lgtm

lgtm

This might work and existing test cases should be able to make sure the
behavior is not regressed.

My profile shows that with the 20000 phi case, the cycles are cut in half
with the limit.

Done. Added comment.

Can you add a new test case testing the new emitted code patterns?

lgtm

The behavior will change if the shared libraries are not 'dlopen'ed but loaded at program startup time.

With this change, gcov_flush will resolve to the copy in the main executable for each shared library -- this is not the desired the behavior.

lgtm

lgtm

lgtm

I added Easwaran to get a closer look at the change.

lgtm

lgtm

Also add an dead vararg elimination test case for better coverage.

lgtm

Can this be done under an option?

LGTM

lgtm

LGTM

Please add a few new test cases first.

lgtm

The branch probability depends on other factors which is not considered here.

LGTM + incorporating Ben's suggestion to use 'ult' if it can simplify the condition.

lgtm

looks like the naming follows the node iteration interfaces. lgtm.

lgtm

I actually mean the former.

lgtm

Add a minimal option test (new heat-color option) in test/Other/2007-06-05-PassID.ll ?

I think 'tryToPromote' method should be moved to the indirect call promotion pass so that the callback to inline cost/benefit is exposed there (a more general cost/benefit model needs to be developed for indirect call promotion). Initially, I think we can limit the use of the getInlineCost callback to cases where profile data is not available, this will achieve what this patch is doing without affecting existing promotion heuristics.

lgtm

Perhaps change the new interface documentation slightly to reflect that it is for real profile data.

I prefer hasProfileData to be hasRealProfileData -- unless we want to make hasProfileData to mean has real profile data and ignores synthetic entry count later.

lgtm

The instrumentation point will be before the no-return call, so there is probably no issue with instrumenting direct successor of the entry.

address comment: remove unnecessary early return check.

lgtm

Ping ..

there are two small suggestions. otherwise LGTM

I suggest making this a multiple stage change.

Is it enough to simply replace StringMap with std::map<std::string, ..> for small maps? Do you have memory data with that?

lgtm