[clang] [llvm] [HLSL] Implement dot2add intrinsic (PR #131237)

Sumit Agarwal via llvm-commits llvm-commits at lists.llvm.org
Thu Mar 27 10:16:35 PDT 2025


https://github.com/sumitsays updated https://github.com/llvm/llvm-project/pull/131237

>From 6d5c4053c90975d64e378e52779dab9c3ffb64cd Mon Sep 17 00:00:00 2001
From: Sumit Agarwal <sumitagarwal at microsoft.com>
Date: Thu, 13 Mar 2025 16:02:32 -0700
Subject: [PATCH 01/12] dot2add working for dxil without sema check

---
 clang/include/clang/Basic/Builtins.td      |   6 +
 clang/lib/CodeGen/CGBuiltin.cpp            |  10 +
 clang/lib/CodeGen/CGHLSLRuntime.h          |   1 +
 clang/lib/Headers/hlsl/hlsl_intrinsics.h   | 366 +++++++++++++++++++++
 llvm/include/llvm/IR/IntrinsicsDirectX.td  |   4 +
 llvm/include/llvm/IR/IntrinsicsSPIRV.td    |   5 +
 llvm/lib/Target/DirectX/DXIL.td            |  11 +
 llvm/lib/Target/DirectX/DXILOpLowering.cpp |  33 ++
 8 files changed, 436 insertions(+)

diff --git a/clang/include/clang/Basic/Builtins.td b/clang/include/clang/Basic/Builtins.td
index 72a5e495c4059..8a0237171f7df 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4891,6 +4891,12 @@ def HLSLDotProduct : LangBuiltin<"HLSL_LANG"> {
   let Prototype = "void(...)";
 }
 
+def HLSLDot2Add : LangBuiltin<"HLSL_LANG"> {
+  let Spellings = ["__builtin_hlsl_dot2add"];
+  let Attributes = [NoThrow, Const];
+  let Prototype = "void(...)";
+}
+
 def HLSLDot4AddI8Packed : LangBuiltin<"HLSL_LANG"> {
   let Spellings = ["__builtin_hlsl_dot4add_i8packed"];
   let Attributes = [NoThrow, Const];
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index c126f88b9e3a5..8c0f42e0c85f7 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -19681,6 +19681,16 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
         getDotProductIntrinsic(CGM.getHLSLRuntime(), VecTy0->getElementType()),
         ArrayRef<Value *>{Op0, Op1}, nullptr, "hlsl.dot");
   }
+  case Builtin::BI__builtin_hlsl_dot2add: {
+    Value *A = EmitScalarExpr(E->getArg(0));
+    Value *B = EmitScalarExpr(E->getArg(1));
+    Value *C = EmitScalarExpr(E->getArg(2));
+
+    Intrinsic::ID ID = CGM.getHLSLRuntime().getDot2AddIntrinsic();
+    return Builder.CreateIntrinsic(
+        /*ReturnType=*/C->getType(), ID, ArrayRef<Value *>{A, B, C}, nullptr,
+        "hlsl.dot2add");
+  }
   case Builtin::BI__builtin_hlsl_dot4add_i8packed: {
     Value *A = EmitScalarExpr(E->getArg(0));
     Value *B = EmitScalarExpr(E->getArg(1));
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h b/clang/lib/CodeGen/CGHLSLRuntime.h
index 23f187b24284f..5ba69819e5431 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.h
+++ b/clang/lib/CodeGen/CGHLSLRuntime.h
@@ -99,6 +99,7 @@ class CGHLSLRuntime {
   GENERATE_HLSL_INTRINSIC_FUNCTION(FDot, fdot)
   GENERATE_HLSL_INTRINSIC_FUNCTION(SDot, sdot)
   GENERATE_HLSL_INTRINSIC_FUNCTION(UDot, udot)
+  GENERATE_HLSL_INTRINSIC_FUNCTION(Dot2Add, dot2add)
   GENERATE_HLSL_INTRINSIC_FUNCTION(Dot4AddI8Packed, dot4add_i8packed)
   GENERATE_HLSL_INTRINSIC_FUNCTION(Dot4AddU8Packed, dot4add_u8packed)
   GENERATE_HLSL_INTRINSIC_FUNCTION(WaveActiveAllTrue, wave_all)
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index 5459cbeb34fd0..38422229e018d 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -118,6 +118,372 @@ const inline float distance(__detail::HLSL_FIXED_VECTOR<float, N> X,
 }
 
 //===----------------------------------------------------------------------===//
+<<<<<<< Updated upstream
+=======
+// dot product builtins
+//===----------------------------------------------------------------------===//
+
+/// \fn K dot(T X, T Y)
+/// \brief Return the dot product (a scalar value) of \a X and \a Y.
+/// \param X The X input value.
+/// \param Y The Y input value.
+
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half, half);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half2, half2);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half3, half3);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+half dot(half4, half4);
+
+#ifdef __HLSL_ENABLE_16_BIT
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t, int16_t);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t2, int16_t2);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t3, int16_t3);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int16_t dot(int16_t4, int16_t4);
+
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint16_t dot(uint16_t, uint16_t);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint16_t dot(uint16_t2, uint16_t2);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint16_t dot(uint16_t3, uint16_t3);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint16_t dot(uint16_t4, uint16_t4);
+#endif
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+float dot(float, float);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+float dot(float2, float2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+float dot(float3, float3);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+float dot(float4, float4);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+double dot(double, double);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int dot(int, int);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int dot(int2, int2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int dot(int3, int3);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int dot(int4, int4);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint dot(uint, uint);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint dot(uint2, uint2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint dot(uint3, uint3);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint dot(uint4, uint4);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int64_t dot(int64_t, int64_t);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int64_t dot(int64_t2, int64_t2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int64_t dot(int64_t3, int64_t3);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+int64_t dot(int64_t4, int64_t4);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint64_t dot(uint64_t, uint64_t);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint64_t dot(uint64_t2, uint64_t2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint64_t dot(uint64_t3, uint64_t3);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
+uint64_t dot(uint64_t4, uint64_t4);
+
+//===----------------------------------------------------------------------===//
+// dot4add builtins
+//===----------------------------------------------------------------------===//
+
+/// \fn int dot4add_i8packed(uint A, uint B, int C)
+
+_HLSL_AVAILABILITY(shadermodel, 6.4)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot4add_i8packed)
+int dot4add_i8packed(uint, uint, int);
+
+/// \fn uint dot4add_u8packed(uint A, uint B, uint C)
+
+_HLSL_AVAILABILITY(shadermodel, 6.4)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot4add_u8packed)
+uint dot4add_u8packed(uint, uint, uint);
+
+//===----------------------------------------------------------------------===//
+// dot2add builtins
+//===----------------------------------------------------------------------===//
+
+/// \fn float dot2add(half2 a, half2 b, float c)
+
+_HLSL_AVAILABILITY(shadermodel, 6.4)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot2add)
+float dot2add(half2, half2, float);
+
+//===----------------------------------------------------------------------===//
+// exp builtins
+//===----------------------------------------------------------------------===//
+
+/// \fn T exp(T x)
+/// \brief Returns the base-e exponential, or \a e**x, of the specified value.
+/// \param x The specified input value.
+///
+/// The return value is the base-e exponential of the \a x parameter.
+
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_elementwise_exp)
+half exp(half);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_elementwise_exp)
+half2 exp(half2);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_elementwise_exp)
+half3 exp(half3);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_elementwise_exp)
+half4 exp(half4);
+
+_HLSL_BUILTIN_ALIAS(__builtin_elementwise_exp)
+float exp(float);
+_HLSL_BUILTIN_ALIAS(__builtin_elementwise_exp)
+float2 exp(float2);
+_HLSL_BUILTIN_ALIAS(__builtin_elementwise_exp)
+float3 exp(float3);
+_HLSL_BUILTIN_ALIAS(__builtin_elementwise_exp)
+float4 exp(float4);
+
+//===----------------------------------------------------------------------===//
+// exp2 builtins
+//===----------------------------------------------------------------------===//
+
+/// \fn T exp2(T x)
+/// \brief Returns the base 2 exponential, or \a 2**x, of the specified value.
+/// \param x The specified input value.
+///
+/// The base 2 exponential of the \a x parameter.
+
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_elementwise_exp2)
+half exp2(half);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_elementwise_exp2)
+half2 exp2(half2);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_elementwise_exp2)
+half3 exp2(half3);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_elementwise_exp2)
+half4 exp2(half4);
+
+_HLSL_BUILTIN_ALIAS(__builtin_elementwise_exp2)
+float exp2(float);
+_HLSL_BUILTIN_ALIAS(__builtin_elementwise_exp2)
+float2 exp2(float2);
+_HLSL_BUILTIN_ALIAS(__builtin_elementwise_exp2)
+float3 exp2(float3);
+_HLSL_BUILTIN_ALIAS(__builtin_elementwise_exp2)
+float4 exp2(float4);
+
+//===----------------------------------------------------------------------===//
+// firstbithigh builtins
+//===----------------------------------------------------------------------===//
+
+/// \fn T firstbithigh(T Val)
+/// \brief Returns the location of the first set bit starting from the highest
+/// order bit and working downward, per component.
+/// \param Val the input value.
+
+#ifdef __HLSL_ENABLE_16_BIT
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
+uint firstbithigh(int16_t);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
+uint2 firstbithigh(int16_t2);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
+uint3 firstbithigh(int16_t3);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
+uint4 firstbithigh(int16_t4);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
+uint firstbithigh(uint16_t);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
+uint2 firstbithigh(uint16_t2);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
+uint3 firstbithigh(uint16_t3);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
+uint4 firstbithigh(uint16_t4);
+#endif
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
+uint firstbithigh(int);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
+uint2 firstbithigh(int2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
+uint3 firstbithigh(int3);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
+uint4 firstbithigh(int4);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
+uint firstbithigh(uint);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
+uint2 firstbithigh(uint2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
+uint3 firstbithigh(uint3);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
+uint4 firstbithigh(uint4);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
+uint firstbithigh(int64_t);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
+uint2 firstbithigh(int64_t2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
+uint3 firstbithigh(int64_t3);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
+uint4 firstbithigh(int64_t4);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
+uint firstbithigh(uint64_t);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
+uint2 firstbithigh(uint64_t2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
+uint3 firstbithigh(uint64_t3);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
+uint4 firstbithigh(uint64_t4);
+
+//===----------------------------------------------------------------------===//
+// firstbitlow builtins
+//===----------------------------------------------------------------------===//
+
+/// \fn T firstbitlow(T Val)
+/// \brief Returns the location of the first set bit starting from the lowest
+/// order bit and working upward, per component.
+/// \param Val the input value.
+
+#ifdef __HLSL_ENABLE_16_BIT
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
+uint firstbitlow(int16_t);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
+uint2 firstbitlow(int16_t2);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
+uint3 firstbitlow(int16_t3);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
+uint4 firstbitlow(int16_t4);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
+uint firstbitlow(uint16_t);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
+uint2 firstbitlow(uint16_t2);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
+uint3 firstbitlow(uint16_t3);
+_HLSL_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
+uint4 firstbitlow(uint16_t4);
+#endif
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
+uint firstbitlow(int);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
+uint2 firstbitlow(int2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
+uint3 firstbitlow(int3);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
+uint4 firstbitlow(int4);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
+uint firstbitlow(uint);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
+uint2 firstbitlow(uint2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
+uint3 firstbitlow(uint3);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
+uint4 firstbitlow(uint4);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
+uint firstbitlow(int64_t);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
+uint2 firstbitlow(int64_t2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
+uint3 firstbitlow(int64_t3);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
+uint4 firstbitlow(int64_t4);
+
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
+uint firstbitlow(uint64_t);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
+uint2 firstbitlow(uint64_t2);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
+uint3 firstbitlow(uint64_t3);
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
+uint4 firstbitlow(uint64_t4);
+
+//===----------------------------------------------------------------------===//
+// floor builtins
+//===----------------------------------------------------------------------===//
+
+/// \fn T floor(T Val)
+/// \brief Returns the largest integer that is less than or equal to the input
+/// value, \a Val.
+/// \param Val The input value.
+
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_elementwise_floor)
+half floor(half);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_elementwise_floor)
+half2 floor(half2);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_elementwise_floor)
+half3 floor(half3);
+_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
+_HLSL_BUILTIN_ALIAS(__builtin_elementwise_floor)
+half4 floor(half4);
+
+_HLSL_BUILTIN_ALIAS(__builtin_elementwise_floor)
+float floor(float);
+_HLSL_BUILTIN_ALIAS(__builtin_elementwise_floor)
+float2 floor(float2);
+_HLSL_BUILTIN_ALIAS(__builtin_elementwise_floor)
+float3 floor(float3);
+_HLSL_BUILTIN_ALIAS(__builtin_elementwise_floor)
+float4 floor(float4);
+
+//===----------------------------------------------------------------------===//
+>>>>>>> Stashed changes
 // fmod builtins
 //===----------------------------------------------------------------------===//
 
