[HLSL] Add load overload with status #166449

bob80905 · 2025-11-04T21:36:06Z

This PR adds a new overload for resources, which takes an additional parameter by reference, status. It fills the status parameter with a 1 or 0, depending on whether or not the resource access was mapped. CheckAccessFullyMapped is also added as an intrinsic, and called in the production of this status bit.

llvmbot · 2025-11-04T21:36:41Z

@llvm/pr-subscribers-backend-x86
@llvm/pr-subscribers-clang
@llvm/pr-subscribers-backend-directx

@llvm/pr-subscribers-hlsl

Author: Joshua Batista (bob80905)

Changes

This PR adds a new overload for resources, which takes an additional parameter by reference, status. It fills the status parameter with a 1 or 0, depending on whether or not the resource access was mapped. CheckAccessFullyMapped is also added as an intrinsic, and called in the production of this status bit.

Patch is 20.86 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/166449.diff

11 Files Affected:

(modified) clang/include/clang/Basic/Builtins.td (+6)
(modified) clang/lib/CodeGen/CGHLSLBuiltins.cpp (+58)
(modified) clang/lib/CodeGen/CGHLSLRuntime.h (+3)
(modified) clang/lib/Sema/HLSLBuiltinTypeDeclBuilder.cpp (+35)
(modified) clang/lib/Sema/HLSLBuiltinTypeDeclBuilder.h (+2)
(modified) clang/lib/Sema/SemaHLSL.cpp (+21)
(modified) clang/test/CodeGenHLSL/resources/StructuredBuffers-methods-lib.hlsl (+44)
(modified) clang/test/CodeGenHLSL/resources/StructuredBuffers-methods-ps.hlsl (+43)
(modified) clang/test/CodeGenHLSL/resources/TypedBuffers-methods.hlsl (+44)
(modified) llvm/include/llvm/IR/IntrinsicsDirectX.td (+3)
(modified) llvm/include/llvm/IR/IntrinsicsSPIRV.td (+3)

diff --git a/clang/include/clang/Basic/Builtins.td b/clang/include/clang/Basic/Builtins.td
index 2b400b012d6ed..f2b4c54a5ba6b 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4934,6 +4934,12 @@ def HLSLResourceGetPointer : LangBuiltin<"HLSL_LANG"> {
   let Prototype = "void(...)";
 }
 
+def HLSLResourceLoadWithStatus : LangBuiltin<"HLSL_LANG"> {
+  let Spellings = ["__builtin_hlsl_resource_load_with_status"];
+  let Attributes = [NoThrow];
+  let Prototype = "void(...)";
+}
+
 def HLSLResourceUninitializedHandle : LangBuiltin<"HLSL_LANG"> {
   let Spellings = ["__builtin_hlsl_resource_uninitializedhandle"];
   let Attributes = [NoThrow];
diff --git a/clang/lib/CodeGen/CGHLSLBuiltins.cpp b/clang/lib/CodeGen/CGHLSLBuiltins.cpp
index fbf4a5722caed..ff3d906d4bf11 100644
--- a/clang/lib/CodeGen/CGHLSLBuiltins.cpp
+++ b/clang/lib/CodeGen/CGHLSLBuiltins.cpp
@@ -353,6 +353,64 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
         RetTy, CGM.getHLSLRuntime().getCreateResourceGetPointerIntrinsic(),
         ArrayRef<Value *>{HandleOp, IndexOp});
   }
+  case Builtin::BI__builtin_hlsl_resource_load_with_status: {
+    Value *HandleOp = EmitScalarExpr(E->getArg(0));
+    Value *IndexOp = EmitScalarExpr(E->getArg(1));
+
+    // Get the *address* of the status argument (since it's a reference)
+    LValue StatusLVal = EmitLValue(E->getArg(2));
+    Address StatusAddr = StatusLVal.getAddress();
+
+    QualType HandleTy = E->getArg(0)->getType();
+    const HLSLAttributedResourceType *RT =
+        HandleTy->getAs<HLSLAttributedResourceType>();
+    assert(RT && "Expected a resource type as first parameter");
+
+    Intrinsic::ID IntrID = RT->getAttrs().RawBuffer
+                               ? llvm::Intrinsic::dx_resource_load_rawbuffer
+                               : llvm::Intrinsic::dx_resource_load_typedbuffer;
+
+    llvm::Type *DataTy = ConvertType(E->getType());
+    llvm::Type *RetTy = llvm::StructType::get(Builder.getContext(),
+                                              {DataTy, Builder.getInt1Ty()});
+
+    SmallVector<Value *, 3> Args;
+    Args.push_back(HandleOp);
+    Args.push_back(IndexOp);
+
+    if (RT->getAttrs().RawBuffer) {
+      Args.push_back(Builder.getInt32(0)); // dummy offset
+    }
+
+    // Call the intrinsic (returns a struct)
+    Value *ResRet =
+        Builder.CreateIntrinsic(RetTy, IntrID, Args, {}, "ld.struct");
+
+    // Extract the loaded data (first element of the struct)
+    Value *LoadedValue = Builder.CreateExtractValue(ResRet, {0}, "ld.value");
+
+    // Extract the status bit (second element of the struct)
+    Value *StatusBit = Builder.CreateExtractValue(ResRet, {1}, "ld.status");
+
+    // Extend the 1-bit status to 32-bit, because the check access fully mapped
+    // intrinsic expects i32 input
+    Value *StatusInt =
+        Builder.CreateZExt(StatusBit, Builder.getInt32Ty(), "ld.status.i32");
+
+    // Call CheckAccessFullyMapped(i32)
+    llvm::Function *CheckAccessFn =
+        CGM.getIntrinsic(llvm::Intrinsic::dx_check_access_fully_mapped);
+
+    Value *StatusBool =
+        Builder.CreateCall(CheckAccessFn, {StatusInt}, "ld.status.checked");
+
+    // Extend the boolean to 32-bit uint for storing in user�s var
+    Value *ExtendedStatus =
+        Builder.CreateZExt(StatusBool, Builder.getInt32Ty(), "ld.status.ext");
+
+    Builder.CreateStore(ExtendedStatus, StatusAddr);
+    return LoadedValue;
+  }
   case Builtin::BI__builtin_hlsl_resource_uninitializedhandle: {
     llvm::Type *HandleTy = CGM.getTypes().ConvertType(E->getType());
     return llvm::PoisonValue::get(HandleTy);
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h b/clang/lib/CodeGen/CGHLSLRuntime.h
index d35df524fdc84..f6c654bd61ff5 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.h
+++ b/clang/lib/CodeGen/CGHLSLRuntime.h
@@ -126,6 +126,9 @@ class CGHLSLRuntime {
 
   GENERATE_HLSL_INTRINSIC_FUNCTION(CreateResourceGetPointer,
                                    resource_getpointer)
+
+  GENERATE_HLSL_INTRINSIC_FUNCTION(CreateResourceLoadTypedBuffer,
+                                   resource_load_typedbuffer)
   GENERATE_HLSL_INTRINSIC_FUNCTION(CreateHandleFromBinding,
                                    resource_handlefrombinding)
   GENERATE_HLSL_INTRINSIC_FUNCTION(CreateHandleFromImplicitBinding,
diff --git a/clang/lib/Sema/HLSLBuiltinTypeDeclBuilder.cpp b/clang/lib/Sema/HLSLBuiltinTypeDeclBuilder.cpp
index 066acf6f01a90..bdbc2d4ce87b3 100644
--- a/clang/lib/Sema/HLSLBuiltinTypeDeclBuilder.cpp
+++ b/clang/lib/Sema/HLSLBuiltinTypeDeclBuilder.cpp
@@ -1140,6 +1140,7 @@ BuiltinTypeDeclBuilder &BuiltinTypeDeclBuilder::addLoadMethods() {
   DeclarationName Load(&II);
   // TODO: We also need versions with status for CheckAccessFullyMapped.
   addHandleAccessFunction(Load, /*IsConst=*/false, /*IsRef=*/false);
+  addLoadWithStatusFunction(Load, /*IsConst=*/false, /*IsRef=*/false);
 
   return *this;
 }
@@ -1232,6 +1233,40 @@ BuiltinTypeDeclBuilder &BuiltinTypeDeclBuilder::addDecrementCounterMethod() {
       .finalize();
 }
 
