diff options
| author | Amir Ayupov <aaupov@fb.com> | 2022-10-19 18:23:43 -0700 |
|---|---|---|
| committer | Amir Ayupov <amir.aupov@gmail.com> | 2022-10-19 18:24:05 -0700 |
| commit | 076240fa062415b6470b79413559aff2bf5bf208 (patch) | |
| tree | 8036c6a7d91314df13cce853335b1b303758898d /llvm/lib/Support/X86TargetParser.cpp | |
| parent | 647ca2ee700b00e27218103b5fcfd2cb3956f6f1 (diff) | |
| download | llvm-076240fa062415b6470b79413559aff2bf5bf208.zip llvm-076240fa062415b6470b79413559aff2bf5bf208.tar.gz llvm-076240fa062415b6470b79413559aff2bf5bf208.tar.bz2 | |
[CMake] Disable BOLT instrumentation of Clang on instrumented build
This enables multi-stage PGO build optimized by BOLT using BOLT.cmake cache.
The issue is that `-DPGO_BUILD_CONFIGURATION` cache file is passed to both
stage2-instrumented and stage2-optimized builds (for them to be identical),
but in case of BOLT.cmake, it doesn't make sense to BOLT-instrument the
instrumented binary (it's not going to be optimized). Hence turn off
`CLANG_BOLT_INSTRUMENT` code if `LLVM_BUILD_INSTRUMENTED` is enabled.
The final workflow that enables multi-stage InstrPGO+ThinLTO+BOLT Clang build:
```
cmake <llvm-project>/llvm -GNinja -DLLVM_ENABLE_LLD=ON \
-DBOOTSTRAP_LLVM_ENABLE_LLD=ON -DBOOTSTRAP_BOOTSTRAP_LLVM_ENABLE_LLD=ON \
-DPGO_INSTRUMENT_LTO=Thin -C llvm-project/clang/cmake/caches/BOLT-PGO.cmake
ninja stage2-clang++-bolt
```
Reviewed By: rafauler
Differential Revision: https://reviews.llvm.org/D136023
Diffstat (limited to 'llvm/lib/Support/X86TargetParser.cpp')
0 files changed, 0 insertions, 0 deletions