diff --git a/llvm/include/llvm/IR/IntrinsicsDirectX.td b/llvm/include/llvm/IR/IntrinsicsDirectX.td
index ead7286f4311c..775d325feeb14 100644
--- a/llvm/include/llvm/IR/IntrinsicsDirectX.td
+++ b/llvm/include/llvm/IR/IntrinsicsDirectX.td
@@ -100,6 +100,10 @@ def int_dx_udot :
     DefaultAttrsIntrinsic<[LLVMVectorElementType<0>],
     [llvm_anyint_ty, LLVMScalarOrSameVectorWidth<0, LLVMVectorElementType<0>>],
     [IntrNoMem, Commutative] >;
+def int_dx_dot2add : 
+    DefaultAttrsIntrinsic<[llvm_float_ty], 
+    [llvm_anyfloat_ty, LLVMMatchType<0>, llvm_float_ty], 
+    [IntrNoMem, Commutative]>;
 def int_dx_dot4add_i8packed : DefaultAttrsIntrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
 def int_dx_dot4add_u8packed : DefaultAttrsIntrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
 
diff --git a/llvm/include/llvm/IR/IntrinsicsSPIRV.td b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
index df3e137c80980..240c36513f0b6 100644
--- a/llvm/include/llvm/IR/IntrinsicsSPIRV.td
+++ b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
@@ -87,6 +87,11 @@ let TargetPrefix = "spv" in {
     DefaultAttrsIntrinsic<[LLVMVectorElementType<0>],
     [llvm_anyint_ty, LLVMScalarOrSameVectorWidth<0, LLVMVectorElementType<0>>],
     [IntrNoMem, Commutative] >;
+
+  def int_spv_dot2add : DefaultAttrsIntrinsic<[llvm_float_ty], 
+                    [llvm_anyfloat_ty, LLVMMatchType<0>, llvm_float_ty], 
+                    [IntrNoMem, Commutative]>;
+
   def int_spv_dot4add_i8packed : DefaultAttrsIntrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
   def int_spv_dot4add_u8packed : DefaultAttrsIntrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
   def int_spv_wave_active_countbits : DefaultAttrsIntrinsic<[llvm_i32_ty], [llvm_i1_ty], [IntrConvergent, IntrNoMem]>;
diff --git a/llvm/lib/Target/DirectX/DXIL.td b/llvm/lib/Target/DirectX/DXIL.td
index ebe1d876d58b1..193b592a525a0 100644
--- a/llvm/lib/Target/DirectX/DXIL.td
+++ b/llvm/lib/Target/DirectX/DXIL.td
@@ -1098,6 +1098,17 @@ def RawBufferStore : DXILOp<140, rawBufferStore> {
   let stages = [Stages<DXIL1_2, [all_stages]>];
 }
 
+def Dot2AddHalf : DXILOp<162, dot2AddHalf> {
+  let Doc = "dot product of 2 vectors of half having size = 2, returns "
+            "float";
+  let intrinsics = [IntrinSelect<int_dx_dot2add>];
+  let arguments = [FloatTy, HalfTy, HalfTy, HalfTy, HalfTy];
+  let result = FloatTy;
+  let overloads = [Overloads<DXIL1_0, []>];
+  let stages = [Stages<DXIL1_0, [all_stages]>];
+  let attributes = [Attributes<DXIL1_0, [ReadNone]>];
+}
+
 def Dot4AddI8Packed : DXILOp<163, dot4AddPacked> {
   let Doc = "signed dot product of 4 x i8 vectors packed into i32, with "
             "accumulate to i32";
diff --git a/llvm/lib/Target/DirectX/DXILOpLowering.cpp b/llvm/lib/Target/DirectX/DXILOpLowering.cpp
index dff9f3e03079e..10a726aba5ff3 100644
--- a/llvm/lib/Target/DirectX/DXILOpLowering.cpp
+++ b/llvm/lib/Target/DirectX/DXILOpLowering.cpp
@@ -75,6 +75,27 @@ static SmallVector<Value *> argVectorFlatten(CallInst *Orig,
   return NewOperands;
 }
 
+static SmallVector<Value *> argVectorFlattenExcludeLastElement(CallInst *Orig,
+                                                               IRBuilder<> &Builder) {
+  // Note: arg[NumOperands-1] is a pointer and is not needed by our flattening.
+  unsigned NumOperands = Orig->getNumOperands() - 2;
+  assert(NumOperands > 0);
+  Value *Arg0 = Orig->getOperand(0);
+  [[maybe_unused]] auto *VecArg0 = dyn_cast<FixedVectorType>(Arg0->getType());
+  assert(VecArg0);
+  SmallVector<Value *> NewOperands = populateOperands(Arg0, Builder);
+  for (unsigned I = 1; I < NumOperands; ++I) {
+    Value *Arg = Orig->getOperand(I);
+    [[maybe_unused]] auto *VecArg = dyn_cast<FixedVectorType>(Arg->getType());
+    assert(VecArg);
+    assert(VecArg0->getElementType() == VecArg->getElementType());
+    assert(VecArg0->getNumElements() == VecArg->getNumElements());
+    auto NextOperandList = populateOperands(Arg, Builder);
+    NewOperands.append(NextOperandList.begin(), NextOperandList.end());
+  }
+  return NewOperands;
+}
+
 namespace {
 class OpLowerer {
   Module &M;
@@ -168,6 +189,18 @@ class OpLowerer {
         }
       } else if (IsVectorArgExpansion) {
         Args = argVectorFlatten(CI, OpBuilder.getIRB());
+      } else if (F.getIntrinsicID() == Intrinsic::dx_dot2add) {
+        unsigned NumOperands = CI->getNumOperands() - 1;
+        assert(NumOperands > 0);
+        Value *LastArg = CI->getOperand(NumOperands - 1);
+        
+        Args.push_back(LastArg);
+        
+        //dbgs() << "Value of LastArg" << LastArg->getName() << "\n";
+        
+        
+        //Args = populateOperands(LastArg, OpBuilder.getIRB());
+        Args.append(argVectorFlattenExcludeLastElement(CI, OpBuilder.getIRB()));
       } else {
         Args.append(CI->arg_begin(), CI->arg_end());
       }

>From 3e265ffc7ba3486027e63105acf2064699e58807 Mon Sep 17 00:00:00 2001
From: Sumit Agarwal <sumitagarwal at microsoft.com>
Date: Tue, 18 Mar 2025 17:33:30 -0700
Subject: [PATCH 02/12] WIP

---
 .../lib/Headers/hlsl/hlsl_alias_intrinsics.h  |  10 +
 clang/lib/Headers/hlsl/hlsl_intrinsics.h      | 366 ------------------
 clang/test/CodeGenHLSL/builtins/dot2add.hlsl  |  15 +
 llvm/test/CodeGen/DirectX/dot2add.ll          |   8 +
 4 files changed, 33 insertions(+), 366 deletions(-)
 create mode 100644 clang/test/CodeGenHLSL/builtins/dot2add.hlsl
 create mode 100644 llvm/test/CodeGen/DirectX/dot2add.ll

diff --git a/clang/lib/Headers/hlsl/hlsl_alias_intrinsics.h b/clang/lib/Headers/hlsl/hlsl_alias_intrinsics.h
index 62054b368691d..b7b9369b3ace9 100644
--- a/clang/lib/Headers/hlsl/hlsl_alias_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_alias_intrinsics.h
@@ -982,6 +982,16 @@ uint64_t dot(uint64_t3, uint64_t3);
 _HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
 uint64_t dot(uint64_t4, uint64_t4);
 
+//===----------------------------------------------------------------------===//
+// dot2add builtins
+//===----------------------------------------------------------------------===//
+
+/// \fn float dot2add(half2 a, half2 b, float c)
+
+_HLSL_AVAILABILITY(shadermodel, 6.4)
+_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot2add)
+float dot2add(half2, half2, float);
+
 //===----------------------------------------------------------------------===//
 // dot4add builtins
 //===----------------------------------------------------------------------===//
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index 38422229e018d..5459cbeb34fd0 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -118,372 +118,6 @@ const inline float distance(__detail::HLSL_FIXED_VECTOR<float, N> X,
 }
 
