address Justin

Author: bob80905

clang/lib/CodeGen/CGHLSLBuiltins.cpp

@@ -365,6 +365,8 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
     const HLSLAttributedResourceType *RT =
         HandleTy->getAs<HLSLAttributedResourceType>();
     assert(RT && "Expected a resource type as first parameter");
+    assert(CGM.getTarget().getTriple().getArch() == llvm::Triple::dxil &&
+           "Only DXIL currently implements load with status");
 
     Intrinsic::ID IntrID = RT->getAttrs().RawBuffer
                                ? llvm::Intrinsic::dx_resource_load_rawbuffer
@@ -379,24 +381,25 @@ Value *CodeGenFunction::EmitHLSLBuiltinExpr(unsigned BuiltinID,
     Args.push_back(IndexOp);
 
     if (RT->getAttrs().RawBuffer) {
-      Args.push_back(Builder.getInt32(0)); // dummy offset
+      Value *Offset = Builder.getInt32(0);
+      Args.push_back(Offset);
     }
 
-    // Call the intrinsic (returns a struct)
+    // Call the intrinsic (returns a struct),
+    // Extract the loaded value and status bit (elements within the struct)
+    // Extend the status bit to a 32-bit integer
+    // Store the extended status into the user's reference variable
+    // Return the loaded value
     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 status bit to a 32-bit integer
     Value *ExtendedStatus =
         Builder.CreateZExt(StatusBit, Builder.getInt32Ty(), "ld.status.ext");
 
-    // Store the extended status into the user's reference variable
     Builder.CreateStore(ExtendedStatus, StatusAddr);
 
     return LoadedValue;

llvm/include/llvm/IR/IntrinsicsDirectX.td

@@ -40,10 +40,6 @@ def int_dx_resource_getpointer
     : DefaultAttrsIntrinsic<[llvm_anyptr_ty], [llvm_any_ty, llvm_i32_ty],
                             [IntrNoMem]>;
 
-def int_dx_resource_load_with_status
-    : DefaultAttrsIntrinsic<[llvm_anyptr_ty], [llvm_any_ty, llvm_i32_ty, llvm_i32_ty],
-                            [IntrNoMem]>;
-
 def int_dx_resource_nonuniformindex
     : DefaultAttrsIntrinsic<[llvm_i32_ty], [llvm_i32_ty], [IntrNoMem]>;
 

llvm/lib/Target/DirectX/DXILOpLowering.cpp

@@ -657,14 +657,6 @@ class OpLowerer {
     return false;
   }
 
-  [[nodiscard]] bool lowerLoadWithStatus(Function &F) {
-    // These should have already been handled in DXILResourceAccess, so we can
-    // just clean up the dead prototype.
-    assert(F.user_empty() && "getpointer operations should have been removed");
-    F.eraseFromParent();
-    return false;
-  }
-
   [[nodiscard]] bool lowerBufferStore(Function &F, bool IsRaw) {
     const DataLayout &DL = F.getDataLayout();
     IRBuilder<> &IRB = OpBuilder.getIRB();
@@ -941,9 +933,6 @@ class OpLowerer {
       case Intrinsic::dx_resource_getpointer:
         HasErrors |= lowerGetPointer(F);
         break;
-      case Intrinsic::dx_resource_load_with_status:
-        HasErrors |= lowerLoadWithStatus(F);
-        break;
       case Intrinsic::dx_resource_nonuniformindex:
         assert(!CleanupNURI &&
                "overloaded llvm.dx.resource.nonuniformindex intrinsics?");

llvm/lib/Target/DirectX/DXILShaderFlags.cpp

@@ -105,16 +105,17 @@ static bool checkWaveOps(Intrinsic::ID IID) {
 // This is our proof that the module requires TiledResources
 // to be set, as if check access fully mapped was used.
 bool checkIfStatusIsExtracted(const Instruction &I) {
-  // Iterate over all uses of the instruction
-  for (const Use &U : I.uses()) {
-    const User *UserInst = U.getUser();
-
-    // Check if the user is an ExtractValue instruction
-    if (const ExtractValueInst *EVI = dyn_cast<ExtractValueInst>(UserInst)) {
-      // ExtractValueInst has a list of indices; check if it extracts index 1
-      if (EVI->getNumIndices() == 1 && EVI->getIndices()[0] == 1) {
+  auto *II = dyn_cast<IntrinsicInst>(&I);
+  assert(II);
+  auto IID = II->getIntrinsicID();
+  assert(IID == Intrinsic::dx_resource_load_typedbuffer ||
+         IID == Intrinsic::dx_resource_load_rawbuffer);
+  for (const User *U : I.users()) {
+    if (const ExtractValueInst *EVI = dyn_cast<ExtractValueInst>(U)) {
+      // Resource load operations return a {result, status} pair
+      // check if we extract the status
+      if (EVI->getNumIndices() == 1 && EVI->getIndices()[0] == 1)
         return true;
-      }
     }
   }