (hit.IsHit()) + { + uint instanceIndex = hit.GetInstanceIndex(); + uint instanceID = hit.GetInstanceID(); + uint geometryIndex = hit.GetGeometryIndex(); + uint primitiveIndex = hit.GetPrimitiveIndex(); + + SomeValues objSomeValues = hit.GetAttributes<SomeValues>(); + + r += instanceIndex; + r += instanceID; + r += geometryIndex; + r += primitiveIndex; + r += objSomeValues.a; + } + + return r; +} + +void rayGenerationMain() +{ + int2 launchID = int2(DispatchRaysIndex().xy); + int2 launchSize = int2(DispatchRaysDimensions().xy); + + int idx = launchID.x; + + float currentTime = idx / 4; + + SomeValues someValues = { idx, idx * 2.0f }; + + RayDesc ray; + ray.Origin = float3(idx, 0, 0); + ray.TMin = 0.01f; + ray.Direction = float3(0, 1, 0); + ray.TMax = 1e4f; + + RAY_FLAG rayFlags = RAY_FLAG_ACCEPT_FIRST_HIT_AND_END_SEARCH | RAY_FLAG_CULL_BACK_FACING_TRIANGLES; + uint instanceInclusionMask = 0xff; + uint rayContributionToHitGroupIndex = 0; + uint multiplierForGeometryContributionToHitGroupIndex = 4; + uint missShaderIndex = 0; + + HitObject hit = HitObject::TraceMotionRay(scene, + rayFlags, + instanceInclusionMask, + rayContributionToHitGroupIndex, + multiplierForGeometryContributionToHitGroupIndex, + missShaderIndex, + ray, + currentTime, + someValues); + + outputBuffer[idx] = calcValue(hit); +} diff --git a/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-trace-motion-ray.slang.1.expected b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-trace-motion-ray.slang.1.expected new file mode 100644 index 000000000..2d1ef23ed --- /dev/null +++ b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-trace-motion-ray.slang.1.expected @@ -0,0 +1,89 @@ +result code = 0 +standard error = { +} +standard output = { +#version 460 +#extension GL_EXT_ray_tracing : require +#extension GL_NV_shader_invocation_reorder : require +#extension GLSL_NV_ray_tracing_motion_blur : require +layout(row_major) uniform; +layout(row_major) buffer; +struct SomeValues_0 +{ + int a_0; + float b_0; +}; + +layout(location = 0) +rayPayloadEXT +SomeValues_0 p_0; + +layout(binding = 0) +uniform accelerationStructureEXT scene_0; + +layout(location = 0) +hitObjectAttributeNV +SomeValues_0 t_0; + +SomeValues_0 HitObject_GetAttributes_0(hitObjectNV this_0) +{ + hitObjectGetAttributesNV((this_0), ((0))); + return t_0; +} + +uint calcValue_0(hitObjectNV hit_0) +{ + bool _S1 = (hitObjectIsHitNV((hit_0))); + uint r_0; + if(_S1) + { + uint instanceIndex_0 = (hitObjectGetInstanceCustomIndexNV((hit_0))); + uint instanceID_0 = (hitObjectGetInstanceIdNV((hit_0))); + uint geometryIndex_0 = (hitObjectGetGeometryIndexNV((hit_0))); + uint primitiveIndex_0 = (hitObjectGetPrimitiveIndexNV((hit_0))); + SomeValues_0 objSomeValues_0 = HitObject_GetAttributes_0(hit_0); + r_0 = 0U + instanceIndex_0 + instanceID_0 + geometryIndex_0 + primitiveIndex_0 + uint(objSomeValues_0.a_0); + } + else + { + r_0 = 0U; + } + return r_0; +} + +layout(std430, binding = 1) buffer _S2 { + uint _data[]; +} outputBuffer_0; +struct RayDesc_0 +{ + vec3 Origin_0; + float TMin_0; + vec3 Direction_0; + float TMax_0; +}; + +void main() +{ + uvec3 _S3 = ((gl_LaunchIDEXT)); + ivec2 launchID_0 = ivec2(_S3.xy); + uvec3 _S4 = ((gl_LaunchSizeEXT)); + int idx_0 = launchID_0.x; + int _S5 = idx_0 / 4; + float currentTime_0 = float(_S5); + SomeValues_0 someValues_0 = { idx_0, float(idx_0) * 2.00000000000000000000 }; + RayDesc_0 ray_0; + ray_0.Origin_0 = vec3(float(idx_0), 0.00000000000000000000, 0.00000000000000000000); + ray_0.TMin_0 = 0.00999999977648258209; + ray_0.Direction_0 = vec3(0.00000000000000000000, 1.00000000000000000000, 0.00000000000000000000); + ray_0.TMax_0 = 10000.00000000000000000000; + RayDesc_0 _S6 = ray_0; + p_0 = someValues_0; + hitObjectNV