 //===----------------------------------------------------------------------===//
-<<<<<<< Updated upstream
-=======
-// dot product builtins
-//===----------------------------------------------------------------------===//
-
-/// \fn K dot(T X, T Y)
-/// \brief Return the dot product (a scalar value) of \a X and \a Y.
-/// \param X The X input value.
-/// \param Y The Y input value.
-
-_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
-half dot(half, half);
-_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
-half dot(half2, half2);
-_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
-half dot(half3, half3);
-_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
-half dot(half4, half4);
-
-#ifdef __HLSL_ENABLE_16_BIT
-_HLSL_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
-int16_t dot(int16_t, int16_t);
-_HLSL_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
-int16_t dot(int16_t2, int16_t2);
-_HLSL_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
-int16_t dot(int16_t3, int16_t3);
-_HLSL_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
-int16_t dot(int16_t4, int16_t4);
-
-_HLSL_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
-uint16_t dot(uint16_t, uint16_t);
-_HLSL_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
-uint16_t dot(uint16_t2, uint16_t2);
-_HLSL_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
-uint16_t dot(uint16_t3, uint16_t3);
-_HLSL_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
-uint16_t dot(uint16_t4, uint16_t4);
-#endif
-
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
-float dot(float, float);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
-float dot(float2, float2);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
-float dot(float3, float3);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
-float dot(float4, float4);
-
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
-double dot(double, double);
-
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
-int dot(int, int);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
-int dot(int2, int2);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
-int dot(int3, int3);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
-int dot(int4, int4);
-
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
-uint dot(uint, uint);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
-uint dot(uint2, uint2);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
-uint dot(uint3, uint3);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
-uint dot(uint4, uint4);
-
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
-int64_t dot(int64_t, int64_t);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
-int64_t dot(int64_t2, int64_t2);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
-int64_t dot(int64_t3, int64_t3);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
-int64_t dot(int64_t4, int64_t4);
-
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
-uint64_t dot(uint64_t, uint64_t);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
-uint64_t dot(uint64_t2, uint64_t2);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
-uint64_t dot(uint64_t3, uint64_t3);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
-uint64_t dot(uint64_t4, uint64_t4);
-
-//===----------------------------------------------------------------------===//
-// dot4add builtins
-//===----------------------------------------------------------------------===//
-
-/// \fn int dot4add_i8packed(uint A, uint B, int C)
-
-_HLSL_AVAILABILITY(shadermodel, 6.4)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot4add_i8packed)
-int dot4add_i8packed(uint, uint, int);
-
-/// \fn uint dot4add_u8packed(uint A, uint B, uint C)
-
-_HLSL_AVAILABILITY(shadermodel, 6.4)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot4add_u8packed)
-uint dot4add_u8packed(uint, uint, uint);
-
-//===----------------------------------------------------------------------===//
-// dot2add builtins
-//===----------------------------------------------------------------------===//
-
-/// \fn float dot2add(half2 a, half2 b, float c)
-
-_HLSL_AVAILABILITY(shadermodel, 6.4)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot2add)
-float dot2add(half2, half2, float);
-
-//===----------------------------------------------------------------------===//
-// exp builtins
-//===----------------------------------------------------------------------===//
-
-/// \fn T exp(T x)
-/// \brief Returns the base-e exponential, or \a e**x, of the specified value.
-/// \param x The specified input value.
-///
-/// The return value is the base-e exponential of the \a x parameter.
-
-_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_elementwise_exp)
-half exp(half);
-_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_elementwise_exp)
-half2 exp(half2);
-_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_elementwise_exp)
-half3 exp(half3);
-_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_elementwise_exp)
-half4 exp(half4);
-
-_HLSL_BUILTIN_ALIAS(__builtin_elementwise_exp)
-float exp(float);
-_HLSL_BUILTIN_ALIAS(__builtin_elementwise_exp)
-float2 exp(float2);
-_HLSL_BUILTIN_ALIAS(__builtin_elementwise_exp)
-float3 exp(float3);
-_HLSL_BUILTIN_ALIAS(__builtin_elementwise_exp)
-float4 exp(float4);
-
-//===----------------------------------------------------------------------===//
-// exp2 builtins
-//===----------------------------------------------------------------------===//
-
-/// \fn T exp2(T x)
-/// \brief Returns the base 2 exponential, or \a 2**x, of the specified value.
-/// \param x The specified input value.
-///
-/// The base 2 exponential of the \a x parameter.
-
-_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_elementwise_exp2)
-half exp2(half);
-_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_elementwise_exp2)
-half2 exp2(half2);
-_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_elementwise_exp2)
-half3 exp2(half3);
-_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_elementwise_exp2)
-half4 exp2(half4);
-
-_HLSL_BUILTIN_ALIAS(__builtin_elementwise_exp2)
-float exp2(float);
-_HLSL_BUILTIN_ALIAS(__builtin_elementwise_exp2)
-float2 exp2(float2);
-_HLSL_BUILTIN_ALIAS(__builtin_elementwise_exp2)
-float3 exp2(float3);
-_HLSL_BUILTIN_ALIAS(__builtin_elementwise_exp2)
-float4 exp2(float4);
-
-//===----------------------------------------------------------------------===//
-// firstbithigh builtins
-//===----------------------------------------------------------------------===//
-
-/// \fn T firstbithigh(T Val)
-/// \brief Returns the location of the first set bit starting from the highest
-/// order bit and working downward, per component.
-/// \param Val the input value.
-
-#ifdef __HLSL_ENABLE_16_BIT
-_HLSL_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint firstbithigh(int16_t);
-_HLSL_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint2 firstbithigh(int16_t2);
-_HLSL_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint3 firstbithigh(int16_t3);
-_HLSL_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint4 firstbithigh(int16_t4);
-_HLSL_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint firstbithigh(uint16_t);
-_HLSL_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint2 firstbithigh(uint16_t2);
-_HLSL_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint3 firstbithigh(uint16_t3);
-_HLSL_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint4 firstbithigh(uint16_t4);
-#endif
-
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint firstbithigh(int);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint2 firstbithigh(int2);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint3 firstbithigh(int3);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint4 firstbithigh(int4);
-
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint firstbithigh(uint);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint2 firstbithigh(uint2);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint3 firstbithigh(uint3);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint4 firstbithigh(uint4);
-
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint firstbithigh(int64_t);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint2 firstbithigh(int64_t2);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint3 firstbithigh(int64_t3);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint4 firstbithigh(int64_t4);
-
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint firstbithigh(uint64_t);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint2 firstbithigh(uint64_t2);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint3 firstbithigh(uint64_t3);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbithigh)
-uint4 firstbithigh(uint64_t4);
-
-//===----------------------------------------------------------------------===//
-// firstbitlow builtins
-//===----------------------------------------------------------------------===//
-
-/// \fn T firstbitlow(T Val)
-/// \brief Returns the location of the first set bit starting from the lowest
-/// order bit and working upward, per component.
-/// \param Val the input value.
-
-#ifdef __HLSL_ENABLE_16_BIT
-_HLSL_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
-uint firstbitlow(int16_t);
-_HLSL_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
-uint2 firstbitlow(int16_t2);
-_HLSL_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
-uint3 firstbitlow(int16_t3);
-_HLSL_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
-uint4 firstbitlow(int16_t4);
-_HLSL_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
-uint firstbitlow(uint16_t);
-_HLSL_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
-uint2 firstbitlow(uint16_t2);
-_HLSL_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
-uint3 firstbitlow(uint16_t3);
-_HLSL_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
-uint4 firstbitlow(uint16_t4);
-#endif
-
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
-uint firstbitlow(int);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
-uint2 firstbitlow(int2);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
-uint3 firstbitlow(int3);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
-uint4 firstbitlow(int4);
-
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
-uint firstbitlow(uint);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
-uint2 firstbitlow(uint2);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
-uint3 firstbitlow(uint3);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
-uint4 firstbitlow(uint4);
-
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
-uint firstbitlow(int64_t);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
-uint2 firstbitlow(int64_t2);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
-uint3 firstbitlow(int64_t3);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
-uint4 firstbitlow(int64_t4);
-
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
-uint firstbitlow(uint64_t);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
-uint2 firstbitlow(uint64_t2);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
-uint3 firstbitlow(uint64_t3);
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_elementwise_firstbitlow)
-uint4 firstbitlow(uint64_t4);
-
-//===----------------------------------------------------------------------===//
-// floor builtins
-//===----------------------------------------------------------------------===//
-
-/// \fn T floor(T Val)
-/// \brief Returns the largest integer that is less than or equal to the input
-/// value, \a Val.
-/// \param Val The input value.
-
-_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_elementwise_floor)
-half floor(half);
-_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_elementwise_floor)
-half2 floor(half2);
-_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_elementwise_floor)
-half3 floor(half3);
-_HLSL_16BIT_AVAILABILITY(shadermodel, 6.2)
-_HLSL_BUILTIN_ALIAS(__builtin_elementwise_floor)
-half4 floor(half4);
-
-_HLSL_BUILTIN_ALIAS(__builtin_elementwise_floor)
-float floor(float);
-_HLSL_BUILTIN_ALIAS(__builtin_elementwise_floor)
-float2 floor(float2);
-_HLSL_BUILTIN_ALIAS(__builtin_elementwise_floor)
-float3 floor(float3);
-_HLSL_BUILTIN_ALIAS(__builtin_elementwise_floor)
-float4 floor(float4);
-
-//===----------------------------------------------------------------------===//
->>>>>>> Stashed changes
 // fmod builtins
 //===----------------------------------------------------------------------===//
 
diff --git a/clang/test/CodeGenHLSL/builtins/dot2add.hlsl b/clang/test/CodeGenHLSL/builtins/dot2add.hlsl
new file mode 100644
index 0000000000000..07e438063c29d
--- /dev/null
+++ b/clang/test/CodeGenHLSL/builtins/dot2add.hlsl
@@ -0,0 +1,15 @@
+// RUN: %clang_cc1 -finclude-default-header -triple \
+// RUN:   dxil-pc-shadermodel6.3-compute %s -emit-llvm -disable-llvm-passes -o - | \
+// RUN:   FileCheck %s -DTARGET=dx
+// RUN: %clang_cc1 -finclude-default-header -triple \
+// RUN:   spirv-pc-vulkan-compute %s -emit-llvm -disable-llvm-passes -o - | \
+// RUN:   FileCheck %s -DTARGET=spv
+
+// Test basic lowering to runtime function call.
+
+// CHECK-LABEL: test
+float test(half2 p1, half2 p2, float p3) {
+  return dot2add(p1, p2, p3);
+}
+
+// CHECK: declare [[TY]] @llvm.[[TARGET]].dot4add.i8packed([[TY]], [[TY]], [[TY]])
diff --git a/llvm/test/CodeGen/DirectX/dot2add.ll b/llvm/test/CodeGen/DirectX/dot2add.ll
new file mode 100644
index 0000000000000..b1019c36b56e8
--- /dev/null
+++ b/llvm/test/CodeGen/DirectX/dot2add.ll
@@ -0,0 +1,8 @@
+; RUN: opt -S -dxil-op-lower -mtriple=dxil-pc-shadermodel6.3-compute %s | FileCheck %s
+
+define noundef float @dot2add_simple(<2 x half> noundef %a, <2 x half> noundef %b, float %c) {
+entry:
+; CHECK: call float @dx.op.dot2AddHalf(i32 162, float %c, half %0, half %1, half %2, half %3)
+  %ret = call float @llvm.dx.dot2add(<2 x half> %a, <2 x half> %b, float %c)
+  ret float %ret
+}
\ No newline at end of file

>From 7baee968d04f37b63072994f60cb20185dadb11a Mon Sep 17 00:00:00 2001
From: Sumit Agarwal <sumitagarwal at microsoft.com>
Date: Wed, 19 Mar 2025 02:53:09 -0700
Subject: [PATCH 03/12] WIP

---
 clang/test/CodeGenHLSL/builtins/dot2add.hlsl | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

diff --git a/clang/test/CodeGenHLSL/builtins/dot2add.hlsl b/clang/test/CodeGenHLSL/builtins/dot2add.hlsl
index 07e438063c29d..ee7cb79bb26c4 100644
--- a/clang/test/CodeGenHLSL/builtins/dot2add.hlsl
+++ b/clang/test/CodeGenHLSL/builtins/dot2add.hlsl
@@ -1,14 +1,13 @@
 // RUN: %clang_cc1 -finclude-default-header -triple \
 // RUN:   dxil-pc-shadermodel6.3-compute %s -emit-llvm -disable-llvm-passes -o - | \
 // RUN:   FileCheck %s -DTARGET=dx
-// RUN: %clang_cc1 -finclude-default-header -triple \
-// RUN:   spirv-pc-vulkan-compute %s -emit-llvm -disable-llvm-passes -o - | \
-// RUN:   FileCheck %s -DTARGET=spv
 
 // Test basic lowering to runtime function call.
 
 // CHECK-LABEL: test
 float test(half2 p1, half2 p2, float p3) {
+  // CHECK-DXIL:  %hlsl.dot2add = call reassoc nnan ninf nsz arcp afn float @llvm.dx.dot2add.v2f32(<2 x float> %0, <2 x float> %1, float %2)
+  // CHECK: ret float %hlsl.dot2add
   return dot2add(p1, p2, p3);
 }
 

>From 0c5a6af1cc975cb4253f61d54622b40a62d3c373 Mon Sep 17 00:00:00 2001
From: Sumit Agarwal <sumitagarwal at microsoft.com>
Date: Wed, 19 Mar 2025 19:30:56 -0700
Subject: [PATCH 04/12] WIP: Mostly wrote everything.

---
 clang/lib/CodeGen/CGBuiltin.cpp               |  7 ++-
 clang/lib/CodeGen/CGHLSLRuntime.h             |  1 -
 .../lib/Headers/hlsl/hlsl_alias_intrinsics.h  | 10 -----
 .../lib/Headers/hlsl/hlsl_intrinsic_helpers.h |  8 ++++
 clang/lib/Headers/hlsl/hlsl_intrinsics.h      | 12 +++++
 clang/lib/Sema/SemaHLSL.cpp                   | 45 +++++++++++++++----
 .../SemaHLSL/BuiltIns/Dot2Add-errors.hlsl     | 11 +++++
 llvm/include/llvm/IR/IntrinsicsSPIRV.td       |  5 ---
 llvm/lib/Target/DirectX/DXILOpLowering.cpp    | 22 ++++++---
 9 files changed, 90 insertions(+), 31 deletions(-)
 create mode 100644 clang/test/SemaHLSL/BuiltIns/Dot2Add-errors.hlsl

diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 8c0f42e0c85f7..b3d9db5be7d8d 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -19682,11 +19682,16 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
         ArrayRef<Value *>{Op0, Op1}, nullptr, "hlsl.dot");
   }
   case Builtin::BI__builtin_hlsl_dot2add: {
+    llvm::Triple::ArchType Arch = CGM.getTarget().getTriple().getArch();
+    if (Arch != llvm::Triple::dxil) {
+      llvm_unreachable("Intrinsic dot2add can be executed as a builtin only on dxil");
+    }
     Value *A = EmitScalarExpr(E->getArg(0));
     Value *B = EmitScalarExpr(E->getArg(1));
     Value *C = EmitScalarExpr(E->getArg(2));
 
