adding clang codegen

clang/include/clang/Basic/Builtins.td

@@ -4788,6 +4788,12 @@ def HLSLStep: LangBuiltin<"HLSL_LANG"> {
   let Prototype = "void(...)";
 }
 
+def HLSLAsUintSplitDouble: LangBuiltin<"HLSL_LANG"> {
+  let Spellings = ["__builtin_hlsl_asuint_splitdouble"];
+  let Attributes = [NoThrow, Const];
+  let Prototype = "void(...)";
+}
+
 // Builtins for XRay.
 def XRayCustomEvent : Builtin {
   let Spellings = ["__xray_customevent"];

clang/lib/CodeGen/CGBuiltin.cpp

@@ -11,6 +11,7 @@
 //===----------------------------------------------------------------------===//
 
 #include "ABIInfo.h"
+#include "Address.h"
 #include "CGCUDARuntime.h"
 #include "CGCXXABI.h"
 #include "CGHLSLRuntime.h"
@@ -18824,6 +18825,33 @@ case Builtin::BI__builtin_hlsl_elementwise_isinf: {
         retType, CGM.getHLSLRuntime().getSignIntrinsic(),
         ArrayRef<Value *>{Op0}, nullptr, "hlsl.sign");
   }
+  // This should only be called when targeting DXIL
+  case Builtin::BI__builtin_hlsl_asuint_splitdouble: {
+
+    assert((E->getArg(0)->getType()->isDoubleType() &&
+            E->getArg(1)->getType()->isUnsignedIntegerType() &&
+            E->getArg(2)->getType()->isUnsignedIntegerType()) &&
+           "asuint operands types mismatch");
+
+    Value *Op0 = EmitScalarExpr(E->getArg(0));
+    llvm::Type *retType = llvm::StructType::get(Int32Ty, Int32Ty);
+    if (Op0->getType()->isVectorTy()) {
+      auto *XVecTy = E->getArg(0)->getType()->getAs<VectorType>();
+
+      llvm::VectorType *i32VecTy = llvm::VectorType::get(
+          Int32Ty, ElementCount::getFixed(XVecTy->getNumElements()));
+
+      retType = llvm::StructType::get(i32VecTy, i32VecTy);
+    }
+
+    CallInst *CI =
+        Builder.CreateIntrinsic(retType, llvm::Intrinsic::dx_asuint_splitdouble,
+                                {Op0}, nullptr, "hlsl.asuint");
+
+    Value *arg1 = Builder.CreateExtractValue(CI, 1);
+    auto addr = EmitPointerWithAlignment(E->getArg(1));
+    return Builder.CreateStore(arg1, addr);
+  }
   }
   return nullptr;
 }

clang/lib/CodeGen/CGExprScalar.cpp

@@ -43,6 +43,7 @@
 #include "llvm/IR/IntrinsicsPowerPC.h"
 #include "llvm/IR/MatrixBuilder.h"
 #include "llvm/IR/Module.h"
+#include "llvm/Support/ErrorHandling.h"
 #include "llvm/Support/TypeSize.h"
 #include <cstdarg>
 #include <optional>
@@ -476,6 +477,10 @@ class ScalarExprEmitter
     return Visit(E->getSubExpr());
   }
 
+  Value *VisitHLSLOutArgExpr(HLSLOutArgExpr *E) {
+    return Visit(E->getArgLValue());
+  }
+
   // Leaves.
   Value *VisitIntegerLiteral(const IntegerLiteral *E) {
     return Builder.getInt(E->getValue());

clang/lib/CodeGen/CGHLSLRuntime.h

@@ -87,7 +87,6 @@ class CGHLSLRuntime {
   GENERATE_HLSL_INTRINSIC_FUNCTION(SDot, sdot)
   GENERATE_HLSL_INTRINSIC_FUNCTION(UDot, udot)
   GENERATE_HLSL_INTRINSIC_FUNCTION(WaveIsFirstLane, wave_is_first_lane)
-
   //===----------------------------------------------------------------------===//
   // End of reserved area for HLSL intrinsic getters.
   //===----------------------------------------------------------------------===//

clang/lib/Headers/hlsl/hlsl_intrinsics.h

@@ -423,6 +423,26 @@ template <typename T> _HLSL_INLINE uint asuint(T F) {
   return __detail::bit_cast<uint, T>(F);
 }
 