+BuiltinTypeDeclBuilder &
+BuiltinTypeDeclBuilder::addLoadWithStatusFunction(DeclarationName &Name,
+                                                  bool IsConst, bool IsRef) {
+  assert(!Record->isCompleteDefinition() && "record is already complete");
+  ASTContext &AST = SemaRef.getASTContext();
+  using PH = BuiltinTypeMethodBuilder::PlaceHolder;
+
+  QualType ElemTy = getHandleElementType();
+  QualType AddrSpaceElemTy =
+      AST.getAddrSpaceQualType(ElemTy, LangAS::hlsl_device);
+  QualType ElemPtrTy = AST.getPointerType(AddrSpaceElemTy);
+  QualType ReturnTy;
+
+  if (IsRef) {
+    ReturnTy = AddrSpaceElemTy;
+    if (IsConst)
+      ReturnTy.addConst();
+    ReturnTy = AST.getLValueReferenceType(ReturnTy);
+  } else {
+    ReturnTy = ElemTy;
+    if (IsConst)
+      ReturnTy.addConst();
+  }
+
+  QualType StatusRefTy = AST.getLValueReferenceType(AST.UnsignedIntTy);
+  return BuiltinTypeMethodBuilder(*this, Name, ReturnTy, IsConst)
+      .addParam("Index", AST.UnsignedIntTy)
+      .addParam("Status", StatusRefTy)
+      .callBuiltin("__builtin_hlsl_resource_load_with_status", ElemPtrTy,
+                   PH::Handle, PH::_0, PH::_1)
+      .dereference(PH::LastStmt)
+      .finalize();
+}
+
 BuiltinTypeDeclBuilder &
 BuiltinTypeDeclBuilder::addHandleAccessFunction(DeclarationName &Name,
                                                 bool IsConst, bool IsRef) {
diff --git a/clang/lib/Sema/HLSLBuiltinTypeDeclBuilder.h b/clang/lib/Sema/HLSLBuiltinTypeDeclBuilder.h
index 95e3a6c4fb2f1..4941b8b7952a2 100644
--- a/clang/lib/Sema/HLSLBuiltinTypeDeclBuilder.h
+++ b/clang/lib/Sema/HLSLBuiltinTypeDeclBuilder.h
@@ -91,6 +91,8 @@ class BuiltinTypeDeclBuilder {
   BuiltinTypeDeclBuilder &addDecrementCounterMethod();
   BuiltinTypeDeclBuilder &addHandleAccessFunction(DeclarationName &Name,
                                                   bool IsConst, bool IsRef);
+  BuiltinTypeDeclBuilder &addLoadWithStatusFunction(DeclarationName &Name,
+                                                    bool IsConst, bool IsRef);
   BuiltinTypeDeclBuilder &addAppendMethod();
   BuiltinTypeDeclBuilder &addConsumeMethod();
 
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index 94a490a8f68dc..f95a647b5679f 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -3010,6 +3010,27 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
 
     break;
   }
+  case Builtin::BI__builtin_hlsl_resource_load_with_status: {
+    if (SemaRef.checkArgCount(TheCall, 3) ||
+        CheckResourceHandle(&SemaRef, TheCall, 0) ||
+        CheckArgTypeMatches(&SemaRef, TheCall->getArg(1),
+                            SemaRef.getASTContext().UnsignedIntTy) ||
+        CheckArgTypeMatches(&SemaRef, TheCall->getArg(2),
+                            SemaRef.getASTContext().UnsignedIntTy))
+      return true;
+
+    auto *ResourceTy =
+        TheCall->getArg(0)->getType()->castAs<HLSLAttributedResourceType>();
+    QualType ContainedTy = ResourceTy->getContainedType();
+    auto ReturnType =
+        SemaRef.Context.getAddrSpaceQualType(ContainedTy, LangAS::hlsl_device);
+    ReturnType = SemaRef.Context.getPointerType(ReturnType);
+    TheCall->setType(ReturnType);
+    TheCall->setValueKind(VK_LValue);
+
+    break;
+  }
+
   case Builtin::BI__builtin_hlsl_resource_uninitializedhandle: {
     if (SemaRef.checkArgCount(TheCall, 1) ||
         CheckResourceHandle(&SemaRef, TheCall, 0))
diff --git a/clang/test/CodeGenHLSL/resources/StructuredBuffers-methods-lib.hlsl b/clang/test/CodeGenHLSL/resources/StructuredBuffers-methods-lib.hlsl
index 1f248d0560006..e692a14ecc168 100644
--- a/clang/test/CodeGenHLSL/resources/StructuredBuffers-methods-lib.hlsl
+++ b/clang/test/CodeGenHLSL/resources/StructuredBuffers-methods-lib.hlsl
@@ -104,6 +104,50 @@ export float TestLoad() {
 // CHECK-NEXT: %[[VAL:.*]] = load float, ptr %[[PTR]]
 // CHECK-NEXT: ret float %[[VAL]]
 
