move dot2add to DirectX.cpp TargetBuiltins

Author: farzonl

.github/new-prs-labeler.yml

@@ -663,6 +663,7 @@ backend:DirectX:
   - clang/lib/Sema/SemaDirectX.cpp
   - clang/include/clang/Sema/SemaDirectX.h
   - clang/include/clang/Basic/BuiltinsDirectX.td
+  - clang/lib/CodeGen/TargetBuiltins/DirectX.cpp
   - clang/test/CodeGenDirectX/**
   - clang/test/SemaDirectX/**
 

clang/include/clang/Basic/Builtins.td

@@ -4891,12 +4891,6 @@ def HLSLDotProduct : LangBuiltin<"HLSL_LANG"> {
   let Prototype = "void(...)";
 }
 
-def HLSLDot2Add : LangBuiltin<"HLSL_LANG"> {
-  let Spellings = ["__builtin_hlsl_dot2add"];
-  let Attributes = [NoThrow, Const];
-  let Prototype = "float(_ExtVector<2, _Float16>, _ExtVector<2, _Float16>, float)";
-}
-
 def HLSLDot4AddI8Packed : LangBuiltin<"HLSL_LANG"> {
   let Spellings = ["__builtin_hlsl_dot4add_i8packed"];
   let Attributes = [NoThrow, Const];

clang/include/clang/Basic/BuiltinsDirectX.td

@@ -11,5 +11,5 @@ include "clang/Basic/BuiltinsBase.td"
 def DxDot2Add : Builtin { 
   let Spellings = ["__builtin_dx_dot2add"];
   let Attributes = [NoThrow, Const];
-  let Prototype = "float(_ExtVector<2, __fp16>,_ExtVector<2, __fp16>,float)";
+  let Prototype = "float(_ExtVector<2, _Float16>, _ExtVector<2, _Float16>, float)";
 }

clang/include/clang/Sema/SemaDirectX.h

@@ -1,5 +1,5 @@
 //===----- SemaDirectX.h ----- Semantic Analysis for DirectX
-//constructs--------===//
+// constructs--------===//
 //
 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
 // See https://llvm.org/LICENSE.txt for license information.

clang/lib/CodeGen/CGHLSLBuiltins.cpp

@@ -380,18 +380,6 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
         getDotProductIntrinsic(CGM.getHLSLRuntime(), VecTy0->getElementType()),
         ArrayRef<Value *>{Op0, Op1}, nullptr, "hlsl.dot");
   }
-  case Builtin::BI__builtin_hlsl_dot2add: {
-    assert(CGM.getTarget().getTriple().getArch() == llvm::Triple::dxil &&
-           "Intrinsic dot2add is only allowed for dxil architecture");
-    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));

clang/lib/CodeGen/TargetBuiltins/DirectX.cpp

@@ -20,6 +20,18 @@ using namespace CodeGen;
 using namespace llvm;
 
 Value *CodeGenFunction::EmitDirectXBuiltinExpr(unsigned BuiltinID,
-       const CallExpr *E) {
-return nullptr;
+                                               const CallExpr *E) {
+  switch (BuiltinID) {
+  case DirectX::BI__builtin_dx_dot2add: {
+    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");
+  }
+  }
+  return nullptr;
 }

clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h

@@ -46,8 +46,8 @@ distance_vec_impl(vector<T, N> X, vector<T, N> Y) {
 }
 
 constexpr float dot2add_impl(half2 a, half2 b, float c) {
-#if defined(__DIRECTX__)
-  return __builtin_hlsl_dot2add(a, b, c);
+#if (__has_builtin(__builtin_dx_dot2add))
+  return __builtin_dx_dot2add(a, b, c);
 #else
   return dot(a, b) + c;
 #endif

clang/lib/Sema/SemaDirectX.cpp

@@ -18,11 +18,6 @@ SemaDirectX::SemaDirectX(Sema &S) : SemaBase(S) {}
 
 bool SemaDirectX::CheckDirectXBuiltinFunctionCall(unsigned BuiltinID,
                                                   CallExpr *TheCall) {
-  switch (BuiltinID) {
-  case DirectX::BI__builtin_dx_dot2add: {
-  }
-  }
-
   return false;
 }
 } // namespace clang

clang/test/CodeGenDirectX/Builtins/dot2add.c

@@ -0,0 +1,23 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 5
+
+// RUN: %clang_cc1 -triple dxil-pc-shadermodel6.3-library %s -emit-llvm -o - | FileCheck %s
+
+typedef _Float16 half;
+typedef half half2 __attribute__((ext_vector_type(2)));
+
+// CHECK-LABEL: define float @test_dot2add(
+// CHECK-SAME: <2 x half> noundef [[X:%.*]], <2 x half> noundef [[Y:%.*]], float noundef [[Z:%.*]]) #[[ATTR0:[0-9]+]] {
+// CHECK-NEXT:  [[ENTRY:.*:]]
+// CHECK-NEXT:    [[X_ADDR:%.*]] = alloca <2 x half>, align 4
+// CHECK-NEXT:    [[Y_ADDR:%.*]] = alloca <2 x half>, align 4
+// CHECK-NEXT:    [[Z_ADDR:%.*]] = alloca float, align 4
+// CHECK-NEXT:    store <2 x half> [[X]], ptr [[X_ADDR]], align 4
+// CHECK-NEXT:    store <2 x half> [[Y]], ptr [[Y_ADDR]], align 4
+// CHECK-NEXT:    store float [[Z]], ptr [[Z_ADDR]], align 4
+// CHECK-NEXT:    [[TMP0:%.*]] = load <2 x half>, ptr [[X_ADDR]], align 4
+// CHECK-NEXT:    [[TMP1:%.*]] = load <2 x half>, ptr [[Y_ADDR]], align 4
+// CHECK-NEXT:    [[TMP2:%.*]] = load float, ptr [[Z_ADDR]], align 4
+// CHECK-NEXT:    [[DX_DOT2ADD:%.*]] = call float @llvm.dx.dot2add.v2f16(<2 x half> [[TMP0]], <2 x half> [[TMP1]], float [[TMP2]])
+// CHECK-NEXT:    ret float [[DX_DOT2ADD]]
+//
+float test_dot2add(half2 X, half2 Y, float Z) { return __builtin_dx_dot2add(X, Y, Z); }