[memprof] Use std::move in toMemProfRecord

[kazutakahirata]

std::move and reserve here result in a measurable speed-up in
llvm-profdata modified to deserialize all MemProfRecords. The cycle
count goes down by 7.1% while the instruction count goes down by 21%.

[snehasish]

This leaves the current IndexedMemProfRecord in a bad state. How can we ensure that users don't accidentally try to read after this method has been called?

I'm also skeptical how much impact this has in real usage since we don't expect to read all records for a single module. Perhaps we can try safer optimizations such as reserve() for the AllocSites vector (and the Callsites vector too)?

[kazutakahirata]

This leaves the current IndexedMemProfRecord in a bad state. How can we ensure that users don't accidentally try to read after this method has been called?

I'm also skeptical how much impact this has in real usage since we don't expect to read all records for a single module. Perhaps we can try safer optimizations such as reserve() for the AllocSites vector (and the Callsites vector too)?

AI here is a local variable. This patch does not modify the current IndexedMemProfRecord in any way. In fact, toMemProfRecord is a const function.

[snehasish]

AI here is a local variable. This patch does not modify the current IndexedMemProfRecord in any way. In fact, toMemProfRecord is a const function.

Thanks for pointing out local variable. Clearly I didn't spend much time reading the rest of the function. Can you try out the reserve calls and commit this along with it?

[snehasish]

lgtm