-    Intrinsic::ID ID = CGM.getHLSLRuntime().getDot2AddIntrinsic();
+    //llvm::Intrinsic::dx_##IntrinsicPostfix
+    Intrinsic::ID ID = llvm ::Intrinsic::dx_dot2add;
     return Builder.CreateIntrinsic(
         /*ReturnType=*/C->getType(), ID, ArrayRef<Value *>{A, B, C}, nullptr,
         "hlsl.dot2add");
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h b/clang/lib/CodeGen/CGHLSLRuntime.h
index 5ba69819e5431..23f187b24284f 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.h
+++ b/clang/lib/CodeGen/CGHLSLRuntime.h
@@ -99,7 +99,6 @@ class CGHLSLRuntime {
   GENERATE_HLSL_INTRINSIC_FUNCTION(FDot, fdot)
   GENERATE_HLSL_INTRINSIC_FUNCTION(SDot, sdot)
   GENERATE_HLSL_INTRINSIC_FUNCTION(UDot, udot)
-  GENERATE_HLSL_INTRINSIC_FUNCTION(Dot2Add, dot2add)
   GENERATE_HLSL_INTRINSIC_FUNCTION(Dot4AddI8Packed, dot4add_i8packed)
   GENERATE_HLSL_INTRINSIC_FUNCTION(Dot4AddU8Packed, dot4add_u8packed)
   GENERATE_HLSL_INTRINSIC_FUNCTION(WaveActiveAllTrue, wave_all)
diff --git a/clang/lib/Headers/hlsl/hlsl_alias_intrinsics.h b/clang/lib/Headers/hlsl/hlsl_alias_intrinsics.h
index b7b9369b3ace9..62054b368691d 100644
--- a/clang/lib/Headers/hlsl/hlsl_alias_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_alias_intrinsics.h
@@ -982,16 +982,6 @@ uint64_t dot(uint64_t3, uint64_t3);
 _HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot)
 uint64_t dot(uint64_t4, uint64_t4);
 
-//===----------------------------------------------------------------------===//
-// dot2add builtins
-//===----------------------------------------------------------------------===//
-
-/// \fn float dot2add(half2 a, half2 b, float c)
-
-_HLSL_AVAILABILITY(shadermodel, 6.4)
-_HLSL_BUILTIN_ALIAS(__builtin_hlsl_dot2add)
-float dot2add(half2, half2, float);
-
 //===----------------------------------------------------------------------===//
 // dot4add builtins
 //===----------------------------------------------------------------------===//
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h b/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h
index 5f7c047dbf340..46653d7b295b2 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h
@@ -45,6 +45,14 @@ distance_vec_impl(vector<T, N> X, vector<T, N> Y) {
   return length_vec_impl(X - Y);
 }
 
+constexpr float dot2add_impl(half2 a, half2 b, float c) {
+#if defined(__DIRECTX__)
+  return __builtin_hlsl_dot2add(a, b, c);
+#else
+  return dot(a, b) + c;
+#endif
+}
+
 template <typename T> constexpr T reflect_impl(T I, T N) {
   return I - 2 * N * I * N;
 }
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index 5459cbeb34fd0..b1c1335ce3328 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -117,6 +117,18 @@ const inline float distance(__detail::HLSL_FIXED_VECTOR<float, N> X,
   return __detail::distance_vec_impl(X, Y);
 }
 
+//===----------------------------------------------------------------------===//
+// dot2add builtins
+//===----------------------------------------------------------------------===//
+
+/// \fn float dot2add(half2 a, half2 b, float c)
+/// \brief Dot product of 2 vector of type half and add a float scalar value.
+
+_HLSL_AVAILABILITY(shadermodel, 6.4)
+const inline float dot2add(half2 a, half2 b, float c) {
+  return __detail::dot2add_impl(a, b, c);
+}
+
 //===----------------------------------------------------------------------===//
 // fmod builtins
 //===----------------------------------------------------------------------===//
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index 36de110e75e8a..98dc1b3fe4e77 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -1989,7 +1989,7 @@ void SemaHLSL::diagnoseAvailabilityViolations(TranslationUnitDecl *TU) {
 }
 
 // Helper function for CheckHLSLBuiltinFunctionCall
-static bool CheckVectorElementCallArgs(Sema *S, CallExpr *TheCall) {
+static bool CheckVectorElementCallArgs(Sema *S, CallExpr *TheCall, unsigned NumArgs) {
   assert(TheCall->getNumArgs() > 1);
   ExprResult A = TheCall->getArg(0);
 
@@ -1999,7 +1999,7 @@ static bool CheckVectorElementCallArgs(Sema *S, CallExpr *TheCall) {
   SourceLocation BuiltinLoc = TheCall->getBeginLoc();
 
   bool AllBArgAreVectors = true;
-  for (unsigned i = 1; i < TheCall->getNumArgs(); ++i) {
+  for (unsigned i = 1; i < NumArgs; ++i) {
     ExprResult B = TheCall->getArg(i);
     QualType ArgTyB = B.get()->getType();
     auto *VecTyB = ArgTyB->getAs<VectorType>();
@@ -2049,6 +2049,10 @@ static bool CheckVectorElementCallArgs(Sema *S, CallExpr *TheCall) {
   return false;
 }
 
+static bool CheckVectorElementCallArgs(Sema *S, CallExpr *TheCall) {
+  return CheckVectorElementCallArgs(S, TheCall, TheCall->getNumArgs());
+}
+
 static bool CheckAllArgsHaveSameType(Sema *S, CallExpr *TheCall) {
   assert(TheCall->getNumArgs() > 1);
   QualType ArgTy0 = TheCall->getArg(0)->getType();
@@ -2091,10 +2095,10 @@ static bool CheckArgTypeIsCorrect(
   return false;
 }
 
-static bool CheckAllArgTypesAreCorrect(
-    Sema *S, CallExpr *TheCall, QualType ExpectedType,
+static bool CheckArgTypesAreCorrect(
+    Sema *S, CallExpr *TheCall, unsigned NumArgs, QualType ExpectedType,
     llvm::function_ref<bool(clang::QualType PassedType)> Check) {
-  for (unsigned i = 0; i < TheCall->getNumArgs(); ++i) {
+  for (unsigned i = 0; i < NumArgs; ++i) {
     Expr *Arg = TheCall->getArg(i);
     if (CheckArgTypeIsCorrect(S, Arg, ExpectedType, Check)) {
       return true;
@@ -2103,6 +2107,13 @@ static bool CheckAllArgTypesAreCorrect(
   return false;
 }
 
+static bool CheckAllArgTypesAreCorrect(
+  Sema *S, CallExpr *TheCall, QualType ExpectedType,
+  llvm::function_ref<bool(clang::QualType PassedType)> Check) {
+    return CheckArgTypesAreCorrect(S, TheCall, TheCall->getNumArgs(),           
+                                   ExpectedType, Check);
+}
+
 static bool CheckAllArgsHaveFloatRepresentation(Sema *S, CallExpr *TheCall) {
   auto checkAllFloatTypes = [](clang::QualType PassedType) -> bool {
     return !PassedType->hasFloatingRepresentation();
@@ -2146,15 +2157,17 @@ static bool CheckModifiableLValue(Sema *S, CallExpr *TheCall,
   return true;
 }
 
-static bool CheckNoDoubleVectors(Sema *S, CallExpr *TheCall) {
+static bool CheckNoDoubleVectors(Sema *S, CallExpr *TheCall, 
+                                 unsigned NumArgs, QualType ExpectedType) {
   auto checkDoubleVector = [](clang::QualType PassedType) -> bool {
     if (const auto *VecTy = PassedType->getAs<VectorType>())
       return VecTy->getElementType()->isDoubleType();
     return false;
   };
-  return CheckAllArgTypesAreCorrect(S, TheCall, S->Context.FloatTy,
-                                    checkDoubleVector);
+  return CheckArgTypesAreCorrect(S, TheCall, NumArgs, 
+                                 ExpectedType, checkDoubleVector);
 }
+
 static bool CheckFloatingOrIntRepresentation(Sema *S, CallExpr *TheCall) {
   auto checkAllSignedTypes = [](clang::QualType PassedType) -> bool {
     return !PassedType->hasIntegerRepresentation() &&
@@ -2468,7 +2481,21 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
       return true;
     if (SemaRef.BuiltinVectorToScalarMath(TheCall))
       return true;
-    if (CheckNoDoubleVectors(&SemaRef, TheCall))
+    if (CheckNoDoubleVectors(&SemaRef, TheCall, 
+                             TheCall->getNumArgs(), SemaRef.Context.FloatTy))
+      return true;
+    break;
+  }
+  case Builtin::BI__builtin_hlsl_dot2add: {
+    if (SemaRef.checkArgCount(TheCall, 3))
+      return true;
+    if (CheckVectorElementCallArgs(&SemaRef, TheCall, TheCall->getNumArgs() - 1))
+      return true;
+    if (CheckArgTypeMatches(&SemaRef, TheCall->getArg(2), SemaRef.getASTContext().FloatTy))
+      return true;
+    if (CheckNoDoubleVectors(&SemaRef, TheCall,
+                             TheCall->getNumArgs() - 1,
+                             SemaRef.Context.HalfTy))
       return true;
     break;
   }
diff --git a/clang/test/SemaHLSL/BuiltIns/Dot2Add-errors.hlsl b/clang/test/SemaHLSL/BuiltIns/Dot2Add-errors.hlsl
new file mode 100644
index 0000000000000..61282a319dafd
--- /dev/null
+++ b/clang/test/SemaHLSL/BuiltIns/Dot2Add-errors.hlsl
@@ -0,0 +1,11 @@
+// RUN: %clang_cc1 -finclude-default-header -triple dxil-pc-shadermodel6.6-library %s -emit-llvm-only -disable-llvm-passes -verify
+
+bool test_too_few_arg() {
+  return __builtin_hlsl_dot2add();
+  // expected-error at -1 {{too few arguments to function call, expected 3, have 0}}
+}
+
+bool test_too_many_arg(half2 p1, half2 p2, float p3) {
+  return __builtin_hlsl_dot2add(p1, p2, p3, p1);
+  // expected-error at -1 {{too many arguments to function call, expected 3, have 4}}
+}
diff --git a/llvm/include/llvm/IR/IntrinsicsSPIRV.td b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
index 240c36513f0b6..df3e137c80980 100644
--- a/llvm/include/llvm/IR/IntrinsicsSPIRV.td
+++ b/llvm/include/llvm/IR/IntrinsicsSPIRV.td
@@ -87,11 +87,6 @@ let TargetPrefix = "spv" in {
     DefaultAttrsIntrinsic<[LLVMVectorElementType<0>],
     [llvm_anyint_ty, LLVMScalarOrSameVectorWidth<0, LLVMVectorElementType<0>>],
     [IntrNoMem, Commutative] >;
-
-  def int_spv_dot2add : DefaultAttrsIntrinsic<[llvm_float_ty], 
-                    [llvm_anyfloat_ty, LLVMMatchType<0>, llvm_float_ty], 
-                    [IntrNoMem, Commutative]>;
-
   def int_spv_dot4add_i8packed : DefaultAttrsIntrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
   def int_spv_dot4add_u8packed : DefaultAttrsIntrinsic<[llvm_i32_ty], [llvm_i32_ty, llvm_i32_ty, llvm_i32_ty], [IntrNoMem]>;
   def int_spv_wave_active_countbits : DefaultAttrsIntrinsic<[llvm_i32_ty], [llvm_i1_ty], [IntrConvergent, IntrNoMem]>;
diff --git a/llvm/lib/Target/DirectX/DXILOpLowering.cpp b/llvm/lib/Target/DirectX/DXILOpLowering.cpp
index 10a726aba5ff3..f7ed0c5071d75 100644
--- a/llvm/lib/Target/DirectX/DXILOpLowering.cpp
+++ b/llvm/lib/Target/DirectX/DXILOpLowering.cpp
@@ -55,9 +55,8 @@ static SmallVector<Value *> populateOperands(Value *Arg, IRBuilder<> &Builder) {
 }
 
 static SmallVector<Value *> argVectorFlatten(CallInst *Orig,
-                                             IRBuilder<> &Builder) {
-  // Note: arg[NumOperands-1] is a pointer and is not needed by our flattening.
-  unsigned NumOperands = Orig->getNumOperands() - 1;
+                                             IRBuilder<> &Builder,
+                                             unsigned NumOperands) {
   assert(NumOperands > 0);
   Value *Arg0 = Orig->getOperand(0);
   [[maybe_unused]] auto *VecArg0 = dyn_cast<FixedVectorType>(Arg0->getType());
@@ -75,6 +74,12 @@ static SmallVector<Value *> argVectorFlatten(CallInst *Orig,
   return NewOperands;
 }
 
