Merge 'release-1.8.2505' to 'staging-sm6.9' (#7461)

Merge 'release-1.8.2505' to 'staging-sm6.9'.  No conflicts.

Author: damyanp

docs/DXIL.rst

@@ -4227,7 +4227,7 @@ codegen for Vulkan:
 - ``-fvk-use-dx-layout``: Uses DirectX layout rules for resources.
 - ``-fvk-invert-y``: Negates (additively inverts) SV_Position.y before writing
   to stage output. Used to accommodate the difference between Vulkan's
-  coordinate system and DirectX's. Only allowed in VS/DS/GS.
+  coordinate system and DirectX's. Only allowed in VS/DS/GS/MS/Lib.
 - ``-fvk-use-dx-position-w``: Reciprocates (multiplicatively inverts)
   SV_Position.w after reading from stage input. Used to accommodate the
   difference between Vulkan DirectX: the w component of SV_Position in PS is

docs/SPIR-V.rst

@@ -1583,6 +1583,10 @@ const unsigned kMSStoreOutputColOpIdx = 3;
 const unsigned kMSStoreOutputVIdxOpIdx = 4;
 const unsigned kMSStoreOutputValOpIdx = 5;
 
+// HitObject::MakeMiss
+const unsigned kHitObjectMakeMiss_RayDescOpIdx = 3;
+const unsigned kHitObjectMakeMiss_NumOp = 11;
+
 // HitObject::TraceRay
 const unsigned kHitObjectTraceRay_RayDescOpIdx = 7;
 const unsigned kHitObjectTraceRay_PayloadOpIdx = 15;
@@ -1609,6 +1613,7 @@ const unsigned kMatVecMulAddIsOutputUnsignedIdx = 15;
 // Outer Product Accumulate
 const unsigned kOuterProdAccMatrixInterpretation = 5;
 const unsigned kOuterProdAccMatrixLayout = 6;
+const unsigned kOuterProdAccMatrixStride = 7;
 
 // TODO: add operand index for all the OpCodeClass.
 } // namespace OperandIndex

include/dxc/DXIL/DxilConstants.h

@@ -396,7 +396,12 @@ const unsigned kAnnotateHandleResourceTypeOpIdx = 3;
 
 // TraceRay.
 const unsigned kTraceRayRayDescOpIdx = 7;
-const unsigned kTraceRayPayLoadOpIdx = 8;
+// kTraceRayPayloadPreOpIdx is before flattening the RayDesc
+const unsigned kTraceRayPayloadPreOpIdx = 8;
+// kTraceRayPayloadOpIdx is after flattening the RayDesc
+const unsigned kTraceRayPayloadOpIdx = 11;
+const unsigned kTraceRay_PreNumOp = 9;
+const unsigned kTraceRay_NumOp = 12;
 
 // AllocateRayQuery
 const unsigned kAllocateRayQueryRayFlagsIdx = 1;
@@ -407,6 +412,10 @@ const unsigned kCallShaderPayloadOpIdx = 2;
 
 // TraceRayInline.
 const unsigned kTraceRayInlineRayDescOpIdx = 5;
+// kTraceRayInlinePayloadPreOpIdx is before flattening the RayDesc
+const unsigned kTraceRayInlinePayloadPreOpIdx = 6;
+// kTraceRayInlinePayloadOpIdx is after flattening the RayDesc
+const unsigned kTraceRayInlinePayloadOpIdx = 9;
 
 // ReportIntersection.
 const unsigned kReportIntersectionAttributeOpIdx = 3;
@@ -435,11 +444,19 @@ const unsigned kAnnotateNodeRecordHandleNodeRecordPropIdx = 2;
 
 // HitObject::MakeMiss
 const unsigned kHitObjectMakeMiss_NumOp = 8;
-const unsigned kHitObjectMakeMissRayDescOpIdx = 4;
+const unsigned kHitObjectMakeMiss_RayDescOpIdx = 4;
 
