[CIR][CUDA] Support for inbuilt texture types

clang/lib/CIR/CodeGen/CIRGenModule.cpp

@@ -569,14 +569,11 @@ bool CIRGenModule::shouldEmitCUDAGlobalVar(const VarDecl *global) const {
   // size and host-side address in order to provide access to
   // their device-side incarnations.
 
-  if (global->getType()->isCUDADeviceBuiltinTextureType()) {
-    llvm_unreachable("NYI");
-  }
-
   return !langOpts.CUDAIsDevice || global->hasAttr<CUDADeviceAttr>() ||
          global->hasAttr<CUDAConstantAttr>() ||
          global->hasAttr<CUDASharedAttr>() ||
-         global->getType()->isCUDADeviceBuiltinSurfaceType();
+         global->getType()->isCUDADeviceBuiltinSurfaceType() ||
+         global->getType()->isCUDADeviceBuiltinTextureType();
 }
 
 void CIRGenModule::emitGlobal(GlobalDecl gd) {
@@ -1212,7 +1209,6 @@ CIRGenModule::getOrCreateCIRGlobal(StringRef mangledName, mlir::Type ty,
   // initializer.
   if (gv.isDeclaration()) {
     getTargetCIRGenInfo().setTargetAttributes(d, gv, *this);
-
     // External HIP managed variables needed to be recorded for transformation
     // in both device and host compilations.
     if (getLangOpts().CUDA && d && d->hasAttr<HIPManagedAttr>() &&
@@ -1493,7 +1489,8 @@ void CIRGenModule::emitGlobalVarDefinition(const clang::VarDecl *d,
     // because they must not be initialized.
     if (linkage != cir::GlobalLinkageKind::InternalLinkage &&
         (d->hasAttr<CUDADeviceAttr>() || d->hasAttr<CUDAConstantAttr>() ||
-         d->getType()->isCUDADeviceBuiltinSurfaceType())) {
+         d->getType()->isCUDADeviceBuiltinSurfaceType() ||
+         d->getType()->isCUDADeviceBuiltinTextureType())) {
       gv->setAttr(CUDAExternallyInitializedAttr::getMnemonic(),
                   CUDAExternallyInitializedAttr::get(&getMLIRContext()));
     }

clang/lib/CIR/CodeGen/CIRGenTypes.cpp

@@ -357,7 +357,11 @@ mlir::Type CIRGenTypes::convertType(QualType T) {
       if (mlir::Type Ty =
               CGM.getTargetCIRGenInfo().getCUDADeviceBuiltinSurfaceDeviceType())
         return Ty;
+      llvm_unreachable("NYI");
     } else if (T->isCUDADeviceBuiltinTextureType()) {
+      if (mlir::Type Ty =
+              CGM.getTargetCIRGenInfo().getCUDADeviceBuiltinTextureDeviceType())
+        return Ty;
       llvm_unreachable("NYI");
     }
   }

clang/lib/CIR/CodeGen/TargetInfo.cpp

@@ -345,13 +345,16 @@ class NVPTXTargetCIRGenInfo : public TargetCIRGenInfo {
 public:
   NVPTXTargetCIRGenInfo(CIRGenTypes &cgt)
       : TargetCIRGenInfo(std::make_unique<NVPTXABIInfo>(cgt)) {}
-
   mlir::Type getCUDADeviceBuiltinSurfaceDeviceType() const override {
+    // On the device side, surface reference is represented as an object handle
+    // in 64-bit integer.
+    return cir::IntType::get(&getABIInfo().CGT.getMLIRContext(), 64, true);
+  }
+  mlir::Type getCUDADeviceBuiltinTextureDeviceType() const override {
     // On the device side, texture reference is represented as an object handle
     // in 64-bit integer.
     return cir::IntType::get(&getABIInfo().CGT.getMLIRContext(), 64, true);
   }
-
   void setTargetAttributes(const clang::Decl *decl, mlir::Operation *global,
                            CIRGenModule &cgm) const override {
     if (const auto *vd = clang::dyn_cast_or_null<clang::VarDecl>(decl)) {

clang/lib/CIR/CodeGen/TargetInfo.h

@@ -132,6 +132,9 @@ class TargetCIRGenInfo {
   virtual mlir::Type getCUDADeviceBuiltinSurfaceDeviceType() const {
     return nullptr;
   }
+  virtual mlir::Type getCUDADeviceBuiltinTextureDeviceType() const {
+    return nullptr;
+  }
   virtual ~TargetCIRGenInfo() {}
 };
 

clang/test/CIR/CodeGen/CUDA/texture.cu

@@ -0,0 +1,24 @@
+// REQUIRES: x86-registered-target
+// REQUIRES: nvptx-registered-target
+
+// RUN: %clang_cc1 -fclangir -std=c++11 -fcuda-is-device -triple nvptx64-nvidia-cuda -emit-llvm -o - %s | FileCheck --check-prefix=DEVICE-LLVM %s
+// RUN: %clang_cc1 -fclangir -std=c++11 -fcuda-is-device -triple nvptx64-nvidia-cuda -emit-cir -o - %s | FileCheck --check-prefix=DEVICE-CIR %s
+// RUN: echo "GPU binary would be here" > %t
+
+struct textureReference {
+  int desc;
+};
+
+enum ReadMode {
+  ElementType = 0,
+  NormalizedFloat = 1
+};
+
+template <typename T, int dim = 1, enum ReadMode mode = ElementType>
+struct __attribute__((device_builtin_texture_type)) texture : public textureReference {
+};
+
+texture<float, 2, NormalizedFloat> tex;
+
+// DEVICE-LLVM: @tex = addrspace(1) externally_initialized global i64 undef, align 4
+// DEVICE-CIR: cir.global external addrspace(offload_global) @tex = #cir.undef : !s64i {alignment = 4 : i64, cu.externally_initialized = #cir.cu.externally_initialized}