We want all operations that load values from the heap to have value profiling. The Resolve opcodes are the last ones that don't.
Created attachment 108546 [details]
This is pretty much performance neutral. There is a spurious, but totally reproducible, 0.3% slow-down in V8 on my MacBook Pro. There are similarly sized, though not as reproducible, speed-ups in SunSpider and Kraken. I checked and none of the benchmarks that have performance variations make any use of any of the affected opcodes. So these are just insignificant variations due to code layout.
Landed in r95887.