[SYCL][ClangLinkerWrapper] Fix race condition

[maksimsab]

This patch fixes a data race in the following line of the code that is supposed to be run parallelly and needs mutex protection.

WrappedOutput.push_back(*SYCLOutputOrErr);

[asudarsa]

Hi @maksimsab

Can you please provide a very short description? Also, do we need a test to support this? Thanks

[maksimsab]

In the listing below the lambda is being launched parallelly in threads.

llvm/clang/tools/clang-linker-wrapper/ClangLinkerWrapper.cpp

Lines 1568 to 1569 in 5048853

	SmallVector<StringRef> WrappedOutput;
	auto Err = parallelForEachError(InputsForTarget, [&](auto &Input) -> Error {

Mutual modifying operations to Images and WrappedOutput should be performed under mutex protection as it is already done for Images here:

llvm/clang/tools/clang-linker-wrapper/ClangLinkerWrapper.cpp

Lines 1649 to 1660 in 5048853

	std::scoped_lock<decltype(ImageMtx)> Guard(ImageMtx);
	OffloadingImage TheImage{};
	TheImage.TheImageKind =
	Args.hasArg(OPT_embed_bitcode) ? IMG_Bitcode : IMG_Object;
	TheImage.TheOffloadKind = Kind;
	TheImage.StringData["triple"] =
	Args.MakeArgString(LinkerArgs.getLastArgValue(OPT_triple_EQ));
	TheImage.StringData["arch"] =
	Args.MakeArgString(LinkerArgs.getLastArgValue(OPT_arch_EQ));
	TheImage.Image = std::move(*FileOrErr);
	Images[Kind].emplace_back(std::move(TheImage));

We haven't caught this bug because we run our tests without --wrapped-jobs option and the default number of jobs is 1. The long-term plan to add such test in the end.

[asudarsa]

Sorry. Missed this one. Looks ok to me. Might want to make sure this is aligned with #12499