+//===----------------------------------------------------------------------===//
+// asuint splitdouble builtins
+//===----------------------------------------------------------------------===//
+
+/// \fn void asuint(double D, out uint lowbits, out int highbits)
+/// \brief Split and interprets the lowbits and highbits of double D into uints.
+/// \param D The input double.
+/// \param lowbits The output lowbits of D.
+/// \param highbits The highbits lowbits D.
+#if __is_target_arch(dxil)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_asuint_splitdouble)
+void asuint(double, out uint, out uint);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_asuint_splitdouble)
+void asuint(double2, out uint2, out uint2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_asuint_splitdouble)
+void asuint(double3, out uint3, out uint3);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_asuint_splitdouble)
+void asuint(double4, out uint4, out uint4);
+#endif
+
 //===----------------------------------------------------------------------===//
 // atan builtins
 //===----------------------------------------------------------------------===//

clang/lib/Sema/SemaHLSL.cpp

@@ -1467,18 +1467,27 @@ bool CheckVectorElementCallArgs(Sema *S, CallExpr *TheCall) {
   return true;
 }
 
+bool CheckArgTypeIsCorrect(
+    Sema *S, Expr *Arg, QualType ExpectedType,
+    llvm::function_ref<bool(clang::QualType PassedType)> Check) {
+  QualType PassedType = Arg->getType();
+  if (Check(PassedType)) {
+    if (auto *VecTyA = PassedType->getAs<VectorType>())
+      ExpectedType = S->Context.getVectorType(
+          ExpectedType, VecTyA->getNumElements(), VecTyA->getVectorKind());
+    S->Diag(Arg->getBeginLoc(), diag::err_typecheck_convert_incompatible)
+        << PassedType << ExpectedType << 1 << 0 << 0;
+    return true;
+  }
+  return false;
+}
+
 bool CheckArgsTypesAreCorrect(
     Sema *S, CallExpr *TheCall, QualType ExpectedType,
     llvm::function_ref<bool(clang::QualType PassedType)> Check) {
   for (unsigned i = 0; i < TheCall->getNumArgs(); ++i) {
-    QualType PassedType = TheCall->getArg(i)->getType();
-    if (Check(PassedType)) {
-      if (auto *VecTyA = PassedType->getAs<VectorType>())
-        ExpectedType = S->Context.getVectorType(
-            ExpectedType, VecTyA->getNumElements(), VecTyA->getVectorKind());
-      S->Diag(TheCall->getArg(0)->getBeginLoc(),
-              diag::err_typecheck_convert_incompatible)
-          << PassedType << ExpectedType << 1 << 0 << 0;
+    Expr *Arg = TheCall->getArg(i);
+    if (CheckArgTypeIsCorrect(S, Arg, ExpectedType, Check)) {
       return true;
     }
   }
@@ -1762,6 +1771,37 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
       return true;
     break;
   }
+  case Builtin::BI__builtin_hlsl_asuint_splitdouble: {
+    if (SemaRef.checkArgCount(TheCall, 3))
+      return true;
+
+    Expr *Op0 = TheCall->getArg(0);
+
+    auto CheckIsNotDouble = [](clang::QualType PassedType) -> bool {
+      return !PassedType->isDoubleType();
+    };
+
+    if (CheckArgTypeIsCorrect(&SemaRef, Op0, SemaRef.Context.DoubleTy,
+                              CheckIsNotDouble)) {
+      return true;
+    }
+
+    Expr *Op1 = TheCall->getArg(1);
+    Expr *Op2 = TheCall->getArg(2);
+
+    auto CheckIsNotUint = [](clang::QualType PassedType) -> bool {
+      return !PassedType->isUnsignedIntegerType();
+    };
+
+    if (CheckArgTypeIsCorrect(&SemaRef, Op1, SemaRef.Context.UnsignedIntTy,
+                              CheckIsNotUint) ||
+        CheckArgTypeIsCorrect(&SemaRef, Op2, SemaRef.Context.UnsignedIntTy,
+                              CheckIsNotUint)) {
+      return true;
+    }
+
+    break;
+  }
   case Builtin::BI__builtin_elementwise_acos:
   case Builtin::BI__builtin_elementwise_asin:
   case Builtin::BI__builtin_elementwise_atan:

