[Diffusion] rGc93f93b2e3f2: Revert "Revert "Recommit "Revert "[CVP] processSwitch: Remove default case when…
Alexander Kornienko via Phabricator via cfe-commits
cfe-commits at lists.llvm.org
Mon Nov 22 09:24:35 PST 2021
alexfh added a comment.
In rGc93f93b2e3f28997f794265089fb8138dd5b5f13#1039669 <https://reviews.llvm.org/rGc93f93b2e3f28997f794265089fb8138dd5b5f13#1039669>, @bgraur wrote:
> Early heads up that this revision causes a large regression in compilation time for some of our internal source files: we are seeing an almost 20x increase in compilation times for some files (from 42s to 728s).
>
> I'm working on a reproducer which I'm going to upload when ready.
We're still reducing it, but the test case is a different one as the one I posted in https://reviews.llvm.org/D106056#3023708. Similar to https://reviews.llvm.org/D106056, adding `-mllvm -disable-early-taildup=true` reduces the compilation time to normal.
`perf record --call-graph lbr clang -xc++ -std=gnu++17 -c -O3 test.ii` results in the profile below:
- 99.28% 0.00% clang clang [.] llvm::FPPassManager::runOnFunction ◆
- 99.27% llvm::FPPassManager::runOnFunction ▒
- 99.24% llvm::MachineFunctionPass::runOnFunction ▒
- 97.51% (anonymous namespace)::MachineBlockPlacement::runOnMachineFunction ▒
- 97.51% (anonymous namespace)::MachineBlockPlacement::buildCFGChains ▒
- 97.45% (anonymous namespace)::MachineBlockPlacement::buildChain ▒
- 96.82% (anonymous namespace)::MachineBlockPlacement::canTailDuplicateUnplacedPreds ▒
94.66% hasSameSuccessors ▒
Hope this helps finding out the cause of the problem.
It would be helpful, if you could revert the commit while you're investigating.
BRANCHES
fix_asan, main
Users:
junparser (Author)
https://reviews.llvm.org/rGc93f93b2e3f2
More information about the cfe-commits
mailing list