[DirectX] Add target builtins (#134439)

farzonl · web-flow · commit 16c84c4475b9 · 2025-04-07T12:06:57.000-04:00
- fixes #132303
- Moves dot2add from a language builtin to a target builtin.
-  Sets the scaffolding for Sema checks for DX builtins
-  Setup DirectX backend as able to have target builtins
- Adds a DX TargetBuiltins emitter in
`clang/lib/CodeGen/TargetBuiltins/DirectX.cpp`
diff --git a/.github/new-prs-labeler.yml b/.github/new-prs-labeler.yml
@@ -660,6 +660,12 @@ backend:DirectX:
   - '**/*dxil*/**'
   - '**/*DXContainer*'
   - '**/*DXContainer*/**'
+  - 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/**
 
 backend:SPIR-V:
   - clang/lib/Driver/ToolChains/SPIRV.*
diff --git a/clang/include/clang/Basic/Builtins.td b/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];
diff --git a/clang/include/clang/Basic/BuiltinsDirectX.td b/clang/include/clang/Basic/BuiltinsDirectX.td
@@ -0,0 +1,15 @@
+//===--- BuiltinsDirectX.td - DirectX Builtin function database -----------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+include "clang/Basic/BuiltinsBase.td"
+
+def DxDot2Add : Builtin { 
+  let Spellings = ["__builtin_dx_dot2add"];
+  let Attributes = [NoThrow, Const];
+  let Prototype = "float(_ExtVector<2, _Float16>, _ExtVector<2, _Float16>, float)";
+}
diff --git a/clang/include/clang/Basic/CMakeLists.txt b/clang/include/clang/Basic/CMakeLists.txt
@@ -82,6 +82,10 @@ clang_tablegen(BuiltinsBPF.inc -gen-clang-builtins
   SOURCE BuiltinsBPF.td
   TARGET ClangBuiltinsBPF)
 
+clang_tablegen(BuiltinsDirectX.inc -gen-clang-builtins
+  SOURCE BuiltinsDirectX.td
+  TARGET ClangBuiltinsDirectX)
+
 clang_tablegen(BuiltinsHexagon.inc -gen-clang-builtins
   SOURCE BuiltinsHexagon.td
   TARGET ClangBuiltinsHexagon)
diff --git a/clang/include/clang/Basic/TargetBuiltins.h b/clang/include/clang/Basic/TargetBuiltins.h
@@ -141,6 +141,17 @@ namespace clang {
   };
   }
 
+  /// DirectX builtins
+  namespace DirectX {
+  enum {
+    LastTIBuiltin = clang::Builtin::FirstTSBuiltin - 1,
+#define GET_BUILTIN_ENUMERATORS
+#include "clang/Basic/BuiltinsDirectX.inc"
+#undef GET_BUILTIN_ENUMERATORS
+    LastTSBuiltin
+  };
+  } // namespace DirectX
+
   /// SPIRV builtins
   namespace SPIRV {
   enum {
diff --git a/clang/include/clang/Sema/Sema.h b/clang/include/clang/Sema/Sema.h
@@ -160,6 +160,7 @@ class SemaAVR;
 class SemaBPF;
 class SemaCodeCompletion;
 class SemaCUDA;
+class SemaDirectX;
 class SemaHLSL;
 class SemaHexagon;
 class SemaLoongArch;
@@ -1074,6 +1075,11 @@ class Sema final : public SemaBase {
     return *CUDAPtr;
   }
 
+  SemaDirectX &DirectX() {
+    assert(DirectXPtr);
+    return *DirectXPtr;
+  }
+
   SemaHLSL &HLSL() {
     assert(HLSLPtr);
     return *HLSLPtr;
@@ -1212,6 +1218,7 @@ class Sema final : public SemaBase {
   std::unique_ptr<SemaBPF> BPFPtr;
   std::unique_ptr<SemaCodeCompletion> CodeCompletionPtr;
   std::unique_ptr<SemaCUDA> CUDAPtr;
+  std::unique_ptr<SemaDirectX> DirectXPtr;
   std::unique_ptr<SemaHLSL> HLSLPtr;
   std::unique_ptr<SemaHexagon> HexagonPtr;
   std::unique_ptr<SemaLoongArch> LoongArchPtr;
diff --git a/clang/include/clang/Sema/SemaDirectX.h b/clang/include/clang/Sema/SemaDirectX.h
@@ -0,0 +1,28 @@
+//===----- SemaDirectX.h ----- Semantic Analysis for DirectX constructs----===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+/// \file
+/// This file declares semantic analysis for DirectX constructs.
+///
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_CLANG_SEMA_SEMADIRECTX_H
+#define LLVM_CLANG_SEMA_SEMADIRECTX_H
+
+#include "clang/AST/ASTFwd.h"
+#include "clang/Sema/SemaBase.h"
+
+namespace clang {
+class SemaDirectX : public SemaBase {
+public:
+  SemaDirectX(Sema &S);
+
+  bool CheckDirectXBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall);
+};
+} // namespace clang
+
+#endif // LLVM_CLANG_SEMA_SEMADIRECTX_H
diff --git a/clang/lib/Basic/Targets/DirectX.cpp b/clang/lib/Basic/Targets/DirectX.cpp
@@ -12,11 +12,31 @@
 