+export float TestLoadWithStatus() {
+    uint s1;
+    uint s2;
+    float ret = RWSB1.Load(1, s1) + SB1.Load(2, s2);
+    ret += float(s1 + s2);
+    return ret;
+}
+
+// CHECK: define noundef nofpclass(nan inf) float @TestLoadWithStatus()()
+// CHECK: call {{.*}} float @hlsl::RWStructuredBuffer<float>::Load(unsigned int, unsigned int&)(ptr {{.*}} @RWSB1, i32 noundef 1, ptr noundef nonnull align 4 dereferenceable(4) %s1)
+// CHECK: call {{.*}} float @hlsl::StructuredBuffer<float>::Load(unsigned int, unsigned int&)(ptr {{.*}} @SB1, i32 noundef 2, ptr noundef nonnull align 4 dereferenceable(4) %s2)
+// CHECK: add
+// CHECK: ret float
+
+// CHECK: define {{.*}} float @hlsl::RWStructuredBuffer<float>::Load(unsigned int, unsigned int&)(ptr {{.*}} %this, i32 noundef %Index, ptr noundef nonnull align 4 dereferenceable(4) %Status)
+// CHECK: %__handle = getelementptr inbounds nuw %"class.hlsl::RWStructuredBuffer", ptr %{{.*}}, i32 0, i32 0
+// DXIL-NEXT: %[[HANDLE:.*]] = load target("dx.RawBuffer", float, 1, 0), ptr %__handle
+// CHECK-NEXT: %[[INDEX:.*]] = load i32, ptr %Index.addr
+// CHECK-NEXT: %[[STATUS:.*]] = load ptr, ptr %Status.addr,
+// DXIL-NEXT: %[[STRUCT:.*]] = call { ptr, i1 } @llvm.dx.resource.load.rawbuffer.p0.tdx.RawBuffer_f32_1_0t(target("dx.RawBuffer", float, 1, 0) %[[HANDLE]], i32 %[[INDEX]], i32 0)
+// CHECK-NEXT: %[[VALUE:.*]] = extractvalue { ptr, i1 } %[[STRUCT]], 0
+// CHECK-NEXT: %[[STATUS_TEMP:.*]] = extractvalue { ptr, i1 } %[[STRUCT]], 1
+// CHECK-NEXT: %[[STATUS_EXT:.*]] = zext i1 %[[STATUS_TEMP]] to i32
+// CHECK-NEXT: %[[STATUS_CHECKED:.*]] = call i1 @llvm.dx.check.access.fully.mapped(i32 %[[STATUS_EXT:.*]])
+// CHECK-NEXT: %[[STATUS_REEXT:.*]] = zext i1 %[[STATUS_CHECKED]] to i32
+// CHECK-NEXT: store i32 %[[STATUS_REEXT]], ptr %2, align 4
+// CHECK-NEXT: %[[RETVAL:.*]] = load float, ptr %[[VALUE]]
+// CHECK-NEXT: ret float %[[RETVAL]]
+
+// CHECK: define {{.*}} float @hlsl::StructuredBuffer<float>::Load(unsigned int, unsigned int&)(ptr {{.*}} %this, i32 noundef %Index, ptr noundef nonnull align 4 dereferenceable(4) %Status)
+// CHECK: %__handle = getelementptr inbounds nuw %"class.hlsl::StructuredBuffer", ptr %{{.*}}, i32 0, i32 0
+// DXIL-NEXT: %[[HANDLE:.*]] = load target("dx.RawBuffer", float, 0, 0), ptr %__handle
+// CHECK-NEXT: %[[INDEX:.*]] = load i32, ptr %Index.addr
+// CHECK-NEXT: %[[STATUS_HANDLE:.*]] = load ptr, ptr %Status.addr, align 4, !nonnull !3, !align !4
+// DXIL-NEXT: %[[STRUCT:.*]] = call { ptr, i1 } @llvm.dx.resource.load.rawbuffer.p0.tdx.RawBuffer_f32_0_0t(target("dx.RawBuffer", float, 0, 0) %0, i32 %1, i32 0)
+// CHECK-NEXT: %[[VALUE:.*]] = extractvalue { ptr, i1 } %[[STRUCT]], 0
+// CHECK-NEXT: %[[STATUS_TEMP:.*]] = extractvalue { ptr, i1 } %[[STRUCT]], 1
+// CHECK-NEXT: %[[STATUS_EXT:.*]] = zext i1 %[[STATUS_TEMP]] to i32
+// CHECK-NEXT: %[[STATUS_CHECKED:.*]] = call i1 @llvm.dx.check.access.fully.mapped(i32 %[[STATUS_EXT:.*]])
+// CHECK-NEXT: %[[STATUS_REEXT:.*]] = zext i1 %[[STATUS_CHECKED]] to i32
+// CHECK-NEXT: store i32 %[[STATUS_REEXT]], ptr %2, align 4
+// CHECK-NEXT: %[[RETVAL:.*]] = load float, ptr %[[VALUE]]
+// CHECK-NEXT: ret float %[[RETVAL]]
+
 export uint TestGetDimensions() {
     uint dim1, dim2, dim3, stride1, stride2, stride3;
     SB1.GetDimensions(dim1, stride1);
diff --git a/clang/test/CodeGenHLSL/resources/StructuredBuffers-methods-ps.hlsl b/clang/test/CodeGenHLSL/resources/StructuredBuffers-methods-ps.hlsl
index 25fa75965d686..77aaf281fbc39 100644
--- a/clang/test/CodeGenHLSL/resources/StructuredBuffers-methods-ps.hlsl
+++ b/clang/test/CodeGenHLSL/resources/StructuredBuffers-methods-ps.hlsl
@@ -65,6 +65,49 @@ export float TestLoad() {
 // CHECK-NEXT: %[[VAL:.*]] = load <2 x i32>, ptr %[[BUFPTR]]
 // CHECK-NEXT: ret <2 x i32> %[[VAL]]
 
+export float TestLoadWithStatus() {
+    uint status;
+    uint status2;
+    float val = ROSB1.Load(10, status).x + ROSB2.Load(20, status2).x;
+    return val + float(status + status2);
+}
+
+// CHECK: define {{.*}} float @TestLoadWithStatus()()
+// CHECK: call {{.*}} float @hlsl::RasterizerOrderedStructuredBuffer<float>::Load(unsigned int, unsigned int&)(ptr {{.*}} @ROSB1, i32 noundef 10, ptr noundef nonnull align 4 dereferenceable(4) %status)
+// CHECK: call {{.*}} <2 x i32> @hlsl::RasterizerOrderedStructuredBuffer<int vector[2]>::Load(unsigned int, unsigned int&)(ptr {{.*}} @ROSB2, i32 noundef 20, ptr noundef nonnull align 4 dereferenceable(4) %status2)
+// CHECK: ret
+
+// CHECK: define {{.*}} float @hlsl::RasterizerOrderedStructuredBuffer<float>::Load(unsigned int, unsigned int&)(ptr {{.*}} %Index, ptr noundef nonnull align 4 dereferenceable(4) %Status)
+// CHECK: %__handle = getelementptr inbounds nuw %"class.hlsl::RasterizerOrderedStructuredBuffer", ptr {{.*}}, i32 0, i32 0
+// CHECK-NEXT: %[[HANDLE:.*]] = load target("dx.RawBuffer", float, 1, 1), ptr %__handle
+// CHECK-NEXT: %[[INDEX:.*]] = load i32, ptr %Index.addr
+// CHECK-NEXT: %[[STATUS:.*]] = load ptr, ptr %Status.addr,
+// DXIL-NEXT: %[[STRUCT:.*]] = call { ptr, i1 } @llvm.dx.resource.load.rawbuffer.p0.tdx.RawBuffer_f32_1_1t(target("dx.RawBuffer", float, 1, 1) %[[HANDLE]], i32 %[[INDEX]], i32 0)
+// CHECK-NEXT: %[[VALUE:.*]] = extractvalue { ptr, i1 } %[[STRUCT]], 0
+// CHECK-NEXT: %[[STATUS_TEMP:.*]] = extractvalue { ptr, i1 } %[[STRUCT]], 1
+// CHECK-NEXT: %[[STATUS_EXT:.*]] = zext i1 %[[STATUS_TEMP]] to i32
+// CHECK-NEXT: %[[STATUS_CHECKED:.*]] = call i1 @llvm.dx.check.access.fully.mapped(i32 %[[STATUS_EXT:.*]])
+// CHECK-NEXT: %[[STATUS_REEXT:.*]] = zext i1 %[[STATUS_CHECKED]] to i32
+// CHECK-NEXT: store i32 %[[STATUS_REEXT]], ptr %2, align 4
+// CHECK-NEXT: %[[RETVAL:.*]] = load float, ptr %[[VALUE]]
+// CHECK-NEXT: ret float %[[RETVAL]]
+
+// CHECK: define {{.*}} <2 x i32> @hlsl::RasterizerOrderedStructuredBuffer<int vector[2]>::Load(unsigned int, unsigned int&)(ptr {{.*}} %Index, ptr noundef nonnull align 4 dereferenceable(4) %Status)
+// CHECK: %__handle = getelementptr inbounds nuw %"class.hlsl::RasterizerOrderedStructuredBuffer.0", ptr {{.*}}, i32 0, i32 0
+// CHECK-NEXT: %[[HANDLE:.*]] = load target("dx.RawBuffer", <2 x i32>, 1, 1), ptr %__handle
+// CHECK-NEXT: %[[INDEX:.*]] = load i32, ptr %Index.addr
+// CHECK-NEXT: %[[STATUS_HANDLE:.*]] = load ptr, ptr %Status.addr, align 4, !nonnull !3, !align !4
+// DXIL-NEXT: %[[STRUCT:.*]] = call { ptr, i1 } @llvm.dx.resource.load.rawbuffer.p0.tdx.RawBuffer_v2i32_1_1t(target("dx.RawBuffer", <2 x i32>, 1, 1) %0, i32 %1, i32 0)
+// CHECK-NEXT: %[[VALUE:.*]] = extractvalue { ptr, i1 } %[[STRUCT]], 0
+// CHECK-NEXT: %[[STATUS_TEMP:.*]] = extractvalue { ptr, i1 } %[[STRUCT]], 1
+// CHECK-NEXT: %[[STATUS_EXT:.*]] = zext i1 %[[STATUS_TEMP]] to i32
+// CHECK-NEXT: %[[STATUS_CHECKED:.*]] = call i1 @llvm.dx.check.access.fully.mapped(i32 %[[STATUS_EXT:.*]])
+// CHECK-NEXT: %[[STATUS_REEXT:.*]] = zext i1 %[[STATUS_CHECKED]] to i32
+// CHECK-NEXT: store i32 %[[STATUS_REEXT]], ptr %2, align 4
+// CHECK-NEXT: %[[RETVAL:.*]] = load float, ptr %[[VALUE]]
+// CHECK-NEXT: ret float %[[RETVAL]]
+
+
 export uint TestGetDimensions() {
     uint dim1, dim2, stride1, stride2;
     ROSB1.GetDimensions(dim1, stride1);
diff --git a/clang/test/CodeGenHLSL/resources/TypedBuffers-methods.hlsl b/clang/test/CodeGenHLSL/resources/TypedBuffers-methods.hlsl
index fdc1ef08b7c2c..14607b58c67d9 100644
--- a/clang/test/CodeGenHLSL/resources/TypedBuffers-methods.hlsl
+++ b/clang/test/CodeGenHLSL/resources/TypedBuffers-methods.hlsl
@@ -38,6 +38,50 @@ export float TestLoad() {
 // CHECK-NEXT: %[[VEC:.*]] = load <4 x i32>, ptr %[[PTR]]
 // CHECK-NEXT: ret <4 x i32> %[[VEC]]
 
