2 files changed, 49 insertions, 3 deletions
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUAlwaysInlinePass.cpp b/llvm/lib/Target/AMDGPU/AMDGPUAlwaysInlinePass.cpp
index 6f3742ed039..c27425443ab 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUAlwaysInlinePass.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUAlwaysInlinePass.cpp
@@ -21,6 +21,12 @@ using namespace llvm;
 
 namespace {
 
+static cl::opt<bool> StressCalls(
+  "amdgpu-stress-function-calls",
+  cl::Hidden,
+  cl::desc("Force all functions to be noinline"),
+  cl::init(false));
+
 class AMDGPUAlwaysInline : public ModulePass {
   bool GlobalOpt;
 
@@ -57,9 +63,13 @@ bool AMDGPUAlwaysInline::runOnModule(Module &M) {
     }
   }
 
+  auto NewAttr = StressCalls ? Attribute::NoInline : Attribute::AlwaysInline;
+  auto IncompatAttr
+    = StressCalls ? Attribute::AlwaysInline : Attribute::NoInline;
+
   for (Function &F : M) {
     if (!F.hasLocalLinkage() && !F.isDeclaration() && !F.use_empty() &&
-        !F.hasFnAttribute(Attribute::NoInline))
+        !F.hasFnAttribute(IncompatAttr))
       FuncsToClone.push_back(&F);
   }
 
@@ -71,8 +81,8 @@ bool AMDGPUAlwaysInline::runOnModule(Module &M) {
   }
 
   for (Function &F : M) {
-    if (F.hasLocalLinkage() && !F.hasFnAttribute(Attribute::NoInline)) {
-      F.addFnAttr(Attribute::AlwaysInline);
+    if (F.hasLocalLinkage() && !F.hasFnAttribute(IncompatAttr)) {
+      F.addFnAttr(NewAttr);
     }
   }
   return false;
diff --git a/llvm/test/CodeGen/AMDGPU/stress-calls.ll b/llvm/test/CodeGen/AMDGPU/stress-calls.ll
new file mode 100644
index 00000000000..480d40d67d4
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/stress-calls.ll
@@ -0,0 +1,36 @@
+; RUN: opt -S -amdgpu-stress-function-calls -amdgpu-always-inline %s | FileCheck %s
+
+; CHECK: define internal fastcc i32 @alwaysinline_func(i32 %a) #0 {
+define internal fastcc i32 @alwaysinline_func(i32 %a) alwaysinline {
+entry:
+  %tmp0 = add i32 %a, 1
+  ret i32 %tmp0
+}
+
+; CHECK: define internal fastcc i32 @noinline_func(i32 %a) #1 {
+define internal fastcc i32 @noinline_func(i32 %a) noinline {
+entry:
+  %tmp0 = add i32 %a, 2
+  ret i32 %tmp0
+}
+
+; CHECK: define internal fastcc i32 @unmarked_func(i32 %a) #1 {
+define internal fastcc i32 @unmarked_func(i32 %a) {
+entry:
+  %tmp0 = add i32 %a, 3
+  ret i32 %tmp0
+}
+
+define amdgpu_kernel void @kernel(i32 addrspace(1)* %out) {
+entry:
+  %tmp0 = call i32 @alwaysinline_func(i32 1)
+  store volatile i32 %tmp0, i32 addrspace(1)* %out
+  %tmp1 = call i32 @noinline_func(i32 1)
+  store volatile i32 %tmp1, i32 addrspace(1)* %out
+  %tmp2 = call i32 @unmarked_func(i32 1)
+  store volatile i32 %tmp2, i32 addrspace(1)* %out
+  ret void
+}
+
+; CHECK: attributes #0 = { alwaysinline }
+; CHECK: attributes #1 = { noinline }