 #include "DirectX.h"
 #include "Targets.h"
+#include "clang/Basic/TargetBuiltins.h"
 
 using namespace clang;
 using namespace clang::targets;
 
+static constexpr int NumBuiltins =
+    clang::DirectX::LastTSBuiltin - Builtin::FirstTSBuiltin;
+
+#define GET_BUILTIN_STR_TABLE
+#include "clang/Basic/BuiltinsDirectX.inc"
+#undef GET_BUILTIN_STR_TABLE
+
+static constexpr Builtin::Info BuiltinInfos[] = {
+#define GET_BUILTIN_INFOS
+#include "clang/Basic/BuiltinsDirectX.inc"
+#undef GET_BUILTIN_INFOS
+};
+static_assert(std::size(BuiltinInfos) == NumBuiltins);
+
 void DirectXTargetInfo::getTargetDefines(const LangOptions &Opts,
                                          MacroBuilder &Builder) const {
   DefineStd(Builder, "DIRECTX", Opts);
 }
+
+llvm::SmallVector<Builtin::InfosShard>
+DirectXTargetInfo::getTargetBuiltins() const {
+  return {{&BuiltinStrings, BuiltinInfos}};
+}
diff --git a/clang/lib/Basic/Targets/DirectX.h b/clang/lib/Basic/Targets/DirectX.h
@@ -73,9 +73,7 @@ class LLVM_LIBRARY_VISIBILITY DirectXTargetInfo : public TargetInfo {
     return Feature == "directx";
   }
 
-  llvm::SmallVector<Builtin::InfosShard> getTargetBuiltins() const override {
-    return {};
-  }
+  llvm::SmallVector<Builtin::InfosShard> getTargetBuiltins() const override;
 