+export float TestLoadWithStatus() {
+    uint s1;
+    uint s2;
+    float ret = Buf.Load(1, s1) + float(RWBuf.Load(2, s2).y);
+    ret += float(s1 + s2);
+    return ret;
+}
+
+// CHECK: define noundef nofpclass(nan inf) float @TestLoadWithStatus()()
+// CHECK: call {{.*}} float @hlsl::Buffer<float>::Load(unsigned int, unsigned int&)(ptr {{.*}} @Buf, i32 noundef 1, ptr noundef nonnull align 4 dereferenceable(4) %s1)
+// CHECK: call {{.*}} <4 x i32> @hlsl::RWBuffer<unsigned int vector[4]>::Load(unsigned int, unsigned int&)(ptr {{.*}} @RWBuf, i32 noundef 2, ptr noundef nonnull align 4 dereferenceable(4) %s2)
+// CHECK: add
+// CHECK: ret float
+
+// CHECK: define {{.*}} float @hlsl::Buffer<float>::Load(unsigned int, unsigned int&)(ptr {{.*}} %this, i32 noundef %Index, ptr noundef nonnull align 4 dereferenceable(4) %Status)
+// CHECK: %__handle = getelementptr inbounds nuw %"class.hlsl::Buffer", ptr %{{.*}}, i32 0, i32 0
+// DXIL-NEXT: %[[HANDLE:.*]] = load target("dx.TypedBuffer", float, 0, 0, 0), ptr %__handle
+// CHECK-NEXT: %[[INDEX:.*]] = load i32, ptr %Index.addr
+// CHECK-NEXT: %[[STATUS:.*]] = load ptr, ptr %Status.addr,
+// DXIL-NEXT: %[[STRUCT:.*]] = call { ptr, i1 } @llvm.dx.resource.load.typedbuffer.p0.tdx.TypedBuffer_f32_0_0_0t(target("dx.TypedBuffer", float, 0, 0, 0) %[[HANDLE]], i32 %[[INDEX]])
+// CHECK-NEXT: %[[VALUE:.*]] = extractvalue { ptr, i1 } %[[STRUCT]], 0
+// CHECK-NEXT: %[[STATUS_TEMP:.*]] = extractvalue { ptr, i1 } %[[STRUCT]], 1
+// CHECK-NEXT: %[[STATUS_EXT:.*]] = zext i1 %[[STATUS_TEMP]] to i32
+// CHECK-NEXT: %[[STATUS_CHECKED:.*]] = call i1 @llvm.dx.check.access.fully.mapped(i32 %[[STATUS_EXT:.*]])
+// CHECK-NEXT: %[[STATUS_REEXT:.*]] = zext i1 %[[STATUS_CHECKED]] to i32
+// CHECK-NEXT: store i32 %[[STATUS_REEXT]], ptr %2, align 4
+// CHECK-NEXT: %[[RETVAL:.*]] = load float, ptr %[[VALUE]]
+// CHECK-NEXT: ret float %[[RETVAL]]
+
+// CHECK: define {{.*}}  <4 x i32> @hlsl::RWBuffer<unsigned int vector[4]>::Load(unsigned int, unsigned int&)(ptr {{.*}} %this, i32 noundef %Index, ptr noundef nonnull align 4 dereferenceable(4) %Status)
+// CHECK: %__handle = getelementptr inbounds nuw %"class.hlsl::RWBuffer", ptr %{{.*}}, i32 0, i32 0
+// DXIL-NEXT: %[[HANDLE:.*]] = load target("dx.TypedBuffer", <4 x i32>, 1, 0, 0), ptr %__handle
+// CHECK-NEXT: %[[INDEX:.*]] = load i32, ptr %Index.addr
+// CHECK-NEXT: %[[STATUS_HANDLE:.*]] = load ptr, ptr %Status.addr, align 4, !nonnull !3, !align !4
+// DXIL-NEXT: %[[STRUCT:.*]] = call { ptr, i1 } @llvm.dx.resource.load.typedbuffer.p0.tdx.TypedBuffer_v4i32_1_0_0t(target("dx.TypedBuffer", <4 x i32>, 1, 0, 0) %0, i32 %1)
+// CHECK-NEXT: %[[VALUE:.*]] = extractvalue { ptr, i1 } %[[STRUCT]], 0
+// CHECK-NEXT: %[[STATUS_TEMP:.*]] = extractvalue { ptr, i1 } %[[STRUCT]], 1
+// CHECK-NEXT: %[[STATUS_EXT:.*]] = zext i1 %[[STATUS_TEMP]] to i32
+// CHECK-NEXT: %[[STATUS_CHECKED:.*]] = call i1 @llvm.dx.check.access.fully.mapped(i32 %[[STATUS_EXT:.*]])
+// CHECK-NEXT: %[[STATUS_REEXT:.*]] = zext i1 %[[STATUS_CHECKED]] to i32
+// CHECK-NEXT: store i32 %[[STATUS_REEXT]], ptr %2, align 4
+// CHECK-NEXT: %[[RETVAL:.*]] = load <4 x i32>, ptr %[[VALUE]]
+// CHECK-NEXT: ret <4 x i32> %[[RETVAL]]
+
 export uint TestGetDimensions() {
     uint dim1, dim2;
     Buf.GetDimensions(dim1);
diff --git a/llvm/include/llvm/IR/IntrinsicsDirectX.td b/llvm/include/llvm/IR/IntrinsicsDirectX.td
index d6b85630eb979..6959978678540 100644
--- a/llvm/include/llvm/IR/IntrinsicsDirectX.td
+++ b/llvm/include/llvm/IR/IntrinsicsDirectX.td
@@ -40,6 +40,9 @@ def int_dx_resource_getpointer
     : DefaultAttrsIntrinsic<[llvm_anyptr_ty], [llvm_any_ty, llvm_i32_ty],
                             [IntrNoMem]>;
 
+def int_dx_check_access_fully_mapped
+    : DefaultAttrsIntrinsic<[llvm_i1_ty], [llvm_i32_ty], [IntrNoMem]>;
+
 def int_...
[truncated]

llvmbot · 2025-11-04T21:36:42Z

@llvm/pr-subscribers-clang-codegen

Author: Joshua Batista (bob80905)

Changes

This PR adds a new overload for resources, which takes an additional parameter by reference, status. It fills the status parameter with a 1 or 0, depending on whether or not the resource access was mapped. CheckAccessFullyMapped is also added as an intrinsic, and called in the production of this status bit.

