[NewPM] Introduce MFAnalysisGetter for a common analysis getter #116166
Conversation
|
Warning This pull request is not mergeable via GitHub because a downstack PR is open. Once all requirements are satisfied, merge this PR as a stack on Graphite.
This stack of pull requests is managed by Graphite. Learn more about stacking. |
optimisan
force-pushed
the
users/Akshat-Oke/11-14-_newpm_introduce_mfanalysisgetter_for_a_common_analysis_getter
branch
from
197b28c to
7e31c7d
Compare
| MFAnalysisGetter(Pass *LegacyPass) : LegacyPass(LegacyPass) {} | ||
| MFAnalysisGetter(MachineFunctionAnalysisManager *MFAM) : MFAM(MFAM) {} | ||
|
|
||
| /// Outer analyses requested from NPM will be cached results and can be null |
| @@ -236,6 +238,82 @@ using MachineFunctionPassManager = PassManager<MachineFunction>; | |||
| /// preserve. | |||
| PreservedAnalyses getMachineFunctionPassPreservedAnalyses(); | |||
|
|
|||
| /// For migrating to new pass manager | |||
There was a problem hiding this comment.
Full stop at the end of the comment.
| template <typename AnalysisT> | ||
| typename AnalysisT::Result *getAnalysis(MachineFunction &MF) { | ||
| if (MFAM) { | ||
| // need a proxy to get the result for outer analyses |
There was a problem hiding this comment.
Capitalize the first character.
optimisan
force-pushed
the
users/Akshat-Oke/10-30-_codegen_newpm_port_machinecycleinfo_to_npm
branch
from
68bcb36 to
be67756
Compare
optimisan
force-pushed
the
users/Akshat-Oke/11-14-_newpm_introduce_mfanalysisgetter_for_a_common_analysis_getter
branch
from
125b82d to
fa301f7
Compare
|
Updated MachineSink to use this #115434 |
optimisan
force-pushed
the
users/Akshat-Oke/10-30-_codegen_newpm_port_machinecycleinfo_to_npm
branch
from
be67756 to
751c8bf
Compare
optimisan
force-pushed
the
users/Akshat-Oke/11-14-_newpm_introduce_mfanalysisgetter_for_a_common_analysis_getter
branch
from
fa301f7 to
9343a47
Compare
There was a problem hiding this comment.
I'd really rather not do this. With the new pass manager transition for the optimization pipeline, it was fine to not have this abstraction, and other similar abstractions between the legacy/new pass manager (CallGraphUpdater) have caused issues due to subtle differences between the legacy/new pass manager. We're just going to have to tear it down anyway if we remove the legacy pass manager.
|
I see. I saw this kind of abstraction being used in Attributor so thought I'd make it available here. |
optimisan
deleted the
users/Akshat-Oke/11-14-_newpm_introduce_mfanalysisgetter_for_a_common_analysis_getter
branch
With this, there is no need to both
getAnalysis<AnaT>()andMFAM.getResult<AnaT>(MF)to pass in analysis from legacy/new pass managers.Instead just pass the
MFAnalysisGetterinto the pass implementation and useAG.getAnalysis<AnaT>(MF)once. It also detects outer analyses and routes it through the proxies itself.