From: Artem Belevich <tra@google.com>
Date: Mon, 10 Aug 2015 20:57:02 +0000 (+0000)
Subject: [CUDA] Add implicit __attribute__((used)) to all __global__ functions.
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=830e9302b923713e8b98382dbea094a1c9a50db9;p=clang

[CUDA] Add implicit __attribute__((used)) to all __global__ functions.

This allows emitting kernels that were instantiated from the host code
and which would never be explicitly referenced otherwise.

Differential Revision: http://reviews.llvm.org/D11666

git-svn-id: https://llvm.org/svn/llvm-project/cfe/trunk@244501 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/Sema/SemaDeclAttr.cpp b/lib/Sema/SemaDeclAttr.cpp
index 31ddd503af..8e04b69f81 100644
--- a/lib/Sema/SemaDeclAttr.cpp
+++ b/lib/Sema/SemaDeclAttr.cpp
@@ -3350,6 +3350,10 @@ static void handleGlobalAttr(Sema &S, Decl *D, const AttributeList &Attr) {
   D->addAttr(::new (S.Context)
               CUDAGlobalAttr(Attr.getRange(), S.Context,
                              Attr.getAttributeSpellingListIndex()));
+
+  // Add implicit attribute((used)) so we don't eliminate kernels
+  // because there is nothing referencing them on device side.
+  D->addAttr(UsedAttr::CreateImplicit(S.Context));
 }
 
 static void handleGNUInlineAttr(Sema &S, Decl *D, const AttributeList &Attr) {
diff --git a/test/CodeGenCUDA/ptx-kernels.cu b/test/CodeGenCUDA/ptx-kernels.cu
index 658b3488fc..bf3b14d3a2 100644
--- a/test/CodeGenCUDA/ptx-kernels.cu
+++ b/test/CodeGenCUDA/ptx-kernels.cu
@@ -1,7 +1,16 @@
+// Make sure that __global__ functions are emitted along with correct
+// annotations and are added to @llvm.used to prevent their elimination.
+// REQUIRES: nvptx-registered-target
+//
 // RUN: %clang_cc1 %s -triple nvptx-unknown-unknown -fcuda-is-device -emit-llvm -o - | FileCheck %s
 
 #include "Inputs/cuda.h"
 
+// Make sure that all __global__ functiona are added to @llvm.used
+// CHECK: @llvm.used = appending global
+// CHECK-SAME: @global_function
+// CHECK-SAME: @_Z16templated_kernelIiEvT_
+
 // CHECK-LABEL: define void @device_function
 extern "C"
 __device__ void device_function() {}
@@ -13,4 +22,10 @@ __global__ void global_function() {
   device_function();
 }
 
+// Make sure host-instantiated kernels are preserved on device side.
+template <typename T> __global__ void templated_kernel(T param) {}
+// CHECK-LABEL: define linkonce_odr void @_Z16templated_kernelIiEvT_
+void host_function() { templated_kernel<<<0,0>>>(0); }
+
 // CHECK: !{{[0-9]+}} = !{void ()* @global_function, !"kernel", i32 1}
+// CHECK: !{{[0-9]+}} = !{void (i32)* @_Z16templated_kernelIiEvT_, !"kernel", i32 1}