hitObj_0; + hitObjectTraceRayMotionNV(hitObj_0, scene_0, 20U, 255U, 0U, 4U, 0U, _S6.Origin_0, _S6.TMin_0, _S6.Direction_0, _S6.TMax_0, currentTime_0, (0)); + uint _S7 = uint(idx_0); + uint _S8 = calcValue_0(hitObj_0); + ((outputBuffer_0)._data[(_S7)]) = _S8; + return; +} + +} diff --git a/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-trace-ray.slang b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-trace-ray.slang index 54f45e35b..63ae4c957 100644 --- a/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-trace-ray.slang +++ b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-trace-ray.slang @@ -1,6 +1,7 @@ // hit-object-trace-ray.slang -//DISABLE_TEST:SIMPLE: -target dxil -entry rayGenerationMain -stage raygeneration -profile sm_6_5 -DNV_SHADER_EXTN_SLOT=u0 +//TEST:SIMPLE: -target dxil -entry rayGenerationMain -stage raygeneration -profile sm_6_5 -DNV_SHADER_EXTN_SLOT=u0 +//TEST:SIMPLE: -target glsl -entry rayGenerationMain -stage raygeneration -profile sm_6_5 -line-directive-mode none //DISABLE_TEST(compute):COMPARE_COMPUTE:-d3d12 -output-using-type -use-dxil -profile sm_6_6 -render-feature ray-query //DISABLE_TEST(compute):COMPARE_COMPUTE:-vk -output-using-type -render-feature ray-query diff --git a/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-trace-ray.slang.1.expected b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-trace-ray.slang.1.expected new file mode 100644 index 000000000..38ddbf233 --- /dev/null +++ b/tests/hlsl-intrinsic/shader-execution-reordering/hit-object-trace-ray.slang.1.expected @@ -0,0 +1,86 @@ +result code = 0 +standard error = { +} +standard output = { +#version 460 +#extension GL_EXT_ray_tracing : require +#extension GL_NV_shader_invocation_reorder : require +layout(row_major) uniform; +layout(row_major) buffer; +struct SomeValues_0 +{ + int a_0; + float b_0; +}; + +layout(location = 0) +rayPayloadEXT +SomeValues_0 p_0; + +layout(binding = 0) +uniform accelerationStructureEXT scene_0; + +layout(location = 0) +hitObjectAttributeNV +SomeValues_0 t_0; + +SomeValues_0 HitObject_GetAttributes_0(hitObjectNV this_0) +{ + hitObjectGetAttributesNV((this_0), ((0))); + return t_0; +} + +uint calcValue_0(hitObjectNV hit_0) +{ + bool _S1 = (hitObjectIsHitNV((hit_0))); + uint r_0; + if(_S1) + { + uint instanceIndex_0 = (hitObjectGetInstanceCustomIndexNV((hit_0))); + uint instanceID_0 = (hitObjectGetInstanceIdNV((hit_0))); + uint geometryIndex_0 = (hitObjectGetGeometryIndexNV((hit_0))); + uint primitiveIndex_0 = (hitObjectGetPrimitiveIndexNV((hit_0))); + SomeValues_0 objSomeValues_0 = HitObject_GetAttributes_0(hit_0); + r_0 = 0U + instanceIndex_0 + instanceID_0 + geometryIndex_0 + primitiveIndex_0 + uint(objSomeValues_0.a_0); + } + else + { + r_0 = 0U; + } + return r_0; +} + +layout(std430, binding = 1) buffer _S2 { + uint _data[]; +} outputBuffer_0; +struct RayDesc_0 +{ + vec3 Origin_0; + float TMin_0; + vec3 Direction_0; + float TMax_0; +}; + +void main() +{ + uvec3 _S3 = ((gl_LaunchIDEXT)); + ivec2 launchID_0 = ivec2(_S3.xy); + uvec3 _S4 = ((gl_LaunchSizeEXT)); + int idx_0 = launchID_0.x; + SomeValues_0 someValues_0 = { idx_0, float(idx_0) * 2.00000000000000000000 }; + RayDesc_0 ray_0; + ray_0.Origin_0 = vec3(float(idx_0), 0.00000000000000000000, 0.00000000000000000000); + ray_0.TMin_0 = 0.00999999977648258209; + ray_0.Direction_0 = vec3(0.00000000000000000000, 1.00000000000000000000, 0.00000000000000000000); + ray_0.TMax_0 = 10000.00000000000000000000; + RayDesc_0 _S5 = ray_0; + p_0 = someValues_0; + hitObjectNV hitObj_0; + hitObjectTraceRayNV(hitObj_0, scene_0, 20U, 255U, 0U, 4U, 0U, _S5.Origin_0, _S5.TMin_0, _S5.Direction_0, _S5.TMax_0, (0)); + uint _S6 = uint(idx_0); + uint _S7 = calcValue_0(hitObj_0); + ((outputBuffer_0)._data[(_S6)]) = _S7; + return; +} + +} |