gh-116772: Cache getenv result in optimizer.c to reduce calls
#126439
Conversation
Python/optimizer.c
Outdated
| static int uop_optimize_initialized = 0; | ||
| static int uop_optimize_flag = 0; |
There was a problem hiding this comment.
This isn't thread safe or subinterpreter safe. You might want to store the value in a PyThreadState / PyInterpreterState in pylifecycle.c instead.
There was a problem hiding this comment.
I gave it a try, but it didn’t work out in the end. I guess I don’t have enough experience yet. I'm really sorry, I might not be able to handle this on my own. Maybe someone else could take it, or maybe you could help improve what I have so far :(
|
@mdboom did you have a change planned for this? |
No, I don't have a plan. Thanks to @Wulian233 for taking this on. |
| initialize_uops_optimize_flag(void) { | ||
| if (!uops_optimize_initialized) { | ||
| PyInterpreterState *interp = _PyInterpreterState_GET(); | ||
| char *env_var = Py_GETENV("PYTHON_UOPS_OPTIMIZE"); |
There was a problem hiding this comment.
Can you please do this instead at where we check for PYTHON_JIT ? So in pylifecycle.c.
Set the thread state/interp state there as well.
4 participants
This is my first time PR for part C, which looks like it might be a small improvement, hopefully I'm not doing anything wrong🙂
About 0.5% faster? (from issue)
getenvon every optimization run #116772