Reduce phase has been parallelized and a execution time was reduced by 60% with this.
The reading of bitcode (bitcode -> Info) was moved to this segment of code parallelized so it now happens just before reducing.
Details
Details
Diff Detail
Diff Detail
Event Timeline
clang-tools-extra/clang-doc/tool/ClangDocMain.cpp | ||
---|---|---|
72 | Can we use the pre-existing concurrency flag instead (https://github.com/llvm/llvm-project/blob/91e5cdfc93729c61c757db4efd4a82670ac7f929/clang/lib/Tooling/AllTUsExecution.cpp#L150)? It seems counterintuitive to have to set two different concurrency flags for one tool. You'll have to put up a separate patch to expose that flag in the AllTUs header, so that you can access it (see rL344335, as an example). |
Comment Actions
LGTM (after comment nit addressed)
clang-tools-extra/clang-doc/tool/ClangDocMain.cpp | ||
---|---|---|
235 | nit: comment about where this var is coming from |
Can we use the pre-existing concurrency flag instead (https://github.com/llvm/llvm-project/blob/91e5cdfc93729c61c757db4efd4a82670ac7f929/clang/lib/Tooling/AllTUsExecution.cpp#L150)? It seems counterintuitive to have to set two different concurrency flags for one tool.
You'll have to put up a separate patch to expose that flag in the AllTUs header, so that you can access it (see rL344335, as an example).