[PATCH] D102463: Making Instrumentation aware of LoopNest Pass
Arthur Eubanks via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Tue May 18 09:00:03 PDT 2021
aeubanks added inline comments.
================
Comment at: llvm/include/llvm/Transforms/Scalar/LoopPassManager.h:363-365
+ const Loop *L = any_isa<const LoopNest *>(&IR)
+ ? &any_cast<const LoopNest *>(&IR)->getOutermostLoop()
+ : any_cast<const Loop *>(&IR);
----------------
raghesh wrote:
> aeubanks wrote:
> > `IR` is either a `Loop&` or a `LoopNest&` right? not an `Any`
> > so we need to extract a `Loop` out of either `Loop` or `LoopNest`. with templates, it should be as simple as creating two functions with the same name that both return a `Loop&`, where one takes `Loop&` and the other takes a `LoopNest&`
> Sorry!! I dont really get you. Are you suggesting something similar to the following
>
> @@ -183,6 +183,9 @@ protected:
> PreservedAnalyses runWithoutLoopNestPasses(Loop &L, LoopAnalysisManager &AM,
> LoopStandardAnalysisResults &AR,
> LPMUpdater &U);
> +private:
> + const Loop& getLoopFromIR(Loop &L) { return L; }
> + const Loop& getLoopFromIR(LoopNest &LN) { return LN.getOutermostLoop(); }
> };
>
> /// The Loop pass manager.
> @@ -360,12 +363,10 @@ Optional<PreservedAnalyses> LoopPassManager::runSinglePass(
> LoopStandardAnalysisResults &AR, LPMUpdater &U, PassInstrumentation &PI) {
> // Pass the outermost loop to BeforePass and AfterPass callbacks in case of
> // LoopNest Pass.
> + const Loop &L = getLoopFromIR(IR);
> // Check the PassInstrumentation's BeforePass callbacks before running the
> // pass, skip its execution completely if asked to (callback returns false).
> - if (!PI.runBeforePass<Loop>(*Pass, *L))
> + if (!PI.runBeforePass<Loop>(*Pass, L))
> return None;
>
> PreservedAnalyses PA;
> @@ -378,7 +379,7 @@ Optional<PreservedAnalyses> LoopPassManager::runSinglePass(
> if (U.skipCurrentLoop())
> PI.runAfterPassInvalidated<IRUnitT>(*Pass, PA);
> else
> - PI.runAfterPass<Loop>(*Pass, *L, PA);
> + PI.runAfterPass<Loop>(*Pass, L, PA);
> return PA;
> }
>
yup that's right
I don't think the current patch does anything because there's no `Any` in this code (a test case would be nice)
Repository:
rG LLVM Github Monorepo
CHANGES SINCE LAST ACTION
https://reviews.llvm.org/D102463/new/
https://reviews.llvm.org/D102463
More information about the llvm-commits
mailing list