Legacy Loop Pass Manager question


If we add multiple loop passes to the pass manager in PassManagerBuilder.cpp consecutively without any func/module pass in between, I used to think they would belong to the same loop pass manager. But it does not seem to be the case.

For example for this code snippet

PM.add(createIndVarSimplifyPass()); // Canonicalize indvars
MPM.add(createLoopIdiomPass()); // Recognize idioms like memset.
addExtensionsToPM(EP_LateLoopOptimizations, MPM);
MPM.add(createLoopDeletionPass()); // Delete dead loops

if (!DisableUnrollLoops)
MPM.add(createSimpleLoopUnrollPass(OptLevel)); // Unroll small loops

I see -debug-pass=Structure output:

Scalar Evolution Analysis
Loop Pass Manager
Induction Variable Simplification
Recognize loop idioms
Delete dead loops
Unroll loops

Which is in line to what I thought. However for this code snippet:

if (EnableUnrollAndJam) {
// Unroll and Jam. We do this before unroll but need to be in a separate
// loop pass manager in order for the outer loop to be processed by
// unroll and jam before the inner loop is unrolled.

MPM.add(createLoopUnrollPass(OptLevel)); // Unroll small loops

I see:

Loop-Closed SSA Form Pass
Loop Pass Manager
Unroll and Jam loops
Loop Pass Manager
Unroll loops
Lazy Branch Probability Analysis

Here two different loop pass managers were created. What is the difference in between these two cases? How was the loop pass manager split up in the second case?


This usually happens if an analysis required by LoopInfo is not
preserved. However, LoopDeletion, LoopUnrollAndJam and LoopUnroll's
getAnalysisUsage all call getLoopAnalysisUsage, which add the same
pass preservations.

I'd debug and step through the loop manager's code to find what makes
the cases different.


Thanks for the pointers Michael. I’ll investigate that and report back with my findings.