================
@@ -1641,6 +1642,24 @@
PassBuilder::buildModuleOptimizationPipeline(OptimizationLevel Level,
if (!LTOPreLink)
MPM.addPass(RelLookupTableConverterPass());
+ if (PTO.DevirtualizeSpeculatively && LTOPhase == ThinOrFullLTOPhase::None) {
+ MPM.addPass(WholeProgramDevirtPass(
+ /*ExportSummary*/ nullptr,
+ /*ImportSummary*/ nullptr,
+ /*DevirtSpeculatively*/ PTO.DevirtualizeSpeculatively));
+ MPM.addPass(LowerTypeTestsPass(nullptr, nullptr,
+ lowertypetests::DropTestKind::Assume));
+ if (EnableModuleInliner) {
----------------
hassnaaHamdi wrote:
Okay, I see that integrating the WPD pass into another pipeline is non-trivial
and would require changes across multiple places. Additionally, I'm uncertain
if this approach will succeed given that WPD is a module pass.
Maybe (maybe) it could be better to have a different pass for non-to
speculative devirtualization isolated from all the LTO stuff and in that case
it's easier to add any needed changes into the pass without touching LTO
things.
But anyway, I'll need to investigate the suggested approach further before
implementing.
Does it make sense to keep the current opt-in speculative devirtualization
changes for now, and revisit the pipeline integration as a follow-up?
Of course, I'm still open to pursuing the correct approach if the current
implementation is not good enough.
https://github.com/llvm/llvm-project/pull/159685
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
