We need some level of minimal profiling. Basic block (non-branching code) profilers report a number of machine clock cycles devoted to executing each line of code. Memory profiling may include cache hits and misses. Some minimal bar needs to be defined with regards to new instructions, and some context for what we expect an extension (and new instructions) to show. Simply bringing support to a tool (like Valgrind) isn't enough. We need to ensure the folks who depend on profiling have the tools they need.