+static SmallVector<Value *> argVectorFlatten(CallInst *Orig,
+                                             IRBuilder<> &Builder) {
+  // Note: arg[NumOperands-1] is a pointer and is not needed by our flattening.
+    return argVectorFlatten(Orig, Builder, Orig->getNumOperands() - 1);
+}
+/*
 static SmallVector<Value *> argVectorFlattenExcludeLastElement(CallInst *Orig,
                                                                IRBuilder<> &Builder) {
   // Note: arg[NumOperands-1] is a pointer and is not needed by our flattening.
@@ -95,7 +100,7 @@ static SmallVector<Value *> argVectorFlattenExcludeLastElement(CallInst *Orig,
   }
   return NewOperands;
 }
-
+*/
 namespace {
 class OpLowerer {
   Module &M;
@@ -190,7 +195,13 @@ class OpLowerer {
       } else if (IsVectorArgExpansion) {
         Args = argVectorFlatten(CI, OpBuilder.getIRB());
       } else if (F.getIntrinsicID() == Intrinsic::dx_dot2add) {
-        unsigned NumOperands = CI->getNumOperands() - 1;
+        // arg[NumOperands-1] is a pointer and is not needed by our flattening.
+        // arg[NumOperands-2] also does not need to be flattened because it is a scalar.
+        unsigned NumOperands = CI->getNumOperands() - 2;
+        Args.push_back(CI->getArgOperand(NumOperands));
+        Args.append(argVectorFlatten(CI, OpBuilder.getIRB(), NumOperands));
+
+        /*unsigned NumOperands = CI->getNumOperands() - 1;
         assert(NumOperands > 0);
         Value *LastArg = CI->getOperand(NumOperands - 1);
         
@@ -201,6 +212,7 @@ class OpLowerer {
         
         //Args = populateOperands(LastArg, OpBuilder.getIRB());
         Args.append(argVectorFlattenExcludeLastElement(CI, OpBuilder.getIRB()));
+        */
       } else {
         Args.append(CI->arg_begin(), CI->arg_end());
       }

>From 9599ba2646edf73c4efef0f3fe02b07577374276 Mon Sep 17 00:00:00 2001
From: Sumit Agarwal <sumitagarwal at microsoft.com>
Date: Thu, 20 Mar 2025 13:22:33 -0700
Subject: [PATCH 05/12] End to end working code for dot2add

---
 clang/include/clang/Basic/Builtins.td        |  4 ++--
 clang/lib/Sema/SemaHLSL.cpp                  |  2 ++
 clang/test/CodeGenHLSL/builtins/dot2add.hlsl | 17 ++++++++++-------
 3 files changed, 14 insertions(+), 9 deletions(-)

diff --git a/clang/include/clang/Basic/Builtins.td b/clang/include/clang/Basic/Builtins.td
index 8a0237171f7df..0357f02ead1c0 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4893,8 +4893,8 @@ def HLSLDotProduct : LangBuiltin<"HLSL_LANG"> {
 
 def HLSLDot2Add : LangBuiltin<"HLSL_LANG"> {
   let Spellings = ["__builtin_hlsl_dot2add"];
-  let Attributes = [NoThrow, Const];
-  let Prototype = "void(...)";
+  let Attributes = [NoThrow, Const, CustomTypeChecking];
+  let Prototype = "float(...)";
 }
 
 def HLSLDot4AddI8Packed : LangBuiltin<"HLSL_LANG"> {
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index 98dc1b3fe4e77..1283ecb82b5f4 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -2487,8 +2487,10 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
     break;
   }
   case Builtin::BI__builtin_hlsl_dot2add: {
+    // Check number of arguments should be 3
     if (SemaRef.checkArgCount(TheCall, 3))
       return true;
+    // Check first two arguments should be vectors of same length
     if (CheckVectorElementCallArgs(&SemaRef, TheCall, TheCall->getNumArgs() - 1))
       return true;
     if (CheckArgTypeMatches(&SemaRef, TheCall->getArg(2), SemaRef.getASTContext().FloatTy))
diff --git a/clang/test/CodeGenHLSL/builtins/dot2add.hlsl b/clang/test/CodeGenHLSL/builtins/dot2add.hlsl
index ee7cb79bb26c4..ce325327a01b5 100644
--- a/clang/test/CodeGenHLSL/builtins/dot2add.hlsl
+++ b/clang/test/CodeGenHLSL/builtins/dot2add.hlsl
@@ -1,14 +1,17 @@
 // RUN: %clang_cc1 -finclude-default-header -triple \
 // RUN:   dxil-pc-shadermodel6.3-compute %s -emit-llvm -disable-llvm-passes -o - | \
-// RUN:   FileCheck %s -DTARGET=dx
+// RUN:   FileCheck %s --check-prefixes=CHECK,CHECK-DXIL
+// RUN: %clang_cc1 -finclude-default-header -triple \
+// RUN:   spirv-pc-vulkan-compute %s -emit-llvm -disable-llvm-passes -o - | \
+// RUN:   FileCheck %s --check-prefixes=CHECK,CHECK-SPIRV
 
 // Test basic lowering to runtime function call.
 
-// CHECK-LABEL: test
 float test(half2 p1, half2 p2, float p3) {
-  // CHECK-DXIL:  %hlsl.dot2add = call reassoc nnan ninf nsz arcp afn float @llvm.dx.dot2add.v2f32(<2 x float> %0, <2 x float> %1, float %2)
-  // CHECK: ret float %hlsl.dot2add
+  // CHECK-SPIRV:  %[[MUL:.*]] = call {{.*}} float @llvm.spv.fdot.v2f32(<2 x float> %1, <2 x float> %2)
+  // CHECK-SPIRV:  %[[C:.*]] = load float, ptr %c.addr, align 4
+  // CHECK-SPIRV:  %[[RES:.*]] = fadd {{.*}} float %[[MUL]], %[[C]]
+  // CHECK-DXIL:  %[[RES:.*]] = call {{.*}} float @llvm.dx.dot2add.v2f32(<2 x float> %0, <2 x float> %1, float %2)
+  // CHECK:  ret float %[[RES]]
   return dot2add(p1, p2, p3);
-}
-
-// CHECK: declare [[TY]] @llvm.[[TARGET]].dot4add.i8packed([[TY]], [[TY]], [[TY]])
+}
\ No newline at end of file

>From 2c9cee530b1abac5fdd9309fa05d60940e9b7cc7 Mon Sep 17 00:00:00 2001
From: Sumit Agarwal <sumitagarwal at microsoft.com>
Date: Thu, 20 Mar 2025 15:06:37 -0700
Subject: [PATCH 06/12] Update SemaCheck and set return type and make void as
 return type in Builtin

---
 clang/include/clang/Basic/Builtins.td |  2 +-
 clang/lib/Sema/SemaHLSL.cpp           | 24 ++++++++++++++++++------
 2 files changed, 19 insertions(+), 7 deletions(-)

diff --git a/clang/include/clang/Basic/Builtins.td b/clang/include/clang/Basic/Builtins.td
index 0357f02ead1c0..76ab463ca0ed6 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4894,7 +4894,7 @@ def HLSLDotProduct : LangBuiltin<"HLSL_LANG"> {
 def HLSLDot2Add : LangBuiltin<"HLSL_LANG"> {
   let Spellings = ["__builtin_hlsl_dot2add"];
   let Attributes = [NoThrow, Const, CustomTypeChecking];
-  let Prototype = "float(...)";
+  let Prototype = "void(...)";
 }
 
 def HLSLDot4AddI8Packed : LangBuiltin<"HLSL_LANG"> {
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index 1283ecb82b5f4..399371c4ae2f6 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -2490,15 +2490,27 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
     // Check number of arguments should be 3
     if (SemaRef.checkArgCount(TheCall, 3))
       return true;
-    // Check first two arguments should be vectors of same length
-    if (CheckVectorElementCallArgs(&SemaRef, TheCall, TheCall->getNumArgs() - 1))
+
+    // Check first two arguments are vector of length 2 with half data type
+    auto checkHalfVectorOfSize2 = [](clang::QualType PassedType) -> bool {
+      if (const auto *VecTy = PassedType->getAs<VectorType>())
+        return !(VecTy->getNumElements() == 2 &&
+            VecTy->getElementType()->isHalfType());
       return true;
-    if (CheckArgTypeMatches(&SemaRef, TheCall->getArg(2), SemaRef.getASTContext().FloatTy))
+    };
+    if(CheckArgTypeIsCorrect(&SemaRef, TheCall->getArg(0),
+                             SemaRef.getASTContext().HalfTy,
+                             checkHalfVectorOfSize2))
       return true;
-    if (CheckNoDoubleVectors(&SemaRef, TheCall,
-                             TheCall->getNumArgs() - 1,
-                             SemaRef.Context.HalfTy))
+    if(CheckArgTypeIsCorrect(&SemaRef, TheCall->getArg(1),
+                             SemaRef.getASTContext().HalfTy,
+                             checkHalfVectorOfSize2))
+      return true;
+
+    // Check third argument is a float
+    if (CheckArgTypeMatches(&SemaRef, TheCall->getArg(2), SemaRef.getASTContext().FloatTy))
       return true;
+    TheCall->setType(TheCall->getArg(2)->getType());
     break;
   }
   case Builtin::BI__builtin_hlsl_elementwise_firstbithigh:

>From 2e4fe63bfb8a6f88cc5bf621df550b6501a89115 Mon Sep 17 00:00:00 2001
From: Sumit Agarwal <sumitagarwal at microsoft.com>
Date: Thu, 20 Mar 2025 15:25:01 -0700
Subject: [PATCH 07/12] Revert not required change

---
 clang/lib/Sema/SemaHLSL.cpp | 32 +++++++++-----------------------
 1 file changed, 9 insertions(+), 23 deletions(-)

diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index 399371c4ae2f6..9cd7a0dd7aad9 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -1989,7 +1989,7 @@ void SemaHLSL::diagnoseAvailabilityViolations(TranslationUnitDecl *TU) {
 }
 
 // Helper function for CheckHLSLBuiltinFunctionCall
-static bool CheckVectorElementCallArgs(Sema *S, CallExpr *TheCall, unsigned NumArgs) {
+static bool CheckVectorElementCallArgs(Sema *S, CallExpr *TheCall) {
   assert(TheCall->getNumArgs() > 1);
   ExprResult A = TheCall->getArg(0);
 
@@ -1999,7 +1999,7 @@ static bool CheckVectorElementCallArgs(Sema *S, CallExpr *TheCall, unsigned NumA
   SourceLocation BuiltinLoc = TheCall->getBeginLoc();
 
   bool AllBArgAreVectors = true;
-  for (unsigned i = 1; i < NumArgs; ++i) {
+  for (unsigned i = 1; i < TheCall->getNumArgs(); ++i) {
     ExprResult B = TheCall->getArg(i);
     QualType ArgTyB = B.get()->getType();
     auto *VecTyB = ArgTyB->getAs<VectorType>();
@@ -2049,10 +2049,6 @@ static bool CheckVectorElementCallArgs(Sema *S, CallExpr *TheCall, unsigned NumA
   return false;
 }
 
-static bool CheckVectorElementCallArgs(Sema *S, CallExpr *TheCall) {
-  return CheckVectorElementCallArgs(S, TheCall, TheCall->getNumArgs());
-}
-
 static bool CheckAllArgsHaveSameType(Sema *S, CallExpr *TheCall) {
   assert(TheCall->getNumArgs() > 1);
   QualType ArgTy0 = TheCall->getArg(0)->getType();
@@ -2095,10 +2091,10 @@ static bool CheckArgTypeIsCorrect(
   return false;
 }
 
-static bool CheckArgTypesAreCorrect(
-    Sema *S, CallExpr *TheCall, unsigned NumArgs, QualType ExpectedType,
+static bool CheckAllArgTypesAreCorrect(
+    Sema *S, CallExpr *TheCall, QualType ExpectedType,
     llvm::function_ref<bool(clang::QualType PassedType)> Check) {
-  for (unsigned i = 0; i < NumArgs; ++i) {
+  for (unsigned i = 0; i < TheCall->getNumArgs(); ++i) {
     Expr *Arg = TheCall->getArg(i);
     if (CheckArgTypeIsCorrect(S, Arg, ExpectedType, Check)) {
       return true;
@@ -2107,13 +2103,6 @@ static bool CheckArgTypesAreCorrect(
   return false;
 }
 