Patch is 20.86 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/166449.diff

11 Files Affected:

(modified) clang/include/clang/Basic/Builtins.td (+6)
(modified) clang/lib/CodeGen/CGHLSLBuiltins.cpp (+58)
(modified) clang/lib/CodeGen/CGHLSLRuntime.h (+3)
(modified) clang/lib/Sema/HLSLBuiltinTypeDeclBuilder.cpp (+35)
(modified) clang/lib/Sema/HLSLBuiltinTypeDeclBuilder.h (+2)
(modified) clang/lib/Sema/SemaHLSL.cpp (+21)
(modified) clang/test/CodeGenHLSL/resources/StructuredBuffers-methods-lib.hlsl (+44)
(modified) clang/test/CodeGenHLSL/resources/StructuredBuffers-methods-ps.hlsl (+43)
(modified) clang/test/CodeGenHLSL/resources/TypedBuffers-methods.hlsl (+44)
(modified) llvm/include/llvm/IR/IntrinsicsDirectX.td (+3)
(modified) llvm/include/llvm/IR/IntrinsicsSPIRV.td (+3)

diff --git a/clang/include/clang/Basic/Builtins.td b/clang/include/clang/Basic/Builtins.td
index 2b400b012d6ed..f2b4c54a5ba6b 100644
--- a/clang/include/clang/Basic/Builtins.td
+++ b/clang/include/clang/Basic/Builtins.td
@@ -4934,6 +4934,12 @@ def HLSLResourceGetPointer : LangBuiltin<"HLSL_LANG"> {
   let Prototype = "void(...)";
 }
 
+def HLSLResourceLoadWithStatus : LangBuiltin<"HLSL_LANG"> {
+  let Spellings = ["__builtin_hlsl_resource_load_with_status"];
+  let Attributes = [NoThrow];
+  let Prototype = "void(...)";
+}
+
 def HLSLResourceUninitializedHandle : LangBuiltin<"HLSL_LANG"> {
   let Spellings = ["__builtin_hlsl_resource_uninitializedhandle"];
   let Attributes = [NoThrow];
diff --git a/clang/lib/CodeGen/CGHLSLBuiltins.cpp b/clang/lib/CodeGen/CGHLSLBuiltins.cpp
index fbf4a5722caed..ff3d906d4bf11 100644
--- a/clang/lib/CodeGen/CGHLSLBuiltins.cpp
+++ b/clang/lib/CodeGen/CGHLSLBuiltins.cpp
@@ -353,6 +353,64 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
         RetTy, CGM.getHLSLRuntime().getCreateResourceGetPointerIntrinsic(),
         ArrayRef<Value *>{HandleOp, IndexOp});
   }
+  case Builtin::BI__builtin_hlsl_resource_load_with_status: {
+    Value *HandleOp = EmitScalarExpr(E->getArg(0));
+    Value *IndexOp = EmitScalarExpr(E->getArg(1));
+
+    // Get the *address* of the status argument (since it's a reference)
+    LValue StatusLVal = EmitLValue(E->getArg(2));
+    Address StatusAddr = StatusLVal.getAddress();
+
+    QualType HandleTy = E->getArg(0)->getType();
+    const HLSLAttributedResourceType *RT =
+        HandleTy->getAs<HLSLAttributedResourceType>();
+    assert(RT && "Expected a resource type as first parameter");
+
+    Intrinsic::ID IntrID = RT->getAttrs().RawBuffer
+                               ? llvm::Intrinsic::dx_resource_load_rawbuffer
+                               : llvm::Intrinsic::dx_resource_load_typedbuffer;
+
+    llvm::Type *DataTy = ConvertType(E->getType());
+    llvm::Type *RetTy = llvm::StructType::get(Builder.getContext(),
+                                              {DataTy, Builder.getInt1Ty()});
+
+    SmallVector<Value *, 3> Args;
+    Args.push_back(HandleOp);
+    Args.push_back(IndexOp);
+
+    if (RT->getAttrs().RawBuffer) {
+      Args.push_back(Builder.getInt32(0)); // dummy offset
+    }
+
+    // Call the intrinsic (returns a struct)
+    Value *ResRet =
+        Builder.CreateIntrinsic(RetTy, IntrID, Args, {}, "ld.struct");
+
+    // Extract the loaded data (first element of the struct)
+    Value *LoadedValue = Builder.CreateExtractValue(ResRet, {0}, "ld.value");
+
+    // Extract the status bit (second element of the struct)
+    Value *StatusBit = Builder.CreateExtractValue(ResRet, {1}, "ld.status");
+
+    // Extend the 1-bit status to 32-bit, because the check access fully mapped
+    // intrinsic expects i32 input
+    Value *StatusInt =
+        Builder.CreateZExt(StatusBit, Builder.getInt32Ty(), "ld.status.i32");
+
+    // Call CheckAccessFullyMapped(i32)
+    llvm::Function *CheckAccessFn =
+        CGM.getIntrinsic(llvm::Intrinsic::dx_check_access_fully_mapped);
+
+    Value *StatusBool =
+        Builder.CreateCall(CheckAccessFn, {StatusInt}, "ld.status.checked");
+
+    // Extend the boolean to 32-bit uint for storing in user�s var
+    Value *ExtendedStatus =
+        Builder.CreateZExt(StatusBool, Builder.getInt32Ty(), "ld.status.ext");
+
+    Builder.CreateStore(ExtendedStatus, StatusAddr);
+    return LoadedValue;
+  }
   case Builtin::BI__builtin_hlsl_resource_uninitializedhandle: {
     llvm::Type *HandleTy = CGM.getTypes().ConvertType(E->getType());
     return llvm::PoisonValue::get(HandleTy);
diff --git a/clang/lib/CodeGen/CGHLSLRuntime.h b/clang/lib/CodeGen/CGHLSLRuntime.h
index d35df524fdc84..f6c654bd61ff5 100644
--- a/clang/lib/CodeGen/CGHLSLRuntime.h
+++ b/clang/lib/CodeGen/CGHLSLRuntime.h
@@ -126,6 +126,9 @@ class CGHLSLRuntime {
 
   GENERATE_HLSL_INTRINSIC_FUNCTION(CreateResourceGetPointer,
                                    resource_getpointer)
+
+  GENERATE_HLSL_INTRINSIC_FUNCTION(CreateResourceLoadTypedBuffer,
+                                   resource_load_typedbuffer)
   GENERATE_HLSL_INTRINSIC_FUNCTION(CreateHandleFromBinding,
                                    resource_handlefrombinding)
   GENERATE_HLSL_INTRINSIC_FUNCTION(CreateHandleFromImplicitBinding,
diff --git a/clang/lib/Sema/HLSLBuiltinTypeDeclBuilder.cpp b/clang/lib/Sema/HLSLBuiltinTypeDeclBuilder.cpp
index 066acf6f01a90..bdbc2d4ce87b3 100644
--- a/clang/lib/Sema/HLSLBuiltinTypeDeclBuilder.cpp
+++ b/clang/lib/Sema/HLSLBuiltinTypeDeclBuilder.cpp
@@ -1140,6 +1140,7 @@ BuiltinTypeDeclBuilder &BuiltinTypeDeclBuilder::addLoadMethods() {
   DeclarationName Load(&II);
   // TODO: We also need versions with status for CheckAccessFullyMapped.
   addHandleAccessFunction(Load, /*IsConst=*/false, /*IsRef=*/false);
+  addLoadWithStatusFunction(Load, /*IsConst=*/false, /*IsRef=*/false);
 
   return *this;
 }
@@ -1232,6 +1233,40 @@ BuiltinTypeDeclBuilder &BuiltinTypeDeclBuilder::addDecrementCounterMethod() {
       .finalize();
 }
 
