[Driver][OpenMP] Update actions builder to create unbundling action when necessary.

[sfantao]

Summary

Each time that offloading support is requested by the user and the input file is not a source file, an action OffloadUnbundlingAction is created to signal that the input file may contain bundles, so that the proper tool is then invoked to attempt to extract the components of the bundle. This patch adds the logic to create that action in offload action builder.

The job creation for the new action will be proposed in a separate patch.

[ABataev]

include/clang/Driver/Action.h

498

'final'?

504

1. 'final'?
2. Default initializers for fields

include/clang/Driver/Types.h

86

'const'?

[sfantao]

• Mark classes 'final' and set defalt initializers.

[sfantao]

Hi Alexey,

Thanks for the review!

include/clang/Driver/Types.h

86

isSrcFile is not member of a class. It is an externally visible function of the namespace types, therefore can't mark it const.

[sfantao]

• Add Ctor dropped by mistake in the previous diff.

[sfantao]

• Rebase

[sfantao]

• Rebase.

[sfantao]

• Rebase.

[ABataev]

LG

[hfinkel]

include/clang/Driver/Action.h

504

Don't need 'Ty' in the name of this struct.

include/clang/Driver/Types.h

84

decided the first -> decides if the first

85

preprocessor one -> preprocessing one

lib/Driver/Driver.cpp

2103

This checks that the file needs to be preprocessed. What does preprocessing have to do with this? I don't imagine that providing a preprocessed source file as input should invoke the unbundler .

test/Driver/openmp-offload.c

309

Oh, are you using .i to indicate a bundle instead of a preprocessed file? Don't do that. Please use a different suffix -- the bundler has its own file format.

[hfinkel]

lib/Driver/Driver.cpp

2103

On second thought, this is okay. It does not make sense to have a non-bundled preprocessed source for the input there, as the host and device compilation don't share a common preprocessor state.

We do need to be careful, perhaps, about .s files (which don't need preprocessing as .S files do) -- we should probably assume that all non-bundled .s files are host assembly code.

test/Driver/openmp-offload.c

309

Never mind; this is okay too.

[sfantao]

Hi Hal,

Thanks for the review! Comments inlined.

include/clang/Driver/Action.h

504

Ok, using DependentActionInfo now.

lib/Driver/Driver.cpp

2103

Yes, that is what we do. If the bundler tool detects that the input is not a bundle, it assumes it is host code/bits. In either case, we still generate the unbundling tool as the driver doesn't check the contents of the files.

test/Driver/openmp-offload.c

309

Ok, there is no particular suffix to indicate a file is a bundle. The (un)bundler, however, has the machinery to detect if a given file is a bundle, it just uses the extension to understand if it is a human readable file, bitcode file, or object file, because the bundle format is different in those three cases.

[sfantao]

• Fix typos and use StringRef() instead of const char * to follow what the Driver does today when it comes to specify the bound architectures.

[hfinkel]

LGTM

lib/Driver/Driver.cpp

2103

Please add to the comment above that if the unbundler tool detects that the input is not a bundle, it assumes it is host input.

[sfantao]

• Add comment explaing that the bundler tool can detect if the input file is a bundle or not.