-static bool CheckAllArgTypesAreCorrect(
-  Sema *S, CallExpr *TheCall, QualType ExpectedType,
-  llvm::function_ref<bool(clang::QualType PassedType)> Check) {
-    return CheckArgTypesAreCorrect(S, TheCall, TheCall->getNumArgs(),           
-                                   ExpectedType, Check);
-}
-
 static bool CheckAllArgsHaveFloatRepresentation(Sema *S, CallExpr *TheCall) {
   auto checkAllFloatTypes = [](clang::QualType PassedType) -> bool {
     return !PassedType->hasFloatingRepresentation();
@@ -2157,17 +2146,15 @@ static bool CheckModifiableLValue(Sema *S, CallExpr *TheCall,
   return true;
 }
 
-static bool CheckNoDoubleVectors(Sema *S, CallExpr *TheCall, 
-                                 unsigned NumArgs, QualType ExpectedType) {
+static bool CheckNoDoubleVectors(Sema *S, CallExpr *TheCall) {
   auto checkDoubleVector = [](clang::QualType PassedType) -> bool {
     if (const auto *VecTy = PassedType->getAs<VectorType>())
       return VecTy->getElementType()->isDoubleType();
     return false;
   };
-  return CheckArgTypesAreCorrect(S, TheCall, NumArgs, 
-                                 ExpectedType, checkDoubleVector);
+  return CheckAllArgTypesAreCorrect(S, TheCall, S->Context.FloatTy,
+                                    checkDoubleVector);
 }
-
 static bool CheckFloatingOrIntRepresentation(Sema *S, CallExpr *TheCall) {
   auto checkAllSignedTypes = [](clang::QualType PassedType) -> bool {
     return !PassedType->hasIntegerRepresentation() &&
@@ -2481,8 +2468,7 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
       return true;
     if (SemaRef.BuiltinVectorToScalarMath(TheCall))
       return true;
-    if (CheckNoDoubleVectors(&SemaRef, TheCall, 
-                             TheCall->getNumArgs(), SemaRef.Context.FloatTy))
+    if (CheckNoDoubleVectors(&SemaRef, TheCall))
       return true;
     break;
   }

>From b5ee3cb1fc9fe7093b08c4ee49d112fcf675a60a Mon Sep 17 00:00:00 2001
From: Sumit Agarwal <sumitagarwal at microsoft.com>
Date: Thu, 20 Mar 2025 15:31:15 -0700
Subject: [PATCH 08/12] Remove commented code

---
 llvm/lib/Target/DirectX/DXILOpLowering.cpp | 36 +---------------------
 1 file changed, 1 insertion(+), 35 deletions(-)

diff --git a/llvm/lib/Target/DirectX/DXILOpLowering.cpp b/llvm/lib/Target/DirectX/DXILOpLowering.cpp
index f7ed0c5071d75..cecd012011419 100644
--- a/llvm/lib/Target/DirectX/DXILOpLowering.cpp
+++ b/llvm/lib/Target/DirectX/DXILOpLowering.cpp
@@ -79,28 +79,7 @@ static SmallVector<Value *> argVectorFlatten(CallInst *Orig,
   // Note: arg[NumOperands-1] is a pointer and is not needed by our flattening.
     return argVectorFlatten(Orig, Builder, Orig->getNumOperands() - 1);
 }
-/*
-static SmallVector<Value *> argVectorFlattenExcludeLastElement(CallInst *Orig,
-                                                               IRBuilder<> &Builder) {
-  // Note: arg[NumOperands-1] is a pointer and is not needed by our flattening.
-  unsigned NumOperands = Orig->getNumOperands() - 2;
-  assert(NumOperands > 0);
-  Value *Arg0 = Orig->getOperand(0);
-  [[maybe_unused]] auto *VecArg0 = dyn_cast<FixedVectorType>(Arg0->getType());
-  assert(VecArg0);
-  SmallVector<Value *> NewOperands = populateOperands(Arg0, Builder);
-  for (unsigned I = 1; I < NumOperands; ++I) {
-    Value *Arg = Orig->getOperand(I);
-    [[maybe_unused]] auto *VecArg = dyn_cast<FixedVectorType>(Arg->getType());
-    assert(VecArg);
-    assert(VecArg0->getElementType() == VecArg->getElementType());
-    assert(VecArg0->getNumElements() == VecArg->getNumElements());
-    auto NextOperandList = populateOperands(Arg, Builder);
-    NewOperands.append(NextOperandList.begin(), NextOperandList.end());
-  }
-  return NewOperands;
-}
-*/
+
 namespace {
 class OpLowerer {
   Module &M;
@@ -200,19 +179,6 @@ class OpLowerer {
         unsigned NumOperands = CI->getNumOperands() - 2;
         Args.push_back(CI->getArgOperand(NumOperands));
         Args.append(argVectorFlatten(CI, OpBuilder.getIRB(), NumOperands));
-
-        /*unsigned NumOperands = CI->getNumOperands() - 1;
-        assert(NumOperands > 0);
-        Value *LastArg = CI->getOperand(NumOperands - 1);
-        
-        Args.push_back(LastArg);
-        
-        //dbgs() << "Value of LastArg" << LastArg->getName() << "\n";
-        
-        
-        //Args = populateOperands(LastArg, OpBuilder.getIRB());
-        Args.append(argVectorFlattenExcludeLastElement(CI, OpBuilder.getIRB()));
-        */
       } else {
         Args.append(CI->arg_begin(), CI->arg_end());
       }

>From 43f17ed4a0f7e475f1e0e9e2480ede4b09fb2490 Mon Sep 17 00:00:00 2001
From: Sumit Agarwal <sumitagarwal at microsoft.com>
Date: Fri, 21 Mar 2025 17:32:42 -0700
Subject: [PATCH 09/12] Addressed PR feedback

---
 clang/include/clang/Basic/Builtins.td         |  2 +-
 clang/lib/CodeGen/CGBuiltin.cpp               |  7 ++---
 clang/lib/Headers/hlsl/hlsl_intrinsics.h      |  9 ++++--
 clang/lib/Sema/SemaHLSL.cpp                   |  5 +--
 clang/test/CodeGenHLSL/builtins/dot2add.hlsl  | 13 ++++----
 .../SemaHLSL/BuiltIns/Dot2Add-errors.hlsl     | 31 +++++++++++++++++--
 llvm/test/CodeGen/DirectX/dot2add.ll          |  2 +-
 7 files changed, 46 insertions(+), 23 deletions(-)

diff --git a/clang/include/clang/Basic/Builtins.td b/clang/include/clang/Basic/Builtins.td
index 76ab463ca0ed6..96ca64a9a8022 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4894,7 +4894,7 @@ def HLSLDotProduct : LangBuiltin<"HLSL_LANG"> {
 def HLSLDot2Add : LangBuiltin<"HLSL_LANG"> {
   let Spellings = ["__builtin_hlsl_dot2add"];
   let Attributes = [NoThrow, Const, CustomTypeChecking];
-  let Prototype = "void(...)";
+  let Prototype = "float(_ExtVector<2, _Float16>,_ExtVector<2, _Float16>, float)";
 }
 
 def HLSLDot4AddI8Packed : LangBuiltin<"HLSL_LANG"> {
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index b3d9db5be7d8d..6e1ec33b197bd 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -19683,18 +19683,15 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
   }
   case Builtin::BI__builtin_hlsl_dot2add: {
     llvm::Triple::ArchType Arch = CGM.getTarget().getTriple().getArch();
-    if (Arch != llvm::Triple::dxil) {
-      llvm_unreachable("Intrinsic dot2add can be executed as a builtin only on dxil");
-    }
+    assert(Arch == llvm::Triple::dxil && "Intrinsic dot2add can be executed as a builtin only on dxil");
     Value *A = EmitScalarExpr(E->getArg(0));
     Value *B = EmitScalarExpr(E->getArg(1));
     Value *C = EmitScalarExpr(E->getArg(2));
 
-    //llvm::Intrinsic::dx_##IntrinsicPostfix
     Intrinsic::ID ID = llvm ::Intrinsic::dx_dot2add;
     return Builder.CreateIntrinsic(
         /*ReturnType=*/C->getType(), ID, ArrayRef<Value *>{A, B, C}, nullptr,
-        "hlsl.dot2add");
+        "dx.dot2add");
   }
   case Builtin::BI__builtin_hlsl_dot4add_i8packed: {
     Value *A = EmitScalarExpr(E->getArg(0));
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsics.h b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
index b1c1335ce3328..af9990ed1398c 100644
--- a/clang/lib/Headers/hlsl/hlsl_intrinsics.h
+++ b/clang/lib/Headers/hlsl/hlsl_intrinsics.h
@@ -121,12 +121,15 @@ const inline float distance(__detail::HLSL_FIXED_VECTOR<float, N> X,
 // dot2add builtins
 //===----------------------------------------------------------------------===//
 
-/// \fn float dot2add(half2 a, half2 b, float c)
+/// \fn float dot2add(half2 A, half2 B, float C)
 /// \brief Dot product of 2 vector of type half and add a float scalar value.
+/// \param A The first input value to dot product. 
+/// \param B The second input value to dot product.
+/// \param C The input value added to the dot product.
 
 _HLSL_AVAILABILITY(shadermodel, 6.4)
-const inline float dot2add(half2 a, half2 b, float c) {
-  return __detail::dot2add_impl(a, b, c);
+const inline float dot2add(half2 A, half2 B, float C) {
+  return __detail::dot2add_impl(A, B, C);
 }
 
 //===----------------------------------------------------------------------===//
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index 9cd7a0dd7aad9..c2ef77a563829 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -2473,11 +2473,9 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
     break;
   }
   case Builtin::BI__builtin_hlsl_dot2add: {
-    // Check number of arguments should be 3
     if (SemaRef.checkArgCount(TheCall, 3))
       return true;
 
-    // Check first two arguments are vector of length 2 with half data type
     auto checkHalfVectorOfSize2 = [](clang::QualType PassedType) -> bool {
       if (const auto *VecTy = PassedType->getAs<VectorType>())
         return !(VecTy->getNumElements() == 2 &&
@@ -2493,10 +2491,9 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
                              checkHalfVectorOfSize2))
       return true;
 
-    // Check third argument is a float
     if (CheckArgTypeMatches(&SemaRef, TheCall->getArg(2), SemaRef.getASTContext().FloatTy))
       return true;
-    TheCall->setType(TheCall->getArg(2)->getType());
+    TheCall->setType(SemaRef.getASTContext().FloatTy);
     break;
   }
   case Builtin::BI__builtin_hlsl_elementwise_firstbithigh:
diff --git a/clang/test/CodeGenHLSL/builtins/dot2add.hlsl b/clang/test/CodeGenHLSL/builtins/dot2add.hlsl
index ce325327a01b5..cea399f6028d2 100644
--- a/clang/test/CodeGenHLSL/builtins/dot2add.hlsl
+++ b/clang/test/CodeGenHLSL/builtins/dot2add.hlsl
@@ -1,17 +1,18 @@
-// RUN: %clang_cc1 -finclude-default-header -triple \
+// RUN: %clang_cc1 -finclude-default-header -fnative-half-type -triple \
 // RUN:   dxil-pc-shadermodel6.3-compute %s -emit-llvm -disable-llvm-passes -o - | \
 // RUN:   FileCheck %s --check-prefixes=CHECK,CHECK-DXIL
-// RUN: %clang_cc1 -finclude-default-header -triple \
+// RUN: %clang_cc1 -finclude-default-header -fnative-half-type -triple \
 // RUN:   spirv-pc-vulkan-compute %s -emit-llvm -disable-llvm-passes -o - | \
 // RUN:   FileCheck %s --check-prefixes=CHECK,CHECK-SPIRV
 
 // Test basic lowering to runtime function call.
 
