[CUDA][HIP] Make template implicitly host device (#70369)

Added option -foffload-implicit-host-device-templates which is off by
default.

When the option is on, template functions and specializations without
host/device attributes have implicit host device attributes.

They can be overridden by device template functions with the same
signagure.
They are emitted on device side only if they are used on device side.

This feature is added as an extension.
`__has_extension(cuda_implicit_host_device_templates)` can be used to
check whether it is enabled.

This is to facilitate using standard C++ headers for device.

Fixes: https://github.com/llvm/llvm-project/issues/69956

Fixes: SWDEV-428314

1 files changed, 41 insertions, 1 deletions

diff --git a/clang/lib/Sema/SemaCUDA.cpp b/clang/lib/Sema/SemaCUDA.cpp
index d993499..318174f 100644
--- a/clang/lib/Sema/SemaCUDA.cpp
+++ b/clang/lib/Sema/SemaCUDA.cpp
@@ -678,6 +678,27 @@ void Sema::checkAllowedCUDAInitializer(VarDecl *VD) {
   }
 }
 
+void Sema::CUDARecordImplicitHostDeviceFuncUsedByDevice(
+    const FunctionDecl *Callee) {
+  FunctionDecl *Caller = getCurFunctionDecl(/*AllowLambda=*/true);
+  if (!Caller)
+    return;
+
+  if (!isCUDAImplicitHostDeviceFunction(Callee))
+    return;
+
+  CUDAFunctionTarget CallerTarget = IdentifyCUDATarget(Caller);
+
+  // Record whether an implicit host device function is used on device side.
+  if (CallerTarget != CFT_Device && CallerTarget != CFT_Global &&
+      (CallerTarget != CFT_HostDevice ||
+       (isCUDAImplicitHostDeviceFunction(Caller) &&
+        !getASTContext().CUDAImplicitHostDeviceFunUsedByDevice.count(Caller))))
+    return;
+
+  getASTContext().CUDAImplicitHostDeviceFunUsedByDevice.insert(Callee);
+}
+
 // With -fcuda-host-device-constexpr, an unattributed constexpr function is
 // treated as implicitly __host__ __device__, unless:
 //  * it is a variadic function (device-side variadic functions are not
@@ -702,6 +723,18 @@ void Sema::maybeAddCUDAHostDeviceAttrs(FunctionDecl *NewD,
     return;
   }
 
+  // If a template function has no host/device/global attributes,
+  // make it implicitly host device function.
+  if (getLangOpts().OffloadImplicitHostDeviceTemplates &&
+      !NewD->hasAttr<CUDAHostAttr>() && !NewD->hasAttr<CUDADeviceAttr>() &&
+      !NewD->hasAttr<CUDAGlobalAttr>() &&
+      (NewD->getDescribedFunctionTemplate() ||
+       NewD->isFunctionTemplateSpecialization())) {
+    NewD->addAttr(CUDAHostAttr::CreateImplicit(Context));
+    NewD->addAttr(CUDADeviceAttr::CreateImplicit(Context));
+    return;
+  }
+
   if (!getLangOpts().CUDAHostDeviceConstexpr || !NewD->isConstexpr() ||
       NewD->isVariadic() || NewD->hasAttr<CUDAHostAttr>() ||
       NewD->hasAttr<CUDADeviceAttr>() || NewD->hasAttr<CUDAGlobalAttr>())
@@ -950,7 +983,14 @@ void Sema::checkCUDATargetOverload(FunctionDecl *NewFD,
     // HD/global functions "exist" in some sense on both the host and device, so
     // should have the same implementation on both sides.
     if (NewTarget != OldTarget &&
-        ((NewTarget == CFT_HostDevice) || (OldTarget == CFT_HostDevice) ||
+        ((NewTarget == CFT_HostDevice &&
+          !(LangOpts.OffloadImplicitHostDeviceTemplates &&
+            isCUDAImplicitHostDeviceFunction(NewFD) &&
+            OldTarget == CFT_Device)) ||
+         (OldTarget == CFT_HostDevice &&
+          !(LangOpts.OffloadImplicitHostDeviceTemplates &&
+            isCUDAImplicitHostDeviceFunction(OldFD) &&
+            NewTarget == CFT_Device)) ||
          (NewTarget == CFT_Global) || (OldTarget == CFT_Global)) &&
         !IsOverload(NewFD, OldFD, /* UseMemberUsingDeclRules = */ false,
                     /* ConsiderCudaAttrs = */ false)) {