+BuiltinTypeDeclBuilder &
+BuiltinTypeDeclBuilder::addLoadWithStatusFunction(DeclarationName &Name,
+                                                  bool IsConst, bool IsRef) {
+  assert(!Record->isCompleteDefinition() && "record is already complete");
+  ASTContext &AST = SemaRef.getASTContext();
+  using PH = BuiltinTypeMethodBuilder::PlaceHolder;
+
+  QualType ElemTy = getHandleElementType();
+  QualType AddrSpaceElemTy =
+      AST.getAddrSpaceQualType(ElemTy, LangAS::hlsl_device);
+  QualType ElemPtrTy = AST.getPointerType(AddrSpaceElemTy);
+  QualType ReturnTy;
+
+  if (IsRef) {
+    ReturnTy = AddrSpaceElemTy;
+    if (IsConst)
+      ReturnTy.addConst();
+    ReturnTy = AST.getLValueReferenceType(ReturnTy);
+  } else {
+    ReturnTy = ElemTy;
+    if (IsConst)
+      ReturnTy.addConst();
+  }
+
+  QualType StatusRefTy = AST.getLValueReferenceType(AST.UnsignedIntTy);
+  return BuiltinTypeMethodBuilder(*this, Name, ReturnTy, IsConst)
+      .addParam("Index", AST.UnsignedIntTy)
+      .addParam("Status", StatusRefTy)
+      .callBuiltin("__builtin_hlsl_resource_load_with_status", ElemPtrTy,
+                   PH::Handle, PH::_0, PH::_1)
+      .dereference(PH::LastStmt)
+      .finalize();
+}
+
 BuiltinTypeDeclBuilder &
 BuiltinTypeDeclBuilder::addHandleAccessFunction(DeclarationName &Name,
                                                 bool IsConst, bool IsRef) {
diff --git a/clang/lib/Sema/HLSLBuiltinTypeDeclBuilder.h b/clang/lib/Sema/HLSLBuiltinTypeDeclBuilder.h
index 95e3a6c4fb2f1..4941b8b7952a2 100644
--- a/clang/lib/Sema/HLSLBuiltinTypeDeclBuilder.h
+++ b/clang/lib/Sema/HLSLBuiltinTypeDeclBuilder.h
@@ -91,6 +91,8 @@ class BuiltinTypeDeclBuilder {
   BuiltinTypeDeclBuilder &addDecrementCounterMethod();
   BuiltinTypeDeclBuilder &addHandleAccessFunction(DeclarationName &Name,
                                                   bool IsConst, bool IsRef);
+  BuiltinTypeDeclBuilder &addLoadWithStatusFunction(DeclarationName &Name,
+                                                    bool IsConst, bool IsRef);
   BuiltinTypeDeclBuilder &addAppendMethod();
   BuiltinTypeDeclBuilder &addConsumeMethod();
 
diff --git a/clang/lib/Sema/SemaHLSL.cpp b/clang/lib/Sema/SemaHLSL.cpp
index 94a490a8f68dc..f95a647b5679f 100644
--- a/clang/lib/Sema/SemaHLSL.cpp
+++ b/clang/lib/Sema/SemaHLSL.cpp
@@ -3010,6 +3010,27 @@ bool SemaHLSL::CheckBuiltinFunctionCall(unsigned BuiltinID, CallExpr *TheCall) {
 
     break;
   }
+  case Builtin::BI__builtin_hlsl_resource_load_with_status: {
+    if (SemaRef.checkArgCount(TheCall, 3) ||
+        CheckResourceHandle(&SemaRef, TheCall, 0) ||
+        CheckArgTypeMatches(&SemaRef, TheCall->getArg(1),
+                            SemaRef.getASTContext().UnsignedIntTy) ||
+        CheckArgTypeMatches(&SemaRef, TheCall->getArg(2),
+                            SemaRef.getASTContext().UnsignedIntTy))
+      return true;
+
+    auto *ResourceTy =
+        TheCall->getArg(0)->getType()->castAs<HLSLAttributedResourceType>();
+    QualType ContainedTy = ResourceTy->getContainedType();
+    auto ReturnType =
+        SemaRef.Context.getAddrSpaceQualType(ContainedTy, LangAS::hlsl_device);
+    ReturnType = SemaRef.Context.getPointerType(ReturnType);
+    TheCall->setType(ReturnType);
+    TheCall->setValueKind(VK_LValue);
+
+    break;
+  }
+
   case Builtin::BI__builtin_hlsl_resource_uninitializedhandle: {
     if (SemaRef.checkArgCount(TheCall, 1) ||
         CheckResourceHandle(&SemaRef, TheCall, 0))
diff --git a/clang/test/CodeGenHLSL/resources/StructuredBuffers-methods-lib.hlsl b/clang/test/CodeGenHLSL/resources/StructuredBuffers-methods-lib.hlsl
index 1f248d0560006..e692a14ecc168 100644
--- a/clang/test/CodeGenHLSL/resources/StructuredBuffers-methods-lib.hlsl
+++ b/clang/test/CodeGenHLSL/resources/StructuredBuffers-methods-lib.hlsl
@@ -104,6 +104,50 @@ export float TestLoad() {
 // CHECK-NEXT: %[[VAL:.*]] = load float, ptr %[[PTR]]
 // CHECK-NEXT: ret float %[[VAL]]
 
+export float TestLoadWithStatus() {
+    uint s1;
+    uint s2;
+    float ret = RWSB1.Load(1, s1) + SB1.Load(2, s2);
+    ret += float(s1 + s2);
+    return ret;
+}
+
+// CHECK: define noundef nofpclass(nan inf) float @TestLoadWithStatus()()
+// CHECK: call {{.*}} float @hlsl::RWStructuredBuffer<float>::Load(unsigned int, unsigned int&)(ptr {{.*}} @RWSB1, i32 noundef 1, ptr noundef nonnull align 4 dereferenceable(4) %s1)
+// CHECK: call {{.*}} float @hlsl::StructuredBuffer<float>::Load(unsigned int, unsigned int&)(ptr {{.*}} @SB1, i32 noundef 2, ptr noundef nonnull align 4 dereferenceable(4) %s2)
+// CHECK: add
+// CHECK: ret float
+
+// CHECK: define {{.*}} float @hlsl::RWStructuredBuffer<float>::Load(unsigned int, unsigned int&)(ptr {{.*}} %this, i32 noundef %Index, ptr noundef nonnull align 4 dereferenceable(4) %Status)
+// CHECK: %__handle = getelementptr inbounds nuw %"class.hlsl::RWStructuredBuffer", ptr %{{.*}}, i32 0, i32 0
+// DXIL-NEXT: %[[HANDLE:.*]] = load target("dx.RawBuffer", float, 1, 0), ptr %__handle
+// CHECK-NEXT: %[[INDEX:.*]] = load i32, ptr %Index.addr
+// CHECK-NEXT: %[[STATUS:.*]] = load ptr, ptr %Status.addr,
+// DXIL-NEXT: %[[STRUCT:.*]] = call { ptr, i1 } @llvm.dx.resource.load.rawbuffer.p0.tdx.RawBuffer_f32_1_0t(target("dx.RawBuffer", float, 1, 0) %[[HANDLE]], i32 %[[INDEX]], i32 0)
+// CHECK-NEXT: %[[VALUE:.*]] = extractvalue { ptr, i1 } %[[STRUCT]], 0
+// CHECK-NEXT: %[[STATUS_TEMP:.*]] = extractvalue { ptr, i1 } %[[STRUCT]], 1
+// CHECK-NEXT: %[[STATUS_EXT:.*]] = zext i1 %[[STATUS_TEMP]] to i32
+// CHECK-NEXT: %[[STATUS_CHECKED:.*]] = call i1 @llvm.dx.check.access.fully.mapped(i32 %[[STATUS_EXT:.*]])
+// CHECK-NEXT: %[[STATUS_REEXT:.*]] = zext i1 %[[STATUS_CHECKED]] to i32
+// CHECK-NEXT: store i32 %[[STATUS_REEXT]], ptr %2, align 4
+// CHECK-NEXT: %[[RETVAL:.*]] = load float, ptr %[[VALUE]]
+// CHECK-NEXT: ret float %[[RETVAL]]
+
+// CHECK: define {{.*}} float @hlsl::StructuredBuffer<float>::Load(unsigned int, unsigned int&)(ptr {{.*}} %this, i32 noundef %Index, ptr noundef nonnull align 4 dereferenceable(4) %Status)
+// CHECK: %__handle = getelementptr inbounds nuw %"class.hlsl::StructuredBuffer", ptr %{{.*}}, i32 0, i32 0
+// DXIL-NEXT: %[[HANDLE:.*]] = load target("dx.RawBuffer", float, 0, 0), ptr %__handle
+// CHECK-NEXT: %[[INDEX:.*]] = load i32, ptr %Index.addr
+// CHECK-NEXT: %[[STATUS_HANDLE:.*]] = load ptr, ptr %Status.addr, align 4, !nonnull !3, !align !4
+// DXIL-NEXT: %[[STRUCT:.*]] = call { ptr, i1 } @llvm.dx.resource.load.rawbuffer.p0.tdx.RawBuffer_f32_0_0t(target("dx.RawBuffer", float, 0, 0) %0, i32 %1, i32 0)
+// CHECK-NEXT: %[[VALUE:.*]] = extractvalue { ptr, i1 } %[[STRUCT]], 0
+// CHECK-NEXT: %[[STATUS_TEMP:.*]] = extractvalue { ptr, i1 } %[[STRUCT]], 1
+// CHECK-NEXT: %[[STATUS_EXT:.*]] = zext i1 %[[STATUS_TEMP]] to i32
+// CHECK-NEXT: %[[STATUS_CHECKED:.*]] = call i1 @llvm.dx.check.access.fully.mapped(i32 %[[STATUS_EXT:.*]])
+// CHECK-NEXT: %[[STATUS_REEXT:.*]] = zext i1 %[[STATUS_CHECKED]] to i32
+// CHECK-NEXT: store i32 %[[STATUS_REEXT]], ptr %2, align 4
+// CHECK-NEXT: %[[RETVAL:.*]] = load float, ptr %[[VALUE]]
+// CHECK-NEXT: ret float %[[RETVAL]]
+
 export uint TestGetDimensions() {
     uint dim1, dim2, dim3, stride1, stride2, stride3;
     SB1.GetDimensions(dim1, stride1);
diff --git a/clang/test/CodeGenHLSL/resources/StructuredBuffers-methods-ps.hlsl b/clang/test/CodeGenHLSL/resources/StructuredBuffers-methods-ps.hlsl
index 25fa75965d686..77aaf281fbc39 100644
--- a/clang/test/CodeGenHLSL/resources/StructuredBuffers-methods-ps.hlsl
+++ b/clang/test/CodeGenHLSL/resources/StructuredBuffers-methods-ps.hlsl
@@ -65,6 +65,49 @@ export float TestLoad() {
 // CHECK-NEXT: %[[VAL:.*]] = load <2 x i32>, ptr %[[BUFPTR]]
 // CHECK-NEXT: ret <2 x i32> %[[VAL]]
 
