Don't bail out on side-effects in assertionprop #108418
Conversation
|
Tagging subscribers to this area: @JulieLeeMSFT, @jakobbotsch |
|
@dotnet/jit-contrib @AndyAyersMS PTAL, small change, Diffs TLDR: We have In order to allow arg0 to have side-effects. we simply create a temp inside LateArg and use that temp. Thus, PS: I recommend reviewing with "Hide whitespaces" option. |
| if ((castObjArg->gtFlags & GTF_ALL_EFFECT) != 0) | ||
| { | ||
| // It won't be trivial to properly extract side-effects from the call node. | ||
| // Ideally, we only need side effects from the castClsArg argument as the call itself | ||
| // won't throw any exceptions. But we should not forget about the EarlyNode (setup args) | ||
| return nullptr; | ||
| } |
There was a problem hiding this comment.
Why would gtExtractSideEffList(call, ..., /* ignoreRoot */ true) not be sufficient here? I don't think extracting the side effects of the operands of a call is all that hard.
There was a problem hiding this comment.
@jakobbotsch I am not sure I understand how it can work, say I have
call(arg0, arg1, arg2)
I want to fold it to just arg1 (or to e.g. arg1*arg1 in case of Math.Pow(x, 2)) and say arg1 has a side-effect in it.
if we just return arg1 node and wrap with gtExtractSideEffList we'll just duplicate the side effect
There was a problem hiding this comment.
Ah okay, I see -- that makes sense.
No description provided.