 float test(half2 p1, half2 p2, float p3) {
-  // CHECK-SPIRV:  %[[MUL:.*]] = call {{.*}} float @llvm.spv.fdot.v2f32(<2 x float> %1, <2 x float> %2)
+  // CHECK-SPIRV:  %[[MUL:.*]] = call {{.*}} half @llvm.spv.fdot.v2f16(<2 x half> %1, <2 x half> %2)
+  // CHECK-SPIRV:  %[[CONVERT:.*]] = fpext {{.*}} half %[[MUL:.*]] to float
   // CHECK-SPIRV:  %[[C:.*]] = load float, ptr %c.addr, align 4
-  // CHECK-SPIRV:  %[[RES:.*]] = fadd {{.*}} float %[[MUL]], %[[C]]
-  // CHECK-DXIL:  %[[RES:.*]] = call {{.*}} float @llvm.dx.dot2add.v2f32(<2 x float> %0, <2 x float> %1, float %2)
+  // CHECK-SPIRV:  %[[RES:.*]] = fadd {{.*}} float %[[CONVERT:.*]], %[[C:.*]]
+  // CHECK-DXIL:  %[[RES:.*]] = call {{.*}} float @llvm.dx.dot2add.v2f16(<2 x half> %0, <2 x half> %1, float %2)
   // CHECK:  ret float %[[RES]]
   return dot2add(p1, p2, p3);
-}
\ No newline at end of file
+}
diff --git a/clang/test/SemaHLSL/BuiltIns/Dot2Add-errors.hlsl b/clang/test/SemaHLSL/BuiltIns/Dot2Add-errors.hlsl
index 61282a319dafd..87bcb931aabe5 100644
--- a/clang/test/SemaHLSL/BuiltIns/Dot2Add-errors.hlsl
+++ b/clang/test/SemaHLSL/BuiltIns/Dot2Add-errors.hlsl
@@ -1,11 +1,36 @@
-// RUN: %clang_cc1 -finclude-default-header -triple dxil-pc-shadermodel6.6-library %s -emit-llvm-only -disable-llvm-passes -verify
+// RUN: %clang_cc1 -finclude-default-header -triple dxil-pc-shadermodel6.6-library %s -fnative-half-type -emit-llvm-only -disable-llvm-passes -verify
 
-bool test_too_few_arg() {
+float test_too_few_arg() {
   return __builtin_hlsl_dot2add();
   // expected-error at -1 {{too few arguments to function call, expected 3, have 0}}
 }
 
-bool test_too_many_arg(half2 p1, half2 p2, float p3) {
+float test_too_many_arg(half2 p1, half2 p2, float p3) {
   return __builtin_hlsl_dot2add(p1, p2, p3, p1);
   // expected-error at -1 {{too many arguments to function call, expected 3, have 4}}
 }
+
+float test_float_arg2_type(half2 p1, float2 p2, float p3) {
+  return __builtin_hlsl_dot2add(p1, p2, p3);
+  // expected-error at -1 {{passing 'float2' (aka 'vector<float, 2>') to parameter of incompatible type '__attribute__((__vector_size__(2 * sizeof(half)))) half' (vector of 2 'half' values)}}
+}
+
+float test_float_arg1_type(float2 p1, half2 p2, float p3) {
+  return __builtin_hlsl_dot2add(p1, p2, p3);
+  // expected-error at -1 {{passing 'float2' (aka 'vector<float, 2>') to parameter of incompatible type '__attribute__((__vector_size__(2 * sizeof(half)))) half' (vector of 2 'half' values)}}
+}
+
+float test_double_arg3_type(half2 p1, half2 p2, double p3) {
+  return __builtin_hlsl_dot2add(p1, p2, p3);
+  // expected-error at -1 {{passing 'double' to parameter of incompatible type 'float'}}
+}
+
+float test_float_arg1_arg2_type(float2 p1, float2 p2, float p3) {
+  return __builtin_hlsl_dot2add(p1, p2, p3);
+  // expected-error at -1 {{passing 'float2' (aka 'vector<float, 2>') to parameter of incompatible type '__attribute__((__vector_size__(2 * sizeof(half)))) half' (vector of 2 'half' values)}}
+}
+
+float test_int16_arg1_arg2_type(int16_t2 p1, int16_t2 p2, float p3) {
+  return __builtin_hlsl_dot2add(p1, p2, p3);
+  // expected-error at -1 {{passing 'int16_t2' (aka 'vector<int16_t, 2>') to parameter of incompatible type '__attribute__((__vector_size__(2 * sizeof(half)))) half' (vector of 2 'half' values)}}
+}
diff --git a/llvm/test/CodeGen/DirectX/dot2add.ll b/llvm/test/CodeGen/DirectX/dot2add.ll
index b1019c36b56e8..40c6cdafc83da 100644
--- a/llvm/test/CodeGen/DirectX/dot2add.ll
+++ b/llvm/test/CodeGen/DirectX/dot2add.ll
@@ -5,4 +5,4 @@ entry:
 ; CHECK: call float @dx.op.dot2AddHalf(i32 162, float %c, half %0, half %1, half %2, half %3)
   %ret = call float @llvm.dx.dot2add(<2 x half> %a, <2 x half> %b, float %c)
   ret float %ret
-}
\ No newline at end of file
+}

>From 3fe546f34ef9962ed00bf4059573ae11a51b887f Mon Sep 17 00:00:00 2001
From: Sumit Agarwal <sumitagarwal at microsoft.com>
Date: Mon, 24 Mar 2025 12:30:39 -0700
Subject: [PATCH 10/12] update sema

---
 clang/lib/Sema/SemaHLSL.cpp | 22 +++++++++++++++++++++-
 1 file changed, 21 insertions(+), 1 deletion(-)

diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index c2ef77a563829..ccc8a6ac15248 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -2482,6 +2482,22 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
             VecTy->getElementType()->isHalfType());
       return true;
     };
+
+    auto CheckArgTypeIsCorrect = [](
+      Sema *S, Expr *Arg, QualType ExpectedType,
+      llvm::function_ref<bool(clang::QualType PassedType)> Check) -> bool {
+      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;
+    };
+
     if(CheckArgTypeIsCorrect(&SemaRef, TheCall->getArg(0),
                              SemaRef.getASTContext().HalfTy,
                              checkHalfVectorOfSize2))
@@ -2491,8 +2507,12 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
                              checkHalfVectorOfSize2))
       return true;
 
-    if (CheckArgTypeMatches(&SemaRef, TheCall->getArg(2), SemaRef.getASTContext().FloatTy))
+    if (!SemaRef.getASTContext().hasSameUnqualifiedType(TheCall->getArg(2)->getType(), SemaRef.getASTContext().FloatTy)) {
+      SemaRef.Diag(TheCall->getArg(2)->getBeginLoc(), diag::err_typecheck_convert_incompatible)
+          << TheCall->getArg(2)->getType() << SemaRef.getASTContext().FloatTy << 1 << 0 << 0;
       return true;
+    }
+
     TheCall->setType(SemaRef.getASTContext().FloatTy);
     break;
   }

>From 8b6de1a8725a4f0ea90446311118555416c40cfd Mon Sep 17 00:00:00 2001
From: Sumit Agarwal <sumitagarwal at microsoft.com>
Date: Wed, 26 Mar 2025 17:38:54 -0700
Subject: [PATCH 11/12] Addressed PR feedback

---
 clang/include/clang/Basic/Builtins.td         |  4 +-
 clang/lib/Sema/SemaHLSL.cpp                   | 44 ------------
 clang/test/CodeGenHLSL/builtins/dot2add.hlsl  | 70 +++++++++++++++++--
 .../SemaHLSL/BuiltIns/Dot2Add-errors.hlsl     | 36 ----------
 .../SemaHLSL/BuiltIns/dot2add-errors.hlsl     | 13 ++++
 5 files changed, 80 insertions(+), 87 deletions(-)
 delete mode 100644 clang/test/SemaHLSL/BuiltIns/Dot2Add-errors.hlsl
 create mode 100644 clang/test/SemaHLSL/BuiltIns/dot2add-errors.hlsl

diff --git a/clang/include/clang/Basic/Builtins.td b/clang/include/clang/Basic/Builtins.td
index 96ca64a9a8022..9f44ac6886e4f 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4893,8 +4893,8 @@ def HLSLDotProduct : LangBuiltin<"HLSL_LANG"> {
 
 def HLSLDot2Add : LangBuiltin<"HLSL_LANG"> {
   let Spellings = ["__builtin_hlsl_dot2add"];
-  let Attributes = [NoThrow, Const, CustomTypeChecking];
-  let Prototype = "float(_ExtVector<2, _Float16>,_ExtVector<2, _Float16>, float)";
+  let Attributes = [NoThrow, Const];
+  let Prototype = "float(_ExtVector<2, __fp16>,_ExtVector<2, __fp16>, float)";
 }
 
 def HLSLDot4AddI8Packed : LangBuiltin<"HLSL_LANG"> {
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index ccc8a6ac15248..36de110e75e8a 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -2472,50 +2472,6 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
       return true;
     break;
   }
-  case Builtin::BI__builtin_hlsl_dot2add: {
-    if (SemaRef.checkArgCount(TheCall, 3))
-      return true;
-
-    auto checkHalfVectorOfSize2 = [](clang::QualType PassedType) -> bool {
-      if (const auto *VecTy = PassedType->getAs<VectorType>())
-        return !(VecTy->getNumElements() == 2 &&
-            VecTy->getElementType()->isHalfType());
-      return true;
-    };
-
-    auto CheckArgTypeIsCorrect = [](
-      Sema *S, Expr *Arg, QualType ExpectedType,
-      llvm::function_ref<bool(clang::QualType PassedType)> Check) -> bool {
-      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;
-    };
-
-    if(CheckArgTypeIsCorrect(&SemaRef, TheCall->getArg(0),
-                             SemaRef.getASTContext().HalfTy,
-                             checkHalfVectorOfSize2))
-      return true;
-    if(CheckArgTypeIsCorrect(&SemaRef, TheCall->getArg(1),
-                             SemaRef.getASTContext().HalfTy,
-                             checkHalfVectorOfSize2))
-      return true;
-
-    if (!SemaRef.getASTContext().hasSameUnqualifiedType(TheCall->getArg(2)->getType(), SemaRef.getASTContext().FloatTy)) {
-      SemaRef.Diag(TheCall->getArg(2)->getBeginLoc(), diag::err_typecheck_convert_incompatible)
-          << TheCall->getArg(2)->getType() << SemaRef.getASTContext().FloatTy << 1 << 0 << 0;
-      return true;
-    }
-
-    TheCall->setType(SemaRef.getASTContext().FloatTy);
-    break;
-  }
   case Builtin::BI__builtin_hlsl_elementwise_firstbithigh:
   case Builtin::BI__builtin_hlsl_elementwise_firstbitlow: {
     if (SemaRef.PrepareBuiltinElementwiseMathOneArgCall(TheCall))
diff --git a/clang/test/CodeGenHLSL/builtins/dot2add.hlsl b/clang/test/CodeGenHLSL/builtins/dot2add.hlsl
index cea399f6028d2..8421c6c9b31be 100644
--- a/clang/test/CodeGenHLSL/builtins/dot2add.hlsl
+++ b/clang/test/CodeGenHLSL/builtins/dot2add.hlsl
@@ -7,12 +7,72 @@
 
 // Test basic lowering to runtime function call.
 
