[PATCH] D49526: Updated llvm-proto-fuzzer to execute the compiled code

[Emmett Neyman via Phabricator via llvm-commits]

emmettneyman added inline comments.


================
Comment at: clang/tools/clang-fuzzer/handle-llvm/handle_llvm.cpp:89
+  Builder.Inliner = createFunctionInliningPass(OptLevel, SizeLevel, false);
+  Builder.LoopVectorize = true;
+  Builder.populateFunctionPassManager(FPM);
----------------
morehouse wrote:
> If we do this, do we need to explicitly add the vectorizer pass below?
No we don't. I've deleted the line below.


================
Comment at: clang/tools/clang-fuzzer/handle-llvm/handle_llvm.cpp:110
+  TargetLibraryInfoImpl TLII(ModuleTriple);
+  Passes.add(new TargetLibraryInfoWrapperPass(TLII));
+  Passes.add(createTargetTransformInfoWrapperPass(TargetIRAnalysis()));
----------------
morehouse wrote:
> Can simplify to `Passes.add(new TargetLibraryInfoWrapperPass(M->getTargetTriple))`.
Contrary to the function's name, `getTargetTriple()` actually returns a `string`, not a `Triple`. But I changed it to `new TargetLibraryInfoWrapperPass(ModuleTriple)` and deleted line 109.


================
Comment at: clang/tools/clang-fuzzer/handle-llvm/handle_llvm.cpp:115
+      make_unique<legacy::FunctionPassManager>(M.get());
+  FPasses->add(createTargetTransformInfoWrapperPass(TargetIRAnalysis()));
+
----------------
morehouse wrote:
> morehouse wrote:
> > Why do we add a `TargetTransformInfoWrapperPass` to both `Passes` and `FPasses`?
> Do we need both `Passes` and `FPasses`?
I think because we're just adding a loop vectorize pass, we don't need `FPasses`. The loop vectorize pass lives within the regular `ModulePassManager`, not the `FunctionPassManager`. I decided to put it in the code so, in the future, it would be easier to add different kinds of passes to the fuzz target. Should I remove it?


================
Comment at: clang/tools/clang-fuzzer/handle-llvm/handle_llvm.cpp:159
+      std::unique_ptr<RTDyldMemoryManager>(RTDyldMM));
+  builder.setOptLevel(OLvl);
+  builder.setTargetOptions(InitTargetOptionsFromCodeGenFlags());
----------------
morehouse wrote:
> If the JIT does optimization already, do we need to call `OptLLVM` at all?  Will the vectorizer kick in without `OptLLVM`?
I'll look into this more. I couldn't find an answer quickly.


Repository:
  rC Clang

https://reviews.llvm.org/D49526