[CSSPGO] Enable stale profile matching by default for CSSPGO

[wlei]

Summary

We tested the stale profile matching on several Meta's internal services, all results are positive, for instance, in one service that refreshed its profile every one or two weeks, it consistently gave 1~2% performance improvement. We also observed an instance that a trivial refactoring caused a 2% regression and the matching can successfully recover the whole regression. Therefore, we'd like to turn it on by default for CSSPGO.

[wenlei]

lgtm, thanks!

[hoy]

Thanks for the change. Would be good to call out some rough numbers measured with our internal services.

llvm/lib/Transforms/IPO/SampleProfile.cpp

2052

Wrap this under a check of profileIsPseudoProbed?

[wlei]

addressing feedback.

[wenlei]

llvm/lib/Transforms/IPO/SampleProfile.cpp

2053

nit: combine the two ifs.

also add a comment why pseudo probe is needed.

[wlei]

addressing feedback, updated the comments.

llvm/lib/Transforms/IPO/SampleProfile.cpp

if (Reader->profileIsCS() || Reader->profileIsPreInlined() ||
     if (!AllowRecursiveInline.getNumOccurrences())
       AllowRecursiveInline = true;
 
     if (Reader->profileIsPreInlined()) {
       if (!UsePreInlinerDecision.getNumOccurrences())
         UsePreInlinerDecision = true;
     }
 
+    // Enable stale profile matching by default for probe-based profile.
+    // Currently the matching relies on if the checksum mismatch is detected,
+    // which is currently only available for pseudo-probe mode. Removing the
+    // checksum check could cause regressions for some cases, so further tuning
+    // might be needed if we want to enable it for all cases.
+    if (Reader->profileIsProbeBased() &&
+        !SalvageStaleProfile.getNumOccurrences()) {
+      SalvageStaleProfile = true;
+    }
+
     if (!Reader->profileIsCS()) {
       // Non-CS profile should be fine without a function size budget for the
       // inliner since the contexts in the profile are either all from inlining
       // in the prevoius build or pre-computed by the preinliner with a size
       // cap, thus they are bounded.
       if (!ProfileInlineLimitMin.getNumOccurrences())
         ProfileInlineLimitMin = std::numeric_limits<unsigned>::max();
       if (!ProfileInlineLimitMax.getNumOccurrences())