+export float TestLoadWithStatus() {
+    uint status;
+    uint status2;
+    float val = ROSB1.Load(10, status).x + ROSB2.Load(20, status2).x;
+    return val + float(status + status2);
+}
+
+// CHECK: define {{.*}} float @TestLoadWithStatus()()
+// CHECK: call {{.*}} float @hlsl::RasterizerOrderedStructuredBuffer<float>::Load(unsigned int, unsigned int&)(ptr {{.*}} @ROSB1, i32 noundef 10, ptr noundef nonnull align 4 dereferenceable(4) %status)
+// CHECK: call {{.*}} <2 x i32> @hlsl::RasterizerOrderedStructuredBuffer<int vector[2]>::Load(unsigned int, unsigned int&)(ptr {{.*}} @ROSB2, i32 noundef 20, ptr noundef nonnull align 4 dereferenceable(4) %status2)
+// CHECK: ret
+
+// CHECK: define {{.*}} float @hlsl::RasterizerOrderedStructuredBuffer<float>::Load(unsigned int, unsigned int&)(ptr {{.*}} %Index, ptr noundef nonnull align 4 dereferenceable(4) %Status)
+// CHECK: %__handle = getelementptr inbounds nuw %"class.hlsl::RasterizerOrderedStructuredBuffer", ptr {{.*}}, i32 0, i32 0
+// CHECK-NEXT: %[[HANDLE:.*]] = load target("dx.RawBuffer", float, 1, 1), ptr %__handle
+// CHECK-NEXT: %[[INDEX:.*]] = load i32, ptr %Index.addr
+// CHECK-NEXT: %[[STATUS:.*]] = load ptr, ptr %Status.addr,
+// DXIL-NEXT: %[[STRUCT:.*]] = call { ptr, i1 } @llvm.dx.resource.load.rawbuffer.p0.tdx.RawBuffer_f32_1_1t(target("dx.RawBuffer", float, 1, 1) %[[HANDLE]], i32 %[[INDEX]], i32 0)
+// CHECK-NEXT: %[[VALUE:.*]] = extractvalue { ptr, i1 } %[[STRUCT]], 0
+// CHECK-NEXT: %[[STATUS_TEMP:.*]] = extractvalue { ptr, i1 } %[[STRUCT]], 1
+// CHECK-NEXT: %[[STATUS_EXT:.*]] = zext i1 %[[STATUS_TEMP]] to i32
+// CHECK-NEXT: %[[STATUS_CHECKED:.*]] = call i1 @llvm.dx.check.access.fully.mapped(i32 %[[STATUS_EXT:.*]])
+// CHECK-NEXT: %[[STATUS_REEXT:.*]] = zext i1 %[[STATUS_CHECKED]] to i32
+// CHECK-NEXT: store i32 %[[STATUS_REEXT]], ptr %2, align 4
+// CHECK-NEXT: %[[RETVAL:.*]] = load float, ptr %[[VALUE]]
+// CHECK-NEXT: ret float %[[RETVAL]]
+
+// CHECK: define {{.*}} <2 x i32> @hlsl::RasterizerOrderedStructuredBuffer<int vector[2]>::Load(unsigned int, unsigned int&)(ptr {{.*}} %Index, ptr noundef nonnull align 4 dereferenceable(4) %Status)
+// CHECK: %__handle = getelementptr inbounds nuw %"class.hlsl::RasterizerOrderedStructuredBuffer.0", ptr {{.*}}, i32 0, i32 0
+// CHECK-NEXT: %[[HANDLE:.*]] = load target("dx.RawBuffer", <2 x i32>, 1, 1), ptr %__handle
+// CHECK-NEXT: %[[INDEX:.*]] = load i32, ptr %Index.addr
+// CHECK-NEXT: %[[STATUS_HANDLE:.*]] = load ptr, ptr %Status.addr, align 4, !nonnull !3, !align !4
+// DXIL-NEXT: %[[STRUCT:.*]] = call { ptr, i1 } @llvm.dx.resource.load.rawbuffer.p0.tdx.RawBuffer_v2i32_1_1t(target("dx.RawBuffer", <2 x i32>, 1, 1) %0, i32 %1, i32 0)
+// CHECK-NEXT: %[[VALUE:.*]] = extractvalue { ptr, i1 } %[[STRUCT]], 0
+// CHECK-NEXT: %[[STATUS_TEMP:.*]] = extractvalue { ptr, i1 } %[[STRUCT]], 1
+// CHECK-NEXT: %[[STATUS_EXT:.*]] = zext i1 %[[STATUS_TEMP]] to i32
+// CHECK-NEXT: %[[STATUS_CHECKED:.*]] = call i1 @llvm.dx.check.access.fully.mapped(i32 %[[STATUS_EXT:.*]])
+// CHECK-NEXT: %[[STATUS_REEXT:.*]] = zext i1 %[[STATUS_CHECKED]] to i32
+// CHECK-NEXT: store i32 %[[STATUS_REEXT]], ptr %2, align 4
+// CHECK-NEXT: %[[RETVAL:.*]] = load float, ptr %[[VALUE]]
+// CHECK-NEXT: ret float %[[RETVAL]]
+
+
 export uint TestGetDimensions() {
     uint dim1, dim2, stride1, stride2;
     ROSB1.GetDimensions(dim1, stride1);
diff --git a/clang/test/CodeGenHLSL/resources/TypedBuffers-methods.hlsl b/clang/test/CodeGenHLSL/resources/TypedBuffers-methods.hlsl
index fdc1ef08b7c2c..14607b58c67d9 100644
--- a/clang/test/CodeGenHLSL/resources/TypedBuffers-methods.hlsl
+++ b/clang/test/CodeGenHLSL/resources/TypedBuffers-methods.hlsl
@@ -38,6 +38,50 @@ export float TestLoad() {
 // CHECK-NEXT: %[[VEC:.*]] = load <4 x i32>, ptr %[[PTR]]
 // CHECK-NEXT: ret <4 x i32> %[[VEC]]
 
