[llvm] 4fa3280 - [NewPM][Pipeline] Add PipelineTuningOption to set inliner threshold

[Arthur Eubanks via llvm-commits]

Author: Arthur Eubanks
Date: 2022-11-02T10:47:51-07:00
New Revision: 4fa328074efd7eefdbb314b8f6e9f855e443ca20

URL: https://github.com/llvm/llvm-project/commit/4fa328074efd7eefdbb314b8f6e9f855e443ca20
DIFF: https://github.com/llvm/llvm-project/commit/4fa328074efd7eefdbb314b8f6e9f855e443ca20.diff

LOG: [NewPM][Pipeline] Add PipelineTuningOption to set inliner threshold

The legacy PM allowed you to set a custom inliner threshold via
  builder.Inliner = llvm::createFunctionInliningPass(inline_threshold);

This allows the same thing to be done with the new PM optimization pipelines.

Reviewed By: asbirlea

Differential Revision: https://reviews.llvm.org/D137038

Added: 
    

Modified: 
    llvm/include/llvm/Passes/PassBuilder.h
    llvm/lib/Passes/PassBuilderPipelines.cpp

Removed: 
    


################################################################################
diff  --git a/llvm/include/llvm/Passes/PassBuilder.h b/llvm/include/llvm/Passes/PassBuilder.h
index 68f07f39afa63..d7d76c30d1a6b 100644
--- a/llvm/include/llvm/Passes/PassBuilder.h
+++ b/llvm/include/llvm/Passes/PassBuilder.h
@@ -75,6 +75,9 @@ class PipelineTuningOptions {
   /// false.
   bool MergeFunctions;
 
+  /// Tuning option to override the default inliner threshold.
+  int InlinerThreshold;
+
   // Experimental option to eagerly invalidate more analyses. This has the
   // potential to decrease max memory usage in exchange for more compile time.
   // This may affect codegen due to either passes using analyses only when

diff  --git a/llvm/lib/Passes/PassBuilderPipelines.cpp b/llvm/lib/Passes/PassBuilderPipelines.cpp
index 4f1ef22775db8..b32d53d8dfaf8 100644
--- a/llvm/lib/Passes/PassBuilderPipelines.cpp
+++ b/llvm/lib/Passes/PassBuilderPipelines.cpp
@@ -200,6 +200,7 @@ PipelineTuningOptions::PipelineTuningOptions() {
   LicmMssaNoAccForPromotionCap = SetLicmMssaNoAccForPromotionCap;
   CallGraphProfile = true;
   MergeFunctions = EnableMergeFunctions;
+  InlinerThreshold = -1;
   EagerlyInvalidateAnalyses = EnableEagerlyInvalidateAnalyses;
 }
 
@@ -719,7 +720,11 @@ static InlineParams getInlineParamsFromOptLevel(OptimizationLevel Level) {
 ModuleInlinerWrapperPass
 PassBuilder::buildInlinerPipeline(OptimizationLevel Level,
                                   ThinOrFullLTOPhase Phase) {
-  InlineParams IP = getInlineParamsFromOptLevel(Level);
+  InlineParams IP;
+  if (PTO.InlinerThreshold == -1)
+    IP = getInlineParamsFromOptLevel(Level);
+  else
+    IP = getInlineParams(PTO.InlinerThreshold);
   // For PreLinkThinLTO + SamplePGO, set hot-caller threshold to 0 to
   // disable hot callsite inline (as much as possible [1]) because it makes
   // profile annotation in the backend inaccurate.