 // HitObject::TraceRay
 const unsigned kHitObjectTraceRay_RayDescOpIdx = 8;
-const unsigned kHitObjectTraceRay_NumOp = 10;
+// kHitObjectTraceRay_PayloadPreOpIdx is before flattening the RayDesc
+const unsigned kHitObjectTraceRay_PayloadPreOpIdx = 9;
+// kHitObjectTraceRay_PayloadOpIdx is after flattening the RayDesc
+const unsigned kHitObjectTraceRay_PayloadOpIdx = 12;
+const unsigned kHitObjectTraceRay_PreNumOp = 10;
+const unsigned kHitObjectTraceRay_NumOp = 13;
+
+// HitObject::Invoke
+const unsigned kHitObjectInvoke_PayloadOpIdx = 2;
 
 // HitObject::FromRayQuery
 const unsigned kHitObjectFromRayQuery_WithAttrs_AttributeOpIdx = 4;

include/dxc/HLSL/HLOperations.h

@@ -368,7 +368,7 @@ def fvk_bind_register : MultiArg<["-"], "fvk-bind-register", 4>, MetaVarName<"<t
   HelpText<"Specify Vulkan descriptor set and binding for a specific register">;
 def vkbr : MultiArg<["-"], "vkbr", 4>, Flags<[CoreOption, DriverOption]>, Alias<fvk_bind_register>;
 def fvk_invert_y: Flag<["-"], "fvk-invert-y">, Group<spirv_Group>, Flags<[CoreOption, DriverOption]>,
-  HelpText<"Negate SV_Position.y before writing to stage output in VS/DS/GS to accommodate Vulkan's coordinate system">;
+  HelpText<"Negate SV_Position.y before writing to stage output in VS/DS/GS/MS/Lib to accommodate Vulkan's coordinate system">;
 def fvk_use_dx_position_w: Flag<["-"], "fvk-use-dx-position-w">, Group<spirv_Group>, Flags<[CoreOption, DriverOption]>,
   HelpText<"Reciprocate SV_Position.w after reading from stage input in PS to accommodate the difference between Vulkan and DirectX">;
 def fvk_support_nonzero_base_instance: Flag<["-"], "fvk-support-nonzero-base-instance">, Group<spirv_Group>, Flags<[CoreOption, DriverOption]>,

include/dxc/Support/HLSLOptions.td

@@ -133,11 +133,15 @@ enum LEGAL_INTRINSIC_COMPTYPES {
   LICOMPTYPE_HIT_OBJECT = 51,
   LICOMPTYPE_RAY_QUERY = 52,
 
+  LICOMPTYPE_LINALG = 53, // f32, partial-precision-f32, f16,
+                          // i32, i16, u32, u16,
+                          // int8_4packed, uint8_4packed
+
 #ifdef ENABLE_SPIRV_CODEGEN
-  LICOMPTYPE_VK_BUFFER_POINTER = 53,
-  LICOMPTYPE_COUNT = 54
+  LICOMPTYPE_VK_BUFFER_POINTER = 54,
+  LICOMPTYPE_COUNT = 55
 #else
-  LICOMPTYPE_COUNT = 53
+  LICOMPTYPE_COUNT = 54
 #endif
 };
 

include/dxc/dxcapi.internal.h

@@ -165,7 +165,8 @@ ValidateSignatureAccess(Instruction *I, DxilSignature &Sig, Value *SigId,
 
 static DxilResourceProperties GetResourceFromHandle(Value *Handle,
                                                     ValidationContext &ValCtx) {
-  if (!isa<CallInst>(Handle)) {
+  CallInst *HandleCall = dyn_cast<CallInst>(Handle);
+  if (!HandleCall) {
     if (Instruction *I = dyn_cast<Instruction>(Handle))
       ValCtx.EmitInstrError(I, ValidationRule::InstrHandleNotFromCreateHandle);
     else
@@ -175,10 +176,13 @@ static DxilResourceProperties GetResourceFromHandle(Value *Handle,
   }
 
