This patch adds the offload kind to the embedded section name in
preparation for offloading to different kinda like CUDA or HIP.
Depends on D120288
Paths
| Differential D120271
[Clang] Add offload kind to embedded offload object ClosedPublic Authored by jhuber6 on Feb 21 2022, 11:45 AM.
Details Summary This patch adds the offload kind to the embedded section name in Depends on D120288
Diff Detail
Event Timelinejhuber6 added a child revision: D120272: [CUDA] Add driver support for compiling CUDA with the new driver.Feb 21 2022, 11:46 AM Comment Actions The failing test needs to be adjusted, right?
Comment Actions
Yes, forgot about that.
Comment Actions Will it be possible to use Bundle Entry ID format [1] for naming sections and being sort of the the primary key to refer an offload object? [1] https://clang.llvm.org/docs/ClangOffloadBundler.html#bundle-entry-id Comment Actions
It might also be useful for PTX, considering when we do LTO we need to know the target features (e.g. +ptx74). As long as it's a valid section name anything goes, especially as things get more complex like when we may want to embed both PTX and cubins, or maybe always embed bitcode for option JIT use. This change definitely brings us closer to what we had in the old offload bundler so we could unify that somewhat. That could be the topic of a future patch. Comment Actions Adjust commit message. LG, we can address outstanding comments for different encodings as we go. This revision is now accepted and ready to land.Mar 14 2022, 4:28 PM This revision was landed with ongoing or failed builds.Mar 14 2022, 5:09 PM Closed by commit rG9f89769cd775: [Clang] Add offload kind to embedded offload object (authored by jhuber6). · Explain Why This revision was automatically updated to reflect the committed changes.
Revision Contents
Diff 415278 clang/lib/Driver/ToolChains/Clang.cpp
clang/test/Driver/openmp-offload-gpu.c
clang/tools/clang-linker-wrapper/ClangLinkerWrapper.cpp
|
Why not filename in the hash and equal method?
Can we split this change off into a separate patch please.