[mypyc] feat: add rare kwarg to IRBuilder.add_bool_branch
#20099
Conversation
|
I imagine it would be nicer if we bundled a change like this with optimization (mark branches with pragmas). Otherwise I feel like complexity doesn't automatically help! Also that would let the actual perf impacts be looked at (branch misprediction counts?). |
|
Well, it exists for this one case so it seems someone else decided this case was worthy enough to write a bunch of code for. I'm not the one who wrote that optimization, I'm just writing a tiny bit more code so other parts of the codebase can reuse it |
|
Ah I see, I missed that. |
|
@A5rocks originally I was trying to figure out how to use this I failed at that for the time being but that can still come down the pipe after this once I have more time to experiment with it |
| builder.read(instance), builder.none_object(), "is", line | ||
| ) | ||
| builder.add_bool_branch(comparison, class_block, instance_block) | ||
| builder.add_bool_branch(comparison, class_block, instance_block, rare=True) |
There was a problem hiding this comment.
this one might be questionable
2 participants
This doesn't do much, currently it enables just one optimization which already exists in the codebase: it helps mypyc determine whether to use CPyTagged_XDECREF or CPyTagged_XDecRef within a particular block
But it's semantically reasonable to accept a rare flag for any boolean branch, even if it doesn't do too much optimizing, and it opens the door for us to implement further optimizations in this area