+// CHECK-LABEL: define {{.*}}test
 float test(half2 p1, half2 p2, float p3) {
-  // CHECK-SPIRV:  %[[MUL:.*]] = call {{.*}} half @llvm.spv.fdot.v2f16(<2 x half> %1, <2 x half> %2)
-  // CHECK-SPIRV:  %[[CONVERT:.*]] = fpext {{.*}} half %[[MUL:.*]] to float
-  // CHECK-SPIRV:  %[[C:.*]] = load float, ptr %c.addr, align 4
-  // CHECK-SPIRV:  %[[RES:.*]] = fadd {{.*}} float %[[CONVERT:.*]], %[[C:.*]]
-  // CHECK-DXIL:  %[[RES:.*]] = call {{.*}} float @llvm.dx.dot2add.v2f16(<2 x half> %0, <2 x half> %1, float %2)
+  // CHECK-SPIRV:  %[[RES:.*]] = call {{.*}} float @_ZN4hlsl7dot2addEDv2_DhS0_f(<2 x half> {{.*}} %1, <2 x half> {{.*}} %2, float {{.*}} %3) #3 {{.*}}
+  // CHECK-DXIL:  %[[RES:.*]] = call {{.*}} float @_ZN4hlsl7dot2addEDv2_DhS0_f(<2 x half> {{.*}} %0, <2 x half> {{.*}} %1, float {{.*}} %2) #2
   // CHECK:  ret float %[[RES]]
   return dot2add(p1, p2, p3);
 }
+
+// CHECK-LABEL: define {{.*}}test_float_arg2_type
+float test_float_arg2_type(half2 p1, float2 p2, float p3) {
+  // CHECK-SPIRV:  %conv = fptrunc reassoc nnan ninf nsz arcp afn <2 x float> %2 to <2 x half>
+  // CHECK-SPIRV:  %[[RES:.*]] = call {{.*}} float @_ZN4hlsl7dot2addEDv2_DhS0_f(<2 x half> {{.*}} %1, <2 x half> {{.*}} %conv, float {{.*}} %3) #3 {{.*}}
+  // CHECK-DXIL:  %conv = fptrunc reassoc nnan ninf nsz arcp afn <2 x float> %1 to <2 x half>
+  // CHECK-DXIL:  %[[RES:.*]] = call {{.*}} float @_ZN4hlsl7dot2addEDv2_DhS0_f(<2 x half> {{.*}} %0, <2 x half> {{.*}} %conv, float {{.*}} %2) #2
+  // CHECK:  ret float %[[RES]]
+  return dot2add(p1, p2, p3);
+}
+
+// CHECK-LABEL: define {{.*}}test_float_arg1_type
+float test_float_arg1_type(float2 p1, half2 p2, float p3) {
+  // CHECK-SPIRV:  %conv = fptrunc reassoc nnan ninf nsz arcp afn <2 x float> %1 to <2 x half>
+  // CHECK-SPIRV:  %[[RES:.*]] = call {{.*}} float @_ZN4hlsl7dot2addEDv2_DhS0_f(<2 x half> {{.*}} %conv, <2 x half> {{.*}} %2, float {{.*}} %3) #3 {{.*}}
+  // CHECK-DXIL:  %conv = fptrunc reassoc nnan ninf nsz arcp afn <2 x float> %0 to <2 x half>
+  // CHECK-DXIL:  %[[RES:.*]] = call {{.*}} float @_ZN4hlsl7dot2addEDv2_DhS0_f(<2 x half> {{.*}} %conv, <2 x half> {{.*}} %1, float {{.*}} %2) #2
+  // CHECK:  ret float %[[RES]]
+  return dot2add(p1, p2, p3);
+}
+
+// CHECK-LABEL: define {{.*}}test_double_arg3_type
+float test_double_arg3_type(half2 p1, half2 p2, double p3) {
+  // CHECK-SPIRV:  %conv = fptrunc reassoc nnan ninf nsz arcp afn double %3 to float
+  // CHECK-SPIRV:  %[[RES:.*]] = call {{.*}} float @_ZN4hlsl7dot2addEDv2_DhS0_f(<2 x half> {{.*}} %1, <2 x half> {{.*}} %2, float {{.*}} %conv) #3 {{.*}}
+  // CHECK-DXIL:  %conv = fptrunc reassoc nnan ninf nsz arcp afn double %2 to float
+  // CHECK-DXIL:  %[[RES:.*]] = call {{.*}} float @_ZN4hlsl7dot2addEDv2_DhS0_f(<2 x half> {{.*}} %0, <2 x half> {{.*}} %1, float {{.*}} %conv) #2
+  // CHECK:  ret float %[[RES]]
+  return dot2add(p1, p2, p3);
+}
+
+// CHECK-LABEL: define {{.*}}test_float_arg1_arg2_type
+float test_float_arg1_arg2_type(float2 p1, float2 p2, float p3) {
+  // CHECK-SPIRV:  %conv = fptrunc reassoc nnan ninf nsz arcp afn <2 x float> %1 to <2 x half>
+  // CHECK-SPIRV:  %conv1 = fptrunc reassoc nnan ninf nsz arcp afn <2 x float> %2 to <2 x half>
+  // CHECK-SPIRV:  %[[RES:.*]] = call {{.*}} float @_ZN4hlsl7dot2addEDv2_DhS0_f(<2 x half> {{.*}} %conv, <2 x half> {{.*}} %conv1, float {{.*}} %3) #3 {{.*}}
+  // CHECK-DXIL:  %conv = fptrunc reassoc nnan ninf nsz arcp afn <2 x float> %0 to <2 x half>
+  // CHECK-DXIL:  %conv1 = fptrunc reassoc nnan ninf nsz arcp afn <2 x float> %1 to <2 x half>
+  // CHECK-DXIL:  %[[RES:.*]] = call {{.*}} float @_ZN4hlsl7dot2addEDv2_DhS0_f(<2 x half> {{.*}} %conv, <2 x half> {{.*}} %conv1, float {{.*}} %2) #2
+  // CHECK:  ret float %[[RES]]
+  return dot2add(p1, p2, p3);
+}
+
+// CHECK-LABEL: define {{.*}}test_int16_arg1_arg2_type
+float test_int16_arg1_arg2_type(int16_t2 p1, int16_t2 p2, float p3) {
+  // CHECK-SPIRV:  %conv = sitofp <2 x i16> %1 to <2 x half>
+  // CHECK-SPIRV:  %conv1 = sitofp <2 x i16> %2 to <2 x half>
+  // CHECK-SPIRV:  %[[RES:.*]] = call {{.*}} float @_ZN4hlsl7dot2addEDv2_DhS0_f(<2 x half> {{.*}} %conv, <2 x half> {{.*}} %conv1, float {{.*}} %3) #3 {{.*}}
+  // CHECK-DXIL:  %conv = sitofp <2 x i16> %0 to <2 x half>
+  // CHECK-DXIL:  %conv1 = sitofp <2 x i16> %1 to <2 x half>
+  // CHECK-DXIL:  %[[RES:.*]] = call {{.*}} float @_ZN4hlsl7dot2addEDv2_DhS0_f(<2 x half> {{.*}} %conv, <2 x half> {{.*}} %conv1, float {{.*}} %2) #2
+  // CHECK:  ret float %[[RES]]
+  return dot2add(p1, p2, p3);
+}
+
+// CHECK-LABEL: define {{.*}}dot2add_impl
+// CHECK-SPIRV:  %[[MUL:.*]] = call reassoc nnan ninf nsz arcp afn half @llvm.spv.fdot.v2f16(<2 x half> %1, <2 x half> %2)
+// CHECK-SPIRV:  %[[CONV:.*]] = fpext reassoc nnan ninf nsz arcp afn half %[[MUL]] to float
+// CHECK-SPIRV:  %[[C:.*]] = load float, ptr %c.addr, align 4
+// CHECK-SPIRV:  %[[RES:.*]] = fadd reassoc nnan ninf nsz arcp afn float %[[CONV]], %[[C]]
+// CHECK-DXIL:  %[[RES:.*]] = call {{.*}} float @llvm.dx.dot2add.v2f16(<2 x half> %0, <2 x half> %1, float %2)
+// CHECK:  ret float %[[RES]]
diff --git a/clang/test/SemaHLSL/BuiltIns/Dot2Add-errors.hlsl b/clang/test/SemaHLSL/BuiltIns/Dot2Add-errors.hlsl
deleted file mode 100644
index 87bcb931aabe5..0000000000000
--- a/clang/test/SemaHLSL/BuiltIns/Dot2Add-errors.hlsl
+++ /dev/null
@@ -1,36 +0,0 @@
-// RUN: %clang_cc1 -finclude-default-header -triple dxil-pc-shadermodel6.6-library %s -fnative-half-type -emit-llvm-only -disable-llvm-passes -verify
-
-float test_too_few_arg() {
-  return __builtin_hlsl_dot2add();
-  // expected-error at -1 {{too few arguments to function call, expected 3, have 0}}
-}
-
-float test_too_many_arg(half2 p1, half2 p2, float p3) {
-  return __builtin_hlsl_dot2add(p1, p2, p3, p1);
-  // expected-error at -1 {{too many arguments to function call, expected 3, have 4}}
-}
-
-float test_float_arg2_type(half2 p1, float2 p2, float p3) {
-  return __builtin_hlsl_dot2add(p1, p2, p3);
-  // expected-error at -1 {{passing 'float2' (aka 'vector<float, 2>') to parameter of incompatible type '__attribute__((__vector_size__(2 * sizeof(half)))) half' (vector of 2 'half' values)}}
-}
-
-float test_float_arg1_type(float2 p1, half2 p2, float p3) {
-  return __builtin_hlsl_dot2add(p1, p2, p3);
-  // expected-error at -1 {{passing 'float2' (aka 'vector<float, 2>') to parameter of incompatible type '__attribute__((__vector_size__(2 * sizeof(half)))) half' (vector of 2 'half' values)}}
-}
-
-float test_double_arg3_type(half2 p1, half2 p2, double p3) {
-  return __builtin_hlsl_dot2add(p1, p2, p3);
-  // expected-error at -1 {{passing 'double' to parameter of incompatible type 'float'}}
-}
-
-float test_float_arg1_arg2_type(float2 p1, float2 p2, float p3) {
-  return __builtin_hlsl_dot2add(p1, p2, p3);
-  // expected-error at -1 {{passing 'float2' (aka 'vector<float, 2>') to parameter of incompatible type '__attribute__((__vector_size__(2 * sizeof(half)))) half' (vector of 2 'half' values)}}
-}
-
-float test_int16_arg1_arg2_type(int16_t2 p1, int16_t2 p2, float p3) {
-  return __builtin_hlsl_dot2add(p1, p2, p3);
-  // expected-error at -1 {{passing 'int16_t2' (aka 'vector<int16_t, 2>') to parameter of incompatible type '__attribute__((__vector_size__(2 * sizeof(half)))) half' (vector of 2 'half' values)}}
-}
diff --git a/clang/test/SemaHLSL/BuiltIns/dot2add-errors.hlsl b/clang/test/SemaHLSL/BuiltIns/dot2add-errors.hlsl
new file mode 100644
index 0000000000000..262ceecbf1d90
--- /dev/null
+++ b/clang/test/SemaHLSL/BuiltIns/dot2add-errors.hlsl
@@ -0,0 +1,13 @@
+// RUN: %clang_cc1 -finclude-default-header -triple dxil-pc-shadermodel6.6-library %s -fnative-half-type -emit-llvm-only -disable-llvm-passes -verify
+
+float test_too_few_arg() {
+  return dot2add();
+  // expected-error at -1 {{no matching function for call to 'dot2add'}}
+  // expected-note at hlsl/hlsl_intrinsics.h:* {{candidate function not viable: requires 3 arguments, but 0 were provided}}
+}
+
+float test_too_many_arg(half2 p1, half2 p2, float p3) {
+  return dot2add(p1, p2, p3, p1);
+  // expected-error at -1 {{no matching function for call to 'dot2add'}}
+  // expected-note at hlsl/hlsl_intrinsics.h:* {{candidate function not viable: requires 3 arguments, but 4 were provided}}
+}

>From 5f14db1edb0632a3a19772f051d6f370c0620edb Mon Sep 17 00:00:00 2001
From: Sumit Agarwal <sumitagarwal at microsoft.com>
Date: Wed, 26 Mar 2025 17:54:39 -0700
Subject: [PATCH 12/12] Moved the hlsl codegen to target specific path

---
 clang/lib/CodeGen/TargetBuiltins/AMDGPU.cpp | 12 ++++++++++++
 1 file changed, 12 insertions(+)

diff --git a/clang/lib/CodeGen/TargetBuiltins/AMDGPU.cpp b/clang/lib/CodeGen/TargetBuiltins/AMDGPU.cpp
index 72f9e5a8174d2..0212ff8d668d2 100644
--- a/clang/lib/CodeGen/TargetBuiltins/AMDGPU.cpp
+++ b/clang/lib/CodeGen/TargetBuiltins/AMDGPU.cpp
@@ -639,6 +639,18 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
         getDotProductIntrinsic(CGM.getHLSLRuntime(), VecTy0->getElementType()),
         ArrayRef<Value *>{Op0, Op1}, nullptr, "hlsl.dot");
   }
+  case Builtin::BI__builtin_hlsl_dot2add: {
+    llvm::Triple::ArchType Arch = CGM.getTarget().getTriple().getArch();
+    assert(Arch == llvm::Triple::dxil && "Intrinsic dot2add can be executed as a builtin only on dxil");
+    Value *A = EmitScalarExpr(E->getArg(0));
+    Value *B = EmitScalarExpr(E->getArg(1));
+    Value *C = EmitScalarExpr(E->getArg(2));
+
+    Intrinsic::ID ID = llvm ::Intrinsic::dx_dot2add;
+    return Builder.CreateIntrinsic(
+        /*ReturnType=*/C->getType(), ID, ArrayRef<Value *>{A, B, C}, nullptr,
+        "dx.dot2add");
+  }
   case Builtin::BI__builtin_hlsl_dot4add_i8packed: {
     Value *A = EmitScalarExpr(E->getArg(0));
     Value *B = EmitScalarExpr(E->getArg(1));



More information about the llvm-commits mailing list