[Polly] [PPCGCodeGeneration] [NFC] [WIP] Add option to fail if a store to a scalar occursin a kernel.

[bollu]

Summary

• Having such a store prevents us from running the kernel since the final value of

the scalar on running the kernel would be undefined.

• Useful to have a flag that fails if this condition occurs.

[bollu]

• [NFC] style fixes

[singam-sanjay]

lib/CodeGen/PPCGCodeGeneration.cpp

2075

Would having if (FailOnStoredScalar) behind this if be more appropriate ? The stored scalar is an issue only if this condition evaluates to true, right ?

[grosser]

lib/CodeGen/PPCGCodeGeneration.cpp

2075

True. Sanjay is right here.

[bollu]

lib/CodeGen/PPCGCodeGeneration.cpp

2075

Yep, you're right, that does make more sense. I'll change it out.

[grosser]

Otherwise, LGTM.

lib/CodeGen/PPCGCodeGeneration.cpp

static cl::opt<bool> PrivateMemory("polly-acc-use-private",
                                    cl::cat(PollyCategory));
 
 static cl::opt<bool> ManagedMemory("polly-acc-codegen-managed-memory",
                                    cl::desc("Generate Host kernel code assuming"
                                             " that all memory has been"
                                             " declared as managed memory"),
                                    cl::Hidden, cl::init(false), cl::ZeroOrMore,
                                    cl::cat(PollyCategory));
+static cl::opt<bool>
+    FailOnStoredScalar("polly-acc-fail-on-stored-scalar",
+                       cl::desc("Fail and generate a backtrace if"
+                                " we have scalar stores in a kernel that "
+                                " prevents us from parallelising "),
+                       cl::Hidden, cl::init(false), cl::ZeroOrMore,
+                       cl::cat(PollyCategory));
 
 static cl::opt<bool>
     FailOnVerifyModuleFailure("polly-acc-fail-on-verify-module-failure",
                               cl::desc("Fail and generate a backtrace if"
                                        " verifyModule fails on the GPU "
                                        " kernel module."),
                               cl::Hidden, cl::init(false), cl::ZeroOrMore,
                               cl::cat(PollyCategory));
for (long i = 0; i < Kernel->n_array; i++) {
 
     Value *Alloca = BlockGen.getOrCreateAlloca(SAI);
     Value *ArgPtr = &*Arg;
     Type *TypePtr = SAI->getElementType()->getPointerTo();
     Value *TypedArgPtr = Builder.CreatePointerCast(ArgPtr, TypePtr);
     Value *Val = Builder.CreateLoad(Alloca);
     Builder.CreateStore(Val, TypedArgPtr);
     StoredScalar = true;
+    if (FailOnStoredScalar)
+      report_fatal_error("Failed because of stored Scalar in Scop: " +
+                         S.getFunction().getName() +
+                         " | Region: " + S.getName());
 
     Arg++;
   }
 
   if (StoredScalar)
     /// In case more than one thread contains scalar stores, the generated
     /// code might be incorrect, if we only store at the end of the kernel.
     /// To support this case we need to store these scalars back at each
     /// memory store or at least before each kernel barrier.
     if (Kernel->n_block != 0 || Kernel->n_grid != 0)
       BuildSuccessful = 0;
 }
 
 void GPUNodeBuilder::createKernelVariables(ppcg_kernel *Kernel, Function *FN) {
   Module *M = Builder.GetInsertBlock()->getParent()->getParent();
 
   for (int i = 0; i < Kernel->n_var; ++i) {
     struct ppcg_kernel_var &Var = Kernel->var[i];