AMDGPU: Add option to stress calls

[arsenm]

Summary

This inverts the behavior of the AlwaysInline pass to mark
every function not already marked alwaysinline as noinline.

[rampitec]

LGTM

[arsenm]

r313865

lib/Target/AMDGPU/AMDGPUAlwaysInlinePass.cpp

 #include "AMDGPU.h"
 #include "llvm/IR/Module.h"
 #include "llvm/Transforms/Utils/Cloning.h"
 
 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;
 
 public:
   static char ID;
 
   AMDGPUAlwaysInline(bool GlobalOpt = false) :
     ModulePass(ID), GlobalOpt(GlobalOpt) { }
bool AMDGPUAlwaysInline::runOnModule(Module &M) {
   }
 
   if (GlobalOpt) {
     for (GlobalAlias* A : AliasesToRemove) {
       A->eraseFromParent();
     }
   }
 
+  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);
   }
 
   for (Function *F : FuncsToClone) {
     ValueToValueMapTy VMap;
     Function *NewFunc = CloneFunction(F, VMap);
     NewFunc->setLinkage(GlobalValue::InternalLinkage);
     F->replaceAllUsesWith(NewFunc);
   }
 
   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;
 }
 
 ModulePass *llvm::createAMDGPUAlwaysInlinePass(bool GlobalOpt) {
   return new AMDGPUAlwaysInline(GlobalOpt);
 }

test/CodeGen/AMDGPU/stress-calls.ll

+; 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 }