clang/test/CodeGenHLSL/builtins/asuint-splitdouble.hlsl

@@ -0,0 +1,9 @@
+// RUN: %clang_cc1 -finclude-default-header -x hlsl -triple dxil-pc-shadermodel6.3-library %s -fnative-half-type -emit-llvm -O0 -o - | FileCheck %s
+
+// CHECK: define {{.*}}test_scalar{{.*}}(double {{.*}} [[VAL1:%.*]], i32 {{.*}} [[VAL2:%.*]], i32 {{.*}} [[VAL3:%.*]]){{.*}} 
+// CHECK: [[VALD:%.*]] = load double, ptr [[VAL1]].addr{{.*}}
+// CHECK: call { i32, i32 } @llvm.dx.asuint.splitdouble.{{.*}}(double [[VALD]])
+uint test_scalar(double d, uint lb, uint hb) {
+  asuint(d, lb, hb);
+  return lb;
+}

llvm/include/llvm/IR/IntrinsicsDirectX.td

@@ -88,4 +88,9 @@ def int_dx_rsqrt  : DefaultAttrsIntrinsic<[llvm_anyfloat_ty], [LLVMMatchType<0>]
 def int_dx_wave_is_first_lane : DefaultAttrsIntrinsic<[llvm_i1_ty], [], [IntrConvergent]>;
 def int_dx_sign : DefaultAttrsIntrinsic<[LLVMScalarOrSameVectorWidth<0, llvm_i32_ty>], [llvm_any_ty], [IntrNoMem]>;
 def int_dx_step : DefaultAttrsIntrinsic<[LLVMMatchType<0>], [llvm_anyfloat_ty, LLVMMatchType<0>], [IntrNoMem]>;
+
+def int_dx_asuint_splitdouble : DefaultAttrsIntrinsic<
+    [llvm_anyint_ty, LLVMMatchType<0>], 
+    [LLVMScalarOrSameVectorWidth<0, llvm_double_ty>], 
+    [IntrNoMem, IntrWillReturn]>;
 }

llvm/lib/Target/DirectX/DXIL.td

@@ -778,6 +778,7 @@ def FlattenedThreadIdInGroup :  DXILOp<96, flattenedThreadIdInGroup> {
   let stages = [Stages<DXIL1_0, [compute, mesh, amplification, node]>];
   let attributes = [Attributes<DXIL1_0, [ReadNone]>];
 }
+//
 
 def AnnotateHandle : DXILOp<217, annotateHandle> {
   let Doc = "annotate handle with resource properties";

llvm/lib/Target/DirectX/DXILIntrinsicExpansion.cpp

@@ -12,6 +12,7 @@
 
 #include "DXILIntrinsicExpansion.h"
 #include "DirectX.h"
+#include "llvm-c/Core.h"
 #include "llvm/ADT/STLExtras.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/Analysis/DXILResource.h"
@@ -346,6 +347,15 @@ static Value *expandStepIntrinsic(CallInst *Orig) {
   return Builder.CreateSelect(Cond, Zero, One);
 }
 
+// static Value *expandSplitdoubleIntrinsic(CallInst *Orig) {
+//   Value *X = Orig->getOperand(0);
+//   Type *Ty = X->getType();
+//   IRBuilder<> Builder(Orig);
+
+//   Builder.CreateIntrinsic()
+
+// }
+
 static Intrinsic::ID getMaxForClamp(Type *ElemTy,
                                     Intrinsic::ID ClampIntrinsic) {
   if (ClampIntrinsic == Intrinsic::dx_uclamp)
@@ -459,6 +469,9 @@ static bool expandIntrinsic(Function &F, CallInst *Orig) {
     break;
   case Intrinsic::dx_step:
     Result = expandStepIntrinsic(Orig);
+    break;
+    // case Intrinsic::dx_asuint_splitdouble:
+    //   Result = expandSplitdoubleIntrinsic(Orig);
   }
   if (Result) {
     Orig->replaceAllUsesWith(Result);