   DxilResourceProperties RP = ValCtx.GetResourceFromVal(Handle);
-  if (RP.getResourceClass() == DXIL::ResourceClass::Invalid) {
+  if (RP.getResourceClass() == DXIL::ResourceClass::Invalid)
     ValCtx.EmitInstrError(cast<CallInst>(Handle),
                           ValidationRule::InstrHandleNotFromCreateHandle);
-  }
+  if (RP.Basic.IsReorderCoherent &&
+      !ValCtx.DxilMod.GetShaderModel()->IsSM69Plus())
+    ValCtx.EmitInstrError(HandleCall,
+                          ValidationRule::InstrReorderCoherentRequiresSM69);
 
   return RP;
 }
@@ -1229,6 +1233,32 @@ static void ValidateImmOperandsForOuterProdAcc(CallInst *CI,
         {"MatrixLayout"});
     return;
   }
+  ConstantInt *ML = cast<ConstantInt>(MatrixLayout);
+  uint64_t MLValue = ML->getLimitedValue();
+  if (MLValue !=
+      static_cast<unsigned>(DXIL::LinalgMatrixLayout::OuterProductOptimal))
+    ValCtx.EmitInstrFormatError(
+        CI,
+        ValidationRule::
+            InstrLinalgInvalidMatrixLayoutValueForOuterProductAccumulate,
+        {GetMatrixLayoutStr(MLValue),
+         GetMatrixLayoutStr(static_cast<unsigned>(
+             DXIL::LinalgMatrixLayout::OuterProductOptimal))});
+
+  llvm::Value *MatrixStride =
+      CI->getOperand(DXIL::OperandIndex::kOuterProdAccMatrixStride);
+  if (!llvm::isa<llvm::Constant>(MatrixStride)) {
+    ValCtx.EmitInstrError(
+        CI, ValidationRule::InstrLinalgMatrixStrideZeroForOptimalLayouts);
+    return;
+  }
+  ConstantInt *MS = cast<ConstantInt>(MatrixStride);
+  uint64_t MSValue = MS->getLimitedValue();
+  if (MSValue != 0) {
+    ValCtx.EmitInstrError(
+        CI, ValidationRule::InstrLinalgMatrixStrideZeroForOptimalLayouts);
+    return;
+  }
 }
 
 // Validate the type-defined mask compared to the store value mask which
@@ -4182,6 +4212,9 @@ static void ValidateResourceOverlap(
 
 static void ValidateResource(hlsl::DxilResource &Res,
                              ValidationContext &ValCtx) {
+  if (Res.IsReorderCoherent() && !ValCtx.DxilMod.GetShaderModel()->IsSM69Plus())
+    ValCtx.EmitResourceError(&Res,
+                             ValidationRule::InstrReorderCoherentRequiresSM69);
   switch (Res.GetKind()) {
   case DXIL::ResourceKind::RawBuffer:
   case DXIL::ResourceKind::TypedBuffer:
@@ -4413,10 +4446,13 @@ static void ValidateResources(ValidationContext &ValCtx) {
       ValCtx.EmitResourceError(Uav.get(),
                                ValidationRule::SmCounterOnlyOnStructBuf);
     }
-    if (Uav->HasCounter() && Uav->IsGloballyCoherent())
-      ValCtx.EmitResourceFormatError(Uav.get(),
-                                     ValidationRule::MetaGlcNotOnAppendConsume,
-                                     {ValCtx.GetResourceName(Uav.get())});
+    const bool UavIsCoherent =
+        Uav->IsGloballyCoherent() || Uav->IsReorderCoherent();
+    if (Uav->HasCounter() && UavIsCoherent) {
+      StringRef Prefix = Uav->IsGloballyCoherent() ? "globally" : "reorder";
+      ValCtx.EmitResourceFormatError(
+          Uav.get(), ValidationRule::MetaCoherenceNotOnAppendConsume, {Prefix});
+    }
 
     ValidateResource(*Uav, ValCtx);
     ValidateResourceOverlap(*Uav, UavAllocator, ValCtx);