   std::string_view getClobbers() const override { return ""; }
 
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -70,6 +70,8 @@ static Value *EmitTargetArchBuiltinExpr(CodeGenFunction *CGF,
   case llvm::Triple::bpfeb:
   case llvm::Triple::bpfel:
     return CGF->EmitBPFBuiltinExpr(BuiltinID, E);
+  case llvm::Triple::dxil:
+    return CGF->EmitDirectXBuiltinExpr(BuiltinID, E);
   case llvm::Triple::x86:
   case llvm::Triple::x86_64:
     return CGF->EmitX86BuiltinExpr(BuiltinID, E);
diff --git a/clang/lib/CodeGen/CGHLSLBuiltins.cpp b/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));
diff --git a/clang/lib/CodeGen/CMakeLists.txt b/clang/lib/CodeGen/CMakeLists.txt
@@ -118,6 +118,7 @@ add_clang_library(clangCodeGen
   SwiftCallingConv.cpp
   TargetBuiltins/ARM.cpp
   TargetBuiltins/AMDGPU.cpp
+  TargetBuiltins/DirectX.cpp
   TargetBuiltins/Hexagon.cpp
   TargetBuiltins/NVPTX.cpp
   TargetBuiltins/PPC.cpp
diff --git a/clang/lib/CodeGen/CodeGenFunction.h b/clang/lib/CodeGen/CodeGenFunction.h
@@ -4809,6 +4809,7 @@ class CodeGenFunction : public CodeGenTypeCache {
   llvm::Value *EmitAMDGPUBuiltinExpr(unsigned BuiltinID, const CallExpr *E);
   llvm::Value *EmitHLSLBuiltinExpr(unsigned BuiltinID, const CallExpr *E,
                                    ReturnValueSlot ReturnValue);
+  llvm::Value *EmitDirectXBuiltinExpr(unsigned BuiltinID, const CallExpr *E);
   llvm::Value *EmitSPIRVBuiltinExpr(unsigned BuiltinID, const CallExpr *E);
   llvm::Value *EmitScalarOrConstFoldImmArg(unsigned ICEArguments, unsigned Idx,
                                            const CallExpr *E);
diff --git a/clang/lib/CodeGen/TargetBuiltins/DirectX.cpp b/clang/lib/CodeGen/TargetBuiltins/DirectX.cpp
@@ -0,0 +1,37 @@
+//===--------- DirectX.cpp - Emit LLVM Code for builtins ------------------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+//
+// This contains code to emit Builtin calls as LLVM code.
+//
+//===----------------------------------------------------------------------===//
+
+#include "CGHLSLRuntime.h"
+#include "CodeGenFunction.h"
+#include "clang/Basic/TargetBuiltins.h"
+#include "llvm/IR/Intrinsics.h"
+
+using namespace clang;
+using namespace CodeGen;
+using namespace llvm;
+
+Value *CodeGenFunction::EmitDirectXBuiltinExpr(unsigned BuiltinID,
+                                               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;
+}
diff --git a/clang/lib/Headers/hlsl/hlsl_intrinsic_helpers.h b/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
diff --git a/clang/lib/Sema/CMakeLists.txt b/clang/lib/Sema/CMakeLists.txt
@@ -47,6 +47,7 @@ add_clang_library(clangSema
   SemaConsumer.cpp
   SemaCoroutine.cpp
   SemaCUDA.cpp
+  SemaDirectX.cpp
   SemaDecl.cpp
   SemaDeclAttr.cpp
   SemaDeclCXX.cpp
diff --git a/clang/lib/Sema/Sema.cpp b/clang/lib/Sema/Sema.cpp
@@ -47,6 +47,7 @@
 #include "clang/Sema/SemaCUDA.h"
 #include "clang/Sema/SemaCodeCompletion.h"
 #include "clang/Sema/SemaConsumer.h"
+#include "clang/Sema/SemaDirectX.h"
 #include "clang/Sema/SemaHLSL.h"
 #include "clang/Sema/SemaHexagon.h"
 #include "clang/Sema/SemaLoongArch.h"
@@ -226,6 +227,7 @@ Sema::Sema(Preprocessor &pp, ASTContext &ctxt, ASTConsumer &consumer,
       CodeCompletionPtr(
           std::make_unique<SemaCodeCompletion>(*this, CodeCompleter)),
       CUDAPtr(std::make_unique<SemaCUDA>(*this)),
+      DirectXPtr(std::make_unique<SemaDirectX>(*this)),
       HLSLPtr(std::make_unique<SemaHLSL>(*this)),
       HexagonPtr(std::make_unique<SemaHexagon>(*this)),
       LoongArchPtr(std::make_unique<SemaLoongArch>(*this)),
diff --git a/clang/lib/Sema/SemaChecking.cpp b/clang/lib/Sema/SemaChecking.cpp
@@ -61,6 +61,7 @@
 #include "clang/Sema/SemaAMDGPU.h"
 #include "clang/Sema/SemaARM.h"
 #include "clang/Sema/SemaBPF.h"
+#include "clang/Sema/SemaDirectX.h"
 #include "clang/Sema/SemaHLSL.h"
 #include "clang/Sema/SemaHexagon.h"
 #include "clang/Sema/SemaLoongArch.h"
@@ -1930,6 +1931,8 @@ bool Sema::CheckTSBuiltinFunctionCall(const TargetInfo &TI, unsigned BuiltinID,
   case llvm::Triple::bpfeb:
   case llvm::Triple::bpfel:
     return BPF().CheckBPFBuiltinFunctionCall(BuiltinID, TheCall);
+  case llvm::Triple::dxil:
+    return DirectX().CheckDirectXBuiltinFunctionCall(BuiltinID, TheCall);
   case llvm::Triple::hexagon:
     return Hexagon().CheckHexagonBuiltinFunctionCall(BuiltinID, TheCall);
   case llvm::Triple::mips:
diff --git a/clang/lib/Sema/SemaDirectX.cpp b/clang/lib/Sema/SemaDirectX.cpp
@@ -0,0 +1,23 @@
+//===- SemaDirectX.cpp - Semantic Analysis for DirectX constructs----------===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+// This implements Semantic Analysis for DirectX constructs.
+//===----------------------------------------------------------------------===//
+
+#include "clang/Sema/SemaDirectX.h"
+#include "clang/Basic/TargetBuiltins.h"
+#include "clang/Sema/Sema.h"
+
+namespace clang {
+
+SemaDirectX::SemaDirectX(Sema &S) : SemaBase(S) {}
+
+bool SemaDirectX::CheckDirectXBuiltinFunctionCall(unsigned BuiltinID,
+                                                  CallExpr *TheCall) {
+  return false;
+}
+} // namespace clang
diff --git a/clang/test/CodeGenDirectX/Builtins/dot2add.c b/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); }

Original file line number	Diff line number	Diff line change
`@@ -73,9 +73,7 @@ class LLVM_LIBRARY_VISIBILITY DirectXTargetInfo : public TargetInfo {`
`73`	`73`	`return Feature == "directx";`
`74`	`74`	`}`
`75`	`75`
`76`		`- llvm::SmallVector<Builtin::InfosShard> getTargetBuiltins() const override {`
`77`		`- return {};`
`78`		`- }`
	`76`	`+ llvm::SmallVector<Builtin::InfosShard> getTargetBuiltins() const override;`
`79`	`77`
`80`	`78`	`std::string_view getClobbers() const override { return ""; }`
`81`	`79`