+export float TestLoadWithStatus() {
+    uint s1;
+    uint s2;
+    float ret = Buf.Load(1, s1) + float(RWBuf.Load(2, s2).y);
+    ret += float(s1 + s2);
+    return ret;
+}
+
+// CHECK: define noundef nofpclass(nan inf) float @TestLoadWithStatus()()
+// CHECK: call {{.*}} float @hlsl::Buffer<float>::Load(unsigned int, unsigned int&)(ptr {{.*}} @Buf, i32 noundef 1, ptr noundef nonnull align 4 dereferenceable(4) %s1)
+// CHECK: call {{.*}} <4 x i32> @hlsl::RWBuffer<unsigned int vector[4]>::Load(unsigned int, unsigned int&)(ptr {{.*}} @RWBuf, i32 noundef 2, ptr noundef nonnull align 4 dereferenceable(4) %s2)
+// CHECK: add
+// CHECK: ret float
+
+// CHECK: define {{.*}} float @hlsl::Buffer<float>::Load(unsigned int, unsigned int&)(ptr {{.*}} %this, i32 noundef %Index, ptr noundef nonnull align 4 dereferenceable(4) %Status)
+// CHECK: %__handle = getelementptr inbounds nuw %"class.hlsl::Buffer", ptr %{{.*}}, i32 0, i32 0
+// DXIL-NEXT: %[[HANDLE:.*]] = load target("dx.TypedBuffer", float, 0, 0, 0), ptr %__handle
+// CHECK-NEXT: %[[INDEX:.*]] = load i32, ptr %Index.addr
+// CHECK-NEXT: %[[STATUS:.*]] = load ptr, ptr %Status.addr,
+// DXIL-NEXT: %[[STRUCT:.*]] = call { ptr, i1 } @llvm.dx.resource.load.typedbuffer.p0.tdx.TypedBuffer_f32_0_0_0t(target("dx.TypedBuffer", float, 0, 0, 0) %[[HANDLE]], i32 %[[INDEX]])
+// CHECK-NEXT: %[[VALUE:.*]] = extractvalue { ptr, i1 } %[[STRUCT]], 0
+// CHECK-NEXT: %[[STATUS_TEMP:.*]] = extractvalue { ptr, i1 } %[[STRUCT]], 1
+// CHECK-NEXT: %[[STATUS_EXT:.*]] = zext i1 %[[STATUS_TEMP]] to i32
+// CHECK-NEXT: %[[STATUS_CHECKED:.*]] = call i1 @llvm.dx.check.access.fully.mapped(i32 %[[STATUS_EXT:.*]])
+// CHECK-NEXT: %[[STATUS_REEXT:.*]] = zext i1 %[[STATUS_CHECKED]] to i32
+// CHECK-NEXT: store i32 %[[STATUS_REEXT]], ptr %2, align 4
+// CHECK-NEXT: %[[RETVAL:.*]] = load float, ptr %[[VALUE]]
+// CHECK-NEXT: ret float %[[RETVAL]]
+
+// CHECK: define {{.*}}  <4 x i32> @hlsl::RWBuffer<unsigned int vector[4]>::Load(unsigned int, unsigned int&)(ptr {{.*}} %this, i32 noundef %Index, ptr noundef nonnull align 4 dereferenceable(4) %Status)
+// CHECK: %__handle = getelementptr inbounds nuw %"class.hlsl::RWBuffer", ptr %{{.*}}, i32 0, i32 0
+// DXIL-NEXT: %[[HANDLE:.*]] = load target("dx.TypedBuffer", <4 x i32>, 1, 0, 0), ptr %__handle
+// CHECK-NEXT: %[[INDEX:.*]] = load i32, ptr %Index.addr
+// CHECK-NEXT: %[[STATUS_HANDLE:.*]] = load ptr, ptr %Status.addr, align 4, !nonnull !3, !align !4
+// DXIL-NEXT: %[[STRUCT:.*]] = call { ptr, i1 } @llvm.dx.resource.load.typedbuffer.p0.tdx.TypedBuffer_v4i32_1_0_0t(target("dx.TypedBuffer", <4 x i32>, 1, 0, 0) %0, i32 %1)
+// CHECK-NEXT: %[[VALUE:.*]] = extractvalue { ptr, i1 } %[[STRUCT]], 0
+// CHECK-NEXT: %[[STATUS_TEMP:.*]] = extractvalue { ptr, i1 } %[[STRUCT]], 1
+// CHECK-NEXT: %[[STATUS_EXT:.*]] = zext i1 %[[STATUS_TEMP]] to i32
+// CHECK-NEXT: %[[STATUS_CHECKED:.*]] = call i1 @llvm.dx.check.access.fully.mapped(i32 %[[STATUS_EXT:.*]])
+// CHECK-NEXT: %[[STATUS_REEXT:.*]] = zext i1 %[[STATUS_CHECKED]] to i32
+// CHECK-NEXT: store i32 %[[STATUS_REEXT]], ptr %2, align 4
+// CHECK-NEXT: %[[RETVAL:.*]] = load <4 x i32>, ptr %[[VALUE]]
+// CHECK-NEXT: ret <4 x i32> %[[RETVAL]]
+
 export uint TestGetDimensions() {
     uint dim1, dim2;
     Buf.GetDimensions(dim1);
diff --git a/llvm/include/llvm/IR/IntrinsicsDirectX.td b/llvm/include/llvm/IR/IntrinsicsDirectX.td
index d6b85630eb979..6959978678540 100644
--- a/llvm/include/llvm/IR/IntrinsicsDirectX.td
+++ b/llvm/include/llvm/IR/IntrinsicsDirectX.td
@@ -40,6 +40,9 @@ def int_dx_resource_getpointer
     : DefaultAttrsIntrinsic<[llvm_anyptr_ty], [llvm_any_ty, llvm_i32_ty],
                             [IntrNoMem]>;
 
+def int_dx_check_access_fully_mapped
+    : DefaultAttrsIntrinsic<[llvm_i1_ty], [llvm_i32_ty], [IntrNoMem]>;
+
 def int_...
[truncated]

s-perron · 2025-11-07T15:59:44Z

llvm/include/llvm/IR/IntrinsicsSPIRV.td

+  def int_spv_resource_load_with_status
+      : DefaultAttrsIntrinsic<[llvm_anyptr_ty], [llvm_any_ty, llvm_i32_ty, llvm_i32_ty],
+                              [IntrNoMem]>;


I opened #166954 to track the implementing this in the SPIR-V backend.

bob80905 added 2 commits October 30, 2025 17:06

as far as I can go without adding CAFM

99c9788

works without cafm

06ad11a

llvmbot added clang Clang issues not falling into any other category clang:frontend Language frontend issues, e.g. anything involving "Sema" clang:codegen IR generation bugs: mangling, exceptions, etc. backend:DirectX HLSL HLSL Language Support llvm:ir labels Nov 4, 2025

furthest checkpoint yet

ec4e3a9

bob80905 force-pushed the add_load_overload_with_status branch from 3b0f1fa to ec4e3a9 Compare November 6, 2025 20:50

llvmbot added backend:X86 clang:headers Headers provided by Clang, e.g. for intrinsics labels Nov 6, 2025

s-perron mentioned this pull request Nov 7, 2025

[HLSL][SPIRV] Implement int_spv_resource_load_with_status in the SPIR-V backend. #166954

Open

s-perron reviewed Nov 7, 2025

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[HLSL] Add load overload with status #166449

[HLSL] Add load overload with status #166449

bob80905 commented Nov 4, 2025

Uh oh!

llvmbot commented Nov 4, 2025 •

edited

Loading

Uh oh!

llvmbot commented Nov 4, 2025

Uh oh!

s-perron Nov 7, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

[HLSL] Add load overload with status #166449

Are you sure you want to change the base?

[HLSL] Add load overload with status #166449

Conversation

bob80905 commented Nov 4, 2025

Uh oh!

llvmbot commented Nov 4, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

llvmbot commented Nov 4, 2025

Uh oh!

s-perron Nov 7, 2025

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

llvmbot commented Nov 4, 2025 •

edited

Loading