diff options
| author | Yingying Wang <3171290993@qq.com> | 2025-11-04 13:25:49 +0800 |
|---|---|---|
| committer | GitHub <noreply@github.com> | 2025-11-04 13:25:49 +0800 |
| commit | 0ba7bfc34fa292a34b31f6338464744b5cad58a3 (patch) | |
| tree | 726e33f56fc156c8736828b086e3d6013e27656d /clang/lib/CodeGen/CodeGenModule.cpp | |
| parent | 93e860e694770f52a9eeecda88ba11173c291ef8 (diff) | |
| download | llvm-0ba7bfc34fa292a34b31f6338464744b5cad58a3.zip llvm-0ba7bfc34fa292a34b31f6338464744b5cad58a3.tar.gz llvm-0ba7bfc34fa292a34b31f6338464744b5cad58a3.tar.bz2 | |
[DFAJumpThreading] Enable DFAJumpThread by default. (#157646)
We recommend setting `dfa-jump-thread` to be enabled by default. It’s a
mature optimization that’s been supported since GCC 9.1.0. At the `-O2`
opt level, both the GCC and ICX compilers have this optimization enabled
by default.
Once it’s enabled, we saw a **13% performance improvement** in the
CoreMark benchmark on the X86 platform (Intel i9-11900K Rocket Lake),
and even a **15% increase** on the KunMingHu FPGA. Additionally, we
verified the correctness of this pass using SPEC 2017.
Co-authored-by: YinZd <104915588+nothiny@users.noreply.github.com>
Co-authored-by: ict-ql <168183727+ict-ql@users.noreply.github.com>
Co-authored-by: Lin Wang <wanglulin@ict.ac.cn>
Diffstat (limited to 'clang/lib/CodeGen/CodeGenModule.cpp')
0 files changed, 0 insertions, 0 deletions