LLVM: Run std-compile-opts on code in the execution engine
Bug #782006 reported by
Matt Giuca
This bug affects 1 person
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
Mars |
Triaged
|
Low
|
Matt Giuca |
Bug Description
Currently, code is not run through any LLVM optimisation passes (although if using marsc, the user can run it through 'opt' afterwards). Mars and Marsc should automatically run std-compile-opts (which can be done through the llvm_optimize module). There should be a compiler flag to turn this on or off.
I haven't found an easy way to ask for "all of the passes in std-compile-opts". Hopefully I don't need to explicitly enumerate all of the optimisation passes.
To post a comment you must log in.
I'm not sure if this is necessary. I recall doing some basic speed tests and found that the interpreter code was actually faster than the compiled code, due to JIT. I think the performance characteristics are simply different, and you don't need to explicitly turn on compiler optimization. I could be wrong though; keeping this bug open as a low priority.